blob: 84d817b842a8f0cf01525f7508b4e078472bed03 [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>
Amit Aroraa2df2a62007-07-17 21:42:41 -040040#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070041#include <asm/uaccess.h>
Eric Sandeen6873fa02008-10-07 00:46:36 -040042#include <linux/fiemap.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040043#include "ext4_jbd2.h"
Theodore Ts'o4a092d72012-11-28 13:03:30 -050044#include "ext4_extents.h"
Tao Maf19d5872012-12-10 14:05:51 -050045#include "xattr.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070046
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040047#include <trace/events/ext4.h>
48
Lukas Czerner5f95d212012-03-19 23:03:19 -040049/*
50 * used by extent splitting.
51 */
52#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 due to ENOSPC */
54#define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
55#define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
56
Dmitry Monakhovdee1f972012-10-10 01:04:58 -040057#define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58#define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59
Darrick J. Wong7ac59902012-04-29 18:37:10 -040060static __le32 ext4_extent_block_csum(struct inode *inode,
61 struct ext4_extent_header *eh)
62{
63 struct ext4_inode_info *ei = EXT4_I(inode);
64 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
65 __u32 csum;
66
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
68 EXT4_EXTENT_TAIL_OFFSET(eh));
69 return cpu_to_le32(csum);
70}
71
72static int ext4_extent_block_csum_verify(struct inode *inode,
73 struct ext4_extent_header *eh)
74{
75 struct ext4_extent_tail *et;
76
77 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
78 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
79 return 1;
80
81 et = find_ext4_extent_tail(eh);
82 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
83 return 0;
84 return 1;
85}
86
87static void ext4_extent_block_csum_set(struct inode *inode,
88 struct ext4_extent_header *eh)
89{
90 struct ext4_extent_tail *et;
91
92 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
93 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
94 return;
95
96 et = find_ext4_extent_tail(eh);
97 et->et_checksum = ext4_extent_block_csum(inode, eh);
98}
99
Allison Hendersond583fb82011-05-25 07:41:43 -0400100static int ext4_split_extent(handle_t *handle,
101 struct inode *inode,
102 struct ext4_ext_path *path,
103 struct ext4_map_blocks *map,
104 int split_flag,
105 int flags);
106
Lukas Czerner5f95d212012-03-19 23:03:19 -0400107static int ext4_split_extent_at(handle_t *handle,
108 struct inode *inode,
109 struct ext4_ext_path *path,
110 ext4_lblk_t split,
111 int split_flag,
112 int flags);
113
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500114static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -0500115 struct extent_status *newes);
Lukas Czerner91dd8c12012-11-28 12:32:26 -0500116
Jan Kara487caee2009-08-17 22:17:20 -0400117static int ext4_ext_truncate_extend_restart(handle_t *handle,
118 struct inode *inode,
119 int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -0700120{
121 int err;
122
Frank Mayhar03901312009-01-07 00:06:22 -0500123 if (!ext4_handle_valid(handle))
124 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700125 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -0400126 return 0;
127 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -0400128 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -0400129 return err;
Jan Kara487caee2009-08-17 22:17:20 -0400130 err = ext4_truncate_restart_trans(handle, inode, needed);
Dmitry Monakhov0617b832010-05-17 01:00:00 -0400131 if (err == 0)
132 err = -EAGAIN;
Jan Kara487caee2009-08-17 22:17:20 -0400133
134 return err;
Alex Tomasa86c6182006-10-11 01:21:03 -0700135}
136
137/*
138 * could return:
139 * - EROFS
140 * - ENOMEM
141 */
142static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
143 struct ext4_ext_path *path)
144{
145 if (path->p_bh) {
146 /* path points to block */
147 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;
164 if (path->p_bh) {
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400165 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
Alex Tomasa86c6182006-10-11 01:21:03 -0700166 /* path points to block */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400167 err = __ext4_handle_dirty_metadata(where, line, handle,
168 inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700169 } else {
170 /* path points to leaf/index in inode body */
171 err = ext4_mark_inode_dirty(handle, inode);
172 }
173 return err;
174}
175
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700176static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700177 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500178 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700179{
Alex Tomasa86c6182006-10-11 01:21:03 -0700180 if (path) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400181 int depth = path->p_depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700182 struct ext4_extent *ex;
Alex Tomasa86c6182006-10-11 01:21:03 -0700183
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500184 /*
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
Tao Mab8d65682011-01-21 23:21:31 +0800198 * especially if the latter case turns out to be
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500199 * common.
200 */
Avantika Mathur7e028972006-12-06 20:41:33 -0800201 ex = path[depth].p_ext;
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500202 if (ex) {
203 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
204 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
205
206 if (block > ext_block)
207 return ext_pblk + (block - ext_block);
208 else
209 return ext_pblk - (ext_block - block);
210 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700211
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700212 /* it looks like index is empty;
213 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700214 if (path[depth].p_bh)
215 return path[depth].p_bh->b_blocknr;
216 }
217
218 /* OK. use inode's group */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400219 return ext4_inode_to_goal_block(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700220}
221
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400222/*
223 * Allocation for a meta data block
224 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700225static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400226ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700227 struct ext4_ext_path *path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400228 struct ext4_extent *ex, int *err, unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700229{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700230 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700231
232 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Allison Henderson55f020d2011-05-25 07:41:26 -0400233 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
234 NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700235 return newblock;
236}
237
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400238static inline int ext4_ext_space_block(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700239{
240 int size;
241
242 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100244#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400245 if (!check && size > 6)
246 size = 6;
Alex Tomasa86c6182006-10-11 01:21:03 -0700247#endif
248 return size;
249}
250
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400251static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700252{
253 int size;
254
255 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
256 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100257#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400258 if (!check && size > 5)
259 size = 5;
Alex Tomasa86c6182006-10-11 01:21:03 -0700260#endif
261 return size;
262}
263
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400264static inline int ext4_ext_space_root(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700265{
266 int size;
267
268 size = sizeof(EXT4_I(inode)->i_data);
269 size -= sizeof(struct ext4_extent_header);
270 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100271#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400272 if (!check && size > 3)
273 size = 3;
Alex Tomasa86c6182006-10-11 01:21:03 -0700274#endif
275 return size;
276}
277
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400278static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700279{
280 int size;
281
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100285#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400286 if (!check && size > 4)
287 size = 4;
Alex Tomasa86c6182006-10-11 01:21:03 -0700288#endif
289 return size;
290}
291
Mingming Caod2a17632008-07-14 17:52:37 -0400292/*
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
296 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500297int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400298{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500299 struct ext4_inode_info *ei = EXT4_I(inode);
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400300 int idxs;
Mingming Caod2a17632008-07-14 17:52:37 -0400301
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500302 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
303 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400304
305 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400312 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500313 if (ei->i_da_metadata_calc_len &&
314 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400315 int num = 0;
316
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500317 if ((ei->i_da_metadata_calc_len % idxs) == 0)
318 num++;
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
320 num++;
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
322 num++;
323 ei->i_da_metadata_calc_len = 0;
324 } else
325 ei->i_da_metadata_calc_len++;
326 ei->i_da_metadata_calc_last_lblock++;
327 return num;
328 }
Mingming Caod2a17632008-07-14 17:52:37 -0400329
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500330 /*
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
333 */
334 ei->i_da_metadata_calc_len = 1;
335 ei->i_da_metadata_calc_last_lblock = lblock;
336 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400337}
338
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400339static int
340ext4_ext_max_entries(struct inode *inode, int depth)
341{
342 int max;
343
344 if (depth == ext_depth(inode)) {
345 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400346 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400347 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400348 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400349 } else {
350 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400351 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400352 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400353 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400354 }
355
356 return max;
357}
358
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400359static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
360{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400361 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400362 int len = ext4_ext_get_actual_len(ext);
Eryu Guanae21dda2013-12-03 21:22:21 -0500363 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
364 ext4_lblk_t last = lblock + len - 1;
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400365
Eryu Guanae21dda2013-12-03 21:22:21 -0500366 if (lblock > last)
Theodore Ts'o31d4f3a2012-03-11 23:30:16 -0400367 return 0;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400368 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400369}
370
371static int ext4_valid_extent_idx(struct inode *inode,
372 struct ext4_extent_idx *ext_idx)
373{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400374 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400375
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400376 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400377}
378
379static int ext4_valid_extent_entries(struct inode *inode,
380 struct ext4_extent_header *eh,
381 int depth)
382{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400383 unsigned short entries;
384 if (eh->eh_entries == 0)
385 return 1;
386
387 entries = le16_to_cpu(eh->eh_entries);
388
389 if (depth == 0) {
390 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400391 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Eryu Guanae21dda2013-12-03 21:22:21 -0500392 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
393 ext4_fsblk_t pblock = 0;
394 ext4_lblk_t lblock = 0;
395 ext4_lblk_t prev = 0;
396 int len = 0;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400397 while (entries) {
398 if (!ext4_valid_extent(inode, ext))
399 return 0;
Eryu Guanae21dda2013-12-03 21:22:21 -0500400
401 /* Check for overlapping extents */
402 lblock = le32_to_cpu(ext->ee_block);
403 len = ext4_ext_get_actual_len(ext);
404 if ((lblock <= prev) && prev) {
405 pblock = ext4_ext_pblock(ext);
406 es->s_last_error_block = cpu_to_le64(pblock);
407 return 0;
408 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400409 ext++;
410 entries--;
Eryu Guanae21dda2013-12-03 21:22:21 -0500411 prev = lblock + len - 1;
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400412 }
413 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400414 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400415 while (entries) {
416 if (!ext4_valid_extent_idx(inode, ext_idx))
417 return 0;
418 ext_idx++;
419 entries--;
420 }
421 }
422 return 1;
423}
424
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400425static int __ext4_ext_check(const char *function, unsigned int line,
426 struct inode *inode, struct ext4_extent_header *eh,
427 int depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400428{
429 const char *error_msg;
430 int max = 0;
431
432 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
433 error_msg = "invalid magic";
434 goto corrupted;
435 }
436 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
437 error_msg = "unexpected eh_depth";
438 goto corrupted;
439 }
440 if (unlikely(eh->eh_max == 0)) {
441 error_msg = "invalid eh_max";
442 goto corrupted;
443 }
444 max = ext4_ext_max_entries(inode, depth);
445 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
446 error_msg = "too large eh_max";
447 goto corrupted;
448 }
449 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
450 error_msg = "invalid eh_entries";
451 goto corrupted;
452 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400453 if (!ext4_valid_extent_entries(inode, eh, depth)) {
454 error_msg = "invalid extent entries";
455 goto corrupted;
456 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400457 /* Verify checksum on non-root extent tree nodes */
458 if (ext_depth(inode) != depth &&
459 !ext4_extent_block_csum_verify(inode, eh)) {
460 error_msg = "extent tree corrupted";
461 goto corrupted;
462 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400463 return 0;
464
465corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400466 ext4_error_inode(inode, function, line, 0,
Theodore Ts'o24676da2010-05-16 21:00:00 -0400467 "bad header/extent: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400468 "entries %u, max %u(%u), depth %u(%u)",
Theodore Ts'o24676da2010-05-16 21:00:00 -0400469 error_msg, le16_to_cpu(eh->eh_magic),
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400470 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
471 max, le16_to_cpu(eh->eh_depth), depth);
472
473 return -EIO;
474}
475
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400476#define ext4_ext_check(inode, eh, depth) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400477 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400478
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400479int ext4_ext_check_inode(struct inode *inode)
480{
481 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
482}
483
Darrick J. Wongf8489122012-04-29 18:21:10 -0400484static int __ext4_ext_check_block(const char *function, unsigned int line,
485 struct inode *inode,
486 struct ext4_extent_header *eh,
487 int depth,
488 struct buffer_head *bh)
489{
490 int ret;
491
492 if (buffer_verified(bh))
493 return 0;
494 ret = ext4_ext_check(inode, eh, depth);
495 if (ret)
496 return ret;
497 set_buffer_verified(bh);
498 return ret;
499}
500
501#define ext4_ext_check_block(inode, eh, depth, bh) \
502 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
503
Alex Tomasa86c6182006-10-11 01:21:03 -0700504#ifdef EXT_DEBUG
505static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
506{
507 int k, l = path->p_depth;
508
509 ext_debug("path:");
510 for (k = 0; k <= l; k++, path++) {
511 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700512 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400513 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700514 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400515 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700516 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400517 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400518 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400519 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700520 } else
521 ext_debug(" []");
522 }
523 ext_debug("\n");
524}
525
526static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
527{
528 int depth = ext_depth(inode);
529 struct ext4_extent_header *eh;
530 struct ext4_extent *ex;
531 int i;
532
533 if (!path)
534 return;
535
536 eh = path[depth].p_hdr;
537 ex = EXT_FIRST_EXTENT(eh);
538
Mingming553f9002009-09-18 13:34:55 -0400539 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
540
Alex Tomasa86c6182006-10-11 01:21:03 -0700541 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400542 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
543 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400544 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700545 }
546 ext_debug("\n");
547}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400548
549static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
550 ext4_fsblk_t newblock, int level)
551{
552 int depth = ext_depth(inode);
553 struct ext4_extent *ex;
554
555 if (depth != level) {
556 struct ext4_extent_idx *idx;
557 idx = path[level].p_idx;
558 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
559 ext_debug("%d: move %d:%llu in new index %llu\n", level,
560 le32_to_cpu(idx->ei_block),
561 ext4_idx_pblock(idx),
562 newblock);
563 idx++;
564 }
565
566 return;
567 }
568
569 ex = path[depth].p_ext;
570 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
571 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
572 le32_to_cpu(ex->ee_block),
573 ext4_ext_pblock(ex),
574 ext4_ext_is_uninitialized(ex),
575 ext4_ext_get_actual_len(ex),
576 newblock);
577 ex++;
578 }
579}
580
Alex Tomasa86c6182006-10-11 01:21:03 -0700581#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400582#define ext4_ext_show_path(inode, path)
583#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400584#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700585#endif
586
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500587void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700588{
589 int depth = path->p_depth;
590 int i;
591
592 for (i = 0; i <= depth; i++, path++)
593 if (path->p_bh) {
594 brelse(path->p_bh);
595 path->p_bh = NULL;
596 }
597}
598
599/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700600 * ext4_ext_binsearch_idx:
601 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400602 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700603 */
604static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500605ext4_ext_binsearch_idx(struct inode *inode,
606 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700607{
608 struct ext4_extent_header *eh = path->p_hdr;
609 struct ext4_extent_idx *r, *l, *m;
610
Alex Tomasa86c6182006-10-11 01:21:03 -0700611
Eric Sandeenbba90742008-01-28 23:58:27 -0500612 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700613
614 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400615 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700616 while (l <= r) {
617 m = l + (r - l) / 2;
618 if (block < le32_to_cpu(m->ei_block))
619 r = m - 1;
620 else
621 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400622 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
623 m, le32_to_cpu(m->ei_block),
624 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700625 }
626
627 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400628 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400629 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700630
631#ifdef CHECK_BINSEARCH
632 {
633 struct ext4_extent_idx *chix, *ix;
634 int k;
635
636 chix = ix = EXT_FIRST_INDEX(eh);
637 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
638 if (k != 0 &&
639 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400640 printk(KERN_DEBUG "k=%d, ix=0x%p, "
641 "first=0x%p\n", k,
642 ix, EXT_FIRST_INDEX(eh));
643 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700644 le32_to_cpu(ix->ei_block),
645 le32_to_cpu(ix[-1].ei_block));
646 }
647 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400648 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700649 if (block < le32_to_cpu(ix->ei_block))
650 break;
651 chix = ix;
652 }
653 BUG_ON(chix != path->p_idx);
654 }
655#endif
656
657}
658
659/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700660 * ext4_ext_binsearch:
661 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400662 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700663 */
664static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500665ext4_ext_binsearch(struct inode *inode,
666 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700667{
668 struct ext4_extent_header *eh = path->p_hdr;
669 struct ext4_extent *r, *l, *m;
670
Alex Tomasa86c6182006-10-11 01:21:03 -0700671 if (eh->eh_entries == 0) {
672 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700673 * this leaf is empty:
674 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700675 */
676 return;
677 }
678
Eric Sandeenbba90742008-01-28 23:58:27 -0500679 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700680
681 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400682 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700683
684 while (l <= r) {
685 m = l + (r - l) / 2;
686 if (block < le32_to_cpu(m->ee_block))
687 r = m - 1;
688 else
689 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400690 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
691 m, le32_to_cpu(m->ee_block),
692 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700693 }
694
695 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400696 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400697 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400698 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400699 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400700 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700701
702#ifdef CHECK_BINSEARCH
703 {
704 struct ext4_extent *chex, *ex;
705 int k;
706
707 chex = ex = EXT_FIRST_EXTENT(eh);
708 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
709 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400710 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700711 if (block < le32_to_cpu(ex->ee_block))
712 break;
713 chex = ex;
714 }
715 BUG_ON(chex != path->p_ext);
716 }
717#endif
718
719}
720
721int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
722{
723 struct ext4_extent_header *eh;
724
725 eh = ext_inode_hdr(inode);
726 eh->eh_depth = 0;
727 eh->eh_entries = 0;
728 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400729 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700730 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700731 return 0;
732}
733
734struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500735ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
736 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700737{
738 struct ext4_extent_header *eh;
739 struct buffer_head *bh;
740 short int depth, i, ppos = 0, alloc = 0;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500741 int ret;
Alex Tomasa86c6182006-10-11 01:21:03 -0700742
743 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400744 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700745
746 /* account possible depth increase */
747 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800748 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700749 GFP_NOFS);
750 if (!path)
751 return ERR_PTR(-ENOMEM);
752 alloc = 1;
753 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700754 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400755 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700756
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400757 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700758 /* walk through the tree */
759 while (i) {
760 ext_debug("depth %d: num %d, max %d\n",
761 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400762
Alex Tomasa86c6182006-10-11 01:21:03 -0700763 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400764 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700765 path[ppos].p_depth = i;
766 path[ppos].p_ext = NULL;
767
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400768 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500769 if (unlikely(!bh)) {
770 ret = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700771 goto err;
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500772 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400773 if (!bh_uptodate_or_lock(bh)) {
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400774 trace_ext4_ext_load_extent(inode, block,
775 path[ppos].p_block);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500776 ret = bh_submit_read(bh);
777 if (ret < 0) {
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400778 put_bh(bh);
779 goto err;
780 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400781 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700782 eh = ext_block_hdr(bh);
783 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500784 if (unlikely(ppos > depth)) {
785 put_bh(bh);
786 EXT4_ERROR_INODE(inode,
787 "ppos %d > depth %d", ppos, depth);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500788 ret = -EIO;
Frank Mayhar273df552010-03-02 11:46:09 -0500789 goto err;
790 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700791 path[ppos].p_bh = bh;
792 path[ppos].p_hdr = eh;
793 i--;
794
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500795 ret = ext4_ext_check_block(inode, eh, i, bh);
796 if (ret < 0)
Alex Tomasa86c6182006-10-11 01:21:03 -0700797 goto err;
798 }
799
800 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700801 path[ppos].p_ext = NULL;
802 path[ppos].p_idx = NULL;
803
Alex Tomasa86c6182006-10-11 01:21:03 -0700804 /* find extent */
805 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400806 /* if not an empty leaf */
807 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400808 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700809
810 ext4_ext_show_path(inode, path);
811
812 return path;
813
814err:
815 ext4_ext_drop_refs(path);
816 if (alloc)
817 kfree(path);
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500818 return ERR_PTR(ret);
Alex Tomasa86c6182006-10-11 01:21:03 -0700819}
820
821/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700822 * ext4_ext_insert_index:
823 * insert new index [@logical;@ptr] into the block at @curp;
824 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700825 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400826static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
827 struct ext4_ext_path *curp,
828 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700829{
830 struct ext4_extent_idx *ix;
831 int len, err;
832
Avantika Mathur7e028972006-12-06 20:41:33 -0800833 err = ext4_ext_get_access(handle, inode, curp);
834 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700835 return err;
836
Frank Mayhar273df552010-03-02 11:46:09 -0500837 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
838 EXT4_ERROR_INODE(inode,
839 "logical %d == ei_block %d!",
840 logical, le32_to_cpu(curp->p_idx->ei_block));
841 return -EIO;
842 }
Robin Dongd4620312011-07-17 23:43:42 -0400843
844 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
845 >= le16_to_cpu(curp->p_hdr->eh_max))) {
846 EXT4_ERROR_INODE(inode,
847 "eh_entries %d >= eh_max %d!",
848 le16_to_cpu(curp->p_hdr->eh_entries),
849 le16_to_cpu(curp->p_hdr->eh_max));
850 return -EIO;
851 }
852
Alex Tomasa86c6182006-10-11 01:21:03 -0700853 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
854 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400855 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700856 ix = curp->p_idx + 1;
857 } else {
858 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400859 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700860 ix = curp->p_idx;
861 }
862
Eric Gouriou80e675f2011-10-27 11:52:18 -0400863 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
864 BUG_ON(len < 0);
865 if (len > 0) {
866 ext_debug("insert new index %d: "
867 "move %d indices from 0x%p to 0x%p\n",
868 logical, len, ix, ix + 1);
869 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
870 }
871
Tao Maf472e022011-10-17 10:13:46 -0400872 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
873 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
874 return -EIO;
875 }
876
Alex Tomasa86c6182006-10-11 01:21:03 -0700877 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700878 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400879 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700880
Frank Mayhar273df552010-03-02 11:46:09 -0500881 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
882 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
883 return -EIO;
884 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700885
886 err = ext4_ext_dirty(handle, inode, curp);
887 ext4_std_error(inode->i_sb, err);
888
889 return err;
890}
891
892/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700893 * ext4_ext_split:
894 * inserts new subtree into the path, using free index entry
895 * at depth @at:
896 * - allocates all needed blocks (new leaf and all intermediate index blocks)
897 * - makes decision where to split
898 * - moves remaining extents and index entries (right to the split point)
899 * into the newly allocated blocks
900 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700901 */
902static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400903 unsigned int flags,
904 struct ext4_ext_path *path,
905 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -0700906{
907 struct buffer_head *bh = NULL;
908 int depth = ext_depth(inode);
909 struct ext4_extent_header *neh;
910 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -0700911 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700912 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700913 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700914 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700915 int err = 0;
916
917 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700918 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700919
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700920 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700921 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -0500922 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
923 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
924 return -EIO;
925 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700926 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
927 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700928 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700929 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400930 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700931 } else {
932 border = newext->ee_block;
933 ext_debug("leaf will be added."
