blob: 0ced78c974e202c90c286b778256f8707db01a83 [file] [log] [blame]
Alex Tomasa86c6182006-10-11 01:21:03 -07001/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
21 */
22
23/*
24 * Extents support for EXT4
25 *
26 * TODO:
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
30 */
31
Alex Tomasa86c6182006-10-11 01:21:03 -070032#include <linux/fs.h>
33#include <linux/time.h>
Mingming Caocd02ff02007-10-16 18:38:25 -040034#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070035#include <linux/highuid.h>
36#include <linux/pagemap.h>
37#include <linux/quotaops.h>
38#include <linux/string.h>
39#include <linux/slab.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070040#include <asm/uaccess.h>
Eric Sandeen6873fa02008-10-07 00:46:36 -040041#include <linux/fiemap.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040042#include "ext4_jbd2.h"
Theodore Ts'o4a092d72012-11-28 13:03:30 -050043#include "ext4_extents.h"
Tao Maf19d5872012-12-10 14:05:51 -050044#include "xattr.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070045
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040046#include <trace/events/ext4.h>
47
Lukas Czerner5f95d212012-03-19 23:03:19 -040048/*
49 * used by extent splitting.
50 */
51#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
52 due to ENOSPC */
Lukas Czerner556615d2014-04-20 23:45:47 -040053#define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54#define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
Lukas Czerner5f95d212012-03-19 23:03:19 -040055
Dmitry Monakhovdee1f972012-10-10 01:04:58 -040056#define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57#define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
58
Darrick J. Wong7ac59902012-04-29 18:37:10 -040059static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
61{
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
64 __u32 csum;
65
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
69}
70
71static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
73{
74 struct ext4_extent_tail *et;
75
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
78 return 1;
79
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
82 return 0;
83 return 1;
84}
85
86static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
88{
89 struct ext4_extent_tail *et;
90
91 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
93 return;
94
95 et = find_ext4_extent_tail(eh);
96 et->et_checksum = ext4_extent_block_csum(inode, eh);
97}
98
Allison Hendersond583fb82011-05-25 07:41:43 -040099static int ext4_split_extent(handle_t *handle,
100 struct inode *inode,
101 struct ext4_ext_path *path,
102 struct ext4_map_blocks *map,
103 int split_flag,
104 int flags);
105
Lukas Czerner5f95d212012-03-19 23:03:19 -0400106static int ext4_split_extent_at(handle_t *handle,
107 struct inode *inode,
108 struct ext4_ext_path *path,
109 ext4_lblk_t split,
110 int split_flag,
111 int flags);
112
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500113static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -0500114 struct extent_status *newes);
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500115
Jan Kara487caee2009-08-17 22:17:20 -0400116static int ext4_ext_truncate_extend_restart(handle_t *handle,
117 struct inode *inode,
118 int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -0700119{
120 int err;
121
Frank Mayhar03901312009-01-07 00:06:22 -0500122 if (!ext4_handle_valid(handle))
123 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700124 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -0400125 return 0;
126 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -0400127 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -0400128 return err;
Jan Kara487caee2009-08-17 22:17:20 -0400129 err = ext4_truncate_restart_trans(handle, inode, needed);
Dmitry Monakhov0617b832010-05-17 01:00:00 -0400130 if (err == 0)
131 err = -EAGAIN;
Jan Kara487caee2009-08-17 22:17:20 -0400132
133 return err;
Alex Tomasa86c6182006-10-11 01:21:03 -0700134}
135
136/*
137 * could return:
138 * - EROFS
139 * - ENOMEM
140 */
141static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
142 struct ext4_ext_path *path)
143{
144 if (path->p_bh) {
145 /* path points to block */
liang xie5d601252014-05-12 22:06:43 -0400146 BUFFER_TRACE(path->p_bh, "get_write_access");
Alex Tomasa86c6182006-10-11 01:21:03 -0700147 return ext4_journal_get_write_access(handle, path->p_bh);
148 }
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
151 return 0;
152}
153
154/*
155 * could return:
156 * - EROFS
157 * - ENOMEM
158 * - EIO
159 */
Darrick J. Wong26564972013-04-19 14:04:12 -0400160int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700162{
163 int err;
Dmitry Monakhov4b1f1662014-07-27 22:28:15 -0400164
165 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
Alex Tomasa86c6182006-10-11 01:21:03 -0700166 if (path->p_bh) {
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400167 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
Alex Tomasa86c6182006-10-11 01:21:03 -0700168 /* path points to block */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400169 err = __ext4_handle_dirty_metadata(where, line, handle,
170 inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700171 } else {
172 /* path points to leaf/index in inode body */
173 err = ext4_mark_inode_dirty(handle, inode);
174 }
175 return err;
176}
177
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700178static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700179 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500180 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700181{
Alex Tomasa86c6182006-10-11 01:21:03 -0700182 if (path) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400183 int depth = path->p_depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700184 struct ext4_extent *ex;
Alex Tomasa86c6182006-10-11 01:21:03 -0700185
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500186 /*
187 * Try to predict block placement assuming that we are
188 * filling in a file which will eventually be
189 * non-sparse --- i.e., in the case of libbfd writing
190 * an ELF object sections out-of-order but in a way
191 * the eventually results in a contiguous object or
192 * executable file, or some database extending a table
193 * space file. However, this is actually somewhat
194 * non-ideal if we are writing a sparse file such as
195 * qemu or KVM writing a raw image file that is going
196 * to stay fairly sparse, since it will end up
197 * fragmenting the file system's free space. Maybe we
198 * should have some hueristics or some way to allow
199 * userspace to pass a hint to file system,
Tao Mab8d65682011-01-21 23:21:31 +0800200 * especially if the latter case turns out to be
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500201 * common.
202 */
Avantika Mathur7e028972006-12-06 20:41:33 -0800203 ex = path[depth].p_ext;
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500204 if (ex) {
205 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
206 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
207
208 if (block > ext_block)
209 return ext_pblk + (block - ext_block);
210 else
211 return ext_pblk - (ext_block - block);
212 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700213
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700214 /* it looks like index is empty;
215 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700216 if (path[depth].p_bh)
217 return path[depth].p_bh->b_blocknr;
218 }
219
220 /* OK. use inode's group */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400221 return ext4_inode_to_goal_block(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700222}
223
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400224/*
225 * Allocation for a meta data block
226 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700227static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400228ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700229 struct ext4_ext_path *path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400230 struct ext4_extent *ex, int *err, unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700231{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700232 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700233
234 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Allison Henderson55f020d2011-05-25 07:41:26 -0400235 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
236 NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700237 return newblock;
238}
239
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400240static inline int ext4_ext_space_block(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700241{
242 int size;
243
244 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
245 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100246#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400247 if (!check && size > 6)
248 size = 6;
Alex Tomasa86c6182006-10-11 01:21:03 -0700249#endif
250 return size;
251}
252
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400253static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700254{
255 int size;
256
257 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
258 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100259#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400260 if (!check && size > 5)
261 size = 5;
Alex Tomasa86c6182006-10-11 01:21:03 -0700262#endif
263 return size;
264}
265
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400266static inline int ext4_ext_space_root(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700267{
268 int size;
269
270 size = sizeof(EXT4_I(inode)->i_data);
271 size -= sizeof(struct ext4_extent_header);
272 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100273#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400274 if (!check && size > 3)
275 size = 3;
Alex Tomasa86c6182006-10-11 01:21:03 -0700276#endif
277 return size;
278}
279
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400280static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700281{
282 int size;
283
284 size = sizeof(EXT4_I(inode)->i_data);
285 size -= sizeof(struct ext4_extent_header);
286 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100287#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400288 if (!check && size > 4)
289 size = 4;
Alex Tomasa86c6182006-10-11 01:21:03 -0700290#endif
291 return size;
292}
293
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -0400294static inline int
295ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
296 struct ext4_ext_path *path, ext4_lblk_t lblk,
297 int nofail)
298{
299 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
300
301 return ext4_split_extent_at(handle, inode, path, lblk, unwritten ?
302 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
303 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
304 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
305}
306
Mingming Caod2a17632008-07-14 17:52:37 -0400307/*
308 * Calculate the number of metadata blocks needed
309 * to allocate @blocks
310 * Worse case is one block per extent
311 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500312int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400313{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500314 struct ext4_inode_info *ei = EXT4_I(inode);
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400315 int idxs;
Mingming Caod2a17632008-07-14 17:52:37 -0400316
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500317 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
318 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400319
320 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500321 * If the new delayed allocation block is contiguous with the
322 * previous da block, it can share index blocks with the
323 * previous block, so we only need to allocate a new index
324 * block every idxs leaf blocks. At ldxs**2 blocks, we need
325 * an additional index block, and at ldxs**3 blocks, yet
326 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400327 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500328 if (ei->i_da_metadata_calc_len &&
329 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400330 int num = 0;
331
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500332 if ((ei->i_da_metadata_calc_len % idxs) == 0)
333 num++;
334 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
335 num++;
336 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
337 num++;
338 ei->i_da_metadata_calc_len = 0;
339 } else
340 ei->i_da_metadata_calc_len++;
341 ei->i_da_metadata_calc_last_lblock++;
342 return num;
343 }
Mingming Caod2a17632008-07-14 17:52:37 -0400344
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500345 /*
346 * In the worst case we need a new set of index blocks at
347 * every level of the inode's extent tree.
348 */
349 ei->i_da_metadata_calc_len = 1;
350 ei->i_da_metadata_calc_last_lblock = lblock;
351 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400352}
353
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400354static int
355ext4_ext_max_entries(struct inode *inode, int depth)
356{
357 int max;
358
359 if (depth == ext_depth(inode)) {
360 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400361 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400362 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400363 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400364 } else {
365 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400366 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400367 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400368 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400369 }
370
371 return max;
372}
373
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400374static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
375{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400376 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400377 int len = ext4_ext_get_actual_len(ext);
Eryu Guan5946d082013-12-03 21:22:21 -0500378 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
379 ext4_lblk_t last = lblock + len - 1;
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400380
Eryu Guan5946d082013-12-03 21:22:21 -0500381 if (lblock > last)
Theodore Ts'o31d4f3a2012-03-11 23:30:16 -0400382 return 0;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400383 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400384}
385
386static int ext4_valid_extent_idx(struct inode *inode,
387 struct ext4_extent_idx *ext_idx)
388{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400389 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400390
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400391 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400392}
393
394static int ext4_valid_extent_entries(struct inode *inode,
395 struct ext4_extent_header *eh,
396 int depth)
397{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400398 unsigned short entries;
399 if (eh->eh_entries == 0)
400 return 1;
401
402 entries = le16_to_cpu(eh->eh_entries);
403
404 if (depth == 0) {
405 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400406 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Eryu Guan5946d082013-12-03 21:22:21 -0500407 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
408 ext4_fsblk_t pblock = 0;
409 ext4_lblk_t lblock = 0;
410 ext4_lblk_t prev = 0;
411 int len = 0;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400412 while (entries) {
413 if (!ext4_valid_extent(inode, ext))
414 return 0;
Eryu Guan5946d082013-12-03 21:22:21 -0500415
416 /* Check for overlapping extents */
417 lblock = le32_to_cpu(ext->ee_block);
418 len = ext4_ext_get_actual_len(ext);
419 if ((lblock <= prev) && prev) {
420 pblock = ext4_ext_pblock(ext);
421 es->s_last_error_block = cpu_to_le64(pblock);
422 return 0;
423 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400424 ext++;
425 entries--;
Eryu Guan5946d082013-12-03 21:22:21 -0500426 prev = lblock + len - 1;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400427 }
428 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400429 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400430 while (entries) {
431 if (!ext4_valid_extent_idx(inode, ext_idx))
432 return 0;
433 ext_idx++;
434 entries--;
435 }
436 }
437 return 1;
438}
439
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400440static int __ext4_ext_check(const char *function, unsigned int line,
441 struct inode *inode, struct ext4_extent_header *eh,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400442 int depth, ext4_fsblk_t pblk)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400443{
444 const char *error_msg;
445 int max = 0;
446
447 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
448 error_msg = "invalid magic";
449 goto corrupted;
450 }
451 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
452 error_msg = "unexpected eh_depth";
453 goto corrupted;
454 }
455 if (unlikely(eh->eh_max == 0)) {
456 error_msg = "invalid eh_max";
457 goto corrupted;
458 }
459 max = ext4_ext_max_entries(inode, depth);
460 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
461 error_msg = "too large eh_max";
462 goto corrupted;
463 }
464 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
465 error_msg = "invalid eh_entries";
466 goto corrupted;
467 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400468 if (!ext4_valid_extent_entries(inode, eh, depth)) {
469 error_msg = "invalid extent entries";
470 goto corrupted;
471 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400472 /* Verify checksum on non-root extent tree nodes */
473 if (ext_depth(inode) != depth &&
474 !ext4_extent_block_csum_verify(inode, eh)) {
475 error_msg = "extent tree corrupted";
476 goto corrupted;
477 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400478 return 0;
479
480corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400481 ext4_error_inode(inode, function, line, 0,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400482 "pblk %llu bad header/extent: %s - magic %x, "
483 "entries %u, max %u(%u), depth %u(%u)",
484 (unsigned long long) pblk, error_msg,
485 le16_to_cpu(eh->eh_magic),
486 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
487 max, le16_to_cpu(eh->eh_depth), depth);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400488 return -EIO;
489}
490
Theodore Ts'oc3491792013-08-16 21:21:41 -0400491#define ext4_ext_check(inode, eh, depth, pblk) \
492 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400493
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400494int ext4_ext_check_inode(struct inode *inode)
495{
Theodore Ts'oc3491792013-08-16 21:21:41 -0400496 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400497}
498
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400499static struct buffer_head *
500__read_extent_tree_block(const char *function, unsigned int line,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400501 struct inode *inode, ext4_fsblk_t pblk, int depth,
502 int flags)
Darrick J. Wongf8489122012-04-29 18:21:10 -0400503{
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400504 struct buffer_head *bh;
505 int err;
Darrick J. Wongf8489122012-04-29 18:21:10 -0400506
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400507 bh = sb_getblk(inode->i_sb, pblk);
508 if (unlikely(!bh))
509 return ERR_PTR(-ENOMEM);
510
511 if (!bh_uptodate_or_lock(bh)) {
512 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
513 err = bh_submit_read(bh);
514 if (err < 0)
515 goto errout;
516 }
Theodore Ts'o7869a4a2013-08-16 22:05:14 -0400517 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400518 return bh;
519 err = __ext4_ext_check(function, line, inode,
Theodore Ts'oc3491792013-08-16 21:21:41 -0400520 ext_block_hdr(bh), depth, pblk);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400521 if (err)
522 goto errout;
Darrick J. Wongf8489122012-04-29 18:21:10 -0400523 set_buffer_verified(bh);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400524 /*
525 * If this is a leaf block, cache all of its entries
526 */
527 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
528 struct ext4_extent_header *eh = ext_block_hdr(bh);
529 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
530 ext4_lblk_t prev = 0;
531 int i;
532
533 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
534 unsigned int status = EXTENT_STATUS_WRITTEN;
535 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
536 int len = ext4_ext_get_actual_len(ex);
537
538 if (prev && (prev != lblk))
539 ext4_es_cache_extent(inode, prev,
540 lblk - prev, ~0,
541 EXTENT_STATUS_HOLE);
542
Lukas Czerner556615d2014-04-20 23:45:47 -0400543 if (ext4_ext_is_unwritten(ex))
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400544 status = EXTENT_STATUS_UNWRITTEN;
545 ext4_es_cache_extent(inode, lblk, len,
546 ext4_ext_pblock(ex), status);
547 prev = lblk + len;
548 }
549 }
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400550 return bh;
551errout:
552 put_bh(bh);
553 return ERR_PTR(err);
554
Darrick J. Wongf8489122012-04-29 18:21:10 -0400555}
556
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400557#define read_extent_tree_block(inode, pblk, depth, flags) \
558 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
559 (depth), (flags))
Darrick J. Wongf8489122012-04-29 18:21:10 -0400560
Theodore Ts'o7869a4a2013-08-16 22:05:14 -0400561/*
562 * This function is called to cache a file's extent information in the
563 * extent status tree
564 */
565int ext4_ext_precache(struct inode *inode)
566{
567 struct ext4_inode_info *ei = EXT4_I(inode);
568 struct ext4_ext_path *path = NULL;
569 struct buffer_head *bh;
570 int i = 0, depth, ret = 0;
571
572 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
573 return 0; /* not an extent-mapped inode */
574
575 down_read(&ei->i_data_sem);
576 depth = ext_depth(inode);
577
578 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
579 GFP_NOFS);
580 if (path == NULL) {
581 up_read(&ei->i_data_sem);
582 return -ENOMEM;
583 }
584
585 /* Don't cache anything if there are no external extent blocks */
586 if (depth == 0)
587 goto out;
588 path[0].p_hdr = ext_inode_hdr(inode);
589 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
590 if (ret)
591 goto out;
592 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
593 while (i >= 0) {
594 /*
595 * If this is a leaf block or we've reached the end of
596 * the index block, go up
597 */
598 if ((i == depth) ||
599 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
600 brelse(path[i].p_bh);
601 path[i].p_bh = NULL;
602 i--;
603 continue;
604 }
605 bh = read_extent_tree_block(inode,
606 ext4_idx_pblock(path[i].p_idx++),
607 depth - i - 1,
608 EXT4_EX_FORCE_CACHE);
609 if (IS_ERR(bh)) {
610 ret = PTR_ERR(bh);
611 break;
612 }
613 i++;
614 path[i].p_bh = bh;
615 path[i].p_hdr = ext_block_hdr(bh);
616 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
617 }
618 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
619out:
620 up_read(&ei->i_data_sem);
621 ext4_ext_drop_refs(path);
622 kfree(path);
623 return ret;
624}
625
Alex Tomasa86c6182006-10-11 01:21:03 -0700626#ifdef EXT_DEBUG
627static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
628{
629 int k, l = path->p_depth;
630
631 ext_debug("path:");
632 for (k = 0; k <= l; k++, path++) {
633 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700634 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400635 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700636 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400637 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700638 le32_to_cpu(path->p_ext->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -0400639 ext4_ext_is_unwritten(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400640 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400641 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700642 } else
643 ext_debug(" []");
644 }
645 ext_debug("\n");
646}
647
648static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
649{
650 int depth = ext_depth(inode);
651 struct ext4_extent_header *eh;
652 struct ext4_extent *ex;
653 int i;
654
655 if (!path)
656 return;
657
658 eh = path[depth].p_hdr;
659 ex = EXT_FIRST_EXTENT(eh);
660
Mingming553f9002009-09-18 13:34:55 -0400661 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
662
Alex Tomasa86c6182006-10-11 01:21:03 -0700663 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400664 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -0400665 ext4_ext_is_unwritten(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400666 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700667 }
668 ext_debug("\n");
669}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400670
671static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
672 ext4_fsblk_t newblock, int level)
673{
674 int depth = ext_depth(inode);
675 struct ext4_extent *ex;
676
677 if (depth != level) {
678 struct ext4_extent_idx *idx;
679 idx = path[level].p_idx;
680 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
681 ext_debug("%d: move %d:%llu in new index %llu\n", level,
682 le32_to_cpu(idx->ei_block),
683 ext4_idx_pblock(idx),
684 newblock);
685 idx++;
686 }
687
688 return;
689 }
690
691 ex = path[depth].p_ext;
692 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
693 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
694 le32_to_cpu(ex->ee_block),
695 ext4_ext_pblock(ex),
Lukas Czerner556615d2014-04-20 23:45:47 -0400696 ext4_ext_is_unwritten(ex),
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400697 ext4_ext_get_actual_len(ex),
698 newblock);
699 ex++;
700 }
701}
702
Alex Tomasa86c6182006-10-11 01:21:03 -0700703#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400704#define ext4_ext_show_path(inode, path)
705#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400706#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700707#endif
708
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500709void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700710{
711 int depth = path->p_depth;
712 int i;
713
714 for (i = 0; i <= depth; i++, path++)
715 if (path->p_bh) {
716 brelse(path->p_bh);
717 path->p_bh = NULL;
718 }
719}
720
721/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700722 * ext4_ext_binsearch_idx:
723 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400724 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700725 */
726static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500727ext4_ext_binsearch_idx(struct inode *inode,
728 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700729{
730 struct ext4_extent_header *eh = path->p_hdr;
731 struct ext4_extent_idx *r, *l, *m;
732
Alex Tomasa86c6182006-10-11 01:21:03 -0700733
Eric Sandeenbba90742008-01-28 23:58:27 -0500734 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700735
736 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400737 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700738 while (l <= r) {
739 m = l + (r - l) / 2;
740 if (block < le32_to_cpu(m->ei_block))
741 r = m - 1;
742 else
743 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400744 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
745 m, le32_to_cpu(m->ei_block),
746 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700747 }
748
749 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400750 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400751 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700752
753#ifdef CHECK_BINSEARCH
754 {
755 struct ext4_extent_idx *chix, *ix;
756 int k;
757
758 chix = ix = EXT_FIRST_INDEX(eh);
759 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
760 if (k != 0 &&
761 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400762 printk(KERN_DEBUG "k=%d, ix=0x%p, "
763 "first=0x%p\n", k,
764 ix, EXT_FIRST_INDEX(eh));
765 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700766 le32_to_cpu(ix->ei_block),
767 le32_to_cpu(ix[-1].ei_block));
768 }
769 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400770 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700771 if (block < le32_to_cpu(ix->ei_block))
772 break;
773 chix = ix;
774 }
775 BUG_ON(chix != path->p_idx);
776 }
777#endif
778
779}
780
781/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700782 * ext4_ext_binsearch:
783 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400784 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700785 */
786static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500787ext4_ext_binsearch(struct inode *inode,
788 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700789{
790 struct ext4_extent_header *eh = path->p_hdr;
791 struct ext4_extent *r, *l, *m;
792
Alex Tomasa86c6182006-10-11 01:21:03 -0700793 if (eh->eh_entries == 0) {
794 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700795 * this leaf is empty:
796 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700797 */
798 return;
799 }
800
Eric Sandeenbba90742008-01-28 23:58:27 -0500801 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700802
803 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400804 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700805
806 while (l <= r) {
807 m = l + (r - l) / 2;
808 if (block < le32_to_cpu(m->ee_block))
809 r = m - 1;
810 else
811 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400812 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
813 m, le32_to_cpu(m->ee_block),
814 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700815 }
816
817 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400818 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400819 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400820 ext4_ext_pblock(path->p_ext),
Lukas Czerner556615d2014-04-20 23:45:47 -0400821 ext4_ext_is_unwritten(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400822 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700823
824#ifdef CHECK_BINSEARCH
825 {
826 struct ext4_extent *chex, *ex;
827 int k;
828
829 chex = ex = EXT_FIRST_EXTENT(eh);
830 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
831 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400832 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700833 if (block < le32_to_cpu(ex->ee_block))
834 break;
835 chex = ex;
836 }
837 BUG_ON(chex != path->p_ext);
838 }
839#endif
840
841}
842
843int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
844{
845 struct ext4_extent_header *eh;
846
847 eh = ext_inode_hdr(inode);
848 eh->eh_depth = 0;
849 eh->eh_entries = 0;
850 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400851 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700852 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700853 return 0;
854}
855
856struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500857ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
Theodore Ts'o705912c2014-09-01 14:34:09 -0400858 struct ext4_ext_path **orig_path, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700859{
860 struct ext4_extent_header *eh;
861 struct buffer_head *bh;
Theodore Ts'o705912c2014-09-01 14:34:09 -0400862 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
863 short int depth, i, ppos = 0;
864 short free_on_err = (flags & EXT4_EX_NOFREE_ON_ERR) == 0;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500865 int ret;
Alex Tomasa86c6182006-10-11 01:21:03 -0700866
867 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400868 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700869
