blob: 5ae1674ec12f16b7eea6635244f9b68f252d9ba2 [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,
115 struct ext4_ext_cache *newex);
116
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 */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400160#define ext4_ext_dirty(handle, inode, path) \
161 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
162static int __ext4_ext_dirty(const char *where, unsigned int line,
163 handle_t *handle, struct inode *inode,
164 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700165{
166 int err;
167 if (path->p_bh) {
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400168 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
Alex Tomasa86c6182006-10-11 01:21:03 -0700169 /* path points to block */
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -0400170 err = __ext4_handle_dirty_metadata(where, line, handle,
171 inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700172 } else {
173 /* path points to leaf/index in inode body */
174 err = ext4_mark_inode_dirty(handle, inode);
175 }
176 return err;
177}
178
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700179static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700180 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500181 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700182{
Alex Tomasa86c6182006-10-11 01:21:03 -0700183 if (path) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400184 int depth = path->p_depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700185 struct ext4_extent *ex;
Alex Tomasa86c6182006-10-11 01:21:03 -0700186
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500187 /*
188 * Try to predict block placement assuming that we are
189 * filling in a file which will eventually be
190 * non-sparse --- i.e., in the case of libbfd writing
191 * an ELF object sections out-of-order but in a way
192 * the eventually results in a contiguous object or
193 * executable file, or some database extending a table
194 * space file. However, this is actually somewhat
195 * non-ideal if we are writing a sparse file such as
196 * qemu or KVM writing a raw image file that is going
197 * to stay fairly sparse, since it will end up
198 * fragmenting the file system's free space. Maybe we
199 * should have some hueristics or some way to allow
200 * userspace to pass a hint to file system,
Tao Mab8d65682011-01-21 23:21:31 +0800201 * especially if the latter case turns out to be
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500202 * common.
203 */
Avantika Mathur7e028972006-12-06 20:41:33 -0800204 ex = path[depth].p_ext;
Kazuya Mioad4fb9c2011-01-10 12:12:28 -0500205 if (ex) {
206 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
207 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
208
209 if (block > ext_block)
210 return ext_pblk + (block - ext_block);
211 else
212 return ext_pblk - (ext_block - block);
213 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700214
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700215 /* it looks like index is empty;
216 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700217 if (path[depth].p_bh)
218 return path[depth].p_bh->b_blocknr;
219 }
220
221 /* OK. use inode's group */
Eric Sandeenf86186b2011-06-28 10:01:31 -0400222 return ext4_inode_to_goal_block(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700223}
224
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400225/*
226 * Allocation for a meta data block
227 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700228static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400229ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700230 struct ext4_ext_path *path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400231 struct ext4_extent *ex, int *err, unsigned int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -0700232{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700233 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700234
235 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Allison Henderson55f020d2011-05-25 07:41:26 -0400236 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
237 NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700238 return newblock;
239}
240
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400241static inline int ext4_ext_space_block(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700242{
243 int size;
244
245 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
246 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100247#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400248 if (!check && size > 6)
249 size = 6;
Alex Tomasa86c6182006-10-11 01:21:03 -0700250#endif
251 return size;
252}
253
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400254static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700255{
256 int size;
257
258 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
259 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100260#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400261 if (!check && size > 5)
262 size = 5;
Alex Tomasa86c6182006-10-11 01:21:03 -0700263#endif
264 return size;
265}
266
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400267static inline int ext4_ext_space_root(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700268{
269 int size;
270
271 size = sizeof(EXT4_I(inode)->i_data);
272 size -= sizeof(struct ext4_extent_header);
273 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100274#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400275 if (!check && size > 3)
276 size = 3;
Alex Tomasa86c6182006-10-11 01:21:03 -0700277#endif
278 return size;
279}
280
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400281static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
Alex Tomasa86c6182006-10-11 01:21:03 -0700282{
283 int size;
284
285 size = sizeof(EXT4_I(inode)->i_data);
286 size -= sizeof(struct ext4_extent_header);
287 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100288#ifdef AGGRESSIVE_TEST
Yongqiang Yang02dc62fb2011-10-29 09:29:11 -0400289 if (!check && size > 4)
290 size = 4;
Alex Tomasa86c6182006-10-11 01:21:03 -0700291#endif
292 return size;
293}
294
Mingming Caod2a17632008-07-14 17:52:37 -0400295/*
296 * Calculate the number of metadata blocks needed
297 * to allocate @blocks
298 * Worse case is one block per extent
299 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500300int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -0400301{
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500302 struct ext4_inode_info *ei = EXT4_I(inode);
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400303 int idxs;
Mingming Caod2a17632008-07-14 17:52:37 -0400304
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500305 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
306 / sizeof(struct ext4_extent_idx));
Mingming Caod2a17632008-07-14 17:52:37 -0400307
308 /*
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500309 * If the new delayed allocation block is contiguous with the
310 * previous da block, it can share index blocks with the
311 * previous block, so we only need to allocate a new index
312 * block every idxs leaf blocks. At ldxs**2 blocks, we need
313 * an additional index block, and at ldxs**3 blocks, yet
314 * another index blocks.
Mingming Caod2a17632008-07-14 17:52:37 -0400315 */
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500316 if (ei->i_da_metadata_calc_len &&
317 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400318 int num = 0;
319
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500320 if ((ei->i_da_metadata_calc_len % idxs) == 0)
321 num++;
322 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
323 num++;
324 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
325 num++;
326 ei->i_da_metadata_calc_len = 0;
327 } else
328 ei->i_da_metadata_calc_len++;
329 ei->i_da_metadata_calc_last_lblock++;
330 return num;
331 }
Mingming Caod2a17632008-07-14 17:52:37 -0400332
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500333 /*
334 * In the worst case we need a new set of index blocks at
335 * every level of the inode's extent tree.
336 */
337 ei->i_da_metadata_calc_len = 1;
338 ei->i_da_metadata_calc_last_lblock = lblock;
339 return ext_depth(inode) + 1;
Mingming Caod2a17632008-07-14 17:52:37 -0400340}
341
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400342static int
343ext4_ext_max_entries(struct inode *inode, int depth)
344{
345 int max;
346
347 if (depth == ext_depth(inode)) {
348 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400349 max = ext4_ext_space_root(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400350 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400351 max = ext4_ext_space_root_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400352 } else {
353 if (depth == 0)
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400354 max = ext4_ext_space_block(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400355 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400356 max = ext4_ext_space_block_idx(inode, 1);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400357 }
358
359 return max;
360}
361
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400362static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
363{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400364 ext4_fsblk_t block = ext4_ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400365 int len = ext4_ext_get_actual_len(ext);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400366
Theodore Ts'o31d4f3a2012-03-11 23:30:16 -0400367 if (len == 0)
368 return 0;
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400369 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400370}
371
372static int ext4_valid_extent_idx(struct inode *inode,
373 struct ext4_extent_idx *ext_idx)
374{
Theodore Ts'obf89d162010-10-27 21:30:14 -0400375 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400376
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400377 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400378}
379
380static int ext4_valid_extent_entries(struct inode *inode,
381 struct ext4_extent_header *eh,
382 int depth)
383{
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400384 unsigned short entries;
385 if (eh->eh_entries == 0)
386 return 1;
387
388 entries = le16_to_cpu(eh->eh_entries);
389
390 if (depth == 0) {
391 /* leaf entries */
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400392 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400393 while (entries) {
394 if (!ext4_valid_extent(inode, ext))
395 return 0;
396 ext++;
397 entries--;
398 }
399 } else {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -0400400 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400401 while (entries) {
402 if (!ext4_valid_extent_idx(inode, ext_idx))
403 return 0;
404 ext_idx++;
405 entries--;
406 }
407 }
408 return 1;
409}
410
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400411static int __ext4_ext_check(const char *function, unsigned int line,
412 struct inode *inode, struct ext4_extent_header *eh,
413 int depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400414{
415 const char *error_msg;
416 int max = 0;
417
418 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
419 error_msg = "invalid magic";
420 goto corrupted;
421 }
422 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
423 error_msg = "unexpected eh_depth";
424 goto corrupted;
425 }
426 if (unlikely(eh->eh_max == 0)) {
427 error_msg = "invalid eh_max";
428 goto corrupted;
429 }
430 max = ext4_ext_max_entries(inode, depth);
431 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
432 error_msg = "too large eh_max";
433 goto corrupted;
434 }
435 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
436 error_msg = "invalid eh_entries";
437 goto corrupted;
438 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400439 if (!ext4_valid_extent_entries(inode, eh, depth)) {
440 error_msg = "invalid extent entries";
441 goto corrupted;
442 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400443 /* Verify checksum on non-root extent tree nodes */
444 if (ext_depth(inode) != depth &&
445 !ext4_extent_block_csum_verify(inode, eh)) {
446 error_msg = "extent tree corrupted";
447 goto corrupted;
448 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400449 return 0;
450
451corrupted:
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400452 ext4_error_inode(inode, function, line, 0,
Theodore Ts'o24676da2010-05-16 21:00:00 -0400453 "bad header/extent: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400454 "entries %u, max %u(%u), depth %u(%u)",
Theodore Ts'o24676da2010-05-16 21:00:00 -0400455 error_msg, le16_to_cpu(eh->eh_magic),
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400456 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
457 max, le16_to_cpu(eh->eh_depth), depth);
458
459 return -EIO;
460}
461
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400462#define ext4_ext_check(inode, eh, depth) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400463 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400464
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400465int ext4_ext_check_inode(struct inode *inode)
466{
467 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
468}
469
Darrick J. Wongf8489122012-04-29 18:21:10 -0400470static int __ext4_ext_check_block(const char *function, unsigned int line,
471 struct inode *inode,
472 struct ext4_extent_header *eh,
473 int depth,
474 struct buffer_head *bh)
475{
476 int ret;
477
478 if (buffer_verified(bh))
479 return 0;
480 ret = ext4_ext_check(inode, eh, depth);
481 if (ret)
482 return ret;
483 set_buffer_verified(bh);
484 return ret;
485}
486
487#define ext4_ext_check_block(inode, eh, depth, bh) \
488 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
489
Alex Tomasa86c6182006-10-11 01:21:03 -0700490#ifdef EXT_DEBUG
491static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
492{
493 int k, l = path->p_depth;
494
495 ext_debug("path:");
496 for (k = 0; k <= l; k++, path++) {
497 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700498 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400499 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700500 } else if (path->p_ext) {
Mingming553f9002009-09-18 13:34:55 -0400501 ext_debug(" %d:[%d]%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700502 le32_to_cpu(path->p_ext->ee_block),
Mingming553f9002009-09-18 13:34:55 -0400503 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400504 ext4_ext_get_actual_len(path->p_ext),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400505 ext4_ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700506 } else
507 ext_debug(" []");
508 }
509 ext_debug("\n");
510}
511
512static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
513{
514 int depth = ext_depth(inode);
515 struct ext4_extent_header *eh;
516 struct ext4_extent *ex;
517 int i;
518
519 if (!path)
520 return;
521
522 eh = path[depth].p_hdr;
523 ex = EXT_FIRST_EXTENT(eh);
524