934 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400935 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700936 }
937
938 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700939 * If error occurs, then we break processing
940 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700941 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700942 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700943 */
944
945 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700946 * Get array to track all allocated blocks.
947 * We need this to handle errors and free blocks
948 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700949 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800950 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700951 if (!ablocks)
952 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700953
954 /* allocate all needed blocks */
955 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
956 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400957 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400958 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -0700959 if (newblock == 0)
960 goto cleanup;
961 ablocks[a] = newblock;
962 }
963
964 /* initialize new leaf */
965 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -0500966 if (unlikely(newblock == 0)) {
967 EXT4_ERROR_INODE(inode, "newblock == 0!");
968 err = -EIO;
969 goto cleanup;
970 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700971 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -0500972 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500973 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700974 goto cleanup;
975 }
976 lock_buffer(bh);
977
Avantika Mathur7e028972006-12-06 20:41:33 -0800978 err = ext4_journal_get_create_access(handle, bh);
979 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700980 goto cleanup;
981
982 neh = ext_block_hdr(bh);
983 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400984 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700985 neh->eh_magic = EXT4_EXT_MAGIC;
986 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700987
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700988 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -0500989 if (unlikely(path[depth].p_hdr->eh_entries !=
990 path[depth].p_hdr->eh_max)) {
991 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
992 path[depth].p_hdr->eh_entries,
993 path[depth].p_hdr->eh_max);
994 err = -EIO;
995 goto cleanup;
996 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700997 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400998 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
999 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07001000 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001001 struct ext4_extent *ex;
1002 ex = EXT_FIRST_EXTENT(neh);
1003 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001004 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001005 }
1006
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001007 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001008 set_buffer_uptodate(bh);
1009 unlock_buffer(bh);
1010
Frank Mayhar03901312009-01-07 00:06:22 -05001011 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001012 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001013 goto cleanup;
1014 brelse(bh);
1015 bh = NULL;
1016
1017 /* correct old leaf */
1018 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001019 err = ext4_ext_get_access(handle, inode, path + depth);
1020 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001021 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001022 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001023 err = ext4_ext_dirty(handle, inode, path + depth);
1024 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001025 goto cleanup;
1026
1027 }
1028
1029 /* create intermediate indexes */
1030 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001031 if (unlikely(k < 0)) {
1032 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1033 err = -EIO;
1034 goto cleanup;
1035 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001036 if (k)
1037 ext_debug("create %d intermediate indices\n", k);
1038 /* insert new index into current index block */
1039 /* current depth stored in i var */
1040 i = depth - 1;
1041 while (k--) {
1042 oldblock = newblock;
1043 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001044 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001045 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001046 err = -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001047 goto cleanup;
1048 }
1049 lock_buffer(bh);
1050
Avantika Mathur7e028972006-12-06 20:41:33 -08001051 err = ext4_journal_get_create_access(handle, bh);
1052 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001053 goto cleanup;
1054
1055 neh = ext_block_hdr(bh);
1056 neh->eh_entries = cpu_to_le16(1);
1057 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001058 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001059 neh->eh_depth = cpu_to_le16(depth - i);
1060 fidx = EXT_FIRST_INDEX(neh);
1061 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001062 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001063
Eric Sandeenbba90742008-01-28 23:58:27 -05001064 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1065 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001066
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001067 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001068 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1069 EXT_LAST_INDEX(path[i].p_hdr))) {
1070 EXT4_ERROR_INODE(inode,
1071 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1072 le32_to_cpu(path[i].p_ext->ee_block));
1073 err = -EIO;
1074 goto cleanup;
1075 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001076 /* start copy indexes */
1077 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1078 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1079 EXT_MAX_INDEX(path[i].p_hdr));
1080 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001081 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001082 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001083 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001084 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001085 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001086 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001087 set_buffer_uptodate(bh);
1088 unlock_buffer(bh);
1089
Frank Mayhar03901312009-01-07 00:06:22 -05001090 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001091 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001092 goto cleanup;
1093 brelse(bh);
1094 bh = NULL;
1095
1096 /* correct old index */
1097 if (m) {
1098 err = ext4_ext_get_access(handle, inode, path + i);
1099 if (err)
1100 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001101 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001102 err = ext4_ext_dirty(handle, inode, path + i);
1103 if (err)
1104 goto cleanup;
1105 }
1106
1107 i--;
1108 }
1109
1110 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001111 err = ext4_ext_insert_index(handle, inode, path + at,
1112 le32_to_cpu(border), newblock);
1113
1114cleanup:
1115 if (bh) {
1116 if (buffer_locked(bh))
1117 unlock_buffer(bh);
1118 brelse(bh);
1119 }
1120
1121 if (err) {
1122 /* free all allocated blocks in error case */
1123 for (i = 0; i < depth; i++) {
1124 if (!ablocks[i])
1125 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001126 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001127 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001128 }
1129 }
1130 kfree(ablocks);
1131
1132 return err;
1133}
1134
1135/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001136 * ext4_ext_grow_indepth:
1137 * implements tree growing procedure:
1138 * - allocates new block
1139 * - moves top-level data (index block or leaf) into the new block
1140 * - initializes new top-level, creating index that points to the
1141 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001142 */
1143static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001144 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001145 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001146{
Alex Tomasa86c6182006-10-11 01:21:03 -07001147 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001148 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001149 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001150 int err = 0;
1151
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001152 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001153 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001154 if (newblock == 0)
1155 return err;
1156
1157 bh = sb_getblk(inode->i_sb, newblock);
Wang Shilongaebf0242013-01-12 16:28:47 -05001158 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001159 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -07001160 lock_buffer(bh);
1161
Avantika Mathur7e028972006-12-06 20:41:33 -08001162 err = ext4_journal_get_create_access(handle, bh);
1163 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001164 unlock_buffer(bh);
1165 goto out;
1166 }
1167
1168 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001169 memmove(bh->b_data, EXT4_I(inode)->i_data,
1170 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001171
1172 /* set size of new block */
1173 neh = ext_block_hdr(bh);
1174 /* old root could have indexes or leaves
1175 * so calculate e_max right way */
1176 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001177 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001178 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001179 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001180 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001181 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001182 set_buffer_uptodate(bh);
1183 unlock_buffer(bh);
1184
Frank Mayhar03901312009-01-07 00:06:22 -05001185 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001186 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001187 goto out;
1188
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001189 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001190 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001191 neh->eh_entries = cpu_to_le16(1);
1192 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1193 if (neh->eh_depth == 0) {
1194 /* Root extent block becomes index block */
1195 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1196 EXT_FIRST_INDEX(neh)->ei_block =
1197 EXT_FIRST_EXTENT(neh)->ee_block;
1198 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001199 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001200 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001201 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001202 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001203
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001204 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001205 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001206out:
1207 brelse(bh);
1208
1209 return err;
1210}
1211
1212/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001213 * ext4_ext_create_new_leaf:
1214 * finds empty index and adds new leaf.
1215 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001216 */
1217static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001218 unsigned int flags,
1219 struct ext4_ext_path *path,
1220 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001221{
1222 struct ext4_ext_path *curp;
1223 int depth, i, err = 0;
1224
1225repeat:
1226 i = depth = ext_depth(inode);
1227
1228 /* walk up to the tree and look for free index entry */
1229 curp = path + depth;
1230 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1231 i--;
1232 curp--;
1233 }
1234
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001235 /* we use already allocated block for index block,
1236 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001237 if (EXT_HAS_FREE_INDEX(curp)) {
1238 /* if we found index with free entry, then use that
1239 * entry: create all needed subtree and add new leaf */
Allison Henderson55f020d2011-05-25 07:41:26 -04001240 err = ext4_ext_split(handle, inode, flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001241 if (err)
1242 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001243
1244 /* refill path */
1245 ext4_ext_drop_refs(path);
1246 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001247 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1248 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001249 if (IS_ERR(path))
1250 err = PTR_ERR(path);
1251 } else {
1252 /* tree is full, time to grow in depth */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001253 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001254 if (err)
1255 goto out;
1256
1257 /* refill path */
1258 ext4_ext_drop_refs(path);
1259 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001260 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1261 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001262 if (IS_ERR(path)) {
1263 err = PTR_ERR(path);
1264 goto out;
1265 }
1266
1267 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001268 * only first (depth 0 -> 1) produces free space;
1269 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001270 */
1271 depth = ext_depth(inode);
1272 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001273 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001274 goto repeat;
1275 }
1276 }
1277
1278out:
1279 return err;
1280}
1281
1282/*
Alex Tomas1988b512008-01-28 23:58:27 -05001283 * search the closest allocated block to the left for *logical
1284 * and returns it at @logical + it's physical address at @phys
1285 * if *logical is the smallest allocated block, the function
1286 * returns 0 at @phys
1287 * return value contains 0 (success) or error code
1288 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001289static int ext4_ext_search_left(struct inode *inode,
1290 struct ext4_ext_path *path,
1291 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001292{
1293 struct ext4_extent_idx *ix;
1294 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001295 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001296
Frank Mayhar273df552010-03-02 11:46:09 -05001297 if (unlikely(path == NULL)) {
1298 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1299 return -EIO;
1300 }
Alex Tomas1988b512008-01-28 23:58:27 -05001301 depth = path->p_depth;
1302 *phys = 0;
1303
1304 if (depth == 0 && path->p_ext == NULL)
1305 return 0;
1306
1307 /* usually extent in the path covers blocks smaller
1308 * then *logical, but it can be that extent is the
1309 * first one in the file */
1310
1311 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001312 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001313 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001314 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1315 EXT4_ERROR_INODE(inode,
1316 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1317 *logical, le32_to_cpu(ex->ee_block));
1318 return -EIO;
1319 }
Alex Tomas1988b512008-01-28 23:58:27 -05001320 while (--depth >= 0) {
1321 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001322 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1323 EXT4_ERROR_INODE(inode,
1324 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001325 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001326 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001327 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001328 depth);
1329 return -EIO;
1330 }
Alex Tomas1988b512008-01-28 23:58:27 -05001331 }
1332 return 0;
1333 }
1334
Frank Mayhar273df552010-03-02 11:46:09 -05001335 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1336 EXT4_ERROR_INODE(inode,
1337 "logical %d < ee_block %d + ee_len %d!",
1338 *logical, le32_to_cpu(ex->ee_block), ee_len);
1339 return -EIO;
1340 }
Alex Tomas1988b512008-01-28 23:58:27 -05001341
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001342 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001343 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001344 return 0;
1345}
1346
1347/*
1348 * search the closest allocated block to the right for *logical
1349 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001350 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001351 * returns 0 at @phys
1352 * return value contains 0 (success) or error code
1353 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001354static int ext4_ext_search_right(struct inode *inode,
1355 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001356 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1357 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001358{
1359 struct buffer_head *bh = NULL;
1360 struct ext4_extent_header *eh;
1361 struct ext4_extent_idx *ix;
1362 struct ext4_extent *ex;
1363 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001364 int depth; /* Note, NOT eh_depth; depth from top of tree */
1365 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001366
Frank Mayhar273df552010-03-02 11:46:09 -05001367 if (unlikely(path == NULL)) {
1368 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1369 return -EIO;
1370 }
Alex Tomas1988b512008-01-28 23:58:27 -05001371 depth = path->p_depth;
1372 *phys = 0;
1373
1374 if (depth == 0 && path->p_ext == NULL)
1375 return 0;
1376
1377 /* usually extent in the path covers blocks smaller
1378 * then *logical, but it can be that extent is the
1379 * first one in the file */
1380
1381 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001382 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001383 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001384 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1385 EXT4_ERROR_INODE(inode,
1386 "first_extent(path[%d].p_hdr) != ex",
1387 depth);
1388 return -EIO;
1389 }
Alex Tomas1988b512008-01-28 23:58:27 -05001390 while (--depth >= 0) {
1391 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001392 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1393 EXT4_ERROR_INODE(inode,
1394 "ix != EXT_FIRST_INDEX *logical %d!",
1395 *logical);
1396 return -EIO;
1397 }
Alex Tomas1988b512008-01-28 23:58:27 -05001398 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001399 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001400 }
1401
Frank Mayhar273df552010-03-02 11:46:09 -05001402 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1403 EXT4_ERROR_INODE(inode,
1404 "logical %d < ee_block %d + ee_len %d!",
1405 *logical, le32_to_cpu(ex->ee_block), ee_len);
1406 return -EIO;
1407 }
Alex Tomas1988b512008-01-28 23:58:27 -05001408
1409 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1410 /* next allocated block in this leaf */
1411 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001412 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001413 }
1414
1415 /* go up and search for index to the right */
1416 while (--depth >= 0) {
1417 ix = path[depth].p_idx;
1418 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001419 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001420 }
1421
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001422 /* we've gone up to the root and found no index to the right */
1423 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001424
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001425got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001426 /* we've found index to the right, let's
1427 * follow it and find the closest allocated
1428 * block to the right */
1429 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001430 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001431 while (++depth < path->p_depth) {
1432 bh = sb_bread(inode->i_sb, block);
1433 if (bh == NULL)
1434 return -EIO;
1435 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001436 /* subtract from p_depth to get proper eh_depth */
Darrick J. Wongf8489122012-04-29 18:21:10 -04001437 if (ext4_ext_check_block(inode, eh,
1438 path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001439 put_bh(bh);
1440 return -EIO;
1441 }
1442 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001443 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001444 put_bh(bh);
1445 }
1446
1447 bh = sb_bread(inode->i_sb, block);
1448 if (bh == NULL)
1449 return -EIO;
1450 eh = ext_block_hdr(bh);
Darrick J. Wongf8489122012-04-29 18:21:10 -04001451 if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001452 put_bh(bh);
1453 return -EIO;
1454 }
1455 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001456found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001457 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001458 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001459 *ret_ex = ex;
1460 if (bh)
1461 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001462 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001463}
1464
1465/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001466 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001467 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001468 * NOTE: it considers block number from index entry as
1469 * allocated block. Thus, index entries have to be consistent
1470 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001471 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001472static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001473ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1474{
1475 int depth;
1476
1477 BUG_ON(path == NULL);
1478 depth = path->p_depth;
1479
1480 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001481 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001482
1483 while (depth >= 0) {
1484 if (depth == path->p_depth) {
1485 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001486 if (path[depth].p_ext &&
1487 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001488 EXT_LAST_EXTENT(path[depth].p_hdr))
1489 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1490 } else {
1491 /* index */
1492 if (path[depth].p_idx !=
1493 EXT_LAST_INDEX(path[depth].p_hdr))
1494 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1495 }
1496 depth--;
1497 }
1498
Lukas Czernerf17722f2011-06-06 00:05:17 -04001499 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001500}
1501
1502/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001503 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001504 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001505 */
Robin Dong57187892011-07-23 21:49:07 -04001506static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001507{
1508 int depth;
1509
1510 BUG_ON(path == NULL);
1511 depth = path->p_depth;
1512
1513 /* zero-tree has no leaf blocks at all */
1514 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001515 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001516
1517 /* go to index block */
1518 depth--;
1519
1520 while (depth >= 0) {
1521 if (path[depth].p_idx !=
1522 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001523 return (ext4_lblk_t)
1524 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001525 depth--;
1526 }
1527
Lukas Czernerf17722f2011-06-06 00:05:17 -04001528 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001529}
1530
1531/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001532 * ext4_ext_correct_indexes:
1533 * if leaf gets modified and modified extent is first in the leaf,
1534 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001535 * TODO: do we need to correct tree in all cases?