870 /* account possible depth increase */
871 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800872 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700873 GFP_NOFS);
Theodore Ts'o19008f62014-08-31 15:03:14 -0400874 if (unlikely(!path))
Alex Tomasa86c6182006-10-11 01:21:03 -0700875 return ERR_PTR(-ENOMEM);
Theodore Ts'o705912c2014-09-01 14:34:09 -0400876 free_on_err = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -0700877 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700878 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400879 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700880
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400881 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700882 /* walk through the tree */
883 while (i) {
884 ext_debug("depth %d: num %d, max %d\n",
885 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400886
Alex Tomasa86c6182006-10-11 01:21:03 -0700887 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400888 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700889 path[ppos].p_depth = i;
890 path[ppos].p_ext = NULL;
891
Theodore Ts'o107a7bd2013-08-16 21:23:41 -0400892 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
893 flags);
Theodore Ts'o19008f62014-08-31 15:03:14 -0400894 if (unlikely(IS_ERR(bh))) {
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400895 ret = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700896 goto err;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500897 }
Theodore Ts'o7d7ea892013-08-16 21:20:41 -0400898
Alex Tomasa86c6182006-10-11 01:21:03 -0700899 eh = ext_block_hdr(bh);
900 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500901 if (unlikely(ppos > depth)) {
902 put_bh(bh);
903 EXT4_ERROR_INODE(inode,
904 "ppos %d > depth %d", ppos, depth);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500905 ret = -EIO;
Frank Mayhar273df552010-03-02 11:46:09 -0500906 goto err;
907 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700908 path[ppos].p_bh = bh;
909 path[ppos].p_hdr = eh;
Alex Tomasa86c6182006-10-11 01:21:03 -0700910 }
911
912 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700913 path[ppos].p_ext = NULL;
914 path[ppos].p_idx = NULL;
915
Alex Tomasa86c6182006-10-11 01:21:03 -0700916 /* find extent */
917 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400918 /* if not an empty leaf */
919 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400920 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700921
922 ext4_ext_show_path(inode, path);
923
924 return path;
925
926err:
927 ext4_ext_drop_refs(path);
Theodore Ts'o705912c2014-09-01 14:34:09 -0400928 if (free_on_err) {
Alex Tomasa86c6182006-10-11 01:21:03 -0700929 kfree(path);
Theodore Ts'o705912c2014-09-01 14:34:09 -0400930 if (orig_path)
931 *orig_path = NULL;
932 }
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500933 return ERR_PTR(ret);
Alex Tomasa86c6182006-10-11 01:21:03 -0700934}
935
936/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700937 * ext4_ext_insert_index:
938 * insert new index [@logical;@ptr] into the block at @curp;
939 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700940 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400941static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
942 struct ext4_ext_path *curp,
943 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700944{
945 struct ext4_extent_idx *ix;
946 int len, err;
947
Avantika Mathur7e028972006-12-06 20:41:33 -0800948 err = ext4_ext_get_access(handle, inode, curp);
949 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700950 return err;
951
Frank Mayhar273df552010-03-02 11:46:09 -0500952 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
953 EXT4_ERROR_INODE(inode,
954 "logical %d == ei_block %d!",
955 logical, le32_to_cpu(curp->p_idx->ei_block));
956 return -EIO;
957 }
Robin Dongd4620312011-07-17 23:43:42 -0400958
959 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
960 >= le16_to_cpu(curp->p_hdr->eh_max))) {
961 EXT4_ERROR_INODE(inode,
962 "eh_entries %d >= eh_max %d!",
963 le16_to_cpu(curp->p_hdr->eh_entries),
964 le16_to_cpu(curp->p_hdr->eh_max));
965 return -EIO;
966 }
967
Alex Tomasa86c6182006-10-11 01:21:03 -0700968 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
969 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400970 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700971 ix = curp->p_idx + 1;
972 } else {
973 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400974 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700975 ix = curp->p_idx;
976 }
977
Eric Gouriou80e675f2011-10-27 11:52:18 -0400978 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
979 BUG_ON(len < 0);
980 if (len > 0) {
981 ext_debug("insert new index %d: "
982 "move %d indices from 0x%p to 0x%p\n",
983 logical, len, ix, ix + 1);
984 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
985 }
986
Tao Maf472e022011-10-17 10:13:46 -0400987 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
988 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
989 return -EIO;
990 }
991
Alex Tomasa86c6182006-10-11 01:21:03 -0700992 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700993 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400994 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700995
Frank Mayhar273df552010-03-02 11:46:09 -0500996 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
997 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
998 return -EIO;
999 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001000
1001 err = ext4_ext_dirty(handle, inode, curp);
1002 ext4_std_error(inode->i_sb, err);
1003
1004 return err;
1005}
1006
1007/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001008 * ext4_ext_split:
1009 * inserts new subtree into the path, using free index entry
1010 * at depth @at:
1011 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1012 * - makes decision where to split
1013 * - moves remaining extents and index entries (right to the split point)
1014 * into the newly allocated blocks
1015 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -07001016 */
1017static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001018 unsigned int flags,
1019 struct ext4_ext_path *path,
1020 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -07001021{
1022 struct buffer_head *bh = NULL;
1023 int depth = ext_depth(inode);
1024 struct ext4_extent_header *neh;
1025 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -07001026 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001027 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001028 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001029 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -07001030 int err = 0;
1031
1032 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001033 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07001034
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001035 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -07001036 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -05001037 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1038 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1039 return -EIO;
1040 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001041 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1042 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001043 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -07001044 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001045 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -07001046 } else {
1047 border = newext->ee_block;
1048 ext_debug("leaf will be added."
1049 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001050 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -07001051 }
1052
1053 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001054 * If error occurs, then we break processing
1055 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -07001056 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001057 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -07001058 */
1059
1060 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001061 * Get array to track all allocated blocks.
1062 * We need this to handle errors and free blocks
1063 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -07001064 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -08001065 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07001066 if (!ablocks)
1067 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001068
1069 /* allocate all needed blocks */
1070 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1071 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -04001072 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -04001073 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001074 if (newblock == 0)
1075 goto cleanup;
1076 ablocks[a] = newblock;
1077 }
1078
1079 /* initialize new leaf */
1080 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -05001081 if (unlikely(newblock == 0)) {
1082 EXT4_ERROR_INODE(inode, "newblock == 0!");
1083 err = -EIO;
1084 goto cleanup;
1085 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001086 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001087 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001088 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001089 goto cleanup;
1090 }
1091 lock_buffer(bh);
1092
Avantika Mathur7e028972006-12-06 20:41:33 -08001093 err = ext4_journal_get_create_access(handle, bh);
1094 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001095 goto cleanup;
1096
1097 neh = ext_block_hdr(bh);
1098 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001099 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001100 neh->eh_magic = EXT4_EXT_MAGIC;
1101 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001102
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001103 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -05001104 if (unlikely(path[depth].p_hdr->eh_entries !=
1105 path[depth].p_hdr->eh_max)) {
1106 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1107 path[depth].p_hdr->eh_entries,
1108 path[depth].p_hdr->eh_max);
1109 err = -EIO;
1110 goto cleanup;
1111 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001112 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001113 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1114 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07001115 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001116 struct ext4_extent *ex;
1117 ex = EXT_FIRST_EXTENT(neh);
1118 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001119 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001120 }
1121
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001122 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001123 set_buffer_uptodate(bh);
1124 unlock_buffer(bh);
1125
Frank Mayhar03901312009-01-07 00:06:22 -05001126 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001127 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001128 goto cleanup;
1129 brelse(bh);
1130 bh = NULL;
1131
1132 /* correct old leaf */
1133 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001134 err = ext4_ext_get_access(handle, inode, path + depth);
1135 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001136 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001137 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001138 err = ext4_ext_dirty(handle, inode, path + depth);
1139 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001140 goto cleanup;
1141
1142 }
1143
1144 /* create intermediate indexes */
1145 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001146 if (unlikely(k < 0)) {
1147 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1148 err = -EIO;
1149 goto cleanup;
1150 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001151 if (k)
1152 ext_debug("create %d intermediate indices\n", k);
1153 /* insert new index into current index block */
1154 /* current depth stored in i var */
1155 i = depth - 1;
1156 while (k--) {
1157 oldblock = newblock;
1158 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001159 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001160 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001161 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001162 goto cleanup;
1163 }
1164 lock_buffer(bh);
1165
Avantika Mathur7e028972006-12-06 20:41:33 -08001166 err = ext4_journal_get_create_access(handle, bh);
1167 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001168 goto cleanup;
1169
1170 neh = ext_block_hdr(bh);
1171 neh->eh_entries = cpu_to_le16(1);
1172 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001173 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001174 neh->eh_depth = cpu_to_le16(depth - i);
1175 fidx = EXT_FIRST_INDEX(neh);
1176 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001177 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001178
Eric Sandeenbba90742008-01-28 23:58:27 -05001179 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1180 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001181
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001182 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001183 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1184 EXT_LAST_INDEX(path[i].p_hdr))) {
1185 EXT4_ERROR_INODE(inode,
1186 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1187 le32_to_cpu(path[i].p_ext->ee_block));
1188 err = -EIO;
1189 goto cleanup;
1190 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001191 /* start copy indexes */
1192 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1193 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1194 EXT_MAX_INDEX(path[i].p_hdr));
1195 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001196 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001197 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001198 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001199 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001200 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001201 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001202 set_buffer_uptodate(bh);
1203 unlock_buffer(bh);
1204
Frank Mayhar03901312009-01-07 00:06:22 -05001205 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001206 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001207 goto cleanup;
1208 brelse(bh);
1209 bh = NULL;
1210
1211 /* correct old index */
1212 if (m) {
1213 err = ext4_ext_get_access(handle, inode, path + i);
1214 if (err)
1215 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001216 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001217 err = ext4_ext_dirty(handle, inode, path + i);
1218 if (err)
1219 goto cleanup;
1220 }
1221
1222 i--;
1223 }
1224
1225 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001226 err = ext4_ext_insert_index(handle, inode, path + at,
1227 le32_to_cpu(border), newblock);
1228
1229cleanup:
1230 if (bh) {
1231 if (buffer_locked(bh))
1232 unlock_buffer(bh);
1233 brelse(bh);
1234 }
1235
1236 if (err) {
1237 /* free all allocated blocks in error case */
1238 for (i = 0; i < depth; i++) {
1239 if (!ablocks[i])
1240 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001241 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001242 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001243 }
1244 }
1245 kfree(ablocks);
1246
1247 return err;
1248}
1249
1250/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001251 * ext4_ext_grow_indepth:
1252 * implements tree growing procedure:
1253 * - allocates new block
1254 * - moves top-level data (index block or leaf) into the new block
1255 * - initializes new top-level, creating index that points to the
1256 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001257 */
1258static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001259 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001260 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001261{
Alex Tomasa86c6182006-10-11 01:21:03 -07001262 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001263 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001264 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001265 int err = 0;
1266
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001267 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001268 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001269 if (newblock == 0)
1270 return err;
1271
1272 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001273 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001274 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001275 lock_buffer(bh);
1276
Avantika Mathur7e028972006-12-06 20:41:33 -08001277 err = ext4_journal_get_create_access(handle, bh);
1278 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001279 unlock_buffer(bh);
1280 goto out;
1281 }
1282
1283 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001284 memmove(bh->b_data, EXT4_I(inode)->i_data,
1285 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001286
1287 /* set size of new block */
1288 neh = ext_block_hdr(bh);
1289 /* old root could have indexes or leaves
1290 * so calculate e_max right way */
1291 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001292 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001293 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001294 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001295 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001296 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001297 set_buffer_uptodate(bh);
1298 unlock_buffer(bh);
1299
Frank Mayhar03901312009-01-07 00:06:22 -05001300 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001301 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001302 goto out;
1303
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001304 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001305 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001306 neh->eh_entries = cpu_to_le16(1);
1307 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1308 if (neh->eh_depth == 0) {
1309 /* Root extent block becomes index block */
1310 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1311 EXT_FIRST_INDEX(neh)->ei_block =
1312 EXT_FIRST_EXTENT(neh)->ee_block;
1313 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001314 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001315 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001316 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001317 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001318
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001319 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001320 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001321out:
1322 brelse(bh);
1323
1324 return err;
1325}
1326
1327/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001328 * ext4_ext_create_new_leaf:
1329 * finds empty index and adds new leaf.
1330 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001331 */
1332static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001333 unsigned int mb_flags,
1334 unsigned int gb_flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001335 struct ext4_ext_path *path,
1336 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001337{
1338 struct ext4_ext_path *curp;
1339 int depth, i, err = 0;
1340
1341repeat:
1342 i = depth = ext_depth(inode);
1343
1344 /* walk up to the tree and look for free index entry */
1345 curp = path + depth;
1346 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1347 i--;
1348 curp--;
1349 }
1350
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001351 /* we use already allocated block for index block,
1352 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001353 if (EXT_HAS_FREE_INDEX(curp)) {
1354 /* if we found index with free entry, then use that
1355 * entry: create all needed subtree and add new leaf */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001356 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001357 if (err)
1358 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001359
1360 /* refill path */
1361 ext4_ext_drop_refs(path);
1362 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001363 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
Theodore Ts'o705912c2014-09-01 14:34:09 -04001364 &path, gb_flags | EXT4_EX_NOFREE_ON_ERR);
Alex Tomasa86c6182006-10-11 01:21:03 -07001365 if (IS_ERR(path))
1366 err = PTR_ERR(path);
1367 } else {
1368 /* tree is full, time to grow in depth */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001369 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001370 if (err)
1371 goto out;
1372
1373 /* refill path */
1374 ext4_ext_drop_refs(path);
1375 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001376 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
Theodore Ts'o705912c2014-09-01 14:34:09 -04001377 &path, gb_flags | EXT4_EX_NOFREE_ON_ERR);
Alex Tomasa86c6182006-10-11 01:21:03 -07001378 if (IS_ERR(path)) {
1379 err = PTR_ERR(path);
1380 goto out;
1381 }
1382
1383 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001384 * only first (depth 0 -> 1) produces free space;
1385 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001386 */
1387 depth = ext_depth(inode);
1388 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001389 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001390 goto repeat;
1391 }
1392 }
1393
1394out:
1395 return err;
1396}
1397
1398/*
Alex Tomas1988b512008-01-28 23:58:27 -05001399 * search the closest allocated block to the left for *logical
1400 * and returns it at @logical + it's physical address at @phys
1401 * if *logical is the smallest allocated block, the function
1402 * returns 0 at @phys
1403 * return value contains 0 (success) or error code
1404 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001405static int ext4_ext_search_left(struct inode *inode,
1406 struct ext4_ext_path *path,
1407 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001408{
1409 struct ext4_extent_idx *ix;
1410 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001411 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001412
Frank Mayhar273df552010-03-02 11:46:09 -05001413 if (unlikely(path == NULL)) {
1414 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1415 return -EIO;
1416 }
Alex Tomas1988b512008-01-28 23:58:27 -05001417 depth = path->p_depth;
1418 *phys = 0;
1419
1420 if (depth == 0 && path->p_ext == NULL)
1421 return 0;
1422
1423 /* usually extent in the path covers blocks smaller
1424 * then *logical, but it can be that extent is the
1425 * first one in the file */
1426
1427 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001428 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001429 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001430 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1431 EXT4_ERROR_INODE(inode,
1432 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1433 *logical, le32_to_cpu(ex->ee_block));
1434 return -EIO;
1435 }
Alex Tomas1988b512008-01-28 23:58:27 -05001436 while (--depth >= 0) {
1437 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001438 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1439 EXT4_ERROR_INODE(inode,
1440 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001441 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001442 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001443 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001444 depth);
1445 return -EIO;
1446 }
Alex Tomas1988b512008-01-28 23:58:27 -05001447 }
1448 return 0;
1449 }
1450
Frank Mayhar273df552010-03-02 11:46:09 -05001451 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1452 EXT4_ERROR_INODE(inode,
1453 "logical %d < ee_block %d + ee_len %d!",
1454 *logical, le32_to_cpu(ex->ee_block), ee_len);
1455 return -EIO;
1456 }
Alex Tomas1988b512008-01-28 23:58:27 -05001457
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001458 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001459 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001460 return 0;
1461}
1462
1463/*
1464 * search the closest allocated block to the right for *logical
1465 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001466 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001467 * returns 0 at @phys
1468 * return value contains 0 (success) or error code
1469 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001470static int ext4_ext_search_right(struct inode *inode,
1471 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001472 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1473 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001474{
1475 struct buffer_head *bh = NULL;
1476 struct ext4_extent_header *eh;
1477 struct ext4_extent_idx *ix;
1478 struct ext4_extent *ex;
1479 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001480 int depth; /* Note, NOT eh_depth; depth from top of tree */
1481 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001482
Frank Mayhar273df552010-03-02 11:46:09 -05001483 if (unlikely(path == NULL)) {
1484 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1485 return -EIO;
1486 }
Alex Tomas1988b512008-01-28 23:58:27 -05001487 depth = path->p_depth;
1488 *phys = 0;
1489
1490 if (depth == 0 && path->p_ext == NULL)
1491 return 0;
1492
1493 /* usually extent in the path covers blocks smaller
1494 * then *logical, but it can be that extent is the
1495 * first one in the file */
1496
1497 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001498 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001499 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001500 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1501 EXT4_ERROR_INODE(inode,
1502 "first_extent(path[%d].p_hdr) != ex",
1503 depth);
1504 return -EIO;
1505 }
Alex Tomas1988b512008-01-28 23:58:27 -05001506 while (--depth >= 0) {
1507 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001508 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1509 EXT4_ERROR_INODE(inode,
1510 "ix != EXT_FIRST_INDEX *logical %d!",
1511 *logical);
1512 return -EIO;
1513 }
Alex Tomas1988b512008-01-28 23:58:27 -05001514 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001515 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001516 }
1517
Frank Mayhar273df552010-03-02 11:46:09 -05001518 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1519 EXT4_ERROR_INODE(inode,
1520 "logical %d < ee_block %d + ee_len %d!",
1521 *logical, le32_to_cpu(ex->ee_block), ee_len);
1522 return -EIO;
1523 }
Alex Tomas1988b512008-01-28 23:58:27 -05001524
1525 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1526 /* next allocated block in this leaf */
1527 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001528 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001529 }
1530
1531 /* go up and search for index to the right */
1532 while (--depth >= 0) {
1533 ix = path[depth].p_idx;
1534 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001535 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001536 }
1537
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001538 /* we've gone up to the root and found no index to the right */
1539 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001540
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001541got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001542 /* we've found index to the right, let's
1543 * follow it and find the closest allocated
1544 * block to the right */
1545 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001546 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001547 while (++depth < path->p_depth) {
Eric Sandeen395a87b2009-03-10 18:18:47 -04001548 /* subtract from p_depth to get proper eh_depth */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001549 bh = read_extent_tree_block(inode, block,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001550 path->p_depth - depth, 0);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001551 if (IS_ERR(bh))
1552 return PTR_ERR(bh);
1553 eh = ext_block_hdr(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001554 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001555 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001556 put_bh(bh);
1557 }
1558
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001559 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04001560 if (IS_ERR(bh))
1561 return PTR_ERR(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001562 eh = ext_block_hdr(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001563 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001564found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001565 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001566 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001567 *ret_ex = ex;
1568 if (bh)
1569 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001570 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001571}
1572
1573/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001574 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001575 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001576 * NOTE: it considers block number from index entry as
1577 * allocated block. Thus, index entries have to be consistent
1578 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001579 */
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04001580ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001581ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1582{
1583 int depth;
1584
1585 BUG_ON(path == NULL);
1586 depth = path->p_depth;
1587
1588 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001589 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001590
1591 while (depth >= 0) {
1592 if (depth == path->p_depth) {
1593 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001594 if (path[depth].p_ext &&
1595 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001596 EXT_LAST_EXTENT(path[depth].p_hdr))
1597 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1598 } else {
1599 /* index */
1600 if (path[depth].p_idx !=
1601 EXT_LAST_INDEX(path[depth].p_hdr))
1602 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1603 }
1604 depth--;
1605 }
1606
Lukas Czernerf17722f2011-06-06 00:05:17 -04001607 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001608}
1609
1610/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001611 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001612 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001613 */
Robin Dong57187892011-07-23 21:49:07 -04001614static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001615{
1616 int depth;
1617
1618 BUG_ON(path == NULL);
1619 depth = path->p_depth;
1620
1621 /* zero-tree has no leaf blocks at all */
1622 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001623 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001624
1625 /* go to index block */
1626 depth--;
1627
1628 while (depth >= 0) {
1629 if (path[depth].p_idx !=
1630 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001631 return (ext4_lblk_t)
1632 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001633 depth--;
1634 }
1635
Lukas Czernerf17722f2011-06-06 00:05:17 -04001636 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001637}
1638
1639/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001640 * ext4_ext_correct_indexes:
1641 * if leaf gets modified and modified extent is first in the leaf,
1642 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001643 * TODO: do we need to correct tree in all cases?
