blob: e889779c9b37d67cc2baeb9f9cb66367c8fb8feb [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Chris Mason8f18cf12008-04-25 16:53:30 -040019#include <linux/kernel.h>
Chris Mason065631f2008-02-20 12:07:25 -050020#include <linux/bio.h>
Chris Mason39279cc2007-06-12 06:35:45 -040021#include <linux/buffer_head.h>
Sage Weilf2eb0a22008-05-02 14:43:14 -040022#include <linux/file.h>
Chris Mason39279cc2007-06-12 06:35:45 -040023#include <linux/fs.h>
24#include <linux/pagemap.h>
25#include <linux/highmem.h>
26#include <linux/time.h>
27#include <linux/init.h>
28#include <linux/string.h>
Chris Mason39279cc2007-06-12 06:35:45 -040029#include <linux/backing-dev.h>
30#include <linux/mpage.h>
31#include <linux/swap.h>
32#include <linux/writeback.h>
33#include <linux/statfs.h>
34#include <linux/compat.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070035#include <linux/aio.h>
Chris Mason9ebefb182007-06-15 13:50:00 -040036#include <linux/bit_spinlock.h>
Josef Bacik5103e942007-11-16 11:45:54 -050037#include <linux/xattr.h>
Josef Bacik33268ea2008-07-24 12:16:36 -040038#include <linux/posix_acl.h>
Yan Zhengd899e052008-10-30 14:25:28 -040039#include <linux/falloc.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090040#include <linux/slab.h>
David Sterba7a36dde2011-05-06 15:33:15 +020041#include <linux/ratelimit.h>
Josef Bacik22c44fe2011-11-30 10:45:38 -050042#include <linux/mount.h>
Filipe Brandenburger55e301f2013-01-29 06:04:50 +000043#include <linux/btrfs.h>
David Woodhouse53b381b2013-01-29 18:40:14 -050044#include <linux/blkdev.h>
Josef Bacikf23b5a52013-06-19 10:16:26 -040045#include <linux/posix_acl_xattr.h>
Chris Mason39279cc2007-06-12 06:35:45 -040046#include "ctree.h"
47#include "disk-io.h"
48#include "transaction.h"
49#include "btrfs_inode.h"
Chris Mason39279cc2007-06-12 06:35:45 -040050#include "print-tree.h"
Chris Masone6dcd2d2008-07-17 12:53:50 -040051#include "ordered-data.h"
Christoph Hellwig95819c02008-08-28 06:21:17 -040052#include "xattr.h"
Chris Masone02119d2008-09-05 16:13:11 -040053#include "tree-log.h"
Jan Schmidt4a54c8c2011-07-22 15:41:52 +020054#include "volumes.h"
Chris Masonc8b97812008-10-29 14:49:59 -040055#include "compression.h"
Chris Masonb4ce94d2009-02-04 09:25:08 -050056#include "locking.h"
Josef Bacikdc89e982011-01-28 17:05:48 -050057#include "free-space-cache.h"
Li Zefan581bb052011-04-20 10:06:11 +080058#include "inode-map.h"
Liu Bo38c227d2013-01-29 03:18:40 +000059#include "backref.h"
Josef Bacikf23b5a52013-06-19 10:16:26 -040060#include "hash.h"
Chris Mason39279cc2007-06-12 06:35:45 -040061
62struct btrfs_iget_args {
63 u64 ino;
64 struct btrfs_root *root;
65};
66
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -070067static const struct inode_operations btrfs_dir_inode_operations;
68static const struct inode_operations btrfs_symlink_inode_operations;
69static const struct inode_operations btrfs_dir_ro_inode_operations;
70static const struct inode_operations btrfs_special_inode_operations;
71static const struct inode_operations btrfs_file_inode_operations;
Alexey Dobriyan7f094102009-09-21 17:01:10 -070072static const struct address_space_operations btrfs_aops;
73static const struct address_space_operations btrfs_symlink_aops;
Alexey Dobriyan828c0952009-10-01 15:43:56 -070074static const struct file_operations btrfs_dir_file_operations;
Chris Masond1310b22008-01-24 16:13:08 -050075static struct extent_io_ops btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -040076
77static struct kmem_cache *btrfs_inode_cachep;
Miao Xie8ccf6f192012-10-25 09:28:04 +000078static struct kmem_cache *btrfs_delalloc_work_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040079struct kmem_cache *btrfs_trans_handle_cachep;
80struct kmem_cache *btrfs_transaction_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040081struct kmem_cache *btrfs_path_cachep;
Josef Bacikdc89e982011-01-28 17:05:48 -050082struct kmem_cache *btrfs_free_space_cachep;
Chris Mason39279cc2007-06-12 06:35:45 -040083
84#define S_SHIFT 12
85static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
86 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
87 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
88 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
89 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
90 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
91 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
92 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
93};
94
Eric Sandeen3972f262013-01-12 02:57:22 +000095static int btrfs_setsize(struct inode *inode, struct iattr *attr);
Josef Bacika41ad392011-01-31 15:30:16 -050096static int btrfs_truncate(struct inode *inode);
Josef Bacik5fd02042012-05-02 14:00:54 -040097static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
Chris Mason771ed682008-11-06 22:02:51 -050098static noinline int cow_file_range(struct inode *inode,
99 struct page *locked_page,
100 u64 start, u64 end, int *page_started,
101 unsigned long *nr_written, int unlock);
Josef Bacik70c8a912012-10-11 16:54:30 -0400102static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
103 u64 len, u64 orig_start,
104 u64 block_start, u64 block_len,
Josef Bacikcc95bef2013-04-04 14:31:27 -0400105 u64 orig_block_len, u64 ram_bytes,
106 int type);
Josef Bacik7b128762008-07-24 12:17:14 -0400107
Eric Sandeen48a3b632013-04-25 20:41:01 +0000108static int btrfs_dirty_inode(struct inode *inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400109
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000110static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
Eric Paris2a7dba32011-02-01 11:05:39 -0500111 struct inode *inode, struct inode *dir,
112 const struct qstr *qstr)
Jim Owens0279b4c2009-02-04 09:29:13 -0500113{
114 int err;
115
Yan, Zhengf34f57a2009-11-12 09:35:27 +0000116 err = btrfs_init_acl(trans, inode, dir);
Jim Owens0279b4c2009-02-04 09:29:13 -0500117 if (!err)
Eric Paris2a7dba32011-02-01 11:05:39 -0500118 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
Jim Owens0279b4c2009-02-04 09:29:13 -0500119 return err;
120}
121
Chris Masond352ac62008-09-29 15:18:18 -0400122/*
Chris Masonc8b97812008-10-29 14:49:59 -0400123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
126 */
Chris Masond3977122009-01-05 21:25:51 -0500127static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
Chris Masonc8b97812008-10-29 14:49:59 -0400128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000130 int compress_type,
Chris Masonc8b97812008-10-29 14:49:59 -0400131 struct page **compressed_pages)
132{
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
137 char *kaddr;
138 unsigned long ptr;
139 struct btrfs_file_extent_item *ei;
140 int err = 0;
141 int ret;
142 size_t cur_size = size;
143 size_t datasize;
144 unsigned long offset;
Chris Masonc8b97812008-10-29 14:49:59 -0400145
Li Zefanfe3f5662011-03-28 08:30:38 +0000146 if (compressed_size && compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400147 cur_size = compressed_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400148
Chris Masond3977122009-01-05 21:25:51 -0500149 path = btrfs_alloc_path();
150 if (!path)
Chris Masonc8b97812008-10-29 14:49:59 -0400151 return -ENOMEM;
152
Chris Masonb9473432009-03-13 11:00:37 -0400153 path->leave_spinning = 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400154
Li Zefan33345d012011-04-20 10:31:50 +0800155 key.objectid = btrfs_ino(inode);
Chris Masonc8b97812008-10-29 14:49:59 -0400156 key.offset = start;
157 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
Chris Masonc8b97812008-10-29 14:49:59 -0400158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159
160 inode_add_bytes(inode, size);
161 ret = btrfs_insert_empty_item(trans, root, path, &key,
162 datasize);
Chris Masonc8b97812008-10-29 14:49:59 -0400163 if (ret) {
164 err = ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400165 goto fail;
166 }
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
176
Li Zefan261507a02010-12-17 14:21:50 +0800177 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400178 struct page *cpage;
179 int i = 0;
Chris Masond3977122009-01-05 21:25:51 -0500180 while (compressed_size > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400181 cpage = compressed_pages[i];
Chris Mason5b050f02008-11-11 09:34:41 -0500182 cur_size = min_t(unsigned long, compressed_size,
Chris Masonc8b97812008-10-29 14:49:59 -0400183 PAGE_CACHE_SIZE);
184
Cong Wang7ac687d2011-11-25 23:14:28 +0800185 kaddr = kmap_atomic(cpage);
Chris Masonc8b97812008-10-29 14:49:59 -0400186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800187 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400188
189 i++;
190 ptr += cur_size;
191 compressed_size -= cur_size;
192 }
193 btrfs_set_file_extent_compression(leaf, ei,
Li Zefan261507a02010-12-17 14:21:50 +0800194 compress_type);
Chris Masonc8b97812008-10-29 14:49:59 -0400195 } else {
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
Cong Wang7ac687d2011-11-25 23:14:28 +0800199 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
Cong Wang7ac687d2011-11-25 23:14:28 +0800202 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400203 page_cache_release(page);
204 }
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_free_path(path);
207
Yan, Zhengc2167752009-11-12 09:34:21 +0000208 /*
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
212 *
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
216 */
Chris Masonc8b97812008-10-29 14:49:59 -0400217 BTRFS_I(inode)->disk_i_size = inode->i_size;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100218 ret = btrfs_update_inode(trans, root, inode);
Yan, Zhengc2167752009-11-12 09:34:21 +0000219
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100220 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400221fail:
222 btrfs_free_path(path);
223 return err;
224}
225
226
227/*
228 * conditionally insert an inline extent into the file. This
229 * does the checks required to make sure the data is small enough
230 * to fit as an inline extent.
231 */
Josef Bacik00361582013-08-14 14:02:47 -0400232static noinline int cow_file_range_inline(struct btrfs_root *root,
233 struct inode *inode, u64 start,
234 u64 end, size_t compressed_size,
235 int compress_type,
236 struct page **compressed_pages)
Chris Masonc8b97812008-10-29 14:49:59 -0400237{
Josef Bacik00361582013-08-14 14:02:47 -0400238 struct btrfs_trans_handle *trans;
Chris Masonc8b97812008-10-29 14:49:59 -0400239 u64 isize = i_size_read(inode);
240 u64 actual_end = min(end + 1, isize);
241 u64 inline_len = actual_end - start;
Qu Wenruofda28322013-02-26 08:10:22 +0000242 u64 aligned_end = ALIGN(end, root->sectorsize);
Chris Masonc8b97812008-10-29 14:49:59 -0400243 u64 data_len = inline_len;
244 int ret;
245
246 if (compressed_size)
247 data_len = compressed_size;
248
249 if (start > 0 ||
Chris Mason70b99e62008-10-31 12:46:39 -0400250 actual_end >= PAGE_CACHE_SIZE ||
Chris Masonc8b97812008-10-29 14:49:59 -0400251 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
252 (!compressed_size &&
253 (actual_end & (root->sectorsize - 1)) == 0) ||
254 end + 1 < isize ||
255 data_len > root->fs_info->max_inline) {
256 return 1;
257 }
258
Josef Bacik00361582013-08-14 14:02:47 -0400259 trans = btrfs_join_transaction(root);
260 if (IS_ERR(trans))
261 return PTR_ERR(trans);
262 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
263
Josef Bacik26714852012-08-29 12:24:27 -0400264 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
Josef Bacik00361582013-08-14 14:02:47 -0400265 if (ret) {
266 btrfs_abort_transaction(trans, root, ret);
267 goto out;
268 }
Chris Masonc8b97812008-10-29 14:49:59 -0400269
270 if (isize > actual_end)
271 inline_len = min_t(u64, isize, actual_end);
272 ret = insert_inline_extent(trans, root, inode, start,
273 inline_len, compressed_size,
Li Zefanfe3f5662011-03-28 08:30:38 +0000274 compress_type, compressed_pages);
Josef Bacik2adcac12012-05-23 16:10:14 -0400275 if (ret && ret != -ENOSPC) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100276 btrfs_abort_transaction(trans, root, ret);
Josef Bacik00361582013-08-14 14:02:47 -0400277 goto out;
Josef Bacik2adcac12012-05-23 16:10:14 -0400278 } else if (ret == -ENOSPC) {
Josef Bacik00361582013-08-14 14:02:47 -0400279 ret = 1;
280 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100281 }
Josef Bacik2adcac12012-05-23 16:10:14 -0400282
Josef Bacikbdc20e62013-02-28 13:23:38 -0500283 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -0400284 btrfs_delalloc_release_metadata(inode, end + 1 - start);
Chris Masona1ed8352009-09-11 12:27:37 -0400285 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Josef Bacik00361582013-08-14 14:02:47 -0400286out:
287 btrfs_end_transaction(trans, root);
288 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400289}
290
Chris Mason771ed682008-11-06 22:02:51 -0500291struct async_extent {
292 u64 start;
293 u64 ram_size;
294 u64 compressed_size;
295 struct page **pages;
296 unsigned long nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800297 int compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500298 struct list_head list;
299};
300
301struct async_cow {
302 struct inode *inode;
303 struct btrfs_root *root;
304 struct page *locked_page;
305 u64 start;
306 u64 end;
307 struct list_head extents;
308 struct btrfs_work work;
309};
310
311static noinline int add_async_extent(struct async_cow *cow,
312 u64 start, u64 ram_size,
313 u64 compressed_size,
314 struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +0800315 unsigned long nr_pages,
316 int compress_type)
Chris Mason771ed682008-11-06 22:02:51 -0500317{
318 struct async_extent *async_extent;
319
320 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100321 BUG_ON(!async_extent); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -0500322 async_extent->start = start;
323 async_extent->ram_size = ram_size;
324 async_extent->compressed_size = compressed_size;
325 async_extent->pages = pages;
326 async_extent->nr_pages = nr_pages;
Li Zefan261507a02010-12-17 14:21:50 +0800327 async_extent->compress_type = compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500328 list_add_tail(&async_extent->list, &cow->extents);
329 return 0;
330}
331
Chris Masonc8b97812008-10-29 14:49:59 -0400332/*
Chris Mason771ed682008-11-06 22:02:51 -0500333 * we create compressed extents in two phases. The first
334 * phase compresses a range of pages that have already been
335 * locked (both pages and state bits are locked).
Chris Masonc8b97812008-10-29 14:49:59 -0400336 *
Chris Mason771ed682008-11-06 22:02:51 -0500337 * This is done inside an ordered work queue, and the compression
338 * is spread across many cpus. The actual IO submission is step
339 * two, and the ordered work queue takes care of making sure that
340 * happens in the same order things were put onto the queue by
341 * writepages and friends.
Chris Masonc8b97812008-10-29 14:49:59 -0400342 *
Chris Mason771ed682008-11-06 22:02:51 -0500343 * If this code finds it can't get good compression, it puts an
344 * entry onto the work queue to write the uncompressed bytes. This
345 * makes sure that both compressed inodes and uncompressed inodes
Artem Bityutskiyb2570312012-07-25 18:12:06 +0300346 * are written in the same order that the flusher thread sent them
347 * down.
Chris Masond352ac62008-09-29 15:18:18 -0400348 */
Chris Mason771ed682008-11-06 22:02:51 -0500349static noinline int compress_file_range(struct inode *inode,
350 struct page *locked_page,
351 u64 start, u64 end,
352 struct async_cow *async_cow,
353 int *num_added)
Chris Masonb888db22007-08-27 16:49:44 -0400354{
355 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masondb945352007-10-15 16:15:53 -0400356 u64 num_bytes;
Chris Masondb945352007-10-15 16:15:53 -0400357 u64 blocksize = root->sectorsize;
Chris Masonc8b97812008-10-29 14:49:59 -0400358 u64 actual_end;
Chris Mason42dc7ba2008-12-15 11:44:56 -0500359 u64 isize = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400360 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400361 struct page **pages = NULL;
362 unsigned long nr_pages;
363 unsigned long nr_pages_ret = 0;
364 unsigned long total_compressed = 0;
365 unsigned long total_in = 0;
366 unsigned long max_compressed = 128 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -0500367 unsigned long max_uncompressed = 128 * 1024;
Chris Masonc8b97812008-10-29 14:49:59 -0400368 int i;
369 int will_compress;
Li Zefan261507a02010-12-17 14:21:50 +0800370 int compress_type = root->fs_info->compress_type;
Chris Mason4adaa612013-03-26 13:07:00 -0400371 int redirty = 0;
Chris Masonb888db22007-08-27 16:49:44 -0400372
Liu Bo4cb13e52012-03-29 09:57:45 -0400373 /* if this is a small write inside eof, kick off a defrag */
374 if ((end - start + 1) < 16 * 1024 &&
375 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Chris Mason4cb53002011-05-24 15:35:30 -0400376 btrfs_add_inode_defrag(NULL, inode);
377
Chris Mason42dc7ba2008-12-15 11:44:56 -0500378 actual_end = min_t(u64, isize, end + 1);
Chris Masonc8b97812008-10-29 14:49:59 -0400379again:
380 will_compress = 0;
381 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
382 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
383
Chris Masonf03d9301f2009-02-04 09:31:06 -0500384 /*
385 * we don't want to send crud past the end of i_size through
386 * compression, that's just a waste of CPU time. So, if the
387 * end of the file is before the start of our current
388 * requested range of bytes, we bail out to the uncompressed
389 * cleanup code that can deal with all of this.
390 *
391 * It isn't really the fastest way to fix things, but this is a
392 * very uncommon corner.
393 */
394 if (actual_end <= start)
395 goto cleanup_and_bail_uncompressed;
396
Chris Masonc8b97812008-10-29 14:49:59 -0400397 total_compressed = actual_end - start;
398
399 /* we want to make sure that amount of ram required to uncompress
400 * an extent is reasonable, so we limit the total size in ram
Chris Mason771ed682008-11-06 22:02:51 -0500401 * of a compressed extent to 128k. This is a crucial number
402 * because it also controls how easily we can spread reads across
403 * cpus for decompression.
404 *
405 * We also want to make sure the amount of IO required to do
406 * a random read is reasonably small, so we limit the size of
407 * a compressed extent to 128k.
Chris Masonc8b97812008-10-29 14:49:59 -0400408 */
409 total_compressed = min(total_compressed, max_uncompressed);
Qu Wenruofda28322013-02-26 08:10:22 +0000410 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Masonbe20aa92007-12-17 20:14:01 -0500411 num_bytes = max(blocksize, num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -0400412 total_in = 0;
413 ret = 0;
Chris Masondb945352007-10-15 16:15:53 -0400414
Chris Mason771ed682008-11-06 22:02:51 -0500415 /*
416 * we do compression for mount -o compress and when the
417 * inode has not been flagged as nocompress. This flag can
418 * change at any time if we discover bad compression ratios.
Chris Masonc8b97812008-10-29 14:49:59 -0400419 */
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200420 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
Chris Mason1e701a32010-03-11 09:42:04 -0500421 (btrfs_test_opt(root, COMPRESS) ||
Liu Bo75e7cb72011-03-22 10:12:20 +0000422 (BTRFS_I(inode)->force_compress) ||
423 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
Chris Masonc8b97812008-10-29 14:49:59 -0400424 WARN_ON(pages);
Chris Masoncfbc2462008-10-30 13:22:14 -0400425 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
Li Zefan560f7d72011-09-08 10:22:01 +0800426 if (!pages) {
427 /* just bail out to the uncompressed code */
428 goto cont;
429 }
Chris Mason179e29e2007-11-01 11:28:41 -0400430
Li Zefan261507a02010-12-17 14:21:50 +0800431 if (BTRFS_I(inode)->force_compress)
432 compress_type = BTRFS_I(inode)->force_compress;
433
Chris Mason4adaa612013-03-26 13:07:00 -0400434 /*
435 * we need to call clear_page_dirty_for_io on each
436 * page in the range. Otherwise applications with the file
437 * mmap'd can wander in and change the page contents while
438 * we are compressing them.
439 *
440 * If the compression fails for any reason, we set the pages
441 * dirty again later on.
442 */
443 extent_range_clear_dirty_for_io(inode, start, end);
444 redirty = 1;
Li Zefan261507a02010-12-17 14:21:50 +0800445 ret = btrfs_compress_pages(compress_type,
446 inode->i_mapping, start,
447 total_compressed, pages,
448 nr_pages, &nr_pages_ret,
449 &total_in,
450 &total_compressed,
451 max_compressed);
Chris Masonc8b97812008-10-29 14:49:59 -0400452
453 if (!ret) {
454 unsigned long offset = total_compressed &
455 (PAGE_CACHE_SIZE - 1);
456 struct page *page = pages[nr_pages_ret - 1];
457 char *kaddr;
458
459 /* zero the tail end of the last page, we might be
460 * sending it down to disk
461 */
462 if (offset) {
Cong Wang7ac687d2011-11-25 23:14:28 +0800463 kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400464 memset(kaddr + offset, 0,
465 PAGE_CACHE_SIZE - offset);
Cong Wang7ac687d2011-11-25 23:14:28 +0800466 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -0400467 }
468 will_compress = 1;
469 }
470 }
Li Zefan560f7d72011-09-08 10:22:01 +0800471cont:
Chris Masonc8b97812008-10-29 14:49:59 -0400472 if (start == 0) {
473 /* lets try to make an inline extent */
Chris Mason771ed682008-11-06 22:02:51 -0500474 if (ret || total_in < (actual_end - start)) {
Chris Masonc8b97812008-10-29 14:49:59 -0400475 /* we didn't compress the entire range, try
Chris Mason771ed682008-11-06 22:02:51 -0500476 * to make an uncompressed inline extent.
Chris Masonc8b97812008-10-29 14:49:59 -0400477 */
Josef Bacik00361582013-08-14 14:02:47 -0400478 ret = cow_file_range_inline(root, inode, start, end,
479 0, 0, NULL);
Chris Masonc8b97812008-10-29 14:49:59 -0400480 } else {
Chris Mason771ed682008-11-06 22:02:51 -0500481 /* try making a compressed inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400482 ret = cow_file_range_inline(root, inode, start, end,
Li Zefanfe3f5662011-03-28 08:30:38 +0000483 total_compressed,
484 compress_type, pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400485 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100486 if (ret <= 0) {
Josef Bacik151a41b2013-07-29 13:22:24 -0400487 unsigned long clear_flags = EXTENT_DELALLOC |
488 EXTENT_DEFRAG;
489 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
490
Chris Mason771ed682008-11-06 22:02:51 -0500491 /*
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100492 * inline extent creation worked or returned error,
493 * we don't need to create any more async work items.
494 * Unlock and free up our temp pages.
Chris Mason771ed682008-11-06 22:02:51 -0500495 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400496 extent_clear_unlock_delalloc(inode, start, end, NULL,
Josef Bacik151a41b2013-07-29 13:22:24 -0400497 clear_flags, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400498 PAGE_CLEAR_DIRTY |
499 PAGE_SET_WRITEBACK |
500 PAGE_END_WRITEBACK);
Chris Masonc8b97812008-10-29 14:49:59 -0400501 goto free_pages_out;
502 }
503 }
504
505 if (will_compress) {
506 /*
507 * we aren't doing an inline extent round the compressed size
508 * up to a block size boundary so the allocator does sane
509 * things
510 */
Qu Wenruofda28322013-02-26 08:10:22 +0000511 total_compressed = ALIGN(total_compressed, blocksize);
Chris Masonc8b97812008-10-29 14:49:59 -0400512
513 /*
514 * one last check to make sure the compression is really a
515 * win, compare the page count read with the blocks on disk
516 */
Qu Wenruofda28322013-02-26 08:10:22 +0000517 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
Chris Masonc8b97812008-10-29 14:49:59 -0400518 if (total_compressed >= total_in) {
519 will_compress = 0;
520 } else {
Chris Masonc8b97812008-10-29 14:49:59 -0400521 num_bytes = total_in;
522 }
523 }
524 if (!will_compress && pages) {
525 /*
526 * the compression code ran but failed to make things smaller,
527 * free any pages it allocated and our page pointer array
528 */
529 for (i = 0; i < nr_pages_ret; i++) {
Chris Mason70b99e62008-10-31 12:46:39 -0400530 WARN_ON(pages[i]->mapping);
Chris Masonc8b97812008-10-29 14:49:59 -0400531 page_cache_release(pages[i]);
532 }
533 kfree(pages);
534 pages = NULL;
535 total_compressed = 0;
536 nr_pages_ret = 0;
537
538 /* flag the file so we don't compress in the future */
Chris Mason1e701a32010-03-11 09:42:04 -0500539 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
540 !(BTRFS_I(inode)->force_compress)) {
Chris Masona555f812010-01-28 16:18:15 -0500541 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
Chris Mason1e701a32010-03-11 09:42:04 -0500542 }
Chris Masonc8b97812008-10-29 14:49:59 -0400543 }
Chris Mason771ed682008-11-06 22:02:51 -0500544 if (will_compress) {
545 *num_added += 1;
546
547 /* the async work queues will take care of doing actual
548 * allocation on disk for these compressed pages,
549 * and will submit them to the elevator.
550 */
551 add_async_extent(async_cow, start, num_bytes,
Li Zefan261507a02010-12-17 14:21:50 +0800552 total_compressed, pages, nr_pages_ret,
553 compress_type);
Chris Mason771ed682008-11-06 22:02:51 -0500554
Yan, Zheng24ae6362010-12-06 07:02:36 +0000555 if (start + num_bytes < end) {
Chris Mason771ed682008-11-06 22:02:51 -0500556 start += num_bytes;
557 pages = NULL;
558 cond_resched();
559 goto again;
560 }
561 } else {
Chris Masonf03d9301f2009-02-04 09:31:06 -0500562cleanup_and_bail_uncompressed:
Chris Mason771ed682008-11-06 22:02:51 -0500563 /*
564 * No compression, but we still need to write the pages in
565 * the file we've been given so far. redirty the locked
566 * page if it corresponds to our extent and set things up
567 * for the async work queue to run cow_file_range to do
568 * the normal delalloc dance
569 */
570 if (page_offset(locked_page) >= start &&
571 page_offset(locked_page) <= end) {
572 __set_page_dirty_nobuffers(locked_page);
573 /* unlocked later on in the async handlers */
574 }
Chris Mason4adaa612013-03-26 13:07:00 -0400575 if (redirty)
576 extent_range_redirty_for_io(inode, start, end);
Li Zefan261507a02010-12-17 14:21:50 +0800577 add_async_extent(async_cow, start, end - start + 1,
578 0, NULL, 0, BTRFS_COMPRESS_NONE);
Chris Mason771ed682008-11-06 22:02:51 -0500579 *num_added += 1;
580 }
581
582out:
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100583 return ret;
Chris Mason771ed682008-11-06 22:02:51 -0500584
585free_pages_out:
586 for (i = 0; i < nr_pages_ret; i++) {
587 WARN_ON(pages[i]->mapping);
588 page_cache_release(pages[i]);
589 }
Chris Masond3977122009-01-05 21:25:51 -0500590 kfree(pages);
Chris Mason771ed682008-11-06 22:02:51 -0500591
592 goto out;
593}
594
595/*
596 * phase two of compressed writeback. This is the ordered portion
597 * of the code, which only gets called in the order the work was
598 * queued. We walk all the async extents created by compress_file_range
599 * and send them down to the disk.
600 */
601static noinline int submit_compressed_extents(struct inode *inode,
602 struct async_cow *async_cow)
603{
604 struct async_extent *async_extent;
605 u64 alloc_hint = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500606 struct btrfs_key ins;
607 struct extent_map *em;
608 struct btrfs_root *root = BTRFS_I(inode)->root;
609 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
610 struct extent_io_tree *io_tree;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500611 int ret = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500612
613 if (list_empty(&async_cow->extents))
614 return 0;
615
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500616again:
Chris Masond3977122009-01-05 21:25:51 -0500617 while (!list_empty(&async_cow->extents)) {
Chris Mason771ed682008-11-06 22:02:51 -0500618 async_extent = list_entry(async_cow->extents.next,
619 struct async_extent, list);
620 list_del(&async_extent->list);
621
622 io_tree = &BTRFS_I(inode)->io_tree;
623
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500624retry:
Chris Mason771ed682008-11-06 22:02:51 -0500625 /* did the compression code fall back to uncompressed IO? */
626 if (!async_extent->pages) {
627 int page_started = 0;
628 unsigned long nr_written = 0;
629
630 lock_extent(io_tree, async_extent->start,
Josef Bacik2ac55d42010-02-03 19:33:23 +0000631 async_extent->start +
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100632 async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500633
634 /* allocate blocks */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500635 ret = cow_file_range(inode, async_cow->locked_page,
636 async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1,
639 &page_started, &nr_written, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500640
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100641 /* JDM XXX */
642
Chris Mason771ed682008-11-06 22:02:51 -0500643 /*
644 * if page_started, cow_file_range inserted an
645 * inline extent and took care of all the unlocking
646 * and IO for us. Otherwise, we need to submit
647 * all those pages down to the drive.
648 */
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500649 if (!page_started && !ret)
Chris Mason771ed682008-11-06 22:02:51 -0500650 extent_write_locked_range(io_tree,
651 inode, async_extent->start,
Chris Masond3977122009-01-05 21:25:51 -0500652 async_extent->start +
Chris Mason771ed682008-11-06 22:02:51 -0500653 async_extent->ram_size - 1,
654 btrfs_get_extent,
655 WB_SYNC_ALL);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500656 else if (ret)
657 unlock_page(async_cow->locked_page);
Chris Mason771ed682008-11-06 22:02:51 -0500658 kfree(async_extent);
659 cond_resched();
660 continue;
661 }
662
663 lock_extent(io_tree, async_extent->start,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100664 async_extent->start + async_extent->ram_size - 1);
Chris Mason771ed682008-11-06 22:02:51 -0500665
Josef Bacik00361582013-08-14 14:02:47 -0400666 ret = btrfs_reserve_extent(root,
Chris Mason771ed682008-11-06 22:02:51 -0500667 async_extent->compressed_size,
668 async_extent->compressed_size,
Josef Bacik81c9ad22012-01-18 10:56:06 -0500669 0, alloc_hint, &ins, 1);
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500670 if (ret) {
671 int i;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500672
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
676 }
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500680
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400681 if (ret == -ENOSPC) {
682 unlock_extent(io_tree, async_extent->start,
683 async_extent->start +
684 async_extent->ram_size - 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100685 goto retry;
Josef Bacikfdf8e2e2013-06-14 16:58:23 -0400686 }
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500687 goto out_free;
Josef Bacikf5a84ee32009-11-10 21:23:48 -0500688 }
689
Yan, Zhengc2167752009-11-12 09:34:21 +0000690 /*
691 * here we're doing allocation and writeback of the
692 * compressed pages
693 */
694 btrfs_drop_extent_cache(inode, async_extent->start,
695 async_extent->start +
696 async_extent->ram_size - 1, 0);
697
David Sterba172ddd62011-04-21 00:48:27 +0200698 em = alloc_extent_map();
Liu Bob9aa55b2013-05-14 02:12:15 +0000699 if (!em) {
700 ret = -ENOMEM;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500701 goto out_free_reserve;
Liu Bob9aa55b2013-05-14 02:12:15 +0000702 }
Chris Mason771ed682008-11-06 22:02:51 -0500703 em->start = async_extent->start;
704 em->len = async_extent->ram_size;
Chris Mason445a6942008-11-10 11:53:33 -0500705 em->orig_start = em->start;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400706 em->mod_start = em->start;
707 em->mod_len = em->len;
Chris Mason771ed682008-11-06 22:02:51 -0500708
709 em->block_start = ins.objectid;
710 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500711 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -0400712 em->ram_bytes = async_extent->ram_size;
Chris Mason771ed682008-11-06 22:02:51 -0500713 em->bdev = root->fs_info->fs_devices->latest_bdev;
Li Zefan261507a02010-12-17 14:21:50 +0800714 em->compress_type = async_extent->compress_type;
Chris Mason771ed682008-11-06 22:02:51 -0500715 set_bit(EXTENT_FLAG_PINNED, &em->flags);
716 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400717 em->generation = -1;
Chris Mason771ed682008-11-06 22:02:51 -0500718
Chris Masond3977122009-01-05 21:25:51 -0500719 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400720 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400721 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -0400722 write_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500723 if (ret != -EEXIST) {
724 free_extent_map(em);
725 break;
726 }
727 btrfs_drop_extent_cache(inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1, 0);
730 }
731
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500732 if (ret)
733 goto out_free_reserve;
734
Li Zefan261507a02010-12-17 14:21:50 +0800735 ret = btrfs_add_ordered_extent_compress(inode,
736 async_extent->start,
737 ins.objectid,
738 async_extent->ram_size,
739 ins.offset,
740 BTRFS_ORDERED_COMPRESSED,
741 async_extent->compress_type);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500742 if (ret)
743 goto out_free_reserve;
Chris Mason771ed682008-11-06 22:02:51 -0500744
Chris Mason771ed682008-11-06 22:02:51 -0500745 /*
746 * clear dirty, set writeback and unlock the pages.
747 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400748 extent_clear_unlock_delalloc(inode, async_extent->start,
Chris Masona791e352009-10-08 11:27:10 -0400749 async_extent->start +
750 async_extent->ram_size - 1,
Josef Bacik151a41b2013-07-29 13:22:24 -0400751 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
752 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -0400753 PAGE_SET_WRITEBACK);
Chris Mason771ed682008-11-06 22:02:51 -0500754 ret = btrfs_submit_compressed_write(inode,
Chris Masond3977122009-01-05 21:25:51 -0500755 async_extent->start,
756 async_extent->ram_size,
757 ins.objectid,
758 ins.offset, async_extent->pages,
759 async_extent->nr_pages);
Chris Mason771ed682008-11-06 22:02:51 -0500760 alloc_hint = ins.objectid + ins.offset;
761 kfree(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500762 if (ret)
763 goto out;
Chris Mason771ed682008-11-06 22:02:51 -0500764 cond_resched();
765 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100766 ret = 0;
767out:
768 return ret;
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500769out_free_reserve:
770 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100771out_free:
Josef Bacikc2790a22013-07-29 11:20:47 -0400772 extent_clear_unlock_delalloc(inode, async_extent->start,
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500773 async_extent->start +
774 async_extent->ram_size - 1,
Josef Bacikc2790a22013-07-29 11:20:47 -0400775 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -0400776 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
777 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
778 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100779 kfree(async_extent);
Josef Bacik3e04e7f2013-02-06 16:49:15 -0500780 goto again;
Chris Mason771ed682008-11-06 22:02:51 -0500781}
782
Josef Bacik4b46fce2010-05-23 11:00:55 -0400783static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
784 u64 num_bytes)
785{
786 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
787 struct extent_map *em;
788 u64 alloc_hint = 0;
789
790 read_lock(&em_tree->lock);
791 em = search_extent_mapping(em_tree, start, num_bytes);
792 if (em) {
793 /*
794 * if block start isn't an actual block number then find the
795 * first block in this inode and use that as a hint. If that
796 * block is also bogus then just don't worry about it.
797 */
798 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
799 free_extent_map(em);
800 em = search_extent_mapping(em_tree, 0, 0);
801 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
802 alloc_hint = em->block_start;
803 if (em)
804 free_extent_map(em);
805 } else {
806 alloc_hint = em->block_start;
807 free_extent_map(em);
808 }
809 }
810 read_unlock(&em_tree->lock);
811
812 return alloc_hint;
813}
814
Chris Mason771ed682008-11-06 22:02:51 -0500815/*
816 * when extent_io.c finds a delayed allocation range in the file,
817 * the call backs end up in this code. The basic idea is to
818 * allocate extents on disk for the range, and create ordered data structs
819 * in ram to track those extents.
820 *
821 * locked_page is the page that writepage had locked already. We use
822 * it to make sure we don't do extra locks or unlocks.
823 *
824 * *page_started is set to one if we unlock locked_page and do everything
825 * required to start IO on it. It may be clean and already done with
826 * IO when we return.
827 */
Josef Bacik00361582013-08-14 14:02:47 -0400828static noinline int cow_file_range(struct inode *inode,
829 struct page *locked_page,
830 u64 start, u64 end, int *page_started,
831 unsigned long *nr_written,
832 int unlock)
Chris Mason771ed682008-11-06 22:02:51 -0500833{
Josef Bacik00361582013-08-14 14:02:47 -0400834 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason771ed682008-11-06 22:02:51 -0500835 u64 alloc_hint = 0;
836 u64 num_bytes;
837 unsigned long ram_size;
838 u64 disk_num_bytes;
839 u64 cur_alloc_size;
840 u64 blocksize = root->sectorsize;
Chris Mason771ed682008-11-06 22:02:51 -0500841 struct btrfs_key ins;
842 struct extent_map *em;
843 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
844 int ret = 0;
845
Josef Bacik02ecd2c2013-10-25 16:19:08 -0400846 if (btrfs_is_free_space_inode(inode)) {
847 WARN_ON_ONCE(1);
848 return -EINVAL;
849 }
Chris Mason771ed682008-11-06 22:02:51 -0500850
Qu Wenruofda28322013-02-26 08:10:22 +0000851 num_bytes = ALIGN(end - start + 1, blocksize);
Chris Mason771ed682008-11-06 22:02:51 -0500852 num_bytes = max(blocksize, num_bytes);
853 disk_num_bytes = num_bytes;
Chris Mason771ed682008-11-06 22:02:51 -0500854
Chris Mason4cb53002011-05-24 15:35:30 -0400855 /* if this is a small write inside eof, kick off defrag */
Liu Bo4cb13e52012-03-29 09:57:45 -0400856 if (num_bytes < 64 * 1024 &&
857 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
Josef Bacik00361582013-08-14 14:02:47 -0400858 btrfs_add_inode_defrag(NULL, inode);
Chris Mason4cb53002011-05-24 15:35:30 -0400859
Chris Mason771ed682008-11-06 22:02:51 -0500860 if (start == 0) {
861 /* lets try to make an inline extent */
Josef Bacik00361582013-08-14 14:02:47 -0400862 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
863 NULL);
Chris Mason771ed682008-11-06 22:02:51 -0500864 if (ret == 0) {
Josef Bacikc2790a22013-07-29 11:20:47 -0400865 extent_clear_unlock_delalloc(inode, start, end, NULL,
866 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -0400867 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -0400868 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
869 PAGE_END_WRITEBACK);
Yan, Zhengc2167752009-11-12 09:34:21 +0000870
Chris Mason771ed682008-11-06 22:02:51 -0500871 *nr_written = *nr_written +
872 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
873 *page_started = 1;
Chris Mason771ed682008-11-06 22:02:51 -0500874 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100875 } else if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100876 goto out_unlock;
Chris Mason771ed682008-11-06 22:02:51 -0500877 }
878 }
Chris Masonc8b97812008-10-29 14:49:59 -0400879
880 BUG_ON(disk_num_bytes >
David Sterba6c417612011-04-13 15:41:04 +0200881 btrfs_super_total_bytes(root->fs_info->super_copy));
Chris Masonc8b97812008-10-29 14:49:59 -0400882
Josef Bacik4b46fce2010-05-23 11:00:55 -0400883 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400884 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
Chris Mason3b951512008-04-17 11:29:12 -0400885
Chris Masond3977122009-01-05 21:25:51 -0500886 while (disk_num_bytes > 0) {
Chris Masona791e352009-10-08 11:27:10 -0400887 unsigned long op;
888
Josef Bacik287a0ab2010-03-19 18:07:23 +0000889 cur_alloc_size = disk_num_bytes;
Josef Bacik00361582013-08-14 14:02:47 -0400890 ret = btrfs_reserve_extent(root, cur_alloc_size,
Chris Mason771ed682008-11-06 22:02:51 -0500891 root->sectorsize, 0, alloc_hint,
Josef Bacik81c9ad22012-01-18 10:56:06 -0500892 &ins, 1);
Josef Bacik00361582013-08-14 14:02:47 -0400893 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100894 goto out_unlock;
Chris Masond3977122009-01-05 21:25:51 -0500895
David Sterba172ddd62011-04-21 00:48:27 +0200896 em = alloc_extent_map();
Liu Bob9aa55b2013-05-14 02:12:15 +0000897 if (!em) {
898 ret = -ENOMEM;
Liu Boace68ba2013-04-22 10:53:47 +0000899 goto out_reserve;
Liu Bob9aa55b2013-05-14 02:12:15 +0000900 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400901 em->start = start;
Chris Mason445a6942008-11-10 11:53:33 -0500902 em->orig_start = em->start;
Chris Mason771ed682008-11-06 22:02:51 -0500903 ram_size = ins.offset;
904 em->len = ins.offset;
Josef Bacik2ab28f32012-10-12 15:27:49 -0400905 em->mod_start = em->start;
906 em->mod_len = em->len;
Chris Masonc8b97812008-10-29 14:49:59 -0400907
Chris Masone6dcd2d2008-07-17 12:53:50 -0400908 em->block_start = ins.objectid;
Chris Masonc8b97812008-10-29 14:49:59 -0400909 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -0500910 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -0400911 em->ram_bytes = ram_size;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400912 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason7f3c74f2008-07-18 12:01:11 -0400913 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -0400914 em->generation = -1;
Chris Masonc8b97812008-10-29 14:49:59 -0400915
Chris Masond3977122009-01-05 21:25:51 -0500916 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -0400917 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -0400918 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -0400919 write_unlock(&em_tree->lock);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400920 if (ret != -EEXIST) {
921 free_extent_map(em);
922 break;
923 }
924 btrfs_drop_extent_cache(inode, start,
Chris Masonc8b97812008-10-29 14:49:59 -0400925 start + ram_size - 1, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400926 }
Liu Boace68ba2013-04-22 10:53:47 +0000927 if (ret)
928 goto out_reserve;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400929
Chris Mason98d20f62008-04-14 09:46:10 -0400930 cur_alloc_size = ins.offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400931 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
Chris Mason771ed682008-11-06 22:02:51 -0500932 ram_size, cur_alloc_size, 0);
Liu Boace68ba2013-04-22 10:53:47 +0000933 if (ret)
934 goto out_reserve;
Chris Masonc8b97812008-10-29 14:49:59 -0400935
Yan Zheng17d217f2008-12-12 10:03:38 -0500936 if (root->root_key.objectid ==
937 BTRFS_DATA_RELOC_TREE_OBJECTID) {
938 ret = btrfs_reloc_clone_csums(inode, start,
939 cur_alloc_size);
Josef Bacik00361582013-08-14 14:02:47 -0400940 if (ret)
Liu Boace68ba2013-04-22 10:53:47 +0000941 goto out_reserve;
Yan Zheng17d217f2008-12-12 10:03:38 -0500942 }
943
Chris Masond3977122009-01-05 21:25:51 -0500944 if (disk_num_bytes < cur_alloc_size)
Chris Mason3b951512008-04-17 11:29:12 -0400945 break;
Chris Masond3977122009-01-05 21:25:51 -0500946
Chris Masonc8b97812008-10-29 14:49:59 -0400947 /* we're not doing compressed IO, don't unlock the first
948 * page (which the caller expects to stay locked), don't
949 * clear any dirty bits and don't set any writeback bits
Chris Mason8b62b722009-09-02 16:53:46 -0400950 *
951 * Do set the Private2 bit so we know this page was properly
952 * setup for writepage
Chris Masonc8b97812008-10-29 14:49:59 -0400953 */
Josef Bacikc2790a22013-07-29 11:20:47 -0400954 op = unlock ? PAGE_UNLOCK : 0;
955 op |= PAGE_SET_PRIVATE2;
Chris Masona791e352009-10-08 11:27:10 -0400956
Josef Bacikc2790a22013-07-29 11:20:47 -0400957 extent_clear_unlock_delalloc(inode, start,
958 start + ram_size - 1, locked_page,
959 EXTENT_LOCKED | EXTENT_DELALLOC,
960 op);
Chris Masonc8b97812008-10-29 14:49:59 -0400961 disk_num_bytes -= cur_alloc_size;
Chris Masonc59f8952007-12-17 20:14:04 -0500962 num_bytes -= cur_alloc_size;
963 alloc_hint = ins.objectid + ins.offset;
964 start += cur_alloc_size;
Chris Masonb888db22007-08-27 16:49:44 -0400965 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100966out:
Chris Masonbe20aa92007-12-17 20:14:01 -0500967 return ret;
Miao Xieb7d5b0a2012-11-01 07:32:18 +0000968
Liu Boace68ba2013-04-22 10:53:47 +0000969out_reserve:
970 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100971out_unlock:
Josef Bacikc2790a22013-07-29 11:20:47 -0400972 extent_clear_unlock_delalloc(inode, start, end, locked_page,
Josef Bacik151a41b2013-07-29 13:22:24 -0400973 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
974 EXTENT_DELALLOC | EXTENT_DEFRAG,
975 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
976 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100977 goto out;
Chris Mason771ed682008-11-06 22:02:51 -0500978}
Chris Masonc8b97812008-10-29 14:49:59 -0400979
Chris Mason771ed682008-11-06 22:02:51 -0500980/*
981 * work queue call back to started compression on a file and pages
982 */
983static noinline void async_cow_start(struct btrfs_work *work)
984{
985 struct async_cow *async_cow;
986 int num_added = 0;
987 async_cow = container_of(work, struct async_cow, work);
988
989 compress_file_range(async_cow->inode, async_cow->locked_page,
990 async_cow->start, async_cow->end, async_cow,
991 &num_added);
Josef Bacik8180ef82012-06-08 15:16:12 -0400992 if (num_added == 0) {
Josef Bacikcb77fcd2012-06-15 12:19:48 -0600993 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -0500994 async_cow->inode = NULL;
Josef Bacik8180ef82012-06-08 15:16:12 -0400995 }
Chris Mason771ed682008-11-06 22:02:51 -0500996}
997
998/*
999 * work queue call back to submit previously compressed pages
1000 */
1001static noinline void async_cow_submit(struct btrfs_work *work)
1002{
1003 struct async_cow *async_cow;
1004 struct btrfs_root *root;
1005 unsigned long nr_pages;
1006
1007 async_cow = container_of(work, struct async_cow, work);
1008
1009 root = async_cow->root;
1010 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1011 PAGE_CACHE_SHIFT;
1012
Josef Bacik66657b32012-08-01 15:36:24 -04001013 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
Liu Bo287082b2012-06-28 04:02:24 -06001014 5 * 1024 * 1024 &&
Chris Mason771ed682008-11-06 22:02:51 -05001015 waitqueue_active(&root->fs_info->async_submit_wait))
1016 wake_up(&root->fs_info->async_submit_wait);
1017
Chris Masond3977122009-01-05 21:25:51 -05001018 if (async_cow->inode)
Chris Mason771ed682008-11-06 22:02:51 -05001019 submit_compressed_extents(async_cow->inode, async_cow);
Chris Mason771ed682008-11-06 22:02:51 -05001020}
Chris Masonc8b97812008-10-29 14:49:59 -04001021
Chris Mason771ed682008-11-06 22:02:51 -05001022static noinline void async_cow_free(struct btrfs_work *work)
1023{
1024 struct async_cow *async_cow;
1025 async_cow = container_of(work, struct async_cow, work);
Josef Bacik8180ef82012-06-08 15:16:12 -04001026 if (async_cow->inode)
Josef Bacikcb77fcd2012-06-15 12:19:48 -06001027 btrfs_add_delayed_iput(async_cow->inode);
Chris Mason771ed682008-11-06 22:02:51 -05001028 kfree(async_cow);
1029}
1030
1031static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032 u64 start, u64 end, int *page_started,
1033 unsigned long *nr_written)
1034{
1035 struct async_cow *async_cow;
1036 struct btrfs_root *root = BTRFS_I(inode)->root;
1037 unsigned long nr_pages;
1038 u64 cur_end;
Liu Bo287082b2012-06-28 04:02:24 -06001039 int limit = 10 * 1024 * 1024;
Chris Mason771ed682008-11-06 22:02:51 -05001040
Chris Masona3429ab2009-10-08 12:30:20 -04001041 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042 1, 0, NULL, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -05001043 while (start < end) {
Chris Mason771ed682008-11-06 22:02:51 -05001044 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001045 BUG_ON(!async_cow); /* -ENOMEM */
Josef Bacik8180ef82012-06-08 15:16:12 -04001046 async_cow->inode = igrab(inode);
Chris Mason771ed682008-11-06 22:02:51 -05001047 async_cow->root = root;
1048 async_cow->locked_page = locked_page;
1049 async_cow->start = start;
1050
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001051 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
Chris Mason771ed682008-11-06 22:02:51 -05001052 cur_end = end;
1053 else
1054 cur_end = min(end, start + 512 * 1024 - 1);
1055
1056 async_cow->end = cur_end;
1057 INIT_LIST_HEAD(&async_cow->extents);
1058
1059 async_cow->work.func = async_cow_start;
1060 async_cow->work.ordered_func = async_cow_submit;
1061 async_cow->work.ordered_free = async_cow_free;
1062 async_cow->work.flags = 0;
1063
Chris Mason771ed682008-11-06 22:02:51 -05001064 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1065 PAGE_CACHE_SHIFT;
1066 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1067
1068 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1069 &async_cow->work);
1070
1071 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072 wait_event(root->fs_info->async_submit_wait,
1073 (atomic_read(&root->fs_info->async_delalloc_pages) <
1074 limit));
1075 }
1076
Chris Masond3977122009-01-05 21:25:51 -05001077 while (atomic_read(&root->fs_info->async_submit_draining) &&
Chris Mason771ed682008-11-06 22:02:51 -05001078 atomic_read(&root->fs_info->async_delalloc_pages)) {
1079 wait_event(root->fs_info->async_submit_wait,
1080 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1081 0));
1082 }
1083
1084 *nr_written += nr_pages;
1085 start = cur_end + 1;
1086 }
1087 *page_started = 1;
1088 return 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05001089}
1090
Chris Masond3977122009-01-05 21:25:51 -05001091static noinline int csum_exist_in_range(struct btrfs_root *root,
Yan Zheng17d217f2008-12-12 10:03:38 -05001092 u64 bytenr, u64 num_bytes)
1093{
1094 int ret;
1095 struct btrfs_ordered_sum *sums;
1096 LIST_HEAD(list);
1097
Yan Zheng07d400a2009-01-06 11:42:00 -05001098 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
Arne Jansena2de7332011-03-08 14:14:00 +01001099 bytenr + num_bytes - 1, &list, 0);
Yan Zheng17d217f2008-12-12 10:03:38 -05001100 if (ret == 0 && list_empty(&list))
1101 return 0;
1102
1103 while (!list_empty(&list)) {
1104 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105 list_del(&sums->list);
1106 kfree(sums);
1107 }
1108 return 1;
1109}
1110
Chris Masond352ac62008-09-29 15:18:18 -04001111/*
1112 * when nowcow writeback call back. This checks for snapshots or COW copies
1113 * of the extents that exist in the file, and COWs the file as required.