Mingming553f9002009-09-18 13:34:55 -0400525 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
526
Alex Tomasa86c6182006-10-11 01:21:03 -0700527 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming553f9002009-09-18 13:34:55 -0400528 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
529 ext4_ext_is_uninitialized(ex),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400530 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700531 }
532 ext_debug("\n");
533}
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400534
535static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
536 ext4_fsblk_t newblock, int level)
537{
538 int depth = ext_depth(inode);
539 struct ext4_extent *ex;
540
541 if (depth != level) {
542 struct ext4_extent_idx *idx;
543 idx = path[level].p_idx;
544 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
545 ext_debug("%d: move %d:%llu in new index %llu\n", level,
546 le32_to_cpu(idx->ei_block),
547 ext4_idx_pblock(idx),
548 newblock);
549 idx++;
550 }
551
552 return;
553 }
554
555 ex = path[depth].p_ext;
556 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
557 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
558 le32_to_cpu(ex->ee_block),
559 ext4_ext_pblock(ex),
560 ext4_ext_is_uninitialized(ex),
561 ext4_ext_get_actual_len(ex),
562 newblock);
563 ex++;
564 }
565}
566
Alex Tomasa86c6182006-10-11 01:21:03 -0700567#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400568#define ext4_ext_show_path(inode, path)
569#define ext4_ext_show_leaf(inode, path)
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400570#define ext4_ext_show_move(inode, path, newblock, level)
Alex Tomasa86c6182006-10-11 01:21:03 -0700571#endif
572
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500573void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700574{
575 int depth = path->p_depth;
576 int i;
577
578 for (i = 0; i <= depth; i++, path++)
579 if (path->p_bh) {
580 brelse(path->p_bh);
581 path->p_bh = NULL;
582 }
583}
584
585/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700586 * ext4_ext_binsearch_idx:
587 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400588 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700589 */
590static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500591ext4_ext_binsearch_idx(struct inode *inode,
592 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700593{
594 struct ext4_extent_header *eh = path->p_hdr;
595 struct ext4_extent_idx *r, *l, *m;
596
Alex Tomasa86c6182006-10-11 01:21:03 -0700597
Eric Sandeenbba90742008-01-28 23:58:27 -0500598 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700599
600 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400601 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700602 while (l <= r) {
603 m = l + (r - l) / 2;
604 if (block < le32_to_cpu(m->ei_block))
605 r = m - 1;
606 else
607 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400608 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
609 m, le32_to_cpu(m->ei_block),
610 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700611 }
612
613 path->p_idx = l - 1;
Zheng Liu4a3c3a52012-05-28 17:55:16 -0400614 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400615 ext4_idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700616
617#ifdef CHECK_BINSEARCH
618 {
619 struct ext4_extent_idx *chix, *ix;
620 int k;
621
622 chix = ix = EXT_FIRST_INDEX(eh);
623 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
624 if (k != 0 &&
625 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400626 printk(KERN_DEBUG "k=%d, ix=0x%p, "
627 "first=0x%p\n", k,
628 ix, EXT_FIRST_INDEX(eh));
629 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700630 le32_to_cpu(ix->ei_block),
631 le32_to_cpu(ix[-1].ei_block));
632 }
633 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400634 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700635 if (block < le32_to_cpu(ix->ei_block))
636 break;
637 chix = ix;
638 }
639 BUG_ON(chix != path->p_idx);
640 }
641#endif
642
643}
644
645/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700646 * ext4_ext_binsearch:
647 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400648 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700649 */
650static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500651ext4_ext_binsearch(struct inode *inode,
652 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700653{
654 struct ext4_extent_header *eh = path->p_hdr;
655 struct ext4_extent *r, *l, *m;
656
Alex Tomasa86c6182006-10-11 01:21:03 -0700657 if (eh->eh_entries == 0) {
658 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700659 * this leaf is empty:
660 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700661 */
662 return;
663 }
664
Eric Sandeenbba90742008-01-28 23:58:27 -0500665 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700666
667 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400668 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700669
670 while (l <= r) {
671 m = l + (r - l) / 2;
672 if (block < le32_to_cpu(m->ee_block))
673 r = m - 1;
674 else
675 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400676 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
677 m, le32_to_cpu(m->ee_block),
678 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700679 }
680
681 path->p_ext = l - 1;
Mingming553f9002009-09-18 13:34:55 -0400682 ext_debug(" -> %d:%llu:[%d]%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400683 le32_to_cpu(path->p_ext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -0400684 ext4_ext_pblock(path->p_ext),
Mingming553f9002009-09-18 13:34:55 -0400685 ext4_ext_is_uninitialized(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400686 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700687
688#ifdef CHECK_BINSEARCH
689 {
690 struct ext4_extent *chex, *ex;
691 int k;
692
693 chex = ex = EXT_FIRST_EXTENT(eh);
694 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
695 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400696 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700697 if (block < le32_to_cpu(ex->ee_block))
698 break;
699 chex = ex;
700 }
701 BUG_ON(chex != path->p_ext);
702 }
703#endif
704
705}
706
707int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
708{
709 struct ext4_extent_header *eh;
710
711 eh = ext_inode_hdr(inode);
712 eh->eh_depth = 0;
713 eh->eh_entries = 0;
714 eh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400715 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700716 ext4_mark_inode_dirty(handle, inode);
717 ext4_ext_invalidate_cache(inode);
718 return 0;
719}
720
721struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500722ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
723 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700724{
725 struct ext4_extent_header *eh;
726 struct buffer_head *bh;
727 short int depth, i, ppos = 0, alloc = 0;
728
729 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400730 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700731
732 /* account possible depth increase */
733 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800734 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700735 GFP_NOFS);
736 if (!path)
737 return ERR_PTR(-ENOMEM);
738 alloc = 1;
739 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700740 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400741 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700742
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400743 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700744 /* walk through the tree */
745 while (i) {
746 ext_debug("depth %d: num %d, max %d\n",
747 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400748
Alex Tomasa86c6182006-10-11 01:21:03 -0700749 ext4_ext_binsearch_idx(inode, path + ppos, block);
Theodore Ts'obf89d162010-10-27 21:30:14 -0400750 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700751 path[ppos].p_depth = i;
752 path[ppos].p_ext = NULL;
753
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400754 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
755 if (unlikely(!bh))
Alex Tomasa86c6182006-10-11 01:21:03 -0700756 goto err;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400757 if (!bh_uptodate_or_lock(bh)) {
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400758 trace_ext4_ext_load_extent(inode, block,
759 path[ppos].p_block);
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400760 if (bh_submit_read(bh) < 0) {
761 put_bh(bh);
762 goto err;
763 }
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400764 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700765 eh = ext_block_hdr(bh);
766 ppos++;
Frank Mayhar273df552010-03-02 11:46:09 -0500767 if (unlikely(ppos > depth)) {
768 put_bh(bh);
769 EXT4_ERROR_INODE(inode,
770 "ppos %d > depth %d", ppos, depth);
771 goto err;
772 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700773 path[ppos].p_bh = bh;
774 path[ppos].p_hdr = eh;
775 i--;
776
Darrick J. Wongf8489122012-04-29 18:21:10 -0400777 if (ext4_ext_check_block(inode, eh, i, bh))
Alex Tomasa86c6182006-10-11 01:21:03 -0700778 goto err;
779 }
780
781 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700782 path[ppos].p_ext = NULL;
783 path[ppos].p_idx = NULL;
784
Alex Tomasa86c6182006-10-11 01:21:03 -0700785 /* find extent */
786 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400787 /* if not an empty leaf */
788 if (path[ppos].p_ext)
Theodore Ts'obf89d162010-10-27 21:30:14 -0400789 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700790
791 ext4_ext_show_path(inode, path);
792
793 return path;
794
795err:
796 ext4_ext_drop_refs(path);
797 if (alloc)
798 kfree(path);
799 return ERR_PTR(-EIO);
800}
801
802/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700803 * ext4_ext_insert_index:
804 * insert new index [@logical;@ptr] into the block at @curp;
805 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700806 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -0400807static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
808 struct ext4_ext_path *curp,
809 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700810{
811 struct ext4_extent_idx *ix;
812 int len, err;
813
Avantika Mathur7e028972006-12-06 20:41:33 -0800814 err = ext4_ext_get_access(handle, inode, curp);
815 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700816 return err;
817
Frank Mayhar273df552010-03-02 11:46:09 -0500818 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
819 EXT4_ERROR_INODE(inode,
820 "logical %d == ei_block %d!",
821 logical, le32_to_cpu(curp->p_idx->ei_block));
822 return -EIO;
823 }
Robin Dongd4620312011-07-17 23:43:42 -0400824
825 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
826 >= le16_to_cpu(curp->p_hdr->eh_max))) {
827 EXT4_ERROR_INODE(inode,
828 "eh_entries %d >= eh_max %d!",
829 le16_to_cpu(curp->p_hdr->eh_entries),
830 le16_to_cpu(curp->p_hdr->eh_max));
831 return -EIO;
832 }
833
Alex Tomasa86c6182006-10-11 01:21:03 -0700834 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
835 /* insert after */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400836 ext_debug("insert new index %d after: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700837 ix = curp->p_idx + 1;
838 } else {
839 /* insert before */
Eric Gouriou80e675f2011-10-27 11:52:18 -0400840 ext_debug("insert new index %d before: %llu\n", logical, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700841 ix = curp->p_idx;
842 }
843
Eric Gouriou80e675f2011-10-27 11:52:18 -0400844 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
845 BUG_ON(len < 0);
846 if (len > 0) {
847 ext_debug("insert new index %d: "
848 "move %d indices from 0x%p to 0x%p\n",
849 logical, len, ix, ix + 1);
850 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
851 }
852
Tao Maf472e022011-10-17 10:13:46 -0400853 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
854 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
855 return -EIO;
856 }
857
Alex Tomasa86c6182006-10-11 01:21:03 -0700858 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700859 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400860 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700861
Frank Mayhar273df552010-03-02 11:46:09 -0500862 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
863 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
864 return -EIO;
865 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700866
867 err = ext4_ext_dirty(handle, inode, curp);
868 ext4_std_error(inode->i_sb, err);
869
870 return err;
871}
872
873/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700874 * ext4_ext_split:
875 * inserts new subtree into the path, using free index entry
876 * at depth @at:
877 * - allocates all needed blocks (new leaf and all intermediate index blocks)
878 * - makes decision where to split
879 * - moves remaining extents and index entries (right to the split point)
880 * into the newly allocated blocks
881 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700882 */
883static int ext4_ext_split(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -0400884 unsigned int flags,
885 struct ext4_ext_path *path,
886 struct ext4_extent *newext, int at)
Alex Tomasa86c6182006-10-11 01:21:03 -0700887{
888 struct buffer_head *bh = NULL;
889 int depth = ext_depth(inode);
890 struct ext4_extent_header *neh;
891 struct ext4_extent_idx *fidx;
Alex Tomasa86c6182006-10-11 01:21:03 -0700892 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700893 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700894 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700895 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700896 int err = 0;
897
898 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700899 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700900
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700901 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700902 * border from split point */
Frank Mayhar273df552010-03-02 11:46:09 -0500903 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
904 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
905 return -EIO;
906 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700907 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
908 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700909 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700910 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400911 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700912 } else {
913 border = newext->ee_block;
914 ext_debug("leaf will be added."
915 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400916 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700917 }
918
919 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700920 * If error occurs, then we break processing
921 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700922 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700923 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700924 */
925
926 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700927 * Get array to track all allocated blocks.