1644 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001645static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001646 struct ext4_ext_path *path)
1647{
1648 struct ext4_extent_header *eh;
1649 int depth = ext_depth(inode);
1650 struct ext4_extent *ex;
1651 __le32 border;
1652 int k, err = 0;
1653
1654 eh = path[depth].p_hdr;
1655 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001656
1657 if (unlikely(ex == NULL || eh == NULL)) {
1658 EXT4_ERROR_INODE(inode,
1659 "ex %p == NULL or eh %p == NULL", ex, eh);
1660 return -EIO;
1661 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001662
1663 if (depth == 0) {
1664 /* there is no tree at all */
1665 return 0;
1666 }
1667
1668 if (ex != EXT_FIRST_EXTENT(eh)) {
1669 /* we correct tree if first leaf got modified only */
1670 return 0;
1671 }
1672
1673 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001674 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001675 */
1676 k = depth - 1;
1677 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001678 err = ext4_ext_get_access(handle, inode, path + k);
1679 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001680 return err;
1681 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001682 err = ext4_ext_dirty(handle, inode, path + k);
1683 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001684 return err;
1685
1686 while (k--) {
1687 /* change all left-side indexes */
1688 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1689 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001690 err = ext4_ext_get_access(handle, inode, path + k);
1691 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001692 break;
1693 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001694 err = ext4_ext_dirty(handle, inode, path + k);
1695 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001696 break;
1697 }
1698
1699 return err;
1700}
1701
Akira Fujita748de672009-06-17 19:24:03 -04001702int
Alex Tomasa86c6182006-10-11 01:21:03 -07001703ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1704 struct ext4_extent *ex2)
1705{
Eric Sandeenda0169b2013-11-04 09:58:26 -05001706 unsigned short ext1_ee_len, ext2_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001707
1708 /*
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001709 * Make sure that both extents are initialized. We don't merge
Lukas Czerner556615d2014-04-20 23:45:47 -04001710 * unwritten extents so that we can be sure that end_io code has
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001711 * the extent that was written properly split out and conversion to
1712 * initialized is trivial.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001713 */
Lukas Czerner556615d2014-04-20 23:45:47 -04001714 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001715 return 0;
1716
1717 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1718 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1719
1720 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001721 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001722 return 0;
1723
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001724 /*
1725 * To allow future support for preallocated extents to be added
1726 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001727 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001728 */
Eric Sandeenda0169b2013-11-04 09:58:26 -05001729 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001730 return 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04001731 if (ext4_ext_is_unwritten(ex1) &&
Darrick J. Wonga9b82412014-02-20 21:17:35 -05001732 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1733 atomic_read(&EXT4_I(inode)->i_unwritten) ||
Lukas Czerner556615d2014-04-20 23:45:47 -04001734 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
Darrick J. Wonga9b82412014-02-20 21:17:35 -05001735 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001736#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001737 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001738 return 0;
1739#endif
1740
Theodore Ts'obf89d162010-10-27 21:30:14 -04001741 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001742 return 1;
1743 return 0;
1744}
1745
1746/*
Amit Arora56055d32007-07-17 21:42:38 -04001747 * This function tries to merge the "ex" extent to the next extent in the tree.
1748 * It always tries to merge towards right. If you want to merge towards
1749 * left, pass "ex - 1" as argument instead of "ex".
1750 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1751 * 1 if they got merged.
1752 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001753static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001754 struct ext4_ext_path *path,
1755 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001756{
1757 struct ext4_extent_header *eh;
1758 unsigned int depth, len;
Lukas Czerner556615d2014-04-20 23:45:47 -04001759 int merge_done = 0, unwritten;
Amit Arora56055d32007-07-17 21:42:38 -04001760
1761 depth = ext_depth(inode);
1762 BUG_ON(path[depth].p_hdr == NULL);
1763 eh = path[depth].p_hdr;
1764
1765 while (ex < EXT_LAST_EXTENT(eh)) {
1766 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1767 break;
1768 /* merge with next extent! */
Lukas Czerner556615d2014-04-20 23:45:47 -04001769 unwritten = ext4_ext_is_unwritten(ex);
Amit Arora56055d32007-07-17 21:42:38 -04001770 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1771 + ext4_ext_get_actual_len(ex + 1));
Lukas Czerner556615d2014-04-20 23:45:47 -04001772 if (unwritten)
1773 ext4_ext_mark_unwritten(ex);
Amit Arora56055d32007-07-17 21:42:38 -04001774
1775 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1776 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1777 * sizeof(struct ext4_extent);
1778 memmove(ex + 1, ex + 2, len);
1779 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001780 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001781 merge_done = 1;
1782 WARN_ON(eh->eh_entries == 0);
1783 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001784 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001785 }
1786
1787 return merge_done;
1788}
1789
1790/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001791 * This function does a very simple check to see if we can collapse
1792 * an extent tree with a single extent tree leaf block into the inode.
1793 */
1794static void ext4_ext_try_to_merge_up(handle_t *handle,
1795 struct inode *inode,
1796 struct ext4_ext_path *path)
1797{
1798 size_t s;
1799 unsigned max_root = ext4_ext_space_root(inode, 0);
1800 ext4_fsblk_t blk;
1801
1802 if ((path[0].p_depth != 1) ||
1803 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1804 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1805 return;
1806
1807 /*
1808 * We need to modify the block allocation bitmap and the block
1809 * group descriptor to release the extent tree block. If we
1810 * can't get the journal credits, give up.
1811 */
1812 if (ext4_journal_extend(handle, 2))
1813 return;
1814
1815 /*
1816 * Copy the extent data up to the inode
1817 */
1818 blk = ext4_idx_pblock(path[0].p_idx);
1819 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1820 sizeof(struct ext4_extent_idx);
1821 s += sizeof(struct ext4_extent_header);
1822
1823 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1824 path[0].p_depth = 0;
1825 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1826 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1827 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1828
1829 brelse(path[1].p_bh);
1830 ext4_free_blocks(handle, inode, NULL, blk, 1,
Theodore Ts'o71d4f7d2014-07-15 06:02:38 -04001831 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001832}
1833
1834/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001835 * This function tries to merge the @ex extent to neighbours in the tree.
1836 * return 1 if merge left else 0.
1837 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001838static void ext4_ext_try_to_merge(handle_t *handle,
1839 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001840 struct ext4_ext_path *path,
1841 struct ext4_extent *ex) {
1842 struct ext4_extent_header *eh;
1843 unsigned int depth;
1844 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001845
1846 depth = ext_depth(inode);
1847 BUG_ON(path[depth].p_hdr == NULL);
1848 eh = path[depth].p_hdr;
1849
1850 if (ex > EXT_FIRST_EXTENT(eh))
1851 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1852
1853 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001854 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001855
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001856 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001857}
1858
1859/*
Amit Arora25d14f92007-05-24 13:04:13 -04001860 * check if a portion of the "newext" extent overlaps with an
1861 * existing extent.
1862 *
1863 * If there is an overlap discovered, it updates the length of the newext
1864 * such that there will be no overlap, and then returns 1.
1865 * If there is no overlap found, it returns 0.
1866 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001867static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1868 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001869 struct ext4_extent *newext,
1870 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001871{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001872 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001873 unsigned int depth, len1;
1874 unsigned int ret = 0;
1875
1876 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001877 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001878 depth = ext_depth(inode);
1879 if (!path[depth].p_ext)
1880 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001881 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
Amit Arora25d14f92007-05-24 13:04:13 -04001882
1883 /*
1884 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001885 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001886 */
1887 if (b2 < b1) {
1888 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001889 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001890 goto out;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05001891 b2 = EXT4_LBLK_CMASK(sbi, b2);
Amit Arora25d14f92007-05-24 13:04:13 -04001892 }
1893
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001894 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001895 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001896 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001897 newext->ee_len = cpu_to_le16(len1);
1898 ret = 1;
1899 }
1900
1901 /* check for overlap */
1902 if (b1 + len1 > b2) {
1903 newext->ee_len = cpu_to_le16(b2 - b1);
1904 ret = 1;
1905 }
1906out:
1907 return ret;
1908}
1909
1910/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001911 * ext4_ext_insert_extent:
1912 * tries to merge requsted extent into the existing extent or
1913 * inserts requested extent as new one into the tree,
1914 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001915 */
1916int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1917 struct ext4_ext_path *path,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001918 struct ext4_extent *newext, int gb_flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07001919{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001920 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001921 struct ext4_extent *ex, *fex;
1922 struct ext4_extent *nearex; /* nearest extent */
1923 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001924 int depth, len, err;
1925 ext4_lblk_t next;
Lukas Czerner556615d2014-04-20 23:45:47 -04001926 int mb_flags = 0, unwritten;
Alex Tomasa86c6182006-10-11 01:21:03 -07001927
Frank Mayhar273df552010-03-02 11:46:09 -05001928 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1929 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1930 return -EIO;
1931 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001932 depth = ext_depth(inode);
1933 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001934 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001935 if (unlikely(path[depth].p_hdr == NULL)) {
1936 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1937 return -EIO;
1938 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001939
1940 /* try to insert block into found extent and return */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04001941 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001942
1943 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001944 * Try to see whether we should rather test the extent on
1945 * right from ex, or from the left of ex. This is because
1946 * ext4_ext_find_extent() can return either extent on the
1947 * left, or on the right from the searched position. This
1948 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001949 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001950 if (ex < EXT_LAST_EXTENT(eh) &&
1951 (le32_to_cpu(ex->ee_block) +
1952 ext4_ext_get_actual_len(ex) <
1953 le32_to_cpu(newext->ee_block))) {
1954 ex += 1;
1955 goto prepend;
1956 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1957 (le32_to_cpu(newext->ee_block) +
1958 ext4_ext_get_actual_len(newext) <
1959 le32_to_cpu(ex->ee_block)))
1960 ex -= 1;
1961
1962 /* Try to append newex to the ex */
1963 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1964 ext_debug("append [%d]%d block to %u:[%d]%d"
1965 "(from %llu)\n",
Lukas Czerner556615d2014-04-20 23:45:47 -04001966 ext4_ext_is_unwritten(newext),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001967 ext4_ext_get_actual_len(newext),
1968 le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04001969 ext4_ext_is_unwritten(ex),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001970 ext4_ext_get_actual_len(ex),
1971 ext4_ext_pblock(ex));
1972 err = ext4_ext_get_access(handle, inode,
1973 path + depth);
1974 if (err)
1975 return err;
Lukas Czerner556615d2014-04-20 23:45:47 -04001976 unwritten = ext4_ext_is_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001977 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001978 + ext4_ext_get_actual_len(newext));
Lukas Czerner556615d2014-04-20 23:45:47 -04001979 if (unwritten)
1980 ext4_ext_mark_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001981 eh = path[depth].p_hdr;
1982 nearex = ex;
1983 goto merge;
1984 }
1985
1986prepend:
1987 /* Try to prepend newex to the ex */
1988 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1989 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1990 "(from %llu)\n",
1991 le32_to_cpu(newext->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04001992 ext4_ext_is_unwritten(newext),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001993 ext4_ext_get_actual_len(newext),
1994 le32_to_cpu(ex->ee_block),
Lukas Czerner556615d2014-04-20 23:45:47 -04001995 ext4_ext_is_unwritten(ex),
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001996 ext4_ext_get_actual_len(ex),
1997 ext4_ext_pblock(ex));
1998 err = ext4_ext_get_access(handle, inode,
1999 path + depth);
2000 if (err)
2001 return err;
2002
Lukas Czerner556615d2014-04-20 23:45:47 -04002003 unwritten = ext4_ext_is_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002004 ex->ee_block = newext->ee_block;
2005 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2006 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2007 + ext4_ext_get_actual_len(newext));
Lukas Czerner556615d2014-04-20 23:45:47 -04002008 if (unwritten)
2009 ext4_ext_mark_unwritten(ex);
Lukas Czernerbe8981b2013-04-03 23:33:28 -04002010 eh = path[depth].p_hdr;
2011 nearex = ex;
2012 goto merge;
2013 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002014 }
2015
Alex Tomasa86c6182006-10-11 01:21:03 -07002016 depth = ext_depth(inode);
2017 eh = path[depth].p_hdr;
2018 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2019 goto has_space;
2020
2021 /* probably next leaf has space for us? */
2022 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04002023 next = EXT_MAX_BLOCKS;
2024 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04002025 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04002026 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002027 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07002028 BUG_ON(npath != NULL);
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002029 npath = ext4_ext_find_extent(inode, next, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002030 if (IS_ERR(npath))
2031 return PTR_ERR(npath);
2032 BUG_ON(npath->p_depth != path->p_depth);
2033 eh = npath[depth].p_hdr;
2034 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002035 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002036 le16_to_cpu(eh->eh_entries));
2037 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04002038 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07002039 }
2040 ext_debug("next leaf has no free space(%d,%d)\n",
2041 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2042 }
2043
2044 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002045 * There is no free space in the found leaf.
2046 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07002047 */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002048 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2049 mb_flags = EXT4_MB_USE_RESERVED;
2050 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2051 path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07002052 if (err)
2053 goto cleanup;
2054 depth = ext_depth(inode);
2055 eh = path[depth].p_hdr;
2056
2057has_space:
2058 nearex = path[depth].p_ext;
2059
Avantika Mathur7e028972006-12-06 20:41:33 -08002060 err = ext4_ext_get_access(handle, inode, path + depth);
2061 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002062 goto cleanup;
2063
2064 if (!nearex) {
2065 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002066 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002067 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002068 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002069 ext4_ext_is_unwritten(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002070 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04002071 nearex = EXT_FIRST_EXTENT(eh);
2072 } else {
2073 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002074 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04002075 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04002076 ext_debug("insert %u:%llu:[%d]%d before: "
2077 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002078 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04002079 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002080 ext4_ext_is_unwritten(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04002081 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002082 nearex);
2083 nearex++;
2084 } else {
2085 /* Insert before */
2086 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04002087 ext_debug("insert %u:%llu:[%d]%d after: "
2088 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04002089 le32_to_cpu(newext->ee_block),
2090 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002091 ext4_ext_is_unwritten(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002092 ext4_ext_get_actual_len(newext),
2093 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002094 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04002095 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2096 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002097 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04002098 "move %d extents from 0x%p to 0x%p\n",
2099 le32_to_cpu(newext->ee_block),
2100 ext4_ext_pblock(newext),
Lukas Czerner556615d2014-04-20 23:45:47 -04002101 ext4_ext_is_unwritten(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04002102 ext4_ext_get_actual_len(newext),
2103 len, nearex, nearex + 1);
2104 memmove(nearex + 1, nearex,
2105 len * sizeof(struct ext4_extent));
2106 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002107 }
2108
Marcin Slusarze8546d02008-04-17 10:38:59 -04002109 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002110 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002111 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002112 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002113 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002114
2115merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002116 /* try to merge extents */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002117 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002118 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002119
Alex Tomasa86c6182006-10-11 01:21:03 -07002120
2121 /* time to correct all indexes above */
2122 err = ext4_ext_correct_indexes(handle, inode, path);
2123 if (err)
2124 goto cleanup;
2125
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002126 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002127
2128cleanup:
2129 if (npath) {
2130 ext4_ext_drop_refs(npath);
2131 kfree(npath);
2132 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002133 return err;
2134}
2135
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002136static int ext4_fill_fiemap_extents(struct inode *inode,
2137 ext4_lblk_t block, ext4_lblk_t num,
2138 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002139{
2140 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002141 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002142 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002143 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002144 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002145 int exists, depth = 0, err = 0;
2146 unsigned int flags = 0;
2147 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002148
Lukas Czernerf17722f2011-06-06 00:05:17 -04002149 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002150 num = last - block;
2151 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002152 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002153
2154 if (path && ext_depth(inode) != depth) {
2155 /* depth was changed. we have to realloc path */
2156 kfree(path);
2157 path = NULL;
2158 }
2159
Theodore Ts'o705912c2014-09-01 14:34:09 -04002160 path = ext4_ext_find_extent(inode, block, &path, 0);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002161 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002162 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002163 err = PTR_ERR(path);
2164 path = NULL;
2165 break;
2166 }
2167
2168 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002169 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002170 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002171 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2172 err = -EIO;
2173 break;
2174 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002175 ex = path[depth].p_ext;
2176 next = ext4_ext_next_allocated_block(path);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002177 ext4_ext_drop_refs(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002178
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002179 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002180 exists = 0;
2181 if (!ex) {
2182 /* there is no extent yet, so try to allocate
2183 * all requested space */
2184 start = block;
2185 end = block + num;
2186 } else if (le32_to_cpu(ex->ee_block) > block) {
2187 /* need to allocate space before found extent */
2188 start = block;
2189 end = le32_to_cpu(ex->ee_block);
2190 if (block + num < end)
2191 end = block + num;
2192 } else if (block >= le32_to_cpu(ex->ee_block)
2193 + ext4_ext_get_actual_len(ex)) {
2194 /* need to allocate space after found extent */
2195 start = block;
2196 end = block + num;
2197 if (end >= next)
2198 end = next;
2199 } else if (block >= le32_to_cpu(ex->ee_block)) {
2200 /*
2201 * some part of requested space is covered
2202 * by found extent
2203 */
2204 start = block;
2205 end = le32_to_cpu(ex->ee_block)
2206 + ext4_ext_get_actual_len(ex);
2207 if (block + num < end)
2208 end = block + num;
2209 exists = 1;
2210 } else {
2211 BUG();
2212 }
2213 BUG_ON(end <= start);
2214
2215 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002216 es.es_lblk = start;
2217 es.es_len = end - start;
2218 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002219 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002220 es.es_lblk = le32_to_cpu(ex->ee_block);
2221 es.es_len = ext4_ext_get_actual_len(ex);
2222 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner556615d2014-04-20 23:45:47 -04002223 if (ext4_ext_is_unwritten(ex))
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002224 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002225 }
2226
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002227 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002228 * Find delayed extent and update es accordingly. We call
2229 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002230 * last existing extent or not.
2231 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002232 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002233 if (!exists && next_del) {
2234 exists = 1;
Jie Liu72dac952013-06-12 23:13:59 -04002235 flags |= (FIEMAP_EXTENT_DELALLOC |
2236 FIEMAP_EXTENT_UNKNOWN);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002237 }
2238 up_read(&EXT4_I(inode)->i_data_sem);
2239
Zheng Liu69eb33d2013-02-18 00:31:07 -05002240 if (unlikely(es.es_len == 0)) {
2241 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002242 err = -EIO;
2243 break;
2244 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002245
Zheng Liuf7fec032013-02-18 00:28:47 -05002246 /*
2247 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2248 * we need to check next == EXT_MAX_BLOCKS because it is
2249 * possible that an extent is with unwritten and delayed
2250 * status due to when an extent is delayed allocated and
2251 * is allocated by fallocate status tree will track both of
2252 * them in a extent.
2253 *
2254 * So we could return a unwritten and delayed extent, and
2255 * its block is equal to 'next'.
2256 */
2257 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002258 flags |= FIEMAP_EXTENT_LAST;
2259 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2260 next != EXT_MAX_BLOCKS)) {
2261 EXT4_ERROR_INODE(inode,
2262 "next extent == %u, next "
2263 "delalloc extent = %u",
2264 next, next_del);
2265 err = -EIO;
2266 break;
2267 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002268 }
2269
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002270 if (exists) {
2271 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002272 (__u64)es.es_lblk << blksize_bits,
2273 (__u64)es.es_pblk << blksize_bits,
2274 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002275 flags);
2276 if (err < 0)
2277 break;
2278 if (err == 1) {
2279 err = 0;
2280 break;
2281 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002282 }
2283
Zheng Liu69eb33d2013-02-18 00:31:07 -05002284 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002285 }
2286
2287 if (path) {
2288 ext4_ext_drop_refs(path);
2289 kfree(path);
2290 }
2291
2292 return err;
2293}
2294
Alex Tomasa86c6182006-10-11 01:21:03 -07002295/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002296 * ext4_ext_put_gap_in_cache:
2297 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002298 * and cache this gap
2299 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002300static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002301ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002302 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002303{
2304 int depth = ext_depth(inode);
Andi Shyti27b1b222013-08-28 14:00:00 -04002305 unsigned long len = 0;
2306 ext4_lblk_t lblock = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002307 struct ext4_extent *ex;
2308
2309 ex = path[depth].p_ext;
2310 if (ex == NULL) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002311 /*
2312 * there is no extent yet, so gap is [0;-] and we
2313 * don't cache it
2314 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002315 ext_debug("cache gap(whole file):");
2316 } else if (block < le32_to_cpu(ex->ee_block)) {
2317 lblock = block;
2318 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002319 ext_debug("cache gap(before): %u [%u:%u]",
2320 block,
2321 le32_to_cpu(ex->ee_block),
2322 ext4_ext_get_actual_len(ex));
Zheng Liud100eef2013-02-18 00:29:59 -05002323 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2324 ext4_es_insert_extent(inode, lblock, len, ~0,
2325 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002326 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002327 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002328 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002329 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002330 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002331
2332 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002333 ext_debug("cache gap(after): [%u:%u] %u",
2334 le32_to_cpu(ex->ee_block),
2335 ext4_ext_get_actual_len(ex),
2336 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002337 BUG_ON(next == lblock);
2338 len = next - lblock;
Zheng Liud100eef2013-02-18 00:29:59 -05002339 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2340 ext4_es_insert_extent(inode, lblock, len, ~0,
2341 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002342 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07002343 BUG();
2344 }
2345
Eric Sandeenbba90742008-01-28 23:58:27 -05002346 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002347}
2348
2349/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002350 * ext4_ext_rm_idx:
2351 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002352 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002353static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002354 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002355{
Alex Tomasa86c6182006-10-11 01:21:03 -07002356 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002357 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002358
2359 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002360 depth--;
2361 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002362 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002363 if (unlikely(path->p_hdr->eh_entries == 0)) {
2364 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2365 return -EIO;
2366 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002367 err = ext4_ext_get_access(handle, inode, path);
2368 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002369 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002370
2371 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2372 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2373 len *= sizeof(struct ext4_extent_idx);
2374 memmove(path->p_idx, path->p_idx + 1, len);
2375 }
2376
Marcin Slusarze8546d02008-04-17 10:38:59 -04002377 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002378 err = ext4_ext_dirty(handle, inode, path);
2379 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002380 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002381 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002382 trace_ext4_ext_rm_idx(inode, leaf);
2383
Peter Huewe7dc57612011-02-21 21:01:42 -05002384 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002385 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002386
2387 while (--depth >= 0) {
2388 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2389 break;
2390 path--;
2391 err = ext4_ext_get_access(handle, inode, path);
2392 if (err)
2393 break;
2394 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2395 err = ext4_ext_dirty(handle, inode, path);
2396 if (err)
2397 break;
2398 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002399 return err;
2400}
2401
2402/*
Mingming Caoee12b632008-08-19 22:16:05 -04002403 * ext4_ext_calc_credits_for_single_extent:
2404 * This routine returns max. credits that needed to insert an extent
2405 * to the extent tree.