1114 *
1115 * If no cow copies or snapshots exist, we write directly to the existing
1116 * blocks on disk
1117 */
Chris Mason7f366cf2009-03-12 20:12:45 -04001118static noinline int run_delalloc_nocow(struct inode *inode,
1119 struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001120 u64 start, u64 end, int *page_started, int force,
1121 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001122{
Chris Masonbe20aa92007-12-17 20:14:01 -05001123 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001124 struct btrfs_trans_handle *trans;
Chris Masonbe20aa92007-12-17 20:14:01 -05001125 struct extent_buffer *leaf;
Chris Masonbe20aa92007-12-17 20:14:01 -05001126 struct btrfs_path *path;
Yan Zheng80ff3852008-10-30 14:20:02 -04001127 struct btrfs_file_extent_item *fi;
Chris Masonbe20aa92007-12-17 20:14:01 -05001128 struct btrfs_key found_key;
Yan Zheng80ff3852008-10-30 14:20:02 -04001129 u64 cow_start;
1130 u64 cur_offset;
1131 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001132 u64 extent_offset;
Yan Zheng80ff3852008-10-30 14:20:02 -04001133 u64 disk_bytenr;
1134 u64 num_bytes;
Josef Bacikb4939682012-12-03 10:31:19 -05001135 u64 disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001136 u64 ram_bytes;
Yan Zheng80ff3852008-10-30 14:20:02 -04001137 int extent_type;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001138 int ret, err;
Yan Zhengd899e052008-10-30 14:25:28 -04001139 int type;
Yan Zheng80ff3852008-10-30 14:20:02 -04001140 int nocow;
1141 int check_prev = 1;
Li Zefan82d59022011-04-20 10:33:24 +08001142 bool nolock;
Li Zefan33345d012011-04-20 10:31:50 +08001143 u64 ino = btrfs_ino(inode);
Chris Masonbe20aa92007-12-17 20:14:01 -05001144
1145 path = btrfs_alloc_path();
Josef Bacik17ca04a2012-05-31 15:58:55 -04001146 if (!path) {
Josef Bacikc2790a22013-07-29 11:20:47 -04001147 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1148 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001149 EXTENT_DO_ACCOUNTING |
1150 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001151 PAGE_CLEAR_DIRTY |
1152 PAGE_SET_WRITEBACK |
1153 PAGE_END_WRITEBACK);
Mark Fashehd8926bb2011-07-13 10:38:47 -07001154 return -ENOMEM;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001155 }
Li Zefan82d59022011-04-20 10:33:24 +08001156
Liu Bo83eea1f2012-07-10 05:28:39 -06001157 nolock = btrfs_is_free_space_inode(inode);
Li Zefan82d59022011-04-20 10:33:24 +08001158
1159 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001160 trans = btrfs_join_transaction_nolock(root);
Li Zefan82d59022011-04-20 10:33:24 +08001161 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04001162 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04001163
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001164 if (IS_ERR(trans)) {
Josef Bacikc2790a22013-07-29 11:20:47 -04001165 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166 EXTENT_LOCKED | EXTENT_DELALLOC |
Josef Bacik151a41b2013-07-29 13:22:24 -04001167 EXTENT_DO_ACCOUNTING |
1168 EXTENT_DEFRAG, PAGE_UNLOCK |
Josef Bacikc2790a22013-07-29 11:20:47 -04001169 PAGE_CLEAR_DIRTY |
1170 PAGE_SET_WRITEBACK |
1171 PAGE_END_WRITEBACK);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001172 btrfs_free_path(path);
1173 return PTR_ERR(trans);
1174 }
1175
Josef Bacik74b21072011-04-13 12:02:53 -04001176 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Chris Masonbe20aa92007-12-17 20:14:01 -05001177
Yan Zheng80ff3852008-10-30 14:20:02 -04001178 cow_start = (u64)-1;
1179 cur_offset = start;
1180 while (1) {
Li Zefan33345d012011-04-20 10:31:50 +08001181 ret = btrfs_lookup_file_extent(trans, root, path, ino,
Yan Zheng80ff3852008-10-30 14:20:02 -04001182 cur_offset, 0);
Josef Bacikd788a342013-10-25 16:55:08 -04001183 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001184 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001185 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186 leaf = path->nodes[0];
1187 btrfs_item_key_to_cpu(leaf, &found_key,
1188 path->slots[0] - 1);
Li Zefan33345d012011-04-20 10:31:50 +08001189 if (found_key.objectid == ino &&
Yan Zheng80ff3852008-10-30 14:20:02 -04001190 found_key.type == BTRFS_EXTENT_DATA_KEY)
1191 path->slots[0]--;
1192 }
1193 check_prev = 0;
1194next_slot:
1195 leaf = path->nodes[0];
1196 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197 ret = btrfs_next_leaf(root, path);
Josef Bacikd788a342013-10-25 16:55:08 -04001198 if (ret < 0)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001199 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001200 if (ret > 0)
1201 break;
1202 leaf = path->nodes[0];
1203 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001204
Yan Zheng80ff3852008-10-30 14:20:02 -04001205 nocow = 0;
1206 disk_bytenr = 0;
Yan Zheng17d217f2008-12-12 10:03:38 -05001207 num_bytes = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001208 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05001209
Li Zefan33345d012011-04-20 10:31:50 +08001210 if (found_key.objectid > ino ||
Yan Zheng80ff3852008-10-30 14:20:02 -04001211 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1212 found_key.offset > end)
1213 break;
Chris Masonbe20aa92007-12-17 20:14:01 -05001214
Yan Zheng80ff3852008-10-30 14:20:02 -04001215 if (found_key.offset > cur_offset) {
1216 extent_end = found_key.offset;
Chris Masone9061e22009-10-09 09:57:45 -04001217 extent_type = 0;
Yan Zheng80ff3852008-10-30 14:20:02 -04001218 goto out_check;
1219 }
Chris Masonc31f8832008-01-08 15:46:31 -05001220
Yan Zheng80ff3852008-10-30 14:20:02 -04001221 fi = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_file_extent_item);
1223 extent_type = btrfs_file_extent_type(leaf, fi);
Chris Masonbe20aa92007-12-17 20:14:01 -05001224
Josef Bacikcc95bef2013-04-04 14:31:27 -04001225 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
Yan Zhengd899e052008-10-30 14:25:28 -04001226 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1227 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001228 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001229 extent_offset = btrfs_file_extent_offset(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001230 extent_end = found_key.offset +
1231 btrfs_file_extent_num_bytes(leaf, fi);
Josef Bacikb4939682012-12-03 10:31:19 -05001232 disk_num_bytes =
1233 btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng80ff3852008-10-30 14:20:02 -04001234 if (extent_end <= start) {
1235 path->slots[0]++;
1236 goto next_slot;
1237 }
Yan Zheng17d217f2008-12-12 10:03:38 -05001238 if (disk_bytenr == 0)
1239 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001240 if (btrfs_file_extent_compression(leaf, fi) ||
1241 btrfs_file_extent_encryption(leaf, fi) ||
1242 btrfs_file_extent_other_encoding(leaf, fi))
1243 goto out_check;
Yan Zhengd899e052008-10-30 14:25:28 -04001244 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1245 goto out_check;
Yan Zhengd2fb3432008-12-11 16:30:39 -05001246 if (btrfs_extent_readonly(root, disk_bytenr))
Yan Zheng80ff3852008-10-30 14:20:02 -04001247 goto out_check;
Li Zefan33345d012011-04-20 10:31:50 +08001248 if (btrfs_cross_ref_exist(trans, root, ino,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001249 found_key.offset -
1250 extent_offset, disk_bytenr))
Yan Zheng17d217f2008-12-12 10:03:38 -05001251 goto out_check;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001252 disk_bytenr += extent_offset;
Yan Zheng17d217f2008-12-12 10:03:38 -05001253 disk_bytenr += cur_offset - found_key.offset;
1254 num_bytes = min(end + 1, extent_end) - cur_offset;
1255 /*
1256 * force cow if csum exists in the range.
1257 * this ensure that csum for a given extent are
1258 * either valid or do not exist.
1259 */
1260 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1261 goto out_check;
Yan Zheng80ff3852008-10-30 14:20:02 -04001262 nocow = 1;
1263 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1264 extent_end = found_key.offset +
1265 btrfs_file_extent_inline_len(leaf, fi);
1266 extent_end = ALIGN(extent_end, root->sectorsize);
1267 } else {
1268 BUG_ON(1);
1269 }
1270out_check:
1271 if (extent_end <= start) {
1272 path->slots[0]++;
1273 goto next_slot;
1274 }
1275 if (!nocow) {
1276 if (cow_start == (u64)-1)
1277 cow_start = cur_offset;
1278 cur_offset = extent_end;
1279 if (cur_offset > end)
1280 break;
1281 path->slots[0]++;
1282 goto next_slot;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001283 }
Chris Masonbe20aa92007-12-17 20:14:01 -05001284
David Sterbab3b4aa72011-04-21 01:20:15 +02001285 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001286 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001287 ret = cow_file_range(inode, locked_page,
1288 cow_start, found_key.offset - 1,
1289 page_started, nr_written, 1);
Josef Bacikd788a342013-10-25 16:55:08 -04001290 if (ret)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001291 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001292 cow_start = (u64)-1;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001293 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001294
Yan Zhengd899e052008-10-30 14:25:28 -04001295 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1296 struct extent_map *em;
1297 struct extent_map_tree *em_tree;
1298 em_tree = &BTRFS_I(inode)->extent_tree;
David Sterba172ddd62011-04-21 00:48:27 +02001299 em = alloc_extent_map();
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001300 BUG_ON(!em); /* -ENOMEM */
Yan Zhengd899e052008-10-30 14:25:28 -04001301 em->start = cur_offset;
Josef Bacik70c8a912012-10-11 16:54:30 -04001302 em->orig_start = found_key.offset - extent_offset;
Yan Zhengd899e052008-10-30 14:25:28 -04001303 em->len = num_bytes;
1304 em->block_len = num_bytes;
1305 em->block_start = disk_bytenr;
Josef Bacikb4939682012-12-03 10:31:19 -05001306 em->orig_block_len = disk_num_bytes;
Josef Bacikcc95bef2013-04-04 14:31:27 -04001307 em->ram_bytes = ram_bytes;
Yan Zhengd899e052008-10-30 14:25:28 -04001308 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacik2ab28f32012-10-12 15:27:49 -04001309 em->mod_start = em->start;
1310 em->mod_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04001311 set_bit(EXTENT_FLAG_PINNED, &em->flags);
Josef Bacikb11e2342012-12-03 10:58:15 -05001312 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik70c8a912012-10-11 16:54:30 -04001313 em->generation = -1;
Yan Zhengd899e052008-10-30 14:25:28 -04001314 while (1) {
Chris Mason890871b2009-09-02 16:24:52 -04001315 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04001316 ret = add_extent_mapping(em_tree, em, 1);
Chris Mason890871b2009-09-02 16:24:52 -04001317 write_unlock(&em_tree->lock);
Yan Zhengd899e052008-10-30 14:25:28 -04001318 if (ret != -EEXIST) {
1319 free_extent_map(em);
1320 break;
1321 }
1322 btrfs_drop_extent_cache(inode, em->start,
1323 em->start + em->len - 1, 0);
1324 }
1325 type = BTRFS_ORDERED_PREALLOC;
1326 } else {
1327 type = BTRFS_ORDERED_NOCOW;
1328 }
Yan Zheng80ff3852008-10-30 14:20:02 -04001329
1330 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Yan Zhengd899e052008-10-30 14:25:28 -04001331 num_bytes, num_bytes, type);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001332 BUG_ON(ret); /* -ENOMEM */
Chris Mason771ed682008-11-06 22:02:51 -05001333
Yan, Zhengefa56462010-05-16 10:49:59 -04001334 if (root->root_key.objectid ==
1335 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1336 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1337 num_bytes);
Josef Bacikd788a342013-10-25 16:55:08 -04001338 if (ret)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001339 goto error;
Yan, Zhengefa56462010-05-16 10:49:59 -04001340 }
1341
Josef Bacikc2790a22013-07-29 11:20:47 -04001342 extent_clear_unlock_delalloc(inode, cur_offset,
1343 cur_offset + num_bytes - 1,
1344 locked_page, EXTENT_LOCKED |
1345 EXTENT_DELALLOC, PAGE_UNLOCK |
1346 PAGE_SET_PRIVATE2);
Yan Zheng80ff3852008-10-30 14:20:02 -04001347 cur_offset = extent_end;
1348 if (cur_offset > end)
1349 break;
Yan Zheng7ea394f2008-08-05 13:05:02 -04001350 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001351 btrfs_release_path(path);
Yan Zheng80ff3852008-10-30 14:20:02 -04001352
Josef Bacik17ca04a2012-05-31 15:58:55 -04001353 if (cur_offset <= end && cow_start == (u64)-1) {
Yan Zheng80ff3852008-10-30 14:20:02 -04001354 cow_start = cur_offset;
Josef Bacik17ca04a2012-05-31 15:58:55 -04001355 cur_offset = end;
1356 }
1357
Yan Zheng80ff3852008-10-30 14:20:02 -04001358 if (cow_start != (u64)-1) {
Josef Bacik00361582013-08-14 14:02:47 -04001359 ret = cow_file_range(inode, locked_page, cow_start, end,
1360 page_started, nr_written, 1);
Josef Bacikd788a342013-10-25 16:55:08 -04001361 if (ret)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001362 goto error;
Yan Zheng80ff3852008-10-30 14:20:02 -04001363 }
1364
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001365error:
Miao Xiea698d0752012-09-20 01:51:59 -06001366 err = btrfs_end_transaction(trans, root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001367 if (!ret)
1368 ret = err;
1369
Josef Bacik17ca04a2012-05-31 15:58:55 -04001370 if (ret && cur_offset < end)
Josef Bacikc2790a22013-07-29 11:20:47 -04001371 extent_clear_unlock_delalloc(inode, cur_offset, end,
1372 locked_page, EXTENT_LOCKED |
Josef Bacik151a41b2013-07-29 13:22:24 -04001373 EXTENT_DELALLOC | EXTENT_DEFRAG |
1374 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1375 PAGE_CLEAR_DIRTY |
Josef Bacikc2790a22013-07-29 11:20:47 -04001376 PAGE_SET_WRITEBACK |
1377 PAGE_END_WRITEBACK);
Yan Zheng7ea394f2008-08-05 13:05:02 -04001378 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001379 return ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05001380}
1381
Chris Masond352ac62008-09-29 15:18:18 -04001382/*
1383 * extent_io.c call back to do delayed allocation processing
1384 */
Chris Masonc8b97812008-10-29 14:49:59 -04001385static int run_delalloc_range(struct inode *inode, struct page *locked_page,
Chris Mason771ed682008-11-06 22:02:51 -05001386 u64 start, u64 end, int *page_started,
1387 unsigned long *nr_written)
Chris Masonbe20aa92007-12-17 20:14:01 -05001388{
Chris Masonbe20aa92007-12-17 20:14:01 -05001389 int ret;
Chris Mason7f366cf2009-03-12 20:12:45 -04001390 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masona2135012008-06-25 16:01:30 -04001391
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001392 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
Chris Masonc8b97812008-10-29 14:49:59 -04001393 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001394 page_started, 1, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001395 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
Yan Zhengd899e052008-10-30 14:25:28 -04001396 ret = run_delalloc_nocow(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001397 page_started, 0, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001398 } else if (!btrfs_test_opt(root, COMPRESS) &&
1399 !(BTRFS_I(inode)->force_compress) &&
1400 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
Chris Mason7f366cf2009-03-12 20:12:45 -04001401 ret = cow_file_range(inode, locked_page, start, end,
1402 page_started, nr_written, 1);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001403 } else {
1404 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1405 &BTRFS_I(inode)->runtime_flags);
Chris Mason771ed682008-11-06 22:02:51 -05001406 ret = cow_file_range_async(inode, locked_page, start, end,
Chris Masond3977122009-01-05 21:25:51 -05001407 page_started, nr_written);
Josef Bacik7ddf5a42012-06-08 15:26:47 -04001408 }
Chris Masonb888db22007-08-27 16:49:44 -04001409 return ret;
1410}
1411
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001412static void btrfs_split_extent_hook(struct inode *inode,
1413 struct extent_state *orig, u64 split)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001414{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001415 /* not delalloc, ignore it */
Josef Bacik9ed74f22009-09-11 16:12:44 -04001416 if (!(orig->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001417 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001418
Josef Bacik9e0baf62011-07-15 15:16:44 +00001419 spin_lock(&BTRFS_I(inode)->lock);
1420 BTRFS_I(inode)->outstanding_extents++;
1421 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001422}
1423
1424/*
1425 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1426 * extents so we can keep track of new extents that are just merged onto old
1427 * extents, such as when we are doing sequential writes, so we can properly
1428 * account for the metadata space we'll need.
1429 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001430static void btrfs_merge_extent_hook(struct inode *inode,
1431 struct extent_state *new,
1432 struct extent_state *other)
Josef Bacik9ed74f22009-09-11 16:12:44 -04001433{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001434 /* not delalloc, ignore it */
1435 if (!(other->state & EXTENT_DELALLOC))
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001436 return;
Josef Bacik9ed74f22009-09-11 16:12:44 -04001437
Josef Bacik9e0baf62011-07-15 15:16:44 +00001438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents--;
1440 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001441}
1442
Miao Xieeb73c1b2013-05-15 07:48:22 +00001443static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1444 struct inode *inode)
1445{
1446 spin_lock(&root->delalloc_lock);
1447 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1448 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1449 &root->delalloc_inodes);
1450 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1451 &BTRFS_I(inode)->runtime_flags);
1452 root->nr_delalloc_inodes++;
1453 if (root->nr_delalloc_inodes == 1) {
1454 spin_lock(&root->fs_info->delalloc_root_lock);
1455 BUG_ON(!list_empty(&root->delalloc_root));
1456 list_add_tail(&root->delalloc_root,
1457 &root->fs_info->delalloc_roots);
1458 spin_unlock(&root->fs_info->delalloc_root_lock);
1459 }
1460 }
1461 spin_unlock(&root->delalloc_lock);
1462}
1463
1464static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1465 struct inode *inode)
1466{
1467 spin_lock(&root->delalloc_lock);
1468 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1470 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1471 &BTRFS_I(inode)->runtime_flags);
1472 root->nr_delalloc_inodes--;
1473 if (!root->nr_delalloc_inodes) {
1474 spin_lock(&root->fs_info->delalloc_root_lock);
1475 BUG_ON(list_empty(&root->delalloc_root));
1476 list_del_init(&root->delalloc_root);
1477 spin_unlock(&root->fs_info->delalloc_root_lock);
1478 }
1479 }
1480 spin_unlock(&root->delalloc_lock);
1481}
1482
Chris Masond352ac62008-09-29 15:18:18 -04001483/*
1484 * extent_io.c set_bit_hook, used to track delayed allocation
1485 * bytes in this file, and to maintain the list of inodes that
1486 * have pending delalloc work to be done.
1487 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001488static void btrfs_set_bit_hook(struct inode *inode,
David Sterba41074882013-04-29 13:38:46 +00001489 struct extent_state *state, unsigned long *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001490{
Josef Bacik9ed74f22009-09-11 16:12:44 -04001491
Chris Mason75eff682008-12-15 15:54:40 -05001492 /*
1493 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001494 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001495 * bit, which is only set or cleared with irqs on
1496 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001497 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001498 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001499 u64 len = state->end + 1 - state->start;
Liu Bo83eea1f2012-07-10 05:28:39 -06001500 bool do_list = !btrfs_is_free_space_inode(inode);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001501
Josef Bacik9e0baf62011-07-15 15:16:44 +00001502 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001503 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001504 } else {
1505 spin_lock(&BTRFS_I(inode)->lock);
1506 BTRFS_I(inode)->outstanding_extents++;
1507 spin_unlock(&BTRFS_I(inode)->lock);
1508 }
Josef Bacik287a0ab2010-03-19 18:07:23 +00001509
Miao Xie963d6782013-01-29 10:10:51 +00001510 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1511 root->fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001512 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001513 BTRFS_I(inode)->delalloc_bytes += len;
Miao Xiedf0af1a2013-01-29 10:11:59 +00001514 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001515 &BTRFS_I(inode)->runtime_flags))
1516 btrfs_add_delalloc_inodes(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001517 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001518 }
Chris Mason291d6732008-01-29 15:55:23 -05001519}
1520
Chris Masond352ac62008-09-29 15:18:18 -04001521/*
1522 * extent_io.c clear_bit_hook, see set_bit_hook for why
1523 */
Jeff Mahoney1bf85042011-07-21 16:56:09 +00001524static void btrfs_clear_bit_hook(struct inode *inode,
David Sterba41074882013-04-29 13:38:46 +00001525 struct extent_state *state,
1526 unsigned long *bits)
Chris Mason291d6732008-01-29 15:55:23 -05001527{
Chris Mason75eff682008-12-15 15:54:40 -05001528 /*
1529 * set_bit and clear bit hooks normally require _irqsave/restore
Sergei Trofimovich27160b62011-05-20 20:20:32 +00001530 * but in this case, we are only testing for the DELALLOC
Chris Mason75eff682008-12-15 15:54:40 -05001531 * bit, which is only set or cleared with irqs on
1532 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001533 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
Chris Mason291d6732008-01-29 15:55:23 -05001534 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001535 u64 len = state->end + 1 - state->start;
Liu Bo83eea1f2012-07-10 05:28:39 -06001536 bool do_list = !btrfs_is_free_space_inode(inode);
Chris Masonbcbfce82008-04-22 13:26:47 -04001537
Josef Bacik9e0baf62011-07-15 15:16:44 +00001538 if (*bits & EXTENT_FIRST_DELALLOC) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001539 *bits &= ~EXTENT_FIRST_DELALLOC;
Josef Bacik9e0baf62011-07-15 15:16:44 +00001540 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541 spin_lock(&BTRFS_I(inode)->lock);
1542 BTRFS_I(inode)->outstanding_extents--;
1543 spin_unlock(&BTRFS_I(inode)->lock);
1544 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001545
Josef Bacikb6d08f02013-09-27 14:57:43 -04001546 /*
1547 * We don't reserve metadata space for space cache inodes so we
1548 * don't need to call dellalloc_release_metadata if there is an
1549 * error.
1550 */
1551 if (*bits & EXTENT_DO_ACCOUNTING &&
1552 root != root->fs_info->tree_root)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001553 btrfs_delalloc_release_metadata(inode, len);
1554
Josef Bacik0cb59c92010-07-02 12:14:14 -04001555 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
Josef Bacik7ee9e442013-06-21 16:37:03 -04001556 && do_list && !(state->state & EXTENT_NORESERVE))
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001557 btrfs_free_reserved_data_space(inode, len);
Josef Bacik9ed74f22009-09-11 16:12:44 -04001558
Miao Xie963d6782013-01-29 10:10:51 +00001559 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1560 root->fs_info->delalloc_batch);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001561 spin_lock(&BTRFS_I(inode)->lock);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04001562 BTRFS_I(inode)->delalloc_bytes -= len;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001563 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
Miao Xiedf0af1a2013-01-29 10:11:59 +00001564 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
Miao Xieeb73c1b2013-05-15 07:48:22 +00001565 &BTRFS_I(inode)->runtime_flags))
1566 btrfs_del_delalloc_inode(root, inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00001567 spin_unlock(&BTRFS_I(inode)->lock);
Chris Mason291d6732008-01-29 15:55:23 -05001568 }
Chris Mason291d6732008-01-29 15:55:23 -05001569}
1570
Chris Masond352ac62008-09-29 15:18:18 -04001571/*
1572 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1573 * we don't create bios that span stripes or chunks
1574 */
David Woodhouse64a16702009-07-15 23:29:37 +01001575int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
Chris Masonc8b97812008-10-29 14:49:59 -04001576 size_t size, struct bio *bio,
1577 unsigned long bio_flags)
Chris Mason239b14b2008-03-24 15:02:07 -04001578{
1579 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
Chris Masona62b9402008-10-03 16:31:08 -04001580 u64 logical = (u64)bio->bi_sector << 9;
Chris Mason239b14b2008-03-24 15:02:07 -04001581 u64 length = 0;
1582 u64 map_length;
Chris Mason239b14b2008-03-24 15:02:07 -04001583 int ret;
1584
Chris Mason771ed682008-11-06 22:02:51 -05001585 if (bio_flags & EXTENT_BIO_COMPRESSED)
1586 return 0;
1587
Chris Masonf2d8d742008-04-21 10:03:05 -04001588 length = bio->bi_size;
Chris Mason239b14b2008-03-24 15:02:07 -04001589 map_length = length;
David Woodhouse64a16702009-07-15 23:29:37 +01001590 ret = btrfs_map_block(root->fs_info, rw, logical,
Chris Masonf1885912008-04-09 16:28:12 -04001591 &map_length, NULL, 0);
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001592 /* Will always return 0 with map_multi == NULL */
Jeff Mahoney3444a972011-10-03 23:23:13 -04001593 BUG_ON(ret < 0);
Chris Masond3977122009-01-05 21:25:51 -05001594 if (map_length < length + size)
Chris Mason239b14b2008-03-24 15:02:07 -04001595 return 1;
Jeff Mahoney3444a972011-10-03 23:23:13 -04001596 return 0;
Chris Mason239b14b2008-03-24 15:02:07 -04001597}
1598
Chris Masond352ac62008-09-29 15:18:18 -04001599/*
1600 * in order to insert checksums into the metadata in large chunks,
1601 * we wait until bio submission time. All the pages in the bio are
1602 * checksummed and sums are attached onto the ordered extent record.
1603 *
1604 * At IO completion time the cums attached on the ordered extent record
1605 * are inserted into the btree
1606 */
Chris Masond3977122009-01-05 21:25:51 -05001607static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1608 struct bio *bio, int mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001609 unsigned long bio_flags,
1610 u64 bio_offset)
Chris Mason065631f2008-02-20 12:07:25 -05001611{
Chris Mason065631f2008-02-20 12:07:25 -05001612 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason065631f2008-02-20 12:07:25 -05001613 int ret = 0;
Chris Masone0156402008-04-16 11:15:20 -04001614
Chris Masond20f7042008-12-08 16:58:54 -05001615 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001616 BUG_ON(ret); /* -ENOMEM */
Chris Mason4a69a412008-11-06 22:03:00 -05001617 return 0;
1618}
Chris Masone0156402008-04-16 11:15:20 -04001619
Chris Mason4a69a412008-11-06 22:03:00 -05001620/*
1621 * in order to insert checksums into the metadata in large chunks,
1622 * we wait until bio submission time. All the pages in the bio are
1623 * checksummed and sums are attached onto the ordered extent record.
1624 *
1625 * At IO completion time the cums attached on the ordered extent record
1626 * are inserted into the btree
1627 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001628static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001629 int mirror_num, unsigned long bio_flags,
1630 u64 bio_offset)
Chris Mason4a69a412008-11-06 22:03:00 -05001631{
1632 struct btrfs_root *root = BTRFS_I(inode)->root;
Stefan Behrens61891922012-11-05 18:51:52 +01001633 int ret;
1634
1635 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1636 if (ret)
1637 bio_endio(bio, ret);
1638 return ret;
Chris Mason44b8bd72008-04-16 11:14:51 -04001639}
1640
Chris Masond352ac62008-09-29 15:18:18 -04001641/*
Chris Masoncad321a2008-12-17 14:51:42 -05001642 * extent_io.c submission hook. This does the right thing for csum calculation
1643 * on write, or reading the csums from the tree before a read
Chris Masond352ac62008-09-29 15:18:18 -04001644 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001645static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
Chris Masoneaf25d92010-05-25 09:48:28 -04001646 int mirror_num, unsigned long bio_flags,
1647 u64 bio_offset)
Chris Mason44b8bd72008-04-16 11:14:51 -04001648{
1649 struct btrfs_root *root = BTRFS_I(inode)->root;
1650 int ret = 0;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001651 int skip_sum;
Jeff Mahoney04173412011-10-03 23:23:12 -04001652 int metadata = 0;
Josef Bacikb812ce22012-11-16 13:56:32 -05001653 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
Chris Mason44b8bd72008-04-16 11:14:51 -04001654
Christoph Hellwig6cbff002009-04-17 10:37:41 +02001655 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masoncad321a2008-12-17 14:51:42 -05001656
Liu Bo83eea1f2012-07-10 05:28:39 -06001657 if (btrfs_is_free_space_inode(inode))
Jeff Mahoney04173412011-10-03 23:23:12 -04001658 metadata = 2;
1659
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001660 if (!(rw & REQ_WRITE)) {
Josef Bacik5fd02042012-05-02 14:00:54 -04001661 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1662 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001663 goto out;
Josef Bacik5fd02042012-05-02 14:00:54 -04001664
Chris Masond20f7042008-12-08 16:58:54 -05001665 if (bio_flags & EXTENT_BIO_COMPRESSED) {
Stefan Behrens61891922012-11-05 18:51:52 +01001666 ret = btrfs_submit_compressed_read(inode, bio,
1667 mirror_num,
1668 bio_flags);
1669 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001670 } else if (!skip_sum) {
1671 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1672 if (ret)
Stefan Behrens61891922012-11-05 18:51:52 +01001673 goto out;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00001674 }
Chris Mason4d1b5fb2008-08-20 09:44:52 -04001675 goto mapit;
Josef Bacikb812ce22012-11-16 13:56:32 -05001676 } else if (async && !skip_sum) {
Yan Zheng17d217f2008-12-12 10:03:38 -05001677 /* csum items have already been cloned */
1678 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1679 goto mapit;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001680 /* we're doing a write, do the async checksumming */
Stefan Behrens61891922012-11-05 18:51:52 +01001681 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Chris Mason44b8bd72008-04-16 11:14:51 -04001682 inode, rw, bio, mirror_num,
Chris Masoneaf25d92010-05-25 09:48:28 -04001683 bio_flags, bio_offset,
1684 __btrfs_submit_bio_start,
Chris Mason4a69a412008-11-06 22:03:00 -05001685 __btrfs_submit_bio_done);
Stefan Behrens61891922012-11-05 18:51:52 +01001686 goto out;
Josef Bacikb812ce22012-11-16 13:56:32 -05001687 } else if (!skip_sum) {
1688 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1689 if (ret)
1690 goto out;
Chris Mason19b9bdb2008-10-30 14:23:13 -04001691 }
1692
Chris Mason0b86a832008-03-24 15:01:56 -04001693mapit:
Stefan Behrens61891922012-11-05 18:51:52 +01001694 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1695
1696out:
1697 if (ret < 0)
1698 bio_endio(bio, ret);
1699 return ret;
Chris Mason065631f2008-02-20 12:07:25 -05001700}
Chris Mason6885f302008-02-20 16:11:05 -05001701
Chris Masond352ac62008-09-29 15:18:18 -04001702/*
1703 * given a list of ordered sums record them in the inode. This happens
1704 * at IO completion time based on sums calculated at bio submission time.
1705 */
Chris Masonba1da2f2008-07-17 12:54:15 -04001706static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
Chris Masone6dcd2d2008-07-17 12:53:50 -04001707 struct inode *inode, u64 file_offset,
1708 struct list_head *list)
1709{
Chris Masone6dcd2d2008-07-17 12:53:50 -04001710 struct btrfs_ordered_sum *sum;
1711
Qinghuang Fengc6e30872009-01-21 10:59:08 -05001712 list_for_each_entry(sum, list, list) {
Miao Xie39847c42013-03-28 08:08:20 +00001713 trans->adding_csums = 1;
Chris Masond20f7042008-12-08 16:58:54 -05001714 btrfs_csum_file_blocks(trans,
1715 BTRFS_I(inode)->root->fs_info->csum_root, sum);
Miao Xie39847c42013-03-28 08:08:20 +00001716 trans->adding_csums = 0;
Chris Masone6dcd2d2008-07-17 12:53:50 -04001717 }
1718 return 0;
1719}
1720
Josef Bacik2ac55d42010-02-03 19:33:23 +00001721int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1722 struct extent_state **cached_state)
Chris Masonea8c2812008-08-04 23:17:27 -04001723{
Julia Lawall6c1500f2012-11-03 20:30:18 +00001724 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
Chris Masonea8c2812008-08-04 23:17:27 -04001725 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
Josef Bacik2ac55d42010-02-03 19:33:23 +00001726 cached_state, GFP_NOFS);
Chris Masonea8c2812008-08-04 23:17:27 -04001727}
1728
Chris Masond352ac62008-09-29 15:18:18 -04001729/* see btrfs_writepage_start_hook for details on why this is required */
Chris Mason247e7432008-07-17 12:53:51 -04001730struct btrfs_writepage_fixup {
1731 struct page *page;
1732 struct btrfs_work work;
1733};
1734
Christoph Hellwigb2950862008-12-02 09:54:17 -05001735static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
Chris Mason247e7432008-07-17 12:53:51 -04001736{
1737 struct btrfs_writepage_fixup *fixup;
1738 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00001739 struct extent_state *cached_state = NULL;
Chris Mason247e7432008-07-17 12:53:51 -04001740 struct page *page;
1741 struct inode *inode;
1742 u64 page_start;
1743 u64 page_end;
Jeff Mahoney87826df2012-02-15 16:23:57 +01001744 int ret;
Chris Mason247e7432008-07-17 12:53:51 -04001745
1746 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1747 page = fixup->page;
Chris Mason4a096752008-07-21 10:29:44 -04001748again:
Chris Mason247e7432008-07-17 12:53:51 -04001749 lock_page(page);
1750 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1751 ClearPageChecked(page);
1752 goto out_page;
1753 }
1754
1755 inode = page->mapping->host;
1756 page_start = page_offset(page);
1757 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1758
Josef Bacik2ac55d42010-02-03 19:33:23 +00001759 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01001760 &cached_state);
Chris Mason4a096752008-07-21 10:29:44 -04001761
1762 /* already ordered? We're done */
Chris Mason8b62b722009-09-02 16:53:46 -04001763 if (PagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001764 goto out;
Chris Mason4a096752008-07-21 10:29:44 -04001765
1766 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1767 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00001768 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1769 page_end, &cached_state, GFP_NOFS);
Chris Mason4a096752008-07-21 10:29:44 -04001770 unlock_page(page);
1771 btrfs_start_ordered_extent(inode, ordered, 1);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001772 btrfs_put_ordered_extent(ordered);
Chris Mason4a096752008-07-21 10:29:44 -04001773 goto again;
1774 }
Chris Mason247e7432008-07-17 12:53:51 -04001775
Jeff Mahoney87826df2012-02-15 16:23:57 +01001776 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1777 if (ret) {
1778 mapping_set_error(page->mapping, ret);
1779 end_extent_writepage(page, ret, page_start, page_end);
1780 ClearPageChecked(page);
1781 goto out;
1782 }
1783
Josef Bacik2ac55d42010-02-03 19:33:23 +00001784 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
Chris Mason247e7432008-07-17 12:53:51 -04001785 ClearPageChecked(page);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001786 set_page_dirty(page);
Chris Mason247e7432008-07-17 12:53:51 -04001787out:
Josef Bacik2ac55d42010-02-03 19:33:23 +00001788 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1789 &cached_state, GFP_NOFS);
Chris Mason247e7432008-07-17 12:53:51 -04001790out_page:
1791 unlock_page(page);
1792 page_cache_release(page);
Miao Xieb897abe2011-01-26 16:19:22 +08001793 kfree(fixup);
Chris Mason247e7432008-07-17 12:53:51 -04001794}
1795
1796/*
1797 * There are a few paths in the higher layers of the kernel that directly
1798 * set the page dirty bit without asking the filesystem if it is a
1799 * good idea. This causes problems because we want to make sure COW
1800 * properly happens and the data=ordered rules are followed.
1801 *
Chris Masonc8b97812008-10-29 14:49:59 -04001802 * In our case any range that doesn't have the ORDERED bit set
Chris Mason247e7432008-07-17 12:53:51 -04001803 * hasn't been properly setup for IO. We kick off an async process
1804 * to fix it up. The async helper will wait for ordered extents, set
1805 * the delalloc bit and make it safe to write the page.
1806 */
Christoph Hellwigb2950862008-12-02 09:54:17 -05001807static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
Chris Mason247e7432008-07-17 12:53:51 -04001808{
1809 struct inode *inode = page->mapping->host;
1810 struct btrfs_writepage_fixup *fixup;
1811 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason247e7432008-07-17 12:53:51 -04001812
Chris Mason8b62b722009-09-02 16:53:46 -04001813 /* this page is properly in the ordered list */
1814 if (TestClearPagePrivate2(page))
Chris Mason247e7432008-07-17 12:53:51 -04001815 return 0;
1816
1817 if (PageChecked(page))
1818 return -EAGAIN;
1819
1820 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1821 if (!fixup)
1822 return -EAGAIN;
Chris Masonf4219502008-07-22 11:18:09 -04001823
Chris Mason247e7432008-07-17 12:53:51 -04001824 SetPageChecked(page);
1825 page_cache_get(page);
1826 fixup->work.func = btrfs_writepage_fixup_worker;
1827 fixup->page = page;
1828 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
Jeff Mahoney87826df2012-02-15 16:23:57 +01001829 return -EBUSY;
Chris Mason247e7432008-07-17 12:53:51 -04001830}
1831
Yan Zhengd899e052008-10-30 14:25:28 -04001832static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1833 struct inode *inode, u64 file_pos,
1834 u64 disk_bytenr, u64 disk_num_bytes,
1835 u64 num_bytes, u64 ram_bytes,
1836 u8 compression, u8 encryption,
1837 u16 other_encoding, int extent_type)
1838{
1839 struct btrfs_root *root = BTRFS_I(inode)->root;
1840 struct btrfs_file_extent_item *fi;
1841 struct btrfs_path *path;
1842 struct extent_buffer *leaf;
1843 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04001844 int ret;
1845
1846 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07001847 if (!path)
1848 return -ENOMEM;
Yan Zhengd899e052008-10-30 14:25:28 -04001849
Chris Masonb9473432009-03-13 11:00:37 -04001850 path->leave_spinning = 1;
Chris Masona1ed8352009-09-11 12:27:37 -04001851
1852 /*
1853 * we may be replacing one extent in the tree with another.
1854 * The new extent is pinned in the extent map, and we don't want
1855 * to drop it from the cache until it is completely in the btree.
1856 *
1857 * So, tell btrfs_drop_extents to leave this extent in the cache.
1858 * the caller is expected to unpin it and allow it to be merged
1859 * with the others.
1860 */
Josef Bacik5dc562c2012-08-17 13:14:17 -04001861 ret = btrfs_drop_extents(trans, root, inode, file_pos,
Josef Bacik26714852012-08-29 12:24:27 -04001862 file_pos + num_bytes, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001863 if (ret)
1864 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04001865
Li Zefan33345d012011-04-20 10:31:50 +08001866 ins.objectid = btrfs_ino(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04001867 ins.offset = file_pos;
1868 ins.type = BTRFS_EXTENT_DATA_KEY;
1869 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001870 if (ret)
1871 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -04001872 leaf = path->nodes[0];
1873 fi = btrfs_item_ptr(leaf, path->slots[0],
1874 struct btrfs_file_extent_item);
1875 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1876 btrfs_set_file_extent_type(leaf, fi, extent_type);
1877 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1878 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1879 btrfs_set_file_extent_offset(leaf, fi, 0);
1880 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1881 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1882 btrfs_set_file_extent_compression(leaf, fi, compression);
1883 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1884 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
Chris Masonb9473432009-03-13 11:00:37 -04001885
Yan Zhengd899e052008-10-30 14:25:28 -04001886 btrfs_mark_buffer_dirty(leaf);
Josef Bacikce195332012-09-25 15:26:16 -04001887 btrfs_release_path(path);
Yan Zhengd899e052008-10-30 14:25:28 -04001888
1889 inode_add_bytes(inode, num_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -04001890
1891 ins.objectid = disk_bytenr;
1892 ins.offset = disk_num_bytes;
1893 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001894 ret = btrfs_alloc_reserved_file_extent(trans, root,
1895 root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08001896 btrfs_ino(inode), file_pos, &ins);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001897out:
Yan Zhengd899e052008-10-30 14:25:28 -04001898 btrfs_free_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04001899
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001900 return ret;
Yan Zhengd899e052008-10-30 14:25:28 -04001901}
1902
Liu Bo38c227d2013-01-29 03:18:40 +00001903/* snapshot-aware defrag */
1904struct sa_defrag_extent_backref {
1905 struct rb_node node;
1906 struct old_sa_defrag_extent *old;
1907 u64 root_id;
1908 u64 inum;
1909 u64 file_pos;
1910 u64 extent_offset;
1911 u64 num_bytes;
1912 u64 generation;
1913};
1914
1915struct old_sa_defrag_extent {
1916 struct list_head list;
1917 struct new_sa_defrag_extent *new;
1918
1919 u64 extent_offset;
1920 u64 bytenr;
1921 u64 offset;
1922 u64 len;
1923 int count;
1924};
1925
1926struct new_sa_defrag_extent {
1927 struct rb_root root;
1928 struct list_head head;
1929 struct btrfs_path *path;
1930 struct inode *inode;
1931 u64 file_pos;
1932 u64 len;
1933 u64 bytenr;
1934 u64 disk_len;
1935 u8 compress_type;
1936};
1937
1938static int backref_comp(struct sa_defrag_extent_backref *b1,
1939 struct sa_defrag_extent_backref *b2)
1940{
1941 if (b1->root_id < b2->root_id)
1942 return -1;
1943 else if (b1->root_id > b2->root_id)
1944 return 1;
1945
1946 if (b1->inum < b2->inum)
1947 return -1;
1948 else if (b1->inum > b2->inum)
1949 return 1;
1950
1951 if (b1->file_pos < b2->file_pos)
1952 return -1;
1953 else if (b1->file_pos > b2->file_pos)
1954 return 1;
1955
1956 /*
1957 * [------------------------------] ===> (a range of space)
1958 * |<--->| |<---->| =============> (fs/file tree A)
1959 * |<---------------------------->| ===> (fs/file tree B)
1960 *
1961 * A range of space can refer to two file extents in one tree while
1962 * refer to only one file extent in another tree.
1963 *
1964 * So we may process a disk offset more than one time(two extents in A)
1965 * and locate at the same extent(one extent in B), then insert two same
1966 * backrefs(both refer to the extent in B).
1967 */
1968 return 0;
1969}
1970
1971static void backref_insert(struct rb_root *root,
1972 struct sa_defrag_extent_backref *backref)
1973{
1974 struct rb_node **p = &root->rb_node;
1975 struct rb_node *parent = NULL;
1976 struct sa_defrag_extent_backref *entry;
1977 int ret;
1978
1979 while (*p) {
1980 parent = *p;
1981 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1982
1983 ret = backref_comp(backref, entry);
1984 if (ret < 0)
1985 p = &(*p)->rb_left;
1986 else
1987 p = &(*p)->rb_right;
1988 }
1989
1990 rb_link_node(&backref->node, parent, p);
1991 rb_insert_color(&backref->node, root);
1992}
1993
1994/*
1995 * Note the backref might has changed, and in this case we just return 0.
1996 */
1997static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
1998 void *ctx)
1999{
2000 struct btrfs_file_extent_item *extent;
2001 struct btrfs_fs_info *fs_info;
2002 struct old_sa_defrag_extent *old = ctx;
2003 struct new_sa_defrag_extent *new = old->new;
2004 struct btrfs_path *path = new->path;
2005 struct btrfs_key key;
2006 struct btrfs_root *root;
2007 struct sa_defrag_extent_backref *backref;
2008 struct extent_buffer *leaf;
2009 struct inode *inode = new->inode;
2010 int slot;
2011 int ret;
2012 u64 extent_offset;
2013 u64 num_bytes;
2014
2015 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2016 inum == btrfs_ino(inode))
2017 return 0;
2018
2019 key.objectid = root_id;
2020 key.type = BTRFS_ROOT_ITEM_KEY;
2021 key.offset = (u64)-1;
2022
2023 fs_info = BTRFS_I(inode)->root->fs_info;
2024 root = btrfs_read_fs_root_no_name(fs_info, &key);
2025 if (IS_ERR(root)) {
2026 if (PTR_ERR(root) == -ENOENT)
2027 return 0;
2028 WARN_ON(1);
2029 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2030 inum, offset, root_id);
2031 return PTR_ERR(root);
2032 }
2033
2034 key.objectid = inum;
2035 key.type = BTRFS_EXTENT_DATA_KEY;
2036 if (offset > (u64)-1 << 32)
2037 key.offset = 0;
2038 else
2039 key.offset = offset;
2040
2041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05302042 if (WARN_ON(ret < 0))
Liu Bo38c227d2013-01-29 03:18:40 +00002043 return ret;
Josef Bacik50f13192013-07-22 12:50:37 -04002044 ret = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002045
2046 while (1) {
2047 cond_resched();
2048
2049 leaf = path->nodes[0];
2050 slot = path->slots[0];
2051
2052 if (slot >= btrfs_header_nritems(leaf)) {
2053 ret = btrfs_next_leaf(root, path);
2054 if (ret < 0) {
2055 goto out;
2056 } else if (ret > 0) {
2057 ret = 0;
2058 goto out;
2059 }
2060 continue;
2061 }
2062
2063 path->slots[0]++;
2064
2065 btrfs_item_key_to_cpu(leaf, &key, slot);
2066
2067 if (key.objectid > inum)
2068 goto out;
2069
2070 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2071 continue;
2072
2073 extent = btrfs_item_ptr(leaf, slot,
2074 struct btrfs_file_extent_item);
2075
2076 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2077 continue;
2078
Liu Boe68afa42013-07-01 22:13:26 +08002079 /*
2080 * 'offset' refers to the exact key.offset,
2081 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2082 * (key.offset - extent_offset).
2083 */
2084 if (key.offset != offset)
Liu Bo38c227d2013-01-29 03:18:40 +00002085 continue;
2086
Liu Boe68afa42013-07-01 22:13:26 +08002087 extent_offset = btrfs_file_extent_offset(leaf, extent);
Liu Bo38c227d2013-01-29 03:18:40 +00002088 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
Liu Boe68afa42013-07-01 22:13:26 +08002089
Liu Bo38c227d2013-01-29 03:18:40 +00002090 if (extent_offset >= old->extent_offset + old->offset +
2091 old->len || extent_offset + num_bytes <=
2092 old->extent_offset + old->offset)
2093 continue;
Liu Bo38c227d2013-01-29 03:18:40 +00002094 break;
2095 }
2096
2097 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2098 if (!backref) {
2099 ret = -ENOENT;
2100 goto out;
2101 }
2102
2103 backref->root_id = root_id;
2104 backref->inum = inum;
Liu Boe68afa42013-07-01 22:13:26 +08002105 backref->file_pos = offset;
Liu Bo38c227d2013-01-29 03:18:40 +00002106 backref->num_bytes = num_bytes;
2107 backref->extent_offset = extent_offset;
2108 backref->generation = btrfs_file_extent_generation(leaf, extent);
2109 backref->old = old;
2110 backref_insert(&new->root, backref);
2111 old->count++;
2112out:
2113 btrfs_release_path(path);
2114 WARN_ON(ret);
2115 return ret;
2116}
2117
2118static noinline bool record_extent_backrefs(struct btrfs_path *path,
2119 struct new_sa_defrag_extent *new)
2120{
2121 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2122 struct old_sa_defrag_extent *old, *tmp;
2123 int ret;
2124
2125 new->path = path;
2126
2127 list_for_each_entry_safe(old, tmp, &new->head, list) {
Liu Boe68afa42013-07-01 22:13:26 +08002128 ret = iterate_inodes_from_logical(old->bytenr +
2129 old->extent_offset, fs_info,
Liu Bo38c227d2013-01-29 03:18:40 +00002130 path, record_one_backref,
2131 old);
Josef Bacik4724b102013-11-05 11:11:40 -05002132 if (ret < 0 && ret != -ENOENT)
2133 return false;
Liu Bo38c227d2013-01-29 03:18:40 +00002134
2135 /* no backref to be processed for this extent */
2136 if (!old->count) {
2137 list_del(&old->list);
2138 kfree(old);
2139 }
2140 }
2141
2142 if (list_empty(&new->head))
2143 return false;
2144
2145 return true;
2146}
2147
2148static int relink_is_mergable(struct extent_buffer *leaf,
2149 struct btrfs_file_extent_item *fi,
Liu Bo116e0022013-08-02 16:30:40 +08002150 struct new_sa_defrag_extent *new)
Liu Bo38c227d2013-01-29 03:18:40 +00002151{
Liu Bo116e0022013-08-02 16:30:40 +08002152 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
Liu Bo38c227d2013-01-29 03:18:40 +00002153 return 0;
2154
2155 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2156 return 0;
2157
Liu Bo116e0022013-08-02 16:30:40 +08002158 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2159 return 0;
2160
2161 if (btrfs_file_extent_encryption(leaf, fi) ||
Liu Bo38c227d2013-01-29 03:18:40 +00002162 btrfs_file_extent_other_encoding(leaf, fi))
2163 return 0;
2164
2165 return 1;
2166}
2167
2168/*
2169 * Note the backref might has changed, and in this case we just return 0.