928 * We need this to handle errors and free blocks
929 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700930 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800931 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700932 if (!ablocks)
933 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700934
935 /* allocate all needed blocks */
936 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
937 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400938 newblock = ext4_ext_new_meta_block(handle, inode, path,
Allison Henderson55f020d2011-05-25 07:41:26 -0400939 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -0700940 if (newblock == 0)
941 goto cleanup;
942 ablocks[a] = newblock;
943 }
944
945 /* initialize new leaf */
946 newblock = ablocks[--a];
Frank Mayhar273df552010-03-02 11:46:09 -0500947 if (unlikely(newblock == 0)) {
948 EXT4_ERROR_INODE(inode, "newblock == 0!");
949 err = -EIO;
950 goto cleanup;
951 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700952 bh = sb_getblk(inode->i_sb, newblock);
953 if (!bh) {
954 err = -EIO;
955 goto cleanup;
956 }
957 lock_buffer(bh);
958
Avantika Mathur7e028972006-12-06 20:41:33 -0800959 err = ext4_journal_get_create_access(handle, bh);
960 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700961 goto cleanup;
962
963 neh = ext_block_hdr(bh);
964 neh->eh_entries = 0;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -0400965 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -0700966 neh->eh_magic = EXT4_EXT_MAGIC;
967 neh->eh_depth = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700968
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700969 /* move remainder of path[depth] to the new leaf */
Frank Mayhar273df552010-03-02 11:46:09 -0500970 if (unlikely(path[depth].p_hdr->eh_entries !=
971 path[depth].p_hdr->eh_max)) {
972 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
973 path[depth].p_hdr->eh_entries,
974 path[depth].p_hdr->eh_max);
975 err = -EIO;
976 goto cleanup;
977 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700978 /* start copy from next extent */
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400979 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
980 ext4_ext_show_move(inode, path, newblock, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -0700981 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -0400982 struct ext4_extent *ex;
983 ex = EXT_FIRST_EXTENT(neh);
984 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400985 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700986 }
987
Darrick J. Wong7ac59902012-04-29 18:37:10 -0400988 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700989 set_buffer_uptodate(bh);
990 unlock_buffer(bh);
991
Frank Mayhar03901312009-01-07 00:06:22 -0500992 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800993 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700994 goto cleanup;
995 brelse(bh);
996 bh = NULL;
997
998 /* correct old leaf */
999 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -08001000 err = ext4_ext_get_access(handle, inode, path + depth);
1001 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001002 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001003 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -08001004 err = ext4_ext_dirty(handle, inode, path + depth);
1005 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001006 goto cleanup;
1007
1008 }
1009
1010 /* create intermediate indexes */
1011 k = depth - at - 1;
Frank Mayhar273df552010-03-02 11:46:09 -05001012 if (unlikely(k < 0)) {
1013 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1014 err = -EIO;
1015 goto cleanup;
1016 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001017 if (k)
1018 ext_debug("create %d intermediate indices\n", k);
1019 /* insert new index into current index block */
1020 /* current depth stored in i var */
1021 i = depth - 1;
1022 while (k--) {
1023 oldblock = newblock;
1024 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -05001025 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001026 if (!bh) {
1027 err = -EIO;
1028 goto cleanup;
1029 }
1030 lock_buffer(bh);
1031
Avantika Mathur7e028972006-12-06 20:41:33 -08001032 err = ext4_journal_get_create_access(handle, bh);
1033 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001034 goto cleanup;
1035
1036 neh = ext_block_hdr(bh);
1037 neh->eh_entries = cpu_to_le16(1);
1038 neh->eh_magic = EXT4_EXT_MAGIC;
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001039 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001040 neh->eh_depth = cpu_to_le16(depth - i);
1041 fidx = EXT_FIRST_INDEX(neh);
1042 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001043 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001044
Eric Sandeenbba90742008-01-28 23:58:27 -05001045 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1046 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001047
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001048 /* move remainder of path[i] to the new index block */
Frank Mayhar273df552010-03-02 11:46:09 -05001049 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1050 EXT_LAST_INDEX(path[i].p_hdr))) {
1051 EXT4_ERROR_INODE(inode,
1052 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1053 le32_to_cpu(path[i].p_ext->ee_block));
1054 err = -EIO;
1055 goto cleanup;
1056 }
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001057 /* start copy indexes */
1058 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1059 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1060 EXT_MAX_INDEX(path[i].p_hdr));
1061 ext4_ext_show_move(inode, path, newblock, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07001062 if (m) {
Yongqiang Yang1b16da72011-05-25 17:41:48 -04001063 memmove(++fidx, path[i].p_idx,
Alex Tomasa86c6182006-10-11 01:21:03 -07001064 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001065 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001066 }
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001067 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001068 set_buffer_uptodate(bh);
1069 unlock_buffer(bh);
1070
Frank Mayhar03901312009-01-07 00:06:22 -05001071 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001072 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001073 goto cleanup;
1074 brelse(bh);
1075 bh = NULL;
1076
1077 /* correct old index */
1078 if (m) {
1079 err = ext4_ext_get_access(handle, inode, path + i);
1080 if (err)
1081 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001082 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -07001083 err = ext4_ext_dirty(handle, inode, path + i);
1084 if (err)
1085 goto cleanup;
1086 }
1087
1088 i--;
1089 }
1090
1091 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001092 err = ext4_ext_insert_index(handle, inode, path + at,
1093 le32_to_cpu(border), newblock);
1094
1095cleanup:
1096 if (bh) {
1097 if (buffer_locked(bh))
1098 unlock_buffer(bh);
1099 brelse(bh);
1100 }
1101
1102 if (err) {
1103 /* free all allocated blocks in error case */
1104 for (i = 0; i < depth; i++) {
1105 if (!ablocks[i])
1106 continue;
Peter Huewe7dc57612011-02-21 21:01:42 -05001107 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05001108 EXT4_FREE_BLOCKS_METADATA);
Alex Tomasa86c6182006-10-11 01:21:03 -07001109 }
1110 }
1111 kfree(ablocks);
1112
1113 return err;
1114}
1115
1116/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001117 * ext4_ext_grow_indepth:
1118 * implements tree growing procedure:
1119 * - allocates new block
1120 * - moves top-level data (index block or leaf) into the new block
1121 * - initializes new top-level, creating index that points to the
1122 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001123 */
1124static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001125 unsigned int flags,
Allison Henderson55f020d2011-05-25 07:41:26 -04001126 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001127{
Alex Tomasa86c6182006-10-11 01:21:03 -07001128 struct ext4_extent_header *neh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001129 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001130 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001131 int err = 0;
1132
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001133 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
Allison Henderson55f020d2011-05-25 07:41:26 -04001134 newext, &err, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07001135 if (newblock == 0)
1136 return err;
1137
1138 bh = sb_getblk(inode->i_sb, newblock);
1139 if (!bh) {
1140 err = -EIO;
1141 ext4_std_error(inode->i_sb, err);
1142 return err;
1143 }
1144 lock_buffer(bh);
1145
Avantika Mathur7e028972006-12-06 20:41:33 -08001146 err = ext4_journal_get_create_access(handle, bh);
1147 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001148 unlock_buffer(bh);
1149 goto out;
1150 }
1151
1152 /* move top-level index/leaf into new block */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001153 memmove(bh->b_data, EXT4_I(inode)->i_data,
1154 sizeof(EXT4_I(inode)->i_data));
Alex Tomasa86c6182006-10-11 01:21:03 -07001155
1156 /* set size of new block */
1157 neh = ext_block_hdr(bh);
1158 /* old root could have indexes or leaves
1159 * so calculate e_max right way */
1160 if (ext_depth(inode))
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001161 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001162 else
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04001163 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07001164 neh->eh_magic = EXT4_EXT_MAGIC;
Darrick J. Wong7ac59902012-04-29 18:37:10 -04001165 ext4_extent_block_csum_set(inode, neh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001166 set_buffer_uptodate(bh);
1167 unlock_buffer(bh);
1168
Frank Mayhar03901312009-01-07 00:06:22 -05001169 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001170 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001171 goto out;
1172
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001173 /* Update top-level index: num,max,pointer */
Alex Tomasa86c6182006-10-11 01:21:03 -07001174 neh = ext_inode_hdr(inode);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001175 neh->eh_entries = cpu_to_le16(1);
1176 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1177 if (neh->eh_depth == 0) {
1178 /* Root extent block becomes index block */
1179 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1180 EXT_FIRST_INDEX(neh)->ei_block =
1181 EXT_FIRST_EXTENT(neh)->ee_block;
1182 }
Mingming Cao2ae02102006-10-11 01:21:11 -07001183 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001184 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Andi Kleen5a0790c2010-06-14 13:28:03 -04001185 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001186 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
Alex Tomasa86c6182006-10-11 01:21:03 -07001187
Wei Yongjunba39ebb2012-09-27 09:37:53 -04001188 le16_add_cpu(&neh->eh_depth, 1);
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001189 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07001190out:
1191 brelse(bh);
1192
1193 return err;
1194}
1195
1196/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001197 * ext4_ext_create_new_leaf:
1198 * finds empty index and adds new leaf.
1199 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001200 */
1201static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
Allison Henderson55f020d2011-05-25 07:41:26 -04001202 unsigned int flags,
1203 struct ext4_ext_path *path,
1204 struct ext4_extent *newext)
Alex Tomasa86c6182006-10-11 01:21:03 -07001205{
1206 struct ext4_ext_path *curp;
1207 int depth, i, err = 0;
1208
1209repeat:
1210 i = depth = ext_depth(inode);
1211
1212 /* walk up to the tree and look for free index entry */
1213 curp = path + depth;
1214 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1215 i--;
1216 curp--;
1217 }
1218
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001219 /* we use already allocated block for index block,
1220 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001221 if (EXT_HAS_FREE_INDEX(curp)) {
1222 /* if we found index with free entry, then use that
1223 * entry: create all needed subtree and add new leaf */
Allison Henderson55f020d2011-05-25 07:41:26 -04001224 err = ext4_ext_split(handle, inode, flags, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001225 if (err)
1226 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001227
1228 /* refill path */
1229 ext4_ext_drop_refs(path);
1230 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001231 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1232 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001233 if (IS_ERR(path))
1234 err = PTR_ERR(path);
1235 } else {
1236 /* tree is full, time to grow in depth */
Dmitry Monakhov1939dd82011-10-22 01:26:05 -04001237 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001238 if (err)
1239 goto out;
1240
1241 /* refill path */
1242 ext4_ext_drop_refs(path);
1243 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001244 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1245 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001246 if (IS_ERR(path)) {
1247 err = PTR_ERR(path);
1248 goto out;
1249 }
1250
1251 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001252 * only first (depth 0 -> 1) produces free space;
1253 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001254 */
1255 depth = ext_depth(inode);
1256 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001257 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001258 goto repeat;
1259 }
1260 }
1261
1262out:
1263 return err;
1264}
1265
1266/*
Alex Tomas1988b512008-01-28 23:58:27 -05001267 * search the closest allocated block to the left for *logical
1268 * and returns it at @logical + it's physical address at @phys
1269 * if *logical is the smallest allocated block, the function
1270 * returns 0 at @phys
1271 * return value contains 0 (success) or error code
1272 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001273static int ext4_ext_search_left(struct inode *inode,
1274 struct ext4_ext_path *path,
1275 ext4_lblk_t *logical, ext4_fsblk_t *phys)
Alex Tomas1988b512008-01-28 23:58:27 -05001276{
1277 struct ext4_extent_idx *ix;
1278 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001279 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001280
Frank Mayhar273df552010-03-02 11:46:09 -05001281 if (unlikely(path == NULL)) {
1282 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1283 return -EIO;
1284 }
Alex Tomas1988b512008-01-28 23:58:27 -05001285 depth = path->p_depth;
1286 *phys = 0;
1287
1288 if (depth == 0 && path->p_ext == NULL)
1289 return 0;
1290
1291 /* usually extent in the path covers blocks smaller
1292 * then *logical, but it can be that extent is the
1293 * first one in the file */
1294
1295 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001296 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001297 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001298 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1299 EXT4_ERROR_INODE(inode,
1300 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1301 *logical, le32_to_cpu(ex->ee_block));
1302 return -EIO;
1303 }
Alex Tomas1988b512008-01-28 23:58:27 -05001304 while (--depth >= 0) {
1305 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001306 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1307 EXT4_ERROR_INODE(inode,
1308 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
Tao Ma6ee3b212011-10-08 16:08:34 -04001309 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001310 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
Tao Ma6ee3b212011-10-08 16:08:34 -04001311 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
Frank Mayhar273df552010-03-02 11:46:09 -05001312 depth);
1313 return -EIO;
1314 }
Alex Tomas1988b512008-01-28 23:58:27 -05001315 }
1316 return 0;
1317 }
1318
Frank Mayhar273df552010-03-02 11:46:09 -05001319 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1320 EXT4_ERROR_INODE(inode,
1321 "logical %d < ee_block %d + ee_len %d!",
1322 *logical, le32_to_cpu(ex->ee_block), ee_len);
1323 return -EIO;
1324 }
Alex Tomas1988b512008-01-28 23:58:27 -05001325