2406 * When pass the actual path, the caller should calculate credits
2407 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002408 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002409int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002410 struct ext4_ext_path *path)
2411{
Alex Tomasa86c6182006-10-11 01:21:03 -07002412 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002413 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002414 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002415
Alex Tomasa86c6182006-10-11 01:21:03 -07002416 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002417 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002418 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2419
2420 /*
2421 * There are some space in the leaf tree, no
2422 * need to account for leaf block credit
2423 *
2424 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002425 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002426 * accounted.
2427 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002428 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002429 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002430 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002431 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002432 }
2433
Mingming Cao525f4ed2008-08-19 22:15:58 -04002434 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002435}
Alex Tomasa86c6182006-10-11 01:21:03 -07002436
Mingming Caoee12b632008-08-19 22:16:05 -04002437/*
Jan Karafffb2732013-06-04 13:01:11 -04002438 * How many index/leaf blocks need to change/allocate to add @extents extents?
Mingming Caoee12b632008-08-19 22:16:05 -04002439 *
Jan Karafffb2732013-06-04 13:01:11 -04002440 * If we add a single extent, then in the worse case, each tree level
2441 * index/leaf need to be changed in case of the tree split.
Mingming Caoee12b632008-08-19 22:16:05 -04002442 *
Jan Karafffb2732013-06-04 13:01:11 -04002443 * If more extents are inserted, they could cause the whole tree split more
2444 * than once, but this is really rare.
Mingming Caoee12b632008-08-19 22:16:05 -04002445 */
Jan Karafffb2732013-06-04 13:01:11 -04002446int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
Mingming Caoee12b632008-08-19 22:16:05 -04002447{
2448 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002449 int depth;
2450
2451 /* If we are converting the inline data, only one is needed here. */
2452 if (ext4_has_inline_data(inode))
2453 return 1;
2454
2455 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002456
Jan Karafffb2732013-06-04 13:01:11 -04002457 if (extents <= 1)
Mingming Caoee12b632008-08-19 22:16:05 -04002458 index = depth * 2;
2459 else
2460 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002461
Mingming Caoee12b632008-08-19 22:16:05 -04002462 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002463}
2464
Theodore Ts'o981250c2013-06-12 11:48:29 -04002465static inline int get_default_free_blocks_flags(struct inode *inode)
2466{
2467 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2468 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2469 else if (ext4_should_journal_data(inode))
2470 return EXT4_FREE_BLOCKS_FORGET;
2471 return 0;
2472}
2473
Alex Tomasa86c6182006-10-11 01:21:03 -07002474static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002475 struct ext4_extent *ex,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002476 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002477 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002478{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002479 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002480 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002481 ext4_fsblk_t pblk;
Theodore Ts'o981250c2013-06-12 11:48:29 -04002482 int flags = get_default_free_blocks_flags(inode);
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002483
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002484 /*
2485 * For bigalloc file systems, we never free a partial cluster
2486 * at the beginning of the extent. Instead, we make a note
2487 * that we tried freeing the cluster, and check to see if we
2488 * need to free it on a subsequent call to ext4_remove_blocks,
2489 * or at the end of the ext4_truncate() operation.
2490 */
2491 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2492
Aditya Kalid8990242011-09-09 19:18:51 -04002493 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002494 /*
2495 * If we have a partial cluster, and it's different from the
2496 * cluster of the last block, we need to explicitly free the
2497 * partial cluster here.
2498 */
2499 pblk = ext4_ext_pblock(ex) + ee_len - 1;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002500 if ((*partial_cluster > 0) &&
2501 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002502 ext4_free_blocks(handle, inode, NULL,
2503 EXT4_C2B(sbi, *partial_cluster),
2504 sbi->s_cluster_ratio, flags);
2505 *partial_cluster = 0;
2506 }
2507
Alex Tomasa86c6182006-10-11 01:21:03 -07002508#ifdef EXTENTS_STATS
2509 {
2510 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002511 spin_lock(&sbi->s_ext_stats_lock);
2512 sbi->s_ext_blocks += ee_len;
2513 sbi->s_ext_extents++;
2514 if (ee_len < sbi->s_ext_min)
2515 sbi->s_ext_min = ee_len;
2516 if (ee_len > sbi->s_ext_max)
2517 sbi->s_ext_max = ee_len;
2518 if (ext_depth(inode) > sbi->s_depth_max)
2519 sbi->s_depth_max = ext_depth(inode);
2520 spin_unlock(&sbi->s_ext_stats_lock);
2521 }
2522#endif
2523 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002524 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002525 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002526 ext4_lblk_t num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002527 unsigned int unaligned;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002528
Amit Aroraa2df2a62007-07-17 21:42:41 -04002529 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002530 pblk = ext4_ext_pblock(ex) + ee_len - num;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002531 /*
2532 * Usually we want to free partial cluster at the end of the
2533 * extent, except for the situation when the cluster is still
2534 * used by any other extent (partial_cluster is negative).
2535 */
2536 if (*partial_cluster < 0 &&
2537 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2538 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2539
2540 ext_debug("free last %u blocks starting %llu partial %lld\n",
2541 num, pblk, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002542 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2543 /*
2544 * If the block range to be freed didn't start at the
2545 * beginning of a cluster, and we removed the entire
Lukas Czernerd23142c2013-05-27 23:33:35 -04002546 * extent and the cluster is not used by any other extent,
2547 * save the partial cluster here, since we might need to
2548 * delete if we determine that the truncate operation has
2549 * removed all of the blocks in the cluster.
2550 *
2551 * On the other hand, if we did not manage to free the whole
2552 * extent, we have to mark the cluster as used (store negative
2553 * cluster number in partial_cluster).
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002554 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05002555 unaligned = EXT4_PBLK_COFF(sbi, pblk);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002556 if (unaligned && (ee_len == num) &&
2557 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002558 *partial_cluster = EXT4_B2C(sbi, pblk);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002559 else if (unaligned)
2560 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2561 else if (*partial_cluster > 0)
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002562 *partial_cluster = 0;
Lukas Czerner78fb9cd2013-05-27 23:32:35 -04002563 } else
2564 ext4_error(sbi->s_sb, "strange request: removal(2) "
2565 "%u-%u from %u:%u\n",
2566 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002567 return 0;
2568}
2569
Allison Hendersond583fb82011-05-25 07:41:43 -04002570
2571/*
2572 * ext4_ext_rm_leaf() Removes the extents associated with the
2573 * blocks appearing between "start" and "end", and splits the extents
2574 * if "start" and "end" appear in the same extent
2575 *
2576 * @handle: The journal handle
2577 * @inode: The files inode
2578 * @path: The path to the leaf
Lukas Czernerd23142c2013-05-27 23:33:35 -04002579 * @partial_cluster: The cluster which we'll have to free if all extents
2580 * has been released from it. It gets negative in case
2581 * that the cluster is still used.
Allison Hendersond583fb82011-05-25 07:41:43 -04002582 * @start: The first block to remove
2583 * @end: The last block to remove
2584 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002585static int
2586ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Lukas Czernerd23142c2013-05-27 23:33:35 -04002587 struct ext4_ext_path *path,
2588 long long *partial_cluster,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002589 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002590{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002591 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002592 int err = 0, correct_index = 0;
2593 int depth = ext_depth(inode), credits;
2594 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002595 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002596 unsigned num;
2597 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002598 unsigned short ex_ee_len;
Lukas Czerner556615d2014-04-20 23:45:47 -04002599 unsigned unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002600 struct ext4_extent *ex;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002601 ext4_fsblk_t pblk;
Alex Tomasa86c6182006-10-11 01:21:03 -07002602
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002603 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002604 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002605 if (!path[depth].p_hdr)
2606 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2607 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002608 if (unlikely(path[depth].p_hdr == NULL)) {
2609 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2610 return -EIO;
2611 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002612 /* find where to start removing */
Ashish Sangwan6ae06ff2013-07-01 08:12:41 -04002613 ex = path[depth].p_ext;
2614 if (!ex)
2615 ex = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002616
2617 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002618 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002619
Eric Whitneyc0634492014-03-13 23:34:16 -04002620 /*
2621 * If we're starting with an extent other than the last one in the
2622 * node, we need to see if it shares a cluster with the extent to
2623 * the right (towards the end of the file). If its leftmost cluster
2624 * is this extent's rightmost cluster and it is not cluster aligned,
2625 * we'll mark it as a partial that is not to be deallocated.
2626 */
2627
2628 if (ex != EXT_LAST_EXTENT(eh)) {
2629 ext4_fsblk_t current_pblk, right_pblk;
2630 long long current_cluster, right_cluster;
2631
2632 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2633 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2634 right_pblk = ext4_ext_pblock(ex + 1);
2635 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2636 if (current_cluster == right_cluster &&
2637 EXT4_PBLK_COFF(sbi, right_pblk))
2638 *partial_cluster = -right_cluster;
2639 }
2640
Aditya Kalid8990242011-09-09 19:18:51 -04002641 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2642
Alex Tomasa86c6182006-10-11 01:21:03 -07002643 while (ex >= EXT_FIRST_EXTENT(eh) &&
2644 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002645
Lukas Czerner556615d2014-04-20 23:45:47 -04002646 if (ext4_ext_is_unwritten(ex))
2647 unwritten = 1;
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002648 else
Lukas Czerner556615d2014-04-20 23:45:47 -04002649 unwritten = 0;
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002650
Mingming553f9002009-09-18 13:34:55 -04002651 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
Lukas Czerner556615d2014-04-20 23:45:47 -04002652 unwritten, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002653 path[depth].p_ext = ex;
2654
2655 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002656 b = ex_ee_block+ex_ee_len - 1 < end ?
2657 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002658
2659 ext_debug(" border %u:%u\n", a, b);
2660
Allison Hendersond583fb82011-05-25 07:41:43 -04002661 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002662 if (end < ex_ee_block) {
Lukas Czernerd23142c2013-05-27 23:33:35 -04002663 /*
2664 * We're going to skip this extent and move to another,
2665 * so if this extent is not cluster aligned we have
2666 * to mark the current cluster as used to avoid
2667 * accidentally freeing it later on
2668 */
2669 pblk = ext4_ext_pblock(ex);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05002670 if (EXT4_PBLK_COFF(sbi, pblk))
Lukas Czernerd23142c2013-05-27 23:33:35 -04002671 *partial_cluster =
2672 -((long long)EXT4_B2C(sbi, pblk));
Allison Hendersond583fb82011-05-25 07:41:43 -04002673 ex--;
2674 ex_ee_block = le32_to_cpu(ex->ee_block);
2675 ex_ee_len = ext4_ext_get_actual_len(ex);
2676 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002677 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002678 EXT4_ERROR_INODE(inode,
2679 "can not handle truncate %u:%u "
2680 "on extent %u:%u",
2681 start, end, ex_ee_block,
2682 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002683 err = -EIO;
2684 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002685 } else if (a != ex_ee_block) {
2686 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002687 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002688 } else {
2689 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002690 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002691 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002692 /*
2693 * 3 for leaf, sb, and inode plus 2 (bmap and group
2694 * descriptor) for each block group; assume two block
2695 * groups plus ex_ee_len/blocks_per_block_group for
2696 * the worst case
2697 */
2698 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002699 if (ex == EXT_FIRST_EXTENT(eh)) {
2700 correct_index = 1;
2701 credits += (ext_depth(inode)) + 1;
2702 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002703 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002704
Jan Kara487caee2009-08-17 22:17:20 -04002705 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002706 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002707 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002708
2709 err = ext4_ext_get_access(handle, inode, path + depth);
2710 if (err)
2711 goto out;
2712
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002713 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2714 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002715 if (err)
2716 goto out;
2717
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002718 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002719 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002720 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002721
Alex Tomasa86c6182006-10-11 01:21:03 -07002722 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002723 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04002724 * Do not mark unwritten if all the blocks in the
Amit Arora749269f2007-07-18 09:02:56 -04002725 * extent have been removed.
2726 */
Lukas Czerner556615d2014-04-20 23:45:47 -04002727 if (unwritten && num)
2728 ext4_ext_mark_unwritten(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002729 /*
2730 * If the extent was completely released,
2731 * we need to remove it from the leaf
2732 */
2733 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002734 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002735 /*
2736 * For hole punching, we need to scoot all the
2737 * extents up when an extent is removed so that
2738 * we dont have blank extents in the middle
2739 */
2740 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2741 sizeof(struct ext4_extent));
2742
2743 /* Now get rid of the one at the end */
2744 memset(EXT_LAST_EXTENT(eh), 0,
2745 sizeof(struct ext4_extent));
2746 }
2747 le16_add_cpu(&eh->eh_entries, -1);
Lukas Czernerd23142c2013-05-27 23:33:35 -04002748 } else if (*partial_cluster > 0)
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002749 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002750
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002751 err = ext4_ext_dirty(handle, inode, path + depth);
2752 if (err)
2753 goto out;
2754
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002755 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002756 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002757 ex--;
2758 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002759 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002760 }
2761
2762 if (correct_index && eh->eh_entries)
2763 err = ext4_ext_correct_indexes(handle, inode, path);
2764
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002765 /*
Eric Whitneyad6599a2014-04-01 19:49:30 -04002766 * If there's a partial cluster and at least one extent remains in
2767 * the leaf, free the partial cluster if it isn't shared with the
2768 * current extent. If there's a partial cluster and no extents
2769 * remain in the leaf, it can't be freed here. It can only be
2770 * freed when it's possible to determine if it's not shared with
2771 * any other extent - when the next leaf is processed or when space
2772 * removal is complete.
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002773 */
Eric Whitneyad6599a2014-04-01 19:49:30 -04002774 if (*partial_cluster > 0 && eh->eh_entries &&
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002775 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2776 *partial_cluster)) {
Theodore Ts'o981250c2013-06-12 11:48:29 -04002777 int flags = get_default_free_blocks_flags(inode);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002778
2779 ext4_free_blocks(handle, inode, NULL,
2780 EXT4_C2B(sbi, *partial_cluster),
2781 sbi->s_cluster_ratio, flags);
2782 *partial_cluster = 0;
2783 }
2784
Alex Tomasa86c6182006-10-11 01:21:03 -07002785 /* if this leaf is free, then we should
2786 * remove it from index block above */
2787 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002788 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002789
2790out:
2791 return err;
2792}
2793
2794/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002795 * ext4_ext_more_to_rm:
2796 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002797 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002798static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002799ext4_ext_more_to_rm(struct ext4_ext_path *path)
2800{
2801 BUG_ON(path->p_idx == NULL);
2802
2803 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2804 return 0;
2805
2806 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002807 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002808 * so we have to consider current index for truncation
2809 */
2810 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2811 return 0;
2812 return 1;
2813}
2814
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002815int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2816 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002817{
2818 struct super_block *sb = inode->i_sb;
2819 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002820 struct ext4_ext_path *path = NULL;
Lukas Czernerd23142c2013-05-27 23:33:35 -04002821 long long partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002822 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002823 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002824
Lukas Czerner5f95d212012-03-19 23:03:19 -04002825 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002826
2827 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002828 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002829 if (IS_ERR(handle))
2830 return PTR_ERR(handle);
2831
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002832again:
Lukas Czerner61801322013-05-27 23:32:35 -04002833 trace_ext4_ext_remove_space(inode, start, end, depth);
Aditya Kalid8990242011-09-09 19:18:51 -04002834
Alex Tomasa86c6182006-10-11 01:21:03 -07002835 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002836 * Check if we are removing extents inside the extent tree. If that
2837 * is the case, we are going to punch a hole inside the extent tree
2838 * so we have to check whether we need to split the extent covering
2839 * the last block to remove so we can easily remove the part of it
2840 * in ext4_ext_rm_leaf().
2841 */
2842 if (end < EXT_MAX_BLOCKS - 1) {
2843 struct ext4_extent *ex;
2844 ext4_lblk_t ee_block;
2845
2846 /* find extent for this block */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002847 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002848 if (IS_ERR(path)) {
2849 ext4_journal_stop(handle);
2850 return PTR_ERR(path);
2851 }
2852 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002853 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002854 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002855 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002856 if (depth) {
2857 EXT4_ERROR_INODE(inode,
2858 "path[%d].p_hdr == NULL",
2859 depth);
2860 err = -EIO;
2861 }
2862 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002863 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002864
2865 ee_block = le32_to_cpu(ex->ee_block);
2866
2867 /*
2868 * See if the last block is inside the extent, if so split
2869 * the extent at 'end' block so we can easily remove the
2870 * tail of the first part of the split extent in
2871 * ext4_ext_rm_leaf().
2872 */
2873 if (end >= ee_block &&
2874 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
Lukas Czerner5f95d212012-03-19 23:03:19 -04002875 /*
2876 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002877 * block in the first new extent. Also we should not
2878 * fail removing space due to ENOSPC so try to use
2879 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002880 */
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04002881 err = ext4_force_split_extent_at(handle, inode, path,
2882 end + 1, 1);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002883 if (err < 0)
2884 goto out;
2885 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002886 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002887 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002888 * We start scanning from right side, freeing all the blocks
2889 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002890 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002891 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002892 if (path) {
2893 int k = i = depth;
2894 while (--k > 0)
2895 path[k].p_block =
2896 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2897 } else {
2898 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2899 GFP_NOFS);
2900 if (path == NULL) {
2901 ext4_journal_stop(handle);
2902 return -ENOMEM;
2903 }
2904 path[0].p_depth = depth;
2905 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002906 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002907
Theodore Ts'oc3491792013-08-16 21:21:41 -04002908 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
Ashish Sangwan968dee72012-07-22 22:49:08 -04002909 err = -EIO;
2910 goto out;
2911 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002912 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002913 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002914
2915 while (i >= 0 && err == 0) {
2916 if (i == depth) {
2917 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002918 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002919 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002920 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002921 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002922 brelse(path[i].p_bh);
2923 path[i].p_bh = NULL;
2924 i--;
2925 continue;
2926 }
2927
2928 /* this is index block */
2929 if (!path[i].p_hdr) {
2930 ext_debug("initialize header\n");
2931 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002932 }
2933
Alex Tomasa86c6182006-10-11 01:21:03 -07002934 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002935 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002936 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2937 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2938 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2939 path[i].p_hdr,
2940 le16_to_cpu(path[i].p_hdr->eh_entries));
2941 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002942 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002943 path[i].p_idx--;
2944 }
2945
2946 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2947 i, EXT_FIRST_INDEX(path[i].p_hdr),
2948 path[i].p_idx);
2949 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002950 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002951 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002952 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002953 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002954 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002955 bh = read_extent_tree_block(inode,
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04002956 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2957 EXT4_EX_NOCACHE);
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002958 if (IS_ERR(bh)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002959 /* should we reset i_size? */
Theodore Ts'o7d7ea892013-08-16 21:20:41 -04002960 err = PTR_ERR(bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002961 break;
2962 }
Theodore Ts'o76828c82013-07-15 12:27:47 -04002963 /* Yield here to deal with large extent trees.
2964 * Should be a no-op if we did IO above. */
2965 cond_resched();
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002966 if (WARN_ON(i + 1 > depth)) {
2967 err = -EIO;
2968 break;
2969 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002970 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002971
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002972 /* save actual number of indexes since this
2973 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002974 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2975 i++;
2976 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002977 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002978 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002979 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002980 * handle must be already prepared by the
2981 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002982 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002983 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002984 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002985 brelse(path[i].p_bh);
2986 path[i].p_bh = NULL;
2987 i--;
2988 ext_debug("return to level %d\n", i);
2989 }
2990 }
2991
Lukas Czerner61801322013-05-27 23:32:35 -04002992 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2993 partial_cluster, path->p_hdr->eh_entries);
Aditya Kalid8990242011-09-09 19:18:51 -04002994
Aditya Kali7b415bf2011-09-09 19:04:51 -04002995 /* If we still have something in the partial cluster and we have removed
2996 * even the first extent, then we should free the blocks in the partial
2997 * cluster as well. */
Lukas Czernerd23142c2013-05-27 23:33:35 -04002998 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
Theodore Ts'o981250c2013-06-12 11:48:29 -04002999 int flags = get_default_free_blocks_flags(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003000
3001 ext4_free_blocks(handle, inode, NULL,
3002 EXT4_C2B(EXT4_SB(sb), partial_cluster),
3003 EXT4_SB(sb)->s_cluster_ratio, flags);
3004 partial_cluster = 0;
3005 }
3006
Alex Tomasa86c6182006-10-11 01:21:03 -07003007 /* TODO: flexible tree reduction should be here */
3008 if (path->p_hdr->eh_entries == 0) {
3009 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003010 * truncate to zero freed all the tree,
3011 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07003012 */
3013 err = ext4_ext_get_access(handle, inode, path);
3014 if (err == 0) {
3015 ext_inode_hdr(inode)->eh_depth = 0;
3016 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04003017 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07003018 err = ext4_ext_dirty(handle, inode, path);
3019 }
3020 }
3021out:
Alex Tomasa86c6182006-10-11 01:21:03 -07003022 ext4_ext_drop_refs(path);
3023 kfree(path);
Ashish Sangwan968dee72012-07-22 22:49:08 -04003024 if (err == -EAGAIN) {
3025 path = NULL;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04003026 goto again;
Ashish Sangwan968dee72012-07-22 22:49:08 -04003027 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003028 ext4_journal_stop(handle);
3029
3030 return err;
3031}
3032
3033/*
3034 * called at mount time
3035 */
3036void ext4_ext_init(struct super_block *sb)
3037{
3038 /*
3039 * possible initialization would be here
3040 */
3041
Theodore Ts'o83982b62009-01-06 14:53:16 -05003042 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04003043#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04003044 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01003045#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04003046 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07003047#endif
3048#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04003049 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07003050#endif
3051#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04003052 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07003053#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04003054 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04003055#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07003056#ifdef EXTENTS_STATS
3057 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3058 EXT4_SB(sb)->s_ext_min = 1 << 30;
3059 EXT4_SB(sb)->s_ext_max = 0;
3060#endif
3061 }
3062}
3063
3064/*
3065 * called at umount time
3066 */
3067void ext4_ext_release(struct super_block *sb)
3068{
Theodore Ts'o83982b62009-01-06 14:53:16 -05003069 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07003070 return;
3071
3072#ifdef EXTENTS_STATS
3073 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3074 struct ext4_sb_info *sbi = EXT4_SB(sb);
3075 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3076 sbi->s_ext_blocks, sbi->s_ext_extents,
3077 sbi->s_ext_blocks / sbi->s_ext_extents);
3078 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3079 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3080 }
3081#endif
3082}
3083
Zheng Liud7b2a002013-08-28 14:47:06 -04003084static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3085{
3086 ext4_lblk_t ee_block;
3087 ext4_fsblk_t ee_pblock;
3088 unsigned int ee_len;
3089
3090 ee_block = le32_to_cpu(ex->ee_block);
3091 ee_len = ext4_ext_get_actual_len(ex);
3092 ee_pblock = ext4_ext_pblock(ex);
3093
3094 if (ee_len == 0)
3095 return 0;
3096
3097 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3098 EXTENT_STATUS_WRITTEN);
3099}
3100
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003101/* FIXME!! we need to try to merge to left or right after zero-out */
3102static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3103{
Lukas Czerner24075182010-10-27 21:30:06 -04003104 ext4_fsblk_t ee_pblock;
3105 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04003106 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003107
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003108 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003109 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003110
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04003111 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04003112 if (ret > 0)
3113 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003114
Lukas Czerner24075182010-10-27 21:30:06 -04003115 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04003116}
3117
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003118/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003119 * ext4_split_extent_at() splits an extent at given block.