2170 */
2171static noinline int relink_extent_backref(struct btrfs_path *path,
2172 struct sa_defrag_extent_backref *prev,
2173 struct sa_defrag_extent_backref *backref)
2174{
2175 struct btrfs_file_extent_item *extent;
2176 struct btrfs_file_extent_item *item;
2177 struct btrfs_ordered_extent *ordered;
2178 struct btrfs_trans_handle *trans;
2179 struct btrfs_fs_info *fs_info;
2180 struct btrfs_root *root;
2181 struct btrfs_key key;
2182 struct extent_buffer *leaf;
2183 struct old_sa_defrag_extent *old = backref->old;
2184 struct new_sa_defrag_extent *new = old->new;
2185 struct inode *src_inode = new->inode;
2186 struct inode *inode;
2187 struct extent_state *cached = NULL;
2188 int ret = 0;
2189 u64 start;
2190 u64 len;
2191 u64 lock_start;
2192 u64 lock_end;
2193 bool merge = false;
2194 int index;
2195
2196 if (prev && prev->root_id == backref->root_id &&
2197 prev->inum == backref->inum &&
2198 prev->file_pos + prev->num_bytes == backref->file_pos)
2199 merge = true;
2200
2201 /* step 1: get root */
2202 key.objectid = backref->root_id;
2203 key.type = BTRFS_ROOT_ITEM_KEY;
2204 key.offset = (u64)-1;
2205
2206 fs_info = BTRFS_I(src_inode)->root->fs_info;
2207 index = srcu_read_lock(&fs_info->subvol_srcu);
2208
2209 root = btrfs_read_fs_root_no_name(fs_info, &key);
2210 if (IS_ERR(root)) {
2211 srcu_read_unlock(&fs_info->subvol_srcu, index);
2212 if (PTR_ERR(root) == -ENOENT)
2213 return 0;
2214 return PTR_ERR(root);
2215 }
Liu Bo38c227d2013-01-29 03:18:40 +00002216
2217 /* step 2: get inode */
2218 key.objectid = backref->inum;
2219 key.type = BTRFS_INODE_ITEM_KEY;
2220 key.offset = 0;
2221
2222 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2223 if (IS_ERR(inode)) {
2224 srcu_read_unlock(&fs_info->subvol_srcu, index);
2225 return 0;
2226 }
2227
2228 srcu_read_unlock(&fs_info->subvol_srcu, index);
2229
2230 /* step 3: relink backref */
2231 lock_start = backref->file_pos;
2232 lock_end = backref->file_pos + backref->num_bytes - 1;
2233 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2234 0, &cached);
2235
2236 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2237 if (ordered) {
2238 btrfs_put_ordered_extent(ordered);
2239 goto out_unlock;
2240 }
2241
2242 trans = btrfs_join_transaction(root);
2243 if (IS_ERR(trans)) {
2244 ret = PTR_ERR(trans);
2245 goto out_unlock;
2246 }
2247
2248 key.objectid = backref->inum;
2249 key.type = BTRFS_EXTENT_DATA_KEY;
2250 key.offset = backref->file_pos;
2251
2252 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2253 if (ret < 0) {
2254 goto out_free_path;
2255 } else if (ret > 0) {
2256 ret = 0;
2257 goto out_free_path;
2258 }
2259
2260 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2261 struct btrfs_file_extent_item);
2262
2263 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2264 backref->generation)
2265 goto out_free_path;
2266
2267 btrfs_release_path(path);
2268
2269 start = backref->file_pos;
2270 if (backref->extent_offset < old->extent_offset + old->offset)
2271 start += old->extent_offset + old->offset -
2272 backref->extent_offset;
2273
2274 len = min(backref->extent_offset + backref->num_bytes,
2275 old->extent_offset + old->offset + old->len);
2276 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2277
2278 ret = btrfs_drop_extents(trans, root, inode, start,
2279 start + len, 1);
2280 if (ret)
2281 goto out_free_path;
2282again:
2283 key.objectid = btrfs_ino(inode);
2284 key.type = BTRFS_EXTENT_DATA_KEY;
2285 key.offset = start;
2286
Liu Boa09a0a72013-03-11 09:20:58 +00002287 path->leave_spinning = 1;
Liu Bo38c227d2013-01-29 03:18:40 +00002288 if (merge) {
2289 struct btrfs_file_extent_item *fi;
2290 u64 extent_len;
2291 struct btrfs_key found_key;
2292
2293 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2294 if (ret < 0)
2295 goto out_free_path;
2296
2297 path->slots[0]--;
2298 leaf = path->nodes[0];
2299 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2300
2301 fi = btrfs_item_ptr(leaf, path->slots[0],
2302 struct btrfs_file_extent_item);
2303 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2304
Liu Bo116e0022013-08-02 16:30:40 +08002305 if (extent_len + found_key.offset == start &&
2306 relink_is_mergable(leaf, fi, new)) {
Liu Bo38c227d2013-01-29 03:18:40 +00002307 btrfs_set_file_extent_num_bytes(leaf, fi,
2308 extent_len + len);
2309 btrfs_mark_buffer_dirty(leaf);
2310 inode_add_bytes(inode, len);
2311
2312 ret = 1;
2313 goto out_free_path;
2314 } else {
2315 merge = false;
2316 btrfs_release_path(path);
2317 goto again;
2318 }
2319 }
2320
2321 ret = btrfs_insert_empty_item(trans, root, path, &key,
2322 sizeof(*extent));
2323 if (ret) {
2324 btrfs_abort_transaction(trans, root, ret);
2325 goto out_free_path;
2326 }
2327
2328 leaf = path->nodes[0];
2329 item = btrfs_item_ptr(leaf, path->slots[0],
2330 struct btrfs_file_extent_item);
2331 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2332 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2333 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2334 btrfs_set_file_extent_num_bytes(leaf, item, len);
2335 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2336 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2337 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2338 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2339 btrfs_set_file_extent_encryption(leaf, item, 0);
2340 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2341
2342 btrfs_mark_buffer_dirty(leaf);
2343 inode_add_bytes(inode, len);
Liu Boa09a0a72013-03-11 09:20:58 +00002344 btrfs_release_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002345
2346 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2347 new->disk_len, 0,
2348 backref->root_id, backref->inum,
2349 new->file_pos, 0); /* start - extent_offset */
2350 if (ret) {
2351 btrfs_abort_transaction(trans, root, ret);
2352 goto out_free_path;
2353 }
2354
2355 ret = 1;
2356out_free_path:
2357 btrfs_release_path(path);
Liu Boa09a0a72013-03-11 09:20:58 +00002358 path->leave_spinning = 0;
Liu Bo38c227d2013-01-29 03:18:40 +00002359 btrfs_end_transaction(trans, root);
2360out_unlock:
2361 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2362 &cached, GFP_NOFS);
2363 iput(inode);
2364 return ret;
2365}
2366
Liu Bo6f519562013-10-29 10:45:05 +08002367static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2368{
2369 struct old_sa_defrag_extent *old, *tmp;
2370
2371 if (!new)
2372 return;
2373
2374 list_for_each_entry_safe(old, tmp, &new->head, list) {
2375 list_del(&old->list);
2376 kfree(old);
2377 }
2378 kfree(new);
2379}
2380
Liu Bo38c227d2013-01-29 03:18:40 +00002381static void relink_file_extents(struct new_sa_defrag_extent *new)
2382{
2383 struct btrfs_path *path;
Liu Bo38c227d2013-01-29 03:18:40 +00002384 struct sa_defrag_extent_backref *backref;
2385 struct sa_defrag_extent_backref *prev = NULL;
2386 struct inode *inode;
2387 struct btrfs_root *root;
2388 struct rb_node *node;
2389 int ret;
2390
2391 inode = new->inode;
2392 root = BTRFS_I(inode)->root;
2393
2394 path = btrfs_alloc_path();
2395 if (!path)
2396 return;
2397
2398 if (!record_extent_backrefs(path, new)) {
2399 btrfs_free_path(path);
2400 goto out;
2401 }
2402 btrfs_release_path(path);
2403
2404 while (1) {
2405 node = rb_first(&new->root);
2406 if (!node)
2407 break;
2408 rb_erase(node, &new->root);
2409
2410 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2411
2412 ret = relink_extent_backref(path, prev, backref);
2413 WARN_ON(ret < 0);
2414
2415 kfree(prev);
2416
2417 if (ret == 1)
2418 prev = backref;
2419 else
2420 prev = NULL;
2421 cond_resched();
2422 }
2423 kfree(prev);
2424
2425 btrfs_free_path(path);
Liu Bo38c227d2013-01-29 03:18:40 +00002426out:
Liu Bo6f519562013-10-29 10:45:05 +08002427 free_sa_defrag_extent(new);
2428
Liu Bo38c227d2013-01-29 03:18:40 +00002429 atomic_dec(&root->fs_info->defrag_running);
2430 wake_up(&root->fs_info->transaction_wait);
Liu Bo38c227d2013-01-29 03:18:40 +00002431}
2432
2433static struct new_sa_defrag_extent *
2434record_old_file_extents(struct inode *inode,
2435 struct btrfs_ordered_extent *ordered)
2436{
2437 struct btrfs_root *root = BTRFS_I(inode)->root;
2438 struct btrfs_path *path;
2439 struct btrfs_key key;
Liu Bo6f519562013-10-29 10:45:05 +08002440 struct old_sa_defrag_extent *old;
Liu Bo38c227d2013-01-29 03:18:40 +00002441 struct new_sa_defrag_extent *new;
2442 int ret;
2443
2444 new = kmalloc(sizeof(*new), GFP_NOFS);
2445 if (!new)
2446 return NULL;
2447
2448 new->inode = inode;
2449 new->file_pos = ordered->file_offset;
2450 new->len = ordered->len;
2451 new->bytenr = ordered->start;
2452 new->disk_len = ordered->disk_len;
2453 new->compress_type = ordered->compress_type;
2454 new->root = RB_ROOT;
2455 INIT_LIST_HEAD(&new->head);
2456
2457 path = btrfs_alloc_path();
2458 if (!path)
2459 goto out_kfree;
2460
2461 key.objectid = btrfs_ino(inode);
2462 key.type = BTRFS_EXTENT_DATA_KEY;
2463 key.offset = new->file_pos;
2464
2465 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2466 if (ret < 0)
2467 goto out_free_path;
2468 if (ret > 0 && path->slots[0] > 0)
2469 path->slots[0]--;
2470
2471 /* find out all the old extents for the file range */
2472 while (1) {
2473 struct btrfs_file_extent_item *extent;
2474 struct extent_buffer *l;
2475 int slot;
2476 u64 num_bytes;
2477 u64 offset;
2478 u64 end;
2479 u64 disk_bytenr;
2480 u64 extent_offset;
2481
2482 l = path->nodes[0];
2483 slot = path->slots[0];
2484
2485 if (slot >= btrfs_header_nritems(l)) {
2486 ret = btrfs_next_leaf(root, path);
2487 if (ret < 0)
Liu Bo6f519562013-10-29 10:45:05 +08002488 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002489 else if (ret > 0)
2490 break;
2491 continue;
2492 }
2493
2494 btrfs_item_key_to_cpu(l, &key, slot);
2495
2496 if (key.objectid != btrfs_ino(inode))
2497 break;
2498 if (key.type != BTRFS_EXTENT_DATA_KEY)
2499 break;
2500 if (key.offset >= new->file_pos + new->len)
2501 break;
2502
2503 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2504
2505 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2506 if (key.offset + num_bytes < new->file_pos)
2507 goto next;
2508
2509 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2510 if (!disk_bytenr)
2511 goto next;
2512
2513 extent_offset = btrfs_file_extent_offset(l, extent);
2514
2515 old = kmalloc(sizeof(*old), GFP_NOFS);
2516 if (!old)
Liu Bo6f519562013-10-29 10:45:05 +08002517 goto out_free_path;
Liu Bo38c227d2013-01-29 03:18:40 +00002518
2519 offset = max(new->file_pos, key.offset);
2520 end = min(new->file_pos + new->len, key.offset + num_bytes);
2521
2522 old->bytenr = disk_bytenr;
2523 old->extent_offset = extent_offset;
2524 old->offset = offset - key.offset;
2525 old->len = end - offset;
2526 old->new = new;
2527 old->count = 0;
2528 list_add_tail(&old->list, &new->head);
2529next:
2530 path->slots[0]++;
2531 cond_resched();
2532 }
2533
2534 btrfs_free_path(path);
2535 atomic_inc(&root->fs_info->defrag_running);
2536
2537 return new;
2538
Liu Bo38c227d2013-01-29 03:18:40 +00002539out_free_path:
2540 btrfs_free_path(path);
2541out_kfree:
Liu Bo6f519562013-10-29 10:45:05 +08002542 free_sa_defrag_extent(new);
Liu Bo38c227d2013-01-29 03:18:40 +00002543 return NULL;
2544}
2545
Chris Mason5d13a982009-03-13 11:41:46 -04002546/*
2547 * helper function for btrfs_finish_ordered_io, this
2548 * just reads in some of the csum leaves to prime them into ram
2549 * before we start the transaction. It limits the amount of btree
2550 * reads required while inside the transaction.
2551 */
Chris Masond352ac62008-09-29 15:18:18 -04002552/* as ordered data IO finishes, this gets called so we can finish
2553 * an ordered extent if the range of bytes in the file it covers are
2554 * fully written.
2555 */
Josef Bacik5fd02042012-05-02 14:00:54 -04002556static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
Chris Masone6dcd2d2008-07-17 12:53:50 -04002557{
Josef Bacik5fd02042012-05-02 14:00:54 -04002558 struct inode *inode = ordered_extent->inode;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002559 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002560 struct btrfs_trans_handle *trans = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002561 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Josef Bacik2ac55d42010-02-03 19:33:23 +00002562 struct extent_state *cached_state = NULL;
Liu Bo38c227d2013-01-29 03:18:40 +00002563 struct new_sa_defrag_extent *new = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08002564 int compress_type = 0;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002565 int ret = 0;
2566 u64 logical_len = ordered_extent->len;
Li Zefan82d59022011-04-20 10:33:24 +08002567 bool nolock;
Josef Bacik77cef2e2013-08-29 13:57:21 -04002568 bool truncated = false;
Chris Masone6dcd2d2008-07-17 12:53:50 -04002569
Liu Bo83eea1f2012-07-10 05:28:39 -06002570 nolock = btrfs_is_free_space_inode(inode);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002571
Josef Bacik5fd02042012-05-02 14:00:54 -04002572 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2573 ret = -EIO;
2574 goto out;
2575 }
2576
Josef Bacik77cef2e2013-08-29 13:57:21 -04002577 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2578 truncated = true;
2579 logical_len = ordered_extent->truncated_len;
2580 /* Truncated the entire extent, don't bother adding */
2581 if (!logical_len)
2582 goto out;
2583 }
2584
Yan, Zhengc2167752009-11-12 09:34:21 +00002585 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002586 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
Josef Bacik6c760c02012-11-09 10:53:21 -05002587 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2588 if (nolock)
2589 trans = btrfs_join_transaction_nolock(root);
2590 else
2591 trans = btrfs_join_transaction(root);
2592 if (IS_ERR(trans)) {
2593 ret = PTR_ERR(trans);
2594 trans = NULL;
2595 goto out;
Yan, Zhengc2167752009-11-12 09:34:21 +00002596 }
Josef Bacik6c760c02012-11-09 10:53:21 -05002597 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2598 ret = btrfs_update_inode_fallback(trans, root, inode);
2599 if (ret) /* -ENOMEM or corruption */
2600 btrfs_abort_transaction(trans, root, ret);
Yan, Zhengc2167752009-11-12 09:34:21 +00002601 goto out;
2602 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04002603
Josef Bacik2ac55d42010-02-03 19:33:23 +00002604 lock_extent_bits(io_tree, ordered_extent->file_offset,
2605 ordered_extent->file_offset + ordered_extent->len - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01002606 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002607
Liu Bo38c227d2013-01-29 03:18:40 +00002608 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2609 ordered_extent->file_offset + ordered_extent->len - 1,
2610 EXTENT_DEFRAG, 1, cached_state);
2611 if (ret) {
2612 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2613 if (last_snapshot >= BTRFS_I(inode)->generation)
2614 /* the inode is shared */
2615 new = record_old_file_extents(inode, ordered_extent);
2616
2617 clear_extent_bit(io_tree, ordered_extent->file_offset,
2618 ordered_extent->file_offset + ordered_extent->len - 1,
2619 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2620 }
2621
Josef Bacik0cb59c92010-07-02 12:14:14 -04002622 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002623 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002624 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04002625 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002626 if (IS_ERR(trans)) {
2627 ret = PTR_ERR(trans);
2628 trans = NULL;
2629 goto out_unlock;
2630 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04002631 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Yan, Zhengc2167752009-11-12 09:34:21 +00002632
Chris Masonc8b97812008-10-29 14:49:59 -04002633 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
Li Zefan261507a02010-12-17 14:21:50 +08002634 compress_type = ordered_extent->compress_type;
Yan Zhengd899e052008-10-30 14:25:28 -04002635 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
Li Zefan261507a02010-12-17 14:21:50 +08002636 BUG_ON(compress_type);
Yan, Zheng920bbbf2009-11-12 09:34:08 +00002637 ret = btrfs_mark_extent_written(trans, inode,
Yan Zhengd899e052008-10-30 14:25:28 -04002638 ordered_extent->file_offset,
2639 ordered_extent->file_offset +
Josef Bacik77cef2e2013-08-29 13:57:21 -04002640 logical_len);
Yan Zhengd899e052008-10-30 14:25:28 -04002641 } else {
Josef Bacik0af3d002010-06-21 14:48:16 -04002642 BUG_ON(root == root->fs_info->tree_root);
Yan Zhengd899e052008-10-30 14:25:28 -04002643 ret = insert_reserved_file_extent(trans, inode,
2644 ordered_extent->file_offset,
2645 ordered_extent->start,
2646 ordered_extent->disk_len,
Josef Bacik77cef2e2013-08-29 13:57:21 -04002647 logical_len, logical_len,
Li Zefan261507a02010-12-17 14:21:50 +08002648 compress_type, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04002649 BTRFS_FILE_EXTENT_REG);
Yan Zhengd899e052008-10-30 14:25:28 -04002650 }
Josef Bacik5dc562c2012-08-17 13:14:17 -04002651 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2652 ordered_extent->file_offset, ordered_extent->len,
2653 trans->transid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002654 if (ret < 0) {
2655 btrfs_abort_transaction(trans, root, ret);
Josef Bacik5fd02042012-05-02 14:00:54 -04002656 goto out_unlock;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002657 }
Josef Bacik2ac55d42010-02-03 19:33:23 +00002658
Chris Masone6dcd2d2008-07-17 12:53:50 -04002659 add_pending_csums(trans, inode, ordered_extent->file_offset,
2660 &ordered_extent->list);
2661
Josef Bacik6c760c02012-11-09 10:53:21 -05002662 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2663 ret = btrfs_update_inode_fallback(trans, root, inode);
2664 if (ret) { /* -ENOMEM or corruption */
2665 btrfs_abort_transaction(trans, root, ret);
2666 goto out_unlock;
Josef Bacik1ef30be2011-04-05 19:25:36 -04002667 }
2668 ret = 0;
Josef Bacik5fd02042012-05-02 14:00:54 -04002669out_unlock:
2670 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2671 ordered_extent->file_offset +
2672 ordered_extent->len - 1, &cached_state, GFP_NOFS);
Yan, Zhengc2167752009-11-12 09:34:21 +00002673out:
Josef Bacik5b0e95b2011-10-06 08:58:24 -04002674 if (root != root->fs_info->tree_root)
Josef Bacik0cb59c92010-07-02 12:14:14 -04002675 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Miao Xiea698d0752012-09-20 01:51:59 -06002676 if (trans)
2677 btrfs_end_transaction(trans, root);
Josef Bacik0cb59c92010-07-02 12:14:14 -04002678
Josef Bacik77cef2e2013-08-29 13:57:21 -04002679 if (ret || truncated) {
2680 u64 start, end;
2681
2682 if (truncated)
2683 start = ordered_extent->file_offset + logical_len;
2684 else
2685 start = ordered_extent->file_offset;
2686 end = ordered_extent->file_offset + ordered_extent->len - 1;
2687 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2688
2689 /* Drop the cache for the part of the extent we didn't write. */
2690 btrfs_drop_extent_cache(inode, start, end, 0);
Josef Bacik5fd02042012-05-02 14:00:54 -04002691
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002692 /*
2693 * If the ordered extent had an IOERR or something else went
2694 * wrong we need to return the space for this ordered extent
Josef Bacik77cef2e2013-08-29 13:57:21 -04002695 * back to the allocator. We only free the extent in the
2696 * truncated case if we didn't write out the extent at all.
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002697 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04002698 if ((ret || !logical_len) &&
2699 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
Josef Bacik0bec9ef2013-01-31 14:58:00 -05002700 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2701 btrfs_free_reserved_extent(root, ordered_extent->start,
2702 ordered_extent->disk_len);
2703 }
2704
2705
Josef Bacik5fd02042012-05-02 14:00:54 -04002706 /*
Liu Bo8bad3c02012-06-18 12:14:23 +08002707 * This needs to be done to make sure anybody waiting knows we are done
2708 * updating everything for this ordered extent.
Josef Bacik5fd02042012-05-02 14:00:54 -04002709 */
2710 btrfs_remove_ordered_extent(inode, ordered_extent);
2711
Liu Bo38c227d2013-01-29 03:18:40 +00002712 /* for snapshot-aware defrag */
Liu Bo6f519562013-10-29 10:45:05 +08002713 if (new) {
2714 if (ret) {
2715 free_sa_defrag_extent(new);
2716 atomic_dec(&root->fs_info->defrag_running);
2717 } else {
2718 relink_file_extents(new);
2719 }
2720 }
Liu Bo38c227d2013-01-29 03:18:40 +00002721
Chris Masone6dcd2d2008-07-17 12:53:50 -04002722 /* once for us */
2723 btrfs_put_ordered_extent(ordered_extent);
2724 /* once for the tree */
2725 btrfs_put_ordered_extent(ordered_extent);
2726
Josef Bacik5fd02042012-05-02 14:00:54 -04002727 return ret;
2728}
2729
2730static void finish_ordered_fn(struct btrfs_work *work)
2731{
2732 struct btrfs_ordered_extent *ordered_extent;
2733 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2734 btrfs_finish_ordered_io(ordered_extent);
Chris Masone6dcd2d2008-07-17 12:53:50 -04002735}
2736
Christoph Hellwigb2950862008-12-02 09:54:17 -05002737static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
Chris Mason211f90e2008-07-18 11:56:15 -04002738 struct extent_state *state, int uptodate)
2739{
Josef Bacik5fd02042012-05-02 14:00:54 -04002740 struct inode *inode = page->mapping->host;
2741 struct btrfs_root *root = BTRFS_I(inode)->root;
2742 struct btrfs_ordered_extent *ordered_extent = NULL;
2743 struct btrfs_workers *workers;
2744
liubo1abe9b82011-03-24 11:18:59 +00002745 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2746
Chris Mason8b62b722009-09-02 16:53:46 -04002747 ClearPagePrivate2(page);
Josef Bacik5fd02042012-05-02 14:00:54 -04002748 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2749 end - start + 1, uptodate))
2750 return 0;
2751
2752 ordered_extent->work.func = finish_ordered_fn;
2753 ordered_extent->work.flags = 0;
2754
Liu Bo83eea1f2012-07-10 05:28:39 -06002755 if (btrfs_is_free_space_inode(inode))
Josef Bacik5fd02042012-05-02 14:00:54 -04002756 workers = &root->fs_info->endio_freespace_worker;
2757 else
2758 workers = &root->fs_info->endio_write_workers;
2759 btrfs_queue_worker(workers, &ordered_extent->work);
2760
2761 return 0;
Chris Mason211f90e2008-07-18 11:56:15 -04002762}
2763
Chris Masond352ac62008-09-29 15:18:18 -04002764/*
Chris Masond352ac62008-09-29 15:18:18 -04002765 * when reads are done, we need to check csums to verify the data is correct
Jan Schmidt4a54c8c2011-07-22 15:41:52 +02002766 * if there's a match, we allow the bio to finish. If not, the code in
2767 * extent_io.c will try to find good copies for us.
Chris Masond352ac62008-09-29 15:18:18 -04002768 */
Miao Xiefacc8a222013-07-25 19:22:34 +08002769static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2770 u64 phy_offset, struct page *page,
2771 u64 start, u64 end, int mirror)
Chris Mason07157aa2007-08-30 08:50:51 -04002772{
Miao Xie4eee4fa2012-12-21 09:17:45 +00002773 size_t offset = start - page_offset(page);
Chris Mason07157aa2007-08-30 08:50:51 -04002774 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05002775 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason07157aa2007-08-30 08:50:51 -04002776 char *kaddr;
Chris Masonff79f812007-10-15 16:22:25 -04002777 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xiefacc8a222013-07-25 19:22:34 +08002778 u32 csum_expected;
Chris Masonff79f812007-10-15 16:22:25 -04002779 u32 csum = ~(u32)0;
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002780 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2781 DEFAULT_RATELIMIT_BURST);
Chris Masond1310b22008-01-24 16:13:08 -05002782
Chris Masond20f7042008-12-08 16:58:54 -05002783 if (PageChecked(page)) {
2784 ClearPageChecked(page);
2785 goto good;
2786 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02002787
2788 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Jan Schmidt08d2f342011-05-04 16:18:50 +02002789 goto good;
Chris Masond20f7042008-12-08 16:58:54 -05002790
Yan Zheng17d217f2008-12-12 10:03:38 -05002791 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
Chris Mason9655d292009-09-02 15:22:30 -04002792 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
Yan Zheng17d217f2008-12-12 10:03:38 -05002793 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2794 GFP_NOFS);
2795 return 0;
2796 }
2797
Miao Xiefacc8a222013-07-25 19:22:34 +08002798 phy_offset >>= inode->i_sb->s_blocksize_bits;
2799 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
Chris Masond3977122009-01-05 21:25:51 -05002800
Miao Xiefacc8a222013-07-25 19:22:34 +08002801 kaddr = kmap_atomic(page);
Liu Bob0496682013-03-14 14:57:45 +00002802 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
Chris Masonff79f812007-10-15 16:22:25 -04002803 btrfs_csum_final(csum, (char *)&csum);
Miao Xiefacc8a222013-07-25 19:22:34 +08002804 if (csum != csum_expected)
Chris Mason07157aa2007-08-30 08:50:51 -04002805 goto zeroit;
Chris Masond3977122009-01-05 21:25:51 -05002806
Cong Wang7ac687d2011-11-25 23:14:28 +08002807 kunmap_atomic(kaddr);
Chris Masond20f7042008-12-08 16:58:54 -05002808good:
Chris Mason07157aa2007-08-30 08:50:51 -04002809 return 0;
2810
2811zeroit:
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002812 if (__ratelimit(&_rs))
Miao Xiefacc8a222013-07-25 19:22:34 +08002813 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02002814 btrfs_ino(page->mapping->host), start, csum, csum_expected);
Chris Masondb945352007-10-15 16:15:53 -04002815 memset(kaddr + offset, 1, end - start + 1);
2816 flush_dcache_page(page);
Cong Wang7ac687d2011-11-25 23:14:28 +08002817 kunmap_atomic(kaddr);
Miao Xiefacc8a222013-07-25 19:22:34 +08002818 if (csum_expected == 0)
Chris Mason3b951512008-04-17 11:29:12 -04002819 return 0;
Chris Mason7e383262008-04-09 16:28:12 -04002820 return -EIO;
Chris Mason07157aa2007-08-30 08:50:51 -04002821}
Chris Masonb888db22007-08-27 16:49:44 -04002822
Yan, Zheng24bbcf042009-11-12 09:36:34 +00002823struct delayed_iput {
2824 struct list_head list;
2825 struct inode *inode;
2826};
2827
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002828/* JDM: If this is fs-wide, why can't we add a pointer to
2829 * btrfs_inode instead and avoid the allocation? */
Yan, Zheng24bbcf042009-11-12 09:36:34 +00002830void btrfs_add_delayed_iput(struct inode *inode)
2831{
2832 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2833 struct delayed_iput *delayed;
2834
2835 if (atomic_add_unless(&inode->i_count, -1, 1))
2836 return;
2837
2838 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2839 delayed->inode = inode;
2840
2841 spin_lock(&fs_info->delayed_iput_lock);
2842 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2843 spin_unlock(&fs_info->delayed_iput_lock);
2844}
2845
2846void btrfs_run_delayed_iputs(struct btrfs_root *root)
2847{
2848 LIST_HEAD(list);
2849 struct btrfs_fs_info *fs_info = root->fs_info;
2850 struct delayed_iput *delayed;
2851 int empty;
2852
2853 spin_lock(&fs_info->delayed_iput_lock);
2854 empty = list_empty(&fs_info->delayed_iputs);
2855 spin_unlock(&fs_info->delayed_iput_lock);
2856 if (empty)
2857 return;
2858
Yan, Zheng24bbcf042009-11-12 09:36:34 +00002859 spin_lock(&fs_info->delayed_iput_lock);
2860 list_splice_init(&fs_info->delayed_iputs, &list);
2861 spin_unlock(&fs_info->delayed_iput_lock);
2862
2863 while (!list_empty(&list)) {
2864 delayed = list_entry(list.next, struct delayed_iput, list);
2865 list_del(&delayed->list);
2866 iput(delayed->inode);
2867 kfree(delayed);
2868 }
Yan, Zheng24bbcf042009-11-12 09:36:34 +00002869}
2870
Yan, Zhengd68fc572010-05-16 10:49:58 -04002871/*
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08002872 * This is called in transaction commit time. If there are no orphan
Yan, Zhengd68fc572010-05-16 10:49:58 -04002873 * files in the subvolume, it removes orphan item and frees block_rsv
2874 * structure.
2875 */
2876void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2877 struct btrfs_root *root)
2878{
Josef Bacik90290e12011-12-02 15:44:12 -05002879 struct btrfs_block_rsv *block_rsv;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002880 int ret;
2881
Josef Bacik8a35d952012-05-23 14:26:42 -04002882 if (atomic_read(&root->orphan_inodes) ||
Yan, Zhengd68fc572010-05-16 10:49:58 -04002883 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2884 return;
2885
Josef Bacik90290e12011-12-02 15:44:12 -05002886 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04002887 if (atomic_read(&root->orphan_inodes)) {
Josef Bacik90290e12011-12-02 15:44:12 -05002888 spin_unlock(&root->orphan_lock);
2889 return;
2890 }
2891
2892 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2893 spin_unlock(&root->orphan_lock);
2894 return;
2895 }
2896
2897 block_rsv = root->orphan_block_rsv;
2898 root->orphan_block_rsv = NULL;
2899 spin_unlock(&root->orphan_lock);
2900
Yan, Zhengd68fc572010-05-16 10:49:58 -04002901 if (root->orphan_item_inserted &&
2902 btrfs_root_refs(&root->root_item) > 0) {
2903 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2904 root->root_key.objectid);
Josef Bacik4ef31a42013-08-13 14:10:08 -04002905 if (ret)
2906 btrfs_abort_transaction(trans, root, ret);
2907 else
2908 root->orphan_item_inserted = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002909 }
2910
Josef Bacik90290e12011-12-02 15:44:12 -05002911 if (block_rsv) {
2912 WARN_ON(block_rsv->size > 0);
2913 btrfs_free_block_rsv(root, block_rsv);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002914 }
2915}
2916
2917/*
Josef Bacik7b128762008-07-24 12:17:14 -04002918 * This creates an orphan entry for the given inode in case something goes
2919 * wrong in the middle of an unlink/truncate.
Yan, Zhengd68fc572010-05-16 10:49:58 -04002920 *
2921 * NOTE: caller of this function should reserve 5 units of metadata for
2922 * this function.
Josef Bacik7b128762008-07-24 12:17:14 -04002923 */
2924int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2925{
2926 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002927 struct btrfs_block_rsv *block_rsv = NULL;
2928 int reserve = 0;
2929 int insert = 0;
2930 int ret;
Josef Bacik7b128762008-07-24 12:17:14 -04002931
Yan, Zhengd68fc572010-05-16 10:49:58 -04002932 if (!root->orphan_block_rsv) {
Miao Xie66d8f3d2012-09-06 04:02:28 -06002933 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Tsutomu Itohb5324022011-07-19 07:27:20 +00002934 if (!block_rsv)
2935 return -ENOMEM;
Josef Bacik7b128762008-07-24 12:17:14 -04002936 }
2937
Yan, Zhengd68fc572010-05-16 10:49:58 -04002938 spin_lock(&root->orphan_lock);
2939 if (!root->orphan_block_rsv) {
2940 root->orphan_block_rsv = block_rsv;
2941 } else if (block_rsv) {
2942 btrfs_free_block_rsv(root, block_rsv);
2943 block_rsv = NULL;
2944 }
Josef Bacik7b128762008-07-24 12:17:14 -04002945
Josef Bacik8a35d952012-05-23 14:26:42 -04002946 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2947 &BTRFS_I(inode)->runtime_flags)) {
Yan, Zhengd68fc572010-05-16 10:49:58 -04002948#if 0
2949 /*
2950 * For proper ENOSPC handling, we should do orphan
2951 * cleanup when mounting. But this introduces backward
2952 * compatibility issue.
2953 */
2954 if (!xchg(&root->orphan_item_inserted, 1))
2955 insert = 2;
2956 else
2957 insert = 1;
2958#endif
2959 insert = 1;
Miao Xie321f0e72012-08-28 22:13:02 -06002960 atomic_inc(&root->orphan_inodes);
Yan, Zhengd68fc572010-05-16 10:49:58 -04002961 }
Josef Bacik7b128762008-07-24 12:17:14 -04002962
Josef Bacik72ac3c02012-05-23 14:13:11 -04002963 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2964 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04002965 reserve = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002966 spin_unlock(&root->orphan_lock);
Josef Bacik7b128762008-07-24 12:17:14 -04002967
Yan, Zhengd68fc572010-05-16 10:49:58 -04002968 /* grab metadata reservation from transaction handle */
2969 if (reserve) {
2970 ret = btrfs_orphan_reserve_metadata(trans, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002971 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
Yan, Zhengd68fc572010-05-16 10:49:58 -04002972 }
2973
2974 /* insert an orphan item to track this unlinked/truncated file */
2975 if (insert >= 1) {
Li Zefan33345d012011-04-20 10:31:50 +08002976 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
Josef Bacik4ef31a42013-08-13 14:10:08 -04002977 if (ret) {
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01002978 atomic_dec(&root->orphan_inodes);
Josef Bacik4ef31a42013-08-13 14:10:08 -04002979 if (reserve) {
2980 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2981 &BTRFS_I(inode)->runtime_flags);
2982 btrfs_orphan_release_metadata(inode);
2983 }
2984 if (ret != -EEXIST) {
Josef Bacike8e7cff2013-08-21 15:54:00 -04002985 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2986 &BTRFS_I(inode)->runtime_flags);
Josef Bacik4ef31a42013-08-13 14:10:08 -04002987 btrfs_abort_transaction(trans, root, ret);
2988 return ret;
2989 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002990 }
2991 ret = 0;
Yan, Zhengd68fc572010-05-16 10:49:58 -04002992 }
2993
2994 /* insert an orphan item to track subvolume contains orphan files */
2995 if (insert >= 2) {
2996 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2997 root->root_key.objectid);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002998 if (ret && ret != -EEXIST) {
2999 btrfs_abort_transaction(trans, root, ret);
3000 return ret;
3001 }
Yan, Zhengd68fc572010-05-16 10:49:58 -04003002 }
3003 return 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003004}
3005
3006/*
3007 * We have done the truncate/delete so we can go ahead and remove the orphan
3008 * item for this particular inode.
3009 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00003010static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3011 struct inode *inode)
Josef Bacik7b128762008-07-24 12:17:14 -04003012{
3013 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003014 int delete_item = 0;
3015 int release_rsv = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003016 int ret = 0;
3017
Yan, Zhengd68fc572010-05-16 10:49:58 -04003018 spin_lock(&root->orphan_lock);
Josef Bacik8a35d952012-05-23 14:26:42 -04003019 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3020 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003021 delete_item = 1;
Josef Bacik7b128762008-07-24 12:17:14 -04003022
Josef Bacik72ac3c02012-05-23 14:13:11 -04003023 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3024 &BTRFS_I(inode)->runtime_flags))
Yan, Zhengd68fc572010-05-16 10:49:58 -04003025 release_rsv = 1;
Yan, Zhengd68fc572010-05-16 10:49:58 -04003026 spin_unlock(&root->orphan_lock);
3027
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003028 if (delete_item) {
Josef Bacik8a35d952012-05-23 14:26:42 -04003029 atomic_dec(&root->orphan_inodes);
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003030 if (trans)
3031 ret = btrfs_del_orphan_item(trans, root,
3032 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04003033 }
Josef Bacik7b128762008-07-24 12:17:14 -04003034
Filipe David Borba Manana703c88e2013-09-22 21:54:55 +01003035 if (release_rsv)
3036 btrfs_orphan_release_metadata(inode);
3037
Josef Bacik4ef31a42013-08-13 14:10:08 -04003038 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003039}
3040
3041/*
3042 * this cleans up any orphans that may be left on the list from the last use
3043 * of this root.
3044 */
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003045int btrfs_orphan_cleanup(struct btrfs_root *root)
Josef Bacik7b128762008-07-24 12:17:14 -04003046{
3047 struct btrfs_path *path;
3048 struct extent_buffer *leaf;
Josef Bacik7b128762008-07-24 12:17:14 -04003049 struct btrfs_key key, found_key;
3050 struct btrfs_trans_handle *trans;
3051 struct inode *inode;
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003052 u64 last_objectid = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003053 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3054
Yan, Zhengd68fc572010-05-16 10:49:58 -04003055 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003056 return 0;
Yan, Zhengc71bf092009-11-12 09:34:40 +00003057
3058 path = btrfs_alloc_path();
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003059 if (!path) {
3060 ret = -ENOMEM;
3061 goto out;
3062 }
Josef Bacik7b128762008-07-24 12:17:14 -04003063 path->reada = -1;
3064
3065 key.objectid = BTRFS_ORPHAN_OBJECTID;
3066 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3067 key.offset = (u64)-1;
3068
Josef Bacik7b128762008-07-24 12:17:14 -04003069 while (1) {
3070 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003071 if (ret < 0)
3072 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003073
3074 /*
3075 * if ret == 0 means we found what we were searching for, which
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003076 * is weird, but possible, so only screw with path if we didn't
Josef Bacik7b128762008-07-24 12:17:14 -04003077 * find the key and see if we have stuff that matches
3078 */
3079 if (ret > 0) {
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003080 ret = 0;
Josef Bacik7b128762008-07-24 12:17:14 -04003081 if (path->slots[0] == 0)
3082 break;
3083 path->slots[0]--;
3084 }
3085
3086 /* pull out the item */
3087 leaf = path->nodes[0];
Josef Bacik7b128762008-07-24 12:17:14 -04003088 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3089
3090 /* make sure the item matches what we want */
3091 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3092 break;
3093 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3094 break;
3095
3096 /* release the path since we're done with it */
David Sterbab3b4aa72011-04-21 01:20:15 +02003097 btrfs_release_path(path);
Josef Bacik7b128762008-07-24 12:17:14 -04003098
3099 /*
3100 * this is where we are basically btrfs_lookup, without the
3101 * crossing root thing. we store the inode number in the
3102 * offset of the orphan item.
3103 */
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003104
3105 if (found_key.offset == last_objectid) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003106 btrfs_err(root->fs_info,
3107 "Error removing orphan entry, stopping orphan cleanup");
Josef Bacik8f6d7f42011-09-26 15:55:20 -04003108 ret = -EINVAL;
3109 goto out;
3110 }
3111
3112 last_objectid = found_key.offset;
3113
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003114 found_key.objectid = found_key.offset;
3115 found_key.type = BTRFS_INODE_ITEM_KEY;
3116 found_key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +00003117 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
Rusty Russell8c6ffba2013-07-15 11:20:32 +09303118 ret = PTR_ERR_OR_ZERO(inode);
Josef Bacika8c9e572011-09-21 16:55:59 -04003119 if (ret && ret != -ESTALE)
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003120 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003121
Arne Jansenf8e9e0b2011-12-14 20:12:02 -05003122 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3123 struct btrfs_root *dead_root;
3124 struct btrfs_fs_info *fs_info = root->fs_info;
3125 int is_dead_root = 0;
3126
3127 /*
3128 * this is an orphan in the tree root. Currently these
3129 * could come from 2 sources:
3130 * a) a snapshot deletion in progress
3131 * b) a free space cache inode
3132 * We need to distinguish those two, as the snapshot
3133 * orphan must not get deleted.
3134 * find_dead_roots already ran before us, so if this
3135 * is a snapshot deletion, we should find the root
3136 * in the dead_roots list
3137 */
3138 spin_lock(&fs_info->trans_lock);
3139 list_for_each_entry(dead_root, &fs_info->dead_roots,
3140 root_list) {
3141 if (dead_root->root_key.objectid ==
3142 found_key.objectid) {
3143 is_dead_root = 1;
3144 break;
3145 }
3146 }
3147 spin_unlock(&fs_info->trans_lock);
3148 if (is_dead_root) {
3149 /* prevent this orphan from being found again */
3150 key.offset = found_key.objectid - 1;
3151 continue;
3152 }
3153 }
Josef Bacika8c9e572011-09-21 16:55:59 -04003154 /*
3155 * Inode is already gone but the orphan item is still there,
3156 * kill the orphan item.
3157 */
3158 if (ret == -ESTALE) {
3159 trans = btrfs_start_transaction(root, 1);
3160 if (IS_ERR(trans)) {
3161 ret = PTR_ERR(trans);
3162 goto out;
3163 }
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003164 btrfs_debug(root->fs_info, "auto deleting %Lu",
3165 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003166 ret = btrfs_del_orphan_item(trans, root,
3167 found_key.objectid);
Josef Bacika8c9e572011-09-21 16:55:59 -04003168 btrfs_end_transaction(trans, root);
Josef Bacik4ef31a42013-08-13 14:10:08 -04003169 if (ret)
3170 goto out;
Josef Bacika8c9e572011-09-21 16:55:59 -04003171 continue;
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003172 }
Josef Bacik7b128762008-07-24 12:17:14 -04003173
Josef Bacik7b128762008-07-24 12:17:14 -04003174 /*
3175 * add this inode to the orphan list so btrfs_orphan_del does
3176 * the proper thing when we hit it
3177 */
Josef Bacik8a35d952012-05-23 14:26:42 -04003178 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3179 &BTRFS_I(inode)->runtime_flags);
Josef Bacik925396e2013-02-01 15:57:47 -05003180 atomic_inc(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04003181
Josef Bacik7b128762008-07-24 12:17:14 -04003182 /* if we have links, this was a truncate, lets do that */
3183 if (inode->i_nlink) {
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303184 if (WARN_ON(!S_ISREG(inode->i_mode))) {
Josef Bacika41ad392011-01-31 15:30:16 -05003185 iput(inode);
3186 continue;
3187 }
Josef Bacik7b128762008-07-24 12:17:14 -04003188 nr_truncate++;
Josef Bacikf3fe8202013-01-07 17:03:21 -05003189
3190 /* 1 for the orphan item deletion. */
3191 trans = btrfs_start_transaction(root, 1);
3192 if (IS_ERR(trans)) {
Josef Bacikc69b26b2013-06-03 16:51:23 -04003193 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003194 ret = PTR_ERR(trans);
3195 goto out;
3196 }
3197 ret = btrfs_orphan_add(trans, inode);
3198 btrfs_end_transaction(trans, root);
Josef Bacikc69b26b2013-06-03 16:51:23 -04003199 if (ret) {
3200 iput(inode);
Josef Bacikf3fe8202013-01-07 17:03:21 -05003201 goto out;
Josef Bacikc69b26b2013-06-03 16:51:23 -04003202 }
Josef Bacikf3fe8202013-01-07 17:03:21 -05003203
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003204 ret = btrfs_truncate(inode);
Josef Bacik4a7d0f62013-02-07 16:27:28 -05003205 if (ret)
3206 btrfs_orphan_del(NULL, inode);
Josef Bacik7b128762008-07-24 12:17:14 -04003207 } else {
3208 nr_unlink++;
3209 }
3210
3211 /* this will do delete_inode and everything for us */
3212 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003213 if (ret)
3214 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003215 }
Miao Xie3254c872011-11-10 20:45:05 -05003216 /* release the path since we're done with it */
3217 btrfs_release_path(path);
3218
Yan, Zhengd68fc572010-05-16 10:49:58 -04003219 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3220
3221 if (root->orphan_block_rsv)
3222 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3223 (u64)-1);
3224
3225 if (root->orphan_block_rsv || root->orphan_item_inserted) {
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003226 trans = btrfs_join_transaction(root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003227 if (!IS_ERR(trans))
3228 btrfs_end_transaction(trans, root);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003229 }
Josef Bacik7b128762008-07-24 12:17:14 -04003230
3231 if (nr_unlink)
David Sterba4884b472013-03-20 13:31:27 +00003232 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
Josef Bacik7b128762008-07-24 12:17:14 -04003233 if (nr_truncate)
David Sterba4884b472013-03-20 13:31:27 +00003234 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003235
3236out:
3237 if (ret)
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003238 btrfs_crit(root->fs_info,
3239 "could not do orphan cleanup %d", ret);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05003240 btrfs_free_path(path);
3241 return ret;
Josef Bacik7b128762008-07-24 12:17:14 -04003242}
3243
Chris Masond352ac62008-09-29 15:18:18 -04003244/*
Chris Mason46a53cc2009-04-27 11:47:50 -04003245 * very simple check to peek ahead in the leaf looking for xattrs. If we
3246 * don't find any xattrs, we know there can't be any acls.
3247 *
3248 * slot is the slot the inode is in, objectid is the objectid of the inode
3249 */
3250static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3251 int slot, u64 objectid)
3252{
3253 u32 nritems = btrfs_header_nritems(leaf);
3254 struct btrfs_key found_key;
Josef Bacikf23b5a52013-06-19 10:16:26 -04003255 static u64 xattr_access = 0;
3256 static u64 xattr_default = 0;
Chris Mason46a53cc2009-04-27 11:47:50 -04003257 int scanned = 0;
3258
Josef Bacikf23b5a52013-06-19 10:16:26 -04003259 if (!xattr_access) {
3260 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3261 strlen(POSIX_ACL_XATTR_ACCESS));
3262 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3263 strlen(POSIX_ACL_XATTR_DEFAULT));
3264 }
3265
Chris Mason46a53cc2009-04-27 11:47:50 -04003266 slot++;
3267 while (slot < nritems) {
3268 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3269
3270 /* we found a different objectid, there must not be acls */
3271 if (found_key.objectid != objectid)
3272 return 0;
3273
3274 /* we found an xattr, assume we've got an acl */
Josef Bacikf23b5a52013-06-19 10:16:26 -04003275 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3276 if (found_key.offset == xattr_access ||
3277 found_key.offset == xattr_default)
3278 return 1;
3279 }
Chris Mason46a53cc2009-04-27 11:47:50 -04003280
3281 /*
3282 * we found a key greater than an xattr key, there can't
3283 * be any acls later on
3284 */
3285 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3286 return 0;
3287
3288 slot++;
3289 scanned++;
3290
3291 /*
3292 * it goes inode, inode backrefs, xattrs, extents,
3293 * so if there are a ton of hard links to an inode there can
3294 * be a lot of backrefs. Don't waste time searching too hard,
3295 * this is just an optimization
3296 */
3297 if (scanned >= 8)
3298 break;
3299 }
3300 /* we hit the end of the leaf before we found an xattr or
3301 * something larger than an xattr. We have to assume the inode
3302 * has acls
3303 */
3304 return 1;
3305}
3306
3307/*
Chris Masond352ac62008-09-29 15:18:18 -04003308 * read an inode from the btree into the in-memory inode
3309 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003310static void btrfs_read_locked_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003311{
3312 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003313 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003314 struct btrfs_inode_item *inode_item;
Chris Mason0b86a832008-03-24 15:01:56 -04003315 struct btrfs_timespec *tspec;
Chris Mason39279cc2007-06-12 06:35:45 -04003316 struct btrfs_root *root = BTRFS_I(inode)->root;
3317 struct btrfs_key location;
Chris Mason46a53cc2009-04-27 11:47:50 -04003318 int maybe_acls;
Josef Bacik618e21d2007-07-11 10:18:17 -04003319 u32 rdev;
Chris Mason39279cc2007-06-12 06:35:45 -04003320 int ret;
Miao Xie2f7e33d2011-06-23 07:27:13 +00003321 bool filled = false;
3322
3323 ret = btrfs_fill_inode(inode, &rdev);
3324 if (!ret)
3325 filled = true;
Chris Mason39279cc2007-06-12 06:35:45 -04003326
3327 path = btrfs_alloc_path();
Mark Fasheh1748f842011-07-12 11:25:31 -07003328 if (!path)
3329 goto make_bad;
3330
Josef Bacikd90c7322011-05-17 09:50:54 -04003331 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04003332 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
Chris Masondc17ff82008-01-08 15:46:30 -05003333
Chris Mason39279cc2007-06-12 06:35:45 -04003334 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04003335 if (ret)
Chris Mason39279cc2007-06-12 06:35:45 -04003336 goto make_bad;
Chris Mason39279cc2007-06-12 06:35:45 -04003337
Chris Mason5f39d392007-10-15 16:14:19 -04003338 leaf = path->nodes[0];
Miao Xie2f7e33d2011-06-23 07:27:13 +00003339
3340 if (filled)
3341 goto cache_acl;
3342
Chris Mason5f39d392007-10-15 16:14:19 -04003343 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3344 struct btrfs_inode_item);
Chris Mason5f39d392007-10-15 16:14:19 -04003345 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
Miklos Szeredibfe86842011-10-28 14:13:29 +02003346 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
Eric W. Biederman2f2f43d2012-02-10 11:05:07 -08003347 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3348 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
Chris Masondbe674a2008-07-17 12:54:05 -04003349 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
Chris Mason5f39d392007-10-15 16:14:19 -04003350
3351 tspec = btrfs_inode_atime(inode_item);
3352 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3353 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3354
3355 tspec = btrfs_inode_mtime(inode_item);
3356 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3357 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3358
3359 tspec = btrfs_inode_ctime(inode_item);
3360 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3361 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3362
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003363 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04003364 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
Josef Bacik5dc562c2012-08-17 13:14:17 -04003365 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3366
3367 /*
3368 * If we were modified in the current generation and evicted from memory
3369 * and then re-read we need to do a full sync since we don't have any
3370 * idea about which extents were modified before we were evicted from
3371 * cache.
3372 */
3373 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3374 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3375 &BTRFS_I(inode)->runtime_flags);
3376
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003377 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
Chris Masone02119d2008-09-05 16:13:11 -04003378 inode->i_generation = BTRFS_I(inode)->generation;
Josef Bacik618e21d2007-07-11 10:18:17 -04003379 inode->i_rdev = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003380 rdev = btrfs_inode_rdev(leaf, inode_item);
3381
Josef Bacikaec74772008-07-24 12:12:38 -04003382 BTRFS_I(inode)->index_cnt = (u64)-1;
Yan Zhengd2fb3432008-12-11 16:30:39 -05003383 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
Miao Xie2f7e33d2011-06-23 07:27:13 +00003384cache_acl:
Chris Mason46a53cc2009-04-27 11:47:50 -04003385 /*
3386 * try to precache a NULL acl entry for files that don't have
3387 * any xattrs or acls
3388 */
Li Zefan33345d012011-04-20 10:31:50 +08003389 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3390 btrfs_ino(inode));
Al Viro72c04902009-06-24 16:58:48 -04003391 if (!maybe_acls)
3392 cache_no_acl(inode);
Chris Mason46a53cc2009-04-27 11:47:50 -04003393
Chris Mason39279cc2007-06-12 06:35:45 -04003394 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003395
Chris Mason39279cc2007-06-12 06:35:45 -04003396 switch (inode->i_mode & S_IFMT) {
Chris Mason39279cc2007-06-12 06:35:45 -04003397 case S_IFREG:
3398 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04003399 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05003400 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04003401 inode->i_fop = &btrfs_file_operations;
3402 inode->i_op = &btrfs_file_inode_operations;
3403 break;
3404 case S_IFDIR:
3405 inode->i_fop = &btrfs_dir_file_operations;
3406 if (root == root->fs_info->tree_root)
3407 inode->i_op = &btrfs_dir_ro_inode_operations;
3408 else
3409 inode->i_op = &btrfs_dir_inode_operations;
3410 break;
3411 case S_IFLNK:
3412 inode->i_op = &btrfs_symlink_inode_operations;
3413 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04003414 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Mason39279cc2007-06-12 06:35:45 -04003415 break;
Josef Bacik618e21d2007-07-11 10:18:17 -04003416 default:
Jim Owens0279b4c2009-02-04 09:29:13 -05003417 inode->i_op = &btrfs_special_inode_operations;
Josef Bacik618e21d2007-07-11 10:18:17 -04003418 init_special_inode(inode, inode->i_mode, rdev);
3419 break;
Chris Mason39279cc2007-06-12 06:35:45 -04003420 }
Christoph Hellwig6cbff002009-04-17 10:37:41 +02003421
3422 btrfs_update_iflags(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003423 return;
3424
3425make_bad:
Chris Mason39279cc2007-06-12 06:35:45 -04003426 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003427 make_bad_inode(inode);
3428}
3429
Chris Masond352ac62008-09-29 15:18:18 -04003430/*
3431 * given a leaf and an inode, copy the inode fields into the leaf
3432 */
Chris Masone02119d2008-09-05 16:13:11 -04003433static void fill_inode_item(struct btrfs_trans_handle *trans,
3434 struct extent_buffer *leaf,
Chris Mason5f39d392007-10-15 16:14:19 -04003435 struct btrfs_inode_item *item,
Chris Mason39279cc2007-06-12 06:35:45 -04003436 struct inode *inode)
3437{
Liu Bo51fab692012-12-27 09:01:21 +00003438 struct btrfs_map_token token;
Chris Mason5f39d392007-10-15 16:14:19 -04003439
Liu Bo51fab692012-12-27 09:01:21 +00003440 btrfs_init_map_token(&token);
Chris Mason5f39d392007-10-15 16:14:19 -04003441
Liu Bo51fab692012-12-27 09:01:21 +00003442 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3443 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3444 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3445 &token);
3446 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3447 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003448
Liu Bo51fab692012-12-27 09:01:21 +00003449 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3450 inode->i_atime.tv_sec, &token);
3451 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3452 inode->i_atime.tv_nsec, &token);
Chris Mason5f39d392007-10-15 16:14:19 -04003453
Liu Bo51fab692012-12-27 09:01:21 +00003454 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3455 inode->i_mtime.tv_sec, &token);
3456 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3457 inode->i_mtime.tv_nsec, &token);
3458
3459 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3460 inode->i_ctime.tv_sec, &token);
3461 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3462 inode->i_ctime.tv_nsec, &token);
3463
3464 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3465 &token);
3466 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3467 &token);
3468 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3469 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3470 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3471 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3472 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
Chris Mason39279cc2007-06-12 06:35:45 -04003473}
3474
Chris Masond352ac62008-09-29 15:18:18 -04003475/*
3476 * copy everything in the in-memory inode into the btree.
3477 */
Chris Mason21151332011-11-10 20:39:08 -05003478static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
Chris Masond3977122009-01-05 21:25:51 -05003479 struct btrfs_root *root, struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04003480{
3481 struct btrfs_inode_item *inode_item;
3482 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003483 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003484 int ret;
3485
3486 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08003487 if (!path)
3488 return -ENOMEM;
3489
Chris Masonb9473432009-03-13 11:00:37 -04003490 path->leave_spinning = 1;
Miao Xie16cdcec2011-04-22 18:12:22 +08003491 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3492 1);
Chris Mason39279cc2007-06-12 06:35:45 -04003493 if (ret) {
3494 if (ret > 0)
3495 ret = -ENOENT;
3496 goto failed;
3497 }
3498
Chris Masonb4ce94d2009-02-04 09:25:08 -05003499 btrfs_unlock_up_safe(path, 1);
Chris Mason5f39d392007-10-15 16:14:19 -04003500 leaf = path->nodes[0];
3501 inode_item = btrfs_item_ptr(leaf, path->slots[0],
Miao Xie16cdcec2011-04-22 18:12:22 +08003502 struct btrfs_inode_item);
Chris Mason39279cc2007-06-12 06:35:45 -04003503
Chris Masone02119d2008-09-05 16:13:11 -04003504 fill_inode_item(trans, leaf, inode_item, inode);
Chris Mason5f39d392007-10-15 16:14:19 -04003505 btrfs_mark_buffer_dirty(leaf);
Josef Bacik15ee9bc2007-08-10 16:22:09 -04003506 btrfs_set_inode_last_trans(trans, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04003507 ret = 0;
3508failed:
Chris Mason39279cc2007-06-12 06:35:45 -04003509 btrfs_free_path(path);
3510 return ret;
3511}
3512
Chris Masond352ac62008-09-29 15:18:18 -04003513/*
Chris Mason21151332011-11-10 20:39:08 -05003514 * copy everything in the in-memory inode into the btree.