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001326 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001327 *phys = ext4_ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001328 return 0;
1329}
1330
1331/*
1332 * search the closest allocated block to the right for *logical
1333 * and returns it at @logical + it's physical address at @phys
Tao Madf3ab172011-10-08 15:53:49 -04001334 * if *logical is the largest allocated block, the function
Alex Tomas1988b512008-01-28 23:58:27 -05001335 * returns 0 at @phys
1336 * return value contains 0 (success) or error code
1337 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001338static int ext4_ext_search_right(struct inode *inode,
1339 struct ext4_ext_path *path,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001340 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1341 struct ext4_extent **ret_ex)
Alex Tomas1988b512008-01-28 23:58:27 -05001342{
1343 struct buffer_head *bh = NULL;
1344 struct ext4_extent_header *eh;
1345 struct ext4_extent_idx *ix;
1346 struct ext4_extent *ex;
1347 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001348 int depth; /* Note, NOT eh_depth; depth from top of tree */
1349 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001350
Frank Mayhar273df552010-03-02 11:46:09 -05001351 if (unlikely(path == NULL)) {
1352 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1353 return -EIO;
1354 }
Alex Tomas1988b512008-01-28 23:58:27 -05001355 depth = path->p_depth;
1356 *phys = 0;
1357
1358 if (depth == 0 && path->p_ext == NULL)
1359 return 0;
1360
1361 /* usually extent in the path covers blocks smaller
1362 * then *logical, but it can be that extent is the
1363 * first one in the file */
1364
1365 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001366 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001367 if (*logical < le32_to_cpu(ex->ee_block)) {
Frank Mayhar273df552010-03-02 11:46:09 -05001368 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1369 EXT4_ERROR_INODE(inode,
1370 "first_extent(path[%d].p_hdr) != ex",
1371 depth);
1372 return -EIO;
1373 }
Alex Tomas1988b512008-01-28 23:58:27 -05001374 while (--depth >= 0) {
1375 ix = path[depth].p_idx;
Frank Mayhar273df552010-03-02 11:46:09 -05001376 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1377 EXT4_ERROR_INODE(inode,
1378 "ix != EXT_FIRST_INDEX *logical %d!",
1379 *logical);
1380 return -EIO;
1381 }
Alex Tomas1988b512008-01-28 23:58:27 -05001382 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001383 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001384 }
1385
Frank Mayhar273df552010-03-02 11:46:09 -05001386 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1387 EXT4_ERROR_INODE(inode,
1388 "logical %d < ee_block %d + ee_len %d!",
1389 *logical, le32_to_cpu(ex->ee_block), ee_len);
1390 return -EIO;
1391 }
Alex Tomas1988b512008-01-28 23:58:27 -05001392
1393 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1394 /* next allocated block in this leaf */
1395 ex++;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001396 goto found_extent;
Alex Tomas1988b512008-01-28 23:58:27 -05001397 }
1398
1399 /* go up and search for index to the right */
1400 while (--depth >= 0) {
1401 ix = path[depth].p_idx;
1402 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001403 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001404 }
1405
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001406 /* we've gone up to the root and found no index to the right */
1407 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001408
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001409got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001410 /* we've found index to the right, let's
1411 * follow it and find the closest allocated
1412 * block to the right */
1413 ix++;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001414 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001415 while (++depth < path->p_depth) {
1416 bh = sb_bread(inode->i_sb, block);
1417 if (bh == NULL)
1418 return -EIO;
1419 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001420 /* subtract from p_depth to get proper eh_depth */
Darrick J. Wongf8489122012-04-29 18:21:10 -04001421 if (ext4_ext_check_block(inode, eh,
1422 path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001423 put_bh(bh);
1424 return -EIO;
1425 }
1426 ix = EXT_FIRST_INDEX(eh);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001427 block = ext4_idx_pblock(ix);
Alex Tomas1988b512008-01-28 23:58:27 -05001428 put_bh(bh);
1429 }
1430
1431 bh = sb_bread(inode->i_sb, block);
1432 if (bh == NULL)
1433 return -EIO;
1434 eh = ext_block_hdr(bh);
Darrick J. Wongf8489122012-04-29 18:21:10 -04001435 if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001436 put_bh(bh);
1437 return -EIO;
1438 }
1439 ex = EXT_FIRST_EXTENT(eh);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001440found_extent:
Alex Tomas1988b512008-01-28 23:58:27 -05001441 *logical = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04001442 *phys = ext4_ext_pblock(ex);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001443 *ret_ex = ex;
1444 if (bh)
1445 put_bh(bh);
Alex Tomas1988b512008-01-28 23:58:27 -05001446 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001447}
1448
1449/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001450 * ext4_ext_next_allocated_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001451 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001452 * NOTE: it considers block number from index entry as
1453 * allocated block. Thus, index entries have to be consistent
1454 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001455 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001456static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001457ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1458{
1459 int depth;
1460
1461 BUG_ON(path == NULL);
1462 depth = path->p_depth;
1463
1464 if (depth == 0 && path->p_ext == NULL)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001465 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001466
1467 while (depth >= 0) {
1468 if (depth == path->p_depth) {
1469 /* leaf */
Curt Wohlgemuth6f8ff532011-10-26 04:38:59 -04001470 if (path[depth].p_ext &&
1471 path[depth].p_ext !=
Alex Tomasa86c6182006-10-11 01:21:03 -07001472 EXT_LAST_EXTENT(path[depth].p_hdr))
1473 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1474 } else {
1475 /* index */
1476 if (path[depth].p_idx !=
1477 EXT_LAST_INDEX(path[depth].p_hdr))
1478 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1479 }
1480 depth--;
1481 }
1482
Lukas Czernerf17722f2011-06-06 00:05:17 -04001483 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001484}
1485
1486/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001487 * ext4_ext_next_leaf_block:
Lukas Czernerf17722f2011-06-06 00:05:17 -04001488 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
Alex Tomasa86c6182006-10-11 01:21:03 -07001489 */
Robin Dong57187892011-07-23 21:49:07 -04001490static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001491{
1492 int depth;
1493
1494 BUG_ON(path == NULL);
1495 depth = path->p_depth;
1496
1497 /* zero-tree has no leaf blocks at all */
1498 if (depth == 0)
Lukas Czernerf17722f2011-06-06 00:05:17 -04001499 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001500
1501 /* go to index block */
1502 depth--;
1503
1504 while (depth >= 0) {
1505 if (path[depth].p_idx !=
1506 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001507 return (ext4_lblk_t)
1508 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001509 depth--;
1510 }
1511
Lukas Czernerf17722f2011-06-06 00:05:17 -04001512 return EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07001513}
1514
1515/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001516 * ext4_ext_correct_indexes:
1517 * if leaf gets modified and modified extent is first in the leaf,
1518 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001519 * TODO: do we need to correct tree in all cases?
1520 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001521static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001522 struct ext4_ext_path *path)
1523{
1524 struct ext4_extent_header *eh;
1525 int depth = ext_depth(inode);
1526 struct ext4_extent *ex;
1527 __le32 border;
1528 int k, err = 0;
1529
1530 eh = path[depth].p_hdr;
1531 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001532
1533 if (unlikely(ex == NULL || eh == NULL)) {
1534 EXT4_ERROR_INODE(inode,
1535 "ex %p == NULL or eh %p == NULL", ex, eh);
1536 return -EIO;
1537 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001538
1539 if (depth == 0) {
1540 /* there is no tree at all */
1541 return 0;
1542 }
1543
1544 if (ex != EXT_FIRST_EXTENT(eh)) {
1545 /* we correct tree if first leaf got modified only */
1546 return 0;
1547 }
1548
1549 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001550 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001551 */
1552 k = depth - 1;
1553 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001554 err = ext4_ext_get_access(handle, inode, path + k);
1555 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001556 return err;
1557 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001558 err = ext4_ext_dirty(handle, inode, path + k);
1559 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001560 return err;
1561
1562 while (k--) {
1563 /* change all left-side indexes */
1564 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1565 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001566 err = ext4_ext_get_access(handle, inode, path + k);
1567 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001568 break;
1569 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001570 err = ext4_ext_dirty(handle, inode, path + k);
1571 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001572 break;
1573 }
1574
1575 return err;
1576}
1577
Akira Fujita748de672009-06-17 19:24:03 -04001578int
Alex Tomasa86c6182006-10-11 01:21:03 -07001579ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1580 struct ext4_extent *ex2)
1581{
Amit Arora749269f2007-07-18 09:02:56 -04001582 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001583
1584 /*
1585 * Make sure that either both extents are uninitialized, or
1586 * both are _not_.
1587 */
1588 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1589 return 0;
1590
Amit Arora749269f2007-07-18 09:02:56 -04001591 if (ext4_ext_is_uninitialized(ex1))
1592 max_len = EXT_UNINIT_MAX_LEN;
1593 else
1594 max_len = EXT_INIT_MAX_LEN;
1595
Amit Aroraa2df2a62007-07-17 21:42:41 -04001596 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1597 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1598
1599 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001600 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001601 return 0;
1602
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001603 /*
1604 * To allow future support for preallocated extents to be added
1605 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001606 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001607 */
Amit Arora749269f2007-07-18 09:02:56 -04001608 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001609 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001610#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001611 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001612 return 0;
1613#endif
1614
Theodore Ts'obf89d162010-10-27 21:30:14 -04001615 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001616 return 1;
1617 return 0;
1618}
1619
1620/*
Amit Arora56055d32007-07-17 21:42:38 -04001621 * This function tries to merge the "ex" extent to the next extent in the tree.
1622 * It always tries to merge towards right. If you want to merge towards
1623 * left, pass "ex - 1" as argument instead of "ex".
1624 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1625 * 1 if they got merged.
1626 */
Yongqiang Yang197217a2011-05-03 11:45:29 -04001627static int ext4_ext_try_to_merge_right(struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001628 struct ext4_ext_path *path,
1629 struct ext4_extent *ex)
Amit Arora56055d32007-07-17 21:42:38 -04001630{
1631 struct ext4_extent_header *eh;
1632 unsigned int depth, len;
1633 int merge_done = 0;
1634 int uninitialized = 0;
1635
1636 depth = ext_depth(inode);
1637 BUG_ON(path[depth].p_hdr == NULL);
1638 eh = path[depth].p_hdr;
1639
1640 while (ex < EXT_LAST_EXTENT(eh)) {
1641 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1642 break;
1643 /* merge with next extent! */
1644 if (ext4_ext_is_uninitialized(ex))
1645 uninitialized = 1;
1646 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1647 + ext4_ext_get_actual_len(ex + 1));
1648 if (uninitialized)
1649 ext4_ext_mark_uninitialized(ex);
1650
1651 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1652 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1653 * sizeof(struct ext4_extent);
1654 memmove(ex + 1, ex + 2, len);
1655 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001656 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001657 merge_done = 1;
1658 WARN_ON(eh->eh_entries == 0);
1659 if (!eh->eh_entries)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001660 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
Amit Arora56055d32007-07-17 21:42:38 -04001661 }
1662
1663 return merge_done;
1664}
1665
1666/*
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001667 * This function does a very simple check to see if we can collapse
1668 * an extent tree with a single extent tree leaf block into the inode.
1669 */
1670static void ext4_ext_try_to_merge_up(handle_t *handle,
1671 struct inode *inode,
1672 struct ext4_ext_path *path)
1673{
1674 size_t s;
1675 unsigned max_root = ext4_ext_space_root(inode, 0);
1676 ext4_fsblk_t blk;
1677
1678 if ((path[0].p_depth != 1) ||
1679 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1680 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1681 return;
1682
1683 /*
1684 * We need to modify the block allocation bitmap and the block
1685 * group descriptor to release the extent tree block. If we
1686 * can't get the journal credits, give up.
1687 */
1688 if (ext4_journal_extend(handle, 2))
1689 return;
1690
1691 /*
1692 * Copy the extent data up to the inode
1693 */
1694 blk = ext4_idx_pblock(path[0].p_idx);
1695 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1696 sizeof(struct ext4_extent_idx);
1697 s += sizeof(struct ext4_extent_header);
1698
1699 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1700 path[0].p_depth = 0;
1701 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1702 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1703 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1704
1705 brelse(path[1].p_bh);
1706 ext4_free_blocks(handle, inode, NULL, blk, 1,
1707 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1708}
1709
1710/*
Yongqiang Yang197217a2011-05-03 11:45:29 -04001711 * This function tries to merge the @ex extent to neighbours in the tree.
1712 * return 1 if merge left else 0.
1713 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001714static void ext4_ext_try_to_merge(handle_t *handle,
1715 struct inode *inode,
Yongqiang Yang197217a2011-05-03 11:45:29 -04001716 struct ext4_ext_path *path,
1717 struct ext4_extent *ex) {
1718 struct ext4_extent_header *eh;
1719 unsigned int depth;
1720 int merge_done = 0;
Yongqiang Yang197217a2011-05-03 11:45:29 -04001721
1722 depth = ext_depth(inode);
1723 BUG_ON(path[depth].p_hdr == NULL);
1724 eh = path[depth].p_hdr;
1725
1726 if (ex > EXT_FIRST_EXTENT(eh))
1727 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1728
1729 if (!merge_done)
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001730 (void) ext4_ext_try_to_merge_right(inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001731
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001732 ext4_ext_try_to_merge_up(handle, inode, path);
Yongqiang Yang197217a2011-05-03 11:45:29 -04001733}
1734
1735/*
Amit Arora25d14f92007-05-24 13:04:13 -04001736 * check if a portion of the "newext" extent overlaps with an
1737 * existing extent.
1738 *
1739 * If there is an overlap discovered, it updates the length of the newext
1740 * such that there will be no overlap, and then returns 1.
1741 * If there is no overlap found, it returns 0.