3120 *
3121 * @handle: the journal handle
3122 * @inode: the file inode
3123 * @path: the path to the extent
3124 * @split: the logical block where the extent is splitted.
3125 * @split_flags: indicates if the extent could be zeroout if split fails, and
Lukas Czerner556615d2014-04-20 23:45:47 -04003126 * the states(init or unwritten) of new extents.
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003127 * @flags: flags used to insert new extent to extent tree.
3128 *
3129 *
3130 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3131 * of which are deterimined by split_flag.
3132 *
3133 * There are two cases:
3134 * a> the extent are splitted into two extent.
3135 * b> split is not needed, and just mark the extent.
3136 *
3137 * return 0 on success.
3138 */
3139static int ext4_split_extent_at(handle_t *handle,
3140 struct inode *inode,
3141 struct ext4_ext_path *path,
3142 ext4_lblk_t split,
3143 int split_flag,
3144 int flags)
3145{
3146 ext4_fsblk_t newblock;
3147 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04003148 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003149 struct ext4_extent *ex2 = NULL;
3150 unsigned int ee_len, depth;
3151 int err = 0;
3152
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003153 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3154 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3155
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003156 ext_debug("ext4_split_extents_at: inode %lu, logical"
3157 "block %llu\n", inode->i_ino, (unsigned long long)split);
3158
3159 ext4_ext_show_leaf(inode, path);
3160
3161 depth = ext_depth(inode);
3162 ex = path[depth].p_ext;
3163 ee_block = le32_to_cpu(ex->ee_block);
3164 ee_len = ext4_ext_get_actual_len(ex);
3165 newblock = split - ee_block + ext4_ext_pblock(ex);
3166
3167 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Lukas Czerner556615d2014-04-20 23:45:47 -04003168 BUG_ON(!ext4_ext_is_unwritten(ex) &&
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003169 split_flag & (EXT4_EXT_MAY_ZEROOUT |
Lukas Czerner556615d2014-04-20 23:45:47 -04003170 EXT4_EXT_MARK_UNWRIT1 |
3171 EXT4_EXT_MARK_UNWRIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003172
3173 err = ext4_ext_get_access(handle, inode, path + depth);
3174 if (err)
3175 goto out;
3176
3177 if (split == ee_block) {
3178 /*
3179 * case b: block @split is the block that the extent begins with
3180 * then we just change the state of the extent, and splitting
3181 * is not needed.
3182 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003183 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3184 ext4_ext_mark_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003185 else
3186 ext4_ext_mark_initialized(ex);
3187
3188 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003189 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003190
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003191 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003192 goto out;
3193 }
3194
3195 /* case a */
3196 memcpy(&orig_ex, ex, sizeof(orig_ex));
3197 ex->ee_len = cpu_to_le16(split - ee_block);
Lukas Czerner556615d2014-04-20 23:45:47 -04003198 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3199 ext4_ext_mark_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003200
3201 /*
3202 * path may lead to new leaf, not to original leaf any more
3203 * after ext4_ext_insert_extent() returns,
3204 */
3205 err = ext4_ext_dirty(handle, inode, path + depth);
3206 if (err)
3207 goto fix_extent_len;
3208
3209 ex2 = &newex;
3210 ex2->ee_block = cpu_to_le32(split);
3211 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3212 ext4_ext_store_pblock(ex2, newblock);
Lukas Czerner556615d2014-04-20 23:45:47 -04003213 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3214 ext4_ext_mark_unwritten(ex2);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003215
3216 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3217 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003218 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003219 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003220 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003221 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003222 zero_ex.ee_len = cpu_to_le16(
3223 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003224 ext4_ext_store_pblock(&zero_ex,
3225 ext4_ext_pblock(ex2));
3226 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003227 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003228 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003229 zero_ex.ee_len = cpu_to_le16(
3230 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003231 ext4_ext_store_pblock(&zero_ex,
3232 ext4_ext_pblock(ex));
3233 }
3234 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003235 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003236 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003237 zero_ex.ee_len = cpu_to_le16(
3238 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003239 ext4_ext_store_pblock(&zero_ex,
3240 ext4_ext_pblock(&orig_ex));
3241 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003242
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003243 if (err)
3244 goto fix_extent_len;
3245 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003246 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003247 ext4_ext_try_to_merge(handle, inode, path, ex);
3248 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003249 if (err)
3250 goto fix_extent_len;
3251
3252 /* update extent status tree */
Zheng Liud7b2a002013-08-28 14:47:06 -04003253 err = ext4_zeroout_es(inode, &zero_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003254
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003255 goto out;
3256 } else if (err)
3257 goto fix_extent_len;
3258
3259out:
3260 ext4_ext_show_leaf(inode, path);
3261 return err;
3262
3263fix_extent_len:
3264 ex->ee_len = orig_ex.ee_len;
Dmitry Monakhov29faed12014-07-27 22:30:29 -04003265 ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003266 return err;
3267}
3268
3269/*
3270 * ext4_split_extents() splits an extent and mark extent which is covered
3271 * by @map as split_flags indicates
3272 *
Anatol Pomozov70261f52013-08-28 14:40:12 -04003273 * It may result in splitting the extent into multiple extents (up to three)
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003274 * There are three possibilities:
3275 * a> There is no split required
3276 * b> Splits in two extents: Split is happening at either end of the extent
3277 * c> Splits in three extents: Somone is splitting in middle of the extent
3278 *
3279 */
3280static int ext4_split_extent(handle_t *handle,
3281 struct inode *inode,
3282 struct ext4_ext_path *path,
3283 struct ext4_map_blocks *map,
3284 int split_flag,
3285 int flags)
3286{
3287 ext4_lblk_t ee_block;
3288 struct ext4_extent *ex;
3289 unsigned int ee_len, depth;
3290 int err = 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04003291 int unwritten;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003292 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003293 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003294
3295 depth = ext_depth(inode);
3296 ex = path[depth].p_ext;
3297 ee_block = le32_to_cpu(ex->ee_block);
3298 ee_len = ext4_ext_get_actual_len(ex);
Lukas Czerner556615d2014-04-20 23:45:47 -04003299 unwritten = ext4_ext_is_unwritten(ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003300
3301 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003302 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003303 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
Lukas Czerner556615d2014-04-20 23:45:47 -04003304 if (unwritten)
3305 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3306 EXT4_EXT_MARK_UNWRIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003307 if (split_flag & EXT4_EXT_DATA_VALID2)
3308 split_flag1 |= EXT4_EXT_DATA_VALID1;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003309 err = ext4_split_extent_at(handle, inode, path,
3310 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003311 if (err)
3312 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003313 } else {
3314 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003315 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003316 /*
3317 * Update path is required because previous ext4_split_extent_at() may
3318 * result in split of original leaf or extent zeroout.
3319 */
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003320 ext4_ext_drop_refs(path);
Theodore Ts'o705912c2014-09-01 14:34:09 -04003321 path = ext4_ext_find_extent(inode, map->m_lblk, &path,
3322 EXT4_EX_NOFREE_ON_ERR);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003323 if (IS_ERR(path))
3324 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003325 depth = ext_depth(inode);
3326 ex = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04003327 if (!ex) {
3328 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3329 (unsigned long) map->m_lblk);
3330 return -EIO;
3331 }
Lukas Czerner556615d2014-04-20 23:45:47 -04003332 unwritten = ext4_ext_is_unwritten(ex);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003333 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003334
3335 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003336 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
Lukas Czerner556615d2014-04-20 23:45:47 -04003337 if (unwritten) {
3338 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003339 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
Lukas Czerner556615d2014-04-20 23:45:47 -04003340 EXT4_EXT_MARK_UNWRIT2);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003341 }
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003342 err = ext4_split_extent_at(handle, inode, path,
3343 map->m_lblk, split_flag1, flags);
3344 if (err)
3345 goto out;
3346 }
3347
3348 ext4_ext_show_leaf(inode, path);
3349out:
Zheng Liu3a225672013-03-10 21:20:23 -04003350 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003351}
3352
Amit Arora56055d32007-07-17 21:42:38 -04003353/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003354 * This function is called by ext4_ext_map_blocks() if someone tries to write
Lukas Czerner556615d2014-04-20 23:45:47 -04003355 * to an unwritten extent. It may result in splitting the unwritten
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003356 * extent into multiple extents (up to three - one initialized and two
Lukas Czerner556615d2014-04-20 23:45:47 -04003357 * unwritten).
Amit Arora56055d32007-07-17 21:42:38 -04003358 * There are three possibilities:
3359 * a> There is no split required: Entire extent should be initialized
3360 * b> Splits in two extents: Write is happening at either end of the extent
3361 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003362 *
3363 * Pre-conditions:
Lukas Czerner556615d2014-04-20 23:45:47 -04003364 * - The extent pointed to by 'path' is unwritten.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003365 * - The extent pointed to by 'path' contains a superset
3366 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3367 *
3368 * Post-conditions on success:
3369 * - the returned value is the number of blocks beyond map->l_lblk
3370 * that are allocated and initialized.
3371 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003372 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003373static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003374 struct inode *inode,
3375 struct ext4_map_blocks *map,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003376 struct ext4_ext_path *path,
3377 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003378{
Zheng Liu67a5da52012-08-17 09:54:17 -04003379 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003380 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003381 struct ext4_map_blocks split_map;
3382 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003383 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003384 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003385 unsigned int ee_len, depth, map_len = map->m_len;
3386 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003387 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003388 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003389
3390 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3391 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003392 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003393
Zheng Liu67a5da52012-08-17 09:54:17 -04003394 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003395 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3396 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003397 if (eof_block < map->m_lblk + map_len)
3398 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003399
3400 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003401 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003402 ex = path[depth].p_ext;
3403 ee_block = le32_to_cpu(ex->ee_block);
3404 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003405 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003406
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003407 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3408
3409 /* Pre-conditions */
Lukas Czerner556615d2014-04-20 23:45:47 -04003410 BUG_ON(!ext4_ext_is_unwritten(ex));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003411 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003412
3413 /*
3414 * Attempt to transfer newly initialized blocks from the currently
Lukas Czerner556615d2014-04-20 23:45:47 -04003415 * unwritten extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003416 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003417 * memmove() calls. Transferring to the left is the common case in
3418 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3419 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003420 *
3421 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003422 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003423 * This would require removing the extent if the transfer
3424 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003425 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003426 * same extent tree node.
3427 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003428 if ((map->m_lblk == ee_block) &&
3429 /* See if we can merge left */
3430 (map_len < ee_len) && /*L1*/
3431 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003432 ext4_lblk_t prev_lblk;
3433 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003434 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003435
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003436 abut_ex = ex - 1;
3437 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3438 prev_len = ext4_ext_get_actual_len(abut_ex);
3439 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003440 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003441
3442 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003443 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003444 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003445 * - C1: abut_ex is initialized,
3446 * - C2: abut_ex is logically abutting ex,
3447 * - C3: abut_ex is physically abutting ex,
3448 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003449 * overflowing the (initialized) length limit.
3450 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003451 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003452 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3453 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003454 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003455 err = ext4_ext_get_access(handle, inode, path + depth);
3456 if (err)
3457 goto out;
3458
3459 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003460 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003461
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003462 /* Shift the start of ex by 'map_len' blocks */
3463 ex->ee_block = cpu_to_le32(ee_block + map_len);
3464 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3465 ex->ee_len = cpu_to_le16(ee_len - map_len);
Lukas Czerner556615d2014-04-20 23:45:47 -04003466 ext4_ext_mark_unwritten(ex); /* Restore the flag */
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003467
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003468 /* Extend abut_ex by 'map_len' blocks */
3469 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003470
3471 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003472 allocated = map_len;
3473 }
3474 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3475 (map_len < ee_len) && /*L1*/
3476 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3477 /* See if we can merge right */
3478 ext4_lblk_t next_lblk;
3479 ext4_fsblk_t next_pblk, ee_pblk;
3480 unsigned int next_len;
3481
3482 abut_ex = ex + 1;
3483 next_lblk = le32_to_cpu(abut_ex->ee_block);
3484 next_len = ext4_ext_get_actual_len(abut_ex);
3485 next_pblk = ext4_ext_pblock(abut_ex);
3486 ee_pblk = ext4_ext_pblock(ex);
3487
3488 /*
3489 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3490 * upon those conditions:
3491 * - C1: abut_ex is initialized,
3492 * - C2: abut_ex is logically abutting ex,
3493 * - C3: abut_ex is physically abutting ex,
3494 * - C4: abut_ex can receive the additional blocks without
3495 * overflowing the (initialized) length limit.
3496 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003497 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003498 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3499 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3500 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3501 err = ext4_ext_get_access(handle, inode, path + depth);
3502 if (err)
3503 goto out;
3504
3505 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3506 map, ex, abut_ex);
3507
3508 /* Shift the start of abut_ex by 'map_len' blocks */
3509 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3510 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3511 ex->ee_len = cpu_to_le16(ee_len - map_len);
Lukas Czerner556615d2014-04-20 23:45:47 -04003512 ext4_ext_mark_unwritten(ex); /* Restore the flag */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003513
3514 /* Extend abut_ex by 'map_len' blocks */
3515 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3516
3517 /* Result: number of initialized blocks past m_lblk */
3518 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003519 }
3520 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003521 if (allocated) {
3522 /* Mark the block containing both extents as dirty */
3523 ext4_ext_dirty(handle, inode, path + depth);
3524
3525 /* Update path to point to the right extent */
3526 path[depth].p_ext = abut_ex;
3527 goto out;
3528 } else
3529 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003530
Yongqiang Yang667eff32011-05-03 12:25:07 -04003531 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003532 /*
3533 * It is safe to convert extent to initialized via explicit
Yongqiang Yang9e740562014-01-06 14:05:23 -05003534 * zeroout only if extent is fully inside i_size or new_size.
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003535 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003536 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003537
Zheng Liu67a5da52012-08-17 09:54:17 -04003538 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3539 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003540 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003541
3542 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3543 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003544 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003545 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003546 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003547 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003548 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003549 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003550
3551 err = ext4_ext_get_access(handle, inode, path + depth);
3552 if (err)
3553 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003554 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003555 ext4_ext_try_to_merge(handle, inode, path, ex);
3556 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003557 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003558 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003559
Amit Arora56055d32007-07-17 21:42:38 -04003560 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003561 * four cases:
3562 * 1. split the extent into three extents.
3563 * 2. split the extent into two extents, zeroout the first half.
3564 * 3. split the extent into two extents, zeroout the second half.
3565 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003566 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003567 split_map.m_lblk = map->m_lblk;
3568 split_map.m_len = map->m_len;
3569
Zheng Liu67a5da52012-08-17 09:54:17 -04003570 if (max_zeroout && (allocated > map->m_len)) {
3571 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003572 /* case 3 */
3573 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003574 cpu_to_le32(map->m_lblk);
3575 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003576 ext4_ext_store_pblock(&zero_ex,
3577 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3578 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003579 if (err)
3580 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003581 split_map.m_lblk = map->m_lblk;
3582 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003583 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003584 /* case 2 */
3585 if (map->m_lblk != ee_block) {
3586 zero_ex.ee_block = ex->ee_block;
3587 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3588 ee_block);
3589 ext4_ext_store_pblock(&zero_ex,
3590 ext4_ext_pblock(ex));
3591 err = ext4_ext_zeroout(inode, &zero_ex);
3592 if (err)
3593 goto out;
3594 }
3595
Yongqiang Yang667eff32011-05-03 12:25:07 -04003596 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003597 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3598 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003599 }
3600 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003601
3602 allocated = ext4_split_extent(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003603 &split_map, split_flag, flags);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003604 if (allocated < 0)
3605 err = allocated;
3606
Amit Arora56055d32007-07-17 21:42:38 -04003607out:
Zheng Liuadb23552013-03-10 21:13:05 -04003608 /* If we have gotten a failure, don't zero out status tree */
3609 if (!err)
Zheng Liud7b2a002013-08-28 14:47:06 -04003610 err = ext4_zeroout_es(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003611 return err ? err : allocated;
3612}
3613
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003614/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003615 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003616 * ext4_get_blocks_dio_write() when DIO to write
Lukas Czerner556615d2014-04-20 23:45:47 -04003617 * to an unwritten extent.
Mingming Cao00314622009-09-28 15:49:08 -04003618 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003619 * Writing to an unwritten extent may result in splitting the unwritten
3620 * extent into multiple initialized/unwritten extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003621 * There are three possibilities:
Lukas Czerner556615d2014-04-20 23:45:47 -04003622 * a> There is no split required: Entire extent should be unwritten
Mingming Cao00314622009-09-28 15:49:08 -04003623 * b> Splits in two extents: Write is happening at either end of the extent
3624 * c> Splits in three extents: Somone is writing in middle of the extent
3625 *
Lukas Czernerb8a86842014-03-18 18:05:35 -04003626 * This works the same way in the case of initialized -> unwritten conversion.
3627 *
Mingming Cao00314622009-09-28 15:49:08 -04003628 * One of more index blocks maybe needed if the extent tree grow after
Lukas Czerner556615d2014-04-20 23:45:47 -04003629 * the unwritten extent split. To prevent ENOSPC occur at the IO
3630 * complete, we need to split the unwritten extent before DIO submit
3631 * the IO. The unwritten extent called at this time will be split
3632 * into three unwritten extent(at most). After IO complete, the part
Mingming Cao00314622009-09-28 15:49:08 -04003633 * being filled will be convert to initialized by the end_io callback function
3634 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003635 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003636 * Returns the size of unwritten extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003637 */
Lukas Czernerb8a86842014-03-18 18:05:35 -04003638static int ext4_split_convert_extents(handle_t *handle,
Mingming Cao00314622009-09-28 15:49:08 -04003639 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003640 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003641 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003642 int flags)
3643{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003644 ext4_lblk_t eof_block;
3645 ext4_lblk_t ee_block;
3646 struct ext4_extent *ex;
3647 unsigned int ee_len;
3648 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003649
Lukas Czernerb8a86842014-03-18 18:05:35 -04003650 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3651 __func__, inode->i_ino,
3652 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003653
3654 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3655 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003656 if (eof_block < map->m_lblk + map->m_len)
3657 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003658 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003659 * It is safe to convert extent to initialized via explicit
3660 * zeroout only if extent is fully insde i_size or new_size.
3661 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003662 depth = ext_depth(inode);
3663 ex = path[depth].p_ext;
3664 ee_block = le32_to_cpu(ex->ee_block);
3665 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003666
Lukas Czernerb8a86842014-03-18 18:05:35 -04003667 /* Convert to unwritten */
3668 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3669 split_flag |= EXT4_EXT_DATA_VALID1;
3670 /* Convert to initialized */
3671 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3672 split_flag |= ee_block + ee_len <= eof_block ?
3673 EXT4_EXT_MAY_ZEROOUT : 0;
Lukas Czerner556615d2014-04-20 23:45:47 -04003674 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
Lukas Czernerb8a86842014-03-18 18:05:35 -04003675 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003676 flags |= EXT4_GET_BLOCKS_PRE_IO;
3677 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003678}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003679
Lukas Czernerb8a86842014-03-18 18:05:35 -04003680static int ext4_convert_initialized_extents(handle_t *handle,
3681 struct inode *inode,
3682 struct ext4_map_blocks *map,
3683 struct ext4_ext_path *path)
3684{
3685 struct ext4_extent *ex;
3686 ext4_lblk_t ee_block;
3687 unsigned int ee_len;
3688 int depth;
3689 int err = 0;
3690
3691 depth = ext_depth(inode);
3692 ex = path[depth].p_ext;
3693 ee_block = le32_to_cpu(ex->ee_block);
3694 ee_len = ext4_ext_get_actual_len(ex);
3695
3696 ext_debug("%s: inode %lu, logical"
3697 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3698 (unsigned long long)ee_block, ee_len);
3699
3700 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3701 err = ext4_split_convert_extents(handle, inode, map, path,
3702 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3703 if (err < 0)
3704 goto out;
3705 ext4_ext_drop_refs(path);
Theodore Ts'o705912c2014-09-01 14:34:09 -04003706 path = ext4_ext_find_extent(inode, map->m_lblk, &path,
3707 EXT4_EX_NOFREE_ON_ERR);
Lukas Czernerb8a86842014-03-18 18:05:35 -04003708 if (IS_ERR(path)) {
3709 err = PTR_ERR(path);
3710 goto out;
3711 }
3712 depth = ext_depth(inode);
3713 ex = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04003714 if (!ex) {
3715 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3716 (unsigned long) map->m_lblk);
3717 err = -EIO;
3718 goto out;
3719 }
Lukas Czernerb8a86842014-03-18 18:05:35 -04003720 }
3721
3722 err = ext4_ext_get_access(handle, inode, path + depth);
3723 if (err)
3724 goto out;
Lukas Czerner556615d2014-04-20 23:45:47 -04003725 /* first mark the extent as unwritten */
3726 ext4_ext_mark_unwritten(ex);
Lukas Czernerb8a86842014-03-18 18:05:35 -04003727
3728 /* note: ext4_ext_correct_indexes() isn't needed here because
3729 * borders are not changed
3730 */
3731 ext4_ext_try_to_merge(handle, inode, path, ex);
3732
3733 /* Mark modified extent as dirty */
3734 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3735out:
3736 ext4_ext_show_leaf(inode, path);
3737 return err;
3738}
3739
3740
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003741static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003742 struct inode *inode,
3743 struct ext4_map_blocks *map,
3744 struct ext4_ext_path *path)
Mingming Cao00314622009-09-28 15:49:08 -04003745{
3746 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003747 ext4_lblk_t ee_block;
3748 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003749 int depth;
3750 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003751
3752 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003753 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003754 ee_block = le32_to_cpu(ex->ee_block);
3755 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003756
Yongqiang Yang197217a2011-05-03 11:45:29 -04003757 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3758 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003759 (unsigned long long)ee_block, ee_len);
3760
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003761 /* If extent is larger than requested it is a clear sign that we still
3762 * have some extent state machine issues left. So extent_split is still
3763 * required.