3515 */
3516noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3517 struct btrfs_root *root, struct inode *inode)
3518{
3519 int ret;
3520
3521 /*
3522 * If the inode is a free space inode, we can deadlock during commit
3523 * if we put it into the delayed code.
3524 *
3525 * The data relocation inode should also be directly updated
3526 * without delay
3527 */
Liu Bo83eea1f2012-07-10 05:28:39 -06003528 if (!btrfs_is_free_space_inode(inode)
Chris Mason21151332011-11-10 20:39:08 -05003529 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
Alexander Block8ea05e32012-07-25 17:35:53 +02003530 btrfs_update_root_times(trans, root);
3531
Chris Mason21151332011-11-10 20:39:08 -05003532 ret = btrfs_delayed_update_inode(trans, root, inode);
3533 if (!ret)
3534 btrfs_set_inode_last_trans(trans, inode);
3535 return ret;
3536 }
3537
3538 return btrfs_update_inode_item(trans, root, inode);
3539}
3540
Josef Bacikbe6aef62012-10-22 15:43:12 -04003541noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3542 struct btrfs_root *root,
3543 struct inode *inode)
Chris Mason21151332011-11-10 20:39:08 -05003544{
3545 int ret;
3546
3547 ret = btrfs_update_inode(trans, root, inode);
3548 if (ret == -ENOSPC)
3549 return btrfs_update_inode_item(trans, root, inode);
3550 return ret;
3551}
3552
3553/*
Chris Masond352ac62008-09-29 15:18:18 -04003554 * unlink helper that gets used here in inode.c and in the tree logging
3555 * recovery code. It remove a link in a directory with a given name, and
3556 * also drops the back refs in the inode to the directory
3557 */
Al Viro92986792011-03-04 17:14:37 +00003558static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3559 struct btrfs_root *root,
3560 struct inode *dir, struct inode *inode,
3561 const char *name, int name_len)
Chris Mason39279cc2007-06-12 06:35:45 -04003562{
3563 struct btrfs_path *path;
Chris Mason39279cc2007-06-12 06:35:45 -04003564 int ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04003565 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003566 struct btrfs_dir_item *di;
Chris Mason5f39d392007-10-15 16:14:19 -04003567 struct btrfs_key key;
Josef Bacikaec74772008-07-24 12:12:38 -04003568 u64 index;
Li Zefan33345d012011-04-20 10:31:50 +08003569 u64 ino = btrfs_ino(inode);
3570 u64 dir_ino = btrfs_ino(dir);
Chris Mason39279cc2007-06-12 06:35:45 -04003571
3572 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04003573 if (!path) {
3574 ret = -ENOMEM;
Tsutomu Itoh554233a2011-02-03 03:16:25 +00003575 goto out;
Chris Mason54aa1f42007-06-22 14:16:25 -04003576 }
3577
Chris Masonb9473432009-03-13 11:00:37 -04003578 path->leave_spinning = 1;
Li Zefan33345d012011-04-20 10:31:50 +08003579 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Chris Mason39279cc2007-06-12 06:35:45 -04003580 name, name_len, -1);
3581 if (IS_ERR(di)) {
3582 ret = PTR_ERR(di);
3583 goto err;
3584 }
3585 if (!di) {
3586 ret = -ENOENT;
3587 goto err;
3588 }
Chris Mason5f39d392007-10-15 16:14:19 -04003589 leaf = path->nodes[0];
3590 btrfs_dir_item_key_to_cpu(leaf, di, &key);
Chris Mason39279cc2007-06-12 06:35:45 -04003591 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Chris Mason54aa1f42007-06-22 14:16:25 -04003592 if (ret)
3593 goto err;
David Sterbab3b4aa72011-04-21 01:20:15 +02003594 btrfs_release_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04003595
Li Zefan33345d012011-04-20 10:31:50 +08003596 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3597 dir_ino, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04003598 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003599 btrfs_info(root->fs_info,
3600 "failed to delete reference to %.*s, inode %llu parent %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003601 name_len, name, ino, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003602 btrfs_abort_transaction(trans, root, ret);
Josef Bacikaec74772008-07-24 12:12:38 -04003603 goto err;
3604 }
3605
Miao Xie16cdcec2011-04-22 18:12:22 +08003606 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003607 if (ret) {
3608 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04003609 goto err;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003610 }
Chris Mason39279cc2007-06-12 06:35:45 -04003611
Chris Masone02119d2008-09-05 16:13:11 -04003612 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
Li Zefan33345d012011-04-20 10:31:50 +08003613 inode, dir_ino);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003614 if (ret != 0 && ret != -ENOENT) {
3615 btrfs_abort_transaction(trans, root, ret);
3616 goto err;
3617 }
Chris Masone02119d2008-09-05 16:13:11 -04003618
3619 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3620 dir, index);
Chris Mason6418c962010-10-30 07:34:24 -04003621 if (ret == -ENOENT)
3622 ret = 0;
Zach Brownd4e39912013-04-02 21:02:16 +00003623 else if (ret)
3624 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04003625err:
3626 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003627 if (ret)
3628 goto out;
3629
3630 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003631 inode_inc_iversion(inode);
3632 inode_inc_iversion(dir);
Chris Masone02119d2008-09-05 16:13:11 -04003633 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
Tsutomu Itohb9959292012-06-25 21:25:22 -06003634 ret = btrfs_update_inode(trans, root, dir);
Chris Masone02119d2008-09-05 16:13:11 -04003635out:
Chris Mason39279cc2007-06-12 06:35:45 -04003636 return ret;
3637}
3638
Al Viro92986792011-03-04 17:14:37 +00003639int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3640 struct btrfs_root *root,
3641 struct inode *dir, struct inode *inode,
3642 const char *name, int name_len)
3643{
3644 int ret;
3645 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3646 if (!ret) {
Zach Brown8b558c52013-10-16 12:10:34 -07003647 drop_nlink(inode);
Al Viro92986792011-03-04 17:14:37 +00003648 ret = btrfs_update_inode(trans, root, inode);
3649 }
3650 return ret;
3651}
Yan, Zhenga22285a2010-05-16 10:48:46 -04003652
3653/*
3654 * helper to start transaction for unlink and rmdir.
3655 *
Josef Bacikd52be812013-05-29 14:54:47 -04003656 * unlink and rmdir are special in btrfs, they do not always free space, so
3657 * if we cannot make our reservations the normal way try and see if there is
3658 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3659 * allow the unlink to occur.
Yan, Zhenga22285a2010-05-16 10:48:46 -04003660 */
Josef Bacikd52be812013-05-29 14:54:47 -04003661static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
Yan, Zhenga22285a2010-05-16 10:48:46 -04003662{
3663 struct btrfs_trans_handle *trans;
3664 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003665 int ret;
3666
Josef Bacike70bea52011-10-11 14:18:24 -04003667 /*
3668 * 1 for the possible orphan item
3669 * 1 for the dir item
3670 * 1 for the dir index
3671 * 1 for the inode ref
Josef Bacike70bea52011-10-11 14:18:24 -04003672 * 1 for the inode
3673 */
Josef Bacik6e137ed2013-03-26 15:26:55 -04003674 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003675 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3676 return trans;
3677
Josef Bacikd52be812013-05-29 14:54:47 -04003678 if (PTR_ERR(trans) == -ENOSPC) {
3679 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003680
Josef Bacikd52be812013-05-29 14:54:47 -04003681 trans = btrfs_start_transaction(root, 0);
3682 if (IS_ERR(trans))
3683 return trans;
3684 ret = btrfs_cond_migrate_bytes(root->fs_info,
3685 &root->fs_info->trans_block_rsv,
3686 num_bytes, 5);
3687 if (ret) {
3688 btrfs_end_transaction(trans, root);
3689 return ERR_PTR(ret);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003690 }
Josef Bacik5a77d762011-11-01 14:32:23 -04003691 trans->block_rsv = &root->fs_info->trans_block_rsv;
Josef Bacikd52be812013-05-29 14:54:47 -04003692 trans->bytes_reserved = num_bytes;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003693 }
Josef Bacikd52be812013-05-29 14:54:47 -04003694 return trans;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003695}
3696
Chris Mason39279cc2007-06-12 06:35:45 -04003697static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3698{
Yan, Zhenga22285a2010-05-16 10:48:46 -04003699 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003700 struct btrfs_trans_handle *trans;
Josef Bacik7b128762008-07-24 12:17:14 -04003701 struct inode *inode = dentry->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -04003702 int ret;
3703
Josef Bacikd52be812013-05-29 14:54:47 -04003704 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003705 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003706 return PTR_ERR(trans);
Chris Mason5f39d392007-10-15 16:14:19 -04003707
Chris Mason12fcfd22009-03-24 10:24:20 -04003708 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3709
Chris Masone02119d2008-09-05 16:13:11 -04003710 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3711 dentry->d_name.name, dentry->d_name.len);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003712 if (ret)
3713 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003714
Yan, Zhenga22285a2010-05-16 10:48:46 -04003715 if (inode->i_nlink == 0) {
Josef Bacik7b128762008-07-24 12:17:14 -04003716 ret = btrfs_orphan_add(trans, inode);
Tsutomu Itohb5324022011-07-19 07:27:20 +00003717 if (ret)
3718 goto out;
Yan, Zhenga22285a2010-05-16 10:48:46 -04003719 }
Josef Bacik7b128762008-07-24 12:17:14 -04003720
Tsutomu Itohb5324022011-07-19 07:27:20 +00003721out:
Josef Bacikd52be812013-05-29 14:54:47 -04003722 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003723 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04003724 return ret;
3725}
3726
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003727int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3728 struct btrfs_root *root,
3729 struct inode *dir, u64 objectid,
3730 const char *name, int name_len)
3731{
3732 struct btrfs_path *path;
3733 struct extent_buffer *leaf;
3734 struct btrfs_dir_item *di;
3735 struct btrfs_key key;
3736 u64 index;
3737 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08003738 u64 dir_ino = btrfs_ino(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003739
3740 path = btrfs_alloc_path();
3741 if (!path)
3742 return -ENOMEM;
3743
Li Zefan33345d012011-04-20 10:31:50 +08003744 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003745 name, name_len, -1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003746 if (IS_ERR_OR_NULL(di)) {
3747 if (!di)
3748 ret = -ENOENT;
3749 else
3750 ret = PTR_ERR(di);
3751 goto out;
3752 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003753
3754 leaf = path->nodes[0];
3755 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3756 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3757 ret = btrfs_delete_one_dir_name(trans, root, path, di);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003758 if (ret) {
3759 btrfs_abort_transaction(trans, root, ret);
3760 goto out;
3761 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003762 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003763
3764 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3765 objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08003766 dir_ino, &index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003767 if (ret < 0) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003768 if (ret != -ENOENT) {
3769 btrfs_abort_transaction(trans, root, ret);
3770 goto out;
3771 }
Li Zefan33345d012011-04-20 10:31:50 +08003772 di = btrfs_search_dir_index_item(root, path, dir_ino,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003773 name, name_len);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003774 if (IS_ERR_OR_NULL(di)) {
3775 if (!di)
3776 ret = -ENOENT;
3777 else
3778 ret = PTR_ERR(di);
3779 btrfs_abort_transaction(trans, root, ret);
3780 goto out;
3781 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003782
3783 leaf = path->nodes[0];
3784 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003785 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003786 index = key.offset;
3787 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003788 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003789
Miao Xie16cdcec2011-04-22 18:12:22 +08003790 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003791 if (ret) {
3792 btrfs_abort_transaction(trans, root, ret);
3793 goto out;
3794 }
Miao Xie16cdcec2011-04-22 18:12:22 +08003795
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003796 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04003797 inode_inc_iversion(dir);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003798 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
Josef Bacik5a24e842012-08-08 10:12:59 -06003799 ret = btrfs_update_inode_fallback(trans, root, dir);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003800 if (ret)
3801 btrfs_abort_transaction(trans, root, ret);
3802out:
Josef Bacik71d7aed2011-06-14 14:24:32 -04003803 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003804 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003805}
3806
Chris Mason39279cc2007-06-12 06:35:45 -04003807static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3808{
3809 struct inode *inode = dentry->d_inode;
Chris Mason1832a6d2007-12-21 16:27:21 -05003810 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003811 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04003812 struct btrfs_trans_handle *trans;
Chris Mason39279cc2007-06-12 06:35:45 -04003813
David Sterbab3ae2442012-09-13 16:04:34 -06003814 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Yan134d4512007-10-25 15:49:25 -04003815 return -ENOTEMPTY;
David Sterbab3ae2442012-09-13 16:04:34 -06003816 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3817 return -EPERM;
Yan134d4512007-10-25 15:49:25 -04003818
Josef Bacikd52be812013-05-29 14:54:47 -04003819 trans = __unlink_start_trans(dir);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003820 if (IS_ERR(trans))
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003821 return PTR_ERR(trans);
Josef Bacik5df6a9f2009-11-10 21:23:48 -05003822
Li Zefan33345d012011-04-20 10:31:50 +08003823 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003824 err = btrfs_unlink_subvol(trans, root, dir,
3825 BTRFS_I(inode)->location.objectid,
3826 dentry->d_name.name,
3827 dentry->d_name.len);
3828 goto out;
3829 }
3830
Josef Bacik7b128762008-07-24 12:17:14 -04003831 err = btrfs_orphan_add(trans, inode);
3832 if (err)
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003833 goto out;
Josef Bacik7b128762008-07-24 12:17:14 -04003834
Chris Mason39279cc2007-06-12 06:35:45 -04003835 /* now the directory is empty */
Chris Masone02119d2008-09-05 16:13:11 -04003836 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3837 dentry->d_name.name, dentry->d_name.len);
Chris Masond3977122009-01-05 21:25:51 -05003838 if (!err)
Chris Masondbe674a2008-07-17 12:54:05 -04003839 btrfs_i_size_write(inode, 0);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04003840out:
Josef Bacikd52be812013-05-29 14:54:47 -04003841 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00003842 btrfs_btree_balance_dirty(root);
Chris Mason39544012007-12-12 14:38:19 -05003843
Chris Mason39279cc2007-06-12 06:35:45 -04003844 return err;
3845}
3846
Chris Mason323ac952008-10-01 19:05:46 -04003847/*
Chris Mason39279cc2007-06-12 06:35:45 -04003848 * this can truncate away extent items, csum items and directory items.
3849 * It starts at a high offset and removes keys until it can't find
Chris Masond352ac62008-09-29 15:18:18 -04003850 * any higher than new_size
Chris Mason39279cc2007-06-12 06:35:45 -04003851 *
3852 * csum items that cross the new i_size are truncated to the new size
3853 * as well.
Josef Bacik7b128762008-07-24 12:17:14 -04003854 *
3855 * min_type is the minimum key type to truncate down to. If set to 0, this
3856 * will kill all the items on this inode, including the INODE_ITEM_KEY.
Chris Mason39279cc2007-06-12 06:35:45 -04003857 */
Yan, Zheng80825102009-11-12 09:35:36 +00003858int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3859 struct btrfs_root *root,
3860 struct inode *inode,
3861 u64 new_size, u32 min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003862{
Chris Mason39279cc2007-06-12 06:35:45 -04003863 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -04003864 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04003865 struct btrfs_file_extent_item *fi;
Yan, Zheng80825102009-11-12 09:35:36 +00003866 struct btrfs_key key;
3867 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04003868 u64 extent_start = 0;
Chris Masondb945352007-10-15 16:15:53 -04003869 u64 extent_num_bytes = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003870 u64 extent_offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003871 u64 item_end = 0;
Josef Bacik7f4f6e02013-08-29 16:43:28 -04003872 u64 last_size = (u64)-1;
Yan, Zheng80825102009-11-12 09:35:36 +00003873 u32 found_type = (u8)-1;
Chris Mason39279cc2007-06-12 06:35:45 -04003874 int found_extent;
3875 int del_item;
Chris Mason85e21ba2008-01-29 15:11:36 -05003876 int pending_del_nr = 0;
3877 int pending_del_slot = 0;
Chris Mason179e29e2007-11-01 11:28:41 -04003878 int extent_type = -1;
Yan, Zheng80825102009-11-12 09:35:36 +00003879 int ret;
3880 int err = 0;
Li Zefan33345d012011-04-20 10:31:50 +08003881 u64 ino = btrfs_ino(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00003882
3883 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
Chris Mason39279cc2007-06-12 06:35:45 -04003884
Mark Fasheh0eb0e192011-07-12 16:44:10 -07003885 path = btrfs_alloc_path();
3886 if (!path)
3887 return -ENOMEM;
3888 path->reada = -1;
3889
Josef Bacik5dc562c2012-08-17 13:14:17 -04003890 /*
3891 * We want to drop from the next block forward in case this new size is
3892 * not block aligned since we will be keeping the last block of the
3893 * extent just the way it is.
3894 */
Josef Bacik0af3d002010-06-21 14:48:16 -04003895 if (root->ref_cows || root == root->fs_info->tree_root)
Qu Wenruofda28322013-02-26 08:10:22 +00003896 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3897 root->sectorsize), (u64)-1, 0);
Yan, Zheng80825102009-11-12 09:35:36 +00003898
Miao Xie16cdcec2011-04-22 18:12:22 +08003899 /*
3900 * This function is also used to drop the items in the log tree before
3901 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3902 * it is used to drop the loged items. So we shouldn't kill the delayed
3903 * items.
3904 */
3905 if (min_type == 0 && root == BTRFS_I(inode)->root)
3906 btrfs_kill_delayed_inode_items(inode);
3907
Li Zefan33345d012011-04-20 10:31:50 +08003908 key.objectid = ino;
Chris Mason39279cc2007-06-12 06:35:45 -04003909 key.offset = (u64)-1;
Chris Mason5f39d392007-10-15 16:14:19 -04003910 key.type = (u8)-1;
3911
Chris Mason85e21ba2008-01-29 15:11:36 -05003912search_again:
Chris Masonb9473432009-03-13 11:00:37 -04003913 path->leave_spinning = 1;
Chris Mason85e21ba2008-01-29 15:11:36 -05003914 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Yan, Zheng80825102009-11-12 09:35:36 +00003915 if (ret < 0) {
3916 err = ret;
3917 goto out;
3918 }
Chris Masond3977122009-01-05 21:25:51 -05003919
Chris Mason85e21ba2008-01-29 15:11:36 -05003920 if (ret > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04003921 /* there are no items in the tree for us to truncate, we're
3922 * done
3923 */
Yan, Zheng80825102009-11-12 09:35:36 +00003924 if (path->slots[0] == 0)
3925 goto out;
Chris Mason85e21ba2008-01-29 15:11:36 -05003926 path->slots[0]--;
3927 }
3928
Chris Masond3977122009-01-05 21:25:51 -05003929 while (1) {
Chris Mason39279cc2007-06-12 06:35:45 -04003930 fi = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -04003931 leaf = path->nodes[0];
3932 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3933 found_type = btrfs_key_type(&found_key);
Chris Mason39279cc2007-06-12 06:35:45 -04003934
Li Zefan33345d012011-04-20 10:31:50 +08003935 if (found_key.objectid != ino)
Chris Mason39279cc2007-06-12 06:35:45 -04003936 break;
Chris Mason5f39d392007-10-15 16:14:19 -04003937
Chris Mason85e21ba2008-01-29 15:11:36 -05003938 if (found_type < min_type)
Chris Mason39279cc2007-06-12 06:35:45 -04003939 break;
3940
Chris Mason5f39d392007-10-15 16:14:19 -04003941 item_end = found_key.offset;
Chris Mason39279cc2007-06-12 06:35:45 -04003942 if (found_type == BTRFS_EXTENT_DATA_KEY) {
Chris Mason5f39d392007-10-15 16:14:19 -04003943 fi = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason39279cc2007-06-12 06:35:45 -04003944 struct btrfs_file_extent_item);
Chris Mason179e29e2007-11-01 11:28:41 -04003945 extent_type = btrfs_file_extent_type(leaf, fi);
3946 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04003947 item_end +=
Chris Masondb945352007-10-15 16:15:53 -04003948 btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason179e29e2007-11-01 11:28:41 -04003949 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason179e29e2007-11-01 11:28:41 -04003950 item_end += btrfs_file_extent_inline_len(leaf,
Chris Masonc8b97812008-10-29 14:49:59 -04003951 fi);
Chris Mason39279cc2007-06-12 06:35:45 -04003952 }
Yan008630c2007-11-07 13:31:09 -05003953 item_end--;
Chris Mason39279cc2007-06-12 06:35:45 -04003954 }
Yan, Zheng80825102009-11-12 09:35:36 +00003955 if (found_type > min_type) {
Chris Mason39279cc2007-06-12 06:35:45 -04003956 del_item = 1;
Yan, Zheng80825102009-11-12 09:35:36 +00003957 } else {
3958 if (item_end < new_size)
3959 break;
3960 if (found_key.offset >= new_size)
3961 del_item = 1;
3962 else
3963 del_item = 0;
3964 }
Chris Mason39279cc2007-06-12 06:35:45 -04003965 found_extent = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04003966 /* FIXME, shrink the extent if the ref count is only 1 */
Chris Mason179e29e2007-11-01 11:28:41 -04003967 if (found_type != BTRFS_EXTENT_DATA_KEY)
3968 goto delete;
3969
Josef Bacik7f4f6e02013-08-29 16:43:28 -04003970 if (del_item)
3971 last_size = found_key.offset;
3972 else
3973 last_size = new_size;
3974
Chris Mason179e29e2007-11-01 11:28:41 -04003975 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
Chris Mason39279cc2007-06-12 06:35:45 -04003976 u64 num_dec;
Chris Masondb945352007-10-15 16:15:53 -04003977 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
Miao Xief70a9a6b2012-01-12 19:10:12 -05003978 if (!del_item) {
Chris Masondb945352007-10-15 16:15:53 -04003979 u64 orig_num_bytes =
3980 btrfs_file_extent_num_bytes(leaf, fi);
Qu Wenruofda28322013-02-26 08:10:22 +00003981 extent_num_bytes = ALIGN(new_size -
3982 found_key.offset,
3983 root->sectorsize);
Chris Masondb945352007-10-15 16:15:53 -04003984 btrfs_set_file_extent_num_bytes(leaf, fi,
3985 extent_num_bytes);
3986 num_dec = (orig_num_bytes -
Chris Mason90692182008-02-08 13:49:28 -05003987 extent_num_bytes);
Chris Masone02119d2008-09-05 16:13:11 -04003988 if (root->ref_cows && extent_start != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04003989 inode_sub_bytes(inode, num_dec);
Chris Mason5f39d392007-10-15 16:14:19 -04003990 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04003991 } else {
Chris Masondb945352007-10-15 16:15:53 -04003992 extent_num_bytes =
3993 btrfs_file_extent_disk_num_bytes(leaf,
3994 fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003995 extent_offset = found_key.offset -
3996 btrfs_file_extent_offset(leaf, fi);
3997
Chris Mason39279cc2007-06-12 06:35:45 -04003998 /* FIXME blocksize != 4096 */
Chris Mason90692182008-02-08 13:49:28 -05003999 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason39279cc2007-06-12 06:35:45 -04004000 if (extent_start != 0) {
4001 found_extent = 1;
Chris Masone02119d2008-09-05 16:13:11 -04004002 if (root->ref_cows)
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004003 inode_sub_bytes(inode, num_dec);
Chris Mason39279cc2007-06-12 06:35:45 -04004004 }
4005 }
Chris Mason90692182008-02-08 13:49:28 -05004006 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -04004007 /*
4008 * we can't truncate inline items that have had
4009 * special encodings
4010 */
4011 if (!del_item &&
4012 btrfs_file_extent_compression(leaf, fi) == 0 &&
4013 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4014 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
Chris Masone02119d2008-09-05 16:13:11 -04004015 u32 size = new_size - found_key.offset;
4016
4017 if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004018 inode_sub_bytes(inode, item_end + 1 -
4019 new_size);
Chris Masone02119d2008-09-05 16:13:11 -04004020 }
4021 size =
4022 btrfs_file_extent_calc_inline_size(size);
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00004023 btrfs_truncate_item(root, path, size, 1);
Chris Masone02119d2008-09-05 16:13:11 -04004024 } else if (root->ref_cows) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -04004025 inode_sub_bytes(inode, item_end + 1 -
4026 found_key.offset);
Chris Mason90692182008-02-08 13:49:28 -05004027 }
Chris Mason39279cc2007-06-12 06:35:45 -04004028 }
Chris Mason179e29e2007-11-01 11:28:41 -04004029delete:
Chris Mason39279cc2007-06-12 06:35:45 -04004030 if (del_item) {
Chris Mason85e21ba2008-01-29 15:11:36 -05004031 if (!pending_del_nr) {
4032 /* no pending yet, add ourselves */
4033 pending_del_slot = path->slots[0];
4034 pending_del_nr = 1;
4035 } else if (pending_del_nr &&
4036 path->slots[0] + 1 == pending_del_slot) {
4037 /* hop on the pending chunk */
4038 pending_del_nr++;
4039 pending_del_slot = path->slots[0];
4040 } else {
Chris Masond3977122009-01-05 21:25:51 -05004041 BUG();
Chris Mason85e21ba2008-01-29 15:11:36 -05004042 }
Chris Mason39279cc2007-06-12 06:35:45 -04004043 } else {
4044 break;
4045 }
Josef Bacik0af3d002010-06-21 14:48:16 -04004046 if (found_extent && (root->ref_cows ||
4047 root == root->fs_info->tree_root)) {
Chris Masonb9473432009-03-13 11:00:37 -04004048 btrfs_set_path_blocking(path);
Chris Mason39279cc2007-06-12 06:35:45 -04004049 ret = btrfs_free_extent(trans, root, extent_start,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004050 extent_num_bytes, 0,
4051 btrfs_header_owner(leaf),
Arne Jansen66d7e7f2011-09-12 15:26:38 +02004052 ino, extent_offset, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04004053 BUG_ON(ret);
4054 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004055
Yan, Zheng80825102009-11-12 09:35:36 +00004056 if (found_type == BTRFS_INODE_ITEM_KEY)
4057 break;
4058
4059 if (path->slots[0] == 0 ||
4060 path->slots[0] != pending_del_slot) {
Yan, Zheng80825102009-11-12 09:35:36 +00004061 if (pending_del_nr) {
4062 ret = btrfs_del_items(trans, root, path,
4063 pending_del_slot,
4064 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004065 if (ret) {
4066 btrfs_abort_transaction(trans,
4067 root, ret);
4068 goto error;
4069 }
Yan, Zheng80825102009-11-12 09:35:36 +00004070 pending_del_nr = 0;
4071 }
David Sterbab3b4aa72011-04-21 01:20:15 +02004072 btrfs_release_path(path);
Chris Mason85e21ba2008-01-29 15:11:36 -05004073 goto search_again;
Yan, Zheng80825102009-11-12 09:35:36 +00004074 } else {
4075 path->slots[0]--;
Chris Mason85e21ba2008-01-29 15:11:36 -05004076 }
Chris Mason39279cc2007-06-12 06:35:45 -04004077 }
Yan, Zheng80825102009-11-12 09:35:36 +00004078out:
Chris Mason85e21ba2008-01-29 15:11:36 -05004079 if (pending_del_nr) {
4080 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4081 pending_del_nr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004082 if (ret)
4083 btrfs_abort_transaction(trans, root, ret);
Chris Mason85e21ba2008-01-29 15:11:36 -05004084 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004085error:
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004086 if (last_size != (u64)-1)
4087 btrfs_ordered_update_i_size(inode, last_size, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04004088 btrfs_free_path(path);
Yan, Zheng80825102009-11-12 09:35:36 +00004089 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004090}
4091
Chris Masona52d9a82007-08-27 16:49:44 -04004092/*
Josef Bacik2aaa6652012-08-29 14:27:18 -04004093 * btrfs_truncate_page - read, zero a chunk and write a page
4094 * @inode - inode that we're zeroing
4095 * @from - the offset to start zeroing
4096 * @len - the length to zero, 0 to zero the entire range respective to the
4097 * offset
4098 * @front - zero up to the offset instead of from the offset on
4099 *
4100 * This will find the page for the "from" offset and cow the page and zero the
4101 * part we want to zero. This is used with truncate and hole punching.
Chris Masona52d9a82007-08-27 16:49:44 -04004102 */
Josef Bacik2aaa6652012-08-29 14:27:18 -04004103int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4104 int front)
Chris Masona52d9a82007-08-27 16:49:44 -04004105{
Josef Bacik2aaa6652012-08-29 14:27:18 -04004106 struct address_space *mapping = inode->i_mapping;
Chris Masondb945352007-10-15 16:15:53 -04004107 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004108 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4109 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004110 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004111 char *kaddr;
Chris Masondb945352007-10-15 16:15:53 -04004112 u32 blocksize = root->sectorsize;
Chris Masona52d9a82007-08-27 16:49:44 -04004113 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4114 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4115 struct page *page;
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004116 gfp_t mask = btrfs_alloc_write_mask(mapping);
Chris Masona52d9a82007-08-27 16:49:44 -04004117 int ret = 0;
4118 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004119 u64 page_end;
Chris Masona52d9a82007-08-27 16:49:44 -04004120
Josef Bacik2aaa6652012-08-29 14:27:18 -04004121 if ((offset & (blocksize - 1)) == 0 &&
4122 (!len || ((len & (blocksize - 1)) == 0)))
Chris Masona52d9a82007-08-27 16:49:44 -04004123 goto out;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004124 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004125 if (ret)
4126 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04004127
Chris Mason211c17f2008-05-15 09:13:45 -04004128again:
Josef Bacik3b16a4e2011-09-21 15:05:58 -04004129 page = find_or_create_page(mapping, index, mask);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004130 if (!page) {
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004131 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Miao Xieac6a2b32012-12-05 10:56:13 +00004132 ret = -ENOMEM;
Chris Masona52d9a82007-08-27 16:49:44 -04004133 goto out;
Josef Bacik5d5e1032009-10-13 16:46:49 -04004134 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04004135
4136 page_start = page_offset(page);
4137 page_end = page_start + PAGE_CACHE_SIZE - 1;
4138
Chris Masona52d9a82007-08-27 16:49:44 -04004139 if (!PageUptodate(page)) {
4140 ret = btrfs_readpage(NULL, page);
4141 lock_page(page);
Chris Mason211c17f2008-05-15 09:13:45 -04004142 if (page->mapping != mapping) {
4143 unlock_page(page);
4144 page_cache_release(page);
4145 goto again;
4146 }
Chris Masona52d9a82007-08-27 16:49:44 -04004147 if (!PageUptodate(page)) {
4148 ret = -EIO;
Chris Mason89642222008-07-24 09:41:53 -04004149 goto out_unlock;
Chris Masona52d9a82007-08-27 16:49:44 -04004150 }
4151 }
Chris Mason211c17f2008-05-15 09:13:45 -04004152 wait_on_page_writeback(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004153
Jeff Mahoneyd0082372012-03-01 14:57:19 +01004154 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004155 set_page_extent_mapped(page);
4156
4157 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4158 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00004159 unlock_extent_cached(io_tree, page_start, page_end,
4160 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004161 unlock_page(page);
4162 page_cache_release(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04004163 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004164 btrfs_put_ordered_extent(ordered);
4165 goto again;
4166 }
4167
Josef Bacik2ac55d42010-02-03 19:33:23 +00004168 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06004169 EXTENT_DIRTY | EXTENT_DELALLOC |
4170 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00004171 0, 0, &cached_state, GFP_NOFS);
Josef Bacik5d5e1032009-10-13 16:46:49 -04004172
Josef Bacik2ac55d42010-02-03 19:33:23 +00004173 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4174 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004175 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00004176 unlock_extent_cached(io_tree, page_start, page_end,
4177 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04004178 goto out_unlock;
4179 }
4180
Chris Masone6dcd2d2008-07-17 12:53:50 -04004181 if (offset != PAGE_CACHE_SIZE) {
Josef Bacik2aaa6652012-08-29 14:27:18 -04004182 if (!len)
4183 len = PAGE_CACHE_SIZE - offset;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004184 kaddr = kmap(page);
Josef Bacik2aaa6652012-08-29 14:27:18 -04004185 if (front)
4186 memset(kaddr, 0, offset);
4187 else
4188 memset(kaddr + offset, 0, len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004189 flush_dcache_page(page);
4190 kunmap(page);
4191 }
Chris Mason247e7432008-07-17 12:53:51 -04004192 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004193 set_page_dirty(page);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004194 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4195 GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -04004196
Chris Mason89642222008-07-24 09:41:53 -04004197out_unlock:
Josef Bacik5d5e1032009-10-13 16:46:49 -04004198 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004199 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason39279cc2007-06-12 06:35:45 -04004200 unlock_page(page);
4201 page_cache_release(page);
4202out:
4203 return ret;
4204}
4205
Josef Bacik16e75492013-10-22 12:18:51 -04004206static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4207 u64 offset, u64 len)
4208{
4209 struct btrfs_trans_handle *trans;
4210 int ret;
4211
4212 /*
4213 * Still need to make sure the inode looks like it's been updated so
4214 * that any holes get logged if we fsync.
4215 */
4216 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4217 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4218 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4219 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4220 return 0;
4221 }
4222
4223 /*
4224 * 1 - for the one we're dropping
4225 * 1 - for the one we're adding
4226 * 1 - for updating the inode.
4227 */
4228 trans = btrfs_start_transaction(root, 3);
4229 if (IS_ERR(trans))
4230 return PTR_ERR(trans);
4231
4232 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4233 if (ret) {
4234 btrfs_abort_transaction(trans, root, ret);
4235 btrfs_end_transaction(trans, root);
4236 return ret;
4237 }
4238
4239 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4240 0, 0, len, 0, len, 0, 0, 0);
4241 if (ret)
4242 btrfs_abort_transaction(trans, root, ret);
4243 else
4244 btrfs_update_inode(trans, root, inode);
4245 btrfs_end_transaction(trans, root);
4246 return ret;
4247}
4248
Josef Bacik695a0d02011-03-04 15:46:53 -05004249/*
4250 * This function puts in dummy file extents for the area we're creating a hole
4251 * for. So if we are truncating this file to a larger size we need to insert
4252 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4253 * the range between oldsize and size
4254 */
Josef Bacika41ad392011-01-31 15:30:16 -05004255int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
Yan Zheng9036c102008-10-30 14:19:41 -04004256{
Yan Zheng9036c102008-10-30 14:19:41 -04004257 struct btrfs_root *root = BTRFS_I(inode)->root;
4258 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Yan, Zhenga22285a2010-05-16 10:48:46 -04004259 struct extent_map *em = NULL;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004260 struct extent_state *cached_state = NULL;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004261 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Qu Wenruofda28322013-02-26 08:10:22 +00004262 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4263 u64 block_end = ALIGN(size, root->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04004264 u64 last_byte;
4265 u64 cur_offset;
4266 u64 hole_size;
Josef Bacik9ed74f22009-09-11 16:12:44 -04004267 int err = 0;
Yan Zheng9036c102008-10-30 14:19:41 -04004268
Josef Bacika71754f2013-06-17 17:14:39 -04004269 /*
4270 * If our size started in the middle of a page we need to zero out the
4271 * rest of the page before we expand the i_size, otherwise we could
4272 * expose stale data.
4273 */
4274 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4275 if (err)
4276 return err;
4277
Yan Zheng9036c102008-10-30 14:19:41 -04004278 if (size <= hole_start)
4279 return 0;
4280
Yan Zheng9036c102008-10-30 14:19:41 -04004281 while (1) {
4282 struct btrfs_ordered_extent *ordered;
Miao Xiefa7c1492013-09-26 13:15:27 +08004283
Josef Bacik2ac55d42010-02-03 19:33:23 +00004284 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01004285 &cached_state);
Miao Xiefa7c1492013-09-26 13:15:27 +08004286 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4287 block_end - hole_start);
Yan Zheng9036c102008-10-30 14:19:41 -04004288 if (!ordered)
4289 break;
Josef Bacik2ac55d42010-02-03 19:33:23 +00004290 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4291 &cached_state, GFP_NOFS);
Miao Xiefa7c1492013-09-26 13:15:27 +08004292 btrfs_start_ordered_extent(inode, ordered, 1);
Yan Zheng9036c102008-10-30 14:19:41 -04004293 btrfs_put_ordered_extent(ordered);
4294 }
4295
Yan Zheng9036c102008-10-30 14:19:41 -04004296 cur_offset = hole_start;
4297 while (1) {
4298 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4299 block_end - cur_offset, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004300 if (IS_ERR(em)) {
4301 err = PTR_ERR(em);
Zach Brownf2767952013-01-08 19:37:58 +00004302 em = NULL;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004303 break;
4304 }
Yan Zheng9036c102008-10-30 14:19:41 -04004305 last_byte = min(extent_map_end(em), block_end);
Qu Wenruofda28322013-02-26 08:10:22 +00004306 last_byte = ALIGN(last_byte , root->sectorsize);
Yan, Zheng80825102009-11-12 09:35:36 +00004307 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
Josef Bacik5dc562c2012-08-17 13:14:17 -04004308 struct extent_map *hole_em;
Yan Zheng9036c102008-10-30 14:19:41 -04004309 hole_size = last_byte - cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004310
Josef Bacik16e75492013-10-22 12:18:51 -04004311 err = maybe_insert_hole(root, inode, cur_offset,
4312 hole_size);
4313 if (err)
Chris Mason771ed682008-11-06 22:02:51 -05004314 break;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004315 btrfs_drop_extent_cache(inode, cur_offset,
4316 cur_offset + hole_size - 1, 0);
4317 hole_em = alloc_extent_map();
4318 if (!hole_em) {
4319 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4320 &BTRFS_I(inode)->runtime_flags);
4321 goto next;
4322 }
4323 hole_em->start = cur_offset;
4324 hole_em->len = hole_size;
4325 hole_em->orig_start = cur_offset;
Yan, Zheng80825102009-11-12 09:35:36 +00004326
Josef Bacik5dc562c2012-08-17 13:14:17 -04004327 hole_em->block_start = EXTENT_MAP_HOLE;
4328 hole_em->block_len = 0;
Josef Bacikb4939682012-12-03 10:31:19 -05004329 hole_em->orig_block_len = 0;
Josef Bacikcc95bef2013-04-04 14:31:27 -04004330 hole_em->ram_bytes = hole_size;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004331 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4332 hole_em->compress_type = BTRFS_COMPRESS_NONE;
Josef Bacik16e75492013-10-22 12:18:51 -04004333 hole_em->generation = root->fs_info->generation;
Josef Bacik5dc562c2012-08-17 13:14:17 -04004334
4335 while (1) {
4336 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04004337 err = add_extent_mapping(em_tree, hole_em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04004338 write_unlock(&em_tree->lock);
4339 if (err != -EEXIST)
4340 break;
4341 btrfs_drop_extent_cache(inode, cur_offset,
4342 cur_offset +
4343 hole_size - 1, 0);
4344 }
4345 free_extent_map(hole_em);
Yan Zheng9036c102008-10-30 14:19:41 -04004346 }
Josef Bacik16e75492013-10-22 12:18:51 -04004347next:
Yan Zheng9036c102008-10-30 14:19:41 -04004348 free_extent_map(em);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004349 em = NULL;
Yan Zheng9036c102008-10-30 14:19:41 -04004350 cur_offset = last_byte;
Yan, Zheng80825102009-11-12 09:35:36 +00004351 if (cur_offset >= block_end)
Yan Zheng9036c102008-10-30 14:19:41 -04004352 break;
4353 }
Yan, Zhenga22285a2010-05-16 10:48:46 -04004354 free_extent_map(em);
Josef Bacik2ac55d42010-02-03 19:33:23 +00004355 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4356 GFP_NOFS);
Yan Zheng9036c102008-10-30 14:19:41 -04004357 return err;
4358}
4359
Eric Sandeen3972f262013-01-12 02:57:22 +00004360static int btrfs_setsize(struct inode *inode, struct iattr *attr)
Yan, Zheng80825102009-11-12 09:35:36 +00004361{
Miao Xief4a2f4c2011-12-14 20:12:01 -05004362 struct btrfs_root *root = BTRFS_I(inode)->root;
4363 struct btrfs_trans_handle *trans;
Josef Bacika41ad392011-01-31 15:30:16 -05004364 loff_t oldsize = i_size_read(inode);
Eric Sandeen3972f262013-01-12 02:57:22 +00004365 loff_t newsize = attr->ia_size;
4366 int mask = attr->ia_valid;
Yan, Zheng80825102009-11-12 09:35:36 +00004367 int ret;
4368
Eric Sandeen3972f262013-01-12 02:57:22 +00004369 /*
4370 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4371 * special case where we need to update the times despite not having
4372 * these flags set. For all other operations the VFS set these flags
4373 * explicitly if it wants a timestamp update.
4374 */
4375 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4376 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4377
Josef Bacika41ad392011-01-31 15:30:16 -05004378 if (newsize > oldsize) {
Kirill A. Shutemov7caef262013-09-12 15:13:56 -07004379 truncate_pagecache(inode, newsize);
Josef Bacika41ad392011-01-31 15:30:16 -05004380 ret = btrfs_cont_expand(inode, oldsize, newsize);
Miao Xief4a2f4c2011-12-14 20:12:01 -05004381 if (ret)
Yan, Zheng80825102009-11-12 09:35:36 +00004382 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00004383
Miao Xief4a2f4c2011-12-14 20:12:01 -05004384 trans = btrfs_start_transaction(root, 1);
4385 if (IS_ERR(trans))
4386 return PTR_ERR(trans);
4387
4388 i_size_write(inode, newsize);
4389 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4390 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05004391 btrfs_end_transaction(trans, root);
Josef Bacika41ad392011-01-31 15:30:16 -05004392 } else {
Yan, Zheng80825102009-11-12 09:35:36 +00004393
Josef Bacika41ad392011-01-31 15:30:16 -05004394 /*
4395 * We're truncating a file that used to have good data down to
4396 * zero. Make sure it gets into the ordered flush list so that
4397 * any new writes get down to disk quickly.
4398 */
4399 if (newsize == 0)
Josef Bacik72ac3c02012-05-23 14:13:11 -04004400 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4401 &BTRFS_I(inode)->runtime_flags);
Yan, Zheng80825102009-11-12 09:35:36 +00004402
Josef Bacikf3fe8202013-01-07 17:03:21 -05004403 /*
4404 * 1 for the orphan item we're going to add
4405 * 1 for the orphan item deletion.
4406 */
4407 trans = btrfs_start_transaction(root, 2);
4408 if (IS_ERR(trans))
4409 return PTR_ERR(trans);
4410
4411 /*
4412 * We need to do this in case we fail at _any_ point during the
4413 * actual truncate. Once we do the truncate_setsize we could
4414 * invalidate pages which forces any outstanding ordered io to
4415 * be instantly completed which will give us extents that need
4416 * to be truncated. If we fail to get an orphan inode down we
4417 * could have left over extents that were never meant to live,
4418 * so we need to garuntee from this point on that everything
4419 * will be consistent.
4420 */
4421 ret = btrfs_orphan_add(trans, inode);
4422 btrfs_end_transaction(trans, root);
4423 if (ret)
4424 return ret;
4425
Josef Bacika41ad392011-01-31 15:30:16 -05004426 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4427 truncate_setsize(inode, newsize);
Miao Xie2e60a512013-02-08 07:01:08 +00004428
4429 /* Disable nonlocked read DIO to avoid the end less truncate */
4430 btrfs_inode_block_unlocked_dio(inode);
4431 inode_dio_wait(inode);
4432 btrfs_inode_resume_unlocked_dio(inode);
4433
Josef Bacika41ad392011-01-31 15:30:16 -05004434 ret = btrfs_truncate(inode);
Josef Bacik7f4f6e02013-08-29 16:43:28 -04004435 if (ret && inode->i_nlink) {
4436 int err;
4437
4438 /*
4439 * failed to truncate, disk_i_size is only adjusted down
4440 * as we remove extents, so it should represent the true
4441 * size of the inode, so reset the in memory size and
4442 * delete our orphan entry.
4443 */
4444 trans = btrfs_join_transaction(root);
4445 if (IS_ERR(trans)) {
4446 btrfs_orphan_del(NULL, inode);
4447 return ret;
4448 }
4449 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4450 err = btrfs_orphan_del(trans, inode);
4451 if (err)
4452 btrfs_abort_transaction(trans, root, err);
4453 btrfs_end_transaction(trans, root);
4454 }
Yan, Zheng80825102009-11-12 09:35:36 +00004455 }
4456
Josef Bacika41ad392011-01-31 15:30:16 -05004457 return ret;
Yan, Zheng80825102009-11-12 09:35:36 +00004458}
4459
Chris Mason39279cc2007-06-12 06:35:45 -04004460static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4461{
4462 struct inode *inode = dentry->d_inode;
Li Zefanb83cc962010-12-20 16:04:08 +08004463 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004464 int err;
4465
Li Zefanb83cc962010-12-20 16:04:08 +08004466 if (btrfs_root_readonly(root))
4467 return -EROFS;
4468
Chris Mason39279cc2007-06-12 06:35:45 -04004469 err = inode_change_ok(inode, attr);
4470 if (err)
4471 return err;
4472
Chris Mason5a3f23d2009-03-31 13:27:11 -04004473 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Eric Sandeen3972f262013-01-12 02:57:22 +00004474 err = btrfs_setsize(inode, attr);
Yan, Zheng80825102009-11-12 09:35:36 +00004475 if (err)
4476 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004477 }
Yan Zheng9036c102008-10-30 14:19:41 -04004478
Christoph Hellwig10257742010-06-04 11:30:02 +02004479 if (attr->ia_valid) {
4480 setattr_copy(inode, attr);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04004481 inode_inc_iversion(inode);
Josef Bacik22c44fe2011-11-30 10:45:38 -05004482 err = btrfs_dirty_inode(inode);
Josef Bacik33268ea2008-07-24 12:16:36 -04004483
Josef Bacik22c44fe2011-11-30 10:45:38 -05004484 if (!err && attr->ia_valid & ATTR_MODE)
Christoph Hellwig10257742010-06-04 11:30:02 +02004485 err = btrfs_acl_chmod(inode);
4486 }
4487
Chris Mason39279cc2007-06-12 06:35:45 -04004488 return err;
4489}
Chris Mason61295eb2008-01-14 16:24:38 -05004490
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004491/*
4492 * While truncating the inode pages during eviction, we get the VFS calling
4493 * btrfs_invalidatepage() against each page of the inode. This is slow because
4494 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4495 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4496 * extent_state structures over and over, wasting lots of time.
4497 *
4498 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4499 * those expensive operations on a per page basis and do only the ordered io
4500 * finishing, while we release here the extent_map and extent_state structures,
4501 * without the excessive merging and splitting.
4502 */
4503static void evict_inode_truncate_pages(struct inode *inode)
4504{
4505 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4506 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4507 struct rb_node *node;
4508
4509 ASSERT(inode->i_state & I_FREEING);
4510 truncate_inode_pages(&inode->i_data, 0);
4511
4512 write_lock(&map_tree->lock);
4513 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4514 struct extent_map *em;
4515
4516 node = rb_first(&map_tree->map);
4517 em = rb_entry(node, struct extent_map, rb_node);
4518 remove_extent_mapping(map_tree, em);
4519 free_extent_map(em);
4520 }
4521 write_unlock(&map_tree->lock);
4522
4523 spin_lock(&io_tree->lock);
4524 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4525 struct extent_state *state;
4526 struct extent_state *cached_state = NULL;
4527
4528 node = rb_first(&io_tree->state);
4529 state = rb_entry(node, struct extent_state, rb_node);
4530 atomic_inc(&state->refs);
4531 spin_unlock(&io_tree->lock);
4532
4533 lock_extent_bits(io_tree, state->start, state->end,
4534 0, &cached_state);
4535 clear_extent_bit(io_tree, state->start, state->end,
4536 EXTENT_LOCKED | EXTENT_DIRTY |
4537 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4538 EXTENT_DEFRAG, 1, 1,
4539 &cached_state, GFP_NOFS);
4540 free_extent_state(state);
4541
4542 spin_lock(&io_tree->lock);
4543 }
4544 spin_unlock(&io_tree->lock);
4545}
4546
Al Virobd555972010-06-07 11:35:40 -04004547void btrfs_evict_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04004548{
4549 struct btrfs_trans_handle *trans;
4550 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik726c35f2011-09-26 15:46:06 -04004551 struct btrfs_block_rsv *rsv, *global_rsv;
Josef Bacik07127182011-08-19 10:29:59 -04004552 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04004553 int ret;
4554
liubo1abe9b82011-03-24 11:18:59 +00004555 trace_btrfs_inode_evict(inode);
4556
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00004557 evict_inode_truncate_pages(inode);
4558
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02004559 if (inode->i_nlink &&
4560 ((btrfs_root_refs(&root->root_item) != 0 &&
4561 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4562 btrfs_is_free_space_inode(inode)))
Al Virobd555972010-06-07 11:35:40 -04004563 goto no_delete;
4564
Chris Mason39279cc2007-06-12 06:35:45 -04004565 if (is_bad_inode(inode)) {
Josef Bacik7b128762008-07-24 12:17:14 -04004566 btrfs_orphan_del(NULL, inode);
Chris Mason39279cc2007-06-12 06:35:45 -04004567 goto no_delete;
4568 }
Al Virobd555972010-06-07 11:35:40 -04004569 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
Chris Mason4a096752008-07-21 10:29:44 -04004570 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Chris Mason5f39d392007-10-15 16:14:19 -04004571
Yan, Zhengc71bf092009-11-12 09:34:40 +00004572 if (root->fs_info->log_root_recovering) {
Liu Bo6bf02312012-06-25 21:59:09 -06004573 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
Josef Bacik8a35d952012-05-23 14:26:42 -04004574 &BTRFS_I(inode)->runtime_flags));
Yan, Zhengc71bf092009-11-12 09:34:40 +00004575 goto no_delete;
4576 }
4577
Yan, Zheng76dda932009-09-21 16:00:26 -04004578 if (inode->i_nlink > 0) {
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02004579 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4580 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
Yan, Zheng76dda932009-09-21 16:00:26 -04004581 goto no_delete;
4582 }
4583
Miao Xie0e8c36a2012-12-19 06:59:51 +00004584 ret = btrfs_commit_inode_delayed_inode(inode);
4585 if (ret) {
4586 btrfs_orphan_del(NULL, inode);
4587 goto no_delete;
4588 }
4589
Miao Xie66d8f3d2012-09-06 04:02:28 -06004590 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacik4289a662011-08-05 13:22:24 -04004591 if (!rsv) {
4592 btrfs_orphan_del(NULL, inode);
4593 goto no_delete;
4594 }
Josef Bacik4a338542011-08-29 11:01:31 -04004595 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04004596 rsv->failfast = 1;
Josef Bacik726c35f2011-09-26 15:46:06 -04004597 global_rsv = &root->fs_info->global_block_rsv;
Josef Bacik4289a662011-08-05 13:22:24 -04004598
Chris Masondbe674a2008-07-17 12:54:05 -04004599 btrfs_i_size_write(inode, 0);
Chris Mason5f39d392007-10-15 16:14:19 -04004600
Josef Bacik4289a662011-08-05 13:22:24 -04004601 /*
Miao Xie8407aa42012-09-07 01:43:32 -06004602 * This is a bit simpler than btrfs_truncate since we've already
4603 * reserved our space for our orphan item in the unlink, so we just
4604 * need to reserve some slack space in case we add bytes and update
4605 * inode item when doing the truncate.
Josef Bacik4289a662011-08-05 13:22:24 -04004606 */
Yan, Zheng80825102009-11-12 09:35:36 +00004607 while (1) {
Miao Xie08e007d2012-10-16 11:33:38 +00004608 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4609 BTRFS_RESERVE_FLUSH_LIMIT);
Yan, Zheng80825102009-11-12 09:35:36 +00004610
Josef Bacik726c35f2011-09-26 15:46:06 -04004611 /*
4612 * Try and steal from the global reserve since we will
4613 * likely not use this space anyway, we want to try as
4614 * hard as possible to get this to work.