1536 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001537static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001538 struct ext4_ext_path *path)
1539{
1540 struct ext4_extent_header *eh;
1541 int depth = ext_depth(inode);
1542 struct ext4_extent *ex;
1543 __le32 border;
1544 int k, err = 0;
1545
1546 eh = path[depth].p_hdr;
1547 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001548
1549 if (unlikely(ex == NULL || eh == NULL)) {
1550 EXT4_ERROR_INODE(inode,
1551 "ex %p == NULL or eh %p == NULL", ex, eh);
1552 return -EIO;
1553 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001554
1555 if (depth == 0) {
1556 /* there is no tree at all */
1557 return 0;
1558 }
1559
1560 if (ex != EXT_FIRST_EXTENT(eh)) {
1561 /* we correct tree if first leaf got modified only */
1562 return 0;
1563 }
1564
1565 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001566 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001567 */
1568 k = depth - 1;
1569 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001570 err = ext4_ext_get_access(handle, inode, path + k);
1571 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001572 return err;
1573 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001574 err = ext4_ext_dirty(handle, inode, path + k);
1575 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001576 return err;
1577
1578 while (k--) {
1579 /* change all left-side indexes */
1580 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1581 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001582 err = ext4_ext_get_access(handle, inode, path + k);
1583 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001584 break;
1585 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001586 err = ext4_ext_dirty(handle, inode, path + k);
1587 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001588 break;
1589 }
1590
1591 return err;
1592}
1593
Akira Fujita748de672009-06-17 19:24:03 -04001594int
Alex Tomasa86c6182006-10-11 01:21:03 -07001595ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1596 struct ext4_extent *ex2)
1597{
Amit Arora749269f2007-07-18 09:02:56 -04001598 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001599
1600 /*
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001601 * Make sure that both extents are initialized. We don't merge
1602 * uninitialized extents so that we can be sure that end_io code has
1603 * the extent that was written properly split out and conversion to
1604 * initialized is trivial.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001605 */
Dmitry Monakhovec22ba82013-03-04 00:36:06 -05001606 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
Amit Aroraa2df2a62007-07-17 21:42:41 -04001607 return 0;
1608
Amit Arora749269f2007-07-18 09:02:56 -04001609 if (ext4_ext_is_uninitialized(ex1))
1610 max_len = EXT_UNINIT_MAX_LEN;
1611 else
1612 max_len = EXT_INIT_MAX_LEN;
1613
Amit Aroraa2df2a62007-07-17 21:42:41 -04001614 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1615 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1616
1617 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001618 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001619 return 0;
1620
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001621 /*
1622 * To allow future support for preallocated extents to be added
1623 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001624 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001625 */
Amit Arora749269f2007-07-18 09:02:56 -04001626 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001627 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001628#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001629 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001630 return 0;
1631#endif
1632
Theodore Ts'obf89d162010-10-27 21:30:14 -04001633 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001634 return 1;
1635 return 0;
1636}
1637
1638/*
Amit Arora56055d32007-07-17 21:42:38 -04001639 * This function tries to merge the "ex" extent to the next extent in the tree.
1640 * It always tries to merge towards right. If you want to merge towards
1641 * left, pass "ex - 1" as argument instead of "ex".
1642 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1643 * 1 if they got merged.
1644 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001645static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001646 struct ext4_ext_path *path,
1647 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001648{
1649 struct ext4_extent_header *eh;
1650 unsigned int depth, len;
1651 int merge_done = 0;
1652 int uninitialized = 0;
1653
1654 depth = ext_depth(inode);
1655 BUG_ON(path[depth].p_hdr == NULL);
1656 eh = path[depth].p_hdr;
1657
1658 while (ex < EXT_LAST_EXTENT(eh)) {
1659 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1660 break;
1661 /* merge with next extent! */
1662 if (ext4_ext_is_uninitialized(ex))
1663 uninitialized = 1;
1664 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1665 + ext4_ext_get_actual_len(ex + 1));
1666 if (uninitialized)
1667 ext4_ext_mark_uninitialized(ex);
1668
1669 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1670 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1671 * sizeof(struct ext4_extent);
1672 memmove(ex + 1, ex + 2, len);
1673 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001674 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001675 merge_done = 1;
1676 WARN_ON(eh->eh_entries == 0);
1677 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001678 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001679 }
1680
1681 return merge_done;
1682}
1683
1684/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001685 * This function does a very simple check to see if we can collapse
1686 * an extent tree with a single extent tree leaf block into the inode.
1687 */
1688static void ext4_ext_try_to_merge_up(handle_t *handle,
1689 struct inode *inode,
1690 struct ext4_ext_path *path)
1691{
1692 size_t s;
1693 unsigned max_root = ext4_ext_space_root(inode, 0);
1694 ext4_fsblk_t blk;
1695
1696 if ((path[0].p_depth != 1) ||
1697 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1698 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1699 return;
1700
1701 /*
1702 * We need to modify the block allocation bitmap and the block
1703 * group descriptor to release the extent tree block. If we
1704 * can't get the journal credits, give up.
1705 */
1706 if (ext4_journal_extend(handle, 2))
1707 return;
1708
1709 /*
1710 * Copy the extent data up to the inode
1711 */
1712 blk = ext4_idx_pblock(path[0].p_idx);
1713 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1714 sizeof(struct ext4_extent_idx);
1715 s += sizeof(struct ext4_extent_header);
1716
1717 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1718 path[0].p_depth = 0;
1719 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1720 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1721 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1722
1723 brelse(path[1].p_bh);
1724 ext4_free_blocks(handle, inode, NULL, blk, 1,
1725 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1726}
1727
1728/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001729 * This function tries to merge the @ex extent to neighbours in the tree.
1730 * return 1 if merge left else 0.
1731 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001732static void ext4_ext_try_to_merge(handle_t *handle,
1733 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001734 struct ext4_ext_path *path,
1735 struct ext4_extent *ex) {
1736 struct ext4_extent_header *eh;
1737 unsigned int depth;
1738 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001739
1740 depth = ext_depth(inode);
1741 BUG_ON(path[depth].p_hdr == NULL);
1742 eh = path[depth].p_hdr;
1743
1744 if (ex > EXT_FIRST_EXTENT(eh))
1745 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1746
1747 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001748 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001749
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001750 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001751}
1752
1753/*
Amit Arora25d14f92007-05-24 13:04:13 -04001754 * check if a portion of the "newext" extent overlaps with an
1755 * existing extent.
1756 *
1757 * If there is an overlap discovered, it updates the length of the newext
1758 * such that there will be no overlap, and then returns 1.
1759 * If there is no overlap found, it returns 0.
1760 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001761static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1762 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001763 struct ext4_extent *newext,
1764 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001765{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001766 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001767 unsigned int depth, len1;
1768 unsigned int ret = 0;
1769
1770 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001771 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001772 depth = ext_depth(inode);
1773 if (!path[depth].p_ext)
1774 goto out;
Theodore Ts'ocd744272013-12-20 09:29:35 -05001775 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
Amit Arora25d14f92007-05-24 13:04:13 -04001776
1777 /*
1778 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001779 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001780 */
1781 if (b2 < b1) {
1782 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001783 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001784 goto out;
Theodore Ts'ocd744272013-12-20 09:29:35 -05001785 b2 = EXT4_LBLK_CMASK(sbi, b2);
Amit Arora25d14f92007-05-24 13:04:13 -04001786 }
1787
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001788 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001789 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001790 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001791 newext->ee_len = cpu_to_le16(len1);
1792 ret = 1;
1793 }
1794
1795 /* check for overlap */
1796 if (b1 + len1 > b2) {
1797 newext->ee_len = cpu_to_le16(b2 - b1);
1798 ret = 1;
1799 }
1800out:
1801 return ret;
1802}
1803
1804/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001805 * ext4_ext_insert_extent:
1806 * tries to merge requsted extent into the existing extent or
1807 * inserts requested extent as new one into the tree,
1808 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001809 */
1810int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1811 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04001812 struct ext4_extent *newext, int flag)
Alex Tomasa86c6182006-10-11 01:21:03 -07001813{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001814 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001815 struct ext4_extent *ex, *fex;
1816 struct ext4_extent *nearex; /* nearest extent */
1817 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001818 int depth, len, err;
1819 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001820 unsigned uninitialized = 0;
Allison Henderson55f020d2011-05-25 07:41:26 -04001821 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001822
Frank Mayhar273df552010-03-02 11:46:09 -05001823 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1824 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1825 return -EIO;
1826 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001827 depth = ext_depth(inode);
1828 ex = path[depth].p_ext;
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001829 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05001830 if (unlikely(path[depth].p_hdr == NULL)) {
1831 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1832 return -EIO;
1833 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001834
1835 /* try to insert block into found extent and return */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001836 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04001837
1838 /*
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001839 * Try to see whether we should rather test the extent on
1840 * right from ex, or from the left of ex. This is because
1841 * ext4_ext_find_extent() can return either extent on the
1842 * left, or on the right from the searched position. This
1843 * will make merging more effective.
Amit Aroraa2df2a62007-07-17 21:42:41 -04001844 */
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001845 if (ex < EXT_LAST_EXTENT(eh) &&
1846 (le32_to_cpu(ex->ee_block) +
1847 ext4_ext_get_actual_len(ex) <
1848 le32_to_cpu(newext->ee_block))) {
1849 ex += 1;
1850 goto prepend;
1851 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1852 (le32_to_cpu(newext->ee_block) +
1853 ext4_ext_get_actual_len(newext) <
1854 le32_to_cpu(ex->ee_block)))
1855 ex -= 1;
1856
1857 /* Try to append newex to the ex */
1858 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1859 ext_debug("append [%d]%d block to %u:[%d]%d"
1860 "(from %llu)\n",
1861 ext4_ext_is_uninitialized(newext),
1862 ext4_ext_get_actual_len(newext),
1863 le32_to_cpu(ex->ee_block),
1864 ext4_ext_is_uninitialized(ex),
1865 ext4_ext_get_actual_len(ex),
1866 ext4_ext_pblock(ex));
1867 err = ext4_ext_get_access(handle, inode,
1868 path + depth);
1869 if (err)
1870 return err;
1871
1872 /*
1873 * ext4_can_extents_be_merged should have checked
1874 * that either both extents are uninitialized, or
1875 * both aren't. Thus we need to check only one of
1876 * them here.
1877 */
1878 if (ext4_ext_is_uninitialized(ex))
1879 uninitialized = 1;
1880 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001881 + ext4_ext_get_actual_len(newext));
Lukas Czernerbe8981b2013-04-03 23:33:28 -04001882 if (uninitialized)
1883 ext4_ext_mark_uninitialized(ex);
1884 eh = path[depth].p_hdr;
1885 nearex = ex;
1886 goto merge;
1887 }
1888
1889prepend:
1890 /* Try to prepend newex to the ex */
1891 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1892 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1893 "(from %llu)\n",
1894 le32_to_cpu(newext->ee_block),
1895 ext4_ext_is_uninitialized(newext),
1896 ext4_ext_get_actual_len(newext),
1897 le32_to_cpu(ex->ee_block),
1898 ext4_ext_is_uninitialized(ex),
1899 ext4_ext_get_actual_len(ex),
1900 ext4_ext_pblock(ex));
1901 err = ext4_ext_get_access(handle, inode,
1902 path + depth);
1903 if (err)
1904 return err;
1905
1906 /*
1907 * ext4_can_extents_be_merged should have checked
1908 * that either both extents are uninitialized, or
1909 * both aren't. Thus we need to check only one of
1910 * them here.
1911 */
1912 if (ext4_ext_is_uninitialized(ex))
1913 uninitialized = 1;
1914 ex->ee_block = newext->ee_block;
1915 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1916 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1917 + ext4_ext_get_actual_len(newext));
1918 if (uninitialized)
1919 ext4_ext_mark_uninitialized(ex);
1920 eh = path[depth].p_hdr;
1921 nearex = ex;
1922 goto merge;
1923 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001924 }
1925
Alex Tomasa86c6182006-10-11 01:21:03 -07001926 depth = ext_depth(inode);
1927 eh = path[depth].p_hdr;
1928 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1929 goto has_space;
1930
1931 /* probably next leaf has space for us? */
1932 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001933 next = EXT_MAX_BLOCKS;
1934 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001935 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001936 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001937 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07001938 BUG_ON(npath != NULL);
1939 npath = ext4_ext_find_extent(inode, next, NULL);
1940 if (IS_ERR(npath))
1941 return PTR_ERR(npath);
1942 BUG_ON(npath->p_depth != path->p_depth);
1943 eh = npath[depth].p_hdr;
1944 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001945 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001946 le16_to_cpu(eh->eh_entries));
1947 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04001948 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07001949 }
1950 ext_debug("next leaf has no free space(%d,%d)\n",
1951 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1952 }
1953
1954 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001955 * There is no free space in the found leaf.