1742 */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001743static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1744 struct inode *inode,
Theodore Ts'o1f109d52010-10-27 21:30:14 -04001745 struct ext4_extent *newext,
1746 struct ext4_ext_path *path)
Amit Arora25d14f92007-05-24 13:04:13 -04001747{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001748 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001749 unsigned int depth, len1;
1750 unsigned int ret = 0;
1751
1752 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001753 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001754 depth = ext_depth(inode);
1755 if (!path[depth].p_ext)
1756 goto out;
1757 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001758 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001759
1760 /*
1761 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001762 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001763 */
1764 if (b2 < b1) {
1765 b2 = ext4_ext_next_allocated_block(path);
Lukas Czernerf17722f2011-06-06 00:05:17 -04001766 if (b2 == EXT_MAX_BLOCKS)
Amit Arora25d14f92007-05-24 13:04:13 -04001767 goto out;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04001768 b2 &= ~(sbi->s_cluster_ratio - 1);
Amit Arora25d14f92007-05-24 13:04:13 -04001769 }
1770
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001771 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001772 if (b1 + len1 < b1) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04001773 len1 = EXT_MAX_BLOCKS - b1;
Amit Arora25d14f92007-05-24 13:04:13 -04001774 newext->ee_len = cpu_to_le16(len1);
1775 ret = 1;
1776 }
1777
1778 /* check for overlap */
1779 if (b1 + len1 > b2) {
1780 newext->ee_len = cpu_to_le16(b2 - b1);
1781 ret = 1;
1782 }
1783out:
1784 return ret;
1785}
1786
1787/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001788 * ext4_ext_insert_extent:
1789 * tries to merge requsted extent into the existing extent or
1790 * inserts requested extent as new one into the tree,
1791 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001792 */
1793int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1794 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04001795 struct ext4_extent *newext, int flag)
Alex Tomasa86c6182006-10-11 01:21:03 -07001796{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001797 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001798 struct ext4_extent *ex, *fex;
1799 struct ext4_extent *nearex; /* nearest extent */
1800 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001801 int depth, len, err;
1802 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001803 unsigned uninitialized = 0;
Allison Henderson55f020d2011-05-25 07:41:26 -04001804 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001805
Frank Mayhar273df552010-03-02 11:46:09 -05001806 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1807 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1808 return -EIO;
1809 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001810 depth = ext_depth(inode);
1811 ex = path[depth].p_ext;
Frank Mayhar273df552010-03-02 11:46:09 -05001812 if (unlikely(path[depth].p_hdr == NULL)) {
1813 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1814 return -EIO;
1815 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001816
1817 /* try to insert block into found extent and return */
Jiaying Zhang744692d2010-03-04 16:14:02 -05001818 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
Mingming Cao00314622009-09-28 15:49:08 -04001819 && ext4_can_extents_be_merged(inode, ex, newext)) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001820 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04001821 ext4_ext_is_uninitialized(newext),
1822 ext4_ext_get_actual_len(newext),
1823 le32_to_cpu(ex->ee_block),
1824 ext4_ext_is_uninitialized(ex),
1825 ext4_ext_get_actual_len(ex),
1826 ext4_ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001827 err = ext4_ext_get_access(handle, inode, path + depth);
1828 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001829 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001830
1831 /*
1832 * ext4_can_extents_be_merged should have checked that either
1833 * both extents are uninitialized, or both aren't. Thus we
1834 * need to check only one of them here.
1835 */
1836 if (ext4_ext_is_uninitialized(ex))
1837 uninitialized = 1;
1838 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1839 + ext4_ext_get_actual_len(newext));
1840 if (uninitialized)
1841 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001842 eh = path[depth].p_hdr;
1843 nearex = ex;
1844 goto merge;
1845 }
1846
Alex Tomasa86c6182006-10-11 01:21:03 -07001847 depth = ext_depth(inode);
1848 eh = path[depth].p_hdr;
1849 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1850 goto has_space;
1851
1852 /* probably next leaf has space for us? */
1853 fex = EXT_LAST_EXTENT(eh);
Robin Dong598dbdf2011-07-11 18:24:01 -04001854 next = EXT_MAX_BLOCKS;
1855 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
Robin Dong57187892011-07-23 21:49:07 -04001856 next = ext4_ext_next_leaf_block(path);
Robin Dong598dbdf2011-07-11 18:24:01 -04001857 if (next != EXT_MAX_BLOCKS) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001858 ext_debug("next leaf block - %u\n", next);
Alex Tomasa86c6182006-10-11 01:21:03 -07001859 BUG_ON(npath != NULL);
1860 npath = ext4_ext_find_extent(inode, next, NULL);
1861 if (IS_ERR(npath))
1862 return PTR_ERR(npath);
1863 BUG_ON(npath->p_depth != path->p_depth);
1864 eh = npath[depth].p_hdr;
1865 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001866 ext_debug("next leaf isn't full(%d)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001867 le16_to_cpu(eh->eh_entries));
1868 path = npath;
Robin Dongffb505f2011-07-11 11:43:59 -04001869 goto has_space;
Alex Tomasa86c6182006-10-11 01:21:03 -07001870 }
1871 ext_debug("next leaf has no free space(%d,%d)\n",
1872 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1873 }
1874
1875 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001876 * There is no free space in the found leaf.
1877 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001878 */
Allison Henderson55f020d2011-05-25 07:41:26 -04001879 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1880 flags = EXT4_MB_USE_ROOT_BLOCKS;
1881 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
Alex Tomasa86c6182006-10-11 01:21:03 -07001882 if (err)
1883 goto cleanup;
1884 depth = ext_depth(inode);
1885 eh = path[depth].p_hdr;
1886
1887has_space:
1888 nearex = path[depth].p_ext;
1889
Avantika Mathur7e028972006-12-06 20:41:33 -08001890 err = ext4_ext_get_access(handle, inode, path + depth);
1891 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001892 goto cleanup;
1893
1894 if (!nearex) {
1895 /* there is no extent in this leaf, create first one */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001896 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001897 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001898 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001899 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001900 ext4_ext_get_actual_len(newext));
Eric Gouriou80e675f2011-10-27 11:52:18 -04001901 nearex = EXT_FIRST_EXTENT(eh);
1902 } else {
1903 if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001904 > le32_to_cpu(nearex->ee_block)) {
Eric Gouriou80e675f2011-10-27 11:52:18 -04001905 /* Insert after */
Yongqiang Yang32de6752011-11-01 18:56:41 -04001906 ext_debug("insert %u:%llu:[%d]%d before: "
1907 "nearest %p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001908 le32_to_cpu(newext->ee_block),
Theodore Ts'obf89d162010-10-27 21:30:14 -04001909 ext4_ext_pblock(newext),
Mingming553f9002009-09-18 13:34:55 -04001910 ext4_ext_is_uninitialized(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001911 ext4_ext_get_actual_len(newext),
Eric Gouriou80e675f2011-10-27 11:52:18 -04001912 nearex);
1913 nearex++;
1914 } else {
1915 /* Insert before */
1916 BUG_ON(newext->ee_block == nearex->ee_block);
Yongqiang Yang32de6752011-11-01 18:56:41 -04001917 ext_debug("insert %u:%llu:[%d]%d after: "
1918 "nearest %p\n",
Eric Gouriou80e675f2011-10-27 11:52:18 -04001919 le32_to_cpu(newext->ee_block),
1920 ext4_ext_pblock(newext),
1921 ext4_ext_is_uninitialized(newext),
1922 ext4_ext_get_actual_len(newext),
1923 nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001924 }
Eric Gouriou80e675f2011-10-27 11:52:18 -04001925 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1926 if (len > 0) {
Yongqiang Yang32de6752011-11-01 18:56:41 -04001927 ext_debug("insert %u:%llu:[%d]%d: "
Eric Gouriou80e675f2011-10-27 11:52:18 -04001928 "move %d extents from 0x%p to 0x%p\n",
1929 le32_to_cpu(newext->ee_block),
1930 ext4_ext_pblock(newext),
1931 ext4_ext_is_uninitialized(newext),
1932 ext4_ext_get_actual_len(newext),
1933 len, nearex, nearex + 1);
1934 memmove(nearex + 1, nearex,
1935 len * sizeof(struct ext4_extent));
1936 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001937 }
1938
Marcin Slusarze8546d02008-04-17 10:38:59 -04001939 le16_add_cpu(&eh->eh_entries, 1);
Eric Gouriou80e675f2011-10-27 11:52:18 -04001940 path[depth].p_ext = nearex;
Alex Tomasa86c6182006-10-11 01:21:03 -07001941 nearex->ee_block = newext->ee_block;
Theodore Ts'obf89d162010-10-27 21:30:14 -04001942 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001943 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001944
1945merge:
HaiboLiue7bcf822012-07-09 16:29:28 -04001946 /* try to merge extents */
Jiaying Zhang744692d2010-03-04 16:14:02 -05001947 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001948 ext4_ext_try_to_merge(handle, inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001949
Alex Tomasa86c6182006-10-11 01:21:03 -07001950
1951 /* time to correct all indexes above */
1952 err = ext4_ext_correct_indexes(handle, inode, path);
1953 if (err)
1954 goto cleanup;
1955
Theodore Ts'oecb94f52012-08-17 09:44:17 -04001956 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07001957
1958cleanup:
1959 if (npath) {
1960 ext4_ext_drop_refs(npath);
1961 kfree(npath);
1962 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001963 ext4_ext_invalidate_cache(inode);
1964 return err;
1965}
1966
Lukas Czerner91dd8c12012-11-28 12:32:26 -05001967static int ext4_fill_fiemap_extents(struct inode *inode,
1968 ext4_lblk_t block, ext4_lblk_t num,
1969 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04001970{
1971 struct ext4_ext_path *path = NULL;
Lukas Czerner06348672012-11-28 12:33:22 -05001972 struct ext4_ext_cache newex;
Eric Sandeen6873fa02008-10-07 00:46:36 -04001973 struct ext4_extent *ex;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05001974 ext4_lblk_t next, next_del, start = 0, end = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04001975 ext4_lblk_t last = block + num;
Lukas Czerner91dd8c12012-11-28 12:32:26 -05001976 int exists, depth = 0, err = 0;
1977 unsigned int flags = 0;
1978 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04001979
Lukas Czernerf17722f2011-06-06 00:05:17 -04001980 while (block < last && block != EXT_MAX_BLOCKS) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04001981 num = last - block;
1982 /* find extent for this block */
Theodore Ts'ofab3a542009-12-09 21:30:02 -05001983 down_read(&EXT4_I(inode)->i_data_sem);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05001984
1985 if (path && ext_depth(inode) != depth) {
1986 /* depth was changed. we have to realloc path */
1987 kfree(path);
1988 path = NULL;
1989 }
1990
Eric Sandeen6873fa02008-10-07 00:46:36 -04001991 path = ext4_ext_find_extent(inode, block, path);
1992 if (IS_ERR(path)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05001993 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04001994 err = PTR_ERR(path);
1995 path = NULL;
1996 break;
1997 }
1998
1999 depth = ext_depth(inode);
Frank Mayhar273df552010-03-02 11:46:09 -05002000 if (unlikely(path[depth].p_hdr == NULL)) {
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002001 up_read(&EXT4_I(inode)->i_data_sem);
Frank Mayhar273df552010-03-02 11:46:09 -05002002 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2003 err = -EIO;
2004 break;
2005 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002006 ex = path[depth].p_ext;
2007 next = ext4_ext_next_allocated_block(path);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002008 ext4_ext_drop_refs(path);
Eric Sandeen6873fa02008-10-07 00:46:36 -04002009
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002010 flags = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002011 exists = 0;
2012 if (!ex) {
2013 /* there is no extent yet, so try to allocate
2014 * all requested space */
2015 start = block;
2016 end = block + num;
2017 } else if (le32_to_cpu(ex->ee_block) > block) {
2018 /* need to allocate space before found extent */
2019 start = block;
2020 end = le32_to_cpu(ex->ee_block);
2021 if (block + num < end)
2022 end = block + num;
2023 } else if (block >= le32_to_cpu(ex->ee_block)
2024 + ext4_ext_get_actual_len(ex)) {
2025 /* need to allocate space after found extent */
2026 start = block;
2027 end = block + num;
2028 if (end >= next)
2029 end = next;
2030 } else if (block >= le32_to_cpu(ex->ee_block)) {
2031 /*
2032 * some part of requested space is covered
2033 * by found extent
2034 */
2035 start = block;
2036 end = le32_to_cpu(ex->ee_block)
2037 + ext4_ext_get_actual_len(ex);
2038 if (block + num < end)
2039 end = block + num;
2040 exists = 1;
2041 } else {
2042 BUG();
2043 }
2044 BUG_ON(end <= start);
2045
2046 if (!exists) {
Lukas Czerner06348672012-11-28 12:33:22 -05002047 newex.ec_block = start;
2048 newex.ec_len = end - start;
2049 newex.ec_start = 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002050 } else {
Lukas Czerner06348672012-11-28 12:33:22 -05002051 newex.ec_block = le32_to_cpu(ex->ee_block);
2052 newex.ec_len = ext4_ext_get_actual_len(ex);
2053 newex.ec_start = ext4_ext_pblock(ex);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002054 if (ext4_ext_is_uninitialized(ex))
2055 flags |= FIEMAP_EXTENT_UNWRITTEN;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002056 }
2057
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002058 /*
Lukas Czerner06348672012-11-28 12:33:22 -05002059 * Find delayed extent and update newex accordingly. We call
2060 * it even in !exists case to find out whether newex is the
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002061 * last existing extent or not.