4615 */
4616 if (ret)
4617 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4618
Yan, Zhengd68fc572010-05-16 10:49:58 -04004619 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004620 btrfs_warn(root->fs_info,
4621 "Could not get space for a delete, will truncate on mount %d",
4622 ret);
Josef Bacik4289a662011-08-05 13:22:24 -04004623 btrfs_orphan_del(NULL, inode);
4624 btrfs_free_block_rsv(root, rsv);
4625 goto no_delete;
Yan, Zhengd68fc572010-05-16 10:49:58 -04004626 }
4627
Miao Xie0e8c36a2012-12-19 06:59:51 +00004628 trans = btrfs_join_transaction(root);
Josef Bacik4289a662011-08-05 13:22:24 -04004629 if (IS_ERR(trans)) {
4630 btrfs_orphan_del(NULL, inode);
4631 btrfs_free_block_rsv(root, rsv);
4632 goto no_delete;
4633 }
4634
4635 trans->block_rsv = rsv;
4636
Yan, Zhengd68fc572010-05-16 10:49:58 -04004637 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
Josef Bacikca7e70f2012-08-27 17:48:15 -04004638 if (ret != -ENOSPC)
Yan, Zheng80825102009-11-12 09:35:36 +00004639 break;
4640
Miao Xie8407aa42012-09-07 01:43:32 -06004641 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00004642 btrfs_end_transaction(trans, root);
4643 trans = NULL;
Liu Bob53d3f52012-11-14 14:34:34 +00004644 btrfs_btree_balance_dirty(root);
Josef Bacik7b128762008-07-24 12:17:14 -04004645 }
4646
Josef Bacik4289a662011-08-05 13:22:24 -04004647 btrfs_free_block_rsv(root, rsv);
4648
Josef Bacik4ef31a42013-08-13 14:10:08 -04004649 /*
4650 * Errors here aren't a big deal, it just means we leave orphan items
4651 * in the tree. They will be cleaned up on the next mount.
4652 */
Yan, Zheng80825102009-11-12 09:35:36 +00004653 if (ret == 0) {
Josef Bacik4289a662011-08-05 13:22:24 -04004654 trans->block_rsv = root->orphan_block_rsv;
Josef Bacik4ef31a42013-08-13 14:10:08 -04004655 btrfs_orphan_del(trans, inode);
4656 } else {
4657 btrfs_orphan_del(NULL, inode);
Yan, Zheng80825102009-11-12 09:35:36 +00004658 }
Chris Mason85e21ba2008-01-29 15:11:36 -05004659
Josef Bacik4289a662011-08-05 13:22:24 -04004660 trans->block_rsv = &root->fs_info->trans_block_rsv;
Li Zefan581bb052011-04-20 10:06:11 +08004661 if (!(root == root->fs_info->tree_root ||
4662 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
Li Zefan33345d012011-04-20 10:31:50 +08004663 btrfs_return_ino(root, btrfs_ino(inode));
Li Zefan581bb052011-04-20 10:06:11 +08004664
Chris Mason54aa1f42007-06-22 14:16:25 -04004665 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00004666 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04004667no_delete:
Miao Xie89042e52013-05-15 07:48:15 +00004668 btrfs_remove_delayed_node(inode);
Jan Karadbd57682012-05-03 14:48:02 +02004669 clear_inode(inode);
Yan, Zheng80825102009-11-12 09:35:36 +00004670 return;
Chris Mason39279cc2007-06-12 06:35:45 -04004671}
4672
4673/*
4674 * this returns the key found in the dir entry in the location pointer.
4675 * If no dir entries were found, location->objectid is 0.
4676 */
4677static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4678 struct btrfs_key *location)
4679{
4680 const char *name = dentry->d_name.name;
4681 int namelen = dentry->d_name.len;
4682 struct btrfs_dir_item *di;
4683 struct btrfs_path *path;
4684 struct btrfs_root *root = BTRFS_I(dir)->root;
Yan0d9f7f32007-10-25 15:48:28 -04004685 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004686
4687 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07004688 if (!path)
4689 return -ENOMEM;
Chris Mason39544012007-12-12 14:38:19 -05004690
Li Zefan33345d012011-04-20 10:31:50 +08004691 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
Chris Mason39279cc2007-06-12 06:35:45 -04004692 namelen, 0);
Yan0d9f7f32007-10-25 15:48:28 -04004693 if (IS_ERR(di))
4694 ret = PTR_ERR(di);
Chris Masond3977122009-01-05 21:25:51 -05004695
David Sterbac7040052011-04-19 18:00:01 +02004696 if (IS_ERR_OR_NULL(di))
Chris Mason39544012007-12-12 14:38:19 -05004697 goto out_err;
Chris Masond3977122009-01-05 21:25:51 -05004698
Chris Mason5f39d392007-10-15 16:14:19 -04004699 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
Chris Mason39279cc2007-06-12 06:35:45 -04004700out:
Chris Mason39279cc2007-06-12 06:35:45 -04004701 btrfs_free_path(path);
4702 return ret;
Chris Mason39544012007-12-12 14:38:19 -05004703out_err:
4704 location->objectid = 0;
4705 goto out;
Chris Mason39279cc2007-06-12 06:35:45 -04004706}
4707
4708/*
4709 * when we hit a tree root in a directory, the btrfs part of the inode
4710 * needs to be changed to reflect the root directory of the tree root. This
4711 * is kind of like crossing a mount point.
4712 */
4713static int fixup_tree_root_location(struct btrfs_root *root,
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004714 struct inode *dir,
4715 struct dentry *dentry,
4716 struct btrfs_key *location,
4717 struct btrfs_root **sub_root)
Chris Mason39279cc2007-06-12 06:35:45 -04004718{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004719 struct btrfs_path *path;
4720 struct btrfs_root *new_root;
4721 struct btrfs_root_ref *ref;
4722 struct extent_buffer *leaf;
4723 int ret;
4724 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004725
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004726 path = btrfs_alloc_path();
4727 if (!path) {
4728 err = -ENOMEM;
4729 goto out;
4730 }
Chris Mason39279cc2007-06-12 06:35:45 -04004731
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004732 err = -ENOENT;
Kelley Nielsen75ac2dd2013-11-04 19:35:58 -08004733 ret = btrfs_find_item(root->fs_info->tree_root, path,
4734 BTRFS_I(dir)->root->root_key.objectid,
4735 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004736 if (ret) {
4737 if (ret < 0)
4738 err = ret;
4739 goto out;
4740 }
Chris Mason39279cc2007-06-12 06:35:45 -04004741
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004742 leaf = path->nodes[0];
4743 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
Li Zefan33345d012011-04-20 10:31:50 +08004744 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004745 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4746 goto out;
4747
4748 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4749 (unsigned long)(ref + 1),
4750 dentry->d_name.len);
4751 if (ret)
4752 goto out;
4753
David Sterbab3b4aa72011-04-21 01:20:15 +02004754 btrfs_release_path(path);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004755
4756 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4757 if (IS_ERR(new_root)) {
4758 err = PTR_ERR(new_root);
4759 goto out;
4760 }
4761
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004762 *sub_root = new_root;
4763 location->objectid = btrfs_root_dirid(&new_root->root_item);
4764 location->type = BTRFS_INODE_ITEM_KEY;
Chris Mason39279cc2007-06-12 06:35:45 -04004765 location->offset = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004766 err = 0;
4767out:
4768 btrfs_free_path(path);
4769 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04004770}
4771
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004772static void inode_tree_add(struct inode *inode)
4773{
4774 struct btrfs_root *root = BTRFS_I(inode)->root;
4775 struct btrfs_inode *entry;
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004776 struct rb_node **p;
4777 struct rb_node *parent;
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01004778 struct rb_node *new = &BTRFS_I(inode)->rb_node;
Li Zefan33345d012011-04-20 10:31:50 +08004779 u64 ino = btrfs_ino(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004780
Al Viro1d3382cb2010-10-23 15:19:20 -04004781 if (inode_unhashed(inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04004782 return;
Miao Xiee1409ce2013-05-15 07:48:16 +00004783 parent = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004784 spin_lock(&root->inode_lock);
Miao Xiee1409ce2013-05-15 07:48:16 +00004785 p = &root->inode_tree.rb_node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004786 while (*p) {
4787 parent = *p;
4788 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4789
Li Zefan33345d012011-04-20 10:31:50 +08004790 if (ino < btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004791 p = &parent->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08004792 else if (ino > btrfs_ino(&entry->vfs_inode))
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004793 p = &parent->rb_right;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004794 else {
4795 WARN_ON(!(entry->vfs_inode.i_state &
Al Viroa4ffdde2010-06-02 17:38:30 -04004796 (I_WILL_FREE | I_FREEING)));
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01004797 rb_replace_node(parent, new, &root->inode_tree);
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004798 RB_CLEAR_NODE(parent);
4799 spin_unlock(&root->inode_lock);
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01004800 return;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004801 }
4802 }
Filipe David Borba Mananacef21932013-09-02 12:19:13 +01004803 rb_link_node(new, parent, p);
4804 rb_insert_color(new, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004805 spin_unlock(&root->inode_lock);
4806}
4807
4808static void inode_tree_del(struct inode *inode)
4809{
4810 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan, Zheng76dda932009-09-21 16:00:26 -04004811 int empty = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004812
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004813 spin_lock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004814 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004815 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004816 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
Yan, Zheng76dda932009-09-21 16:00:26 -04004817 empty = RB_EMPTY_ROOT(&root->inode_tree);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004818 }
From: Nick Piggin03e860b2009-08-21 10:09:44 +02004819 spin_unlock(&root->inode_lock);
Yan, Zheng76dda932009-09-21 16:00:26 -04004820
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02004821 if (empty && btrfs_root_refs(&root->root_item) == 0) {
Yan, Zheng76dda932009-09-21 16:00:26 -04004822 synchronize_srcu(&root->fs_info->subvol_srcu);
4823 spin_lock(&root->inode_lock);
4824 empty = RB_EMPTY_ROOT(&root->inode_tree);
4825 spin_unlock(&root->inode_lock);
4826 if (empty)
4827 btrfs_add_dead_root(root);
4828 }
4829}
4830
Jeff Mahoney143bede2012-03-01 14:56:26 +01004831void btrfs_invalidate_inodes(struct btrfs_root *root)
Yan, Zheng76dda932009-09-21 16:00:26 -04004832{
4833 struct rb_node *node;
4834 struct rb_node *prev;
4835 struct btrfs_inode *entry;
4836 struct inode *inode;
4837 u64 objectid = 0;
4838
4839 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4840
4841 spin_lock(&root->inode_lock);
4842again:
4843 node = root->inode_tree.rb_node;
4844 prev = NULL;
4845 while (node) {
4846 prev = node;
4847 entry = rb_entry(node, struct btrfs_inode, rb_node);
4848
Li Zefan33345d012011-04-20 10:31:50 +08004849 if (objectid < btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04004850 node = node->rb_left;
Li Zefan33345d012011-04-20 10:31:50 +08004851 else if (objectid > btrfs_ino(&entry->vfs_inode))
Yan, Zheng76dda932009-09-21 16:00:26 -04004852 node = node->rb_right;
4853 else
4854 break;
4855 }
4856 if (!node) {
4857 while (prev) {
4858 entry = rb_entry(prev, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08004859 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
Yan, Zheng76dda932009-09-21 16:00:26 -04004860 node = prev;
4861 break;
4862 }
4863 prev = rb_next(prev);
4864 }
4865 }
4866 while (node) {
4867 entry = rb_entry(node, struct btrfs_inode, rb_node);
Li Zefan33345d012011-04-20 10:31:50 +08004868 objectid = btrfs_ino(&entry->vfs_inode) + 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04004869 inode = igrab(&entry->vfs_inode);
4870 if (inode) {
4871 spin_unlock(&root->inode_lock);
4872 if (atomic_read(&inode->i_count) > 1)
4873 d_prune_aliases(inode);
4874 /*
Al Viro45321ac2010-06-07 13:43:19 -04004875 * btrfs_drop_inode will have it removed from
Yan, Zheng76dda932009-09-21 16:00:26 -04004876 * the inode cache when its usage count
4877 * hits zero.
4878 */
4879 iput(inode);
4880 cond_resched();
4881 spin_lock(&root->inode_lock);
4882 goto again;
4883 }
4884
4885 if (cond_resched_lock(&root->inode_lock))
4886 goto again;
4887
4888 node = rb_next(node);
4889 }
4890 spin_unlock(&root->inode_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004891}
4892
Chris Masone02119d2008-09-05 16:13:11 -04004893static int btrfs_init_locked_inode(struct inode *inode, void *p)
4894{
4895 struct btrfs_iget_args *args = p;
4896 inode->i_ino = args->ino;
Chris Masone02119d2008-09-05 16:13:11 -04004897 BTRFS_I(inode)->root = args->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004898 return 0;
4899}
4900
4901static int btrfs_find_actor(struct inode *inode, void *opaque)
4902{
4903 struct btrfs_iget_args *args = opaque;
Li Zefan33345d012011-04-20 10:31:50 +08004904 return args->ino == btrfs_ino(inode) &&
Chris Masond3977122009-01-05 21:25:51 -05004905 args->root == BTRFS_I(inode)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004906}
4907
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004908static struct inode *btrfs_iget_locked(struct super_block *s,
4909 u64 objectid,
4910 struct btrfs_root *root)
Chris Mason39279cc2007-06-12 06:35:45 -04004911{
4912 struct inode *inode;
4913 struct btrfs_iget_args args;
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01004914 unsigned long hashval = btrfs_inode_hash(objectid, root);
4915
Chris Mason39279cc2007-06-12 06:35:45 -04004916 args.ino = objectid;
4917 args.root = root;
4918
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01004919 inode = iget5_locked(s, hashval, btrfs_find_actor,
Chris Mason39279cc2007-06-12 06:35:45 -04004920 btrfs_init_locked_inode,
4921 (void *)&args);
4922 return inode;
4923}
4924
Balaji Rao1a54ef82008-07-21 02:01:04 +05304925/* Get an inode object given its location and corresponding root.
4926 * Returns in *is_new if the inode was read from disk
4927 */
4928struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Josef Bacik73f73412009-12-04 17:38:27 +00004929 struct btrfs_root *root, int *new)
Balaji Rao1a54ef82008-07-21 02:01:04 +05304930{
4931 struct inode *inode;
4932
4933 inode = btrfs_iget_locked(s, location->objectid, root);
4934 if (!inode)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004935 return ERR_PTR(-ENOMEM);
Balaji Rao1a54ef82008-07-21 02:01:04 +05304936
4937 if (inode->i_state & I_NEW) {
4938 BTRFS_I(inode)->root = root;
4939 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4940 btrfs_read_locked_inode(inode);
Mark Fasheh1748f842011-07-12 11:25:31 -07004941 if (!is_bad_inode(inode)) {
4942 inode_tree_add(inode);
4943 unlock_new_inode(inode);
4944 if (new)
4945 *new = 1;
4946 } else {
Sergei Trofimoviche0b6d652011-09-11 10:52:24 -04004947 unlock_new_inode(inode);
4948 iput(inode);
4949 inode = ERR_PTR(-ESTALE);
Mark Fasheh1748f842011-07-12 11:25:31 -07004950 }
4951 }
4952
Balaji Rao1a54ef82008-07-21 02:01:04 +05304953 return inode;
4954}
4955
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004956static struct inode *new_simple_dir(struct super_block *s,
4957 struct btrfs_key *key,
4958 struct btrfs_root *root)
4959{
4960 struct inode *inode = new_inode(s);
4961
4962 if (!inode)
4963 return ERR_PTR(-ENOMEM);
4964
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004965 BTRFS_I(inode)->root = root;
4966 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
Josef Bacik72ac3c02012-05-23 14:13:11 -04004967 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004968
4969 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
Li Zefan848cce02012-02-21 17:04:28 +08004970 inode->i_op = &btrfs_dir_ro_inode_operations;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004971 inode->i_fop = &simple_dir_operations;
4972 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4973 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4974
4975 return inode;
4976}
4977
Chris Mason3de45862008-11-17 21:02:50 -05004978struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04004979{
Chris Masond3977122009-01-05 21:25:51 -05004980 struct inode *inode;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004981 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04004982 struct btrfs_root *sub_root = root;
4983 struct btrfs_key location;
Yan, Zheng76dda932009-09-21 16:00:26 -04004984 int index;
Josef Bacikb4aff1f2011-06-28 16:18:59 -04004985 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04004986
4987 if (dentry->d_name.len > BTRFS_NAME_LEN)
4988 return ERR_PTR(-ENAMETOOLONG);
Chris Mason5f39d392007-10-15 16:14:19 -04004989
Jeff Layton39e3c952012-11-28 11:30:53 -05004990 ret = btrfs_inode_by_name(dir, dentry, &location);
Chris Mason39279cc2007-06-12 06:35:45 -04004991 if (ret < 0)
4992 return ERR_PTR(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04004993
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004994 if (location.objectid == 0)
4995 return NULL;
4996
4997 if (location.type == BTRFS_INODE_ITEM_KEY) {
Josef Bacik73f73412009-12-04 17:38:27 +00004998 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04004999 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04005000 }
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005001
5002 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5003
Yan, Zheng76dda932009-09-21 16:00:26 -04005004 index = srcu_read_lock(&root->fs_info->subvol_srcu);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005005 ret = fixup_tree_root_location(root, dir, dentry,
5006 &location, &sub_root);
5007 if (ret < 0) {
5008 if (ret != -ENOENT)
5009 inode = ERR_PTR(ret);
5010 else
5011 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5012 } else {
Josef Bacik73f73412009-12-04 17:38:27 +00005013 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04005014 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005015 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5016
Julia Lawall34d19ba2011-01-24 19:55:19 +00005017 if (!IS_ERR(inode) && root != sub_root) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00005018 down_read(&root->fs_info->cleanup_work_sem);
5019 if (!(inode->i_sb->s_flags & MS_RDONLY))
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005020 ret = btrfs_orphan_cleanup(sub_root);
Yan, Zhengc71bf092009-11-12 09:34:40 +00005021 up_read(&root->fs_info->cleanup_work_sem);
Josef Bacik01cd3362013-06-03 21:39:49 -04005022 if (ret) {
5023 iput(inode);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05005024 inode = ERR_PTR(ret);
Josef Bacik01cd3362013-06-03 21:39:49 -04005025 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00005026 }
5027
Chris Mason3de45862008-11-17 21:02:50 -05005028 return inode;
5029}
5030
Nick Pigginfe15ce42011-01-07 17:49:23 +11005031static int btrfs_dentry_delete(const struct dentry *dentry)
Yan, Zheng76dda932009-09-21 16:00:26 -04005032{
5033 struct btrfs_root *root;
Li Zefan848cce02012-02-21 17:04:28 +08005034 struct inode *inode = dentry->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04005035
Li Zefan848cce02012-02-21 17:04:28 +08005036 if (!inode && !IS_ROOT(dentry))
5037 inode = dentry->d_parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04005038
Li Zefan848cce02012-02-21 17:04:28 +08005039 if (inode) {
5040 root = BTRFS_I(inode)->root;
Yan, Zhengefefb142009-10-09 09:25:16 -04005041 if (btrfs_root_refs(&root->root_item) == 0)
5042 return 1;
Li Zefan848cce02012-02-21 17:04:28 +08005043
5044 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5045 return 1;
Yan, Zhengefefb142009-10-09 09:25:16 -04005046 }
Yan, Zheng76dda932009-09-21 16:00:26 -04005047 return 0;
5048}
5049
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005050static void btrfs_dentry_release(struct dentry *dentry)
5051{
5052 if (dentry->d_fsdata)
5053 kfree(dentry->d_fsdata);
5054}
5055
Chris Mason3de45862008-11-17 21:02:50 -05005056static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -04005057 unsigned int flags)
Chris Mason3de45862008-11-17 21:02:50 -05005058{
Josef Bacika66e7cc2011-09-18 10:34:03 -04005059 struct dentry *ret;
5060
5061 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
Josef Bacika66e7cc2011-09-18 10:34:03 -04005062 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005063}
5064
Miao Xie16cdcec2011-04-22 18:12:22 +08005065unsigned char btrfs_filetype_table[] = {
Chris Mason39279cc2007-06-12 06:35:45 -04005066 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5067};
5068
Al Viro9cdda8d2013-05-22 16:48:09 -04005069static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
Chris Mason39279cc2007-06-12 06:35:45 -04005070{
Al Viro9cdda8d2013-05-22 16:48:09 -04005071 struct inode *inode = file_inode(file);
Chris Mason39279cc2007-06-12 06:35:45 -04005072 struct btrfs_root *root = BTRFS_I(inode)->root;
5073 struct btrfs_item *item;
5074 struct btrfs_dir_item *di;
5075 struct btrfs_key key;
Chris Mason5f39d392007-10-15 16:14:19 -04005076 struct btrfs_key found_key;
Chris Mason39279cc2007-06-12 06:35:45 -04005077 struct btrfs_path *path;
Miao Xie16cdcec2011-04-22 18:12:22 +08005078 struct list_head ins_list;
5079 struct list_head del_list;
Chris Mason39279cc2007-06-12 06:35:45 -04005080 int ret;
Chris Mason5f39d392007-10-15 16:14:19 -04005081 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04005082 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -04005083 unsigned char d_type;
5084 int over = 0;
5085 u32 di_cur;
5086 u32 di_total;
5087 u32 di_len;
5088 int key_type = BTRFS_DIR_INDEX_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04005089 char tmp_name[32];
5090 char *name_ptr;
5091 int name_len;
Al Viro9cdda8d2013-05-22 16:48:09 -04005092 int is_curr = 0; /* ctx->pos points to the current index? */
Chris Mason39279cc2007-06-12 06:35:45 -04005093
5094 /* FIXME, use a real flag for deciding about the key type */
5095 if (root->fs_info->tree_root == root)
5096 key_type = BTRFS_DIR_ITEM_KEY;
Chris Mason5f39d392007-10-15 16:14:19 -04005097
Al Viro9cdda8d2013-05-22 16:48:09 -04005098 if (!dir_emit_dots(file, ctx))
5099 return 0;
5100
David Woodhouse49593bf2008-08-17 17:08:36 +01005101 path = btrfs_alloc_path();
Miao Xie16cdcec2011-04-22 18:12:22 +08005102 if (!path)
5103 return -ENOMEM;
Chris Masonff5714c2011-05-28 07:00:39 -04005104
Josef Bacik026fd312011-05-13 10:32:11 -04005105 path->reada = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01005106
Miao Xie16cdcec2011-04-22 18:12:22 +08005107 if (key_type == BTRFS_DIR_INDEX_KEY) {
5108 INIT_LIST_HEAD(&ins_list);
5109 INIT_LIST_HEAD(&del_list);
5110 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5111 }
5112
Chris Mason39279cc2007-06-12 06:35:45 -04005113 btrfs_set_key_type(&key, key_type);
Al Viro9cdda8d2013-05-22 16:48:09 -04005114 key.offset = ctx->pos;
Li Zefan33345d012011-04-20 10:31:50 +08005115 key.objectid = btrfs_ino(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005116
Chris Mason39279cc2007-06-12 06:35:45 -04005117 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5118 if (ret < 0)
5119 goto err;
David Woodhouse49593bf2008-08-17 17:08:36 +01005120
5121 while (1) {
Chris Mason5f39d392007-10-15 16:14:19 -04005122 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -04005123 slot = path->slots[0];
Li Zefanb9e03af2011-03-23 10:43:58 +08005124 if (slot >= btrfs_header_nritems(leaf)) {
5125 ret = btrfs_next_leaf(root, path);
5126 if (ret < 0)
5127 goto err;
5128 else if (ret > 0)
5129 break;
5130 continue;
Chris Mason39279cc2007-06-12 06:35:45 -04005131 }
Chris Mason3de45862008-11-17 21:02:50 -05005132
Ross Kirkdd3cc162013-09-16 15:58:09 +01005133 item = btrfs_item_nr(slot);
Chris Mason5f39d392007-10-15 16:14:19 -04005134 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5135
5136 if (found_key.objectid != key.objectid)
Chris Mason39279cc2007-06-12 06:35:45 -04005137 break;
Chris Mason5f39d392007-10-15 16:14:19 -04005138 if (btrfs_key_type(&found_key) != key_type)
Chris Mason39279cc2007-06-12 06:35:45 -04005139 break;
Al Viro9cdda8d2013-05-22 16:48:09 -04005140 if (found_key.offset < ctx->pos)
Li Zefanb9e03af2011-03-23 10:43:58 +08005141 goto next;
Miao Xie16cdcec2011-04-22 18:12:22 +08005142 if (key_type == BTRFS_DIR_INDEX_KEY &&
5143 btrfs_should_delete_dir_index(&del_list,
5144 found_key.offset))
5145 goto next;
Chris Mason5f39d392007-10-15 16:14:19 -04005146
Al Viro9cdda8d2013-05-22 16:48:09 -04005147 ctx->pos = found_key.offset;
Miao Xie16cdcec2011-04-22 18:12:22 +08005148 is_curr = 1;
David Woodhouse49593bf2008-08-17 17:08:36 +01005149
Chris Mason39279cc2007-06-12 06:35:45 -04005150 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5151 di_cur = 0;
Chris Mason5f39d392007-10-15 16:14:19 -04005152 di_total = btrfs_item_size(leaf, item);
David Woodhouse49593bf2008-08-17 17:08:36 +01005153
5154 while (di_cur < di_total) {
Chris Mason5f39d392007-10-15 16:14:19 -04005155 struct btrfs_key location;
5156
Josef Bacik22a94d42011-03-16 16:47:17 -04005157 if (verify_dir_item(root, leaf, di))
5158 break;
5159
Chris Mason5f39d392007-10-15 16:14:19 -04005160 name_len = btrfs_dir_name_len(leaf, di);
David Woodhouse49593bf2008-08-17 17:08:36 +01005161 if (name_len <= sizeof(tmp_name)) {
Chris Mason5f39d392007-10-15 16:14:19 -04005162 name_ptr = tmp_name;
5163 } else {
5164 name_ptr = kmalloc(name_len, GFP_NOFS);
David Woodhouse49593bf2008-08-17 17:08:36 +01005165 if (!name_ptr) {
5166 ret = -ENOMEM;
5167 goto err;
5168 }
Chris Mason5f39d392007-10-15 16:14:19 -04005169 }
5170 read_extent_buffer(leaf, name_ptr,
5171 (unsigned long)(di + 1), name_len);
5172
5173 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5174 btrfs_dir_item_key_to_cpu(leaf, di, &location);
Chris Mason3de45862008-11-17 21:02:50 -05005175
Josef Bacikb4aff1f2011-06-28 16:18:59 -04005176
Chris Mason3de45862008-11-17 21:02:50 -05005177 /* is this a reference to our own snapshot? If so
Arne Jansen8c9c2bf2012-02-25 09:09:30 +01005178 * skip it.
5179 *
5180 * In contrast to old kernels, we insert the snapshot's
5181 * dir item and dir index after it has been created, so
5182 * we won't find a reference to our own snapshot. We
5183 * still keep the following code for backward
5184 * compatibility.
Chris Mason3de45862008-11-17 21:02:50 -05005185 */
5186 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5187 location.objectid == root->root_key.objectid) {
5188 over = 0;
5189 goto skip;
5190 }
Al Viro9cdda8d2013-05-22 16:48:09 -04005191 over = !dir_emit(ctx, name_ptr, name_len,
5192 location.objectid, d_type);
Chris Mason5f39d392007-10-15 16:14:19 -04005193
Chris Mason3de45862008-11-17 21:02:50 -05005194skip:
Chris Mason5f39d392007-10-15 16:14:19 -04005195 if (name_ptr != tmp_name)
5196 kfree(name_ptr);
5197
Chris Mason39279cc2007-06-12 06:35:45 -04005198 if (over)
5199 goto nopos;
Josef Bacik5103e942007-11-16 11:45:54 -05005200 di_len = btrfs_dir_name_len(leaf, di) +
David Woodhouse49593bf2008-08-17 17:08:36 +01005201 btrfs_dir_data_len(leaf, di) + sizeof(*di);
Chris Mason39279cc2007-06-12 06:35:45 -04005202 di_cur += di_len;
5203 di = (struct btrfs_dir_item *)((char *)di + di_len);
5204 }
Li Zefanb9e03af2011-03-23 10:43:58 +08005205next:
5206 path->slots[0]++;
Chris Mason39279cc2007-06-12 06:35:45 -04005207 }
David Woodhouse49593bf2008-08-17 17:08:36 +01005208
Miao Xie16cdcec2011-04-22 18:12:22 +08005209 if (key_type == BTRFS_DIR_INDEX_KEY) {
5210 if (is_curr)
Al Viro9cdda8d2013-05-22 16:48:09 -04005211 ctx->pos++;
5212 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
Miao Xie16cdcec2011-04-22 18:12:22 +08005213 if (ret)
5214 goto nopos;
5215 }
5216
David Woodhouse49593bf2008-08-17 17:08:36 +01005217 /* Reached end of directory/root. Bump pos past the last item. */
Zach Browndb62efb2013-07-11 16:19:42 -07005218 ctx->pos++;
5219
5220 /*
5221 * Stop new entries from being returned after we return the last
5222 * entry.
5223 *
5224 * New directory entries are assigned a strictly increasing
5225 * offset. This means that new entries created during readdir
5226 * are *guaranteed* to be seen in the future by that readdir.
5227 * This has broken buggy programs which operate on names as
5228 * they're returned by readdir. Until we re-use freed offsets
5229 * we have this hack to stop new entries from being returned
5230 * under the assumption that they'll never reach this huge
5231 * offset.
5232 *
5233 * This is being careful not to overflow 32bit loff_t unless the
5234 * last entry requires it because doing so has broken 32bit apps
5235 * in the past.
5236 */
5237 if (key_type == BTRFS_DIR_INDEX_KEY) {
5238 if (ctx->pos >= INT_MAX)
5239 ctx->pos = LLONG_MAX;
5240 else
5241 ctx->pos = INT_MAX;
5242 }
Chris Mason39279cc2007-06-12 06:35:45 -04005243nopos:
5244 ret = 0;
5245err:
Miao Xie16cdcec2011-04-22 18:12:22 +08005246 if (key_type == BTRFS_DIR_INDEX_KEY)
5247 btrfs_put_delayed_items(&ins_list, &del_list);
Chris Mason39279cc2007-06-12 06:35:45 -04005248 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04005249 return ret;
5250}
5251
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005252int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Chris Mason39279cc2007-06-12 06:35:45 -04005253{
5254 struct btrfs_root *root = BTRFS_I(inode)->root;
5255 struct btrfs_trans_handle *trans;
5256 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04005257 bool nolock = false;
Chris Mason39279cc2007-06-12 06:35:45 -04005258
Josef Bacik72ac3c02012-05-23 14:13:11 -04005259 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Chris Mason4ca8b412008-08-05 13:30:48 -04005260 return 0;
5261
Liu Bo83eea1f2012-07-10 05:28:39 -06005262 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
Li Zefan82d59022011-04-20 10:33:24 +08005263 nolock = true;
Josef Bacik0af3d002010-06-21 14:48:16 -04005264
Christoph Hellwiga9185b42010-03-05 09:21:37 +01005265 if (wbc->sync_mode == WB_SYNC_ALL) {
Josef Bacik0af3d002010-06-21 14:48:16 -04005266 if (nolock)
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005267 trans = btrfs_join_transaction_nolock(root);
Josef Bacik0af3d002010-06-21 14:48:16 -04005268 else
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005269 trans = btrfs_join_transaction(root);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00005270 if (IS_ERR(trans))
5271 return PTR_ERR(trans);
Miao Xiea698d0752012-09-20 01:51:59 -06005272 ret = btrfs_commit_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04005273 }
5274 return ret;
5275}
5276
5277/*
Chris Mason54aa1f42007-06-22 14:16:25 -04005278 * This is somewhat expensive, updating the tree every time the
Chris Mason39279cc2007-06-12 06:35:45 -04005279 * inode changes. But, it is most likely to find the inode in cache.
5280 * FIXME, needs more benchmarking...there are no reasons other than performance
5281 * to keep or drop this code.
5282 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00005283static int btrfs_dirty_inode(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04005284{
5285 struct btrfs_root *root = BTRFS_I(inode)->root;
5286 struct btrfs_trans_handle *trans;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005287 int ret;
5288
Josef Bacik72ac3c02012-05-23 14:13:11 -04005289 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
Josef Bacik22c44fe2011-11-30 10:45:38 -05005290 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005291
Josef Bacik7a7eaa42011-04-13 12:54:33 -04005292 trans = btrfs_join_transaction(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005293 if (IS_ERR(trans))
5294 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005295
5296 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04005297 if (ret && ret == -ENOSPC) {
5298 /* whoops, lets try again with the full transaction */
5299 btrfs_end_transaction(trans, root);
5300 trans = btrfs_start_transaction(root, 1);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005301 if (IS_ERR(trans))
5302 return PTR_ERR(trans);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005303
Chris Mason94b60442010-05-26 11:02:00 -04005304 ret = btrfs_update_inode(trans, root, inode);
Chris Mason94b60442010-05-26 11:02:00 -04005305 }
Chris Mason39279cc2007-06-12 06:35:45 -04005306 btrfs_end_transaction(trans, root);
Miao Xie16cdcec2011-04-22 18:12:22 +08005307 if (BTRFS_I(inode)->delayed_node)
5308 btrfs_balance_delayed_items(root);
Josef Bacik22c44fe2011-11-30 10:45:38 -05005309
5310 return ret;
5311}
5312
5313/*
5314 * This is a copy of file_update_time. We need this so we can return error on
5315 * ENOSPC for updating the inode in the case of file write and mmap writes.
5316 */
Josef Bacike41f9412012-03-26 09:46:47 -04005317static int btrfs_update_time(struct inode *inode, struct timespec *now,
5318 int flags)
Josef Bacik22c44fe2011-11-30 10:45:38 -05005319{
Alexander Block2bc5565282012-06-15 09:49:33 +02005320 struct btrfs_root *root = BTRFS_I(inode)->root;
5321
5322 if (btrfs_root_readonly(root))
5323 return -EROFS;
5324
Josef Bacike41f9412012-03-26 09:46:47 -04005325 if (flags & S_VERSION)
Josef Bacik22c44fe2011-11-30 10:45:38 -05005326 inode_inc_iversion(inode);
Josef Bacike41f9412012-03-26 09:46:47 -04005327 if (flags & S_CTIME)
5328 inode->i_ctime = *now;
5329 if (flags & S_MTIME)
5330 inode->i_mtime = *now;
5331 if (flags & S_ATIME)
5332 inode->i_atime = *now;
5333 return btrfs_dirty_inode(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005334}
5335
Chris Masond352ac62008-09-29 15:18:18 -04005336/*
5337 * find the highest existing sequence number in a directory
5338 * and then set the in-memory index_cnt variable to reflect
5339 * free sequence numbers
5340 */
Josef Bacikaec74772008-07-24 12:12:38 -04005341static int btrfs_set_inode_index_count(struct inode *inode)
5342{
5343 struct btrfs_root *root = BTRFS_I(inode)->root;
5344 struct btrfs_key key, found_key;
5345 struct btrfs_path *path;
5346 struct extent_buffer *leaf;
5347 int ret;
5348
Li Zefan33345d012011-04-20 10:31:50 +08005349 key.objectid = btrfs_ino(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04005350 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5351 key.offset = (u64)-1;
5352
5353 path = btrfs_alloc_path();
5354 if (!path)
5355 return -ENOMEM;
5356
5357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5358 if (ret < 0)
5359 goto out;
5360 /* FIXME: we should be able to handle this */
5361 if (ret == 0)
5362 goto out;
5363 ret = 0;
5364
5365 /*
5366 * MAGIC NUMBER EXPLANATION:
5367 * since we search a directory based on f_pos we have to start at 2
5368 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5369 * else has to start at 2
5370 */
5371 if (path->slots[0] == 0) {
5372 BTRFS_I(inode)->index_cnt = 2;
5373 goto out;
5374 }
5375
5376 path->slots[0]--;
5377
5378 leaf = path->nodes[0];
5379 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5380
Li Zefan33345d012011-04-20 10:31:50 +08005381 if (found_key.objectid != btrfs_ino(inode) ||
Josef Bacikaec74772008-07-24 12:12:38 -04005382 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5383 BTRFS_I(inode)->index_cnt = 2;
5384 goto out;
5385 }
5386
5387 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5388out:
5389 btrfs_free_path(path);
5390 return ret;
5391}
5392
Chris Masond352ac62008-09-29 15:18:18 -04005393/*
5394 * helper to find a free sequence number in a given directory. This current
5395 * code is very simple, later versions will do smarter things in the btree
5396 */
Chris Mason3de45862008-11-17 21:02:50 -05005397int btrfs_set_inode_index(struct inode *dir, u64 *index)
Josef Bacikaec74772008-07-24 12:12:38 -04005398{
5399 int ret = 0;
5400
5401 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
Miao Xie16cdcec2011-04-22 18:12:22 +08005402 ret = btrfs_inode_delayed_dir_index_count(dir);
5403 if (ret) {
5404 ret = btrfs_set_inode_index_count(dir);
5405 if (ret)
5406 return ret;
5407 }
Josef Bacikaec74772008-07-24 12:12:38 -04005408 }
5409
Chris Mason00e4e6b2008-08-05 11:18:09 -04005410 *index = BTRFS_I(dir)->index_cnt;
Josef Bacikaec74772008-07-24 12:12:38 -04005411 BTRFS_I(dir)->index_cnt++;
5412
5413 return ret;
5414}
5415
Chris Mason39279cc2007-06-12 06:35:45 -04005416static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5417 struct btrfs_root *root,
Josef Bacikaec74772008-07-24 12:12:38 -04005418 struct inode *dir,
Chris Mason9c583092008-01-29 15:15:18 -05005419 const char *name, int name_len,
Al Viro175a4eb2011-07-26 03:30:54 -04005420 u64 ref_objectid, u64 objectid,
5421 umode_t mode, u64 *index)
Chris Mason39279cc2007-06-12 06:35:45 -04005422{
5423 struct inode *inode;
Chris Mason5f39d392007-10-15 16:14:19 -04005424 struct btrfs_inode_item *inode_item;
Chris Mason39279cc2007-06-12 06:35:45 -04005425 struct btrfs_key *location;
Chris Mason5f39d392007-10-15 16:14:19 -04005426 struct btrfs_path *path;
Chris Mason9c583092008-01-29 15:15:18 -05005427 struct btrfs_inode_ref *ref;
5428 struct btrfs_key key[2];
5429 u32 sizes[2];
5430 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04005431 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005432
Chris Mason5f39d392007-10-15 16:14:19 -04005433 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07005434 if (!path)
5435 return ERR_PTR(-ENOMEM);
Chris Mason5f39d392007-10-15 16:14:19 -04005436
Chris Mason39279cc2007-06-12 06:35:45 -04005437 inode = new_inode(root->fs_info->sb);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00005438 if (!inode) {
5439 btrfs_free_path(path);
Chris Mason39279cc2007-06-12 06:35:45 -04005440 return ERR_PTR(-ENOMEM);
Yoshinori Sano8fb27642011-04-09 02:30:07 +00005441 }
Chris Mason39279cc2007-06-12 06:35:45 -04005442
Li Zefan581bb052011-04-20 10:06:11 +08005443 /*
5444 * we have to initialize this early, so we can reclaim the inode
5445 * number if we fail afterwards in this function.
5446 */
5447 inode->i_ino = objectid;
5448
Josef Bacikaec74772008-07-24 12:12:38 -04005449 if (dir) {
liubo1abe9b82011-03-24 11:18:59 +00005450 trace_btrfs_inode_request(dir);
5451
Chris Mason3de45862008-11-17 21:02:50 -05005452 ret = btrfs_set_inode_index(dir, index);
Shen Feng09771432009-04-02 16:46:06 -04005453 if (ret) {
Yoshinori Sano8fb27642011-04-09 02:30:07 +00005454 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04005455 iput(inode);
Josef Bacikaec74772008-07-24 12:12:38 -04005456 return ERR_PTR(ret);
Shen Feng09771432009-04-02 16:46:06 -04005457 }
Josef Bacikaec74772008-07-24 12:12:38 -04005458 }
5459 /*
5460 * index_cnt is ignored for everything but a dir,
5461 * btrfs_get_inode_index_count has an explanation for the magic
5462 * number
5463 */
5464 BTRFS_I(inode)->index_cnt = 2;
Chris Mason39279cc2007-06-12 06:35:45 -04005465 BTRFS_I(inode)->root = root;
Chris Masone02119d2008-09-05 16:13:11 -04005466 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik76195852010-11-19 02:18:02 +00005467 inode->i_generation = BTRFS_I(inode)->generation;
Chris Masonb888db22007-08-27 16:49:44 -04005468
Josef Bacik5dc562c2012-08-17 13:14:17 -04005469 /*
5470 * We could have gotten an inode number from somebody who was fsynced
5471 * and then removed in this same transaction, so let's just set full
5472 * sync since it will be a full sync anyway and this will blow away the
5473 * old info in the log.
5474 */
5475 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5476
Chris Mason9c583092008-01-29 15:15:18 -05005477 key[0].objectid = objectid;
5478 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5479 key[0].offset = 0;
5480
Mark Fashehf1863732012-08-08 11:32:27 -07005481 /*
5482 * Start new inodes with an inode_ref. This is slightly more
5483 * efficient for small numbers of hard links since they will
5484 * be packed into one item. Extended refs will kick in if we
5485 * add more hard links than can fit in the ref item.
5486 */
Chris Mason9c583092008-01-29 15:15:18 -05005487 key[1].objectid = objectid;
5488 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5489 key[1].offset = ref_objectid;
5490
5491 sizes[0] = sizeof(struct btrfs_inode_item);
5492 sizes[1] = name_len + sizeof(*ref);
5493
Chris Masonb9473432009-03-13 11:00:37 -04005494 path->leave_spinning = 1;
Chris Mason9c583092008-01-29 15:15:18 -05005495 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5496 if (ret != 0)
Chris Mason5f39d392007-10-15 16:14:19 -04005497 goto fail;
5498
Dmitry Monakhovecc11fab2010-03-04 17:31:47 +03005499 inode_init_owner(inode, dir, mode);
Yan Zhenga76a3cd2008-10-09 11:46:29 -04005500 inode_set_bytes(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04005501 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
Chris Mason5f39d392007-10-15 16:14:19 -04005502 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5503 struct btrfs_inode_item);
Li Zefan293f7e02012-07-10 00:58:58 -06005504 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5505 sizeof(*inode_item));
Chris Masone02119d2008-09-05 16:13:11 -04005506 fill_inode_item(trans, path->nodes[0], inode_item, inode);
Chris Mason9c583092008-01-29 15:15:18 -05005507
5508 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5509 struct btrfs_inode_ref);
5510 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
Chris Mason00e4e6b2008-08-05 11:18:09 -04005511 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
Chris Mason9c583092008-01-29 15:15:18 -05005512 ptr = (unsigned long)(ref + 1);
5513 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5514
Chris Mason5f39d392007-10-15 16:14:19 -04005515 btrfs_mark_buffer_dirty(path->nodes[0]);
5516 btrfs_free_path(path);
5517
Chris Mason39279cc2007-06-12 06:35:45 -04005518 location = &BTRFS_I(inode)->location;
5519 location->objectid = objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04005520 location->offset = 0;
5521 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5522
Christoph Hellwig6cbff002009-04-17 10:37:41 +02005523 btrfs_inherit_iflags(inode, dir);
5524
Al Viro569254b2011-07-24 17:08:40 -04005525 if (S_ISREG(mode)) {
Chris Mason94272162009-07-02 12:26:06 -04005526 if (btrfs_test_opt(root, NODATASUM))
5527 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
Liu Bo213490b2012-09-11 08:33:50 -06005528 if (btrfs_test_opt(root, NODATACOW))
Josef Bacikf2bdf9a2013-02-21 15:28:28 -05005529 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5530 BTRFS_INODE_NODATASUM;
Chris Mason94272162009-07-02 12:26:06 -04005531 }
5532
Filipe David Borba Manana778ba822013-10-06 22:22:33 +01005533 btrfs_insert_inode_hash(inode);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005534 inode_tree_add(inode);
liubo1abe9b82011-03-24 11:18:59 +00005535
5536 trace_btrfs_inode_new(inode);
Chris Mason1973f0f2011-06-24 13:13:29 -04005537 btrfs_set_inode_last_trans(trans, inode);
liubo1abe9b82011-03-24 11:18:59 +00005538
Alexander Block8ea05e32012-07-25 17:35:53 +02005539 btrfs_update_root_times(trans, root);
5540
Chris Mason39279cc2007-06-12 06:35:45 -04005541 return inode;
Chris Mason5f39d392007-10-15 16:14:19 -04005542fail:
Josef Bacikaec74772008-07-24 12:12:38 -04005543 if (dir)
5544 BTRFS_I(dir)->index_cnt--;
Chris Mason5f39d392007-10-15 16:14:19 -04005545 btrfs_free_path(path);
Shen Feng09771432009-04-02 16:46:06 -04005546 iput(inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005547 return ERR_PTR(ret);
Chris Mason39279cc2007-06-12 06:35:45 -04005548}
5549
5550static inline u8 btrfs_inode_type(struct inode *inode)
5551{
5552 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5553}
5554
Chris Masond352ac62008-09-29 15:18:18 -04005555/*
5556 * utility function to add 'inode' into 'parent_inode' with
5557 * a give name and a given sequence number.
5558 * if 'add_backref' is true, also insert a backref from the
5559 * inode to the parent directory.
5560 */
Chris Masone02119d2008-09-05 16:13:11 -04005561int btrfs_add_link(struct btrfs_trans_handle *trans,
5562 struct inode *parent_inode, struct inode *inode,
5563 const char *name, int name_len, int add_backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04005564{
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005565 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005566 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04005567 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
Li Zefan33345d012011-04-20 10:31:50 +08005568 u64 ino = btrfs_ino(inode);
5569 u64 parent_ino = btrfs_ino(parent_inode);
Chris Mason5f39d392007-10-15 16:14:19 -04005570
Li Zefan33345d012011-04-20 10:31:50 +08005571 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005572 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5573 } else {
Li Zefan33345d012011-04-20 10:31:50 +08005574 key.objectid = ino;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005575 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5576 key.offset = 0;
5577 }
Chris Mason39279cc2007-06-12 06:35:45 -04005578
Li Zefan33345d012011-04-20 10:31:50 +08005579 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005580 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5581 key.objectid, root->root_key.objectid,
Li Zefan33345d012011-04-20 10:31:50 +08005582 parent_ino, index, name, name_len);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005583 } else if (add_backref) {
Li Zefan33345d012011-04-20 10:31:50 +08005584 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5585 parent_ino, index);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005586 }
5587
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005588 /* Nothing to clean up yet */
5589 if (ret)
5590 return ret;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04005591
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005592 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5593 parent_inode, &key,
5594 btrfs_inode_type(inode), index);
Chris Mason9c520572012-12-17 14:26:57 -05005595 if (ret == -EEXIST || ret == -EOVERFLOW)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005596 goto fail_dir_item;
5597 else if (ret) {
5598 btrfs_abort_transaction(trans, root, ret);
5599 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005600 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005601
5602 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5603 name_len * 2);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04005604 inode_inc_iversion(parent_inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005605 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5606 ret = btrfs_update_inode(trans, root, parent_inode);
5607 if (ret)
5608 btrfs_abort_transaction(trans, root, ret);
Chris Mason39279cc2007-06-12 06:35:45 -04005609 return ret;
Chris Masonfe66a052012-02-20 08:40:56 -05005610
5611fail_dir_item:
5612 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5613 u64 local_index;
5614 int err;
5615 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5616 key.objectid, root->root_key.objectid,
5617 parent_ino, &local_index, name, name_len);
5618
5619 } else if (add_backref) {
5620 u64 local_index;
5621 int err;
5622
5623 err = btrfs_del_inode_ref(trans, root, name, name_len,
5624 ino, parent_ino, &local_index);
5625 }
5626 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04005627}
5628
5629static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
Josef Bacika1b075d2010-11-19 20:36:11 +00005630 struct inode *dir, struct dentry *dentry,
5631 struct inode *inode, int backref, u64 index)
Chris Mason39279cc2007-06-12 06:35:45 -04005632{
Josef Bacika1b075d2010-11-19 20:36:11 +00005633 int err = btrfs_add_link(trans, dir, inode,
5634 dentry->d_name.name, dentry->d_name.len,
5635 backref, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005636 if (err > 0)
5637 err = -EEXIST;
5638 return err;
5639}
5640
Josef Bacik618e21d2007-07-11 10:18:17 -04005641static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04005642 umode_t mode, dev_t rdev)
Josef Bacik618e21d2007-07-11 10:18:17 -04005643{
5644 struct btrfs_trans_handle *trans;
5645 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05005646 struct inode *inode = NULL;
Josef Bacik618e21d2007-07-11 10:18:17 -04005647 int err;
5648 int drop_inode = 0;
5649 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005650 u64 index = 0;
Josef Bacik618e21d2007-07-11 10:18:17 -04005651
5652 if (!new_valid_dev(rdev))
5653 return -EINVAL;
5654
Josef Bacik9ed74f22009-09-11 16:12:44 -04005655 /*
5656 * 2 for inode item and ref
5657 * 2 for dir items
5658 * 1 for xattr if selinux is on
5659 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04005660 trans = btrfs_start_transaction(root, 5);
5661 if (IS_ERR(trans))
5662 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05005663
Li Zefan581bb052011-04-20 10:06:11 +08005664 err = btrfs_find_free_ino(root, &objectid);
5665 if (err)
5666 goto out_unlock;
5667
Josef Bacikaec74772008-07-24 12:12:38 -04005668 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08005669 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04005670 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005671 if (IS_ERR(inode)) {
5672 err = PTR_ERR(inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04005673 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005674 }
Josef Bacik618e21d2007-07-11 10:18:17 -04005675
Eric Paris2a7dba32011-02-01 11:05:39 -05005676 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04005677 if (err) {
5678 drop_inode = 1;
5679 goto out_unlock;
5680 }
5681
Casey Schauflerad19db72011-12-15 10:09:07 -05005682 /*
5683 * If the active LSM wants to access the inode during
5684 * d_instantiate it needs these. Smack checks to see
5685 * if the filesystem supports xattrs by looking at the
5686 * ops vector.
5687 */
5688
5689 inode->i_op = &btrfs_special_inode_operations;
Josef Bacika1b075d2010-11-19 20:36:11 +00005690 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Josef Bacik618e21d2007-07-11 10:18:17 -04005691 if (err)
5692 drop_inode = 1;
5693 else {
Josef Bacik618e21d2007-07-11 10:18:17 -04005694 init_special_inode(inode, inode->i_mode, rdev);
Yan1b4ab1b2007-08-29 09:11:44 -04005695 btrfs_update_inode(trans, root, inode);
Al Viro08c422c2011-12-23 07:58:13 -05005696 d_instantiate(dentry, inode);
Josef Bacik618e21d2007-07-11 10:18:17 -04005697 }
Josef Bacik618e21d2007-07-11 10:18:17 -04005698out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005699 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00005700 btrfs_btree_balance_dirty(root);
Josef Bacik618e21d2007-07-11 10:18:17 -04005701 if (drop_inode) {
5702 inode_dec_link_count(inode);
5703 iput(inode);
5704 }
Josef Bacik618e21d2007-07-11 10:18:17 -04005705 return err;
5706}
5707
Chris Mason39279cc2007-06-12 06:35:45 -04005708static int btrfs_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -04005709 umode_t mode, bool excl)
Chris Mason39279cc2007-06-12 06:35:45 -04005710{
5711 struct btrfs_trans_handle *trans;
5712 struct btrfs_root *root = BTRFS_I(dir)->root;
Chris Mason1832a6d2007-12-21 16:27:21 -05005713 struct inode *inode = NULL;
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005714 int drop_inode_on_err = 0;
Yan, Zhenga22285a2010-05-16 10:48:46 -04005715 int err;
Chris Mason39279cc2007-06-12 06:35:45 -04005716 u64 objectid;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005717 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005718
Josef Bacik9ed74f22009-09-11 16:12:44 -04005719 /*
5720 * 2 for inode item and ref
5721 * 2 for dir items
5722 * 1 for xattr if selinux is on
5723 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04005724 trans = btrfs_start_transaction(root, 5);
5725 if (IS_ERR(trans))
5726 return PTR_ERR(trans);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005727
Li Zefan581bb052011-04-20 10:06:11 +08005728 err = btrfs_find_free_ino(root, &objectid);
5729 if (err)
5730 goto out_unlock;
5731
Josef Bacikaec74772008-07-24 12:12:38 -04005732 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08005733 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04005734 mode, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005735 if (IS_ERR(inode)) {
5736 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04005737 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04005738 }
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005739 drop_inode_on_err = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04005740
Eric Paris2a7dba32011-02-01 11:05:39 -05005741 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005742 if (err)
Josef Bacik33268ea2008-07-24 12:16:36 -04005743 goto out_unlock;
Josef Bacik33268ea2008-07-24 12:16:36 -04005744
Filipe Brandenburger9185aa52012-11-30 03:40:08 +00005745 err = btrfs_update_inode(trans, root, inode);
5746 if (err)
5747 goto out_unlock;
5748
Casey Schauflerad19db72011-12-15 10:09:07 -05005749 /*
5750 * If the active LSM wants to access the inode during
5751 * d_instantiate it needs these. Smack checks to see
5752 * if the filesystem supports xattrs by looking at the
5753 * ops vector.