1956 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001957 */
Lukas Czerner27dd4382013-04-09 22:11:22 -04001958 if (flag & EXT4_GET_BLOCKS_METADATA_NOFAIL)
1959 flags = EXT4_MB_USE_RESERVED;
Allison Henderson55f020d2011-05-25 07:41:26 -04001960 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001961 if (err)
1962 goto cleanup;
1963 depth = ext_depth(inode);
1964 eh = path[depth].p_hdr;
1965
1966has_space:
1967 nearex = path[depth].p_ext;
1968
Avantika Mathur7e028972006-12-06 20:41:33 -08001969 err = ext4_ext_get_access(handle, inode, path + depth);
1970 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001971 goto cleanup;
1972
1973 if (!nearex) {
1974 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001975 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001976 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001977 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001978 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001979 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04001980 nearex = EXT_FIRST_EXTENT(eh);
1981 } else {
1982 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001983 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04001984 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001985 ext_debug("insert %u:%llu:[%d]%d before: "
1986 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001987 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001988 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001989 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001990 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04001991 nearex);
1992 nearex++;
1993 } else {
1994 /* Insert before */
1995 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04001996 ext_debug("insert %u:%llu:[%d]%d after: "
1997 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04001998 le32_to_cpu(newext->ee_block),
1999 ext4_ext_pblock(newext),
2000 ext4_ext_is_uninitialized(newext),
2001 ext4_ext_get_actual_len(newext),
2002 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002003 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04002004 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2005 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04002006 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04002007 "move %d extents from 0x%p to 0x%p\n",
2008 le32_to_cpu(newext->ee_block),
2009 ext4_ext_pblock(newext),
2010 ext4_ext_is_uninitialized(newext),
2011 ext4_ext_get_actual_len(newext),
2012 len, nearex, nearex + 1);
2013 memmove(nearex + 1, nearex,
2014 len * sizeof(struct ext4_extent));
2015 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002016 }
2017
Marcin Slusarze8546d02008-04-17 10:38:59 -04002018 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04002019 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07002020 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002021 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07002022 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07002023
2024merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04002025 /* try to merge extents */
Jiaying Zhang744692d2010-03-04 16:14:02 -05002026 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002027 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002028
Alex Tomasa86c6182006-10-11 01:21:03 -07002029
2030 /* time to correct all indexes above */
2031 err = ext4_ext_correct_indexes(handle, inode, path);
2032 if (err)
2033 goto cleanup;
2034
Theodore Ts'oecb94f52012-08-17 09:44:17 -04002035 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002036
2037cleanup:
2038 if (npath) {
2039 ext4_ext_drop_refs(npath);
2040 kfree(npath);
2041 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002042 return err;
2043}
2044
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002045static int ext4_fill_fiemap_extents(struct inode *inode,
2046 ext4_lblk_t block, ext4_lblk_t num,
2047 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04002048{
2049 struct ext4_ext_path *path = NULL;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002050 struct ext4_extent *ex;
Zheng Liu69eb33d2013-02-18 00:31:07 -05002051 struct extent_status es;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002052 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002053 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002054 int exists, depth = 0, err = 0;
2055 unsigned int flags = 0;
2056 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002057
Lukas Czernerf17722f2011-06-06 00:05:17 -04002058 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04002059 num = last - block;
2060 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05002061 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002062
2063 if (path && ext_depth(inode) != depth) {
2064 /* depth was changed. we have to realloc path */
2065 kfree(path);
2066 path = NULL;
2067 }
2068
Eric Sandeen6873fa02008-10-07 00:46:36 -04002069 path = ext4_ext_find_extent(inode, block, path);
2070 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002071 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002072 err = PTR_ERR(path);
2073 path = NULL;
2074 break;
2075 }
2076
2077 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002078 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002079 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002080 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2081 err = -EIO;
2082 break;
2083 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002084 ex = path[depth].p_ext;
2085 next = ext4_ext_next_allocated_block(path);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002086 ext4_ext_drop_refs(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002087
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002088 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002089 exists = 0;
2090 if (!ex) {
2091 /* there is no extent yet, so try to allocate
2092 * all requested space */
2093 start = block;
2094 end = block + num;
2095 } else if (le32_to_cpu(ex->ee_block) > block) {
2096 /* need to allocate space before found extent */
2097 start = block;
2098 end = le32_to_cpu(ex->ee_block);
2099 if (block + num < end)
2100 end = block + num;
2101 } else if (block >= le32_to_cpu(ex->ee_block)
2102 + ext4_ext_get_actual_len(ex)) {
2103 /* need to allocate space after found extent */
2104 start = block;
2105 end = block + num;
2106 if (end >= next)
2107 end = next;
2108 } else if (block >= le32_to_cpu(ex->ee_block)) {
2109 /*
2110 * some part of requested space is covered
2111 * by found extent
2112 */
2113 start = block;
2114 end = le32_to_cpu(ex->ee_block)
2115 + ext4_ext_get_actual_len(ex);
2116 if (block + num < end)
2117 end = block + num;
2118 exists = 1;
2119 } else {
2120 BUG();
2121 }
2122 BUG_ON(end <= start);
2123
2124 if (!exists) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002125 es.es_lblk = start;
2126 es.es_len = end - start;
2127 es.es_pblk = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002128 } else {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002129 es.es_lblk = le32_to_cpu(ex->ee_block);
2130 es.es_len = ext4_ext_get_actual_len(ex);
2131 es.es_pblk = ext4_ext_pblock(ex);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002132 if (ext4_ext_is_uninitialized(ex))
2133 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002134 }
2135
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002136 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05002137 * Find delayed extent and update es accordingly. We call
2138 * it even in !exists case to find out whether es is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002139 * last existing extent or not.
2140 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05002141 next_del = ext4_find_delayed_extent(inode, &es);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002142 if (!exists && next_del) {
2143 exists = 1;
2144 flags |= FIEMAP_EXTENT_DELALLOC;
2145 }
2146 up_read(&EXT4_I(inode)->i_data_sem);
2147
Zheng Liu69eb33d2013-02-18 00:31:07 -05002148 if (unlikely(es.es_len == 0)) {
2149 EXT4_ERROR_INODE(inode, "es.es_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002150 err = -EIO;
2151 break;
2152 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002153
Zheng Liuf7fec032013-02-18 00:28:47 -05002154 /*
2155 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2156 * we need to check next == EXT_MAX_BLOCKS because it is
2157 * possible that an extent is with unwritten and delayed
2158 * status due to when an extent is delayed allocated and
2159 * is allocated by fallocate status tree will track both of
2160 * them in a extent.
2161 *
2162 * So we could return a unwritten and delayed extent, and
2163 * its block is equal to 'next'.
2164 */
2165 if (next == next_del && next == EXT_MAX_BLOCKS) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002166 flags |= FIEMAP_EXTENT_LAST;
2167 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2168 next != EXT_MAX_BLOCKS)) {
2169 EXT4_ERROR_INODE(inode,
2170 "next extent == %u, next "
2171 "delalloc extent = %u",
2172 next, next_del);
2173 err = -EIO;
2174 break;
2175 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002176 }
2177
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002178 if (exists) {
2179 err = fiemap_fill_next_extent(fieinfo,
Zheng Liu69eb33d2013-02-18 00:31:07 -05002180 (__u64)es.es_lblk << blksize_bits,
2181 (__u64)es.es_pblk << blksize_bits,
2182 (__u64)es.es_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002183 flags);
2184 if (err < 0)
2185 break;
2186 if (err == 1) {
2187 err = 0;
2188 break;
2189 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002190 }
2191
Zheng Liu69eb33d2013-02-18 00:31:07 -05002192 block = es.es_lblk + es.es_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002193 }
2194
2195 if (path) {
2196 ext4_ext_drop_refs(path);
2197 kfree(path);
2198 }
2199
2200 return err;
2201}
2202
Alex Tomasa86c6182006-10-11 01:21:03 -07002203/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002204 * ext4_ext_put_gap_in_cache:
2205 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002206 * and cache this gap
2207 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002208static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002209ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002210 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002211{
2212 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002213 unsigned long len;
2214 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002215 struct ext4_extent *ex;
2216
2217 ex = path[depth].p_ext;
2218 if (ex == NULL) {
Zheng Liu69eb33d2013-02-18 00:31:07 -05002219 /*
2220 * there is no extent yet, so gap is [0;-] and we
2221 * don't cache it
2222 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002223 ext_debug("cache gap(whole file):");
2224 } else if (block < le32_to_cpu(ex->ee_block)) {
2225 lblock = block;
2226 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002227 ext_debug("cache gap(before): %u [%u:%u]",
2228 block,
2229 le32_to_cpu(ex->ee_block),
2230 ext4_ext_get_actual_len(ex));
Zheng Liud100eef2013-02-18 00:29:59 -05002231 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2232 ext4_es_insert_extent(inode, lblock, len, ~0,
2233 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002234 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002235 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002236 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002237 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002238 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002239
2240 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002241 ext_debug("cache gap(after): [%u:%u] %u",
2242 le32_to_cpu(ex->ee_block),
2243 ext4_ext_get_actual_len(ex),
2244 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002245 BUG_ON(next == lblock);
2246 len = next - lblock;
Zheng Liud100eef2013-02-18 00:29:59 -05002247 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2248 ext4_es_insert_extent(inode, lblock, len, ~0,
2249 EXTENT_STATUS_HOLE);
Alex Tomasa86c6182006-10-11 01:21:03 -07002250 } else {
2251 lblock = len = 0;
2252 BUG();
2253 }
2254
Eric Sandeenbba90742008-01-28 23:58:27 -05002255 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002256}
2257
2258/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002259 * ext4_ext_rm_idx:
2260 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002261 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002262static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002263 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002264{
Alex Tomasa86c6182006-10-11 01:21:03 -07002265 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002266 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002267
2268 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002269 depth--;
2270 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002271 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002272 if (unlikely(path->p_hdr->eh_entries == 0)) {
2273 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2274 return -EIO;
2275 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002276 err = ext4_ext_get_access(handle, inode, path);
2277 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002278 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002279
2280 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2281 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2282 len *= sizeof(struct ext4_extent_idx);
2283 memmove(path->p_idx, path->p_idx + 1, len);
2284 }
2285
Marcin Slusarze8546d02008-04-17 10:38:59 -04002286 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002287 err = ext4_ext_dirty(handle, inode, path);
2288 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002289 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002290 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002291 trace_ext4_ext_rm_idx(inode, leaf);
2292
Peter Huewe7dc57612011-02-21 21:01:42 -05002293 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002294 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002295
2296 while (--depth >= 0) {
2297 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2298 break;
2299 path--;
2300 err = ext4_ext_get_access(handle, inode, path);
2301 if (err)
2302 break;
2303 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2304 err = ext4_ext_dirty(handle, inode, path);
2305 if (err)
2306 break;
2307 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002308 return err;
2309}
2310
2311/*
Mingming Caoee12b632008-08-19 22:16:05 -04002312 * ext4_ext_calc_credits_for_single_extent:
2313 * This routine returns max. credits that needed to insert an extent
2314 * to the extent tree.
2315 * When pass the actual path, the caller should calculate credits
2316 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002317 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002318int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002319 struct ext4_ext_path *path)
2320{
Alex Tomasa86c6182006-10-11 01:21:03 -07002321 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002322 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002323 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002324
Alex Tomasa86c6182006-10-11 01:21:03 -07002325 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002326 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002327 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2328
2329 /*
2330 * There are some space in the leaf tree, no
2331 * need to account for leaf block credit
2332 *
2333 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002334 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002335 * accounted.
2336 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002337 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002338 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002339 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002340 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002341 }
2342
Mingming Cao525f4ed2008-08-19 22:15:58 -04002343 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002344}
Alex Tomasa86c6182006-10-11 01:21:03 -07002345
Mingming Caoee12b632008-08-19 22:16:05 -04002346/*
2347 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2348 *
2349 * if nrblocks are fit in a single extent (chunk flag is 1), then
2350 * in the worse case, each tree level index/leaf need to be changed
2351 * if the tree split due to insert a new extent, then the old tree
2352 * index/leaf need to be updated too
2353 *
2354 * If the nrblocks are discontiguous, they could cause
2355 * the whole tree split more than once, but this is really rare.
2356 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002357int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04002358{
2359 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002360 int depth;
2361
2362 /* If we are converting the inline data, only one is needed here. */
2363 if (ext4_has_inline_data(inode))
2364 return 1;
2365
2366 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002367
Mingming Caoee12b632008-08-19 22:16:05 -04002368 if (chunk)
2369 index = depth * 2;
2370 else
2371 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002372
Mingming Caoee12b632008-08-19 22:16:05 -04002373 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002374}
2375
2376static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002377 struct ext4_extent *ex,
2378 ext4_fsblk_t *partial_cluster,
2379 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002380{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002381 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002382 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002383 ext4_fsblk_t pblk;
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002384 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002385
Alex Tomasc9de5602008-01-29 00:19:52 -05002386 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002387 flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2388 else if (ext4_should_journal_data(inode))
2389 flags |= EXT4_FREE_BLOCKS_FORGET;
2390
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002391 /*
2392 * For bigalloc file systems, we never free a partial cluster
2393 * at the beginning of the extent. Instead, we make a note
2394 * that we tried freeing the cluster, and check to see if we
2395 * need to free it on a subsequent call to ext4_remove_blocks,
2396 * or at the end of the ext4_truncate() operation.
2397 */
2398 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2399
Aditya Kalid8990242011-09-09 19:18:51 -04002400 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002401 /*
2402 * If we have a partial cluster, and it's different from the
2403 * cluster of the last block, we need to explicitly free the
2404 * partial cluster here.
2405 */
2406 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2407 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2408 ext4_free_blocks(handle, inode, NULL,
2409 EXT4_C2B(sbi, *partial_cluster),
2410 sbi->s_cluster_ratio, flags);
2411 *partial_cluster = 0;
2412 }
2413
Alex Tomasa86c6182006-10-11 01:21:03 -07002414#ifdef EXTENTS_STATS
2415 {
2416 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002417 spin_lock(&sbi->s_ext_stats_lock);
2418 sbi->s_ext_blocks += ee_len;
2419 sbi->s_ext_extents++;
2420 if (ee_len < sbi->s_ext_min)
2421 sbi->s_ext_min = ee_len;
2422 if (ee_len > sbi->s_ext_max)
2423 sbi->s_ext_max = ee_len;
2424 if (ext_depth(inode) > sbi->s_depth_max)
2425 sbi->s_depth_max = ext_depth(inode);
2426 spin_unlock(&sbi->s_ext_stats_lock);
2427 }
2428#endif
2429 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002430 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002431 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002432 ext4_lblk_t num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002433
Amit Aroraa2df2a62007-07-17 21:42:41 -04002434 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002435 pblk = ext4_ext_pblock(ex) + ee_len - num;
2436 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2437 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2438 /*
2439 * If the block range to be freed didn't start at the
2440 * beginning of a cluster, and we removed the entire
2441 * extent, save the partial cluster here, since we
2442 * might need to delete if we determine that the
2443 * truncate operation has removed all of the blocks in
2444 * the cluster.
2445 */
Theodore Ts'ocd744272013-12-20 09:29:35 -05002446 if (EXT4_PBLK_COFF(sbi, pblk) &&
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002447 (ee_len == num))
2448 *partial_cluster = EXT4_B2C(sbi, pblk);
2449 else
2450 *partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002451 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002452 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002453 /* head removal */
2454 ext4_lblk_t num;
2455 ext4_fsblk_t start;
2456
2457 num = to - from;
2458 start = ext4_ext_pblock(ex);
2459
2460 ext_debug("free first %u blocks starting %llu\n", num, start);
H Hartley Sweetenee90d572011-10-18 11:01:51 -04002461 ext4_free_blocks(handle, inode, NULL, start, num, flags);
Allison Hendersond583fb82011-05-25 07:41:43 -04002462
Alex Tomasa86c6182006-10-11 01:21:03 -07002463 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002464 printk(KERN_INFO "strange request: removal(2) "
2465 "%u-%u from %u:%u\n",
2466 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002467 }
2468 return 0;
2469}
2470
Allison Hendersond583fb82011-05-25 07:41:43 -04002471
2472/*
2473 * ext4_ext_rm_leaf() Removes the extents associated with the
2474 * blocks appearing between "start" and "end", and splits the extents
2475 * if "start" and "end" appear in the same extent
2476 *
2477 * @handle: The journal handle
2478 * @inode: The files inode
2479 * @path: The path to the leaf
2480 * @start: The first block to remove
2481 * @end: The last block to remove
2482 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002483static int
2484ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002485 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2486 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002487{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002488 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002489 int err = 0, correct_index = 0;
2490 int depth = ext_depth(inode), credits;
2491 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002492 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002493 unsigned num;
2494 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002495 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002496 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002497 struct ext4_extent *ex;
2498
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002499 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002500 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002501 if (!path[depth].p_hdr)
2502 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2503 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002504 if (unlikely(path[depth].p_hdr == NULL)) {
2505 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2506 return -EIO;
2507 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002508 /* find where to start removing */
2509 ex = EXT_LAST_EXTENT(eh);
2510
2511 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002512 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002513
Eric Whitney8459d9c2014-03-13 23:34:16 -04002514 /*
2515 * If we're starting with an extent other than the last one in the
2516 * node, we need to see if it shares a cluster with the extent to
2517 * the right (towards the end of the file). If its leftmost cluster
2518 * is this extent's rightmost cluster and it is not cluster aligned,
2519 * we'll mark it as a partial that is not to be deallocated.
2520 */
2521
2522 if (ex != EXT_LAST_EXTENT(eh)) {
2523 ext4_fsblk_t current_pblk, right_pblk;
2524 long long current_cluster, right_cluster;
2525
2526 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2527 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2528 right_pblk = ext4_ext_pblock(ex + 1);
2529 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2530 if (current_cluster == right_cluster &&
2531 EXT4_PBLK_COFF(sbi, right_pblk))
2532 *partial_cluster = -right_cluster;
2533 }
2534
Aditya Kalid8990242011-09-09 19:18:51 -04002535 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2536
Alex Tomasa86c6182006-10-11 01:21:03 -07002537 while (ex >= EXT_FIRST_EXTENT(eh) &&
2538 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002539
2540 if (ext4_ext_is_uninitialized(ex))
2541 uninitialized = 1;
2542 else
2543 uninitialized = 0;
2544
Mingming553f9002009-09-18 13:34:55 -04002545 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2546 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002547 path[depth].p_ext = ex;
2548
2549 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002550 b = ex_ee_block+ex_ee_len - 1 < end ?