3764 * TODO: Once all related issues will be fixed this situation should be
3765 * illegal.
3766 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003767 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003768#ifdef EXT4_DEBUG
3769 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3770 " len %u; IO logical block %llu, len %u\n",
3771 inode->i_ino, (unsigned long long)ee_block, ee_len,
3772 (unsigned long long)map->m_lblk, map->m_len);
3773#endif
Lukas Czernerb8a86842014-03-18 18:05:35 -04003774 err = ext4_split_convert_extents(handle, inode, map, path,
3775 EXT4_GET_BLOCKS_CONVERT);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003776 if (err < 0)
3777 goto out;
3778 ext4_ext_drop_refs(path);
Theodore Ts'o705912c2014-09-01 14:34:09 -04003779 path = ext4_ext_find_extent(inode, map->m_lblk, &path,
3780 EXT4_EX_NOFREE_ON_ERR);
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003781 if (IS_ERR(path)) {
3782 err = PTR_ERR(path);
3783 goto out;
3784 }
3785 depth = ext_depth(inode);
3786 ex = path[depth].p_ext;
3787 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003788
Mingming Cao00314622009-09-28 15:49:08 -04003789 err = ext4_ext_get_access(handle, inode, path + depth);
3790 if (err)
3791 goto out;
3792 /* first mark the extent as initialized */
3793 ext4_ext_mark_initialized(ex);
3794
Yongqiang Yang197217a2011-05-03 11:45:29 -04003795 /* note: ext4_ext_correct_indexes() isn't needed here because
3796 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003797 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003798 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003799
Mingming Cao00314622009-09-28 15:49:08 -04003800 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003801 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003802out:
3803 ext4_ext_show_leaf(inode, path);
3804 return err;
3805}
3806
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003807static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3808 sector_t block, int count)
3809{
3810 int i;
3811 for (i = 0; i < count; i++)
3812 unmap_underlying_metadata(bdev, block + i);
3813}
3814
Theodore Ts'o58590b02010-10-27 21:23:12 -04003815/*
3816 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3817 */
3818static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003819 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003820 struct ext4_ext_path *path,
3821 unsigned int len)
3822{
3823 int i, depth;
3824 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003825 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003826
3827 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3828 return 0;
3829
3830 depth = ext_depth(inode);
3831 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003832
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003833 /*
3834 * We're going to remove EOFBLOCKS_FL entirely in future so we
3835 * do not care for this case anymore. Simply remove the flag
3836 * if there are no extents.
3837 */
3838 if (unlikely(!eh->eh_entries))
3839 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003840 last_ex = EXT_LAST_EXTENT(eh);
3841 /*
3842 * We should clear the EOFBLOCKS_FL flag if we are writing the
3843 * last block in the last extent in the file. We test this by
3844 * first checking to see if the caller to
3845 * ext4_ext_get_blocks() was interested in the last block (or
3846 * a block beyond the last block) in the current extent. If
3847 * this turns out to be false, we can bail out from this
3848 * function immediately.
3849 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003850 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003851 ext4_ext_get_actual_len(last_ex))
3852 return 0;
3853 /*
3854 * If the caller does appear to be planning to write at or
3855 * beyond the end of the current extent, we then test to see
3856 * if the current extent is the last extent in the file, by
3857 * checking to make sure it was reached via the rightmost node
3858 * at each level of the tree.
3859 */
3860 for (i = depth-1; i >= 0; i--)
3861 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3862 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003863out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003864 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3865 return ext4_mark_inode_dirty(handle, inode);
3866}
3867
Aditya Kali7b415bf2011-09-09 19:04:51 -04003868/**
3869 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3870 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003871 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003872 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003873int ext4_find_delalloc_range(struct inode *inode,
3874 ext4_lblk_t lblk_start,
3875 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003876{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003877 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003878
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003879 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003880 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003881 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003882 else if (es.es_lblk <= lblk_start &&
3883 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003884 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003885 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003886 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003887 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003888 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003889}
3890
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003891int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003892{
3893 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3894 ext4_lblk_t lblk_start, lblk_end;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003895 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003896 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3897
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003898 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003899}
3900
3901/**
3902 * Determines how many complete clusters (out of those specified by the 'map')
3903 * are under delalloc and were reserved quota for.
3904 * This function is called when we are writing out the blocks that were
3905 * originally written with their allocation delayed, but then the space was
3906 * allocated using fallocate() before the delayed allocation could be resolved.
3907 * The cases to look for are:
3908 * ('=' indicated delayed allocated blocks
3909 * '-' indicates non-delayed allocated blocks)
3910 * (a) partial clusters towards beginning and/or end outside of allocated range
3911 * are not delalloc'ed.
3912 * Ex:
3913 * |----c---=|====c====|====c====|===-c----|
3914 * |++++++ allocated ++++++|
3915 * ==> 4 complete clusters in above example
3916 *
3917 * (b) partial cluster (outside of allocated range) towards either end is
3918 * marked for delayed allocation. In this case, we will exclude that
3919 * cluster.
3920 * Ex:
3921 * |----====c========|========c========|
3922 * |++++++ allocated ++++++|
3923 * ==> 1 complete clusters in above example
3924 *
3925 * Ex:
3926 * |================c================|
3927 * |++++++ allocated ++++++|
3928 * ==> 0 complete clusters in above example
3929 *
3930 * The ext4_da_update_reserve_space will be called only if we
3931 * determine here that there were some "entire" clusters that span
3932 * this 'allocated' range.
3933 * In the non-bigalloc case, this function will just end up returning num_blks
3934 * without ever calling ext4_find_delalloc_range.
3935 */
3936static unsigned int
3937get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3938 unsigned int num_blks)
3939{
3940 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3941 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3942 ext4_lblk_t lblk_from, lblk_to, c_offset;
3943 unsigned int allocated_clusters = 0;
3944
3945 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3946 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3947
3948 /* max possible clusters for this allocation */
3949 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3950
Aditya Kalid8990242011-09-09 19:18:51 -04003951 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3952
Aditya Kali7b415bf2011-09-09 19:04:51 -04003953 /* Check towards left side */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003954 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003955 if (c_offset) {
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003956 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003957 lblk_to = lblk_from + c_offset - 1;
3958
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003959 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003960 allocated_clusters--;
3961 }
3962
3963 /* Now check towards right. */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05003964 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003965 if (allocated_clusters && c_offset) {
3966 lblk_from = lblk_start + num_blks;
3967 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3968
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003969 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003970 allocated_clusters--;
3971 }
3972
3973 return allocated_clusters;
3974}
3975
Mingming Cao00314622009-09-28 15:49:08 -04003976static int
Lukas Czernerb8a86842014-03-18 18:05:35 -04003977ext4_ext_convert_initialized_extent(handle_t *handle, struct inode *inode,
3978 struct ext4_map_blocks *map,
3979 struct ext4_ext_path *path, int flags,
3980 unsigned int allocated, ext4_fsblk_t newblock)
3981{
3982 int ret = 0;
3983 int err = 0;
3984
3985 /*
3986 * Make sure that the extent is no bigger than we support with
Lukas Czerner556615d2014-04-20 23:45:47 -04003987 * unwritten extent
Lukas Czernerb8a86842014-03-18 18:05:35 -04003988 */
Lukas Czerner556615d2014-04-20 23:45:47 -04003989 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3990 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
Lukas Czernerb8a86842014-03-18 18:05:35 -04003991
3992 ret = ext4_convert_initialized_extents(handle, inode, map,
3993 path);
3994 if (ret >= 0) {
3995 ext4_update_inode_fsync_trans(handle, inode, 1);
3996 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3997 path, map->m_len);
3998 } else
3999 err = ret;
4000 map->m_flags |= EXT4_MAP_UNWRITTEN;
4001 if (allocated > map->m_len)
4002 allocated = map->m_len;
4003 map->m_len = allocated;
4004
4005 return err ? err : allocated;
4006}
4007
4008static int
Lukas Czerner556615d2014-04-20 23:45:47 -04004009ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004010 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04004011 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004012 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04004013{
4014 int ret = 0;
4015 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004016 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04004017
Lukas Czerner556615d2014-04-20 23:45:47 -04004018 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
Zheng Liu88635ca2011-12-28 19:00:25 -05004019 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004020 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04004021 flags, allocated);
4022 ext4_ext_show_leaf(inode, path);
4023
Lukas Czerner27dd4382013-04-09 22:11:22 -04004024 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04004025 * When writing into unwritten space, we should not fail to
Lukas Czerner27dd4382013-04-09 22:11:22 -04004026 * allocate metadata blocks for the new extent block if needed.
4027 */
4028 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4029
Lukas Czerner556615d2014-04-20 23:45:47 -04004030 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
Zheng Liub5645532012-11-08 14:33:43 -05004031 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04004032
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004033 /* get_block() before submit the IO, split the extent */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004034 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
Lukas Czernerb8a86842014-03-18 18:05:35 -04004035 ret = ext4_split_convert_extents(handle, inode, map,
4036 path, flags | EXT4_GET_BLOCKS_CONVERT);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004037 if (ret <= 0)
4038 goto out;
Mingming5f5249502009-11-10 10:48:04 -05004039 /*
4040 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004041 * that this IO needs to conversion to written when IO is
Mingming5f5249502009-11-10 10:48:04 -05004042 * completed
4043 */
Tao Ma0edeb712011-10-31 17:30:44 -04004044 if (io)
4045 ext4_set_io_unwritten_flag(inode, io);
4046 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004047 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05004048 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004049 goto out;
4050 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004051 /* IO end_io complete, convert the filled extent to written */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004052 if (flags & EXT4_GET_BLOCKS_CONVERT) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04004053 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Mingming Cao00314622009-09-28 15:49:08 -04004054 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04004055 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05004056 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05004057 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4058 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04004059 } else
4060 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04004061 map->m_flags |= EXT4_MAP_MAPPED;
Eric Whitney15cc1762014-02-12 10:42:45 -05004062 map->m_pblk = newblock;
Zheng Liucdee7842013-03-10 21:08:52 -04004063 if (allocated > map->m_len)
4064 allocated = map->m_len;
4065 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04004066 goto out2;
4067 }
4068 /* buffered IO case */
4069 /*
4070 * repeat fallocate creation request
4071 * we already have an unwritten extent
4072 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004073 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
Zheng Liua25a4e12013-02-18 00:28:04 -05004074 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004075 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05004076 }
Mingming Cao00314622009-09-28 15:49:08 -04004077
4078 /* buffered READ or buffered write_begin() lookup */
4079 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4080 /*
4081 * We have blocks reserved already. We
4082 * return allocated blocks so that delalloc
4083 * won't do block reservation for us. But
4084 * the buffer head will be unmapped so that
4085 * a read from the block returns 0s.
4086 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004087 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04004088 goto out1;
4089 }
4090
4091 /* buffered write, writepage time, convert*/
Lukas Czerner27dd4382013-04-09 22:11:22 -04004092 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004093 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004094 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04004095out:
4096 if (ret <= 0) {
4097 err = ret;
4098 goto out2;
4099 } else
4100 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004101 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004102 /*
4103 * if we allocated more blocks than requested
4104 * we need to make sure we unmap the extra block
4105 * allocated. The actual needed block will get
4106 * unmapped later when we find the buffer_head marked
4107 * new.
4108 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004109 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004110 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004111 newblock + map->m_len,
4112 allocated - map->m_len);
4113 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05004114 }
Zheng Liu3a225672013-03-10 21:20:23 -04004115 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004116
4117 /*
4118 * If we have done fallocate with the offset that is already
4119 * delayed allocated, we would have block reservation
4120 * and quota reservation done in the delayed write path.
4121 * But fallocate would have already updated quota and block
4122 * count for this offset. So cancel these reservation
4123 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004124 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4125 unsigned int reserved_clusters;
4126 reserved_clusters = get_reserved_cluster_alloc(inode,
4127 map->m_lblk, map->m_len);
4128 if (reserved_clusters)
4129 ext4_da_update_reserve_space(inode,
4130 reserved_clusters,
4131 0);
4132 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004133
Mingming Cao00314622009-09-28 15:49:08 -04004134map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004135 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004136 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4137 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4138 map->m_len);
4139 if (err < 0)
4140 goto out2;
4141 }
Mingming Cao00314622009-09-28 15:49:08 -04004142out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004143 if (allocated > map->m_len)
4144 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04004145 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004146 map->m_pblk = newblock;
4147 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04004148out2:
Mingming Cao00314622009-09-28 15:49:08 -04004149 return err ? err : allocated;
4150}
Theodore Ts'o58590b02010-10-27 21:23:12 -04004151
Mingming Cao00314622009-09-28 15:49:08 -04004152/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004153 * get_implied_cluster_alloc - check to see if the requested
4154 * allocation (in the map structure) overlaps with a cluster already
4155 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04004156 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004157 * @map The requested lblk->pblk mapping
4158 * @ex The extent structure which might contain an implied
4159 * cluster allocation
4160 *
4161 * This function is called by ext4_ext_map_blocks() after we failed to
4162 * find blocks that were already in the inode's extent tree. Hence,
4163 * we know that the beginning of the requested region cannot overlap
4164 * the extent from the inode's extent tree. There are three cases we
4165 * want to catch. The first is this case:
4166 *
4167 * |--- cluster # N--|
4168 * |--- extent ---| |---- requested region ---|
4169 * |==========|
4170 *
4171 * The second case that we need to test for is this one:
4172 *
4173 * |--------- cluster # N ----------------|
4174 * |--- requested region --| |------- extent ----|
4175 * |=======================|
4176 *
4177 * The third case is when the requested region lies between two extents
4178 * within the same cluster:
4179 * |------------- cluster # N-------------|
4180 * |----- ex -----| |---- ex_right ----|
4181 * |------ requested region ------|
4182 * |================|
4183 *
4184 * In each of the above cases, we need to set the map->m_pblk and
4185 * map->m_len so it corresponds to the return the extent labelled as
4186 * "|====|" from cluster #N, since it is already in use for data in
4187 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4188 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4189 * as a new "allocated" block region. Otherwise, we will return 0 and
4190 * ext4_ext_map_blocks() will then allocate one or more new clusters
4191 * by calling ext4_mb_new_blocks().
4192 */
Aditya Kalid8990242011-09-09 19:18:51 -04004193static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004194 struct ext4_map_blocks *map,
4195 struct ext4_extent *ex,
4196 struct ext4_ext_path *path)
4197{
Aditya Kalid8990242011-09-09 19:18:51 -04004198 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004199 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004200 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05004201 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004202 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4203 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4204 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4205
4206 /* The extent passed in that we are trying to match */
4207 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4208 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4209
4210 /* The requested region passed into ext4_map_blocks() */
4211 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004212
4213 if ((rr_cluster_start == ex_cluster_end) ||
4214 (rr_cluster_start == ex_cluster_start)) {
4215 if (rr_cluster_start == ex_cluster_end)
4216 ee_start += ee_len - 1;
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004217 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004218 map->m_len = min(map->m_len,
4219 (unsigned) sbi->s_cluster_ratio - c_offset);
4220 /*
4221 * Check for and handle this case:
4222 *
4223 * |--------- cluster # N-------------|
4224 * |------- extent ----|
4225 * |--- requested region ---|
4226 * |===========|
4227 */
4228
4229 if (map->m_lblk < ee_block)
4230 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4231
4232 /*
4233 * Check for the case where there is already another allocated
4234 * block to the right of 'ex' but before the end of the cluster.
4235 *
4236 * |------------- cluster # N-------------|
4237 * |----- ex -----| |---- ex_right ----|
4238 * |------ requested region ------|
4239 * |================|
4240 */
4241 if (map->m_lblk > ee_block) {
4242 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4243 map->m_len = min(map->m_len, next - map->m_lblk);
4244 }
Aditya Kalid8990242011-09-09 19:18:51 -04004245
4246 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004247 return 1;
4248 }
Aditya Kalid8990242011-09-09 19:18:51 -04004249
4250 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004251 return 0;
4252}
4253
4254
4255/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05004256 * Block allocation/map/preallocation routine for extents based files
4257 *
4258 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004259 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004260 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4261 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004262 *
4263 * return > 0, number of of blocks already mapped/allocated
4264 * if create == 0 and these are pre-allocated blocks
4265 * buffer head is unmapped
4266 * otherwise blocks are mapped
4267 *
4268 * return = 0, if plain look up failed (blocks have not been allocated)
4269 * buffer head is unmapped
4270 *
4271 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004272 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004273int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4274 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004275{
4276 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004277 struct ext4_extent newex, *ex, *ex2;
4278 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004279 ext4_fsblk_t newblock = 0;
Eric Whitneyce37c422014-02-19 18:52:39 -05004280 int free_on_err = 0, err = 0, depth, ret;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004281 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004282 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004283 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004284 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004285 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004286 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07004287
Mingming84fe3be2009-09-01 08:44:37 -04004288 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004289 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004290 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004291
Alex Tomasa86c6182006-10-11 01:21:03 -07004292 /* find extent for this block */
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04004293 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004294 if (IS_ERR(path)) {
4295 err = PTR_ERR(path);
4296 path = NULL;
4297 goto out2;
4298 }
4299
4300 depth = ext_depth(inode);
4301
4302 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004303 * consistent leaf must not be empty;
4304 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07004305 * this is why assert can't be put in ext4_ext_find_extent()
4306 */
Frank Mayhar273df552010-03-02 11:46:09 -05004307 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4308 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004309 "lblock: %lu, depth: %d pblock %lld",
4310 (unsigned long) map->m_lblk, depth,
4311 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004312 err = -EIO;
4313 goto out2;
4314 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004315
Avantika Mathur7e028972006-12-06 20:41:33 -08004316 ex = path[depth].p_ext;
4317 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004318 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004319 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004320 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004321
Lukas Czernerb8a86842014-03-18 18:05:35 -04004322
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004323 /*
Lukas Czerner556615d2014-04-20 23:45:47 -04004324 * unwritten extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004325 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004326 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004327 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004328
4329 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4330
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004331 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004332 if (in_range(map->m_lblk, ee_block, ee_len)) {
4333 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004334 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004335 allocated = ee_len - (map->m_lblk - ee_block);
4336 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4337 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004338
Lukas Czernerb8a86842014-03-18 18:05:35 -04004339 /*
4340 * If the extent is initialized check whether the
4341 * caller wants to convert it to unwritten.
4342 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004343 if ((!ext4_ext_is_unwritten(ex)) &&
Lukas Czernerb8a86842014-03-18 18:05:35 -04004344 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4345 allocated = ext4_ext_convert_initialized_extent(
4346 handle, inode, map, path, flags,
4347 allocated, newblock);
4348 goto out2;
Lukas Czerner556615d2014-04-20 23:45:47 -04004349 } else if (!ext4_ext_is_unwritten(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004350 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004351
Lukas Czerner556615d2014-04-20 23:45:47 -04004352 ret = ext4_ext_handle_unwritten_extents(
Lukas Czerner78771912012-03-19 23:05:43 -04004353 handle, inode, map, path, flags,
4354 allocated, newblock);
Eric Whitneyce37c422014-02-19 18:52:39 -05004355 if (ret < 0)
4356 err = ret;
4357 else
4358 allocated = ret;
Eric Whitney31cf0f22014-03-13 23:14:46 -04004359 goto out2;
Alex Tomasa86c6182006-10-11 01:21:03 -07004360 }
4361 }
4362
Aditya Kali7b415bf2011-09-09 19:04:51 -04004363 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05004364 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04004365 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4366
Alex Tomasa86c6182006-10-11 01:21:03 -07004367 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004368 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004369 * we couldn't try to create block if create flag is zero
4370 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004371 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004372 /*
4373 * put just found gap into cache to speed up
4374 * subsequent requests
4375 */
Zheng Liud100eef2013-02-18 00:29:59 -05004376 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4377 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004378 goto out2;
4379 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004380
Alex Tomasa86c6182006-10-11 01:21:03 -07004381 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004382 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004383 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004384 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004385 newex.ee_block = cpu_to_le32(map->m_lblk);
Eric Whitneyd0abafa2014-01-06 14:00:23 -05004386 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004387
4388 /*
4389 * If we are doing bigalloc, check to see if the extent returned
4390 * by ext4_ext_find_extent() implies a cluster we can use.
4391 */
4392 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004393 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004394 ar.len = allocated = map->m_len;
4395 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004396 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004397 goto got_allocated_blocks;
4398 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004399
Alex Tomasc9de5602008-01-29 00:19:52 -05004400 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004401 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004402 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4403 if (err)
4404 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004405 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004406 ex2 = NULL;
4407 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004408 if (err)
4409 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004410
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004411 /* Check if the extent after searching to the right implies a
4412 * cluster we can use. */
4413 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004414 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004415 ar.len = allocated = map->m_len;
4416 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004417 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004418 goto got_allocated_blocks;
4419 }
4420
Amit Arora749269f2007-07-18 09:02:56 -04004421 /*
4422 * See if request is beyond maximum number of blocks we can have in
4423 * a single extent. For an initialized extent this limit is
Lukas Czerner556615d2014-04-20 23:45:47 -04004424 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4425 * EXT_UNWRITTEN_MAX_LEN.
Amit Arora749269f2007-07-18 09:02:56 -04004426 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004427 if (map->m_len > EXT_INIT_MAX_LEN &&
Lukas Czerner556615d2014-04-20 23:45:47 -04004428 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004429 map->m_len = EXT_INIT_MAX_LEN;
Lukas Czerner556615d2014-04-20 23:45:47 -04004430 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4431 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4432 map->m_len = EXT_UNWRITTEN_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004433
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004434 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004435 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004436 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004437 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004438 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004439 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004440 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004441
4442 /* allocate new block */
4443 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004444 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4445 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004446 /*
4447 * We calculate the offset from the beginning of the cluster
4448 * for the logical block number, since when we allocate a
4449 * physical cluster, the physical block should start at the
4450 * same offset from the beginning of the cluster. This is
4451 * needed so that future calls to get_implied_cluster_alloc()
4452 * work correctly.