5754 */
5755 inode->i_fop = &btrfs_file_operations;
5756 inode->i_op = &btrfs_file_inode_operations;
5757
Josef Bacika1b075d2010-11-19 20:36:11 +00005758 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005759 if (err)
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005760 goto out_unlock;
5761
5762 inode->i_mapping->a_ops = &btrfs_aops;
5763 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5764 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5765 d_instantiate(dentry, inode);
5766
Chris Mason39279cc2007-06-12 06:35:45 -04005767out_unlock:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005768 btrfs_end_transaction(trans, root);
Filipe Brandenburger43baa572012-11-30 03:40:09 +00005769 if (err && drop_inode_on_err) {
Chris Mason39279cc2007-06-12 06:35:45 -04005770 inode_dec_link_count(inode);
5771 iput(inode);
5772 }
Liu Bob53d3f52012-11-14 14:34:34 +00005773 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005774 return err;
5775}
5776
5777static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5778 struct dentry *dentry)
5779{
5780 struct btrfs_trans_handle *trans;
5781 struct btrfs_root *root = BTRFS_I(dir)->root;
5782 struct inode *inode = old_dentry->d_inode;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005783 u64 index;
Chris Mason39279cc2007-06-12 06:35:45 -04005784 int err;
5785 int drop_inode = 0;
5786
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00005787 /* do not allow sys_link's with other subvols of the same device */
5788 if (root->objectid != BTRFS_I(inode)->root->objectid)
Mark Fasheh3ab35642011-03-22 17:20:26 +00005789 return -EXDEV;
TARUISI Hiroaki4a8be422009-11-12 07:14:26 +00005790
Mark Fashehf1863732012-08-08 11:32:27 -07005791 if (inode->i_nlink >= BTRFS_LINK_MAX)
Al Viroc055e992011-03-04 17:15:18 +00005792 return -EMLINK;
Josef Bacik9ed74f22009-09-11 16:12:44 -04005793
Chris Mason3de45862008-11-17 21:02:50 -05005794 err = btrfs_set_inode_index(dir, &index);
Josef Bacikaec74772008-07-24 12:12:38 -04005795 if (err)
5796 goto fail;
5797
Yan, Zhenga22285a2010-05-16 10:48:46 -04005798 /*
Miao Xie7e6b6462011-02-18 09:21:17 +00005799 * 2 items for inode and inode ref
Yan, Zhenga22285a2010-05-16 10:48:46 -04005800 * 2 items for dir items
Miao Xie7e6b6462011-02-18 09:21:17 +00005801 * 1 item for parent inode
Yan, Zhenga22285a2010-05-16 10:48:46 -04005802 */
Miao Xie7e6b6462011-02-18 09:21:17 +00005803 trans = btrfs_start_transaction(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04005804 if (IS_ERR(trans)) {
5805 err = PTR_ERR(trans);
5806 goto fail;
5807 }
Chris Mason5f39d392007-10-15 16:14:19 -04005808
Zach Brown8b558c52013-10-16 12:10:34 -07005809 inc_nlink(inode);
Josef Bacik0c4d2d92012-04-05 15:03:02 -04005810 inode_inc_iversion(inode);
Miao Xie31534952011-04-13 13:19:21 +08005811 inode->i_ctime = CURRENT_TIME;
Al Viro7de9c6ee2010-10-23 11:11:40 -04005812 ihold(inode);
Josef Bacike9976152012-10-11 15:53:56 -04005813 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
Josef Bacikaec74772008-07-24 12:12:38 -04005814
Josef Bacika1b075d2010-11-19 20:36:11 +00005815 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
Chris Mason5f39d392007-10-15 16:14:19 -04005816
Yan, Zhenga5719522009-09-24 09:17:31 -04005817 if (err) {
Chris Mason39279cc2007-06-12 06:35:45 -04005818 drop_inode = 1;
Yan, Zhenga5719522009-09-24 09:17:31 -04005819 } else {
Al Viro10d9f302011-07-16 23:09:10 -04005820 struct dentry *parent = dentry->d_parent;
Yan, Zhenga5719522009-09-24 09:17:31 -04005821 err = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005822 if (err)
5823 goto fail;
Al Viro08c422c2011-12-23 07:58:13 -05005824 d_instantiate(dentry, inode);
Josef Bacik6a912212010-11-20 09:48:00 +00005825 btrfs_log_new_name(trans, inode, NULL, parent);
Yan, Zhenga5719522009-09-24 09:17:31 -04005826 }
Chris Mason39279cc2007-06-12 06:35:45 -04005827
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005828 btrfs_end_transaction(trans, root);
Chris Mason1832a6d2007-12-21 16:27:21 -05005829fail:
Chris Mason39279cc2007-06-12 06:35:45 -04005830 if (drop_inode) {
5831 inode_dec_link_count(inode);
5832 iput(inode);
5833 }
Liu Bob53d3f52012-11-14 14:34:34 +00005834 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005835 return err;
5836}
5837
Al Viro18bb1db2011-07-26 01:41:39 -04005838static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Chris Mason39279cc2007-06-12 06:35:45 -04005839{
Chris Masonb9d86662008-05-02 16:13:49 -04005840 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04005841 struct btrfs_trans_handle *trans;
5842 struct btrfs_root *root = BTRFS_I(dir)->root;
5843 int err = 0;
5844 int drop_on_err = 0;
Chris Masonb9d86662008-05-02 16:13:49 -04005845 u64 objectid = 0;
Chris Mason00e4e6b2008-08-05 11:18:09 -04005846 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005847
Josef Bacik9ed74f22009-09-11 16:12:44 -04005848 /*
5849 * 2 items for inode and ref
5850 * 2 items for dir items
5851 * 1 for xattr if selinux is on
5852 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04005853 trans = btrfs_start_transaction(root, 5);
5854 if (IS_ERR(trans))
5855 return PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04005856
Li Zefan581bb052011-04-20 10:06:11 +08005857 err = btrfs_find_free_ino(root, &objectid);
5858 if (err)
5859 goto out_fail;
5860
Josef Bacikaec74772008-07-24 12:12:38 -04005861 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08005862 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04005863 S_IFDIR | mode, &index);
Chris Mason39279cc2007-06-12 06:35:45 -04005864 if (IS_ERR(inode)) {
5865 err = PTR_ERR(inode);
5866 goto out_fail;
5867 }
Chris Mason5f39d392007-10-15 16:14:19 -04005868
Chris Mason39279cc2007-06-12 06:35:45 -04005869 drop_on_err = 1;
Josef Bacik33268ea2008-07-24 12:16:36 -04005870
Eric Paris2a7dba32011-02-01 11:05:39 -05005871 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04005872 if (err)
5873 goto out_fail;
5874
Chris Mason39279cc2007-06-12 06:35:45 -04005875 inode->i_op = &btrfs_dir_inode_operations;
5876 inode->i_fop = &btrfs_dir_file_operations;
Chris Mason39279cc2007-06-12 06:35:45 -04005877
Chris Masondbe674a2008-07-17 12:54:05 -04005878 btrfs_i_size_write(inode, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04005879 err = btrfs_update_inode(trans, root, inode);
5880 if (err)
5881 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04005882
Josef Bacika1b075d2010-11-19 20:36:11 +00005883 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5884 dentry->d_name.len, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04005885 if (err)
5886 goto out_fail;
Chris Mason5f39d392007-10-15 16:14:19 -04005887
Chris Mason39279cc2007-06-12 06:35:45 -04005888 d_instantiate(dentry, inode);
5889 drop_on_err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04005890
5891out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05005892 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04005893 if (drop_on_err)
5894 iput(inode);
Liu Bob53d3f52012-11-14 14:34:34 +00005895 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04005896 return err;
5897}
5898
Chris Masond352ac62008-09-29 15:18:18 -04005899/* helper for btfs_get_extent. Given an existing extent in the tree,
5900 * and an extent that you want to insert, deal with overlap and insert
5901 * the new extent into the tree.
5902 */
Chris Mason3b951512008-04-17 11:29:12 -04005903static int merge_extent_mapping(struct extent_map_tree *em_tree,
5904 struct extent_map *existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04005905 struct extent_map *em,
5906 u64 map_start, u64 map_len)
Chris Mason3b951512008-04-17 11:29:12 -04005907{
5908 u64 start_diff;
Chris Mason3b951512008-04-17 11:29:12 -04005909
Chris Masone6dcd2d2008-07-17 12:53:50 -04005910 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5911 start_diff = map_start - em->start;
5912 em->start = map_start;
5913 em->len = map_len;
Chris Masonc8b97812008-10-29 14:49:59 -04005914 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5915 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04005916 em->block_start += start_diff;
Chris Masonc8b97812008-10-29 14:49:59 -04005917 em->block_len -= start_diff;
5918 }
Josef Bacik09a2a8f92013-04-05 16:51:15 -04005919 return add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04005920}
5921
Chris Masonc8b97812008-10-29 14:49:59 -04005922static noinline int uncompress_inline(struct btrfs_path *path,
5923 struct inode *inode, struct page *page,
5924 size_t pg_offset, u64 extent_offset,
5925 struct btrfs_file_extent_item *item)
5926{
5927 int ret;
5928 struct extent_buffer *leaf = path->nodes[0];
5929 char *tmp;
5930 size_t max_size;
5931 unsigned long inline_size;
5932 unsigned long ptr;
Li Zefan261507a02010-12-17 14:21:50 +08005933 int compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04005934
5935 WARN_ON(pg_offset != 0);
Li Zefan261507a02010-12-17 14:21:50 +08005936 compress_type = btrfs_file_extent_compression(leaf, item);
Chris Masonc8b97812008-10-29 14:49:59 -04005937 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5938 inline_size = btrfs_file_extent_inline_item_len(leaf,
Ross Kirkdd3cc162013-09-16 15:58:09 +01005939 btrfs_item_nr(path->slots[0]));
Chris Masonc8b97812008-10-29 14:49:59 -04005940 tmp = kmalloc(inline_size, GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04005941 if (!tmp)
5942 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -04005943 ptr = btrfs_file_extent_inline_start(item);
5944
5945 read_extent_buffer(leaf, tmp, ptr, inline_size);
5946
Chris Mason5b050f02008-11-11 09:34:41 -05005947 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
Li Zefan261507a02010-12-17 14:21:50 +08005948 ret = btrfs_decompress(compress_type, tmp, page,
5949 extent_offset, inline_size, max_size);
Chris Masonc8b97812008-10-29 14:49:59 -04005950 if (ret) {
Cong Wang7ac687d2011-11-25 23:14:28 +08005951 char *kaddr = kmap_atomic(page);
Chris Masonc8b97812008-10-29 14:49:59 -04005952 unsigned long copy_size = min_t(u64,
5953 PAGE_CACHE_SIZE - pg_offset,
5954 max_size - extent_offset);
5955 memset(kaddr + pg_offset, 0, copy_size);
Cong Wang7ac687d2011-11-25 23:14:28 +08005956 kunmap_atomic(kaddr);
Chris Masonc8b97812008-10-29 14:49:59 -04005957 }
5958 kfree(tmp);
5959 return 0;
5960}
5961
Chris Masond352ac62008-09-29 15:18:18 -04005962/*
5963 * a bit scary, this does extent mapping from logical file offset to the disk.
Chris Masond3977122009-01-05 21:25:51 -05005964 * the ugly parts come from merging extents from the disk with the in-ram
5965 * representation. This gets more complex because of the data=ordered code,
Chris Masond352ac62008-09-29 15:18:18 -04005966 * where the in-ram extents might be locked pending data=ordered completion.
5967 *
5968 * This also copies inline extents directly into the page.
5969 */
Chris Masond3977122009-01-05 21:25:51 -05005970
Chris Masona52d9a82007-08-27 16:49:44 -04005971struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
Chris Mason70dec802008-01-29 09:59:12 -05005972 size_t pg_offset, u64 start, u64 len,
Chris Masona52d9a82007-08-27 16:49:44 -04005973 int create)
5974{
5975 int ret;
5976 int err = 0;
Chris Masondb945352007-10-15 16:15:53 -04005977 u64 bytenr;
Chris Masona52d9a82007-08-27 16:49:44 -04005978 u64 extent_start = 0;
5979 u64 extent_end = 0;
Li Zefan33345d012011-04-20 10:31:50 +08005980 u64 objectid = btrfs_ino(inode);
Chris Masona52d9a82007-08-27 16:49:44 -04005981 u32 found_type;
Chris Masonf4219502008-07-22 11:18:09 -04005982 struct btrfs_path *path = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -04005983 struct btrfs_root *root = BTRFS_I(inode)->root;
5984 struct btrfs_file_extent_item *item;
Chris Mason5f39d392007-10-15 16:14:19 -04005985 struct extent_buffer *leaf;
5986 struct btrfs_key found_key;
Chris Masona52d9a82007-08-27 16:49:44 -04005987 struct extent_map *em = NULL;
5988 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Chris Masond1310b22008-01-24 16:13:08 -05005989 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04005990 struct btrfs_trans_handle *trans = NULL;
Li Zefan261507a02010-12-17 14:21:50 +08005991 int compress_type;
Chris Masona52d9a82007-08-27 16:49:44 -04005992
Chris Masona52d9a82007-08-27 16:49:44 -04005993again:
Chris Mason890871b2009-09-02 16:24:52 -04005994 read_lock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05005995 em = lookup_extent_mapping(em_tree, start, len);
Chris Masona061fc82008-05-07 11:43:44 -04005996 if (em)
5997 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Mason890871b2009-09-02 16:24:52 -04005998 read_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -05005999
Chris Masona52d9a82007-08-27 16:49:44 -04006000 if (em) {
Chris Masone1c4b742008-04-22 13:26:46 -04006001 if (em->start > start || em->start + em->len <= start)
6002 free_extent_map(em);
6003 else if (em->block_start == EXTENT_MAP_INLINE && page)
Chris Mason70dec802008-01-29 09:59:12 -05006004 free_extent_map(em);
6005 else
6006 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006007 }
David Sterba172ddd62011-04-21 00:48:27 +02006008 em = alloc_extent_map();
Chris Masona52d9a82007-08-27 16:49:44 -04006009 if (!em) {
Chris Masond1310b22008-01-24 16:13:08 -05006010 err = -ENOMEM;
6011 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006012 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006013 em->bdev = root->fs_info->fs_devices->latest_bdev;
Chris Masond1310b22008-01-24 16:13:08 -05006014 em->start = EXTENT_MAP_HOLE;
Chris Mason445a6942008-11-10 11:53:33 -05006015 em->orig_start = EXTENT_MAP_HOLE;
Chris Masond1310b22008-01-24 16:13:08 -05006016 em->len = (u64)-1;
Chris Masonc8b97812008-10-29 14:49:59 -04006017 em->block_len = (u64)-1;
Chris Masonf4219502008-07-22 11:18:09 -04006018
6019 if (!path) {
6020 path = btrfs_alloc_path();
Josef Bacik026fd312011-05-13 10:32:11 -04006021 if (!path) {
6022 err = -ENOMEM;
6023 goto out;
6024 }
6025 /*
6026 * Chances are we'll be called again, so go ahead and do
6027 * readahead
6028 */
6029 path->reada = 1;
Chris Masonf4219502008-07-22 11:18:09 -04006030 }
6031
Chris Mason179e29e2007-11-01 11:28:41 -04006032 ret = btrfs_lookup_file_extent(trans, root, path,
6033 objectid, start, trans != NULL);
Chris Masona52d9a82007-08-27 16:49:44 -04006034 if (ret < 0) {
6035 err = ret;
6036 goto out;
6037 }
6038
6039 if (ret != 0) {
6040 if (path->slots[0] == 0)
6041 goto not_found;
6042 path->slots[0]--;
6043 }
6044
Chris Mason5f39d392007-10-15 16:14:19 -04006045 leaf = path->nodes[0];
6046 item = btrfs_item_ptr(leaf, path->slots[0],
Chris Masona52d9a82007-08-27 16:49:44 -04006047 struct btrfs_file_extent_item);
Chris Masona52d9a82007-08-27 16:49:44 -04006048 /* are we inside the extent that was found? */
Chris Mason5f39d392007-10-15 16:14:19 -04006049 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6050 found_type = btrfs_key_type(&found_key);
6051 if (found_key.objectid != objectid ||
Chris Masona52d9a82007-08-27 16:49:44 -04006052 found_type != BTRFS_EXTENT_DATA_KEY) {
Josef Bacik25a50342013-10-14 12:08:38 -04006053 /*
6054 * If we backup past the first extent we want to move forward
6055 * and see if there is an extent in front of us, otherwise we'll
6056 * say there is a hole for our whole search range which can
6057 * cause problems.
6058 */
6059 extent_end = start;
6060 goto next;
Chris Masona52d9a82007-08-27 16:49:44 -04006061 }
6062
Chris Mason5f39d392007-10-15 16:14:19 -04006063 found_type = btrfs_file_extent_type(leaf, item);
6064 extent_start = found_key.offset;
Li Zefan261507a02010-12-17 14:21:50 +08006065 compress_type = btrfs_file_extent_compression(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04006066 if (found_type == BTRFS_FILE_EXTENT_REG ||
6067 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Masona52d9a82007-08-27 16:49:44 -04006068 extent_end = extent_start +
Chris Masondb945352007-10-15 16:15:53 -04006069 btrfs_file_extent_num_bytes(leaf, item);
Yan Zheng9036c102008-10-30 14:19:41 -04006070 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6071 size_t size;
6072 size = btrfs_file_extent_inline_len(leaf, item);
Qu Wenruofda28322013-02-26 08:10:22 +00006073 extent_end = ALIGN(extent_start + size, root->sectorsize);
Yan Zheng9036c102008-10-30 14:19:41 -04006074 }
Josef Bacik25a50342013-10-14 12:08:38 -04006075next:
Yan Zheng9036c102008-10-30 14:19:41 -04006076 if (start >= extent_end) {
6077 path->slots[0]++;
6078 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6079 ret = btrfs_next_leaf(root, path);
6080 if (ret < 0) {
6081 err = ret;
6082 goto out;
Chris Masona52d9a82007-08-27 16:49:44 -04006083 }
Yan Zheng9036c102008-10-30 14:19:41 -04006084 if (ret > 0)
6085 goto not_found;
6086 leaf = path->nodes[0];
Chris Masona52d9a82007-08-27 16:49:44 -04006087 }
Yan Zheng9036c102008-10-30 14:19:41 -04006088 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6089 if (found_key.objectid != objectid ||
6090 found_key.type != BTRFS_EXTENT_DATA_KEY)
6091 goto not_found;
6092 if (start + len <= found_key.offset)
6093 goto not_found;
6094 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04006095 em->orig_start = start;
Yan Zheng9036c102008-10-30 14:19:41 -04006096 em->len = found_key.offset - start;
6097 goto not_found_em;
6098 }
6099
Josef Bacikcc95bef2013-04-04 14:31:27 -04006100 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
Yan Zhengd899e052008-10-30 14:25:28 -04006101 if (found_type == BTRFS_FILE_EXTENT_REG ||
6102 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng9036c102008-10-30 14:19:41 -04006103 em->start = extent_start;
6104 em->len = extent_end - extent_start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -05006105 em->orig_start = extent_start -
6106 btrfs_file_extent_offset(leaf, item);
Josef Bacikb4939682012-12-03 10:31:19 -05006107 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6108 item);
Chris Masondb945352007-10-15 16:15:53 -04006109 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6110 if (bytenr == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04006111 em->block_start = EXTENT_MAP_HOLE;
Chris Masona52d9a82007-08-27 16:49:44 -04006112 goto insert;
6113 }
Li Zefan261507a02010-12-17 14:21:50 +08006114 if (compress_type != BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04006115 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08006116 em->compress_type = compress_type;
Chris Masonc8b97812008-10-29 14:49:59 -04006117 em->block_start = bytenr;
Josef Bacikb4939682012-12-03 10:31:19 -05006118 em->block_len = em->orig_block_len;
Chris Masonc8b97812008-10-29 14:49:59 -04006119 } else {
6120 bytenr += btrfs_file_extent_offset(leaf, item);
6121 em->block_start = bytenr;
6122 em->block_len = em->len;
Yan Zhengd899e052008-10-30 14:25:28 -04006123 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6124 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonc8b97812008-10-29 14:49:59 -04006125 }
Chris Masona52d9a82007-08-27 16:49:44 -04006126 goto insert;
6127 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason5f39d392007-10-15 16:14:19 -04006128 unsigned long ptr;
Chris Masona52d9a82007-08-27 16:49:44 -04006129 char *map;
Chris Mason3326d1b2007-10-15 16:18:25 -04006130 size_t size;
6131 size_t extent_offset;
6132 size_t copy_size;
Chris Masona52d9a82007-08-27 16:49:44 -04006133
Chris Masona52d9a82007-08-27 16:49:44 -04006134 em->block_start = EXTENT_MAP_INLINE;
Chris Masonc8b97812008-10-29 14:49:59 -04006135 if (!page || create) {
Yan689f9342007-10-29 11:41:07 -04006136 em->start = extent_start;
Yan Zheng9036c102008-10-30 14:19:41 -04006137 em->len = extent_end - extent_start;
Yan689f9342007-10-29 11:41:07 -04006138 goto out;
6139 }
6140
Yan Zheng9036c102008-10-30 14:19:41 -04006141 size = btrfs_file_extent_inline_len(leaf, item);
6142 extent_offset = page_offset(page) + pg_offset - extent_start;
Chris Mason70dec802008-01-29 09:59:12 -05006143 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
Yan689f9342007-10-29 11:41:07 -04006144 size - extent_offset);
Chris Mason3326d1b2007-10-15 16:18:25 -04006145 em->start = extent_start + extent_offset;
Qu Wenruofda28322013-02-26 08:10:22 +00006146 em->len = ALIGN(copy_size, root->sectorsize);
Josef Bacikb4939682012-12-03 10:31:19 -05006147 em->orig_block_len = em->len;
Josef Bacik70c8a912012-10-11 16:54:30 -04006148 em->orig_start = em->start;
Li Zefan261507a02010-12-17 14:21:50 +08006149 if (compress_type) {
Chris Masonc8b97812008-10-29 14:49:59 -04006150 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Li Zefan261507a02010-12-17 14:21:50 +08006151 em->compress_type = compress_type;
6152 }
Yan689f9342007-10-29 11:41:07 -04006153 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
Chris Mason179e29e2007-11-01 11:28:41 -04006154 if (create == 0 && !PageUptodate(page)) {
Li Zefan261507a02010-12-17 14:21:50 +08006155 if (btrfs_file_extent_compression(leaf, item) !=
6156 BTRFS_COMPRESS_NONE) {
Chris Masonc8b97812008-10-29 14:49:59 -04006157 ret = uncompress_inline(path, inode, page,
6158 pg_offset,
6159 extent_offset, item);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006160 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -04006161 } else {
6162 map = kmap(page);
6163 read_extent_buffer(leaf, map + pg_offset, ptr,
6164 copy_size);
Chris Mason93c82d52009-09-11 12:36:29 -04006165 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6166 memset(map + pg_offset + copy_size, 0,
6167 PAGE_CACHE_SIZE - pg_offset -
6168 copy_size);
6169 }
Chris Masonc8b97812008-10-29 14:49:59 -04006170 kunmap(page);
6171 }
Chris Mason179e29e2007-11-01 11:28:41 -04006172 flush_dcache_page(page);
6173 } else if (create && PageUptodate(page)) {
Jan Schmidt6bf7e082011-12-01 14:35:19 +01006174 BUG();
Chris Mason179e29e2007-11-01 11:28:41 -04006175 if (!trans) {
6176 kunmap(page);
6177 free_extent_map(em);
6178 em = NULL;
Chris Masonff5714c2011-05-28 07:00:39 -04006179
David Sterbab3b4aa72011-04-21 01:20:15 +02006180 btrfs_release_path(path);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006181 trans = btrfs_join_transaction(root);
Chris Masonff5714c2011-05-28 07:00:39 -04006182
Tsutomu Itoh3612b492011-01-25 02:51:38 +00006183 if (IS_ERR(trans))
6184 return ERR_CAST(trans);
Chris Mason179e29e2007-11-01 11:28:41 -04006185 goto again;
6186 }
Chris Masonc8b97812008-10-29 14:49:59 -04006187 map = kmap(page);
Chris Mason70dec802008-01-29 09:59:12 -05006188 write_extent_buffer(leaf, map + pg_offset, ptr,
Chris Mason179e29e2007-11-01 11:28:41 -04006189 copy_size);
Chris Masonc8b97812008-10-29 14:49:59 -04006190 kunmap(page);
Chris Mason179e29e2007-11-01 11:28:41 -04006191 btrfs_mark_buffer_dirty(leaf);
Chris Masona52d9a82007-08-27 16:49:44 -04006192 }
Chris Masond1310b22008-01-24 16:13:08 -05006193 set_extent_uptodate(io_tree, em->start,
Arne Jansen507903b2011-04-06 10:02:20 +00006194 extent_map_end(em) - 1, NULL, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -04006195 goto insert;
6196 } else {
Julia Lawall31b1a2b2012-11-03 10:58:34 +00006197 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
Chris Masona52d9a82007-08-27 16:49:44 -04006198 }
6199not_found:
6200 em->start = start;
Josef Bacik70c8a912012-10-11 16:54:30 -04006201 em->orig_start = start;
Chris Masond1310b22008-01-24 16:13:08 -05006202 em->len = len;
Chris Masona52d9a82007-08-27 16:49:44 -04006203not_found_em:
Chris Mason5f39d392007-10-15 16:14:19 -04006204 em->block_start = EXTENT_MAP_HOLE;
Yan Zheng9036c102008-10-30 14:19:41 -04006205 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -04006206insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02006207 btrfs_release_path(path);
Chris Masond1310b22008-01-24 16:13:08 -05006208 if (em->start > start || extent_map_end(em) <= start) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006209 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02006210 em->start, em->len, start, len);
Chris Masona52d9a82007-08-27 16:49:44 -04006211 err = -EIO;
6212 goto out;
6213 }
Chris Masond1310b22008-01-24 16:13:08 -05006214
6215 err = 0;
Chris Mason890871b2009-09-02 16:24:52 -04006216 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006217 ret = add_extent_mapping(em_tree, em, 0);
Chris Mason3b951512008-04-17 11:29:12 -04006218 /* it is possible that someone inserted the extent into the tree
6219 * while we had the lock dropped. It is also possible that
6220 * an overlapping map exists in the tree
6221 */
Chris Masona52d9a82007-08-27 16:49:44 -04006222 if (ret == -EEXIST) {
Chris Mason3b951512008-04-17 11:29:12 -04006223 struct extent_map *existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006224
6225 ret = 0;
6226
Chris Mason3b951512008-04-17 11:29:12 -04006227 existing = lookup_extent_mapping(em_tree, start, len);
Chris Masone1c4b742008-04-22 13:26:46 -04006228 if (existing && (existing->start > start ||
6229 existing->start + existing->len <= start)) {
6230 free_extent_map(existing);
6231 existing = NULL;
6232 }
Chris Mason3b951512008-04-17 11:29:12 -04006233 if (!existing) {
6234 existing = lookup_extent_mapping(em_tree, em->start,
6235 em->len);
6236 if (existing) {
6237 err = merge_extent_mapping(em_tree, existing,
Chris Masone6dcd2d2008-07-17 12:53:50 -04006238 em, start,
6239 root->sectorsize);
Chris Mason3b951512008-04-17 11:29:12 -04006240 free_extent_map(existing);
6241 if (err) {
6242 free_extent_map(em);
6243 em = NULL;
6244 }
6245 } else {
6246 err = -EIO;
Chris Mason3b951512008-04-17 11:29:12 -04006247 free_extent_map(em);
6248 em = NULL;
6249 }
6250 } else {
6251 free_extent_map(em);
6252 em = existing;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006253 err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04006254 }
Chris Masona52d9a82007-08-27 16:49:44 -04006255 }
Chris Mason890871b2009-09-02 16:24:52 -04006256 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -04006257out:
liubo1abe9b82011-03-24 11:18:59 +00006258
Steven Rostedt4cd8587c2013-11-14 22:57:29 -05006259 trace_btrfs_get_extent(root, em);
liubo1abe9b82011-03-24 11:18:59 +00006260
Chris Masonf4219502008-07-22 11:18:09 -04006261 if (path)
6262 btrfs_free_path(path);
Chris Masona52d9a82007-08-27 16:49:44 -04006263 if (trans) {
6264 ret = btrfs_end_transaction(trans, root);
Chris Masond3977122009-01-05 21:25:51 -05006265 if (!err)
Chris Masona52d9a82007-08-27 16:49:44 -04006266 err = ret;
6267 }
Chris Masona52d9a82007-08-27 16:49:44 -04006268 if (err) {
6269 free_extent_map(em);
Chris Masona52d9a82007-08-27 16:49:44 -04006270 return ERR_PTR(err);
6271 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006272 BUG_ON(!em); /* Error is always set */
Chris Masona52d9a82007-08-27 16:49:44 -04006273 return em;
6274}
6275
Chris Masonec29ed52011-02-23 16:23:20 -05006276struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6277 size_t pg_offset, u64 start, u64 len,
6278 int create)
6279{
6280 struct extent_map *em;
6281 struct extent_map *hole_em = NULL;
6282 u64 range_start = start;
6283 u64 end;
6284 u64 found;
6285 u64 found_end;
6286 int err = 0;
6287
6288 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6289 if (IS_ERR(em))
6290 return em;
6291 if (em) {
6292 /*
Liu Bof9e4fb52013-01-07 10:10:12 +00006293 * if our em maps to
6294 * - a hole or
6295 * - a pre-alloc extent,
6296 * there might actually be delalloc bytes behind it.
Chris Masonec29ed52011-02-23 16:23:20 -05006297 */
Liu Bof9e4fb52013-01-07 10:10:12 +00006298 if (em->block_start != EXTENT_MAP_HOLE &&
6299 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
Chris Masonec29ed52011-02-23 16:23:20 -05006300 return em;
6301 else
6302 hole_em = em;
6303 }
6304
6305 /* check to see if we've wrapped (len == -1 or similar) */
6306 end = start + len;
6307 if (end < start)
6308 end = (u64)-1;
6309 else
6310 end -= 1;
6311
6312 em = NULL;
6313
6314 /* ok, we didn't find anything, lets look for delalloc */
6315 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6316 end, len, EXTENT_DELALLOC, 1);
6317 found_end = range_start + found;
6318 if (found_end < range_start)
6319 found_end = (u64)-1;
6320
6321 /*
6322 * we didn't find anything useful, return
6323 * the original results from get_extent()
6324 */
6325 if (range_start > end || found_end <= start) {
6326 em = hole_em;
6327 hole_em = NULL;
6328 goto out;
6329 }
6330
6331 /* adjust the range_start to make sure it doesn't
6332 * go backwards from the start they passed in
6333 */
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05306334 range_start = max(start, range_start);
Chris Masonec29ed52011-02-23 16:23:20 -05006335 found = found_end - range_start;
6336
6337 if (found > 0) {
6338 u64 hole_start = start;
6339 u64 hole_len = len;
6340
David Sterba172ddd62011-04-21 00:48:27 +02006341 em = alloc_extent_map();
Chris Masonec29ed52011-02-23 16:23:20 -05006342 if (!em) {
6343 err = -ENOMEM;
6344 goto out;
6345 }
6346 /*
6347 * when btrfs_get_extent can't find anything it
6348 * returns one huge hole
6349 *
6350 * make sure what it found really fits our range, and
6351 * adjust to make sure it is based on the start from
6352 * the caller
6353 */
6354 if (hole_em) {
6355 u64 calc_end = extent_map_end(hole_em);
6356
6357 if (calc_end <= start || (hole_em->start > end)) {
6358 free_extent_map(hole_em);
6359 hole_em = NULL;
6360 } else {
6361 hole_start = max(hole_em->start, start);
6362 hole_len = calc_end - hole_start;
6363 }
6364 }
6365 em->bdev = NULL;
6366 if (hole_em && range_start > hole_start) {
6367 /* our hole starts before our delalloc, so we
6368 * have to return just the parts of the hole
6369 * that go until the delalloc starts
6370 */
6371 em->len = min(hole_len,
6372 range_start - hole_start);
6373 em->start = hole_start;
6374 em->orig_start = hole_start;
6375 /*
6376 * don't adjust block start at all,
6377 * it is fixed at EXTENT_MAP_HOLE
6378 */
6379 em->block_start = hole_em->block_start;
6380 em->block_len = hole_len;
Liu Bof9e4fb52013-01-07 10:10:12 +00006381 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6382 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Chris Masonec29ed52011-02-23 16:23:20 -05006383 } else {
6384 em->start = range_start;
6385 em->len = found;
6386 em->orig_start = range_start;
6387 em->block_start = EXTENT_MAP_DELALLOC;
6388 em->block_len = found;
6389 }
6390 } else if (hole_em) {
6391 return hole_em;
6392 }
6393out:
6394
6395 free_extent_map(hole_em);
6396 if (err) {
6397 free_extent_map(em);
6398 return ERR_PTR(err);
6399 }
6400 return em;
6401}
6402
Josef Bacik4b46fce2010-05-23 11:00:55 -04006403static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6404 u64 start, u64 len)
6405{
6406 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik70c8a912012-10-11 16:54:30 -04006407 struct extent_map *em;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006408 struct btrfs_key ins;
6409 u64 alloc_hint;
6410 int ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006411
Josef Bacik4b46fce2010-05-23 11:00:55 -04006412 alloc_hint = get_extent_allocation_hint(inode, start, len);
Josef Bacik00361582013-08-14 14:02:47 -04006413 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
Josef Bacik81c9ad22012-01-18 10:56:06 -05006414 alloc_hint, &ins, 1);
Josef Bacik00361582013-08-14 14:02:47 -04006415 if (ret)
6416 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006417
Josef Bacik70c8a912012-10-11 16:54:30 -04006418 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
Josef Bacikcc95bef2013-04-04 14:31:27 -04006419 ins.offset, ins.offset, ins.offset, 0);
Josef Bacik00361582013-08-14 14:02:47 -04006420 if (IS_ERR(em)) {
6421 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6422 return em;
6423 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006424
6425 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6426 ins.offset, ins.offset, 0);
6427 if (ret) {
6428 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
Josef Bacik00361582013-08-14 14:02:47 -04006429 free_extent_map(em);
6430 return ERR_PTR(ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006431 }
Josef Bacik00361582013-08-14 14:02:47 -04006432
Josef Bacik4b46fce2010-05-23 11:00:55 -04006433 return em;
6434}
6435
Chris Mason46bfbb52010-05-26 11:04:10 -04006436/*
6437 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6438 * block must be cow'd
6439 */
Josef Bacik00361582013-08-14 14:02:47 -04006440noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
Josef Bacik7ee9e442013-06-21 16:37:03 -04006441 u64 *orig_start, u64 *orig_block_len,
6442 u64 *ram_bytes)
Chris Mason46bfbb52010-05-26 11:04:10 -04006443{
Josef Bacik00361582013-08-14 14:02:47 -04006444 struct btrfs_trans_handle *trans;
Chris Mason46bfbb52010-05-26 11:04:10 -04006445 struct btrfs_path *path;
6446 int ret;
6447 struct extent_buffer *leaf;
6448 struct btrfs_root *root = BTRFS_I(inode)->root;
6449 struct btrfs_file_extent_item *fi;
6450 struct btrfs_key key;
6451 u64 disk_bytenr;
6452 u64 backref_offset;
6453 u64 extent_end;
6454 u64 num_bytes;
6455 int slot;
6456 int found_type;
Josef Bacik7ee9e442013-06-21 16:37:03 -04006457 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
Chris Mason46bfbb52010-05-26 11:04:10 -04006458 path = btrfs_alloc_path();
6459 if (!path)
6460 return -ENOMEM;
6461
Josef Bacik00361582013-08-14 14:02:47 -04006462 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
Chris Mason46bfbb52010-05-26 11:04:10 -04006463 offset, 0);
6464 if (ret < 0)
6465 goto out;
6466
6467 slot = path->slots[0];
6468 if (ret == 1) {
6469 if (slot == 0) {
6470 /* can't find the item, must cow */
6471 ret = 0;
6472 goto out;
6473 }
6474 slot--;
6475 }
6476 ret = 0;
6477 leaf = path->nodes[0];
6478 btrfs_item_key_to_cpu(leaf, &key, slot);
Li Zefan33345d012011-04-20 10:31:50 +08006479 if (key.objectid != btrfs_ino(inode) ||
Chris Mason46bfbb52010-05-26 11:04:10 -04006480 key.type != BTRFS_EXTENT_DATA_KEY) {
6481 /* not our file or wrong item type, must cow */
6482 goto out;
6483 }
6484
6485 if (key.offset > offset) {
6486 /* Wrong offset, must cow */
6487 goto out;
6488 }
6489
6490 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6491 found_type = btrfs_file_extent_type(leaf, fi);
6492 if (found_type != BTRFS_FILE_EXTENT_REG &&
6493 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6494 /* not a regular extent, must cow */
6495 goto out;
6496 }
Josef Bacik7ee9e442013-06-21 16:37:03 -04006497
6498 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6499 goto out;
6500
Chris Mason46bfbb52010-05-26 11:04:10 -04006501 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Josef Bacik7ee9e442013-06-21 16:37:03 -04006502 if (disk_bytenr == 0)
6503 goto out;
6504
6505 if (btrfs_file_extent_compression(leaf, fi) ||
6506 btrfs_file_extent_encryption(leaf, fi) ||
6507 btrfs_file_extent_other_encoding(leaf, fi))
6508 goto out;
6509
Chris Mason46bfbb52010-05-26 11:04:10 -04006510 backref_offset = btrfs_file_extent_offset(leaf, fi);
6511
Josef Bacik7ee9e442013-06-21 16:37:03 -04006512 if (orig_start) {
6513 *orig_start = key.offset - backref_offset;
6514 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6515 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6516 }
Josef Bacikeb384b52013-04-24 16:32:55 -04006517
Chris Mason46bfbb52010-05-26 11:04:10 -04006518 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
Chris Mason46bfbb52010-05-26 11:04:10 -04006519
6520 if (btrfs_extent_readonly(root, disk_bytenr))
6521 goto out;
Josef Bacik1bda19e2013-10-18 12:10:36 -04006522 btrfs_release_path(path);
Chris Mason46bfbb52010-05-26 11:04:10 -04006523
6524 /*
6525 * look for other files referencing this extent, if we
6526 * find any we must cow
6527 */
Josef Bacik00361582013-08-14 14:02:47 -04006528 trans = btrfs_join_transaction(root);
6529 if (IS_ERR(trans)) {
6530 ret = 0;
Chris Mason46bfbb52010-05-26 11:04:10 -04006531 goto out;
Josef Bacik00361582013-08-14 14:02:47 -04006532 }
6533
6534 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6535 key.offset - backref_offset, disk_bytenr);
6536 btrfs_end_transaction(trans, root);
6537 if (ret) {
6538 ret = 0;
6539 goto out;
6540 }
Chris Mason46bfbb52010-05-26 11:04:10 -04006541
6542 /*
6543 * adjust disk_bytenr and num_bytes to cover just the bytes
6544 * in this extent we are about to write. If there
6545 * are any csums in that range we have to cow in order
6546 * to keep the csums correct
6547 */
6548 disk_bytenr += backref_offset;
6549 disk_bytenr += offset - key.offset;
Josef Bacikeb384b52013-04-24 16:32:55 -04006550 num_bytes = min(offset + *len, extent_end) - offset;
Chris Mason46bfbb52010-05-26 11:04:10 -04006551 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6552 goto out;
6553 /*
6554 * all of the above have passed, it is safe to overwrite this extent
6555 * without cow
6556 */
Josef Bacikeb384b52013-04-24 16:32:55 -04006557 *len = num_bytes;
Chris Mason46bfbb52010-05-26 11:04:10 -04006558 ret = 1;
6559out:
6560 btrfs_free_path(path);
6561 return ret;
6562}
6563
Josef Bacikeb838e72012-07-31 16:28:48 -04006564static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6565 struct extent_state **cached_state, int writing)
6566{
6567 struct btrfs_ordered_extent *ordered;
6568 int ret = 0;
6569
6570 while (1) {
6571 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6572 0, cached_state);
6573 /*
6574 * We're concerned with the entire range that we're going to be
6575 * doing DIO to, so we need to make sure theres no ordered
6576 * extents in this range.
6577 */
6578 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6579 lockend - lockstart + 1);
6580
6581 /*
6582 * We need to make sure there are no buffered pages in this
6583 * range either, we could have raced between the invalidate in
6584 * generic_file_direct_write and locking the extent. The
6585 * invalidate needs to happen so that reads after a write do not
6586 * get stale data.
6587 */
6588 if (!ordered && (!writing ||
6589 !test_range_bit(&BTRFS_I(inode)->io_tree,
6590 lockstart, lockend, EXTENT_UPTODATE, 0,
6591 *cached_state)))
6592 break;
6593
6594 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6595 cached_state, GFP_NOFS);
6596
6597 if (ordered) {
6598 btrfs_start_ordered_extent(inode, ordered, 1);
6599 btrfs_put_ordered_extent(ordered);
6600 } else {
6601 /* Screw you mmap */
6602 ret = filemap_write_and_wait_range(inode->i_mapping,
6603 lockstart,
6604 lockend);
6605 if (ret)
6606 break;
6607
6608 /*
6609 * If we found a page that couldn't be invalidated just
6610 * fall back to buffered.
6611 */
6612 ret = invalidate_inode_pages2_range(inode->i_mapping,
6613 lockstart >> PAGE_CACHE_SHIFT,
6614 lockend >> PAGE_CACHE_SHIFT);
6615 if (ret)
6616 break;
6617 }
6618
6619 cond_resched();
6620 }
6621
6622 return ret;
6623}
6624
Josef Bacik69ffb542012-09-11 15:40:07 -04006625static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6626 u64 len, u64 orig_start,
6627 u64 block_start, u64 block_len,
Josef Bacikcc95bef2013-04-04 14:31:27 -04006628 u64 orig_block_len, u64 ram_bytes,
6629 int type)
Josef Bacik69ffb542012-09-11 15:40:07 -04006630{
6631 struct extent_map_tree *em_tree;
6632 struct extent_map *em;
6633 struct btrfs_root *root = BTRFS_I(inode)->root;
6634 int ret;
6635
6636 em_tree = &BTRFS_I(inode)->extent_tree;
6637 em = alloc_extent_map();
6638 if (!em)
6639 return ERR_PTR(-ENOMEM);
6640
6641 em->start = start;
6642 em->orig_start = orig_start;
Josef Bacik2ab28f32012-10-12 15:27:49 -04006643 em->mod_start = start;
6644 em->mod_len = len;
Josef Bacik69ffb542012-09-11 15:40:07 -04006645 em->len = len;
6646 em->block_len = block_len;
6647 em->block_start = block_start;
6648 em->bdev = root->fs_info->fs_devices->latest_bdev;
Josef Bacikb4939682012-12-03 10:31:19 -05006649 em->orig_block_len = orig_block_len;
Josef Bacikcc95bef2013-04-04 14:31:27 -04006650 em->ram_bytes = ram_bytes;
Josef Bacik70c8a912012-10-11 16:54:30 -04006651 em->generation = -1;
Josef Bacik69ffb542012-09-11 15:40:07 -04006652 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6653 if (type == BTRFS_ORDERED_PREALLOC)
Josef Bacikb11e2342012-12-03 10:58:15 -05006654 set_bit(EXTENT_FLAG_FILLING, &em->flags);
Josef Bacik69ffb542012-09-11 15:40:07 -04006655
6656 do {
6657 btrfs_drop_extent_cache(inode, em->start,
6658 em->start + em->len - 1, 0);
6659 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04006660 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik69ffb542012-09-11 15:40:07 -04006661 write_unlock(&em_tree->lock);
6662 } while (ret == -EEXIST);
6663
6664 if (ret) {
6665 free_extent_map(em);
6666 return ERR_PTR(ret);
6667 }
6668
6669 return em;
6670}
6671
6672
Josef Bacik4b46fce2010-05-23 11:00:55 -04006673static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6674 struct buffer_head *bh_result, int create)
6675{
6676 struct extent_map *em;
6677 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikeb838e72012-07-31 16:28:48 -04006678 struct extent_state *cached_state = NULL;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006679 u64 start = iblock << inode->i_blkbits;
Josef Bacikeb838e72012-07-31 16:28:48 -04006680 u64 lockstart, lockend;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006681 u64 len = bh_result->b_size;
Josef Bacikeb838e72012-07-31 16:28:48 -04006682 int unlock_bits = EXTENT_LOCKED;
Miao Xie09348562013-02-07 10:12:07 +00006683 int ret = 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04006684
Miao Xie172a5042013-02-21 02:48:22 -07006685 if (create)
Josef Bacikeb838e72012-07-31 16:28:48 -04006686 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
Miao Xie172a5042013-02-21 02:48:22 -07006687 else
Josef Bacikc3298612012-08-03 16:49:19 -04006688 len = min_t(u64, len, root->sectorsize);
Josef Bacikeb838e72012-07-31 16:28:48 -04006689
Josef Bacikc3298612012-08-03 16:49:19 -04006690 lockstart = start;
6691 lockend = start + len - 1;
6692
Josef Bacikeb838e72012-07-31 16:28:48 -04006693 /*
6694 * If this errors out it's because we couldn't invalidate pagecache for
6695 * this range and we need to fallback to buffered.
6696 */
6697 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6698 return -ENOTBLK;
6699
Josef Bacik4b46fce2010-05-23 11:00:55 -04006700 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
Josef Bacikeb838e72012-07-31 16:28:48 -04006701 if (IS_ERR(em)) {
6702 ret = PTR_ERR(em);
6703 goto unlock_err;
6704 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006705
6706 /*
6707 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6708 * io. INLINE is special, and we could probably kludge it in here, but
6709 * it's still buffered so for safety lets just fall back to the generic
6710 * buffered path.
6711 *
6712 * For COMPRESSED we _have_ to read the entire extent in so we can
6713 * decompress it, so there will be buffering required no matter what we
6714 * do, so go ahead and fallback to buffered.
6715 *
6716 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6717 * to buffered IO. Don't blame me, this is the price we pay for using
6718 * the generic code.
6719 */
6720 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6721 em->block_start == EXTENT_MAP_INLINE) {
6722 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04006723 ret = -ENOTBLK;
6724 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006725 }
6726
6727 /* Just a good old fashioned hole, return */
6728 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6729 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6730 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04006731 goto unlock_err;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006732 }
6733
6734 /*
6735 * We don't allocate a new extent in the following cases
6736 *
6737 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6738 * existing extent.
6739 * 2) The extent is marked as PREALLOC. We're good to go here and can
6740 * just use the extent.
6741 *
6742 */
Chris Mason46bfbb52010-05-26 11:04:10 -04006743 if (!create) {
Josef Bacikeb838e72012-07-31 16:28:48 -04006744 len = min(len, em->len - (start - em->start));
6745 lockstart = start + len;
6746 goto unlock;
Chris Mason46bfbb52010-05-26 11:04:10 -04006747 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006748
6749 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6750 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6751 em->block_start != EXTENT_MAP_HOLE)) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04006752 int type;
6753 int ret;
Josef Bacikeb384b52013-04-24 16:32:55 -04006754 u64 block_start, orig_start, orig_block_len, ram_bytes;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006755
6756 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6757 type = BTRFS_ORDERED_PREALLOC;
6758 else
6759 type = BTRFS_ORDERED_NOCOW;
Chris Mason46bfbb52010-05-26 11:04:10 -04006760 len = min(len, em->len - (start - em->start));
Josef Bacik4b46fce2010-05-23 11:00:55 -04006761 block_start = em->block_start + (start - em->start);
Chris Mason46bfbb52010-05-26 11:04:10 -04006762
Josef Bacik00361582013-08-14 14:02:47 -04006763 if (can_nocow_extent(inode, start, &len, &orig_start,
Josef Bacik7ee9e442013-06-21 16:37:03 -04006764 &orig_block_len, &ram_bytes) == 1) {
Josef Bacik69ffb542012-09-11 15:40:07 -04006765 if (type == BTRFS_ORDERED_PREALLOC) {
6766 free_extent_map(em);
6767 em = create_pinned_em(inode, start, len,
6768 orig_start,
Josef Bacikb4939682012-12-03 10:31:19 -05006769 block_start, len,
Josef Bacikcc95bef2013-04-04 14:31:27 -04006770 orig_block_len,
6771 ram_bytes, type);
Josef Bacik00361582013-08-14 14:02:47 -04006772 if (IS_ERR(em))
Josef Bacik69ffb542012-09-11 15:40:07 -04006773 goto unlock_err;
Josef Bacik69ffb542012-09-11 15:40:07 -04006774 }
6775
Chris Mason46bfbb52010-05-26 11:04:10 -04006776 ret = btrfs_add_ordered_extent_dio(inode, start,
6777 block_start, len, len, type);
Chris Mason46bfbb52010-05-26 11:04:10 -04006778 if (ret) {
6779 free_extent_map(em);
Josef Bacikeb838e72012-07-31 16:28:48 -04006780 goto unlock_err;
Chris Mason46bfbb52010-05-26 11:04:10 -04006781 }
6782 goto unlock;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006783 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006784 }
Josef Bacik00361582013-08-14 14:02:47 -04006785
Chris Mason46bfbb52010-05-26 11:04:10 -04006786 /*
6787 * this will cow the extent, reset the len in case we changed
6788 * it above
6789 */
6790 len = bh_result->b_size;
Josef Bacik70c8a912012-10-11 16:54:30 -04006791 free_extent_map(em);
6792 em = btrfs_new_extent_direct(inode, start, len);
Josef Bacikeb838e72012-07-31 16:28:48 -04006793 if (IS_ERR(em)) {
6794 ret = PTR_ERR(em);
6795 goto unlock_err;
6796 }
Chris Mason46bfbb52010-05-26 11:04:10 -04006797 len = min(len, em->len - (start - em->start));
6798unlock:
Josef Bacik4b46fce2010-05-23 11:00:55 -04006799 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6800 inode->i_blkbits;
Chris Mason46bfbb52010-05-26 11:04:10 -04006801 bh_result->b_size = len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006802 bh_result->b_bdev = em->bdev;
6803 set_buffer_mapped(bh_result);
Josef Bacikc3473e82012-06-19 10:59:00 -04006804 if (create) {
6805 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6806 set_buffer_new(bh_result);
6807
6808 /*
6809 * Need to update the i_size under the extent lock so buffered
6810 * readers will get the updated i_size when we unlock.
6811 */
6812 if (start + len > i_size_read(inode))
6813 i_size_write(inode, start + len);
Miao Xie09348562013-02-07 10:12:07 +00006814
Miao Xie172a5042013-02-21 02:48:22 -07006815 spin_lock(&BTRFS_I(inode)->lock);
6816 BTRFS_I(inode)->outstanding_extents++;
6817 spin_unlock(&BTRFS_I(inode)->lock);
6818
Miao Xie09348562013-02-07 10:12:07 +00006819 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6820 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6821 &cached_state, GFP_NOFS);
6822 BUG_ON(ret);
Josef Bacikc3473e82012-06-19 10:59:00 -04006823 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006824
Josef Bacikeb838e72012-07-31 16:28:48 -04006825 /*
6826 * In the case of write we need to clear and unlock the entire range,
6827 * in the case of read we need to unlock only the end area that we
6828 * aren't using if there is any left over space.