2062 */
Lukas Czerner06348672012-11-28 12:33:22 -05002063 next_del = ext4_find_delayed_extent(inode, &newex);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002064 if (!exists && next_del) {
2065 exists = 1;
2066 flags |= FIEMAP_EXTENT_DELALLOC;
2067 }
2068 up_read(&EXT4_I(inode)->i_data_sem);
2069
Lukas Czerner06348672012-11-28 12:33:22 -05002070 if (unlikely(newex.ec_len == 0)) {
2071 EXT4_ERROR_INODE(inode, "newex.ec_len == 0");
Frank Mayhar273df552010-03-02 11:46:09 -05002072 err = -EIO;
2073 break;
2074 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002075
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002076 /* This is possible iff next == next_del == EXT_MAX_BLOCKS */
2077 if (next == next_del) {
2078 flags |= FIEMAP_EXTENT_LAST;
2079 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2080 next != EXT_MAX_BLOCKS)) {
2081 EXT4_ERROR_INODE(inode,
2082 "next extent == %u, next "
2083 "delalloc extent = %u",
2084 next, next_del);
2085 err = -EIO;
2086 break;
2087 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002088 }
2089
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002090 if (exists) {
2091 err = fiemap_fill_next_extent(fieinfo,
Lukas Czerner06348672012-11-28 12:33:22 -05002092 (__u64)newex.ec_block << blksize_bits,
2093 (__u64)newex.ec_start << blksize_bits,
2094 (__u64)newex.ec_len << blksize_bits,
Lukas Czerner91dd8c12012-11-28 12:32:26 -05002095 flags);
2096 if (err < 0)
2097 break;
2098 if (err == 1) {
2099 err = 0;
2100 break;
2101 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04002102 }
2103
Lukas Czerner06348672012-11-28 12:33:22 -05002104 block = newex.ec_block + newex.ec_len;
Eric Sandeen6873fa02008-10-07 00:46:36 -04002105 }
2106
2107 if (path) {
2108 ext4_ext_drop_refs(path);
2109 kfree(path);
2110 }
2111
2112 return err;
2113}
2114
Avantika Mathur09b88252006-12-06 20:41:36 -08002115static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002116ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002117 __u32 len, ext4_fsblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07002118{
2119 struct ext4_ext_cache *cex;
2120 BUG_ON(len == 0);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002121 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Aditya Kalid8990242011-09-09 19:18:51 -04002122 trace_ext4_ext_put_in_cache(inode, block, len, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002123 cex = &EXT4_I(inode)->i_cached_extent;
Alex Tomasa86c6182006-10-11 01:21:03 -07002124 cex->ec_block = block;
2125 cex->ec_len = len;
2126 cex->ec_start = start;
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002127 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07002128}
2129
2130/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002131 * ext4_ext_put_gap_in_cache:
2132 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07002133 * and cache this gap
2134 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002135static void
Alex Tomasa86c6182006-10-11 01:21:03 -07002136ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002137 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07002138{
2139 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002140 unsigned long len;
2141 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002142 struct ext4_extent *ex;
2143
2144 ex = path[depth].p_ext;
2145 if (ex == NULL) {
2146 /* there is no extent yet, so gap is [0;-] */
2147 lblock = 0;
Lukas Czernerf17722f2011-06-06 00:05:17 -04002148 len = EXT_MAX_BLOCKS;
Alex Tomasa86c6182006-10-11 01:21:03 -07002149 ext_debug("cache gap(whole file):");
2150 } else if (block < le32_to_cpu(ex->ee_block)) {
2151 lblock = block;
2152 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05002153 ext_debug("cache gap(before): %u [%u:%u]",
2154 block,
2155 le32_to_cpu(ex->ee_block),
2156 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002157 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002158 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002159 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002160 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002161 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002162
2163 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05002164 ext_debug("cache gap(after): [%u:%u] %u",
2165 le32_to_cpu(ex->ee_block),
2166 ext4_ext_get_actual_len(ex),
2167 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002168 BUG_ON(next == lblock);
2169 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07002170 } else {
2171 lblock = len = 0;
2172 BUG();
2173 }
2174
Eric Sandeenbba90742008-01-28 23:58:27 -05002175 ext_debug(" -> %u:%lu\n", lblock, len);
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002176 ext4_ext_put_in_cache(inode, lblock, len, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002177}
2178
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002179/*
Lukas Czerner63fedaf2012-09-26 21:09:06 -04002180 * ext4_ext_in_cache()
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002181 * Checks to see if the given block is in the cache.
2182 * If it is, the cached extent is stored in the given
Lukas Czerner63fedaf2012-09-26 21:09:06 -04002183 * cache extent pointer.
Allison Hendersona4bb6b62011-05-25 07:41:50 -04002184 *
2185 * @inode: The files inode
2186 * @block: The block to look for in the cache
2187 * @ex: Pointer where the cached extent will be stored
2188 * if it contains block
2189 *
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002190 * Return 0 if cache is invalid; 1 if the cache is valid
2191 */
Lukas Czerner63fedaf2012-09-26 21:09:06 -04002192static int
2193ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2194 struct ext4_extent *ex)
2195{
Alex Tomasa86c6182006-10-11 01:21:03 -07002196 struct ext4_ext_cache *cex;
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002197 int ret = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002198
Theodore Ts'o60e66792010-05-17 07:00:00 -04002199 /*
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002200 * We borrow i_block_reservation_lock to protect i_cached_extent
2201 */
2202 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07002203 cex = &EXT4_I(inode)->i_cached_extent;
2204
2205 /* has cache valid data? */
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002206 if (cex->ec_len == 0)
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002207 goto errout;
Alex Tomasa86c6182006-10-11 01:21:03 -07002208
Akinobu Mita731eb1a2010-03-03 23:55:01 -05002209 if (in_range(block, cex->ec_block, cex->ec_len)) {
Lukas Czerner63fedaf2012-09-26 21:09:06 -04002210 ex->ee_block = cpu_to_le32(cex->ec_block);
2211 ext4_ext_store_pblock(ex, cex->ec_start);
2212 ex->ee_len = cpu_to_le16(cex->ec_len);
Eric Sandeenbba90742008-01-28 23:58:27 -05002213 ext_debug("%u cached by %u:%u:%llu\n",
2214 block,
2215 cex->ec_block, cex->ec_len, cex->ec_start);
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05002216 ret = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07002217 }
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002218errout:
Aditya Kalid8990242011-09-09 19:18:51 -04002219 trace_ext4_ext_in_cache(inode, block, ret);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04002220 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2221 return ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002222}
2223
2224/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002225 * ext4_ext_rm_idx:
2226 * removes index from the index block.
Alex Tomasa86c6182006-10-11 01:21:03 -07002227 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002228static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Forrest Liuc36575e2012-12-17 09:55:39 -05002229 struct ext4_ext_path *path, int depth)
Alex Tomasa86c6182006-10-11 01:21:03 -07002230{
Alex Tomasa86c6182006-10-11 01:21:03 -07002231 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002232 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07002233
2234 /* free index block */
Forrest Liuc36575e2012-12-17 09:55:39 -05002235 depth--;
2236 path = path + depth;
Theodore Ts'obf89d162010-10-27 21:30:14 -04002237 leaf = ext4_idx_pblock(path->p_idx);
Frank Mayhar273df552010-03-02 11:46:09 -05002238 if (unlikely(path->p_hdr->eh_entries == 0)) {
2239 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2240 return -EIO;
2241 }
Avantika Mathur7e028972006-12-06 20:41:33 -08002242 err = ext4_ext_get_access(handle, inode, path);
2243 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002244 return err;
Robin Dong0e1147b2011-07-27 21:29:33 -04002245
2246 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2247 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2248 len *= sizeof(struct ext4_extent_idx);
2249 memmove(path->p_idx, path->p_idx + 1, len);
2250 }
2251
Marcin Slusarze8546d02008-04-17 10:38:59 -04002252 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08002253 err = ext4_ext_dirty(handle, inode, path);
2254 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002255 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07002256 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Aditya Kalid8990242011-09-09 19:18:51 -04002257 trace_ext4_ext_rm_idx(inode, leaf);
2258
Peter Huewe7dc57612011-02-21 21:01:42 -05002259 ext4_free_blocks(handle, inode, NULL, leaf, 1,
Theodore Ts'oe6362602009-11-23 07:17:05 -05002260 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
Forrest Liuc36575e2012-12-17 09:55:39 -05002261
2262 while (--depth >= 0) {
2263 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2264 break;
2265 path--;
2266 err = ext4_ext_get_access(handle, inode, path);
2267 if (err)
2268 break;
2269 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2270 err = ext4_ext_dirty(handle, inode, path);
2271 if (err)
2272 break;
2273 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002274 return err;
2275}
2276
2277/*
Mingming Caoee12b632008-08-19 22:16:05 -04002278 * ext4_ext_calc_credits_for_single_extent:
2279 * This routine returns max. credits that needed to insert an extent
2280 * to the extent tree.
2281 * When pass the actual path, the caller should calculate credits
2282 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07002283 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002284int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07002285 struct ext4_ext_path *path)
2286{
Alex Tomasa86c6182006-10-11 01:21:03 -07002287 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04002288 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04002289 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04002290
Alex Tomasa86c6182006-10-11 01:21:03 -07002291 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07002292 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04002293 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2294
2295 /*
2296 * There are some space in the leaf tree, no
2297 * need to account for leaf block credit
2298 *
2299 * bitmaps and block group descriptor blocks
Tao Madf3ab172011-10-08 15:53:49 -04002300 * and other metadata blocks still need to be
Mingming Caoee12b632008-08-19 22:16:05 -04002301 * accounted.
2302 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002303 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04002304 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
Aneesh Kumar K.V5887e982009-07-05 23:12:04 -04002305 return ret;
Mingming Caoee12b632008-08-19 22:16:05 -04002306 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002307 }
2308
Mingming Cao525f4ed2008-08-19 22:15:58 -04002309 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04002310}
Alex Tomasa86c6182006-10-11 01:21:03 -07002311
Mingming Caoee12b632008-08-19 22:16:05 -04002312/*
2313 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2314 *
2315 * if nrblocks are fit in a single extent (chunk flag is 1), then
2316 * in the worse case, each tree level index/leaf need to be changed
2317 * if the tree split due to insert a new extent, then the old tree
2318 * index/leaf need to be updated too
2319 *
2320 * If the nrblocks are discontiguous, they could cause
2321 * the whole tree split more than once, but this is really rare.
2322 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002323int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04002324{
2325 int index;
Tao Maf19d5872012-12-10 14:05:51 -05002326 int depth;
2327
2328 /* If we are converting the inline data, only one is needed here. */
2329 if (ext4_has_inline_data(inode))
2330 return 1;
2331
2332 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002333
Mingming Caoee12b632008-08-19 22:16:05 -04002334 if (chunk)
2335 index = depth * 2;
2336 else
2337 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002338
Mingming Caoee12b632008-08-19 22:16:05 -04002339 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002340}
2341
2342static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002343 struct ext4_extent *ex,
2344 ext4_fsblk_t *partial_cluster,
2345 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002346{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002347 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002348 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002349 ext4_fsblk_t pblk;
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002350 int flags = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002351
Alex Tomasc9de5602008-01-29 00:19:52 -05002352 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
Andrey Sidorov18888cf2012-09-19 14:14:53 -04002353 flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2354 else if (ext4_should_journal_data(inode))
2355 flags |= EXT4_FREE_BLOCKS_FORGET;
2356
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002357 /*
2358 * For bigalloc file systems, we never free a partial cluster
2359 * at the beginning of the extent. Instead, we make a note
2360 * that we tried freeing the cluster, and check to see if we
2361 * need to free it on a subsequent call to ext4_remove_blocks,
2362 * or at the end of the ext4_truncate() operation.
2363 */
2364 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2365
Aditya Kalid8990242011-09-09 19:18:51 -04002366 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002367 /*
2368 * If we have a partial cluster, and it's different from the
2369 * cluster of the last block, we need to explicitly free the
2370 * partial cluster here.
2371 */
2372 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2373 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2374 ext4_free_blocks(handle, inode, NULL,
2375 EXT4_C2B(sbi, *partial_cluster),
2376 sbi->s_cluster_ratio, flags);
2377 *partial_cluster = 0;
2378 }
2379
Alex Tomasa86c6182006-10-11 01:21:03 -07002380#ifdef EXTENTS_STATS
2381 {
2382 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002383 spin_lock(&sbi->s_ext_stats_lock);
2384 sbi->s_ext_blocks += ee_len;
2385 sbi->s_ext_extents++;
2386 if (ee_len < sbi->s_ext_min)
2387 sbi->s_ext_min = ee_len;
2388 if (ee_len > sbi->s_ext_max)
2389 sbi->s_ext_max = ee_len;
2390 if (ext_depth(inode) > sbi->s_depth_max)
2391 sbi->s_depth_max = ext_depth(inode);
2392 spin_unlock(&sbi->s_ext_stats_lock);
2393 }
2394#endif
2395 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002396 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002397 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002398 ext4_lblk_t num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002399
Amit Aroraa2df2a62007-07-17 21:42:41 -04002400 num = le32_to_cpu(ex->ee_block) + ee_len - from;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002401 pblk = ext4_ext_pblock(ex) + ee_len - num;
2402 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2403 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2404 /*
2405 * If the block range to be freed didn't start at the
2406 * beginning of a cluster, and we removed the entire
2407 * extent, save the partial cluster here, since we
2408 * might need to delete if we determine that the
2409 * truncate operation has removed all of the blocks in
2410 * the cluster.