4453 */
Theodore Ts'of5a44db2013-12-20 09:29:35 -05004454 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004455 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4456 ar.goal -= offset;
4457 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004458 if (S_ISREG(inode->i_mode))
4459 ar.flags = EXT4_MB_HINT_DATA;
4460 else
4461 /* disable in-core preallocation for non-regular files */
4462 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004463 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4464 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004465 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004466 if (!newblock)
4467 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004468 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004469 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004470 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004471 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004472 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4473 if (ar.len > allocated)
4474 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004475
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004476got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004477 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004478 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004479 newex.ee_len = cpu_to_le16(ar.len);
Lukas Czerner556615d2014-04-20 23:45:47 -04004480 /* Mark unwritten */
4481 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4482 ext4_ext_mark_unwritten(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004483 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004484 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004485 * io_end structure was created for every IO write to an
Lukas Czerner556615d2014-04-20 23:45:47 -04004486 * unwritten extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004487 * here we flag the IO that really needs the conversion.
Mingming5f5249502009-11-10 10:48:04 -05004488 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004489 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004490 */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04004491 if (flags & EXT4_GET_BLOCKS_PRE_IO)
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004492 set_unwritten = 1;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004493 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004494
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004495 err = 0;
4496 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4497 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4498 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004499 if (!err)
4500 err = ext4_ext_insert_extent(handle, inode, path,
4501 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004502
4503 if (!err && set_unwritten) {
4504 if (io)
4505 ext4_set_io_unwritten_flag(inode, io);
4506 else
4507 ext4_set_inode_state(inode,
4508 EXT4_STATE_DIO_UNWRITTEN);
4509 }
4510
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004511 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004512 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4513 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004514 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004515 /* not a good idea to call discard here directly,
4516 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004517 ext4_discard_preallocations(inode);
Theodore Ts'oc8e15132013-07-15 00:09:37 -04004518 ext4_free_blocks(handle, inode, NULL, newblock,
4519 EXT4_C2B(sbi, allocated_clusters), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004520 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004521 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004522
Alex Tomasa86c6182006-10-11 01:21:03 -07004523 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004524 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004525 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004526 if (allocated > map->m_len)
4527 allocated = map->m_len;
4528 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004529
Jan Karab436b9b2009-12-08 23:51:10 -05004530 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004531 * Update reserved blocks/metadata blocks after successful
4532 * block allocation which had been deferred till now.
4533 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004534 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004535 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004536 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004537 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004538 */
4539 reserved_clusters = get_reserved_cluster_alloc(inode,
4540 map->m_lblk, allocated);
4541 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4542 if (reserved_clusters) {
4543 /*
4544 * We have clusters reserved for this range.
4545 * But since we are not doing actual allocation
4546 * and are simply using blocks from previously
4547 * allocated cluster, we should release the
4548 * reservation and not claim quota.
4549 */
4550 ext4_da_update_reserve_space(inode,
4551 reserved_clusters, 0);
4552 }
4553 } else {
4554 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004555 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f2612011-09-09 19:20:51 -04004556 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004557 int reservation = allocated_clusters -
4558 reserved_clusters;
4559 /*
4560 * It seems we claimed few clusters outside of
4561 * the range of this allocation. We should give
4562 * it back to the reservation pool. This can
4563 * happen in the following case:
4564 *
4565 * * Suppose s_cluster_ratio is 4 (i.e., each
4566 * cluster has 4 blocks. Thus, the clusters
4567 * are [0-3],[4-7],[8-11]...
4568 * * First comes delayed allocation write for
4569 * logical blocks 10 & 11. Since there were no
4570 * previous delayed allocated blocks in the
4571 * range [8-11], we would reserve 1 cluster
4572 * for this write.
4573 * * Next comes write for logical blocks 3 to 8.
4574 * In this case, we will reserve 2 clusters
4575 * (for [0-3] and [4-7]; and not for [8-11] as
4576 * that range has a delayed allocated blocks.
4577 * Thus total reserved clusters now becomes 3.
4578 * * Now, during the delayed allocation writeout
4579 * time, we will first write blocks [3-8] and
4580 * allocate 3 clusters for writing these
4581 * blocks. Also, we would claim all these
4582 * three clusters above.
4583 * * Now when we come here to writeout the
4584 * blocks [10-11], we would expect to claim
4585 * the reservation of 1 cluster we had made
4586 * (and we would claim it since there are no
4587 * more delayed allocated blocks in the range
4588 * [8-11]. But our reserved cluster count had
4589 * already gone to 0.
4590 *
4591 * Thus, at the step 4 above when we determine
4592 * that there are still some unwritten delayed
4593 * allocated blocks outside of our current
4594 * block range, we should increment the
4595 * reserved clusters count so that when the
4596 * remaining blocks finally gets written, we
4597 * could claim them.
4598 */
Aditya Kali5356f2612011-09-09 19:20:51 -04004599 dquot_reserve_block(inode,
4600 EXT4_C2B(sbi, reservation));
4601 spin_lock(&ei->i_block_reservation_lock);
4602 ei->i_reserved_data_blocks += reservation;
4603 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004604 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004605 /*
4606 * We will claim quota for all newly allocated blocks.
4607 * We're updating the reserved space *after* the
4608 * correction above so we do not accidentally free
4609 * all the metadata reservation because we might
4610 * actually need it later on.
4611 */
4612 ext4_da_update_reserve_space(inode, allocated_clusters,
4613 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004614 }
4615 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004616
4617 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004618 * Cache the extent and update transaction to commit on fdatasync only
Lukas Czerner556615d2014-04-20 23:45:47 -04004619 * when it is _not_ an unwritten extent.
Jan Karab436b9b2009-12-08 23:51:10 -05004620 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004621 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004622 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004623 else
Jan Karab436b9b2009-12-08 23:51:10 -05004624 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004625out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004626 if (allocated > map->m_len)
4627 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004628 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004629 map->m_flags |= EXT4_MAP_MAPPED;
4630 map->m_pblk = newblock;
4631 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004632out2:
4633 if (path) {
4634 ext4_ext_drop_refs(path);
4635 kfree(path);
4636 }
Allison Hendersone8613042011-05-25 07:41:46 -04004637
Theodore Ts'o63b99962013-07-16 10:28:47 -04004638 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4639 err ? err : allocated);
4640 ext4_es_lru_add(inode);
Lukas Czerner78771912012-03-19 23:05:43 -04004641 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004642}
4643
Theodore Ts'o819c4922013-04-03 12:47:17 -04004644void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004645{
Alex Tomasa86c6182006-10-11 01:21:03 -07004646 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004647 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004648 int err = 0;
4649
4650 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004651 * TODO: optimization is possible here.
4652 * Probably we need not scan at all,
4653 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004654 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004655
4656 /* we have to know where to truncate from in crash case */
4657 EXT4_I(inode)->i_disksize = inode->i_size;
4658 ext4_mark_inode_dirty(handle, inode);
4659
4660 last_block = (inode->i_size + sb->s_blocksize - 1)
4661 >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004662retry:
Zheng Liu51865fd2012-11-08 21:57:32 -05004663 err = ext4_es_remove_extent(inode, last_block,
4664 EXT_MAX_BLOCKS - last_block);
Theodore Ts'o94eec0f2013-07-29 12:12:56 -04004665 if (err == -ENOMEM) {
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004666 cond_resched();
4667 congestion_wait(BLK_RW_ASYNC, HZ/50);
4668 goto retry;
4669 }
4670 if (err) {
4671 ext4_std_error(inode->i_sb, err);
4672 return;
4673 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04004674 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Theodore Ts'o8acd5e92013-07-15 00:09:19 -04004675 ext4_std_error(inode->i_sb, err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004676}
4677
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004678static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004679 ext4_lblk_t len, loff_t new_size,
4680 int flags, int mode)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004681{
Al Viro496ad9a2013-01-23 17:07:38 -05004682 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004683 handle_t *handle;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004684 int ret = 0;
4685 int ret2 = 0;
4686 int retries = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004687 struct ext4_map_blocks map;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004688 unsigned int credits;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004689 loff_t epos;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004690
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004691 map.m_lblk = offset;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004692 map.m_len = len;
Greg Harm3c6fe772011-10-31 18:41:47 -04004693 /*
4694 * Don't normalize the request if it can fit in one extent so
4695 * that it doesn't get unnecessarily split into multiple
4696 * extents.
4697 */
Lukas Czerner556615d2014-04-20 23:45:47 -04004698 if (len <= EXT_UNWRITTEN_MAX_LEN)
Greg Harm3c6fe772011-10-31 18:41:47 -04004699 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004700
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004701 /*
4702 * credits to insert 1 extent into extent tree
4703 */
4704 credits = ext4_chunk_trans_blocks(inode, len);
4705
Amit Aroraa2df2a62007-07-17 21:42:41 -04004706retry:
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004707 while (ret >= 0 && len) {
Theodore Ts'o9924a922013-02-08 21:59:22 -05004708 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4709 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004710 if (IS_ERR(handle)) {
4711 ret = PTR_ERR(handle);
4712 break;
4713 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004714 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004715 if (ret <= 0) {
Lukas Czernerf282ac12014-03-18 17:44:35 -04004716 ext4_debug("inode #%lu: block %u: len %u: "
4717 "ext4_ext_map_blocks returned %d",
4718 inode->i_ino, map.m_lblk,
4719 map.m_len, ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004720 ext4_mark_inode_dirty(handle, inode);
4721 ret2 = ext4_journal_stop(handle);
4722 break;
4723 }
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004724 map.m_lblk += ret;
4725 map.m_len = len = len - ret;
4726 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4727 inode->i_ctime = ext4_current_time(inode);
4728 if (new_size) {
4729 if (epos > new_size)
4730 epos = new_size;
4731 if (ext4_update_inode_size(inode, epos) & 0x1)
4732 inode->i_mtime = inode->i_ctime;
4733 } else {
4734 if (epos > inode->i_size)
4735 ext4_set_inode_flag(inode,
4736 EXT4_INODE_EOFBLOCKS);
4737 }
4738 ext4_mark_inode_dirty(handle, inode);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004739 ret2 = ext4_journal_stop(handle);
4740 if (ret2)
4741 break;
4742 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004743 if (ret == -ENOSPC &&
4744 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4745 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004746 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004747 }
Lukas Czernerf282ac12014-03-18 17:44:35 -04004748
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004749 return ret > 0 ? ret2 : ret;
4750}
4751
Lukas Czernerb8a86842014-03-18 18:05:35 -04004752static long ext4_zero_range(struct file *file, loff_t offset,
4753 loff_t len, int mode)
4754{
4755 struct inode *inode = file_inode(file);
4756 handle_t *handle = NULL;
4757 unsigned int max_blocks;
4758 loff_t new_size = 0;
4759 int ret = 0;
4760 int flags;
Dmitry Monakhov69dc9532014-08-27 18:33:49 -04004761 int credits;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004762 int partial_begin, partial_end;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004763 loff_t start, end;
4764 ext4_lblk_t lblk;
4765 struct address_space *mapping = inode->i_mapping;
4766 unsigned int blkbits = inode->i_blkbits;
4767
4768 trace_ext4_zero_range(inode, offset, len, mode);
4769
jon ernst6c5e73d2014-04-18 11:50:35 -04004770 if (!S_ISREG(inode->i_mode))
4771 return -EINVAL;
4772
Namjae Jeone1ee60f2014-05-27 12:48:55 -04004773 /* Call ext4_force_commit to flush all data in case of data=journal. */
4774 if (ext4_should_journal_data(inode)) {
4775 ret = ext4_force_commit(inode->i_sb);
4776 if (ret)
4777 return ret;
4778 }
4779
Lukas Czernerb8a86842014-03-18 18:05:35 -04004780 /*
4781 * Write out all dirty pages to avoid race conditions
4782 * Then release them.
4783 */
4784 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4785 ret = filemap_write_and_wait_range(mapping, offset,
4786 offset + len - 1);
4787 if (ret)
4788 return ret;
4789 }
4790
4791 /*
4792 * Round up offset. This is not fallocate, we neet to zero out
4793 * blocks, so convert interior block aligned part of the range to
4794 * unwritten and possibly manually zero out unaligned parts of the
4795 * range.
4796 */
4797 start = round_up(offset, 1 << blkbits);
4798 end = round_down((offset + len), 1 << blkbits);
4799
4800 if (start < offset || end > offset + len)
4801 return -EINVAL;
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004802 partial_begin = offset & ((1 << blkbits) - 1);
4803 partial_end = (offset + len) & ((1 << blkbits) - 1);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004804
4805 lblk = start >> blkbits;
4806 max_blocks = (end >> blkbits);
4807 if (max_blocks < lblk)
4808 max_blocks = 0;
4809 else
4810 max_blocks -= lblk;
4811
Lukas Czerner556615d2014-04-20 23:45:47 -04004812 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
Theodore Ts'o713e8dd2014-09-01 14:32:09 -04004813 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4814 EXT4_EX_NOCACHE;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004815 if (mode & FALLOC_FL_KEEP_SIZE)
4816 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4817
4818 mutex_lock(&inode->i_mutex);
4819
4820 /*
4821 * Indirect files do not support unwritten extnets
4822 */
4823 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4824 ret = -EOPNOTSUPP;
4825 goto out_mutex;
4826 }
4827
4828 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4829 offset + len > i_size_read(inode)) {
4830 new_size = offset + len;
4831 ret = inode_newsize_ok(inode, new_size);
4832 if (ret)
4833 goto out_mutex;
4834 /*
4835 * If we have a partial block after EOF we have to allocate
4836 * the entire block.
4837 */
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004838 if (partial_end)
Lukas Czernerb8a86842014-03-18 18:05:35 -04004839 max_blocks += 1;
4840 }
4841
4842 if (max_blocks > 0) {
4843
4844 /* Now release the pages and zero block aligned part of pages*/
4845 truncate_pagecache_range(inode, start, end - 1);
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004846 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004847
4848 /* Wait all existing dio workers, newcomers will block on i_mutex */
4849 ext4_inode_block_unlocked_dio(inode);
4850 inode_dio_wait(inode);
4851
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004852 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4853 flags, mode);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004854 if (ret)
4855 goto out_dio;
Theodore Ts'o713e8dd2014-09-01 14:32:09 -04004856 /*
4857 * Remove entire range from the extent status tree.
4858 *
4859 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4860 * NOT sufficient. I'm not sure why this is the case,
4861 * but let's be conservative and remove the extent
4862 * status tree for the entire inode. There should be
4863 * no outstanding delalloc extents thanks to the
4864 * filemap_write_and_wait_range() call above.
4865 */
4866 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4867 if (ret)
4868 goto out_dio;
Lukas Czernerb8a86842014-03-18 18:05:35 -04004869 }
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004870 if (!partial_begin && !partial_end)
4871 goto out_dio;
4872
Dmitry Monakhov69dc9532014-08-27 18:33:49 -04004873 /*
4874 * In worst case we have to writeout two nonadjacent unwritten
4875 * blocks and update the inode
4876 */
4877 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4878 if (ext4_should_journal_data(inode))
4879 credits += 2;
4880 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
Lukas Czernerb8a86842014-03-18 18:05:35 -04004881 if (IS_ERR(handle)) {
4882 ret = PTR_ERR(handle);
4883 ext4_std_error(inode->i_sb, ret);
4884 goto out_dio;
4885 }
4886
4887 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
Lukas Czernere5b30412014-04-01 00:59:21 -04004888 if (new_size) {
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04004889 ext4_update_inode_size(inode, new_size);
Lukas Czernere5b30412014-04-01 00:59:21 -04004890 } else {
Lukas Czernerb8a86842014-03-18 18:05:35 -04004891 /*
4892 * Mark that we allocate beyond EOF so the subsequent truncate
4893 * can proceed even if the new size is the same as i_size.
4894 */
4895 if ((offset + len) > i_size_read(inode))
4896 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4897 }
Lukas Czernerb8a86842014-03-18 18:05:35 -04004898 ext4_mark_inode_dirty(handle, inode);
4899
4900 /* Zero out partial block at the edges of the range */
4901 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4902
4903 if (file->f_flags & O_SYNC)
4904 ext4_handle_sync(handle);
4905
4906 ext4_journal_stop(handle);
4907out_dio:
4908 ext4_inode_resume_unlocked_dio(inode);
4909out_mutex:
4910 mutex_unlock(&inode->i_mutex);
4911 return ret;
4912}
4913
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004914/*
4915 * preallocate space for a file. This implements ext4's fallocate file
4916 * operation, which gets called from sys_fallocate system call.
4917 * For block-mapped files, posix_fallocate should fall back to the method
4918 * of writing zeroes to the required new blocks (the same behavior which is
4919 * expected for file systems which do not support fallocate() system call).
4920 */
4921long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4922{
4923 struct inode *inode = file_inode(file);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004924 loff_t new_size = 0;
4925 unsigned int max_blocks;
4926 int ret = 0;
4927 int flags;
4928 ext4_lblk_t lblk;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004929 unsigned int blkbits = inode->i_blkbits;
4930
4931 /* Return error if mode is not supported */
4932 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
Lukas Czernerb8a86842014-03-18 18:05:35 -04004933 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004934 return -EOPNOTSUPP;
4935
4936 if (mode & FALLOC_FL_PUNCH_HOLE)
4937 return ext4_punch_hole(inode, offset, len);
4938
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004939 ret = ext4_convert_inline_data(inode);
4940 if (ret)
4941 return ret;
4942
4943 /*
4944 * currently supporting (pre)allocate mode for extent-based
4945 * files _only_
4946 */
4947 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4948 return -EOPNOTSUPP;
4949
Theodore Ts'o40c406c2014-04-12 22:53:53 -04004950 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4951 return ext4_collapse_range(inode, offset, len);
4952
Lukas Czernerb8a86842014-03-18 18:05:35 -04004953 if (mode & FALLOC_FL_ZERO_RANGE)
4954 return ext4_zero_range(file, offset, len, mode);
4955
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004956 trace_ext4_fallocate_enter(inode, offset, len, mode);
4957 lblk = offset >> blkbits;
4958 /*
4959 * We can't just convert len to max_blocks because
4960 * If blocksize = 4096 offset = 3072 and len = 2048
4961 */
4962 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4963 - lblk;
4964
Lukas Czerner556615d2014-04-20 23:45:47 -04004965 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004966 if (mode & FALLOC_FL_KEEP_SIZE)
4967 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4968
4969 mutex_lock(&inode->i_mutex);
4970
4971 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4972 offset + len > i_size_read(inode)) {
4973 new_size = offset + len;
4974 ret = inode_newsize_ok(inode, new_size);
4975 if (ret)
4976 goto out;
4977 }
4978
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004979 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4980 flags, mode);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004981 if (ret)
4982 goto out;
4983
Dmitry Monakhovc174e6d2014-08-27 18:40:00 -04004984 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4985 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4986 EXT4_I(inode)->i_sync_tid);
Lukas Czernerf282ac12014-03-18 17:44:35 -04004987 }
Lukas Czernerf282ac12014-03-18 17:44:35 -04004988out:
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004989 mutex_unlock(&inode->i_mutex);
Lukas Czerner0e8b6872014-03-18 18:03:51 -04004990 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4991 return ret;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004992}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004993
4994/*
Mingming Cao00314622009-09-28 15:49:08 -04004995 * This function convert a range of blocks to written extents
4996 * The caller of this function will pass the start offset and the size.
4997 * all unwritten extents within this range will be converted to
4998 * written extents.
4999 *
5000 * This function is called from the direct IO end io call back
5001 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05005002 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04005003 */
Jan Kara6b523df2013-06-04 13:21:11 -04005004int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5005 loff_t offset, ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04005006{
Mingming Cao00314622009-09-28 15:49:08 -04005007 unsigned int max_blocks;
5008 int ret = 0;
5009 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005010 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04005011 unsigned int credits, blkbits = inode->i_blkbits;
5012
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005013 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04005014 /*
5015 * We can't just convert len to max_blocks because
5016 * If blocksize = 4096 offset = 3072 and len = 2048
5017 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005018 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5019 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04005020 /*
Jan Kara6b523df2013-06-04 13:21:11 -04005021 * This is somewhat ugly but the idea is clear: When transaction is
5022 * reserved, everything goes into it. Otherwise we rather start several
5023 * smaller transactions for conversion of each extent separately.
Mingming Cao00314622009-09-28 15:49:08 -04005024 */
Jan Kara6b523df2013-06-04 13:21:11 -04005025 if (handle) {
5026 handle = ext4_journal_start_reserved(handle,
5027 EXT4_HT_EXT_CONVERT);
5028 if (IS_ERR(handle))
5029 return PTR_ERR(handle);
5030 credits = 0;
5031 } else {
5032 /*
5033 * credits to insert 1 extent into extent tree
5034 */
5035 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5036 }
Mingming Cao00314622009-09-28 15:49:08 -04005037 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005038 map.m_lblk += ret;
5039 map.m_len = (max_blocks -= ret);
Jan Kara6b523df2013-06-04 13:21:11 -04005040 if (credits) {
5041 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5042 credits);
5043 if (IS_ERR(handle)) {
5044 ret = PTR_ERR(handle);
5045 break;
5046 }
Mingming Cao00314622009-09-28 15:49:08 -04005047 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04005048 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05005049 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05005050 if (ret <= 0)
5051 ext4_warning(inode->i_sb,
5052 "inode #%lu: block %u: len %u: "
5053 "ext4_ext_map_blocks returned %d",
5054 inode->i_ino, map.m_lblk,
5055 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04005056 ext4_mark_inode_dirty(handle, inode);
Jan Kara6b523df2013-06-04 13:21:11 -04005057 if (credits)
5058 ret2 = ext4_journal_stop(handle);
5059 if (ret <= 0 || ret2)
Mingming Cao00314622009-09-28 15:49:08 -04005060 break;
5061 }
Jan Kara6b523df2013-06-04 13:21:11 -04005062 if (!credits)
5063 ret2 = ext4_journal_stop(handle);
Mingming Cao00314622009-09-28 15:49:08 -04005064 return ret > 0 ? ret2 : ret;
5065}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005066
Mingming Cao00314622009-09-28 15:49:08 -04005067/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05005068 * If newes is not existing extent (newes->ec_pblk equals zero) find
5069 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005070 * return start of the next delayed extent.