2551 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002552
2553 ext_debug(" border %u:%u\n", a, b);
2554
Allison Hendersond583fb82011-05-25 07:41:43 -04002555 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002556 if (end < ex_ee_block) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002557 ex--;
2558 ex_ee_block = le32_to_cpu(ex->ee_block);
2559 ex_ee_len = ext4_ext_get_actual_len(ex);
2560 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002561 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002562 EXT4_ERROR_INODE(inode,
2563 "can not handle truncate %u:%u "
2564 "on extent %u:%u",
2565 start, end, ex_ee_block,
2566 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002567 err = -EIO;
2568 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002569 } else if (a != ex_ee_block) {
2570 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002571 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002572 } else {
2573 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002574 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002575 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002576 /*
2577 * 3 for leaf, sb, and inode plus 2 (bmap and group
2578 * descriptor) for each block group; assume two block
2579 * groups plus ex_ee_len/blocks_per_block_group for
2580 * the worst case
2581 */
2582 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002583 if (ex == EXT_FIRST_EXTENT(eh)) {
2584 correct_index = 1;
2585 credits += (ext_depth(inode)) + 1;
2586 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002587 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002588
Jan Kara487caee2009-08-17 22:17:20 -04002589 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002590 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002591 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002592
2593 err = ext4_ext_get_access(handle, inode, path + depth);
2594 if (err)
2595 goto out;
2596
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002597 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2598 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002599 if (err)
2600 goto out;
2601
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002602 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002603 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002604 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002605
Alex Tomasa86c6182006-10-11 01:21:03 -07002606 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002607 /*
2608 * Do not mark uninitialized if all the blocks in the
2609 * extent have been removed.
2610 */
2611 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002612 ext4_ext_mark_uninitialized(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002613 /*
2614 * If the extent was completely released,
2615 * we need to remove it from the leaf
2616 */
2617 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002618 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002619 /*
2620 * For hole punching, we need to scoot all the
2621 * extents up when an extent is removed so that
2622 * we dont have blank extents in the middle
2623 */
2624 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2625 sizeof(struct ext4_extent));
2626
2627 /* Now get rid of the one at the end */
2628 memset(EXT_LAST_EXTENT(eh), 0,
2629 sizeof(struct ext4_extent));
2630 }
2631 le16_add_cpu(&eh->eh_entries, -1);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002632 } else
2633 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002634
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002635 err = ext4_ext_dirty(handle, inode, path + depth);
2636 if (err)
2637 goto out;
2638
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002639 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002640 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002641 ex--;
2642 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002643 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002644 }
2645
2646 if (correct_index && eh->eh_entries)
2647 err = ext4_ext_correct_indexes(handle, inode, path);
2648
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002649 /*
2650 * If there is still a entry in the leaf node, check to see if
2651 * it references the partial cluster. This is the only place
2652 * where it could; if it doesn't, we can free the cluster.
2653 */
2654 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2655 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2656 *partial_cluster)) {
2657 int flags = EXT4_FREE_BLOCKS_FORGET;
2658
2659 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2660 flags |= EXT4_FREE_BLOCKS_METADATA;
2661
2662 ext4_free_blocks(handle, inode, NULL,
2663 EXT4_C2B(sbi, *partial_cluster),
2664 sbi->s_cluster_ratio, flags);
2665 *partial_cluster = 0;
2666 }
2667
Alex Tomasa86c6182006-10-11 01:21:03 -07002668 /* if this leaf is free, then we should
2669 * remove it from index block above */
2670 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002671 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002672
2673out:
2674 return err;
2675}
2676
2677/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002678 * ext4_ext_more_to_rm:
2679 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002680 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002681static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002682ext4_ext_more_to_rm(struct ext4_ext_path *path)
2683{
2684 BUG_ON(path->p_idx == NULL);
2685
2686 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2687 return 0;
2688
2689 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002690 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002691 * so we have to consider current index for truncation
2692 */
2693 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2694 return 0;
2695 return 1;
2696}
2697
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04002698int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2699 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002700{
2701 struct super_block *sb = inode->i_sb;
2702 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002703 struct ext4_ext_path *path = NULL;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002704 ext4_fsblk_t partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002705 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002706 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002707
Lukas Czerner5f95d212012-03-19 23:03:19 -04002708 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002709
2710 /* probably first extent we're gonna free will be last in block */
Theodore Ts'o9924a922013-02-08 21:59:22 -05002711 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002712 if (IS_ERR(handle))
2713 return PTR_ERR(handle);
2714
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002715again:
Aditya Kalid8990242011-09-09 19:18:51 -04002716 trace_ext4_ext_remove_space(inode, start, depth);
2717
Alex Tomasa86c6182006-10-11 01:21:03 -07002718 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002719 * Check if we are removing extents inside the extent tree. If that
2720 * is the case, we are going to punch a hole inside the extent tree
2721 * so we have to check whether we need to split the extent covering
2722 * the last block to remove so we can easily remove the part of it
2723 * in ext4_ext_rm_leaf().
2724 */
2725 if (end < EXT_MAX_BLOCKS - 1) {
2726 struct ext4_extent *ex;
2727 ext4_lblk_t ee_block;
2728
2729 /* find extent for this block */
2730 path = ext4_ext_find_extent(inode, end, NULL);
2731 if (IS_ERR(path)) {
2732 ext4_journal_stop(handle);
2733 return PTR_ERR(path);
2734 }
2735 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002736 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002737 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002738 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002739 if (depth) {
2740 EXT4_ERROR_INODE(inode,
2741 "path[%d].p_hdr == NULL",
2742 depth);
2743 err = -EIO;
2744 }
2745 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002746 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002747
2748 ee_block = le32_to_cpu(ex->ee_block);
2749
2750 /*
2751 * See if the last block is inside the extent, if so split
2752 * the extent at 'end' block so we can easily remove the
2753 * tail of the first part of the split extent in
2754 * ext4_ext_rm_leaf().
2755 */
2756 if (end >= ee_block &&
2757 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2758 int split_flag = 0;
2759
2760 if (ext4_ext_is_uninitialized(ex))
2761 split_flag = EXT4_EXT_MARK_UNINIT1 |
2762 EXT4_EXT_MARK_UNINIT2;
2763
2764 /*
2765 * Split the extent in two so that 'end' is the last
Lukas Czerner27dd4382013-04-09 22:11:22 -04002766 * block in the first new extent. Also we should not
2767 * fail removing space due to ENOSPC so try to use
2768 * reserved block if that happens.
Lukas Czerner5f95d212012-03-19 23:03:19 -04002769 */
2770 err = ext4_split_extent_at(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04002771 end + 1, split_flag,
2772 EXT4_GET_BLOCKS_PRE_IO |
2773 EXT4_GET_BLOCKS_METADATA_NOFAIL);
Lukas Czerner5f95d212012-03-19 23:03:19 -04002774
2775 if (err < 0)
2776 goto out;
2777 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002778 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002779 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002780 * We start scanning from right side, freeing all the blocks
2781 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002782 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002783 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002784 if (path) {
2785 int k = i = depth;
2786 while (--k > 0)
2787 path[k].p_block =
2788 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2789 } else {
2790 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2791 GFP_NOFS);
2792 if (path == NULL) {
2793 ext4_journal_stop(handle);
2794 return -ENOMEM;
2795 }
2796 path[0].p_depth = depth;
2797 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002798 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002799
Ashish Sangwan968dee72012-07-22 22:49:08 -04002800 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2801 err = -EIO;
2802 goto out;
2803 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002804 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002805 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002806
2807 while (i >= 0 && err == 0) {
2808 if (i == depth) {
2809 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002810 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002811 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002812 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002813 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002814 brelse(path[i].p_bh);
2815 path[i].p_bh = NULL;
2816 i--;
2817 continue;
2818 }
2819
2820 /* this is index block */
2821 if (!path[i].p_hdr) {
2822 ext_debug("initialize header\n");
2823 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002824 }
2825
Alex Tomasa86c6182006-10-11 01:21:03 -07002826 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002827 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002828 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2829 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2830 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2831 path[i].p_hdr,
2832 le16_to_cpu(path[i].p_hdr->eh_entries));
2833 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002834 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002835 path[i].p_idx--;
2836 }
2837
2838 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2839 i, EXT_FIRST_INDEX(path[i].p_hdr),
2840 path[i].p_idx);
2841 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002842 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002843 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002844 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002845 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002846 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'obf89d162010-10-27 21:30:14 -04002847 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002848 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002849 /* should we reset i_size? */
2850 err = -EIO;
2851 break;
2852 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002853 if (WARN_ON(i + 1 > depth)) {
2854 err = -EIO;
2855 break;
2856 }
Darrick J. Wongf8489122012-04-29 18:21:10 -04002857 if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2858 depth - i - 1, bh)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002859 err = -EIO;
2860 break;
2861 }
2862 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002863
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002864 /* save actual number of indexes since this
2865 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002866 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2867 i++;
2868 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002869 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002870 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002871 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002872 * handle must be already prepared by the
2873 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002874 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002875 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002876 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002877 brelse(path[i].p_bh);
2878 path[i].p_bh = NULL;
2879 i--;
2880 ext_debug("return to level %d\n", i);
2881 }
2882 }
2883
Aditya Kalid8990242011-09-09 19:18:51 -04002884 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2885 path->p_hdr->eh_entries);
2886
Aditya Kali7b415bf2011-09-09 19:04:51 -04002887 /* If we still have something in the partial cluster and we have removed
2888 * even the first extent, then we should free the blocks in the partial
2889 * cluster as well. */
2890 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2891 int flags = EXT4_FREE_BLOCKS_FORGET;
2892
2893 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2894 flags |= EXT4_FREE_BLOCKS_METADATA;
2895
2896 ext4_free_blocks(handle, inode, NULL,
2897 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2898 EXT4_SB(sb)->s_cluster_ratio, flags);
2899 partial_cluster = 0;
2900 }
2901
Alex Tomasa86c6182006-10-11 01:21:03 -07002902 /* TODO: flexible tree reduction should be here */
2903 if (path->p_hdr->eh_entries == 0) {
2904 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002905 * truncate to zero freed all the tree,
2906 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002907 */
2908 err = ext4_ext_get_access(handle, inode, path);
2909 if (err == 0) {
2910 ext_inode_hdr(inode)->eh_depth = 0;
2911 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002912 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002913 err = ext4_ext_dirty(handle, inode, path);
2914 }
2915 }
2916out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002917 ext4_ext_drop_refs(path);
2918 kfree(path);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002919 if (err == -EAGAIN) {
2920 path = NULL;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002921 goto again;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002922 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002923 ext4_journal_stop(handle);
2924
2925 return err;
2926}
2927
2928/*
2929 * called at mount time
2930 */
2931void ext4_ext_init(struct super_block *sb)
2932{
2933 /*
2934 * possible initialization would be here
2935 */
2936
Theodore Ts'o83982b62009-01-06 14:53:16 -05002937 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002938#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04002939 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002940#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04002941 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07002942#endif
2943#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04002944 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07002945#endif
2946#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04002947 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07002948#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04002949 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04002950#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07002951#ifdef EXTENTS_STATS
2952 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2953 EXT4_SB(sb)->s_ext_min = 1 << 30;
2954 EXT4_SB(sb)->s_ext_max = 0;
2955#endif
2956 }
2957}
2958
2959/*
2960 * called at umount time
2961 */
2962void ext4_ext_release(struct super_block *sb)
2963{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002964 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002965 return;
2966
2967#ifdef EXTENTS_STATS
2968 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2969 struct ext4_sb_info *sbi = EXT4_SB(sb);
2970 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2971 sbi->s_ext_blocks, sbi->s_ext_extents,
2972 sbi->s_ext_blocks / sbi->s_ext_extents);
2973 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2974 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2975 }
2976#endif
2977}
2978
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002979/* FIXME!! we need to try to merge to left or right after zero-out */
2980static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2981{
Lukas Czerner24075182010-10-27 21:30:06 -04002982 ext4_fsblk_t ee_pblock;
2983 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04002984 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002985
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002986 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04002987 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002988
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002989 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04002990 if (ret > 0)
2991 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002992
Lukas Czerner24075182010-10-27 21:30:06 -04002993 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002994}
2995
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002996/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002997 * ext4_split_extent_at() splits an extent at given block.
2998 *
2999 * @handle: the journal handle
3000 * @inode: the file inode
3001 * @path: the path to the extent
3002 * @split: the logical block where the extent is splitted.
3003 * @split_flags: indicates if the extent could be zeroout if split fails, and
3004 * the states(init or uninit) of new extents.
3005 * @flags: flags used to insert new extent to extent tree.
3006 *
3007 *
3008 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3009 * of which are deterimined by split_flag.
3010 *
3011 * There are two cases:
3012 * a> the extent are splitted into two extent.
3013 * b> split is not needed, and just mark the extent.
3014 *
3015 * return 0 on success.
3016 */
3017static int ext4_split_extent_at(handle_t *handle,
3018 struct inode *inode,
3019 struct ext4_ext_path *path,
3020 ext4_lblk_t split,
3021 int split_flag,
3022 int flags)
3023{
3024 ext4_fsblk_t newblock;
3025 ext4_lblk_t ee_block;
Zheng Liuadb23552013-03-10 21:13:05 -04003026 struct ext4_extent *ex, newex, orig_ex, zero_ex;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003027 struct ext4_extent *ex2 = NULL;
3028 unsigned int ee_len, depth;
3029 int err = 0;
3030
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003031 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3032 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3033
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003034 ext_debug("ext4_split_extents_at: inode %lu, logical"
3035 "block %llu\n", inode->i_ino, (unsigned long long)split);
3036
3037 ext4_ext_show_leaf(inode, path);
3038
3039 depth = ext_depth(inode);
3040 ex = path[depth].p_ext;
3041 ee_block = le32_to_cpu(ex->ee_block);
3042 ee_len = ext4_ext_get_actual_len(ex);
3043 newblock = split - ee_block + ext4_ext_pblock(ex);
3044
3045 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003046 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3047 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3048 EXT4_EXT_MARK_UNINIT1 |
3049 EXT4_EXT_MARK_UNINIT2));
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003050
3051 err = ext4_ext_get_access(handle, inode, path + depth);
3052 if (err)
3053 goto out;
3054
3055 if (split == ee_block) {
3056 /*
3057 * case b: block @split is the block that the extent begins with
3058 * then we just change the state of the extent, and splitting
3059 * is not needed.
3060 */
3061 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3062 ext4_ext_mark_uninitialized(ex);
3063 else
3064 ext4_ext_mark_initialized(ex);
3065
3066 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003067 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003068
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003069 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003070 goto out;
3071 }
3072
3073 /* case a */
3074 memcpy(&orig_ex, ex, sizeof(orig_ex));
3075 ex->ee_len = cpu_to_le16(split - ee_block);
3076 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3077 ext4_ext_mark_uninitialized(ex);
3078
3079 /*
3080 * path may lead to new leaf, not to original leaf any more
3081 * after ext4_ext_insert_extent() returns,
3082 */
3083 err = ext4_ext_dirty(handle, inode, path + depth);
3084 if (err)
3085 goto fix_extent_len;
3086
3087 ex2 = &newex;
3088 ex2->ee_block = cpu_to_le32(split);
3089 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3090 ext4_ext_store_pblock(ex2, newblock);
3091 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3092 ext4_ext_mark_uninitialized(ex2);
3093
3094 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3095 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003096 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
Zheng Liuadb23552013-03-10 21:13:05 -04003097 if (split_flag & EXT4_EXT_DATA_VALID1) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003098 err = ext4_ext_zeroout(inode, ex2);
Zheng Liuadb23552013-03-10 21:13:05 -04003099 zero_ex.ee_block = ex2->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003100 zero_ex.ee_len = cpu_to_le16(
3101 ext4_ext_get_actual_len(ex2));
Zheng Liuadb23552013-03-10 21:13:05 -04003102 ext4_ext_store_pblock(&zero_ex,
3103 ext4_ext_pblock(ex2));
3104 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003105 err = ext4_ext_zeroout(inode, ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003106 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003107 zero_ex.ee_len = cpu_to_le16(
3108 ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003109 ext4_ext_store_pblock(&zero_ex,
3110 ext4_ext_pblock(ex));
3111 }
3112 } else {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003113 err = ext4_ext_zeroout(inode, &orig_ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003114 zero_ex.ee_block = orig_ex.ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003115 zero_ex.ee_len = cpu_to_le16(
3116 ext4_ext_get_actual_len(&orig_ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003117 ext4_ext_store_pblock(&zero_ex,
3118 ext4_ext_pblock(&orig_ex));
3119 }
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003120
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003121 if (err)
3122 goto fix_extent_len;
3123 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003124 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003125 ext4_ext_try_to_merge(handle, inode, path, ex);
3126 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Zheng Liuadb23552013-03-10 21:13:05 -04003127 if (err)
3128 goto fix_extent_len;
3129
3130 /* update extent status tree */
3131 err = ext4_es_zeroout(inode, &zero_ex);
3132
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003133 goto out;
3134 } else if (err)
3135 goto fix_extent_len;
3136
3137out:
3138 ext4_ext_show_leaf(inode, path);
3139 return err;
3140
3141fix_extent_len:
3142 ex->ee_len = orig_ex.ee_len;
3143 ext4_ext_dirty(handle, inode, path + depth);
3144 return err;
3145}
3146
3147/*
3148 * ext4_split_extents() splits an extent and mark extent which is covered
3149 * by @map as split_flags indicates
3150 *
3151 * It may result in splitting the extent into multiple extents (upto three)
3152 * There are three possibilities:
3153 * a> There is no split required
3154 * b> Splits in two extents: Split is happening at either end of the extent
3155 * c> Splits in three extents: Somone is splitting in middle of the extent
3156 *
3157 */
3158static int ext4_split_extent(handle_t *handle,
3159 struct inode *inode,
3160 struct ext4_ext_path *path,
3161 struct ext4_map_blocks *map,
3162 int split_flag,
3163 int flags)
3164{
3165 ext4_lblk_t ee_block;
3166 struct ext4_extent *ex;
3167 unsigned int ee_len, depth;
3168 int err = 0;
3169 int uninitialized;
3170 int split_flag1, flags1;
Zheng Liu3a225672013-03-10 21:20:23 -04003171 int allocated = map->m_len;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003172
3173 depth = ext_depth(inode);
3174 ex = path[depth].p_ext;
3175 ee_block = le32_to_cpu(ex->ee_block);
3176 ee_len = ext4_ext_get_actual_len(ex);
3177 uninitialized = ext4_ext_is_uninitialized(ex);
3178
3179 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003180 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003181 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3182 if (uninitialized)
3183 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3184 EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003185 if (split_flag & EXT4_EXT_DATA_VALID2)
3186 split_flag1 |= EXT4_EXT_DATA_VALID1;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003187 err = ext4_split_extent_at(handle, inode, path,
3188 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003189 if (err)
3190 goto out;
Zheng Liu3a225672013-03-10 21:20:23 -04003191 } else {
3192 allocated = ee_len - (map->m_lblk - ee_block);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003193 }
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003194 /*
3195 * Update path is required because previous ext4_split_extent_at() may
3196 * result in split of original leaf or extent zeroout.