6829 */
Liu Bo24c03fa2012-08-22 20:10:38 -06006830 if (lockstart < lockend) {
Miao Xie09348562013-02-07 10:12:07 +00006831 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6832 lockend, unlock_bits, 1, 0,
6833 &cached_state, GFP_NOFS);
Liu Bo24c03fa2012-08-22 20:10:38 -06006834 } else {
Josef Bacikeb838e72012-07-31 16:28:48 -04006835 free_extent_state(cached_state);
Liu Bo24c03fa2012-08-22 20:10:38 -06006836 }
Josef Bacikeb838e72012-07-31 16:28:48 -04006837
Josef Bacik4b46fce2010-05-23 11:00:55 -04006838 free_extent_map(em);
6839
6840 return 0;
Josef Bacikeb838e72012-07-31 16:28:48 -04006841
6842unlock_err:
Josef Bacikeb838e72012-07-31 16:28:48 -04006843 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6844 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6845 return ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006846}
6847
Josef Bacik4b46fce2010-05-23 11:00:55 -04006848static void btrfs_endio_direct_read(struct bio *bio, int err)
6849{
Miao Xiee65e1532010-11-22 03:04:43 +00006850 struct btrfs_dio_private *dip = bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006851 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6852 struct bio_vec *bvec = bio->bi_io_vec;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006853 struct inode *inode = dip->inode;
6854 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason9be33952013-05-17 18:30:14 -04006855 struct bio *dio_bio;
Miao Xiefacc8a222013-07-25 19:22:34 +08006856 u32 *csums = (u32 *)dip->csum;
6857 int index = 0;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006858 u64 start;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006859
6860 start = dip->logical_offset;
6861 do {
6862 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6863 struct page *page = bvec->bv_page;
6864 char *kaddr;
6865 u32 csum = ~(u32)0;
6866 unsigned long flags;
6867
6868 local_irq_save(flags);
Cong Wang7ac687d2011-11-25 23:14:28 +08006869 kaddr = kmap_atomic(page);
Liu Bob0496682013-03-14 14:57:45 +00006870 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
Josef Bacik4b46fce2010-05-23 11:00:55 -04006871 csum, bvec->bv_len);
6872 btrfs_csum_final(csum, (char *)&csum);
Cong Wang7ac687d2011-11-25 23:14:28 +08006873 kunmap_atomic(kaddr);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006874 local_irq_restore(flags);
6875
6876 flush_dcache_page(bvec->bv_page);
Miao Xiefacc8a222013-07-25 19:22:34 +08006877 if (csum != csums[index]) {
6878 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02006879 btrfs_ino(inode), start, csum,
6880 csums[index]);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006881 err = -EIO;
6882 }
6883 }
6884
6885 start += bvec->bv_len;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006886 bvec++;
Miao Xiefacc8a222013-07-25 19:22:34 +08006887 index++;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006888 } while (bvec <= bvec_end);
6889
6890 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
Jeff Mahoneyd0082372012-03-01 14:57:19 +01006891 dip->logical_offset + dip->bytes - 1);
Chris Mason9be33952013-05-17 18:30:14 -04006892 dio_bio = dip->dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006893
Josef Bacik4b46fce2010-05-23 11:00:55 -04006894 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04006895
6896 /* If we had a csum failure make sure to clear the uptodate flag */
6897 if (err)
Chris Mason9be33952013-05-17 18:30:14 -04006898 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6899 dio_end_io(dio_bio, err);
6900 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006901}
6902
6903static void btrfs_endio_direct_write(struct bio *bio, int err)
6904{
6905 struct btrfs_dio_private *dip = bio->bi_private;
6906 struct inode *inode = dip->inode;
6907 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006908 struct btrfs_ordered_extent *ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05006909 u64 ordered_offset = dip->logical_offset;
6910 u64 ordered_bytes = dip->bytes;
Chris Mason9be33952013-05-17 18:30:14 -04006911 struct bio *dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006912 int ret;
6913
6914 if (err)
6915 goto out_done;
Chris Mason163cf092010-11-28 19:56:33 -05006916again:
6917 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6918 &ordered_offset,
Josef Bacik5fd02042012-05-02 14:00:54 -04006919 ordered_bytes, !err);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006920 if (!ret)
Chris Mason163cf092010-11-28 19:56:33 -05006921 goto out_test;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006922
Josef Bacik5fd02042012-05-02 14:00:54 -04006923 ordered->work.func = finish_ordered_fn;
6924 ordered->work.flags = 0;
6925 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6926 &ordered->work);
Chris Mason163cf092010-11-28 19:56:33 -05006927out_test:
6928 /*
6929 * our bio might span multiple ordered extents. If we haven't
6930 * completed the accounting for the whole dio, go back and try again
6931 */
6932 if (ordered_offset < dip->logical_offset + dip->bytes) {
6933 ordered_bytes = dip->logical_offset + dip->bytes -
6934 ordered_offset;
Josef Bacik5fd02042012-05-02 14:00:54 -04006935 ordered = NULL;
Chris Mason163cf092010-11-28 19:56:33 -05006936 goto again;
6937 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04006938out_done:
Chris Mason9be33952013-05-17 18:30:14 -04006939 dio_bio = dip->dio_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04006940
Josef Bacik4b46fce2010-05-23 11:00:55 -04006941 kfree(dip);
Josef Bacikc0da7aa2011-03-22 11:05:07 -04006942
6943 /* If we had an error make sure to clear the uptodate flag */
6944 if (err)
Chris Mason9be33952013-05-17 18:30:14 -04006945 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6946 dio_end_io(dio_bio, err);
6947 bio_put(bio);
Josef Bacik4b46fce2010-05-23 11:00:55 -04006948}
6949
Chris Masoneaf25d92010-05-25 09:48:28 -04006950static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6951 struct bio *bio, int mirror_num,
6952 unsigned long bio_flags, u64 offset)
6953{
6954 int ret;
6955 struct btrfs_root *root = BTRFS_I(inode)->root;
6956 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006957 BUG_ON(ret); /* -ENOMEM */
Chris Masoneaf25d92010-05-25 09:48:28 -04006958 return 0;
6959}
6960
Miao Xiee65e1532010-11-22 03:04:43 +00006961static void btrfs_end_dio_bio(struct bio *bio, int err)
6962{
6963 struct btrfs_dio_private *dip = bio->bi_private;
6964
6965 if (err) {
Li Zefan33345d012011-04-20 10:31:50 +08006966 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
Jan Beulich3dd14622010-12-07 14:54:09 +00006967 "sector %#Lx len %u err no %d\n",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02006968 btrfs_ino(dip->inode), bio->bi_rw,
Jan Beulich3dd14622010-12-07 14:54:09 +00006969 (unsigned long long)bio->bi_sector, bio->bi_size, err);
Miao Xiee65e1532010-11-22 03:04:43 +00006970 dip->errors = 1;
6971
6972 /*
6973 * before atomic variable goto zero, we must make sure
6974 * dip->errors is perceived to be set.
6975 */
6976 smp_mb__before_atomic_dec();
6977 }
6978
6979 /* if there are more bios still pending for this dio, just exit */
6980 if (!atomic_dec_and_test(&dip->pending_bios))
6981 goto out;
6982
Chris Mason9be33952013-05-17 18:30:14 -04006983 if (dip->errors) {
Miao Xiee65e1532010-11-22 03:04:43 +00006984 bio_io_error(dip->orig_bio);
Chris Mason9be33952013-05-17 18:30:14 -04006985 } else {
6986 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
Miao Xiee65e1532010-11-22 03:04:43 +00006987 bio_endio(dip->orig_bio, 0);
6988 }
6989out:
6990 bio_put(bio);
6991}
6992
6993static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6994 u64 first_sector, gfp_t gfp_flags)
6995{
6996 int nr_vecs = bio_get_nr_vecs(bdev);
6997 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6998}
6999
7000static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7001 int rw, u64 file_offset, int skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04007002 int async_submit)
Miao Xiee65e1532010-11-22 03:04:43 +00007003{
Miao Xiefacc8a222013-07-25 19:22:34 +08007004 struct btrfs_dio_private *dip = bio->bi_private;
Miao Xiee65e1532010-11-22 03:04:43 +00007005 int write = rw & REQ_WRITE;
7006 struct btrfs_root *root = BTRFS_I(inode)->root;
7007 int ret;
7008
Josef Bacikb812ce22012-11-16 13:56:32 -05007009 if (async_submit)
7010 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7011
Miao Xiee65e1532010-11-22 03:04:43 +00007012 bio_get(bio);
Josef Bacik5fd02042012-05-02 14:00:54 -04007013
7014 if (!write) {
7015 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7016 if (ret)
7017 goto err;
7018 }
Miao Xiee65e1532010-11-22 03:04:43 +00007019
Josef Bacik1ae39932011-04-06 14:41:34 -04007020 if (skip_sum)
7021 goto map;
7022
7023 if (write && async_submit) {
Miao Xiee65e1532010-11-22 03:04:43 +00007024 ret = btrfs_wq_submit_bio(root->fs_info,
7025 inode, rw, bio, 0, 0,
7026 file_offset,
7027 __btrfs_submit_bio_start_direct_io,
7028 __btrfs_submit_bio_done);
7029 goto err;
Josef Bacik1ae39932011-04-06 14:41:34 -04007030 } else if (write) {
7031 /*
7032 * If we aren't doing async submit, calculate the csum of the
7033 * bio now.
7034 */
7035 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7036 if (ret)
7037 goto err;
Tsutomu Itohc2db1072011-03-01 06:48:31 +00007038 } else if (!skip_sum) {
Miao Xiefacc8a222013-07-25 19:22:34 +08007039 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7040 file_offset);
Tsutomu Itohc2db1072011-03-01 06:48:31 +00007041 if (ret)
7042 goto err;
7043 }
Miao Xiee65e1532010-11-22 03:04:43 +00007044
Josef Bacik1ae39932011-04-06 14:41:34 -04007045map:
7046 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00007047err:
7048 bio_put(bio);
7049 return ret;
7050}
7051
7052static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7053 int skip_sum)
7054{
7055 struct inode *inode = dip->inode;
7056 struct btrfs_root *root = BTRFS_I(inode)->root;
Miao Xiee65e1532010-11-22 03:04:43 +00007057 struct bio *bio;
7058 struct bio *orig_bio = dip->orig_bio;
7059 struct bio_vec *bvec = orig_bio->bi_io_vec;
7060 u64 start_sector = orig_bio->bi_sector;
7061 u64 file_offset = dip->logical_offset;
7062 u64 submit_len = 0;
7063 u64 map_length;
7064 int nr_pages = 0;
Miao Xiee65e1532010-11-22 03:04:43 +00007065 int ret = 0;
Josef Bacik1ae39932011-04-06 14:41:34 -04007066 int async_submit = 0;
Miao Xiee65e1532010-11-22 03:04:43 +00007067
Miao Xiee65e1532010-11-22 03:04:43 +00007068 map_length = orig_bio->bi_size;
David Woodhouse53b381b2013-01-29 18:40:14 -05007069 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00007070 &map_length, NULL, 0);
7071 if (ret) {
Josef Bacik64728bb2011-04-25 19:43:52 -04007072 bio_put(orig_bio);
Miao Xiee65e1532010-11-22 03:04:43 +00007073 return -EIO;
7074 }
Miao Xiefacc8a222013-07-25 19:22:34 +08007075
Josef Bacik02f57c72011-04-06 14:25:44 -04007076 if (map_length >= orig_bio->bi_size) {
7077 bio = orig_bio;
7078 goto submit;
7079 }
7080
David Woodhouse53b381b2013-01-29 18:40:14 -05007081 /* async crcs make it difficult to collect full stripe writes. */
7082 if (btrfs_get_alloc_profile(root, 1) &
7083 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7084 async_submit = 0;
7085 else
7086 async_submit = 1;
7087
Josef Bacik02f57c72011-04-06 14:25:44 -04007088 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7089 if (!bio)
7090 return -ENOMEM;
7091 bio->bi_private = dip;
7092 bio->bi_end_io = btrfs_end_dio_bio;
7093 atomic_inc(&dip->pending_bios);
7094
Miao Xiee65e1532010-11-22 03:04:43 +00007095 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7096 if (unlikely(map_length < submit_len + bvec->bv_len ||
7097 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7098 bvec->bv_offset) < bvec->bv_len)) {
7099 /*
7100 * inc the count before we submit the bio so
7101 * we know the end IO handler won't happen before
7102 * we inc the count. Otherwise, the dip might get freed
7103 * before we're done setting it up
7104 */
7105 atomic_inc(&dip->pending_bios);
7106 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7107 file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04007108 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00007109 if (ret) {
7110 bio_put(bio);
7111 atomic_dec(&dip->pending_bios);
7112 goto out_err;
7113 }
7114
Miao Xiee65e1532010-11-22 03:04:43 +00007115 start_sector += submit_len >> 9;
7116 file_offset += submit_len;
7117
7118 submit_len = 0;
7119 nr_pages = 0;
7120
7121 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7122 start_sector, GFP_NOFS);
7123 if (!bio)
7124 goto out_err;
7125 bio->bi_private = dip;
7126 bio->bi_end_io = btrfs_end_dio_bio;
7127
7128 map_length = orig_bio->bi_size;
David Woodhouse53b381b2013-01-29 18:40:14 -05007129 ret = btrfs_map_block(root->fs_info, rw,
Stefan Behrens3ec706c2012-11-05 15:46:42 +01007130 start_sector << 9,
Miao Xiee65e1532010-11-22 03:04:43 +00007131 &map_length, NULL, 0);
7132 if (ret) {
7133 bio_put(bio);
7134 goto out_err;
7135 }
7136 } else {
7137 submit_len += bvec->bv_len;
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05307138 nr_pages++;
Miao Xiee65e1532010-11-22 03:04:43 +00007139 bvec++;
7140 }
7141 }
7142
Josef Bacik02f57c72011-04-06 14:25:44 -04007143submit:
Miao Xiee65e1532010-11-22 03:04:43 +00007144 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
Josef Bacikc3298612012-08-03 16:49:19 -04007145 async_submit);
Miao Xiee65e1532010-11-22 03:04:43 +00007146 if (!ret)
7147 return 0;
7148
7149 bio_put(bio);
7150out_err:
7151 dip->errors = 1;
7152 /*
7153 * before atomic variable goto zero, we must
7154 * make sure dip->errors is perceived to be set.
7155 */
7156 smp_mb__before_atomic_dec();
7157 if (atomic_dec_and_test(&dip->pending_bios))
7158 bio_io_error(dip->orig_bio);
7159
7160 /* bio_end_io() will handle error, so we needn't return it */
7161 return 0;
7162}
7163
Chris Mason9be33952013-05-17 18:30:14 -04007164static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7165 struct inode *inode, loff_t file_offset)
Josef Bacik4b46fce2010-05-23 11:00:55 -04007166{
7167 struct btrfs_root *root = BTRFS_I(inode)->root;
7168 struct btrfs_dio_private *dip;
Chris Mason9be33952013-05-17 18:30:14 -04007169 struct bio *io_bio;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007170 int skip_sum;
Miao Xiefacc8a222013-07-25 19:22:34 +08007171 int sum_len;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02007172 int write = rw & REQ_WRITE;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007173 int ret = 0;
Miao Xiefacc8a222013-07-25 19:22:34 +08007174 u16 csum_size;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007175
7176 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7177
Chris Mason9be33952013-05-17 18:30:14 -04007178 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04007179 if (!io_bio) {
Josef Bacik4b46fce2010-05-23 11:00:55 -04007180 ret = -ENOMEM;
7181 goto free_ordered;
7182 }
Josef Bacik4b46fce2010-05-23 11:00:55 -04007183
Miao Xiefacc8a222013-07-25 19:22:34 +08007184 if (!skip_sum && !write) {
7185 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7186 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7187 sum_len *= csum_size;
7188 } else {
7189 sum_len = 0;
7190 }
7191
7192 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
Chris Mason9be33952013-05-17 18:30:14 -04007193 if (!dip) {
7194 ret = -ENOMEM;
7195 goto free_io_bio;
7196 }
7197
7198 dip->private = dio_bio->bi_private;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007199 dip->inode = inode;
7200 dip->logical_offset = file_offset;
Miao Xiee6da5d22013-06-19 18:19:17 +08007201 dip->bytes = dio_bio->bi_size;
Chris Mason9be33952013-05-17 18:30:14 -04007202 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7203 io_bio->bi_private = dip;
Miao Xiee65e1532010-11-22 03:04:43 +00007204 dip->errors = 0;
Chris Mason9be33952013-05-17 18:30:14 -04007205 dip->orig_bio = io_bio;
7206 dip->dio_bio = dio_bio;
Miao Xiee65e1532010-11-22 03:04:43 +00007207 atomic_set(&dip->pending_bios, 0);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007208
7209 if (write)
Chris Mason9be33952013-05-17 18:30:14 -04007210 io_bio->bi_end_io = btrfs_endio_direct_write;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007211 else
Chris Mason9be33952013-05-17 18:30:14 -04007212 io_bio->bi_end_io = btrfs_endio_direct_read;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007213
Miao Xiee65e1532010-11-22 03:04:43 +00007214 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7215 if (!ret)
Chris Masoneaf25d92010-05-25 09:48:28 -04007216 return;
Chris Mason9be33952013-05-17 18:30:14 -04007217
7218free_io_bio:
7219 bio_put(io_bio);
7220
Josef Bacik4b46fce2010-05-23 11:00:55 -04007221free_ordered:
7222 /*
7223 * If this is a write, we need to clean up the reserved space and kill
7224 * the ordered extent.
7225 */
7226 if (write) {
7227 struct btrfs_ordered_extent *ordered;
Josef Bacik955256f2010-11-19 09:41:10 -05007228 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007229 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7230 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7231 btrfs_free_reserved_extent(root, ordered->start,
7232 ordered->disk_len);
7233 btrfs_put_ordered_extent(ordered);
7234 btrfs_put_ordered_extent(ordered);
7235 }
Chris Mason9be33952013-05-17 18:30:14 -04007236 bio_endio(dio_bio, ret);
Josef Bacik4b46fce2010-05-23 11:00:55 -04007237}
7238
Chris Mason5a5f79b2010-05-26 21:33:37 -04007239static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7240 const struct iovec *iov, loff_t offset,
7241 unsigned long nr_segs)
7242{
7243 int seg;
Josef Bacika1b75f72011-04-08 15:51:18 +00007244 int i;
Chris Mason5a5f79b2010-05-26 21:33:37 -04007245 size_t size;
7246 unsigned long addr;
7247 unsigned blocksize_mask = root->sectorsize - 1;
7248 ssize_t retval = -EINVAL;
7249 loff_t end = offset;
7250
7251 if (offset & blocksize_mask)
7252 goto out;
7253
7254 /* Check the memory alignment. Blocks cannot straddle pages */
7255 for (seg = 0; seg < nr_segs; seg++) {
7256 addr = (unsigned long)iov[seg].iov_base;
7257 size = iov[seg].iov_len;
7258 end += size;
Josef Bacika1b75f72011-04-08 15:51:18 +00007259 if ((addr & blocksize_mask) || (size & blocksize_mask))
Chris Mason5a5f79b2010-05-26 21:33:37 -04007260 goto out;
Josef Bacika1b75f72011-04-08 15:51:18 +00007261
7262 /* If this is a write we don't need to check anymore */
7263 if (rw & WRITE)
7264 continue;
7265
7266 /*
7267 * Check to make sure we don't have duplicate iov_base's in this
7268 * iovec, if so return EINVAL, otherwise we'll get csum errors
7269 * when reading back.
7270 */
7271 for (i = seg + 1; i < nr_segs; i++) {
7272 if (iov[seg].iov_base == iov[i].iov_base)
7273 goto out;
7274 }
Chris Mason5a5f79b2010-05-26 21:33:37 -04007275 }
7276 retval = 0;
7277out:
7278 return retval;
7279}
Josef Bacikeb838e72012-07-31 16:28:48 -04007280
Chris Mason16432982008-04-10 10:23:21 -04007281static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7282 const struct iovec *iov, loff_t offset,
7283 unsigned long nr_segs)
7284{
Josef Bacik4b46fce2010-05-23 11:00:55 -04007285 struct file *file = iocb->ki_filp;
7286 struct inode *inode = file->f_mapping->host;
Miao Xie09348562013-02-07 10:12:07 +00007287 size_t count = 0;
Miao Xie2e60a512013-02-08 07:01:08 +00007288 int flags = 0;
Miao Xie38851cc2013-02-08 07:04:11 +00007289 bool wakeup = true;
7290 bool relock = false;
Miao Xie09348562013-02-07 10:12:07 +00007291 ssize_t ret;
Josef Bacik4b46fce2010-05-23 11:00:55 -04007292
Chris Mason5a5f79b2010-05-26 21:33:37 -04007293 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
Josef Bacikeb838e72012-07-31 16:28:48 -04007294 offset, nr_segs))
Chris Mason5a5f79b2010-05-26 21:33:37 -04007295 return 0;
Chris Mason5a5f79b2010-05-26 21:33:37 -04007296
Miao Xie38851cc2013-02-08 07:04:11 +00007297 atomic_inc(&inode->i_dio_count);
7298 smp_mb__after_atomic_inc();
7299
Josef Bacik0e267c42013-07-02 10:38:02 -04007300 /*
7301 * The generic stuff only does filemap_write_and_wait_range, which isn't
7302 * enough if we've written compressed pages to this area, so we need to
7303 * call btrfs_wait_ordered_range to make absolutely sure that any
7304 * outstanding dirty pages are on disk.
7305 */
7306 count = iov_length(iov, nr_segs);
Josef Bacik0ef8b722013-10-25 16:13:35 -04007307 ret = btrfs_wait_ordered_range(inode, offset, count);
7308 if (ret)
7309 return ret;
Josef Bacik0e267c42013-07-02 10:38:02 -04007310
Miao Xie09348562013-02-07 10:12:07 +00007311 if (rw & WRITE) {
Miao Xie38851cc2013-02-08 07:04:11 +00007312 /*
7313 * If the write DIO is beyond the EOF, we need update
7314 * the isize, but it is protected by i_mutex. So we can
7315 * not unlock the i_mutex at this case.
7316 */
7317 if (offset + count <= inode->i_size) {
7318 mutex_unlock(&inode->i_mutex);
7319 relock = true;
7320 }
Miao Xie09348562013-02-07 10:12:07 +00007321 ret = btrfs_delalloc_reserve_space(inode, count);
7322 if (ret)
Miao Xie38851cc2013-02-08 07:04:11 +00007323 goto out;
7324 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7325 &BTRFS_I(inode)->runtime_flags))) {
7326 inode_dio_done(inode);
7327 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7328 wakeup = false;
Miao Xie09348562013-02-07 10:12:07 +00007329 }
7330
7331 ret = __blockdev_direct_IO(rw, iocb, inode,
7332 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7333 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
Miao Xie2e60a512013-02-08 07:01:08 +00007334 btrfs_submit_direct, flags);
Miao Xie09348562013-02-07 10:12:07 +00007335 if (rw & WRITE) {
7336 if (ret < 0 && ret != -EIOCBQUEUED)
7337 btrfs_delalloc_release_space(inode, count);
Miao Xie172a5042013-02-21 02:48:22 -07007338 else if (ret >= 0 && (size_t)ret < count)
Miao Xie09348562013-02-07 10:12:07 +00007339 btrfs_delalloc_release_space(inode,
7340 count - (size_t)ret);
Miao Xie172a5042013-02-21 02:48:22 -07007341 else
7342 btrfs_delalloc_release_metadata(inode, 0);
Miao Xie09348562013-02-07 10:12:07 +00007343 }
Miao Xie38851cc2013-02-08 07:04:11 +00007344out:
Miao Xie2e60a512013-02-08 07:01:08 +00007345 if (wakeup)
7346 inode_dio_done(inode);
Miao Xie38851cc2013-02-08 07:04:11 +00007347 if (relock)
7348 mutex_lock(&inode->i_mutex);
Miao Xie09348562013-02-07 10:12:07 +00007349
7350 return ret;
Chris Mason16432982008-04-10 10:23:21 -04007351}
7352
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00007353#define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7354
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05007355static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7356 __u64 start, __u64 len)
7357{
Tsutomu Itoh05dadc02012-11-29 05:08:26 +00007358 int ret;
7359
7360 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7361 if (ret)
7362 return ret;
7363
Chris Masonec29ed52011-02-23 16:23:20 -05007364 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05007365}
7366
Chris Mason9ebefb182007-06-15 13:50:00 -04007367int btrfs_readpage(struct file *file, struct page *page)
7368{
Chris Masond1310b22008-01-24 16:13:08 -05007369 struct extent_io_tree *tree;
7370 tree = &BTRFS_I(page->mapping->host)->io_tree;
Jan Schmidt8ddc7d92011-06-13 20:02:58 +02007371 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
Chris Mason39279cc2007-06-12 06:35:45 -04007372}
Chris Mason1832a6d2007-12-21 16:27:21 -05007373
Chris Mason39279cc2007-06-12 06:35:45 -04007374static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7375{
Chris Masond1310b22008-01-24 16:13:08 -05007376 struct extent_io_tree *tree;
Chris Masonb888db22007-08-27 16:49:44 -04007377
7378
7379 if (current->flags & PF_MEMALLOC) {
7380 redirty_page_for_writepage(wbc, page);
7381 unlock_page(page);
7382 return 0;
7383 }
Chris Masond1310b22008-01-24 16:13:08 -05007384 tree = &BTRFS_I(page->mapping->host)->io_tree;
Chris Masona52d9a82007-08-27 16:49:44 -04007385 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7386}
Chris Mason39279cc2007-06-12 06:35:45 -04007387
Eric Sandeen48a3b632013-04-25 20:41:01 +00007388static int btrfs_writepages(struct address_space *mapping,
7389 struct writeback_control *wbc)
Chris Masonb293f022007-11-01 19:45:34 -04007390{
Chris Masond1310b22008-01-24 16:13:08 -05007391 struct extent_io_tree *tree;
Chris Mason771ed682008-11-06 22:02:51 -05007392
Chris Masond1310b22008-01-24 16:13:08 -05007393 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Masonb293f022007-11-01 19:45:34 -04007394 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7395}
7396
Chris Mason3ab2fb52007-11-08 10:59:22 -05007397static int
7398btrfs_readpages(struct file *file, struct address_space *mapping,
7399 struct list_head *pages, unsigned nr_pages)
7400{
Chris Masond1310b22008-01-24 16:13:08 -05007401 struct extent_io_tree *tree;
7402 tree = &BTRFS_I(mapping->host)->io_tree;
Chris Mason3ab2fb52007-11-08 10:59:22 -05007403 return extent_readpages(tree, mapping, pages, nr_pages,
7404 btrfs_get_extent);
7405}
Chris Masone6dcd2d2008-07-17 12:53:50 -04007406static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
Chris Masona52d9a82007-08-27 16:49:44 -04007407{
Chris Masond1310b22008-01-24 16:13:08 -05007408 struct extent_io_tree *tree;
7409 struct extent_map_tree *map;
Chris Masona52d9a82007-08-27 16:49:44 -04007410 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -04007411
Chris Masond1310b22008-01-24 16:13:08 -05007412 tree = &BTRFS_I(page->mapping->host)->io_tree;
7413 map = &BTRFS_I(page->mapping->host)->extent_tree;
Chris Mason70dec802008-01-29 09:59:12 -05007414 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
Chris Masona52d9a82007-08-27 16:49:44 -04007415 if (ret == 1) {
7416 ClearPagePrivate(page);
7417 set_page_private(page, 0);
7418 page_cache_release(page);
7419 }
7420 return ret;
7421}
Chris Mason39279cc2007-06-12 06:35:45 -04007422
Chris Masone6dcd2d2008-07-17 12:53:50 -04007423static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7424{
Chris Mason98509cfc2008-09-11 15:51:43 -04007425 if (PageWriteback(page) || PageDirty(page))
7426 return 0;
Yan Zhengb335b002009-02-12 10:06:04 -05007427 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007428}
7429
Lukas Czernerd47992f2013-05-21 23:17:23 -04007430static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7431 unsigned int length)
Chris Masona52d9a82007-08-27 16:49:44 -04007432{
Josef Bacik5fd02042012-05-02 14:00:54 -04007433 struct inode *inode = page->mapping->host;
Chris Masond1310b22008-01-24 16:13:08 -05007434 struct extent_io_tree *tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007435 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00007436 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007437 u64 page_start = page_offset(page);
7438 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00007439 int inode_evicting = inode->i_state & I_FREEING;
Chris Masona52d9a82007-08-27 16:49:44 -04007440
Chris Mason8b62b722009-09-02 16:53:46 -04007441 /*
7442 * we have the page locked, so new writeback can't start,
7443 * and the dirty bit won't be cleared while we are here.
7444 *
7445 * Wait for IO on this page so that we can safely clear
7446 * the PagePrivate2 bit and do ordered accounting
7447 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04007448 wait_on_page_writeback(page);
Chris Mason8b62b722009-09-02 16:53:46 -04007449
Josef Bacik5fd02042012-05-02 14:00:54 -04007450 tree = &BTRFS_I(inode)->io_tree;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007451 if (offset) {
7452 btrfs_releasepage(page, GFP_NOFS);
7453 return;
7454 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00007455
7456 if (!inode_evicting)
7457 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7458 ordered = btrfs_lookup_ordered_extent(inode, page_start);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007459 if (ordered) {
Chris Masoneb84ae02008-07-17 13:53:27 -04007460 /*
7461 * IO on this page will never be started, so we need
7462 * to account for any ordered extents now
7463 */
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00007464 if (!inode_evicting)
7465 clear_extent_bit(tree, page_start, page_end,
7466 EXTENT_DIRTY | EXTENT_DELALLOC |
7467 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7468 EXTENT_DEFRAG, 1, 0, &cached_state,
7469 GFP_NOFS);
Chris Mason8b62b722009-09-02 16:53:46 -04007470 /*
7471 * whoever cleared the private bit is responsible
7472 * for the finish_ordered_io
7473 */
Josef Bacik77cef2e2013-08-29 13:57:21 -04007474 if (TestClearPagePrivate2(page)) {
7475 struct btrfs_ordered_inode_tree *tree;
7476 u64 new_len;
7477
7478 tree = &BTRFS_I(inode)->ordered_tree;
7479
7480 spin_lock_irq(&tree->lock);
7481 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7482 new_len = page_start - ordered->file_offset;
7483 if (new_len < ordered->truncated_len)
7484 ordered->truncated_len = new_len;
7485 spin_unlock_irq(&tree->lock);
7486
7487 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7488 page_start,
7489 PAGE_CACHE_SIZE, 1))
7490 btrfs_finish_ordered_io(ordered);
Chris Mason8b62b722009-09-02 16:53:46 -04007491 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04007492 btrfs_put_ordered_extent(ordered);
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00007493 if (!inode_evicting) {
7494 cached_state = NULL;
7495 lock_extent_bits(tree, page_start, page_end, 0,
7496 &cached_state);
7497 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04007498 }
Filipe David Borba Manana131e4042013-11-19 22:29:35 +00007499
7500 if (!inode_evicting) {
7501 clear_extent_bit(tree, page_start, page_end,
7502 EXTENT_LOCKED | EXTENT_DIRTY |
7503 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7504 EXTENT_DEFRAG, 1, 1,
7505 &cached_state, GFP_NOFS);
7506
7507 __btrfs_releasepage(page, GFP_NOFS);
7508 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04007509
Chris Mason4a096752008-07-21 10:29:44 -04007510 ClearPageChecked(page);
Chris Mason9ad6b7b2008-04-18 16:11:30 -04007511 if (PagePrivate(page)) {
Chris Mason9ad6b7b2008-04-18 16:11:30 -04007512 ClearPagePrivate(page);
7513 set_page_private(page, 0);
7514 page_cache_release(page);
7515 }
Chris Mason39279cc2007-06-12 06:35:45 -04007516}
7517
Chris Mason9ebefb182007-06-15 13:50:00 -04007518/*
7519 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7520 * called from a page fault handler when a page is first dirtied. Hence we must
7521 * be careful to check for EOF conditions here. We set the page up correctly
7522 * for a written page which means we get ENOSPC checking when writing into
7523 * holes and correct delalloc and unwritten extent mapping on filesystems that
7524 * support these features.
7525 *
7526 * We are not allowed to take the i_mutex here so we have to play games to
7527 * protect against truncate races as the page could now be beyond EOF. Because
7528 * vmtruncate() writes the inode size before removing pages, once we have the
7529 * page lock we can determine safely if the page is beyond EOF. If it is not
7530 * beyond EOF, then the page is guaranteed safe against truncation until we
7531 * unlock the page.
7532 */
Nick Pigginc2ec1752009-03-31 15:23:21 -07007533int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Chris Mason9ebefb182007-06-15 13:50:00 -04007534{
Nick Pigginc2ec1752009-03-31 15:23:21 -07007535 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05007536 struct inode *inode = file_inode(vma->vm_file);
Chris Mason1832a6d2007-12-21 16:27:21 -05007537 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007538 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7539 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +00007540 struct extent_state *cached_state = NULL;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007541 char *kaddr;
7542 unsigned long zero_start;
Chris Mason9ebefb182007-06-15 13:50:00 -04007543 loff_t size;
Chris Mason1832a6d2007-12-21 16:27:21 -05007544 int ret;
Chris Mason9998eb72012-01-25 13:47:40 -05007545 int reserved = 0;
Chris Masona52d9a82007-08-27 16:49:44 -04007546 u64 page_start;
Chris Masone6dcd2d2008-07-17 12:53:50 -04007547 u64 page_end;
Chris Mason9ebefb182007-06-15 13:50:00 -04007548
Jan Karab2b5ef52012-06-12 16:20:45 +02007549 sb_start_pagefault(inode->i_sb);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04007550 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05007551 if (!ret) {
Josef Bacike41f9412012-03-26 09:46:47 -04007552 ret = file_update_time(vma->vm_file);
Chris Mason9998eb72012-01-25 13:47:40 -05007553 reserved = 1;
7554 }
Nick Piggin56a76f82009-03-31 15:23:23 -07007555 if (ret) {
7556 if (ret == -ENOMEM)
7557 ret = VM_FAULT_OOM;
7558 else /* -ENOSPC, -EIO, etc */
7559 ret = VM_FAULT_SIGBUS;
Chris Mason9998eb72012-01-25 13:47:40 -05007560 if (reserved)
7561 goto out;
7562 goto out_noreserve;
Nick Piggin56a76f82009-03-31 15:23:23 -07007563 }
Chris Mason1832a6d2007-12-21 16:27:21 -05007564
Nick Piggin56a76f82009-03-31 15:23:23 -07007565 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
Chris Masone6dcd2d2008-07-17 12:53:50 -04007566again:
Chris Mason9ebefb182007-06-15 13:50:00 -04007567 lock_page(page);
Chris Mason9ebefb182007-06-15 13:50:00 -04007568 size = i_size_read(inode);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007569 page_start = page_offset(page);
7570 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masona52d9a82007-08-27 16:49:44 -04007571
Chris Mason9ebefb182007-06-15 13:50:00 -04007572 if ((page->mapping != inode->i_mapping) ||
Chris Masone6dcd2d2008-07-17 12:53:50 -04007573 (page_start >= size)) {
Chris Mason9ebefb182007-06-15 13:50:00 -04007574 /* page got truncated out from underneath us */
7575 goto out_unlock;
7576 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04007577 wait_on_page_writeback(page);
7578
Jeff Mahoneyd0082372012-03-01 14:57:19 +01007579 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007580 set_page_extent_mapped(page);
7581
Chris Masoneb84ae02008-07-17 13:53:27 -04007582 /*
7583 * we can't set the delalloc bits if there are pending ordered
7584 * extents. Drop our locks and wait for them to finish
7585 */
Chris Masone6dcd2d2008-07-17 12:53:50 -04007586 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7587 if (ordered) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00007588 unlock_extent_cached(io_tree, page_start, page_end,
7589 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007590 unlock_page(page);
Chris Masoneb84ae02008-07-17 13:53:27 -04007591 btrfs_start_ordered_extent(inode, ordered, 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007592 btrfs_put_ordered_extent(ordered);
7593 goto again;
7594 }
7595
Josef Bacikfbf19082009-10-01 17:10:23 -04007596 /*
7597 * XXX - page_mkwrite gets called every time the page is dirtied, even
7598 * if it was already dirty, so for space accounting reasons we need to
7599 * clear any delalloc bits for the range we are fixing to save. There
7600 * is probably a better way to do this, but for now keep consistent with
7601 * prepare_pages in the normal write path.
7602 */
Josef Bacik2ac55d42010-02-03 19:33:23 +00007603 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
Liu Bo9e8a4a82012-09-05 19:10:51 -06007604 EXTENT_DIRTY | EXTENT_DELALLOC |
7605 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
Josef Bacik2ac55d42010-02-03 19:33:23 +00007606 0, 0, &cached_state, GFP_NOFS);
Josef Bacikfbf19082009-10-01 17:10:23 -04007607
Josef Bacik2ac55d42010-02-03 19:33:23 +00007608 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7609 &cached_state);
Josef Bacik9ed74f22009-09-11 16:12:44 -04007610 if (ret) {
Josef Bacik2ac55d42010-02-03 19:33:23 +00007611 unlock_extent_cached(io_tree, page_start, page_end,
7612 &cached_state, GFP_NOFS);
Josef Bacik9ed74f22009-09-11 16:12:44 -04007613 ret = VM_FAULT_SIGBUS;
7614 goto out_unlock;
7615 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04007616 ret = 0;
Chris Mason9ebefb182007-06-15 13:50:00 -04007617
7618 /* page is wholly or partially inside EOF */
Chris Masona52d9a82007-08-27 16:49:44 -04007619 if (page_start + PAGE_CACHE_SIZE > size)
Chris Masone6dcd2d2008-07-17 12:53:50 -04007620 zero_start = size & ~PAGE_CACHE_MASK;
Chris Mason9ebefb182007-06-15 13:50:00 -04007621 else
Chris Masone6dcd2d2008-07-17 12:53:50 -04007622 zero_start = PAGE_CACHE_SIZE;
Chris Mason9ebefb182007-06-15 13:50:00 -04007623
Chris Masone6dcd2d2008-07-17 12:53:50 -04007624 if (zero_start != PAGE_CACHE_SIZE) {
7625 kaddr = kmap(page);
7626 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7627 flush_dcache_page(page);
7628 kunmap(page);
7629 }
Chris Mason247e7432008-07-17 12:53:51 -04007630 ClearPageChecked(page);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007631 set_page_dirty(page);
Chris Mason50a9b212009-09-11 12:33:12 -04007632 SetPageUptodate(page);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007633
Chris Mason257c62e2009-10-13 13:21:08 -04007634 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7635 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
Liu Bo46d8bc32012-08-29 01:07:55 -06007636 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
Chris Mason257c62e2009-10-13 13:21:08 -04007637
Josef Bacik2ac55d42010-02-03 19:33:23 +00007638 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
Chris Mason9ebefb182007-06-15 13:50:00 -04007639
7640out_unlock:
Jan Karab2b5ef52012-06-12 16:20:45 +02007641 if (!ret) {
7642 sb_end_pagefault(inode->i_sb);
Chris Mason50a9b212009-09-11 12:33:12 -04007643 return VM_FAULT_LOCKED;
Jan Karab2b5ef52012-06-12 16:20:45 +02007644 }
Chris Mason9ebefb182007-06-15 13:50:00 -04007645 unlock_page(page);
Chris Mason1832a6d2007-12-21 16:27:21 -05007646out:
Josef Bacikec39e182012-01-12 19:10:12 -05007647 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
Chris Mason9998eb72012-01-25 13:47:40 -05007648out_noreserve:
Jan Karab2b5ef52012-06-12 16:20:45 +02007649 sb_end_pagefault(inode->i_sb);
Chris Mason9ebefb182007-06-15 13:50:00 -04007650 return ret;
7651}
7652
Josef Bacika41ad392011-01-31 15:30:16 -05007653static int btrfs_truncate(struct inode *inode)
Chris Mason39279cc2007-06-12 06:35:45 -04007654{
7655 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikfcb80c22011-05-03 10:40:22 -04007656 struct btrfs_block_rsv *rsv;
Josef Bacika71754f2013-06-17 17:14:39 -04007657 int ret = 0;
Josef Bacik3893e332011-01-31 16:03:11 -05007658 int err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007659 struct btrfs_trans_handle *trans;
Chris Masondbe674a2008-07-17 12:54:05 -04007660 u64 mask = root->sectorsize - 1;
Josef Bacik07127182011-08-19 10:29:59 -04007661 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
Chris Mason39279cc2007-06-12 06:35:45 -04007662
Josef Bacik0ef8b722013-10-25 16:13:35 -04007663 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7664 (u64)-1);
7665 if (ret)
7666 return ret;
Chris Mason39279cc2007-06-12 06:35:45 -04007667
Josef Bacikfcb80c22011-05-03 10:40:22 -04007668 /*
7669 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7670 * 3 things going on here
7671 *
7672 * 1) We need to reserve space for our orphan item and the space to
7673 * delete our orphan item. Lord knows we don't want to have a dangling
7674 * orphan item because we didn't reserve space to remove it.
7675 *
7676 * 2) We need to reserve space to update our inode.
7677 *
7678 * 3) We need to have something to cache all the space that is going to
7679 * be free'd up by the truncate operation, but also have some slack
7680 * space reserved in case it uses space during the truncate (thank you
7681 * very much snapshotting).
7682 *
7683 * And we need these to all be seperate. The fact is we can use alot of
7684 * space doing the truncate, and we have no earthly idea how much space
7685 * we will use, so we need the truncate reservation to be seperate so it
7686 * doesn't end up using space reserved for updating the inode or
7687 * removing the orphan item. We also need to be able to stop the
7688 * transaction and start a new one, which means we need to be able to
7689 * update the inode several times, and we have no idea of knowing how
7690 * many times that will be, so we can't just reserve 1 item for the
7691 * entirety of the opration, so that has to be done seperately as well.
7692 * Then there is the orphan item, which does indeed need to be held on
7693 * to for the whole operation, and we need nobody to touch this reserved
7694 * space except the orphan code.
7695 *
7696 * So that leaves us with
7697 *
7698 * 1) root->orphan_block_rsv - for the orphan deletion.
7699 * 2) rsv - for the truncate reservation, which we will steal from the
7700 * transaction reservation.
7701 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7702 * updating the inode.
7703 */
Miao Xie66d8f3d2012-09-06 04:02:28 -06007704 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
Josef Bacikfcb80c22011-05-03 10:40:22 -04007705 if (!rsv)
7706 return -ENOMEM;
Josef Bacik4a338542011-08-29 11:01:31 -04007707 rsv->size = min_size;
Josef Bacikca7e70f2012-08-27 17:48:15 -04007708 rsv->failfast = 1;
Josef Bacikf0cd8462011-03-04 14:37:08 -05007709
Josef Bacik907cbce2011-08-08 13:46:15 -04007710 /*
Josef Bacik07127182011-08-19 10:29:59 -04007711 * 1 for the truncate slack space
Josef Bacik907cbce2011-08-08 13:46:15 -04007712 * 1 for updating the inode.
7713 */
Josef Bacikf3fe8202013-01-07 17:03:21 -05007714 trans = btrfs_start_transaction(root, 2);
Josef Bacikfcb80c22011-05-03 10:40:22 -04007715 if (IS_ERR(trans)) {
7716 err = PTR_ERR(trans);
7717 goto out;
7718 }
Josef Bacikf0cd8462011-03-04 14:37:08 -05007719
Josef Bacik907cbce2011-08-08 13:46:15 -04007720 /* Migrate the slack space for the truncate to our reserve */
7721 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7722 min_size);
Josef Bacikfcb80c22011-05-03 10:40:22 -04007723 BUG_ON(ret);
Josef Bacikf0cd8462011-03-04 14:37:08 -05007724
Chris Mason5a3f23d2009-03-31 13:27:11 -04007725 /*
7726 * setattr is responsible for setting the ordered_data_close flag,
7727 * but that is only tested during the last file release. That
7728 * could happen well after the next commit, leaving a great big
7729 * window where new writes may get lost if someone chooses to write
7730 * to this file after truncating to zero
7731 *
7732 * The inode doesn't have any dirty data here, and so if we commit
7733 * this is a noop. If someone immediately starts writing to the inode
7734 * it is very likely we'll catch some of their writes in this
7735 * transaction, and the commit will find this file on the ordered
7736 * data list with good things to send down.
7737 *
7738 * This is a best effort solution, there is still a window where
7739 * using truncate to replace the contents of the file will
7740 * end up with a zero length file after a crash.
7741 */
Josef Bacik72ac3c02012-05-23 14:13:11 -04007742 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7743 &BTRFS_I(inode)->runtime_flags))
Chris Mason5a3f23d2009-03-31 13:27:11 -04007744 btrfs_add_ordered_operation(trans, root, inode);
7745
Josef Bacik5dc562c2012-08-17 13:14:17 -04007746 /*
7747 * So if we truncate and then write and fsync we normally would just
7748 * write the extents that changed, which is a problem if we need to
7749 * first truncate that entire inode. So set this flag so we write out
7750 * all of the extents in the inode to the sync log so we're completely
7751 * safe.
7752 */
7753 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
Josef Bacikca7e70f2012-08-27 17:48:15 -04007754 trans->block_rsv = rsv;
Josef Bacik5dc562c2012-08-17 13:14:17 -04007755
Yan, Zheng80825102009-11-12 09:35:36 +00007756 while (1) {
7757 ret = btrfs_truncate_inode_items(trans, root, inode,
7758 inode->i_size,
7759 BTRFS_EXTENT_DATA_KEY);
Josef Bacikca7e70f2012-08-27 17:48:15 -04007760 if (ret != -ENOSPC) {
Josef Bacik3893e332011-01-31 16:03:11 -05007761 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00007762 break;
Josef Bacik3893e332011-01-31 16:03:11 -05007763 }
Chris Mason39279cc2007-06-12 06:35:45 -04007764
Josef Bacikfcb80c22011-05-03 10:40:22 -04007765 trans->block_rsv = &root->fs_info->trans_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00007766 ret = btrfs_update_inode(trans, root, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05007767 if (ret) {
7768 err = ret;
7769 break;
7770 }
Josef Bacikca7e70f2012-08-27 17:48:15 -04007771
Yan, Zheng80825102009-11-12 09:35:36 +00007772 btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00007773 btrfs_btree_balance_dirty(root);
Josef Bacikca7e70f2012-08-27 17:48:15 -04007774
7775 trans = btrfs_start_transaction(root, 2);
7776 if (IS_ERR(trans)) {
7777 ret = err = PTR_ERR(trans);
7778 trans = NULL;
7779 break;
7780 }
7781
7782 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7783 rsv, min_size);
7784 BUG_ON(ret); /* shouldn't happen */
7785 trans->block_rsv = rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00007786 }
7787
7788 if (ret == 0 && inode->i_nlink > 0) {
Josef Bacikfcb80c22011-05-03 10:40:22 -04007789 trans->block_rsv = root->orphan_block_rsv;
Yan, Zheng80825102009-11-12 09:35:36 +00007790 ret = btrfs_orphan_del(trans, inode);
Josef Bacik3893e332011-01-31 16:03:11 -05007791 if (ret)
7792 err = ret;
Yan, Zheng80825102009-11-12 09:35:36 +00007793 }
7794
Chris Mason917c16b2011-11-08 14:49:59 -05007795 if (trans) {
7796 trans->block_rsv = &root->fs_info->trans_block_rsv;
7797 ret = btrfs_update_inode(trans, root, inode);
7798 if (ret && !err)
7799 err = ret;
Josef Bacik7b128762008-07-24 12:17:14 -04007800
Josef Bacik7ad85bb2012-01-12 19:10:12 -05007801 ret = btrfs_end_transaction(trans, root);
Liu Bob53d3f52012-11-14 14:34:34 +00007802 btrfs_btree_balance_dirty(root);
Chris Mason917c16b2011-11-08 14:49:59 -05007803 }
Josef Bacikfcb80c22011-05-03 10:40:22 -04007804
7805out:
7806 btrfs_free_block_rsv(root, rsv);
7807
Josef Bacik3893e332011-01-31 16:03:11 -05007808 if (ret && !err)
7809 err = ret;
Josef Bacika41ad392011-01-31 15:30:16 -05007810
Josef Bacik3893e332011-01-31 16:03:11 -05007811 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04007812}
7813
Sven Wegener3b963622008-06-09 21:57:42 -04007814/*
Chris Masond352ac62008-09-29 15:18:18 -04007815 * create a new subvolume directory/inode (helper for the ioctl).
7816 */
Yan Zhengd2fb3432008-12-11 16:30:39 -05007817int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
Josef Bacikd82a6f12011-05-11 15:26:06 -04007818 struct btrfs_root *new_root, u64 new_dirid)
Chris Mason39279cc2007-06-12 06:35:45 -04007819{
Chris Mason39279cc2007-06-12 06:35:45 -04007820 struct inode *inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04007821 int err;
Chris Mason00e4e6b2008-08-05 11:18:09 -04007822 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04007823
Florian Albrechtskirchinger12fc9d02012-02-10 22:15:54 +01007824 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7825 new_dirid, new_dirid,
7826 S_IFDIR | (~current_umask() & S_IRWXUGO),
7827 &index);
Chris Mason54aa1f42007-06-22 14:16:25 -04007828 if (IS_ERR(inode))
Christoph Hellwigf46b5a62008-06-11 21:53:53 -04007829 return PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04007830 inode->i_op = &btrfs_dir_inode_operations;
7831 inode->i_fop = &btrfs_dir_file_operations;
7832
Miklos Szeredibfe86842011-10-28 14:13:29 +02007833 set_nlink(inode, 1);
Chris Masondbe674a2008-07-17 12:54:05 -04007834 btrfs_i_size_write(inode, 0);
Sven Wegener3b963622008-06-09 21:57:42 -04007835
Yan, Zheng76dda932009-09-21 16:00:26 -04007836 err = btrfs_update_inode(trans, new_root, inode);
Christoph Hellwigcb8e7092008-10-09 13:39:39 -04007837
Yan, Zheng76dda932009-09-21 16:00:26 -04007838 iput(inode);
Mark Fashehce598972011-07-26 11:32:23 -07007839 return err;
Chris Mason39279cc2007-06-12 06:35:45 -04007840}
7841
Chris Mason39279cc2007-06-12 06:35:45 -04007842struct inode *btrfs_alloc_inode(struct super_block *sb)
7843{
7844 struct btrfs_inode *ei;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007845 struct inode *inode;
Chris Mason39279cc2007-06-12 06:35:45 -04007846
7847 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7848 if (!ei)
7849 return NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007850
7851 ei->root = NULL;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007852 ei->generation = 0;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04007853 ei->last_trans = 0;
Chris Mason257c62e2009-10-13 13:21:08 -04007854 ei->last_sub_trans = 0;
Chris Masone02119d2008-09-05 16:13:11 -04007855 ei->logged_trans = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007856 ei->delalloc_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007857 ei->disk_i_size = 0;
7858 ei->flags = 0;
Josef Bacik7709cde2011-08-04 10:25:02 -04007859 ei->csum_bytes = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007860 ei->index_cnt = (u64)-1;
7861 ei->last_unlink_trans = 0;
Liu Bo46d8bc32012-08-29 01:07:55 -06007862 ei->last_log_commit = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007863
Josef Bacik9e0baf62011-07-15 15:16:44 +00007864 spin_lock_init(&ei->lock);
7865 ei->outstanding_extents = 0;
7866 ei->reserved_extents = 0;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007867
Josef Bacik72ac3c02012-05-23 14:13:11 -04007868 ei->runtime_flags = 0;
Li Zefan261507a02010-12-17 14:21:50 +08007869 ei->force_compress = BTRFS_COMPRESS_NONE;
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007870
Miao Xie16cdcec2011-04-22 18:12:22 +08007871 ei->delayed_node = NULL;
7872
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007873 inode = &ei->vfs_inode;
David Sterbaa8067e02011-04-21 00:34:43 +02007874 extent_map_tree_init(&ei->extent_tree);
David Sterbaf993c882011-04-20 23:35:57 +02007875 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7876 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
Josef Bacik0b32f4b2012-03-13 09:38:00 -04007877 ei->io_tree.track_uptodate = 1;
7878 ei->io_failure_tree.track_uptodate = 1;
Josef Bacikb812ce22012-11-16 13:56:32 -05007879 atomic_set(&ei->sync_writers, 0);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007880 mutex_init(&ei->log_mutex);
Josef Bacikf2486792012-01-13 12:09:22 -05007881 mutex_init(&ei->delalloc_mutex);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007882 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007883 INIT_LIST_HEAD(&ei->delalloc_inodes);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007884 INIT_LIST_HEAD(&ei->ordered_operations);
Yan, Zheng2ead6ae2010-05-16 10:46:25 -04007885 RB_CLEAR_NODE(&ei->rb_node);
7886
7887 return inode;
Chris Mason39279cc2007-06-12 06:35:45 -04007888}
7889
Josef Bacikaaedb552013-10-11 14:44:09 -04007890#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7891void btrfs_test_destroy_inode(struct inode *inode)
7892{
7893 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7894 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7895}
7896#endif
7897
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11007898static void btrfs_i_callback(struct rcu_head *head)
7899{
7900 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11007901 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7902}
7903
Chris Mason39279cc2007-06-12 06:35:45 -04007904void btrfs_destroy_inode(struct inode *inode)
7905{
Chris Masone6dcd2d2008-07-17 12:53:50 -04007906 struct btrfs_ordered_extent *ordered;
Chris Mason5a3f23d2009-03-31 13:27:11 -04007907 struct btrfs_root *root = BTRFS_I(inode)->root;
7908
Al Virob3d9b7a2012-06-09 13:51:19 -04007909 WARN_ON(!hlist_empty(&inode->i_dentry));
Chris Mason39279cc2007-06-12 06:35:45 -04007910 WARN_ON(inode->i_data.nrpages);
Josef Bacik9e0baf62011-07-15 15:16:44 +00007911 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7912 WARN_ON(BTRFS_I(inode)->reserved_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04007913 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7914 WARN_ON(BTRFS_I(inode)->csum_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04007915
Chris Mason5a3f23d2009-03-31 13:27:11 -04007916 /*
Josef Bacika6dbd422009-11-11 15:53:34 -05007917 * This can happen where we create an inode, but somebody else also
7918 * created the same inode and we need to destroy the one we already
7919 * created.