5071 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05005072 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005073 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05005074 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04005075 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005076static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05005077 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04005078{
Zheng Liub3aff3e2012-11-08 21:57:37 -05005079 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05005080 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005081
Zheng Liu69eb33d2013-02-18 00:31:07 -05005082 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04005083 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5084 newes->es_lblk + newes->es_len - 1, &es);
5085
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005086 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05005087 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005088 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005089 */
Zheng Liu06b0c882013-02-18 00:26:51 -05005090 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05005091 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005092 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005093
Zheng Liu69eb33d2013-02-18 00:31:07 -05005094 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05005095 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05005096 newes->es_len = min(es.es_lblk - newes->es_lblk,
5097 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005098 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005099 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05005100
Zheng Liu69eb33d2013-02-18 00:31:07 -05005101 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005102 }
5103
Zheng Liu69eb33d2013-02-18 00:31:07 -05005104 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04005105 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05005106 if (es.es_len == 0)
5107 next_del = EXT_MAX_BLOCKS;
5108 else
5109 next_del = es.es_lblk;
5110
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005111 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005112}
Eric Sandeen6873fa02008-10-07 00:46:36 -04005113/* fiemap flags we can handle specified here */
5114#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5115
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05005116static int ext4_xattr_fiemap(struct inode *inode,
5117 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04005118{
5119 __u64 physical = 0;
5120 __u64 length;
5121 __u32 flags = FIEMAP_EXTENT_LAST;
5122 int blockbits = inode->i_sb->s_blocksize_bits;
5123 int error = 0;
5124
5125 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005126 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04005127 struct ext4_iloc iloc;
5128 int offset; /* offset of xattr in inode */
5129
5130 error = ext4_get_inode_loc(inode, &iloc);
5131 if (error)
5132 return error;
Jan Karaa60697f2013-05-31 19:38:56 -04005133 physical = (__u64)iloc.bh->b_blocknr << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005134 offset = EXT4_GOOD_OLD_INODE_SIZE +
5135 EXT4_I(inode)->i_extra_isize;
5136 physical += offset;
5137 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5138 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04005139 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005140 } else { /* external block */
Jan Karaa60697f2013-05-31 19:38:56 -04005141 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005142 length = inode->i_sb->s_blocksize;
5143 }
5144
5145 if (physical)
5146 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5147 length, flags);
5148 return (error < 0 ? error : 0);
5149}
5150
5151int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5152 __u64 start, __u64 len)
5153{
5154 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005155 int error = 0;
5156
Tao Ma94191982012-12-10 14:06:02 -05005157 if (ext4_has_inline_data(inode)) {
5158 int has_inline = 1;
5159
5160 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5161
5162 if (has_inline)
5163 return error;
5164 }
5165
Theodore Ts'o7869a4a2013-08-16 22:05:14 -04005166 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5167 error = ext4_ext_precache(inode);
5168 if (error)
5169 return error;
5170 }
5171
Eric Sandeen6873fa02008-10-07 00:46:36 -04005172 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005173 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04005174 return generic_block_fiemap(inode, fieinfo, start, len,
5175 ext4_get_block);
5176
5177 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5178 return -EBADR;
5179
5180 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5181 error = ext4_xattr_fiemap(inode, fieinfo);
5182 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005183 ext4_lblk_t len_blks;
5184 __u64 last_blk;
5185
Eric Sandeen6873fa02008-10-07 00:46:36 -04005186 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005187 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04005188 if (last_blk >= EXT_MAX_BLOCKS)
5189 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05005190 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04005191
5192 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005193 * Walk the extent tree gathering extent information
5194 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04005195 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05005196 error = ext4_fill_fiemap_extents(inode, start_blk,
5197 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005198 }
Theodore Ts'o107a7bd2013-08-16 21:23:41 -04005199 ext4_es_lru_add(inode);
Eric Sandeen6873fa02008-10-07 00:46:36 -04005200 return error;
5201}
Namjae Jeon9eb79482014-02-23 15:18:59 -05005202
5203/*
5204 * ext4_access_path:
5205 * Function to access the path buffer for marking it dirty.
5206 * It also checks if there are sufficient credits left in the journal handle
5207 * to update path.
5208 */
5209static int
5210ext4_access_path(handle_t *handle, struct inode *inode,
5211 struct ext4_ext_path *path)
5212{
5213 int credits, err;
5214
5215 if (!ext4_handle_valid(handle))
5216 return 0;
5217
5218 /*
5219 * Check if need to extend journal credits
5220 * 3 for leaf, sb, and inode plus 2 (bmap and group
5221 * descriptor) for each block group; assume two block
5222 * groups
5223 */
5224 if (handle->h_buffer_credits < 7) {
5225 credits = ext4_writepage_trans_blocks(inode);
5226 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5227 /* EAGAIN is success */
5228 if (err && err != -EAGAIN)
5229 return err;
5230 }
5231
5232 err = ext4_ext_get_access(handle, inode, path);
5233 return err;
5234}
5235
5236/*
5237 * ext4_ext_shift_path_extents:
5238 * Shift the extents of a path structure lying between path[depth].p_ext
5239 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5240 * from starting block for each extent.
5241 */
5242static int
5243ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5244 struct inode *inode, handle_t *handle,
5245 ext4_lblk_t *start)
5246{
5247 int depth, err = 0;
5248 struct ext4_extent *ex_start, *ex_last;
5249 bool update = 0;
5250 depth = path->p_depth;
5251
5252 while (depth >= 0) {
5253 if (depth == path->p_depth) {
5254 ex_start = path[depth].p_ext;
5255 if (!ex_start)
5256 return -EIO;
5257
5258 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5259 if (!ex_last)
5260 return -EIO;
5261
5262 err = ext4_access_path(handle, inode, path + depth);
5263 if (err)
5264 goto out;
5265
5266 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5267 update = 1;
5268
Zheng Liu847c6c42014-04-12 12:45:55 -04005269 *start = le32_to_cpu(ex_last->ee_block) +
Namjae Jeon9eb79482014-02-23 15:18:59 -05005270 ext4_ext_get_actual_len(ex_last);
5271
5272 while (ex_start <= ex_last) {
Zheng Liu847c6c42014-04-12 12:45:55 -04005273 le32_add_cpu(&ex_start->ee_block, -shift);
Lukas Czerner6dd834e2014-04-18 10:55:24 -04005274 /* Try to merge to the left. */
5275 if ((ex_start >
5276 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5277 ext4_ext_try_to_merge_right(inode,
5278 path, ex_start - 1))
5279 ex_last--;
5280 else
5281 ex_start++;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005282 }
5283 err = ext4_ext_dirty(handle, inode, path + depth);
5284 if (err)
5285 goto out;
5286
5287 if (--depth < 0 || !update)
5288 break;
5289 }
5290
5291 /* Update index too */
5292 err = ext4_access_path(handle, inode, path + depth);
5293 if (err)
5294 goto out;
5295
Zheng Liu847c6c42014-04-12 12:45:55 -04005296 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005297 err = ext4_ext_dirty(handle, inode, path + depth);
5298 if (err)
5299 goto out;
5300
5301 /* we are done if current index is not a starting index */
5302 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5303 break;
5304
5305 depth--;
5306 }
5307
5308out:
5309 return err;
5310}
5311
5312/*
5313 * ext4_ext_shift_extents:
5314 * All the extents which lies in the range from start to the last allocated
5315 * block for the file are shifted downwards by shift blocks.
5316 * On success, 0 is returned, error otherwise.
5317 */
5318static int
5319ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5320 ext4_lblk_t start, ext4_lblk_t shift)
5321{
5322 struct ext4_ext_path *path;
5323 int ret = 0, depth;
5324 struct ext4_extent *extent;
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005325 ext4_lblk_t stop_block;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005326 ext4_lblk_t ex_start, ex_end;
5327
5328 /* Let path point to the last extent */
5329 path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5330 if (IS_ERR(path))
5331 return PTR_ERR(path);
5332
5333 depth = path->p_depth;
5334 extent = path[depth].p_ext;
5335 if (!extent) {
5336 ext4_ext_drop_refs(path);
5337 kfree(path);
5338 return ret;
5339 }
5340
Zheng Liu847c6c42014-04-12 12:45:55 -04005341 stop_block = le32_to_cpu(extent->ee_block) +
5342 ext4_ext_get_actual_len(extent);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005343 ext4_ext_drop_refs(path);
5344 kfree(path);
5345
5346 /* Nothing to shift, if hole is at the end of file */
5347 if (start >= stop_block)
5348 return ret;
5349
5350 /*
5351 * Don't start shifting extents until we make sure the hole is big
5352 * enough to accomodate the shift.
5353 */
5354 path = ext4_ext_find_extent(inode, start - 1, NULL, 0);
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005355 if (IS_ERR(path))
5356 return PTR_ERR(path);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005357 depth = path->p_depth;
5358 extent = path[depth].p_ext;
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005359 if (extent) {
5360 ex_start = le32_to_cpu(extent->ee_block);
5361 ex_end = le32_to_cpu(extent->ee_block) +
Zheng Liu847c6c42014-04-12 12:45:55 -04005362 ext4_ext_get_actual_len(extent);
Dmitry Monakhov8dc79ec2014-04-13 15:05:42 -04005363 } else {
5364 ex_start = 0;
5365 ex_end = 0;
5366 }
Namjae Jeon9eb79482014-02-23 15:18:59 -05005367 ext4_ext_drop_refs(path);
5368 kfree(path);
5369
5370 if ((start == ex_start && shift > ex_start) ||
5371 (shift > start - ex_end))
5372 return -EINVAL;
5373
5374 /* Its safe to start updating extents */
5375 while (start < stop_block) {
5376 path = ext4_ext_find_extent(inode, start, NULL, 0);
5377 if (IS_ERR(path))
5378 return PTR_ERR(path);
5379 depth = path->p_depth;
5380 extent = path[depth].p_ext;
Dmitry Monakhova18ed352014-04-13 15:41:13 -04005381 if (!extent) {
5382 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5383 (unsigned long) start);
5384 return -EIO;
5385 }
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005386 if (start > le32_to_cpu(extent->ee_block)) {
Namjae Jeon9eb79482014-02-23 15:18:59 -05005387 /* Hole, move to the next extent */
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005388 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5389 path[depth].p_ext++;
5390 } else {
5391 start = ext4_ext_next_allocated_block(path);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005392 ext4_ext_drop_refs(path);
5393 kfree(path);
Dmitry Monakhovf8fb4f42014-08-30 23:50:56 -04005394 continue;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005395 }
5396 }
5397 ret = ext4_ext_shift_path_extents(path, shift, inode,
5398 handle, &start);
5399 ext4_ext_drop_refs(path);
5400 kfree(path);
5401 if (ret)
5402 break;
5403 }
5404
5405 return ret;
5406}
5407
5408/*
5409 * ext4_collapse_range:
5410 * This implements the fallocate's collapse range functionality for ext4
5411 * Returns: 0 and non-zero on error.
5412 */
5413int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5414{
5415 struct super_block *sb = inode->i_sb;
5416 ext4_lblk_t punch_start, punch_stop;
5417 handle_t *handle;
5418 unsigned int credits;
Namjae Jeona8680e02014-04-19 16:37:31 -04005419 loff_t new_size, ioffset;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005420 int ret;
5421
Namjae Jeon9eb79482014-02-23 15:18:59 -05005422 /* Collapse range works only on fs block size aligned offsets. */
Namjae Jeonee98fa32014-07-29 10:45:31 -04005423 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5424 len & (EXT4_CLUSTER_SIZE(sb) - 1))
Namjae Jeon9eb79482014-02-23 15:18:59 -05005425 return -EINVAL;
5426
5427 if (!S_ISREG(inode->i_mode))
Theodore Ts'o86f1ca32014-04-18 11:52:11 -04005428 return -EINVAL;
Namjae Jeon9eb79482014-02-23 15:18:59 -05005429
5430 trace_ext4_collapse_range(inode, offset, len);
5431
5432 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5433 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5434
Namjae Jeon1ce01c42014-04-10 22:58:20 -04005435 /* Call ext4_force_commit to flush all data in case of data=journal. */
5436 if (ext4_should_journal_data(inode)) {
5437 ret = ext4_force_commit(inode->i_sb);
5438 if (ret)
5439 return ret;
5440 }
5441
Namjae Jeona8680e02014-04-19 16:37:31 -04005442 /*
5443 * Need to round down offset to be aligned with page size boundary
5444 * for page size > block size.
5445 */
5446 ioffset = round_down(offset, PAGE_SIZE);
5447
Namjae Jeon9eb79482014-02-23 15:18:59 -05005448 /* Write out all dirty pages */
Namjae Jeona8680e02014-04-19 16:37:31 -04005449 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5450 LLONG_MAX);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005451 if (ret)
5452 return ret;
5453
5454 /* Take mutex lock */
5455 mutex_lock(&inode->i_mutex);
5456
Lukas Czerner23fffa92014-04-12 09:56:41 -04005457 /*
5458 * There is no need to overlap collapse range with EOF, in which case
5459 * it is effectively a truncate operation
5460 */
5461 if (offset + len >= i_size_read(inode)) {
5462 ret = -EINVAL;
5463 goto out_mutex;
5464 }
5465
Namjae Jeon9eb79482014-02-23 15:18:59 -05005466 /* Currently just for extent based files */
5467 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5468 ret = -EOPNOTSUPP;
5469 goto out_mutex;
5470 }
5471
Namjae Jeona8680e02014-04-19 16:37:31 -04005472 truncate_pagecache(inode, ioffset);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005473
5474 /* Wait for existing dio to complete */
5475 ext4_inode_block_unlocked_dio(inode);
5476 inode_dio_wait(inode);
5477
5478 credits = ext4_writepage_trans_blocks(inode);
5479 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5480 if (IS_ERR(handle)) {
5481 ret = PTR_ERR(handle);
5482 goto out_dio;
5483 }
5484
5485 down_write(&EXT4_I(inode)->i_data_sem);
5486 ext4_discard_preallocations(inode);
5487
5488 ret = ext4_es_remove_extent(inode, punch_start,
Lukas Czerner2c1d2322014-04-18 10:43:21 -04005489 EXT_MAX_BLOCKS - punch_start);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005490 if (ret) {
5491 up_write(&EXT4_I(inode)->i_data_sem);
5492 goto out_stop;
5493 }
5494
5495 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5496 if (ret) {
5497 up_write(&EXT4_I(inode)->i_data_sem);
5498 goto out_stop;
5499 }
Lukas Czerneref24f6c2014-04-18 10:50:23 -04005500 ext4_discard_preallocations(inode);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005501
5502 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5503 punch_stop - punch_start);
5504 if (ret) {
5505 up_write(&EXT4_I(inode)->i_data_sem);
5506 goto out_stop;
5507 }
5508
5509 new_size = i_size_read(inode) - len;
Lukas Czerner9337d5d2014-04-18 10:48:25 -04005510 i_size_write(inode, new_size);
Namjae Jeon9eb79482014-02-23 15:18:59 -05005511 EXT4_I(inode)->i_disksize = new_size;
5512
Namjae Jeon9eb79482014-02-23 15:18:59 -05005513 up_write(&EXT4_I(inode)->i_data_sem);
5514 if (IS_SYNC(inode))
5515 ext4_handle_sync(handle);
5516 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5517 ext4_mark_inode_dirty(handle, inode);
5518
5519out_stop:
5520 ext4_journal_stop(handle);
5521out_dio:
5522 ext4_inode_resume_unlocked_dio(inode);
5523out_mutex:
5524 mutex_unlock(&inode->i_mutex);
5525 return ret;
5526}
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005527
5528/**
5529 * ext4_swap_extents - Swap extents between two inodes
5530 *
5531 * @inode1: First inode
5532 * @inode2: Second inode
5533 * @lblk1: Start block for first inode
5534 * @lblk2: Start block for second inode
5535 * @count: Number of blocks to swap
5536 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5537 * @erp: Pointer to save error value
5538 *
5539 * This helper routine does exactly what is promise "swap extents". All other
5540 * stuff such as page-cache locking consistency, bh mapping consistency or
5541 * extent's data copying must be performed by caller.
5542 * Locking:
5543 * i_mutex is held for both inodes
5544 * i_data_sem is locked for write for both inodes
5545 * Assumptions:
5546 * All pages from requested range are locked for both inodes
5547 */
5548int
5549ext4_swap_extents(handle_t *handle, struct inode *inode1,
5550 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5551 ext4_lblk_t count, int unwritten, int *erp)
5552{
5553 struct ext4_ext_path *path1 = NULL;
5554 struct ext4_ext_path *path2 = NULL;
5555 int replaced_count = 0;
5556
5557 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5558 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5559 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5560 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5561
5562 *erp = ext4_es_remove_extent(inode1, lblk1, count);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005563 if (unlikely(*erp))
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005564 return 0;
5565 *erp = ext4_es_remove_extent(inode2, lblk2, count);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005566 if (unlikely(*erp))
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005567 return 0;
5568
5569 while (count) {
5570 struct ext4_extent *ex1, *ex2, tmp_ex;
5571 ext4_lblk_t e1_blk, e2_blk;
5572 int e1_len, e2_len, len;
5573 int split = 0;
5574
5575 path1 = ext4_ext_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005576 if (unlikely(IS_ERR(path1))) {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005577 *erp = PTR_ERR(path1);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005578 path1 = NULL;
5579 finish:
5580 count = 0;
5581 goto repeat;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005582 }
5583 path2 = ext4_ext_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005584 if (unlikely(IS_ERR(path2))) {
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005585 *erp = PTR_ERR(path2);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005586 path2 = NULL;
5587 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005588 }
5589 ex1 = path1[path1->p_depth].p_ext;
5590 ex2 = path2[path2->p_depth].p_ext;
5591 /* Do we have somthing to swap ? */
5592 if (unlikely(!ex2 || !ex1))
Theodore Ts'o19008f62014-08-31 15:03:14 -04005593 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005594
5595 e1_blk = le32_to_cpu(ex1->ee_block);
5596 e2_blk = le32_to_cpu(ex2->ee_block);
5597 e1_len = ext4_ext_get_actual_len(ex1);
5598 e2_len = ext4_ext_get_actual_len(ex2);
5599
5600 /* Hole handling */
5601 if (!in_range(lblk1, e1_blk, e1_len) ||
5602 !in_range(lblk2, e2_blk, e2_len)) {
5603 ext4_lblk_t next1, next2;
5604
5605 /* if hole after extent, then go to next extent */
5606 next1 = ext4_ext_next_allocated_block(path1);
5607 next2 = ext4_ext_next_allocated_block(path2);
5608 /* If hole before extent, then shift to that extent */
5609 if (e1_blk > lblk1)
5610 next1 = e1_blk;
5611 if (e2_blk > lblk2)
5612 next2 = e1_blk;
5613 /* Do we have something to swap */
5614 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
Theodore Ts'o19008f62014-08-31 15:03:14 -04005615 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005616 /* Move to the rightest boundary */
5617 len = next1 - lblk1;
5618 if (len < next2 - lblk2)
5619 len = next2 - lblk2;
5620 if (len > count)
5621 len = count;
5622 lblk1 += len;
5623 lblk2 += len;
5624 count -= len;
5625 goto repeat;
5626 }
5627
5628 /* Prepare left boundary */
5629 if (e1_blk < lblk1) {
5630 split = 1;
5631 *erp = ext4_force_split_extent_at(handle, inode1,
5632 path1, lblk1, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005633 if (unlikely(*erp))
5634 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005635 }
5636 if (e2_blk < lblk2) {
5637 split = 1;
5638 *erp = ext4_force_split_extent_at(handle, inode2,
5639 path2, lblk2, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005640 if (unlikely(*erp))
5641 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005642 }
5643 /* ext4_split_extent_at() may retult in leaf extent split,
5644 * path must to be revalidated. */
5645 if (split)
5646 goto repeat;
5647
5648 /* Prepare right boundary */
5649 len = count;
5650 if (len > e1_blk + e1_len - lblk1)
5651 len = e1_blk + e1_len - lblk1;
5652 if (len > e2_blk + e2_len - lblk2)
5653 len = e2_blk + e2_len - lblk2;
5654
5655 if (len != e1_len) {
5656 split = 1;
5657 *erp = ext4_force_split_extent_at(handle, inode1,
5658 path1, lblk1 + len, 0);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005659 if (unlikely(*erp))
5660 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005661 }
5662 if (len != e2_len) {
5663 split = 1;
5664 *erp = ext4_force_split_extent_at(handle, inode2,
5665 path2, lblk2 + len, 0);
5666 if (*erp)
Theodore Ts'o19008f62014-08-31 15:03:14 -04005667 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005668 }
5669 /* ext4_split_extent_at() may retult in leaf extent split,
5670 * path must to be revalidated. */
5671 if (split)
5672 goto repeat;
5673
5674 BUG_ON(e2_len != e1_len);
5675 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005676 if (unlikely(*erp))
5677 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005678 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005679 if (unlikely(*erp))
5680 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005681
5682 /* Both extents are fully inside boundaries. Swap it now */
5683 tmp_ex = *ex1;
5684 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5685 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5686 ex1->ee_len = cpu_to_le16(e2_len);
5687 ex2->ee_len = cpu_to_le16(e1_len);
5688 if (unwritten)
5689 ext4_ext_mark_unwritten(ex2);
5690 if (ext4_ext_is_unwritten(&tmp_ex))
5691 ext4_ext_mark_unwritten(ex1);
5692
5693 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5694 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5695 *erp = ext4_ext_dirty(handle, inode2, path2 +
5696 path2->p_depth);
Theodore Ts'o19008f62014-08-31 15:03:14 -04005697 if (unlikely(*erp))
5698 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005699 *erp = ext4_ext_dirty(handle, inode1, path1 +
5700 path1->p_depth);
5701 /*
5702 * Looks scarry ah..? second inode already points to new blocks,
5703 * and it was successfully dirtied. But luckily error may happen
5704 * only due to journal error, so full transaction will be
5705 * aborted anyway.
5706 */
Theodore Ts'o19008f62014-08-31 15:03:14 -04005707 if (unlikely(*erp))
5708 goto finish;
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005709 lblk1 += len;
5710 lblk2 += len;
5711 replaced_count += len;
5712 count -= len;
5713
5714 repeat:
5715 if (path1) {
5716 ext4_ext_drop_refs(path1);
5717 kfree(path1);
5718 path1 = NULL;
5719 }
5720 if (path2) {
5721 ext4_ext_drop_refs(path2);
5722 kfree(path2);
5723 path2 = NULL;
5724 }
5725 }
Dmitry Monakhovfcf6b1b72014-08-30 23:52:19 -04005726 return replaced_count;
5727}