2411 */
2412 if (pblk & (sbi->s_cluster_ratio - 1) &&
2413 (ee_len == num))
2414 *partial_cluster = EXT4_B2C(sbi, pblk);
2415 else
2416 *partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002417 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002418 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002419 /* head removal */
2420 ext4_lblk_t num;
2421 ext4_fsblk_t start;
2422
2423 num = to - from;
2424 start = ext4_ext_pblock(ex);
2425
2426 ext_debug("free first %u blocks starting %llu\n", num, start);
H Hartley Sweetenee90d572011-10-18 11:01:51 -04002427 ext4_free_blocks(handle, inode, NULL, start, num, flags);
Allison Hendersond583fb82011-05-25 07:41:43 -04002428
Alex Tomasa86c6182006-10-11 01:21:03 -07002429 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002430 printk(KERN_INFO "strange request: removal(2) "
2431 "%u-%u from %u:%u\n",
2432 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002433 }
2434 return 0;
2435}
2436
Allison Hendersond583fb82011-05-25 07:41:43 -04002437
2438/*
2439 * ext4_ext_rm_leaf() Removes the extents associated with the
2440 * blocks appearing between "start" and "end", and splits the extents
2441 * if "start" and "end" appear in the same extent
2442 *
2443 * @handle: The journal handle
2444 * @inode: The files inode
2445 * @path: The path to the leaf
2446 * @start: The first block to remove
2447 * @end: The last block to remove
2448 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002449static int
2450ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002451 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2452 ext4_lblk_t start, ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002453{
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002454 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002455 int err = 0, correct_index = 0;
2456 int depth = ext_depth(inode), credits;
2457 struct ext4_extent_header *eh;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002458 ext4_lblk_t a, b;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002459 unsigned num;
2460 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002461 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002462 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002463 struct ext4_extent *ex;
2464
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002465 /* the header must be checked already in ext4_ext_remove_space() */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002466 ext_debug("truncate since %u in leaf to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002467 if (!path[depth].p_hdr)
2468 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2469 eh = path[depth].p_hdr;
Frank Mayhar273df552010-03-02 11:46:09 -05002470 if (unlikely(path[depth].p_hdr == NULL)) {
2471 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2472 return -EIO;
2473 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002474 /* find where to start removing */
2475 ex = EXT_LAST_EXTENT(eh);
2476
2477 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002478 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002479
Aditya Kalid8990242011-09-09 19:18:51 -04002480 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2481
Alex Tomasa86c6182006-10-11 01:21:03 -07002482 while (ex >= EXT_FIRST_EXTENT(eh) &&
2483 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002484
2485 if (ext4_ext_is_uninitialized(ex))
2486 uninitialized = 1;
2487 else
2488 uninitialized = 0;
2489
Mingming553f9002009-09-18 13:34:55 -04002490 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2491 uninitialized, ex_ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002492 path[depth].p_ext = ex;
2493
2494 a = ex_ee_block > start ? ex_ee_block : start;
Allison Hendersond583fb82011-05-25 07:41:43 -04002495 b = ex_ee_block+ex_ee_len - 1 < end ?
2496 ex_ee_block+ex_ee_len - 1 : end;
Alex Tomasa86c6182006-10-11 01:21:03 -07002497
2498 ext_debug(" border %u:%u\n", a, b);
2499
Allison Hendersond583fb82011-05-25 07:41:43 -04002500 /* If this extent is beyond the end of the hole, skip it */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002501 if (end < ex_ee_block) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002502 ex--;
2503 ex_ee_block = le32_to_cpu(ex->ee_block);
2504 ex_ee_len = ext4_ext_get_actual_len(ex);
2505 continue;
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002506 } else if (b != ex_ee_block + ex_ee_len - 1) {
Lukas Czernerdc1841d2012-03-19 23:07:43 -04002507 EXT4_ERROR_INODE(inode,
2508 "can not handle truncate %u:%u "
2509 "on extent %u:%u",
2510 start, end, ex_ee_block,
2511 ex_ee_block + ex_ee_len - 1);
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002512 err = -EIO;
2513 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002514 } else if (a != ex_ee_block) {
2515 /* remove tail of the extent */
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002516 num = a - ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002517 } else {
2518 /* remove whole extent: excellent! */
Alex Tomasa86c6182006-10-11 01:21:03 -07002519 num = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002520 }
Theodore Ts'o34071da2008-08-01 21:59:19 -04002521 /*
2522 * 3 for leaf, sb, and inode plus 2 (bmap and group
2523 * descriptor) for each block group; assume two block
2524 * groups plus ex_ee_len/blocks_per_block_group for
2525 * the worst case
2526 */
2527 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002528 if (ex == EXT_FIRST_EXTENT(eh)) {
2529 correct_index = 1;
2530 credits += (ext_depth(inode)) + 1;
2531 }
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002532 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002533
Jan Kara487caee2009-08-17 22:17:20 -04002534 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
Shen Feng9102e4f2008-07-11 19:27:31 -04002535 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002536 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002537
2538 err = ext4_ext_get_access(handle, inode, path + depth);
2539 if (err)
2540 goto out;
2541
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002542 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2543 a, b);
Alex Tomasa86c6182006-10-11 01:21:03 -07002544 if (err)
2545 goto out;
2546
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002547 if (num == 0)
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002548 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002549 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002550
Alex Tomasa86c6182006-10-11 01:21:03 -07002551 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002552 /*
2553 * Do not mark uninitialized if all the blocks in the
2554 * extent have been removed.
2555 */
2556 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002557 ext4_ext_mark_uninitialized(ex);
Allison Hendersond583fb82011-05-25 07:41:43 -04002558 /*
2559 * If the extent was completely released,
2560 * we need to remove it from the leaf
2561 */
2562 if (num == 0) {
Lukas Czernerf17722f2011-06-06 00:05:17 -04002563 if (end != EXT_MAX_BLOCKS - 1) {
Allison Hendersond583fb82011-05-25 07:41:43 -04002564 /*
2565 * For hole punching, we need to scoot all the
2566 * extents up when an extent is removed so that
2567 * we dont have blank extents in the middle
2568 */
2569 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2570 sizeof(struct ext4_extent));
2571
2572 /* Now get rid of the one at the end */
2573 memset(EXT_LAST_EXTENT(eh), 0,
2574 sizeof(struct ext4_extent));
2575 }
2576 le16_add_cpu(&eh->eh_entries, -1);
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002577 } else
2578 *partial_cluster = 0;
Allison Hendersond583fb82011-05-25 07:41:43 -04002579
Dmitry Monakhov750c9c42011-10-25 05:35:05 -04002580 err = ext4_ext_dirty(handle, inode, path + depth);
2581 if (err)
2582 goto out;
2583
Yongqiang Yangbf52c6f2011-11-01 18:59:26 -04002584 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
Theodore Ts'obf89d162010-10-27 21:30:14 -04002585 ext4_ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002586 ex--;
2587 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002588 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002589 }
2590
2591 if (correct_index && eh->eh_entries)
2592 err = ext4_ext_correct_indexes(handle, inode, path);
2593
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002594 /*
2595 * If there is still a entry in the leaf node, check to see if
2596 * it references the partial cluster. This is the only place
2597 * where it could; if it doesn't, we can free the cluster.
2598 */
2599 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2600 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2601 *partial_cluster)) {
2602 int flags = EXT4_FREE_BLOCKS_FORGET;
2603
2604 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2605 flags |= EXT4_FREE_BLOCKS_METADATA;
2606
2607 ext4_free_blocks(handle, inode, NULL,
2608 EXT4_C2B(sbi, *partial_cluster),
2609 sbi->s_cluster_ratio, flags);
2610 *partial_cluster = 0;
2611 }
2612
Alex Tomasa86c6182006-10-11 01:21:03 -07002613 /* if this leaf is free, then we should
2614 * remove it from index block above */
2615 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
Forrest Liuc36575e2012-12-17 09:55:39 -05002616 err = ext4_ext_rm_idx(handle, inode, path, depth);
Alex Tomasa86c6182006-10-11 01:21:03 -07002617
2618out:
2619 return err;
2620}
2621
2622/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002623 * ext4_ext_more_to_rm:
2624 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002625 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002626static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002627ext4_ext_more_to_rm(struct ext4_ext_path *path)
2628{
2629 BUG_ON(path->p_idx == NULL);
2630
2631 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2632 return 0;
2633
2634 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002635 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002636 * so we have to consider current index for truncation
2637 */
2638 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2639 return 0;
2640 return 1;
2641}
2642
Lukas Czerner5f95d212012-03-19 23:03:19 -04002643static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2644 ext4_lblk_t end)
Alex Tomasa86c6182006-10-11 01:21:03 -07002645{
2646 struct super_block *sb = inode->i_sb;
2647 int depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002648 struct ext4_ext_path *path = NULL;
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002649 ext4_fsblk_t partial_cluster = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002650 handle_t *handle;
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002651 int i = 0, err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002652
Lukas Czerner5f95d212012-03-19 23:03:19 -04002653 ext_debug("truncate since %u to %u\n", start, end);
Alex Tomasa86c6182006-10-11 01:21:03 -07002654
2655 /* probably first extent we're gonna free will be last in block */
2656 handle = ext4_journal_start(inode, depth + 1);
2657 if (IS_ERR(handle))
2658 return PTR_ERR(handle);
2659
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002660again:
Alex Tomasa86c6182006-10-11 01:21:03 -07002661 ext4_ext_invalidate_cache(inode);
2662
Aditya Kalid8990242011-09-09 19:18:51 -04002663 trace_ext4_ext_remove_space(inode, start, depth);
2664
Alex Tomasa86c6182006-10-11 01:21:03 -07002665 /*
Lukas Czerner5f95d212012-03-19 23:03:19 -04002666 * Check if we are removing extents inside the extent tree. If that
2667 * is the case, we are going to punch a hole inside the extent tree
2668 * so we have to check whether we need to split the extent covering
2669 * the last block to remove so we can easily remove the part of it
2670 * in ext4_ext_rm_leaf().
2671 */
2672 if (end < EXT_MAX_BLOCKS - 1) {
2673 struct ext4_extent *ex;
2674 ext4_lblk_t ee_block;
2675
2676 /* find extent for this block */
2677 path = ext4_ext_find_extent(inode, end, NULL);
2678 if (IS_ERR(path)) {
2679 ext4_journal_stop(handle);
2680 return PTR_ERR(path);
2681 }
2682 depth = ext_depth(inode);
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002683 /* Leaf not may not exist only if inode has no blocks at all */
Lukas Czerner5f95d212012-03-19 23:03:19 -04002684 ex = path[depth].p_ext;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002685 if (!ex) {
Dmitry Monakhov6f2080e2012-09-30 23:03:50 -04002686 if (depth) {
2687 EXT4_ERROR_INODE(inode,
2688 "path[%d].p_hdr == NULL",
2689 depth);
2690 err = -EIO;
2691 }
2692 goto out;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002693 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002694
2695 ee_block = le32_to_cpu(ex->ee_block);
2696
2697 /*
2698 * See if the last block is inside the extent, if so split
2699 * the extent at 'end' block so we can easily remove the
2700 * tail of the first part of the split extent in
2701 * ext4_ext_rm_leaf().