7920 */
7921 if (!root)
7922 goto free;
7923
7924 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04007925 * Make sure we're properly removed from the ordered operation
7926 * lists.
7927 */
7928 smp_mb();
7929 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
Miao Xie199c2a92013-05-15 07:48:23 +00007930 spin_lock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007931 list_del_init(&BTRFS_I(inode)->ordered_operations);
Miao Xie199c2a92013-05-15 07:48:23 +00007932 spin_unlock(&root->fs_info->ordered_root_lock);
Chris Mason5a3f23d2009-03-31 13:27:11 -04007933 }
7934
Josef Bacik8a35d952012-05-23 14:26:42 -04007935 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7936 &BTRFS_I(inode)->runtime_flags)) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00007937 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02007938 btrfs_ino(inode));
Josef Bacik8a35d952012-05-23 14:26:42 -04007939 atomic_dec(&root->orphan_inodes);
Josef Bacik7b128762008-07-24 12:17:14 -04007940 }
Josef Bacik7b128762008-07-24 12:17:14 -04007941
Chris Masond3977122009-01-05 21:25:51 -05007942 while (1) {
Chris Masone6dcd2d2008-07-17 12:53:50 -04007943 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7944 if (!ordered)
7945 break;
7946 else {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00007947 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02007948 ordered->file_offset, ordered->len);
Chris Masone6dcd2d2008-07-17 12:53:50 -04007949 btrfs_remove_ordered_extent(inode, ordered);
7950 btrfs_put_ordered_extent(ordered);
7951 btrfs_put_ordered_extent(ordered);
7952 }
7953 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007954 inode_tree_del(inode);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04007955 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
Josef Bacika6dbd422009-11-11 15:53:34 -05007956free:
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11007957 call_rcu(&inode->i_rcu, btrfs_i_callback);
Chris Mason39279cc2007-06-12 06:35:45 -04007958}
7959
Al Viro45321ac2010-06-07 13:43:19 -04007960int btrfs_drop_inode(struct inode *inode)
Yan, Zheng76dda932009-09-21 16:00:26 -04007961{
7962 struct btrfs_root *root = BTRFS_I(inode)->root;
Al Viro45321ac2010-06-07 13:43:19 -04007963
Naohiro Aota6379ef92013-06-06 09:56:34 +00007964 if (root == NULL)
7965 return 1;
7966
Liu Bofa6ac872013-02-20 14:10:23 +00007967 /* the snap/subvol tree is on deleting */
Stefan Behrens69e9c6c2013-09-05 16:58:43 +02007968 if (btrfs_root_refs(&root->root_item) == 0)
Al Viro45321ac2010-06-07 13:43:19 -04007969 return 1;
Yan, Zheng76dda932009-09-21 16:00:26 -04007970 else
Al Viro45321ac2010-06-07 13:43:19 -04007971 return generic_drop_inode(inode);
Yan, Zheng76dda932009-09-21 16:00:26 -04007972}
7973
Sven Wegener0ee0fda2008-07-30 16:54:26 -04007974static void init_once(void *foo)
Chris Mason39279cc2007-06-12 06:35:45 -04007975{
7976 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7977
7978 inode_init_once(&ei->vfs_inode);
7979}
7980
7981void btrfs_destroy_cachep(void)
7982{
Kirill A. Shutemov8c0a8532012-09-26 11:33:07 +10007983 /*
7984 * Make sure all delayed rcu free inodes are flushed before we
7985 * destroy cache.
7986 */
7987 rcu_barrier();
Chris Mason39279cc2007-06-12 06:35:45 -04007988 if (btrfs_inode_cachep)
7989 kmem_cache_destroy(btrfs_inode_cachep);
7990 if (btrfs_trans_handle_cachep)
7991 kmem_cache_destroy(btrfs_trans_handle_cachep);
7992 if (btrfs_transaction_cachep)
7993 kmem_cache_destroy(btrfs_transaction_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04007994 if (btrfs_path_cachep)
7995 kmem_cache_destroy(btrfs_path_cachep);
Josef Bacikdc89e982011-01-28 17:05:48 -05007996 if (btrfs_free_space_cachep)
7997 kmem_cache_destroy(btrfs_free_space_cachep);
Miao Xie8ccf6f192012-10-25 09:28:04 +00007998 if (btrfs_delalloc_work_cachep)
7999 kmem_cache_destroy(btrfs_delalloc_work_cachep);
Chris Mason39279cc2007-06-12 06:35:45 -04008000}
8001
8002int btrfs_init_cachep(void)
8003{
David Sterba837e1972012-09-07 03:00:48 -06008004 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008005 sizeof(struct btrfs_inode), 0,
8006 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
Chris Mason39279cc2007-06-12 06:35:45 -04008007 if (!btrfs_inode_cachep)
8008 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008009
David Sterba837e1972012-09-07 03:00:48 -06008010 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008011 sizeof(struct btrfs_trans_handle), 0,
8012 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04008013 if (!btrfs_trans_handle_cachep)
8014 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008015
David Sterba837e1972012-09-07 03:00:48 -06008016 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008017 sizeof(struct btrfs_transaction), 0,
8018 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04008019 if (!btrfs_transaction_cachep)
8020 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008021
David Sterba837e1972012-09-07 03:00:48 -06008022 btrfs_path_cachep = kmem_cache_create("btrfs_path",
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008023 sizeof(struct btrfs_path), 0,
8024 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
Chris Mason39279cc2007-06-12 06:35:45 -04008025 if (!btrfs_path_cachep)
8026 goto fail;
Christoph Hellwig9601e3f2009-04-13 15:33:09 +02008027
David Sterba837e1972012-09-07 03:00:48 -06008028 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
Josef Bacikdc89e982011-01-28 17:05:48 -05008029 sizeof(struct btrfs_free_space), 0,
8030 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8031 if (!btrfs_free_space_cachep)
8032 goto fail;
8033
Miao Xie8ccf6f192012-10-25 09:28:04 +00008034 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8035 sizeof(struct btrfs_delalloc_work), 0,
8036 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8037 NULL);
8038 if (!btrfs_delalloc_work_cachep)
8039 goto fail;
8040
Chris Mason39279cc2007-06-12 06:35:45 -04008041 return 0;
8042fail:
8043 btrfs_destroy_cachep();
8044 return -ENOMEM;
8045}
8046
8047static int btrfs_getattr(struct vfsmount *mnt,
8048 struct dentry *dentry, struct kstat *stat)
8049{
Miao Xiedf0af1a2013-01-29 10:11:59 +00008050 u64 delalloc_bytes;
Chris Mason39279cc2007-06-12 06:35:45 -04008051 struct inode *inode = dentry->d_inode;
David Sterbafadc0d82011-11-20 07:33:38 -05008052 u32 blocksize = inode->i_sb->s_blocksize;
8053
Chris Mason39279cc2007-06-12 06:35:45 -04008054 generic_fillattr(inode, stat);
Al Viro0ee5dc62011-07-07 15:44:25 -04008055 stat->dev = BTRFS_I(inode)->root->anon_dev;
Chris Masond6667462008-01-03 14:51:00 -05008056 stat->blksize = PAGE_CACHE_SIZE;
Miao Xiedf0af1a2013-01-29 10:11:59 +00008057
8058 spin_lock(&BTRFS_I(inode)->lock);
8059 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8060 spin_unlock(&BTRFS_I(inode)->lock);
David Sterbafadc0d82011-11-20 07:33:38 -05008061 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
Miao Xiedf0af1a2013-01-29 10:11:59 +00008062 ALIGN(delalloc_bytes, blocksize)) >> 9;
Chris Mason39279cc2007-06-12 06:35:45 -04008063 return 0;
8064}
8065
Chris Masond3977122009-01-05 21:25:51 -05008066static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8067 struct inode *new_dir, struct dentry *new_dentry)
Chris Mason39279cc2007-06-12 06:35:45 -04008068{
8069 struct btrfs_trans_handle *trans;
8070 struct btrfs_root *root = BTRFS_I(old_dir)->root;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008071 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
Chris Mason39279cc2007-06-12 06:35:45 -04008072 struct inode *new_inode = new_dentry->d_inode;
8073 struct inode *old_inode = old_dentry->d_inode;
8074 struct timespec ctime = CURRENT_TIME;
Chris Mason00e4e6b2008-08-05 11:18:09 -04008075 u64 index = 0;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008076 u64 root_objectid;
Chris Mason39279cc2007-06-12 06:35:45 -04008077 int ret;
Li Zefan33345d012011-04-20 10:31:50 +08008078 u64 old_ino = btrfs_ino(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008079
Li Zefan33345d012011-04-20 10:31:50 +08008080 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
Yan, Zhengf679a842009-09-24 09:17:31 -04008081 return -EPERM;
8082
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008083 /* we only allow rename subvolume link between subvolumes */
Li Zefan33345d012011-04-20 10:31:50 +08008084 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
Chris Mason3394e162008-11-17 20:42:26 -05008085 return -EXDEV;
8086
Li Zefan33345d012011-04-20 10:31:50 +08008087 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8088 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008089 return -ENOTEMPTY;
8090
Chris Mason39279cc2007-06-12 06:35:45 -04008091 if (S_ISDIR(old_inode->i_mode) && new_inode &&
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008092 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
Chris Mason39279cc2007-06-12 06:35:45 -04008093 return -ENOTEMPTY;
Chris Mason9c520572012-12-17 14:26:57 -05008094
8095
8096 /* check for collisions, even if the name isn't there */
Josef Bacik4871c152013-10-09 12:24:04 -04008097 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
Chris Mason9c520572012-12-17 14:26:57 -05008098 new_dentry->d_name.name,
8099 new_dentry->d_name.len);
8100
8101 if (ret) {
8102 if (ret == -EEXIST) {
8103 /* we shouldn't get
8104 * eexist without a new_inode */
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05308105 if (WARN_ON(!new_inode)) {
Chris Mason9c520572012-12-17 14:26:57 -05008106 return ret;
8107 }
8108 } else {
8109 /* maybe -EOVERFLOW */
8110 return ret;
8111 }
8112 }
8113 ret = 0;
8114
Chris Mason5a3f23d2009-03-31 13:27:11 -04008115 /*
8116 * we're using rename to replace one file with another.
8117 * and the replacement file is large. Start IO on it now so
8118 * we don't add too much work to the end of the transaction
8119 */
Bartlomiej Zolnierkiewicz4baf8c92009-08-07 13:47:08 -04008120 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Chris Mason5a3f23d2009-03-31 13:27:11 -04008121 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8122 filemap_flush(old_inode->i_mapping);
8123
Yan, Zheng76dda932009-09-21 16:00:26 -04008124 /* close the racy window with snapshot create/destroy ioctl */
Li Zefan33345d012011-04-20 10:31:50 +08008125 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04008126 down_read(&root->fs_info->subvol_sem);
Yan, Zhenga22285a2010-05-16 10:48:46 -04008127 /*
8128 * We want to reserve the absolute worst case amount of items. So if
8129 * both inodes are subvols and we need to unlink them then that would
8130 * require 4 item modifications, but if they are both normal inodes it
8131 * would require 5 item modifications, so we'll assume their normal
8132 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8133 * should cover the worst case number of items we'll modify.
8134 */
Josef Bacik6e137ed2013-03-26 15:26:55 -04008135 trans = btrfs_start_transaction(root, 11);
Johann Lombardib44c59a2011-03-31 13:23:47 +00008136 if (IS_ERR(trans)) {
8137 ret = PTR_ERR(trans);
8138 goto out_notrans;
8139 }
Chris Mason5f39d392007-10-15 16:14:19 -04008140
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008141 if (dest != root)
8142 btrfs_record_root_in_trans(trans, dest);
8143
Yan, Zhenga5719522009-09-24 09:17:31 -04008144 ret = btrfs_set_inode_index(new_dir, &index);
8145 if (ret)
8146 goto out_fail;
Chris Mason5a3f23d2009-03-31 13:27:11 -04008147
Li Zefan33345d012011-04-20 10:31:50 +08008148 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008149 /* force full log commit if subvolume involved. */
8150 root->fs_info->last_trans_log_full_commit = trans->transid;
8151 } else {
Yan, Zhenga5719522009-09-24 09:17:31 -04008152 ret = btrfs_insert_inode_ref(trans, dest,
8153 new_dentry->d_name.name,
8154 new_dentry->d_name.len,
Li Zefan33345d012011-04-20 10:31:50 +08008155 old_ino,
8156 btrfs_ino(new_dir), index);
Yan, Zhenga5719522009-09-24 09:17:31 -04008157 if (ret)
8158 goto out_fail;
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008159 /*
8160 * this is an ugly little race, but the rename is required
8161 * to make sure that if we crash, the inode is either at the
8162 * old name or the new one. pinning the log transaction lets
8163 * us make sure we don't allow a log commit to come in after
8164 * we unlink the name but before we add the new name back in.
8165 */
8166 btrfs_pin_log_trans(root);
8167 }
Chris Mason12fcfd22009-03-24 10:24:20 -04008168 /*
Chris Mason5a3f23d2009-03-31 13:27:11 -04008169 * make sure the inode gets flushed if it is replacing
8170 * something.
8171 */
Li Zefan33345d012011-04-20 10:31:50 +08008172 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
Chris Mason5a3f23d2009-03-31 13:27:11 -04008173 btrfs_add_ordered_operation(trans, root, old_inode);
Chris Mason5a3f23d2009-03-31 13:27:11 -04008174
Josef Bacik0c4d2d92012-04-05 15:03:02 -04008175 inode_inc_iversion(old_dir);
8176 inode_inc_iversion(new_dir);
8177 inode_inc_iversion(old_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008178 old_dir->i_ctime = old_dir->i_mtime = ctime;
8179 new_dir->i_ctime = new_dir->i_mtime = ctime;
8180 old_inode->i_ctime = ctime;
Chris Mason5f39d392007-10-15 16:14:19 -04008181
Chris Mason12fcfd22009-03-24 10:24:20 -04008182 if (old_dentry->d_parent != new_dentry->d_parent)
8183 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8184
Li Zefan33345d012011-04-20 10:31:50 +08008185 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008186 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8187 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8188 old_dentry->d_name.name,
8189 old_dentry->d_name.len);
8190 } else {
Al Viro92986792011-03-04 17:14:37 +00008191 ret = __btrfs_unlink_inode(trans, root, old_dir,
8192 old_dentry->d_inode,
8193 old_dentry->d_name.name,
8194 old_dentry->d_name.len);
8195 if (!ret)
8196 ret = btrfs_update_inode(trans, root, old_inode);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008197 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008198 if (ret) {
8199 btrfs_abort_transaction(trans, root, ret);
8200 goto out_fail;
8201 }
Chris Mason39279cc2007-06-12 06:35:45 -04008202
8203 if (new_inode) {
Josef Bacik0c4d2d92012-04-05 15:03:02 -04008204 inode_inc_iversion(new_inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008205 new_inode->i_ctime = CURRENT_TIME;
Li Zefan33345d012011-04-20 10:31:50 +08008206 if (unlikely(btrfs_ino(new_inode) ==
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008207 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8208 root_objectid = BTRFS_I(new_inode)->location.objectid;
8209 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8210 root_objectid,
8211 new_dentry->d_name.name,
8212 new_dentry->d_name.len);
8213 BUG_ON(new_inode->i_nlink == 0);
8214 } else {
8215 ret = btrfs_unlink_inode(trans, dest, new_dir,
8216 new_dentry->d_inode,
8217 new_dentry->d_name.name,
8218 new_dentry->d_name.len);
8219 }
Josef Bacik4ef31a42013-08-13 14:10:08 -04008220 if (!ret && new_inode->i_nlink == 0)
Chris Masone02119d2008-09-05 16:13:11 -04008221 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008222 if (ret) {
8223 btrfs_abort_transaction(trans, root, ret);
8224 goto out_fail;
8225 }
Chris Mason39279cc2007-06-12 06:35:45 -04008226 }
Josef Bacikaec74772008-07-24 12:12:38 -04008227
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008228 ret = btrfs_add_link(trans, new_dir, old_inode,
8229 new_dentry->d_name.name,
Yan, Zhenga5719522009-09-24 09:17:31 -04008230 new_dentry->d_name.len, 0, index);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008231 if (ret) {
8232 btrfs_abort_transaction(trans, root, ret);
8233 goto out_fail;
8234 }
Chris Mason39279cc2007-06-12 06:35:45 -04008235
Li Zefan33345d012011-04-20 10:31:50 +08008236 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
Al Viro10d9f302011-07-16 23:09:10 -04008237 struct dentry *parent = new_dentry->d_parent;
Josef Bacik6a912212010-11-20 09:48:00 +00008238 btrfs_log_new_name(trans, old_inode, old_dir, parent);
Yan, Zheng4df27c4d2009-09-21 15:56:00 -04008239 btrfs_end_log_trans(root);
8240 }
Chris Mason39279cc2007-06-12 06:35:45 -04008241out_fail:
Josef Bacik7ad85bb2012-01-12 19:10:12 -05008242 btrfs_end_transaction(trans, root);
Johann Lombardib44c59a2011-03-31 13:23:47 +00008243out_notrans:
Li Zefan33345d012011-04-20 10:31:50 +08008244 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
Yan, Zheng76dda932009-09-21 16:00:26 -04008245 up_read(&root->fs_info->subvol_sem);
Josef Bacik9ed74f22009-09-11 16:12:44 -04008246
Chris Mason39279cc2007-06-12 06:35:45 -04008247 return ret;
8248}
8249
Miao Xie8ccf6f192012-10-25 09:28:04 +00008250static void btrfs_run_delalloc_work(struct btrfs_work *work)
8251{
8252 struct btrfs_delalloc_work *delalloc_work;
Josef Bacik9f23e282013-10-28 15:03:41 -04008253 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00008254
8255 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8256 work);
Josef Bacik9f23e282013-10-28 15:03:41 -04008257 inode = delalloc_work->inode;
8258 if (delalloc_work->wait) {
8259 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8260 } else {
8261 filemap_flush(inode->i_mapping);
8262 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8263 &BTRFS_I(inode)->runtime_flags))
8264 filemap_flush(inode->i_mapping);
8265 }
Miao Xie8ccf6f192012-10-25 09:28:04 +00008266
8267 if (delalloc_work->delay_iput)
Josef Bacik9f23e282013-10-28 15:03:41 -04008268 btrfs_add_delayed_iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +00008269 else
Josef Bacik9f23e282013-10-28 15:03:41 -04008270 iput(inode);
Miao Xie8ccf6f192012-10-25 09:28:04 +00008271 complete(&delalloc_work->completion);
8272}
8273
8274struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8275 int wait, int delay_iput)
8276{
8277 struct btrfs_delalloc_work *work;
8278
8279 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8280 if (!work)
8281 return NULL;
8282
8283 init_completion(&work->completion);
8284 INIT_LIST_HEAD(&work->list);
8285 work->inode = inode;
8286 work->wait = wait;
8287 work->delay_iput = delay_iput;
8288 work->work.func = btrfs_run_delalloc_work;
8289
8290 return work;
8291}
8292
8293void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8294{
8295 wait_for_completion(&work->completion);
8296 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8297}
8298
Chris Masond352ac62008-09-29 15:18:18 -04008299/*
8300 * some fairly slow code that needs optimization. This walks the list
8301 * of all the inodes with pending delalloc and forces them to disk.
8302 */
Miao Xieeb73c1b2013-05-15 07:48:22 +00008303static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
Chris Masonea8c2812008-08-04 23:17:27 -04008304{
Chris Masonea8c2812008-08-04 23:17:27 -04008305 struct btrfs_inode *binode;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04008306 struct inode *inode;
Miao Xie8ccf6f192012-10-25 09:28:04 +00008307 struct btrfs_delalloc_work *work, *next;
8308 struct list_head works;
Miao Xie1eafa6c2013-01-22 10:49:00 +00008309 struct list_head splice;
Miao Xie8ccf6f192012-10-25 09:28:04 +00008310 int ret = 0;
Chris Masonea8c2812008-08-04 23:17:27 -04008311
Miao Xie8ccf6f192012-10-25 09:28:04 +00008312 INIT_LIST_HEAD(&works);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008313 INIT_LIST_HEAD(&splice);
Miao Xie63607cc2013-01-22 10:50:35 +00008314
Miao Xieeb73c1b2013-05-15 07:48:22 +00008315 spin_lock(&root->delalloc_lock);
8316 list_splice_init(&root->delalloc_inodes, &splice);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008317 while (!list_empty(&splice)) {
8318 binode = list_entry(splice.next, struct btrfs_inode,
Chris Masonea8c2812008-08-04 23:17:27 -04008319 delalloc_inodes);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008320
Miao Xieeb73c1b2013-05-15 07:48:22 +00008321 list_move_tail(&binode->delalloc_inodes,
8322 &root->delalloc_inodes);
Zheng Yan5b21f2e2008-09-26 10:05:38 -04008323 inode = igrab(&binode->vfs_inode);
Miao Xiedf0af1a2013-01-29 10:11:59 +00008324 if (!inode) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00008325 cond_resched_lock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008326 continue;
Miao Xiedf0af1a2013-01-29 10:11:59 +00008327 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00008328 spin_unlock(&root->delalloc_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008329
8330 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8331 if (unlikely(!work)) {
Josef Bacikf4ab9ea2013-09-17 11:25:44 -04008332 if (delay_iput)
8333 btrfs_add_delayed_iput(inode);
8334 else
8335 iput(inode);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008336 ret = -ENOMEM;
8337 goto out;
Zheng Yan5b21f2e2008-09-26 10:05:38 -04008338 }
Miao Xie1eafa6c2013-01-22 10:49:00 +00008339 list_add_tail(&work->list, &works);
8340 btrfs_queue_worker(&root->fs_info->flush_workers,
8341 &work->work);
8342
Zheng Yan5b21f2e2008-09-26 10:05:38 -04008343 cond_resched();
Miao Xieeb73c1b2013-05-15 07:48:22 +00008344 spin_lock(&root->delalloc_lock);
Chris Masonea8c2812008-08-04 23:17:27 -04008345 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00008346 spin_unlock(&root->delalloc_lock);
Chris Mason8c8bee12008-09-29 11:19:10 -04008347
Miao Xie1eafa6c2013-01-22 10:49:00 +00008348 list_for_each_entry_safe(work, next, &works, list) {
8349 list_del_init(&work->list);
8350 btrfs_wait_and_free_delalloc_work(work);
8351 }
Miao Xieeb73c1b2013-05-15 07:48:22 +00008352 return 0;
8353out:
8354 list_for_each_entry_safe(work, next, &works, list) {
8355 list_del_init(&work->list);
8356 btrfs_wait_and_free_delalloc_work(work);
8357 }
Miao Xie1eafa6c2013-01-22 10:49:00 +00008358
Miao Xieeb73c1b2013-05-15 07:48:22 +00008359 if (!list_empty_careful(&splice)) {
8360 spin_lock(&root->delalloc_lock);
8361 list_splice_tail(&splice, &root->delalloc_inodes);
8362 spin_unlock(&root->delalloc_lock);
8363 }
8364 return ret;
8365}
8366
8367int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8368{
8369 int ret;
8370
8371 if (root->fs_info->sb->s_flags & MS_RDONLY)
8372 return -EROFS;
8373
8374 ret = __start_delalloc_inodes(root, delay_iput);
8375 /*
8376 * the filemap_flush will queue IO into the worker threads, but
Chris Mason8c8bee12008-09-29 11:19:10 -04008377 * we have to make sure the IO is actually started and that
8378 * ordered extents get created before we return
8379 */
8380 atomic_inc(&root->fs_info->async_submit_draining);
Chris Masond3977122009-01-05 21:25:51 -05008381 while (atomic_read(&root->fs_info->nr_async_submits) ||
Chris Mason771ed682008-11-06 22:02:51 -05008382 atomic_read(&root->fs_info->async_delalloc_pages)) {
Chris Mason8c8bee12008-09-29 11:19:10 -04008383 wait_event(root->fs_info->async_submit_wait,
Chris Mason771ed682008-11-06 22:02:51 -05008384 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8385 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
Chris Mason8c8bee12008-09-29 11:19:10 -04008386 }
8387 atomic_dec(&root->fs_info->async_submit_draining);
Miao Xieeb73c1b2013-05-15 07:48:22 +00008388 return ret;
8389}
8390
Miao Xie91aef862013-11-04 23:13:26 +08008391int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
Miao Xieeb73c1b2013-05-15 07:48:22 +00008392{
8393 struct btrfs_root *root;
8394 struct list_head splice;
8395 int ret;
8396
8397 if (fs_info->sb->s_flags & MS_RDONLY)
8398 return -EROFS;
8399
8400 INIT_LIST_HEAD(&splice);
8401
8402 spin_lock(&fs_info->delalloc_root_lock);
8403 list_splice_init(&fs_info->delalloc_roots, &splice);
8404 while (!list_empty(&splice)) {
8405 root = list_first_entry(&splice, struct btrfs_root,
8406 delalloc_root);
8407 root = btrfs_grab_fs_root(root);
8408 BUG_ON(!root);
8409 list_move_tail(&root->delalloc_root,
8410 &fs_info->delalloc_roots);
8411 spin_unlock(&fs_info->delalloc_root_lock);
8412
8413 ret = __start_delalloc_inodes(root, delay_iput);
8414 btrfs_put_fs_root(root);
8415 if (ret)
8416 goto out;
8417
8418 spin_lock(&fs_info->delalloc_root_lock);
8419 }
8420 spin_unlock(&fs_info->delalloc_root_lock);
8421
8422 atomic_inc(&fs_info->async_submit_draining);
8423 while (atomic_read(&fs_info->nr_async_submits) ||
8424 atomic_read(&fs_info->async_delalloc_pages)) {
8425 wait_event(fs_info->async_submit_wait,
8426 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8427 atomic_read(&fs_info->async_delalloc_pages) == 0));
8428 }
8429 atomic_dec(&fs_info->async_submit_draining);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008430 return 0;
Miao Xie8ccf6f192012-10-25 09:28:04 +00008431out:
Miao Xie1eafa6c2013-01-22 10:49:00 +00008432 if (!list_empty_careful(&splice)) {
Miao Xieeb73c1b2013-05-15 07:48:22 +00008433 spin_lock(&fs_info->delalloc_root_lock);
8434 list_splice_tail(&splice, &fs_info->delalloc_roots);
8435 spin_unlock(&fs_info->delalloc_root_lock);
Miao Xie1eafa6c2013-01-22 10:49:00 +00008436 }
Miao Xie8ccf6f192012-10-25 09:28:04 +00008437 return ret;
Chris Masonea8c2812008-08-04 23:17:27 -04008438}
8439
Chris Mason39279cc2007-06-12 06:35:45 -04008440static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8441 const char *symname)
8442{
8443 struct btrfs_trans_handle *trans;
8444 struct btrfs_root *root = BTRFS_I(dir)->root;
8445 struct btrfs_path *path;
8446 struct btrfs_key key;
Chris Mason1832a6d2007-12-21 16:27:21 -05008447 struct inode *inode = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04008448 int err;
8449 int drop_inode = 0;
8450 u64 objectid;
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05308451 u64 index = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04008452 int name_len;
8453 int datasize;
Chris Mason5f39d392007-10-15 16:14:19 -04008454 unsigned long ptr;
Chris Mason39279cc2007-06-12 06:35:45 -04008455 struct btrfs_file_extent_item *ei;
Chris Mason5f39d392007-10-15 16:14:19 -04008456 struct extent_buffer *leaf;
Chris Mason39279cc2007-06-12 06:35:45 -04008457
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +01008458 name_len = strlen(symname);
Chris Mason39279cc2007-06-12 06:35:45 -04008459 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8460 return -ENAMETOOLONG;
Chris Mason1832a6d2007-12-21 16:27:21 -05008461
Josef Bacik9ed74f22009-09-11 16:12:44 -04008462 /*
8463 * 2 items for inode item and ref
8464 * 2 items for dir items
8465 * 1 item for xattr if selinux is on
8466 */
Yan, Zhenga22285a2010-05-16 10:48:46 -04008467 trans = btrfs_start_transaction(root, 5);
8468 if (IS_ERR(trans))
8469 return PTR_ERR(trans);
Chris Mason1832a6d2007-12-21 16:27:21 -05008470
Li Zefan581bb052011-04-20 10:06:11 +08008471 err = btrfs_find_free_ino(root, &objectid);
8472 if (err)
8473 goto out_unlock;
8474
Josef Bacikaec74772008-07-24 12:12:38 -04008475 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
Li Zefan33345d012011-04-20 10:31:50 +08008476 dentry->d_name.len, btrfs_ino(dir), objectid,
Josef Bacikd82a6f12011-05-11 15:26:06 -04008477 S_IFLNK|S_IRWXUGO, &index);
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04008478 if (IS_ERR(inode)) {
8479 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008480 goto out_unlock;
Tsutomu Itoh7cf96da2011-04-25 19:43:53 -04008481 }
Chris Mason39279cc2007-06-12 06:35:45 -04008482
Eric Paris2a7dba32011-02-01 11:05:39 -05008483 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
Josef Bacik33268ea2008-07-24 12:16:36 -04008484 if (err) {
8485 drop_inode = 1;
8486 goto out_unlock;
8487 }
8488
Casey Schauflerad19db72011-12-15 10:09:07 -05008489 /*
8490 * If the active LSM wants to access the inode during
8491 * d_instantiate it needs these. Smack checks to see
8492 * if the filesystem supports xattrs by looking at the
8493 * ops vector.
8494 */
8495 inode->i_fop = &btrfs_file_operations;
8496 inode->i_op = &btrfs_file_inode_operations;
8497
Josef Bacika1b075d2010-11-19 20:36:11 +00008498 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
Chris Mason39279cc2007-06-12 06:35:45 -04008499 if (err)
8500 drop_inode = 1;
8501 else {
8502 inode->i_mapping->a_ops = &btrfs_aops;
Chris Mason04160082008-03-26 10:28:07 -04008503 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Chris Masond1310b22008-01-24 16:13:08 -05008504 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
Chris Mason39279cc2007-06-12 06:35:45 -04008505 }
Chris Mason39279cc2007-06-12 06:35:45 -04008506 if (drop_inode)
8507 goto out_unlock;
8508
8509 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07008510 if (!path) {
8511 err = -ENOMEM;
8512 drop_inode = 1;
8513 goto out_unlock;
8514 }
Li Zefan33345d012011-04-20 10:31:50 +08008515 key.objectid = btrfs_ino(inode);
Chris Mason39279cc2007-06-12 06:35:45 -04008516 key.offset = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04008517 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8518 datasize = btrfs_file_extent_calc_inline_size(name_len);
8519 err = btrfs_insert_empty_item(trans, root, path, &key,
8520 datasize);
Chris Mason54aa1f42007-06-22 14:16:25 -04008521 if (err) {
8522 drop_inode = 1;
Julia Lawallb0839162011-05-14 07:10:51 +00008523 btrfs_free_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04008524 goto out_unlock;
8525 }
Chris Mason5f39d392007-10-15 16:14:19 -04008526 leaf = path->nodes[0];
8527 ei = btrfs_item_ptr(leaf, path->slots[0],
8528 struct btrfs_file_extent_item);
8529 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8530 btrfs_set_file_extent_type(leaf, ei,
Chris Mason39279cc2007-06-12 06:35:45 -04008531 BTRFS_FILE_EXTENT_INLINE);
Chris Masonc8b97812008-10-29 14:49:59 -04008532 btrfs_set_file_extent_encryption(leaf, ei, 0);
8533 btrfs_set_file_extent_compression(leaf, ei, 0);
8534 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8535 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8536
Chris Mason39279cc2007-06-12 06:35:45 -04008537 ptr = btrfs_file_extent_inline_start(ei);
Chris Mason5f39d392007-10-15 16:14:19 -04008538 write_extent_buffer(leaf, symname, ptr, name_len);
8539 btrfs_mark_buffer_dirty(leaf);
Chris Mason39279cc2007-06-12 06:35:45 -04008540 btrfs_free_path(path);
Chris Mason5f39d392007-10-15 16:14:19 -04008541
Chris Mason39279cc2007-06-12 06:35:45 -04008542 inode->i_op = &btrfs_symlink_inode_operations;
8543 inode->i_mapping->a_ops = &btrfs_symlink_aops;
Chris Mason04160082008-03-26 10:28:07 -04008544 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Yan Zhengd899e052008-10-30 14:25:28 -04008545 inode_set_bytes(inode, name_len);
Filipe David Borba Mananaf06becc2013-09-16 09:53:28 +01008546 btrfs_i_size_write(inode, name_len);
Chris Mason54aa1f42007-06-22 14:16:25 -04008547 err = btrfs_update_inode(trans, root, inode);
8548 if (err)
8549 drop_inode = 1;
Chris Mason39279cc2007-06-12 06:35:45 -04008550
8551out_unlock:
Al Viro08c422c2011-12-23 07:58:13 -05008552 if (!err)
8553 d_instantiate(dentry, inode);
Josef Bacik7ad85bb2012-01-12 19:10:12 -05008554 btrfs_end_transaction(trans, root);
Chris Mason39279cc2007-06-12 06:35:45 -04008555 if (drop_inode) {
8556 inode_dec_link_count(inode);
8557 iput(inode);
8558 }
Liu Bob53d3f52012-11-14 14:34:34 +00008559 btrfs_btree_balance_dirty(root);
Chris Mason39279cc2007-06-12 06:35:45 -04008560 return err;
8561}
Chris Mason16432982008-04-10 10:23:21 -04008562
Josef Bacik0af3d002010-06-21 14:48:16 -04008563static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8564 u64 start, u64 num_bytes, u64 min_size,
8565 loff_t actual_len, u64 *alloc_hint,
8566 struct btrfs_trans_handle *trans)
Yan Zhengd899e052008-10-30 14:25:28 -04008567{
Josef Bacik5dc562c2012-08-17 13:14:17 -04008568 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8569 struct extent_map *em;
Yan Zhengd899e052008-10-30 14:25:28 -04008570 struct btrfs_root *root = BTRFS_I(inode)->root;
8571 struct btrfs_key ins;
Yan Zhengd899e052008-10-30 14:25:28 -04008572 u64 cur_offset = start;
Josef Bacik55a61d12010-11-22 18:50:32 +00008573 u64 i_size;
Chris Mason154ea282013-03-05 11:11:26 -05008574 u64 cur_bytes;
Yan Zhengd899e052008-10-30 14:25:28 -04008575 int ret = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -04008576 bool own_trans = true;
Yan Zhengd899e052008-10-30 14:25:28 -04008577
Josef Bacik0af3d002010-06-21 14:48:16 -04008578 if (trans)
8579 own_trans = false;
Yan Zhengd899e052008-10-30 14:25:28 -04008580 while (num_bytes > 0) {
Josef Bacik0af3d002010-06-21 14:48:16 -04008581 if (own_trans) {
8582 trans = btrfs_start_transaction(root, 3);
8583 if (IS_ERR(trans)) {
8584 ret = PTR_ERR(trans);
8585 break;
8586 }
Yan Zhengd899e052008-10-30 14:25:28 -04008587 }
Yan, Zheng5a303d52009-11-12 09:34:52 +00008588
Chris Mason154ea282013-03-05 11:11:26 -05008589 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8590 cur_bytes = max(cur_bytes, min_size);
Josef Bacik00361582013-08-14 14:02:47 -04008591 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8592 *alloc_hint, &ins, 1);
Yan, Zheng5a303d52009-11-12 09:34:52 +00008593 if (ret) {
Josef Bacik0af3d002010-06-21 14:48:16 -04008594 if (own_trans)
8595 btrfs_end_transaction(trans, root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04008596 break;
Yan, Zheng5a303d52009-11-12 09:34:52 +00008597 }
8598
Yan Zhengd899e052008-10-30 14:25:28 -04008599 ret = insert_reserved_file_extent(trans, inode,
8600 cur_offset, ins.objectid,
8601 ins.offset, ins.offset,
Yan, Zheng920bbbf2009-11-12 09:34:08 +00008602 ins.offset, 0, 0, 0,
Yan Zhengd899e052008-10-30 14:25:28 -04008603 BTRFS_FILE_EXTENT_PREALLOC);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008604 if (ret) {
Josef Bacik857cc2f2013-10-07 15:21:08 -04008605 btrfs_free_reserved_extent(root, ins.objectid,
8606 ins.offset);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008607 btrfs_abort_transaction(trans, root, ret);
8608 if (own_trans)
8609 btrfs_end_transaction(trans, root);
8610 break;
8611 }
Chris Masona1ed8352009-09-11 12:27:37 -04008612 btrfs_drop_extent_cache(inode, cur_offset,
8613 cur_offset + ins.offset -1, 0);
Yan, Zheng5a303d52009-11-12 09:34:52 +00008614
Josef Bacik5dc562c2012-08-17 13:14:17 -04008615 em = alloc_extent_map();
8616 if (!em) {
8617 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8618 &BTRFS_I(inode)->runtime_flags);
8619 goto next;
8620 }
8621
8622 em->start = cur_offset;
8623 em->orig_start = cur_offset;
8624 em->len = ins.offset;
8625 em->block_start = ins.objectid;
8626 em->block_len = ins.offset;
Josef Bacikb4939682012-12-03 10:31:19 -05008627 em->orig_block_len = ins.offset;
Josef Bacikcc95bef2013-04-04 14:31:27 -04008628 em->ram_bytes = ins.offset;
Josef Bacik5dc562c2012-08-17 13:14:17 -04008629 em->bdev = root->fs_info->fs_devices->latest_bdev;
8630 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8631 em->generation = trans->transid;
8632
8633 while (1) {
8634 write_lock(&em_tree->lock);
Josef Bacik09a2a8f92013-04-05 16:51:15 -04008635 ret = add_extent_mapping(em_tree, em, 1);
Josef Bacik5dc562c2012-08-17 13:14:17 -04008636 write_unlock(&em_tree->lock);
8637 if (ret != -EEXIST)
8638 break;
8639 btrfs_drop_extent_cache(inode, cur_offset,
8640 cur_offset + ins.offset - 1,
8641 0);
8642 }
8643 free_extent_map(em);
8644next:
Yan Zhengd899e052008-10-30 14:25:28 -04008645 num_bytes -= ins.offset;
8646 cur_offset += ins.offset;
Yan, Zhengefa56462010-05-16 10:49:59 -04008647 *alloc_hint = ins.objectid + ins.offset;
Yan, Zheng5a303d52009-11-12 09:34:52 +00008648
Josef Bacik0c4d2d92012-04-05 15:03:02 -04008649 inode_inc_iversion(inode);
Yan Zhengd899e052008-10-30 14:25:28 -04008650 inode->i_ctime = CURRENT_TIME;
Christoph Hellwig6cbff002009-04-17 10:37:41 +02008651 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Yan Zhengd899e052008-10-30 14:25:28 -04008652 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
Yan, Zhengefa56462010-05-16 10:49:59 -04008653 (actual_len > inode->i_size) &&
8654 (cur_offset > inode->i_size)) {
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00008655 if (cur_offset > actual_len)
Josef Bacik55a61d12010-11-22 18:50:32 +00008656 i_size = actual_len;
Aneesh Kumar K.Vd1ea6a62010-01-20 07:28:54 +00008657 else
Josef Bacik55a61d12010-11-22 18:50:32 +00008658 i_size = cur_offset;
8659 i_size_write(inode, i_size);
8660 btrfs_ordered_update_i_size(inode, i_size, NULL);
Yan, Zheng5a303d52009-11-12 09:34:52 +00008661 }
8662
Yan Zhengd899e052008-10-30 14:25:28 -04008663 ret = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008664
8665 if (ret) {
8666 btrfs_abort_transaction(trans, root, ret);
8667 if (own_trans)
8668 btrfs_end_transaction(trans, root);
8669 break;
8670 }
Yan Zhengd899e052008-10-30 14:25:28 -04008671
Josef Bacik0af3d002010-06-21 14:48:16 -04008672 if (own_trans)
8673 btrfs_end_transaction(trans, root);
Yan, Zheng5a303d52009-11-12 09:34:52 +00008674 }
Yan Zhengd899e052008-10-30 14:25:28 -04008675 return ret;
8676}
8677
Josef Bacik0af3d002010-06-21 14:48:16 -04008678int btrfs_prealloc_file_range(struct inode *inode, int mode,
8679 u64 start, u64 num_bytes, u64 min_size,
8680 loff_t actual_len, u64 *alloc_hint)
8681{
8682 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8683 min_size, actual_len, alloc_hint,
8684 NULL);
8685}
8686
8687int btrfs_prealloc_file_range_trans(struct inode *inode,
8688 struct btrfs_trans_handle *trans, int mode,
8689 u64 start, u64 num_bytes, u64 min_size,
8690 loff_t actual_len, u64 *alloc_hint)
8691{
8692 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8693 min_size, actual_len, alloc_hint, trans);
8694}
8695
Chris Masone6dcd2d2008-07-17 12:53:50 -04008696static int btrfs_set_page_dirty(struct page *page)
8697{
Chris Masone6dcd2d2008-07-17 12:53:50 -04008698 return __set_page_dirty_nobuffers(page);
8699}
8700
Al Viro10556cb2011-06-20 19:28:19 -04008701static int btrfs_permission(struct inode *inode, int mask)
Yanfdebe2b2008-01-14 13:26:08 -05008702{
Li Zefanb83cc962010-12-20 16:04:08 +08008703 struct btrfs_root *root = BTRFS_I(inode)->root;
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00008704 umode_t mode = inode->i_mode;
Li Zefanb83cc962010-12-20 16:04:08 +08008705
Jeff Mahoneycb6db4e2011-08-15 17:27:21 +00008706 if (mask & MAY_WRITE &&
8707 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8708 if (btrfs_root_readonly(root))
8709 return -EROFS;
8710 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8711 return -EACCES;
8712 }
Al Viro2830ba72011-06-20 19:16:29 -04008713 return generic_permission(inode, mask);
Yanfdebe2b2008-01-14 13:26:08 -05008714}
Chris Mason39279cc2007-06-12 06:35:45 -04008715
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008716static const struct inode_operations btrfs_dir_inode_operations = {
Chris Mason3394e162008-11-17 20:42:26 -05008717 .getattr = btrfs_getattr,
Chris Mason39279cc2007-06-12 06:35:45 -04008718 .lookup = btrfs_lookup,
8719 .create = btrfs_create,
8720 .unlink = btrfs_unlink,
8721 .link = btrfs_link,
8722 .mkdir = btrfs_mkdir,
8723 .rmdir = btrfs_rmdir,
8724 .rename = btrfs_rename,
8725 .symlink = btrfs_symlink,
8726 .setattr = btrfs_setattr,
Josef Bacik618e21d2007-07-11 10:18:17 -04008727 .mknod = btrfs_mknod,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008728 .setxattr = btrfs_setxattr,
8729 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05008730 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008731 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05008732 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008733 .get_acl = btrfs_get_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -07008734 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04008735};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008736static const struct inode_operations btrfs_dir_ro_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04008737 .lookup = btrfs_lookup,
Yanfdebe2b2008-01-14 13:26:08 -05008738 .permission = btrfs_permission,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008739 .get_acl = btrfs_get_acl,
Guangyu Sun93fd63c2013-09-16 10:42:03 -07008740 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04008741};
Yan, Zheng76dda932009-09-21 16:00:26 -04008742
Alexey Dobriyan828c0952009-10-01 15:43:56 -07008743static const struct file_operations btrfs_dir_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04008744 .llseek = generic_file_llseek,
8745 .read = generic_read_dir,
Al Viro9cdda8d2013-05-22 16:48:09 -04008746 .iterate = btrfs_real_readdir,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04008747 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04008748#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04008749 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04008750#endif
Sage Weil6bf13c02008-06-10 10:07:39 -04008751 .release = btrfs_release_file,
Chris Masone02119d2008-09-05 16:13:11 -04008752 .fsync = btrfs_sync_file,
Chris Mason39279cc2007-06-12 06:35:45 -04008753};
8754
Chris Masond1310b22008-01-24 16:13:08 -05008755static struct extent_io_ops btrfs_extent_io_ops = {
Chris Mason07157aa2007-08-30 08:50:51 -04008756 .fill_delalloc = run_delalloc_range,
Chris Mason065631f2008-02-20 12:07:25 -05008757 .submit_bio_hook = btrfs_submit_bio_hook,
Chris Mason239b14b2008-03-24 15:02:07 -04008758 .merge_bio_hook = btrfs_merge_bio_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04008759 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
Chris Masone6dcd2d2008-07-17 12:53:50 -04008760 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
Chris Mason247e7432008-07-17 12:53:51 -04008761 .writepage_start_hook = btrfs_writepage_start_hook,
Chris Masonb0c68f82008-01-31 11:05:37 -05008762 .set_bit_hook = btrfs_set_bit_hook,
8763 .clear_bit_hook = btrfs_clear_bit_hook,
Josef Bacik9ed74f22009-09-11 16:12:44 -04008764 .merge_extent_hook = btrfs_merge_extent_hook,
8765 .split_extent_hook = btrfs_split_extent_hook,
Chris Mason07157aa2007-08-30 08:50:51 -04008766};
8767
Chris Mason35054392009-01-21 13:11:13 -05008768/*
8769 * btrfs doesn't support the bmap operation because swapfiles
8770 * use bmap to make a mapping of extents in the file. They assume
8771 * these extents won't change over the life of the file and they
8772 * use the bmap result to do IO directly to the drive.
8773 *
8774 * the btrfs bmap call would return logical addresses that aren't
8775 * suitable for IO and they also will change frequently as COW
8776 * operations happen. So, swapfile + btrfs == corruption.
8777 *
8778 * For now we're avoiding this by dropping bmap.
8779 */
Alexey Dobriyan7f094102009-09-21 17:01:10 -07008780static const struct address_space_operations btrfs_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04008781 .readpage = btrfs_readpage,
8782 .writepage = btrfs_writepage,
Chris Masonb293f022007-11-01 19:45:34 -04008783 .writepages = btrfs_writepages,
Chris Mason3ab2fb52007-11-08 10:59:22 -05008784 .readpages = btrfs_readpages,
Chris Mason16432982008-04-10 10:23:21 -04008785 .direct_IO = btrfs_direct_IO,
Chris Masona52d9a82007-08-27 16:49:44 -04008786 .invalidatepage = btrfs_invalidatepage,
8787 .releasepage = btrfs_releasepage,
Chris Masone6dcd2d2008-07-17 12:53:50 -04008788 .set_page_dirty = btrfs_set_page_dirty,
Andi Kleen465fdd92009-09-16 11:50:18 +02008789 .error_remove_page = generic_error_remove_page,
Chris Mason39279cc2007-06-12 06:35:45 -04008790};
8791
Alexey Dobriyan7f094102009-09-21 17:01:10 -07008792static const struct address_space_operations btrfs_symlink_aops = {
Chris Mason39279cc2007-06-12 06:35:45 -04008793 .readpage = btrfs_readpage,
8794 .writepage = btrfs_writepage,
Chris Mason2bf5a722007-08-30 11:54:02 -04008795 .invalidatepage = btrfs_invalidatepage,
8796 .releasepage = btrfs_releasepage,
Chris Mason39279cc2007-06-12 06:35:45 -04008797};
8798
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008799static const struct inode_operations btrfs_file_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04008800 .getattr = btrfs_getattr,
8801 .setattr = btrfs_setattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008802 .setxattr = btrfs_setxattr,
8803 .getxattr = btrfs_getxattr,
Josef Bacik5103e942007-11-16 11:45:54 -05008804 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008805 .removexattr = btrfs_removexattr,
Yanfdebe2b2008-01-14 13:26:08 -05008806 .permission = btrfs_permission,
Yehuda Sadeh1506fcc2009-01-21 14:39:14 -05008807 .fiemap = btrfs_fiemap,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008808 .get_acl = btrfs_get_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04008809 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04008810};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008811static const struct inode_operations btrfs_special_inode_operations = {
Josef Bacik618e21d2007-07-11 10:18:17 -04008812 .getattr = btrfs_getattr,
8813 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05008814 .permission = btrfs_permission,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008815 .setxattr = btrfs_setxattr,
8816 .getxattr = btrfs_getxattr,
Josef Bacik33268ea2008-07-24 12:16:36 -04008817 .listxattr = btrfs_listxattr,
Christoph Hellwig95819c02008-08-28 06:21:17 -04008818 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008819 .get_acl = btrfs_get_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04008820 .update_time = btrfs_update_time,
Josef Bacik618e21d2007-07-11 10:18:17 -04008821};
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07008822static const struct inode_operations btrfs_symlink_inode_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04008823 .readlink = generic_readlink,
8824 .follow_link = page_follow_link_light,
8825 .put_link = page_put_link,
Li Zefanf2095612010-11-19 02:05:24 +00008826 .getattr = btrfs_getattr,
Josef Bacik22c44fe2011-11-30 10:45:38 -05008827 .setattr = btrfs_setattr,
Yanfdebe2b2008-01-14 13:26:08 -05008828 .permission = btrfs_permission,
Jim Owens0279b4c2009-02-04 09:29:13 -05008829 .setxattr = btrfs_setxattr,
8830 .getxattr = btrfs_getxattr,
8831 .listxattr = btrfs_listxattr,
8832 .removexattr = btrfs_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +02008833 .get_acl = btrfs_get_acl,
Josef Bacike41f9412012-03-26 09:46:47 -04008834 .update_time = btrfs_update_time,
Chris Mason39279cc2007-06-12 06:35:45 -04008835};
Yan, Zheng76dda932009-09-21 16:00:26 -04008836
Alexey Dobriyan82d339d2009-10-09 09:54:36 -04008837const struct dentry_operations btrfs_dentry_operations = {
Yan, Zheng76dda932009-09-21 16:00:26 -04008838 .d_delete = btrfs_dentry_delete,
Josef Bacikb4aff1f2011-06-28 16:18:59 -04008839 .d_release = btrfs_dentry_release,
Yan, Zheng76dda932009-09-21 16:00:26 -04008840};