3197 */
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003198 ext4_ext_drop_refs(path);
3199 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3200 if (IS_ERR(path))
3201 return PTR_ERR(path);
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003202 depth = ext_depth(inode);
3203 ex = path[depth].p_ext;
3204 uninitialized = ext4_ext_is_uninitialized(ex);
3205 split_flag1 = 0;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003206
3207 if (map->m_lblk >= ee_block) {
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003208 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3209 if (uninitialized) {
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003210 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
Dmitry Monakhov357b66f2013-03-04 00:34:34 -05003211 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3212 EXT4_EXT_MARK_UNINIT2);
3213 }
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003214 err = ext4_split_extent_at(handle, inode, path,
3215 map->m_lblk, split_flag1, flags);
3216 if (err)
3217 goto out;
3218 }
3219
3220 ext4_ext_show_leaf(inode, path);
3221out:
Zheng Liu3a225672013-03-10 21:20:23 -04003222 return err ? err : allocated;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003223}
3224
Amit Arora56055d32007-07-17 21:42:38 -04003225/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003226 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04003227 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003228 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04003229 * uninitialized).
3230 * There are three possibilities:
3231 * a> There is no split required: Entire extent should be initialized
3232 * b> Splits in two extents: Write is happening at either end of the extent
3233 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003234 *
3235 * Pre-conditions:
3236 * - The extent pointed to by 'path' is uninitialized.
3237 * - The extent pointed to by 'path' contains a superset
3238 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3239 *
3240 * Post-conditions on success:
3241 * - the returned value is the number of blocks beyond map->l_lblk
3242 * that are allocated and initialized.
3243 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003244 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003245static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003246 struct inode *inode,
3247 struct ext4_map_blocks *map,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003248 struct ext4_ext_path *path,
3249 int flags)
Amit Arora56055d32007-07-17 21:42:38 -04003250{
Zheng Liu67a5da52012-08-17 09:54:17 -04003251 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003252 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003253 struct ext4_map_blocks split_map;
3254 struct ext4_extent zero_ex;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003255 struct ext4_extent *ex, *abut_ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003256 ext4_lblk_t ee_block, eof_block;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003257 unsigned int ee_len, depth, map_len = map->m_len;
3258 int allocated = 0, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003259 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003260 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003261
3262 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3263 "block %llu, max_blocks %u\n", inode->i_ino,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003264 (unsigned long long)map->m_lblk, map_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003265
Zheng Liu67a5da52012-08-17 09:54:17 -04003266 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003267 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3268 inode->i_sb->s_blocksize_bits;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003269 if (eof_block < map->m_lblk + map_len)
3270 eof_block = map->m_lblk + map_len;
Amit Arora56055d32007-07-17 21:42:38 -04003271
3272 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003273 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003274 ex = path[depth].p_ext;
3275 ee_block = le32_to_cpu(ex->ee_block);
3276 ee_len = ext4_ext_get_actual_len(ex);
Zheng Liuadb23552013-03-10 21:13:05 -04003277 zero_ex.ee_len = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003278
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003279 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3280
3281 /* Pre-conditions */
3282 BUG_ON(!ext4_ext_is_uninitialized(ex));
3283 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003284
3285 /*
3286 * Attempt to transfer newly initialized blocks from the currently
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003287 * uninitialized extent to its neighbor. This is much cheaper
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003288 * than an insertion followed by a merge as those involve costly
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003289 * memmove() calls. Transferring to the left is the common case in
3290 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3291 * followed by append writes.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003292 *
3293 * Limitations of the current logic:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003294 * - L1: we do not deal with writes covering the whole extent.
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003295 * This would require removing the extent if the transfer
3296 * is possible.
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003297 * - L2: we only attempt to merge with an extent stored in the
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003298 * same extent tree node.
3299 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003300 if ((map->m_lblk == ee_block) &&
3301 /* See if we can merge left */
3302 (map_len < ee_len) && /*L1*/
3303 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003304 ext4_lblk_t prev_lblk;
3305 ext4_fsblk_t prev_pblk, ee_pblk;
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003306 unsigned int prev_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003307
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003308 abut_ex = ex - 1;
3309 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3310 prev_len = ext4_ext_get_actual_len(abut_ex);
3311 prev_pblk = ext4_ext_pblock(abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003312 ee_pblk = ext4_ext_pblock(ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003313
3314 /*
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003315 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003316 * upon those conditions:
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003317 * - C1: abut_ex is initialized,
3318 * - C2: abut_ex is logically abutting ex,
3319 * - C3: abut_ex is physically abutting ex,
3320 * - C4: abut_ex can receive the additional blocks without
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003321 * overflowing the (initialized) length limit.
3322 */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003323 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003324 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3325 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003326 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003327 err = ext4_ext_get_access(handle, inode, path + depth);
3328 if (err)
3329 goto out;
3330
3331 trace_ext4_ext_convert_to_initialized_fastpath(inode,
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003332 map, ex, abut_ex);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003333
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003334 /* Shift the start of ex by 'map_len' blocks */
3335 ex->ee_block = cpu_to_le32(ee_block + map_len);
3336 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3337 ex->ee_len = cpu_to_le16(ee_len - map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003338 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3339
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003340 /* Extend abut_ex by 'map_len' blocks */
3341 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003342
3343 /* Result: number of initialized blocks past m_lblk */
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003344 allocated = map_len;
3345 }
3346 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3347 (map_len < ee_len) && /*L1*/
3348 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3349 /* See if we can merge right */
3350 ext4_lblk_t next_lblk;
3351 ext4_fsblk_t next_pblk, ee_pblk;
3352 unsigned int next_len;
3353
3354 abut_ex = ex + 1;
3355 next_lblk = le32_to_cpu(abut_ex->ee_block);
3356 next_len = ext4_ext_get_actual_len(abut_ex);
3357 next_pblk = ext4_ext_pblock(abut_ex);
3358 ee_pblk = ext4_ext_pblock(ex);
3359
3360 /*
3361 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3362 * upon those conditions:
3363 * - C1: abut_ex is initialized,
3364 * - C2: abut_ex is logically abutting ex,
3365 * - C3: abut_ex is physically abutting ex,
3366 * - C4: abut_ex can receive the additional blocks without
3367 * overflowing the (initialized) length limit.
3368 */
3369 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3370 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3371 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3372 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3373 err = ext4_ext_get_access(handle, inode, path + depth);
3374 if (err)
3375 goto out;
3376
3377 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3378 map, ex, abut_ex);
3379
3380 /* Shift the start of abut_ex by 'map_len' blocks */
3381 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3382 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3383 ex->ee_len = cpu_to_le16(ee_len - map_len);
3384 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3385
3386 /* Extend abut_ex by 'map_len' blocks */
3387 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3388
3389 /* Result: number of initialized blocks past m_lblk */
3390 allocated = map_len;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003391 }
3392 }
Lukas Czernerbc2d9db2013-04-03 23:33:27 -04003393 if (allocated) {
3394 /* Mark the block containing both extents as dirty */
3395 ext4_ext_dirty(handle, inode, path + depth);
3396
3397 /* Update path to point to the right extent */
3398 path[depth].p_ext = abut_ex;
3399 goto out;
3400 } else
3401 allocated = ee_len - (map->m_lblk - ee_block);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003402
Yongqiang Yang667eff32011-05-03 12:25:07 -04003403 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003404 /*
3405 * It is safe to convert extent to initialized via explicit
3406 * zeroout only if extent is fully insde i_size or new_size.
3407 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003408 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003409
Zheng Liu67a5da52012-08-17 09:54:17 -04003410 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3411 max_zeroout = sbi->s_extent_max_zeroout_kb >>
Lukas Czerner4f42f802013-03-12 12:40:04 -04003412 (inode->i_sb->s_blocksize_bits - 10);
Zheng Liu67a5da52012-08-17 09:54:17 -04003413
3414 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3415 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003416 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003417 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003418 goto out;
Zheng Liuadb23552013-03-10 21:13:05 -04003419 zero_ex.ee_block = ex->ee_block;
Zheng Liu8cde7ad2013-04-03 12:27:18 -04003420 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
Zheng Liuadb23552013-03-10 21:13:05 -04003421 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003422
3423 err = ext4_ext_get_access(handle, inode, path + depth);
3424 if (err)
3425 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003426 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003427 ext4_ext_try_to_merge(handle, inode, path, ex);
3428 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003429 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003430 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003431
Amit Arora56055d32007-07-17 21:42:38 -04003432 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003433 * four cases:
3434 * 1. split the extent into three extents.
3435 * 2. split the extent into two extents, zeroout the first half.
3436 * 3. split the extent into two extents, zeroout the second half.
3437 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003438 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003439 split_map.m_lblk = map->m_lblk;
3440 split_map.m_len = map->m_len;
3441
Zheng Liu67a5da52012-08-17 09:54:17 -04003442 if (max_zeroout && (allocated > map->m_len)) {
3443 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003444 /* case 3 */
3445 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003446 cpu_to_le32(map->m_lblk);
3447 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003448 ext4_ext_store_pblock(&zero_ex,
3449 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3450 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003451 if (err)
3452 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003453 split_map.m_lblk = map->m_lblk;
3454 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003455 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003456 /* case 2 */
3457 if (map->m_lblk != ee_block) {
3458 zero_ex.ee_block = ex->ee_block;
3459 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3460 ee_block);
3461 ext4_ext_store_pblock(&zero_ex,
3462 ext4_ext_pblock(ex));
3463 err = ext4_ext_zeroout(inode, &zero_ex);
3464 if (err)
3465 goto out;
3466 }
3467
Yongqiang Yang667eff32011-05-03 12:25:07 -04003468 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003469 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3470 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003471 }
3472 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003473
3474 allocated = ext4_split_extent(handle, inode, path,
Lukas Czerner27dd4382013-04-09 22:11:22 -04003475 &split_map, split_flag, flags);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003476 if (allocated < 0)
3477 err = allocated;
3478
Amit Arora56055d32007-07-17 21:42:38 -04003479out:
Zheng Liuadb23552013-03-10 21:13:05 -04003480 /* If we have gotten a failure, don't zero out status tree */
3481 if (!err)
3482 err = ext4_es_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003483 return err ? err : allocated;
3484}
3485
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003486/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003487 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003488 * ext4_get_blocks_dio_write() when DIO to write
3489 * to an uninitialized extent.
3490 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003491 * Writing to an uninitialized extent may result in splitting the uninitialized
Wang Sheng-Hui30cb27d2012-08-18 22:38:07 -04003492 * extent into multiple initialized/uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003493 * There are three possibilities:
3494 * a> There is no split required: Entire extent should be uninitialized
3495 * b> Splits in two extents: Write is happening at either end of the extent
3496 * c> Splits in three extents: Somone is writing in middle of the extent
3497 *
3498 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003499 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003500 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003501 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003502 * into three uninitialized extent(at most). After IO complete, the part
3503 * being filled will be convert to initialized by the end_io callback function
3504 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003505 *
3506 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003507 */
3508static int ext4_split_unwritten_extents(handle_t *handle,
3509 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003510 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003511 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003512 int flags)
3513{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003514 ext4_lblk_t eof_block;
3515 ext4_lblk_t ee_block;
3516 struct ext4_extent *ex;
3517 unsigned int ee_len;
3518 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003519
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003520 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3521 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003522 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003523
3524 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3525 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003526 if (eof_block < map->m_lblk + map->m_len)
3527 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003528 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003529 * It is safe to convert extent to initialized via explicit
3530 * zeroout only if extent is fully insde i_size or new_size.
3531 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003532 depth = ext_depth(inode);
3533 ex = path[depth].p_ext;
3534 ee_block = le32_to_cpu(ex->ee_block);
3535 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003536
Yongqiang Yang667eff32011-05-03 12:25:07 -04003537 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3538 split_flag |= EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003539 if (flags & EXT4_GET_BLOCKS_CONVERT)
3540 split_flag |= EXT4_EXT_DATA_VALID2;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003541 flags |= EXT4_GET_BLOCKS_PRE_IO;
3542 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003543}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003544
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003545static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003546 struct inode *inode,
3547 struct ext4_map_blocks *map,
3548 struct ext4_ext_path *path)
Mingming Cao00314622009-09-28 15:49:08 -04003549{
3550 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003551 ext4_lblk_t ee_block;
3552 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003553 int depth;
3554 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003555
3556 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003557 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003558 ee_block = le32_to_cpu(ex->ee_block);
3559 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003560
Yongqiang Yang197217a2011-05-03 11:45:29 -04003561 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3562 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003563 (unsigned long long)ee_block, ee_len);
3564
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003565 /* If extent is larger than requested it is a clear sign that we still
3566 * have some extent state machine issues left. So extent_split is still
3567 * required.
3568 * TODO: Once all related issues will be fixed this situation should be
3569 * illegal.
3570 */
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003571 if (ee_block != map->m_lblk || ee_len > map->m_len) {
Dmitry Monakhovff95ec22013-03-04 00:41:05 -05003572#ifdef EXT4_DEBUG
3573 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3574 " len %u; IO logical block %llu, len %u\n",
3575 inode->i_ino, (unsigned long long)ee_block, ee_len,
3576 (unsigned long long)map->m_lblk, map->m_len);
3577#endif
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003578 err = ext4_split_unwritten_extents(handle, inode, map, path,
3579 EXT4_GET_BLOCKS_CONVERT);
3580 if (err < 0)
3581 goto out;
3582 ext4_ext_drop_refs(path);
3583 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3584 if (IS_ERR(path)) {
3585 err = PTR_ERR(path);
3586 goto out;
3587 }
3588 depth = ext_depth(inode);
3589 ex = path[depth].p_ext;
3590 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003591
Mingming Cao00314622009-09-28 15:49:08 -04003592 err = ext4_ext_get_access(handle, inode, path + depth);
3593 if (err)
3594 goto out;
3595 /* first mark the extent as initialized */
3596 ext4_ext_mark_initialized(ex);
3597
Yongqiang Yang197217a2011-05-03 11:45:29 -04003598 /* note: ext4_ext_correct_indexes() isn't needed here because
3599 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003600 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003601 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003602
Mingming Cao00314622009-09-28 15:49:08 -04003603 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003604 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003605out:
3606 ext4_ext_show_leaf(inode, path);
3607 return err;
3608}
3609
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003610static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3611 sector_t block, int count)
3612{
3613 int i;
3614 for (i = 0; i < count; i++)
3615 unmap_underlying_metadata(bdev, block + i);
3616}
3617
Theodore Ts'o58590b02010-10-27 21:23:12 -04003618/*
3619 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3620 */
3621static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003622 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003623 struct ext4_ext_path *path,
3624 unsigned int len)
3625{
3626 int i, depth;
3627 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003628 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003629
3630 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3631 return 0;
3632
3633 depth = ext_depth(inode);
3634 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003635
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003636 /*
3637 * We're going to remove EOFBLOCKS_FL entirely in future so we
3638 * do not care for this case anymore. Simply remove the flag
3639 * if there are no extents.
3640 */
3641 if (unlikely(!eh->eh_entries))
3642 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003643 last_ex = EXT_LAST_EXTENT(eh);
3644 /*
3645 * We should clear the EOFBLOCKS_FL flag if we are writing the
3646 * last block in the last extent in the file. We test this by
3647 * first checking to see if the caller to
3648 * ext4_ext_get_blocks() was interested in the last block (or
3649 * a block beyond the last block) in the current extent. If
3650 * this turns out to be false, we can bail out from this
3651 * function immediately.
3652 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003653 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003654 ext4_ext_get_actual_len(last_ex))
3655 return 0;
3656 /*
3657 * If the caller does appear to be planning to write at or
3658 * beyond the end of the current extent, we then test to see
3659 * if the current extent is the last extent in the file, by
3660 * checking to make sure it was reached via the rightmost node
3661 * at each level of the tree.
3662 */
3663 for (i = depth-1; i >= 0; i--)
3664 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3665 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003666out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003667 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3668 return ext4_mark_inode_dirty(handle, inode);
3669}
3670
Aditya Kali7b415bf2011-09-09 19:04:51 -04003671/**
3672 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3673 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003674 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003675 */
Zheng Liuf7fec032013-02-18 00:28:47 -05003676int ext4_find_delalloc_range(struct inode *inode,
3677 ext4_lblk_t lblk_start,
3678 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003679{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003680 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003681
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04003682 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
Zheng Liu06b0c882013-02-18 00:26:51 -05003683 if (es.es_len == 0)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003684 return 0; /* there is no delay extent in this tree */
Zheng Liu06b0c882013-02-18 00:26:51 -05003685 else if (es.es_lblk <= lblk_start &&
3686 lblk_start < es.es_lblk + es.es_len)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003687 return 1;
Zheng Liu06b0c882013-02-18 00:26:51 -05003688 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003689 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003690 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003691 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003692}
3693
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003694int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003695{
3696 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3697 ext4_lblk_t lblk_start, lblk_end;
Theodore Ts'ocd744272013-12-20 09:29:35 -05003698 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003699 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3700
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003701 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003702}
3703
3704/**
3705 * Determines how many complete clusters (out of those specified by the 'map')
3706 * are under delalloc and were reserved quota for.
3707 * This function is called when we are writing out the blocks that were
3708 * originally written with their allocation delayed, but then the space was
3709 * allocated using fallocate() before the delayed allocation could be resolved.
3710 * The cases to look for are:
3711 * ('=' indicated delayed allocated blocks
3712 * '-' indicates non-delayed allocated blocks)
3713 * (a) partial clusters towards beginning and/or end outside of allocated range
3714 * are not delalloc'ed.
3715 * Ex:
3716 * |----c---=|====c====|====c====|===-c----|
3717 * |++++++ allocated ++++++|
3718 * ==> 4 complete clusters in above example
3719 *
3720 * (b) partial cluster (outside of allocated range) towards either end is
3721 * marked for delayed allocation. In this case, we will exclude that
3722 * cluster.
3723 * Ex:
3724 * |----====c========|========c========|
3725 * |++++++ allocated ++++++|
3726 * ==> 1 complete clusters in above example
3727 *
3728 * Ex:
3729 * |================c================|
3730 * |++++++ allocated ++++++|
3731 * ==> 0 complete clusters in above example
3732 *
3733 * The ext4_da_update_reserve_space will be called only if we
3734 * determine here that there were some "entire" clusters that span
3735 * this 'allocated' range.
3736 * In the non-bigalloc case, this function will just end up returning num_blks
3737 * without ever calling ext4_find_delalloc_range.
3738 */
3739static unsigned int
3740get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3741 unsigned int num_blks)
3742{
3743 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3744 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3745 ext4_lblk_t lblk_from, lblk_to, c_offset;
3746 unsigned int allocated_clusters = 0;
3747
3748 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3749 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3750
3751 /* max possible clusters for this allocation */
3752 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3753
Aditya Kalid8990242011-09-09 19:18:51 -04003754 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3755
Aditya Kali7b415bf2011-09-09 19:04:51 -04003756 /* Check towards left side */
Theodore Ts'ocd744272013-12-20 09:29:35 -05003757 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003758 if (c_offset) {
Theodore Ts'ocd744272013-12-20 09:29:35 -05003759 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003760 lblk_to = lblk_from + c_offset - 1;
3761
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003762 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003763 allocated_clusters--;
3764 }
3765
3766 /* Now check towards right. */
Theodore Ts'ocd744272013-12-20 09:29:35 -05003767 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003768 if (allocated_clusters && c_offset) {
3769 lblk_from = lblk_start + num_blks;
3770 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3771
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003772 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003773 allocated_clusters--;
3774 }
3775
3776 return allocated_clusters;
3777}
3778
Mingming Cao00314622009-09-28 15:49:08 -04003779static int
3780ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003781 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003782 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003783 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003784{
3785 int ret = 0;
3786 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003787 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003788
Zheng Liu88635ca2011-12-28 19:00:25 -05003789 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3790 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003791 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003792 flags, allocated);
3793 ext4_ext_show_leaf(inode, path);
3794
Lukas Czerner27dd4382013-04-09 22:11:22 -04003795 /*
3796 * When writing into uninitialized space, we should not fail to
3797 * allocate metadata blocks for the new extent block if needed.
3798 */
3799 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3800
Zheng Liub5645532012-11-08 14:33:43 -05003801 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3802 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04003803
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003804 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003805 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003806 ret = ext4_split_unwritten_extents(handle, inode, map,
3807 path, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04003808 if (ret <= 0)
3809 goto out;
Mingming5f5249502009-11-10 10:48:04 -05003810 /*
3811 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003812 * that this IO needs to conversion to written when IO is
Mingming5f5249502009-11-10 10:48:04 -05003813 * completed
3814 */
Tao Ma0edeb712011-10-31 17:30:44 -04003815 if (io)
3816 ext4_set_io_unwritten_flag(inode, io);
3817 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003818 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Zheng Liua25a4e12013-02-18 00:28:04 -05003819 map->m_flags |= EXT4_MAP_UNWRITTEN;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003820 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003821 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003822 goto out;
3823 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003824 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003825 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003826 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Mingming Cao00314622009-09-28 15:49:08 -04003827 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003828 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003829 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003830 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3831 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003832 } else
3833 err = ret;
Zheng Liucdee7842013-03-10 21:08:52 -04003834 map->m_flags |= EXT4_MAP_MAPPED;
3835 if (allocated > map->m_len)
3836 allocated = map->m_len;
3837 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003838 goto out2;
3839 }
3840 /* buffered IO case */
3841 /*
3842 * repeat fallocate creation request
3843 * we already have an unwritten extent
3844 */
Zheng Liua25a4e12013-02-18 00:28:04 -05003845 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3846 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003847 goto map_out;
Zheng Liua25a4e12013-02-18 00:28:04 -05003848 }
Mingming Cao00314622009-09-28 15:49:08 -04003849
3850 /* buffered READ or buffered write_begin() lookup */
3851 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3852 /*
3853 * We have blocks reserved already. We
3854 * return allocated blocks so that delalloc
3855 * won't do block reservation for us. But
3856 * the buffer head will be unmapped so that
3857 * a read from the block returns 0s.
3858 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003859 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003860 goto out1;
3861 }
3862
3863 /* buffered write, writepage time, convert*/
Lukas Czerner27dd4382013-04-09 22:11:22 -04003864 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003865 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05003866 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04003867out:
3868 if (ret <= 0) {
3869 err = ret;
3870 goto out2;
3871 } else
3872 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003873 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003874 /*
3875 * if we allocated more blocks than requested
3876 * we need to make sure we unmap the extra block
3877 * allocated. The actual needed block will get
3878 * unmapped later when we find the buffer_head marked
3879 * new.
3880 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003881 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003882 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003883 newblock + map->m_len,
3884 allocated - map->m_len);
3885 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003886 }
Zheng Liu3a225672013-03-10 21:20:23 -04003887 map->m_len = allocated;
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003888
3889 /*
3890 * If we have done fallocate with the offset that is already
3891 * delayed allocated, we would have block reservation
3892 * and quota reservation done in the delayed write path.
3893 * But fallocate would have already updated quota and block
3894 * count for this offset. So cancel these reservation
3895 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003896 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3897 unsigned int reserved_clusters;
3898 reserved_clusters = get_reserved_cluster_alloc(inode,
3899 map->m_lblk, map->m_len);
3900 if (reserved_clusters)
3901 ext4_da_update_reserve_space(inode,
3902 reserved_clusters,
3903 0);
3904 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003905
Mingming Cao00314622009-09-28 15:49:08 -04003906map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003907 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003908 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3909 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3910 map->m_len);
3911 if (err < 0)
3912 goto out2;
3913 }
Mingming Cao00314622009-09-28 15:49:08 -04003914out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003915 if (allocated > map->m_len)
3916 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003917 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003918 map->m_pblk = newblock;
3919 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003920out2:
3921 if (path) {
3922 ext4_ext_drop_refs(path);
3923 kfree(path);
3924 }
3925 return err ? err : allocated;
3926}
Theodore Ts'o58590b02010-10-27 21:23:12 -04003927
Mingming Cao00314622009-09-28 15:49:08 -04003928/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003929 * get_implied_cluster_alloc - check to see if the requested
3930 * allocation (in the map structure) overlaps with a cluster already
3931 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04003932 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003933 * @map The requested lblk->pblk mapping
3934 * @ex The extent structure which might contain an implied
3935 * cluster allocation
3936 *
3937 * This function is called by ext4_ext_map_blocks() after we failed to
3938 * find blocks that were already in the inode's extent tree. Hence,
3939 * we know that the beginning of the requested region cannot overlap
3940 * the extent from the inode's extent tree. There are three cases we
3941 * want to catch. The first is this case:
3942 *
3943 * |--- cluster # N--|
3944 * |--- extent ---| |---- requested region ---|
3945 * |==========|
3946 *
3947 * The second case that we need to test for is this one:
3948 *
3949 * |--------- cluster # N ----------------|
3950 * |--- requested region --| |------- extent ----|
3951 * |=======================|
3952 *
3953 * The third case is when the requested region lies between two extents
3954 * within the same cluster:
3955 * |------------- cluster # N-------------|
3956 * |----- ex -----| |---- ex_right ----|
3957 * |------ requested region ------|
3958 * |================|
3959 *
3960 * In each of the above cases, we need to set the map->m_pblk and
3961 * map->m_len so it corresponds to the return the extent labelled as
3962 * "|====|" from cluster #N, since it is already in use for data in
3963 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3964 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3965 * as a new "allocated" block region. Otherwise, we will return 0 and
3966 * ext4_ext_map_blocks() will then allocate one or more new clusters
3967 * by calling ext4_mb_new_blocks().
3968 */
Aditya Kalid8990242011-09-09 19:18:51 -04003969static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003970 struct ext4_map_blocks *map,
3971 struct ext4_extent *ex,
3972 struct ext4_ext_path *path)
3973{
Aditya Kalid8990242011-09-09 19:18:51 -04003974 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'ocd744272013-12-20 09:29:35 -05003975 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003976 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05003977 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003978 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3979 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3980 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3981
3982 /* The extent passed in that we are trying to match */
3983 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3984 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3985
3986 /* The requested region passed into ext4_map_blocks() */
3987 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003988
3989 if ((rr_cluster_start == ex_cluster_end) ||
3990 (rr_cluster_start == ex_cluster_start)) {
3991 if (rr_cluster_start == ex_cluster_end)
3992 ee_start += ee_len - 1;
Theodore Ts'ocd744272013-12-20 09:29:35 -05003993 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003994 map->m_len = min(map->m_len,
3995 (unsigned) sbi->s_cluster_ratio - c_offset);
3996 /*
3997 * Check for and handle this case:
3998 *
3999 * |--------- cluster # N-------------|
4000 * |------- extent ----|
4001 * |--- requested region ---|
4002 * |===========|
4003 */
4004
4005 if (map->m_lblk < ee_block)
4006 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4007
4008 /*
4009 * Check for the case where there is already another allocated
4010 * block to the right of 'ex' but before the end of the cluster.
4011 *
4012 * |------------- cluster # N-------------|
4013 * |----- ex -----| |---- ex_right ----|
4014 * |------ requested region ------|
4015 * |================|
4016 */
4017 if (map->m_lblk > ee_block) {
4018 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4019 map->m_len = min(map->m_len, next - map->m_lblk);
4020 }
Aditya Kalid8990242011-09-09 19:18:51 -04004021
4022 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004023 return 1;
4024 }
Aditya Kalid8990242011-09-09 19:18:51 -04004025
4026 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004027 return 0;
4028}
4029
4030
4031/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05004032 * Block allocation/map/preallocation routine for extents based files
4033 *
4034 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004035 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004036 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4037 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05004038 *
4039 * return > 0, number of of blocks already mapped/allocated
4040 * if create == 0 and these are pre-allocated blocks
4041 * buffer head is unmapped
4042 * otherwise blocks are mapped
4043 *
4044 * return = 0, if plain look up failed (blocks have not been allocated)
4045 * buffer head is unmapped
4046 *
4047 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05004048 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004049int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4050 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07004051{
4052 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004053 struct ext4_extent newex, *ex, *ex2;
4054 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004055 ext4_fsblk_t newblock = 0;
Eric Whitneycbe099e2014-02-19 18:52:39 -05004056 int free_on_err = 0, err = 0, depth, ret;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004057 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004058 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05004059 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04004060 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004061 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004062 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07004063
Mingming84fe3be2009-09-01 08:44:37 -04004064 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004065 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004066 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004067
Alex Tomasa86c6182006-10-11 01:21:03 -07004068 /* find extent for this block */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004069 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07004070 if (IS_ERR(path)) {
4071 err = PTR_ERR(path);
4072 path = NULL;
4073 goto out2;
4074 }
4075
4076 depth = ext_depth(inode);
4077
4078 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004079 * consistent leaf must not be empty;
4080 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07004081 * this is why assert can't be put in ext4_ext_find_extent()
4082 */
Frank Mayhar273df552010-03-02 11:46:09 -05004083 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4084 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04004085 "lblock: %lu, depth: %d pblock %lld",
4086 (unsigned long) map->m_lblk, depth,
4087 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05004088 err = -EIO;
4089 goto out2;
4090 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004091
Avantika Mathur7e028972006-12-06 20:41:33 -08004092 ex = path[depth].p_ext;
4093 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004094 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04004095 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004096 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004097
4098 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004099 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04004100 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07004101 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004102 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04004103
4104 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4105
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004106 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004107 if (in_range(map->m_lblk, ee_block, ee_len)) {
4108 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004109 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004110 allocated = ee_len - (map->m_lblk - ee_block);
4111 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4112 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04004113
Zheng Liu69eb33d2013-02-18 00:31:07 -05004114 if (!ext4_ext_is_uninitialized(ex))
Lukas Czerner78771912012-03-19 23:05:43 -04004115 goto out;
Zheng Liu69eb33d2013-02-18 00:31:07 -05004116
Eric Whitneycbe099e2014-02-19 18:52:39 -05004117 ret = ext4_ext_handle_uninitialized_extents(
Lukas Czerner78771912012-03-19 23:05:43 -04004118 handle, inode, map, path, flags,
4119 allocated, newblock);
Eric Whitneycbe099e2014-02-19 18:52:39 -05004120 if (ret < 0)
4121 err = ret;
4122 else
4123 allocated = ret;
Zheng Liu37794732012-11-08 14:47:52 -05004124 goto out3;
Alex Tomasa86c6182006-10-11 01:21:03 -07004125 }
4126 }
4127
Aditya Kali7b415bf2011-09-09 19:04:51 -04004128 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05004129 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04004130 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4131
Alex Tomasa86c6182006-10-11 01:21:03 -07004132 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004133 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07004134 * we couldn't try to create block if create flag is zero
4135 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004136 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004137 /*
4138 * put just found gap into cache to speed up
4139 * subsequent requests
4140 */
Zheng Liud100eef2013-02-18 00:29:59 -05004141 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4142 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004143 goto out2;
4144 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004145
Alex Tomasa86c6182006-10-11 01:21:03 -07004146 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004147 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004148 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004149 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004150 newex.ee_block = cpu_to_le32(map->m_lblk);
Eric Whitney70139b62014-01-06 14:00:23 -05004151 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004152
4153 /*
4154 * If we are doing bigalloc, check to see if the extent returned
4155 * by ext4_ext_find_extent() implies a cluster we can use.
4156 */
4157 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004158 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004159 ar.len = allocated = map->m_len;
4160 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004161 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004162 goto got_allocated_blocks;
4163 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004164
Alex Tomasc9de5602008-01-29 00:19:52 -05004165 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004166 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004167 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4168 if (err)
4169 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004170 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004171 ex2 = NULL;
4172 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004173 if (err)
4174 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004175
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004176 /* Check if the extent after searching to the right implies a
4177 * cluster we can use. */
4178 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004179 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004180 ar.len = allocated = map->m_len;
4181 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004182 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004183 goto got_allocated_blocks;
4184 }
4185
Amit Arora749269f2007-07-18 09:02:56 -04004186 /*
4187 * See if request is beyond maximum number of blocks we can have in
4188 * a single extent. For an initialized extent this limit is
4189 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4190 * EXT_UNINIT_MAX_LEN.
4191 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004192 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004193 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004194 map->m_len = EXT_INIT_MAX_LEN;
4195 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004196 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004197 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004198
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004199 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004200 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004201 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004202 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004203 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004204 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004205 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004206
4207 /* allocate new block */
4208 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004209 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4210 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004211 /*
4212 * We calculate the offset from the beginning of the cluster
4213 * for the logical block number, since when we allocate a
4214 * physical cluster, the physical block should start at the
4215 * same offset from the beginning of the cluster. This is
4216 * needed so that future calls to get_implied_cluster_alloc()
4217 * work correctly.
4218 */
Theodore Ts'ocd744272013-12-20 09:29:35 -05004219 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004220 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4221 ar.goal -= offset;
4222 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004223 if (S_ISREG(inode->i_mode))
4224 ar.flags = EXT4_MB_HINT_DATA;
4225 else
4226 /* disable in-core preallocation for non-regular files */
4227 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004228 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4229 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004230 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004231 if (!newblock)
4232 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004233 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004234 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004235 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004236 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004237 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4238 if (ar.len > allocated)
4239 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004240
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004241got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004242 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004243 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004244 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004245 /* Mark uninitialized */
4246 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04004247 ext4_ext_mark_uninitialized(&newex);
Zheng Liua25a4e12013-02-18 00:28:04 -05004248 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004249 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004250 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004251 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004252 * here we flag the IO that really needs the conversion.
Mingming5f5249502009-11-10 10:48:04 -05004253 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004254 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004255 */
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004256 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4257 set_unwritten = 1;
Jiaying Zhang744692d2010-03-04 16:14:02 -05004258 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004259 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004260 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004261
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004262 err = 0;
4263 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4264 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4265 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004266 if (!err)
4267 err = ext4_ext_insert_extent(handle, inode, path,
4268 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004269
4270 if (!err && set_unwritten) {
4271 if (io)
4272 ext4_set_io_unwritten_flag(inode, io);
4273 else
4274 ext4_set_inode_state(inode,
4275 EXT4_STATE_DIO_UNWRITTEN);
4276 }
4277
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004278 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004279 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4280 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004281 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004282 /* not a good idea to call discard here directly,
4283 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004284 ext4_discard_preallocations(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004285 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
Maxim Patlasov7132de72011-07-10 19:37:48 -04004286 ext4_ext_get_actual_len(&newex), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004287 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004288 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004289
Alex Tomasa86c6182006-10-11 01:21:03 -07004290 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004291 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004292 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004293 if (allocated > map->m_len)
4294 allocated = map->m_len;
4295 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004296
Jan Karab436b9b2009-12-08 23:51:10 -05004297 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004298 * Update reserved blocks/metadata blocks after successful
4299 * block allocation which had been deferred till now.
4300 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004301 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004302 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004303 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004304 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004305 */
4306 reserved_clusters = get_reserved_cluster_alloc(inode,
4307 map->m_lblk, allocated);
4308 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4309 if (reserved_clusters) {
4310 /*
4311 * We have clusters reserved for this range.
4312 * But since we are not doing actual allocation
4313 * and are simply using blocks from previously
4314 * allocated cluster, we should release the
4315 * reservation and not claim quota.
4316 */
4317 ext4_da_update_reserve_space(inode,
4318 reserved_clusters, 0);
4319 }
4320 } else {
4321 BUG_ON(allocated_clusters < reserved_clusters);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004322 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f2612011-09-09 19:20:51 -04004323 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004324 int reservation = allocated_clusters -
4325 reserved_clusters;
4326 /*
4327 * It seems we claimed few clusters outside of
4328 * the range of this allocation. We should give
4329 * it back to the reservation pool. This can
4330 * happen in the following case:
4331 *
4332 * * Suppose s_cluster_ratio is 4 (i.e., each
4333 * cluster has 4 blocks. Thus, the clusters
4334 * are [0-3],[4-7],[8-11]...
4335 * * First comes delayed allocation write for
4336 * logical blocks 10 & 11. Since there were no
4337 * previous delayed allocated blocks in the
4338 * range [8-11], we would reserve 1 cluster
4339 * for this write.
4340 * * Next comes write for logical blocks 3 to 8.
4341 * In this case, we will reserve 2 clusters
4342 * (for [0-3] and [4-7]; and not for [8-11] as
4343 * that range has a delayed allocated blocks.
4344 * Thus total reserved clusters now becomes 3.
4345 * * Now, during the delayed allocation writeout
4346 * time, we will first write blocks [3-8] and
4347 * allocate 3 clusters for writing these
4348 * blocks. Also, we would claim all these
4349 * three clusters above.
4350 * * Now when we come here to writeout the
4351 * blocks [10-11], we would expect to claim
4352 * the reservation of 1 cluster we had made
4353 * (and we would claim it since there are no
4354 * more delayed allocated blocks in the range
4355 * [8-11]. But our reserved cluster count had
4356 * already gone to 0.
4357 *
4358 * Thus, at the step 4 above when we determine
4359 * that there are still some unwritten delayed
4360 * allocated blocks outside of our current
4361 * block range, we should increment the
4362 * reserved clusters count so that when the
4363 * remaining blocks finally gets written, we
4364 * could claim them.
4365 */
Aditya Kali5356f2612011-09-09 19:20:51 -04004366 dquot_reserve_block(inode,
4367 EXT4_C2B(sbi, reservation));
4368 spin_lock(&ei->i_block_reservation_lock);
4369 ei->i_reserved_data_blocks += reservation;
4370 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004371 }
Lukas Czerner232ec872013-03-10 22:46:30 -04004372 /*
4373 * We will claim quota for all newly allocated blocks.
4374 * We're updating the reserved space *after* the
4375 * correction above so we do not accidentally free
4376 * all the metadata reservation because we might
4377 * actually need it later on.
4378 */
4379 ext4_da_update_reserve_space(inode, allocated_clusters,
4380 1);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004381 }
4382 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004383
4384 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004385 * Cache the extent and update transaction to commit on fdatasync only
4386 * when it is _not_ an uninitialized extent.
4387 */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004388 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
Jan Karab436b9b2009-12-08 23:51:10 -05004389 ext4_update_inode_fsync_trans(handle, inode, 1);
Zheng Liu69eb33d2013-02-18 00:31:07 -05004390 else
Jan Karab436b9b2009-12-08 23:51:10 -05004391 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004392out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004393 if (allocated > map->m_len)
4394 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004395 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004396 map->m_flags |= EXT4_MAP_MAPPED;
4397 map->m_pblk = newblock;
4398 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004399out2:
4400 if (path) {
4401 ext4_ext_drop_refs(path);
4402 kfree(path);
4403 }
Allison Hendersone8613042011-05-25 07:41:46 -04004404
Zheng Liu37794732012-11-08 14:47:52 -05004405out3:
Zheng Liu19b303d2012-11-08 14:34:04 -05004406 trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated);
Yongqiang Yange7b319e2011-10-29 09:39:51 -04004407
Lukas Czerner78771912012-03-19 23:05:43 -04004408 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004409}
4410
Theodore Ts'o819c4922013-04-03 12:47:17 -04004411void ext4_ext_truncate(handle_t *handle, struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004412{
Alex Tomasa86c6182006-10-11 01:21:03 -07004413 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004414 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004415 int err = 0;
4416
4417 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004418 * TODO: optimization is possible here.
4419 * Probably we need not scan at all,
4420 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004421 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004422
4423 /* we have to know where to truncate from in crash case */
4424 EXT4_I(inode)->i_disksize = inode->i_size;
4425 ext4_mark_inode_dirty(handle, inode);
4426
4427 last_block = (inode->i_size + sb->s_blocksize - 1)
4428 >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'oc07ae682013-07-15 00:09:19 -04004429retry:
Zheng Liu51865fd2012-11-08 21:57:32 -05004430 err = ext4_es_remove_extent(inode, last_block,
4431 EXT_MAX_BLOCKS - last_block);
Theodore Ts'oc7039e92013-07-29 12:12:56 -04004432 if (err == -ENOMEM) {
Theodore Ts'oc07ae682013-07-15 00:09:19 -04004433 cond_resched();
4434 congestion_wait(BLK_RW_ASYNC, HZ/50);
4435 goto retry;
4436 }
4437 if (err) {
4438 ext4_std_error(inode->i_sb, err);
4439 return;
4440 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04004441 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Theodore Ts'oc07ae682013-07-15 00:09:19 -04004442 ext4_std_error(inode->i_sb, err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004443}
4444
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004445static void ext4_falloc_update_inode(struct inode *inode,
4446 int mode, loff_t new_size, int update_ctime)
4447{
4448 struct timespec now;
4449
4450 if (update_ctime) {
4451 now = current_fs_time(inode->i_sb);
4452 if (!timespec_equal(&inode->i_ctime, &now))
4453 inode->i_ctime = now;
4454 }
4455 /*
4456 * Update only when preallocation was requested beyond
4457 * the file size.
4458 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04004459 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4460 if (new_size > i_size_read(inode))
4461 i_size_write(inode, new_size);
4462 if (new_size > EXT4_I(inode)->i_disksize)
4463 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004464 } else {
4465 /*
4466 * Mark that we allocate beyond EOF so the subsequent truncate
4467 * can proceed even if the new size is the same as i_size.
4468 */
4469 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004470 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004471 }
4472
4473}
4474
Amit Aroraa2df2a62007-07-17 21:42:41 -04004475/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004476 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04004477 * operation, which gets called from sys_fallocate system call.
4478 * For block-mapped files, posix_fallocate should fall back to the method
4479 * of writing zeroes to the required new blocks (the same behavior which is
4480 * expected for file systems which do not support fallocate() system call).
4481 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004482long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004483{
Al Viro496ad9a2013-01-23 17:07:38 -05004484 struct inode *inode = file_inode(file);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004485 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004486 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004487 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004488 int ret = 0;
4489 int ret2 = 0;
4490 int retries = 0;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004491 int flags;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004492 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004493 unsigned int credits, blkbits = inode->i_blkbits;
4494
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004495 /* Return error if mode is not supported */
4496 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4497 return -EOPNOTSUPP;
4498
4499 if (mode & FALLOC_FL_PUNCH_HOLE)
4500 return ext4_punch_hole(file, offset, len);
4501
Tao Ma0c8d4142012-12-10 14:06:03 -05004502 ret = ext4_convert_inline_data(inode);
4503 if (ret)
4504 return ret;
4505
Zheng Liu8bad6fc2013-01-28 09:21:37 -05004506 /*
4507 * currently supporting (pre)allocate mode for extent-based
4508 * files _only_
4509 */
4510 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4511 return -EOPNOTSUPP;
4512
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004513 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004514 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004515 /*
4516 * We can't just convert len to max_blocks because
4517 * If blocksize = 4096 offset = 3072 and len = 2048
4518 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004519 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004520 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004521 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04004522 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04004523 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004524 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004525 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004526 ret = inode_newsize_ok(inode, (len + offset));
4527 if (ret) {
4528 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004529 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004530 return ret;
4531 }
Greg Harm3c6fe772011-10-31 18:41:47 -04004532 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004533 if (mode & FALLOC_FL_KEEP_SIZE)
4534 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
Greg Harm3c6fe772011-10-31 18:41:47 -04004535 /*
4536 * Don't normalize the request if it can fit in one extent so
4537 * that it doesn't get unnecessarily split into multiple
4538 * extents.
4539 */
4540 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4541 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004542
Amit Aroraa2df2a62007-07-17 21:42:41 -04004543retry:
4544 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004545 map.m_lblk = map.m_lblk + ret;
4546 map.m_len = max_blocks = max_blocks - ret;
Theodore Ts'o9924a922013-02-08 21:59:22 -05004547 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4548 credits);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004549 if (IS_ERR(handle)) {
4550 ret = PTR_ERR(handle);
4551 break;
4552 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004553 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004554 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004555#ifdef EXT4FS_DEBUG
Lukas Czernerb06acd32013-01-28 21:21:12 -05004556 ext4_warning(inode->i_sb,
4557 "inode #%lu: block %u: len %u: "
4558 "ext4_ext_map_blocks returned %d",
4559 inode->i_ino, map.m_lblk,
4560 map.m_len, ret);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004561#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04004562 ext4_mark_inode_dirty(handle, inode);
4563 ret2 = ext4_journal_stop(handle);
4564 break;
4565 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004566 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004567 blkbits) >> blkbits))
4568 new_size = offset + len;
4569 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04004570 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004571
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004572 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004573 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04004574 ext4_mark_inode_dirty(handle, inode);
Zheng Liuf4e95b32012-06-30 19:12:57 -04004575 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4576 ext4_handle_sync(handle);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004577 ret2 = ext4_journal_stop(handle);
4578 if (ret2)
4579 break;
4580 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004581 if (ret == -ENOSPC &&
4582 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4583 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004584 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004585 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004586 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004587 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4588 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004589 return ret > 0 ? ret2 : ret;
4590}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004591
4592/*
Mingming Cao00314622009-09-28 15:49:08 -04004593 * This function convert a range of blocks to written extents
4594 * The caller of this function will pass the start offset and the size.
4595 * all unwritten extents within this range will be converted to
4596 * written extents.
4597 *
4598 * This function is called from the direct IO end io call back
4599 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004600 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004601 */
4602int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
Eric Sandeena1de02d2010-02-04 23:58:38 -05004603 ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004604{
4605 handle_t *handle;
Mingming Cao00314622009-09-28 15:49:08 -04004606 unsigned int max_blocks;
4607 int ret = 0;
4608 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004609 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004610 unsigned int credits, blkbits = inode->i_blkbits;
4611
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004612 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004613 /*
4614 * We can't just convert len to max_blocks because
4615 * If blocksize = 4096 offset = 3072 and len = 2048
4616 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004617 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4618 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004619 /*
4620 * credits to insert 1 extent into extent tree
4621 */
4622 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4623 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004624 map.m_lblk += ret;
4625 map.m_len = (max_blocks -= ret);
Theodore Ts'o9924a922013-02-08 21:59:22 -05004626 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits);
Mingming Cao00314622009-09-28 15:49:08 -04004627 if (IS_ERR(handle)) {
4628 ret = PTR_ERR(handle);
4629 break;
4630 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004631 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004632 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Lukas Czernerb06acd32013-01-28 21:21:12 -05004633 if (ret <= 0)
4634 ext4_warning(inode->i_sb,
4635 "inode #%lu: block %u: len %u: "
4636 "ext4_ext_map_blocks returned %d",
4637 inode->i_ino, map.m_lblk,
4638 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04004639 ext4_mark_inode_dirty(handle, inode);
4640 ret2 = ext4_journal_stop(handle);
4641 if (ret <= 0 || ret2 )
4642 break;
4643 }
4644 return ret > 0 ? ret2 : ret;
4645}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004646
Mingming Cao00314622009-09-28 15:49:08 -04004647/*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004648 * If newes is not existing extent (newes->ec_pblk equals zero) find
4649 * delayed extent at start of newes and update newes accordingly and
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004650 * return start of the next delayed extent.
4651 *
Zheng Liu69eb33d2013-02-18 00:31:07 -05004652 * If newes is existing extent (newes->ec_pblk is not equal zero)
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004653 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
Zheng Liu69eb33d2013-02-18 00:31:07 -05004654 * extent found. Leave newes unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004655 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004656static int ext4_find_delayed_extent(struct inode *inode,
Zheng Liu69eb33d2013-02-18 00:31:07 -05004657 struct extent_status *newes)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004658{
Zheng Liub3aff3e2012-11-08 21:57:37 -05004659 struct extent_status es;
Zheng Liube401362013-02-18 00:27:26 -05004660 ext4_lblk_t block, next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004661
Zheng Liu69eb33d2013-02-18 00:31:07 -05004662 if (newes->es_pblk == 0) {
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004663 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4664 newes->es_lblk + newes->es_len - 1, &es);
4665
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004666 /*
Zheng Liu69eb33d2013-02-18 00:31:07 -05004667 * No extent in extent-tree contains block @newes->es_pblk,
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004668 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004669 */
Zheng Liu06b0c882013-02-18 00:26:51 -05004670 if (es.es_len == 0)
Zheng Liub3aff3e2012-11-08 21:57:37 -05004671 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004672 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004673
Zheng Liu69eb33d2013-02-18 00:31:07 -05004674 if (es.es_lblk > newes->es_lblk) {
Zheng Liub3aff3e2012-11-08 21:57:37 -05004675 /* A hole found. */
Zheng Liu69eb33d2013-02-18 00:31:07 -05004676 newes->es_len = min(es.es_lblk - newes->es_lblk,
4677 newes->es_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004678 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004679 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004680
Zheng Liu69eb33d2013-02-18 00:31:07 -05004681 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004682 }
4683
Zheng Liu69eb33d2013-02-18 00:31:07 -05004684 block = newes->es_lblk + newes->es_len;
Yan, Zhenge30b5dc2013-05-03 02:15:52 -04004685 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
Zheng Liube401362013-02-18 00:27:26 -05004686 if (es.es_len == 0)
4687 next_del = EXT_MAX_BLOCKS;
4688 else
4689 next_del = es.es_lblk;
4690
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004691 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004692}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004693/* fiemap flags we can handle specified here */
4694#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4695
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004696static int ext4_xattr_fiemap(struct inode *inode,
4697 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004698{
4699 __u64 physical = 0;
4700 __u64 length;
4701 __u32 flags = FIEMAP_EXTENT_LAST;
4702 int blockbits = inode->i_sb->s_blocksize_bits;
4703 int error = 0;
4704
4705 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004706 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004707 struct ext4_iloc iloc;
4708 int offset; /* offset of xattr in inode */
4709
4710 error = ext4_get_inode_loc(inode, &iloc);
4711 if (error)
4712 return error;
Jan Karab9ea8422013-05-31 19:38:56 -04004713 physical = (__u64)iloc.bh->b_blocknr << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004714 offset = EXT4_GOOD_OLD_INODE_SIZE +
4715 EXT4_I(inode)->i_extra_isize;
4716 physical += offset;
4717 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4718 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004719 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004720 } else { /* external block */
Jan Karab9ea8422013-05-31 19:38:56 -04004721 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004722 length = inode->i_sb->s_blocksize;
4723 }
4724
4725 if (physical)
4726 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4727 length, flags);
4728 return (error < 0 ? error : 0);
4729}
4730
4731int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4732 __u64 start, __u64 len)
4733{
4734 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004735 int error = 0;
4736
Tao Ma94191982012-12-10 14:06:02 -05004737 if (ext4_has_inline_data(inode)) {
4738 int has_inline = 1;
4739
4740 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4741
4742 if (has_inline)
4743 return error;
4744 }
4745
Eric Sandeen6873fa02008-10-07 00:46:36 -04004746 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004747 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004748 return generic_block_fiemap(inode, fieinfo, start, len,
4749 ext4_get_block);
4750
4751 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4752 return -EBADR;
4753
4754 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4755 error = ext4_xattr_fiemap(inode, fieinfo);
4756 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004757 ext4_lblk_t len_blks;
4758 __u64 last_blk;
4759
Eric Sandeen6873fa02008-10-07 00:46:36 -04004760 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004761 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004762 if (last_blk >= EXT_MAX_BLOCKS)
4763 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004764 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004765
4766 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004767 * Walk the extent tree gathering extent information
4768 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004769 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004770 error = ext4_fill_fiemap_extents(inode, start_blk,
4771 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004772 }
4773
4774 return error;
4775}