2702 */
2703 if (end >= ee_block &&
2704 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2705 int split_flag = 0;
2706
2707 if (ext4_ext_is_uninitialized(ex))
2708 split_flag = EXT4_EXT_MARK_UNINIT1 |
2709 EXT4_EXT_MARK_UNINIT2;
2710
2711 /*
2712 * Split the extent in two so that 'end' is the last
2713 * block in the first new extent
2714 */
2715 err = ext4_split_extent_at(handle, inode, path,
2716 end + 1, split_flag,
2717 EXT4_GET_BLOCKS_PRE_IO |
2718 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2719
2720 if (err < 0)
2721 goto out;
2722 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002723 }
Lukas Czerner5f95d212012-03-19 23:03:19 -04002724 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002725 * We start scanning from right side, freeing all the blocks
2726 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002727 */
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002728 depth = ext_depth(inode);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002729 if (path) {
2730 int k = i = depth;
2731 while (--k > 0)
2732 path[k].p_block =
2733 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2734 } else {
2735 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2736 GFP_NOFS);
2737 if (path == NULL) {
2738 ext4_journal_stop(handle);
2739 return -ENOMEM;
2740 }
2741 path[0].p_depth = depth;
2742 path[0].p_hdr = ext_inode_hdr(inode);
Theodore Ts'o89a4e482012-08-17 08:54:52 -04002743 i = 0;
Lukas Czerner5f95d212012-03-19 23:03:19 -04002744
Ashish Sangwan968dee72012-07-22 22:49:08 -04002745 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2746 err = -EIO;
2747 goto out;
2748 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002749 }
Ashish Sangwan968dee72012-07-22 22:49:08 -04002750 err = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002751
2752 while (i >= 0 && err == 0) {
2753 if (i == depth) {
2754 /* this is leaf block */
Allison Hendersond583fb82011-05-25 07:41:43 -04002755 err = ext4_ext_rm_leaf(handle, inode, path,
Theodore Ts'o0aa06002011-09-09 18:54:51 -04002756 &partial_cluster, start,
Lukas Czerner5f95d212012-03-19 23:03:19 -04002757 end);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002758 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002759 brelse(path[i].p_bh);
2760 path[i].p_bh = NULL;
2761 i--;
2762 continue;
2763 }
2764
2765 /* this is index block */
2766 if (!path[i].p_hdr) {
2767 ext_debug("initialize header\n");
2768 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002769 }
2770
Alex Tomasa86c6182006-10-11 01:21:03 -07002771 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002772 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002773 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2774 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2775 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2776 path[i].p_hdr,
2777 le16_to_cpu(path[i].p_hdr->eh_entries));
2778 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002779 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002780 path[i].p_idx--;
2781 }
2782
2783 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2784 i, EXT_FIRST_INDEX(path[i].p_hdr),
2785 path[i].p_idx);
2786 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002787 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002788 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002789 ext_debug("move to level %d (block %llu)\n",
Theodore Ts'obf89d162010-10-27 21:30:14 -04002790 i + 1, ext4_idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002791 memset(path + i + 1, 0, sizeof(*path));
Theodore Ts'obf89d162010-10-27 21:30:14 -04002792 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002793 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002794 /* should we reset i_size? */
2795 err = -EIO;
2796 break;
2797 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002798 if (WARN_ON(i + 1 > depth)) {
2799 err = -EIO;
2800 break;
2801 }
Darrick J. Wongf8489122012-04-29 18:21:10 -04002802 if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2803 depth - i - 1, bh)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002804 err = -EIO;
2805 break;
2806 }
2807 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002808
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002809 /* save actual number of indexes since this
2810 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002811 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2812 i++;
2813 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002814 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002815 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002816 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002817 * handle must be already prepared by the
2818 * truncatei_leaf() */
Forrest Liuc36575e2012-12-17 09:55:39 -05002819 err = ext4_ext_rm_idx(handle, inode, path, i);
Alex Tomasa86c6182006-10-11 01:21:03 -07002820 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002821 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002822 brelse(path[i].p_bh);
2823 path[i].p_bh = NULL;
2824 i--;
2825 ext_debug("return to level %d\n", i);
2826 }
2827 }
2828
Aditya Kalid8990242011-09-09 19:18:51 -04002829 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2830 path->p_hdr->eh_entries);
2831
Aditya Kali7b415bf2011-09-09 19:04:51 -04002832 /* If we still have something in the partial cluster and we have removed
2833 * even the first extent, then we should free the blocks in the partial
2834 * cluster as well. */
2835 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2836 int flags = EXT4_FREE_BLOCKS_FORGET;
2837
2838 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2839 flags |= EXT4_FREE_BLOCKS_METADATA;
2840
2841 ext4_free_blocks(handle, inode, NULL,
2842 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2843 EXT4_SB(sb)->s_cluster_ratio, flags);
2844 partial_cluster = 0;
2845 }
2846
Alex Tomasa86c6182006-10-11 01:21:03 -07002847 /* TODO: flexible tree reduction should be here */
2848 if (path->p_hdr->eh_entries == 0) {
2849 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002850 * truncate to zero freed all the tree,
2851 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002852 */
2853 err = ext4_ext_get_access(handle, inode, path);
2854 if (err == 0) {
2855 ext_inode_hdr(inode)->eh_depth = 0;
2856 ext_inode_hdr(inode)->eh_max =
Theodore Ts'o55ad63b2009-08-28 10:40:33 -04002857 cpu_to_le16(ext4_ext_space_root(inode, 0));
Alex Tomasa86c6182006-10-11 01:21:03 -07002858 err = ext4_ext_dirty(handle, inode, path);
2859 }
2860 }
2861out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002862 ext4_ext_drop_refs(path);
2863 kfree(path);
Ashish Sangwan968dee72012-07-22 22:49:08 -04002864 if (err == -EAGAIN) {
2865 path = NULL;
Dmitry Monakhov0617b832010-05-17 01:00:00 -04002866 goto again;
Ashish Sangwan968dee72012-07-22 22:49:08 -04002867 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002868 ext4_journal_stop(handle);
2869
2870 return err;
2871}
2872
2873/*
2874 * called at mount time
2875 */
2876void ext4_ext_init(struct super_block *sb)
2877{
2878 /*
2879 * possible initialization would be here
2880 */
2881
Theodore Ts'o83982b62009-01-06 14:53:16 -05002882 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o90576c02009-09-29 15:51:30 -04002883#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
Theodore Ts'o92b97812012-03-19 23:41:49 -04002884 printk(KERN_INFO "EXT4-fs: file extents enabled"
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002885#ifdef AGGRESSIVE_TEST
Theodore Ts'o92b97812012-03-19 23:41:49 -04002886 ", aggressive tests"
Alex Tomasa86c6182006-10-11 01:21:03 -07002887#endif
2888#ifdef CHECK_BINSEARCH
Theodore Ts'o92b97812012-03-19 23:41:49 -04002889 ", check binsearch"
Alex Tomasa86c6182006-10-11 01:21:03 -07002890#endif
2891#ifdef EXTENTS_STATS
Theodore Ts'o92b97812012-03-19 23:41:49 -04002892 ", stats"
Alex Tomasa86c6182006-10-11 01:21:03 -07002893#endif
Theodore Ts'o92b97812012-03-19 23:41:49 -04002894 "\n");
Theodore Ts'o90576c02009-09-29 15:51:30 -04002895#endif
Alex Tomasa86c6182006-10-11 01:21:03 -07002896#ifdef EXTENTS_STATS
2897 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2898 EXT4_SB(sb)->s_ext_min = 1 << 30;
2899 EXT4_SB(sb)->s_ext_max = 0;
2900#endif
2901 }
2902}
2903
2904/*
2905 * called at umount time
2906 */
2907void ext4_ext_release(struct super_block *sb)
2908{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002909 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002910 return;
2911
2912#ifdef EXTENTS_STATS
2913 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2914 struct ext4_sb_info *sbi = EXT4_SB(sb);
2915 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2916 sbi->s_ext_blocks, sbi->s_ext_extents,
2917 sbi->s_ext_blocks / sbi->s_ext_extents);
2918 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2919 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2920 }
2921#endif
2922}
2923
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002924/* FIXME!! we need to try to merge to left or right after zero-out */
2925static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2926{
Lukas Czerner24075182010-10-27 21:30:06 -04002927 ext4_fsblk_t ee_pblock;
2928 unsigned int ee_len;
Jing Zhangb7203032010-05-12 00:00:00 -04002929 int ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002930
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002931 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'obf89d162010-10-27 21:30:14 -04002932 ee_pblock = ext4_ext_pblock(ex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002933
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002934 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
Lukas Czerner24075182010-10-27 21:30:06 -04002935 if (ret > 0)
2936 ret = 0;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002937
Lukas Czerner24075182010-10-27 21:30:06 -04002938 return ret;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002939}
2940
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002941/*
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002942 * ext4_split_extent_at() splits an extent at given block.
2943 *
2944 * @handle: the journal handle
2945 * @inode: the file inode
2946 * @path: the path to the extent
2947 * @split: the logical block where the extent is splitted.
2948 * @split_flags: indicates if the extent could be zeroout if split fails, and
2949 * the states(init or uninit) of new extents.
2950 * @flags: flags used to insert new extent to extent tree.
2951 *
2952 *
2953 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2954 * of which are deterimined by split_flag.
2955 *
2956 * There are two cases:
2957 * a> the extent are splitted into two extent.
2958 * b> split is not needed, and just mark the extent.
2959 *
2960 * return 0 on success.
2961 */
2962static int ext4_split_extent_at(handle_t *handle,
2963 struct inode *inode,
2964 struct ext4_ext_path *path,
2965 ext4_lblk_t split,
2966 int split_flag,
2967 int flags)
2968{
2969 ext4_fsblk_t newblock;
2970 ext4_lblk_t ee_block;
2971 struct ext4_extent *ex, newex, orig_ex;
2972 struct ext4_extent *ex2 = NULL;
2973 unsigned int ee_len, depth;
2974 int err = 0;
2975
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04002976 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2977 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2978
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04002979 ext_debug("ext4_split_extents_at: inode %lu, logical"
2980 "block %llu\n", inode->i_ino, (unsigned long long)split);
2981
2982 ext4_ext_show_leaf(inode, path);
2983
2984 depth = ext_depth(inode);
2985 ex = path[depth].p_ext;
2986 ee_block = le32_to_cpu(ex->ee_block);
2987 ee_len = ext4_ext_get_actual_len(ex);
2988 newblock = split - ee_block + ext4_ext_pblock(ex);
2989
2990 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2991
2992 err = ext4_ext_get_access(handle, inode, path + depth);
2993 if (err)
2994 goto out;
2995
2996 if (split == ee_block) {
2997 /*
2998 * case b: block @split is the block that the extent begins with
2999 * then we just change the state of the extent, and splitting
3000 * is not needed.
3001 */
3002 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3003 ext4_ext_mark_uninitialized(ex);
3004 else
3005 ext4_ext_mark_initialized(ex);
3006
3007 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003008 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003009
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003010 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003011 goto out;
3012 }
3013
3014 /* case a */
3015 memcpy(&orig_ex, ex, sizeof(orig_ex));
3016 ex->ee_len = cpu_to_le16(split - ee_block);
3017 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3018 ext4_ext_mark_uninitialized(ex);
3019
3020 /*
3021 * path may lead to new leaf, not to original leaf any more
3022 * after ext4_ext_insert_extent() returns,
3023 */
3024 err = ext4_ext_dirty(handle, inode, path + depth);
3025 if (err)
3026 goto fix_extent_len;
3027
3028 ex2 = &newex;
3029 ex2->ee_block = cpu_to_le32(split);
3030 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3031 ext4_ext_store_pblock(ex2, newblock);
3032 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3033 ext4_ext_mark_uninitialized(ex2);
3034
3035 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3036 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003037 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3038 if (split_flag & EXT4_EXT_DATA_VALID1)
3039 err = ext4_ext_zeroout(inode, ex2);
3040 else
3041 err = ext4_ext_zeroout(inode, ex);
3042 } else
3043 err = ext4_ext_zeroout(inode, &orig_ex);
3044
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003045 if (err)
3046 goto fix_extent_len;
3047 /* update the extent length and mark as initialized */
Al Viroaf1584f2012-04-12 20:32:25 -04003048 ex->ee_len = cpu_to_le16(ee_len);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003049 ext4_ext_try_to_merge(handle, inode, path, ex);
3050 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003051 goto out;
3052 } else if (err)
3053 goto fix_extent_len;
3054
3055out:
3056 ext4_ext_show_leaf(inode, path);
3057 return err;
3058
3059fix_extent_len:
3060 ex->ee_len = orig_ex.ee_len;
3061 ext4_ext_dirty(handle, inode, path + depth);
3062 return err;
3063}
3064
3065/*
3066 * ext4_split_extents() splits an extent and mark extent which is covered
3067 * by @map as split_flags indicates
3068 *
3069 * It may result in splitting the extent into multiple extents (upto three)
3070 * There are three possibilities:
3071 * a> There is no split required
3072 * b> Splits in two extents: Split is happening at either end of the extent
3073 * c> Splits in three extents: Somone is splitting in middle of the extent
3074 *
3075 */
3076static int ext4_split_extent(handle_t *handle,
3077 struct inode *inode,
3078 struct ext4_ext_path *path,
3079 struct ext4_map_blocks *map,
3080 int split_flag,
3081 int flags)
3082{
3083 ext4_lblk_t ee_block;
3084 struct ext4_extent *ex;
3085 unsigned int ee_len, depth;
3086 int err = 0;
3087 int uninitialized;
3088 int split_flag1, flags1;
3089
3090 depth = ext_depth(inode);
3091 ex = path[depth].p_ext;
3092 ee_block = le32_to_cpu(ex->ee_block);
3093 ee_len = ext4_ext_get_actual_len(ex);
3094 uninitialized = ext4_ext_is_uninitialized(ex);
3095
3096 if (map->m_lblk + map->m_len < ee_block + ee_len) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003097 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003098 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3099 if (uninitialized)
3100 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3101 EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003102 if (split_flag & EXT4_EXT_DATA_VALID2)
3103 split_flag1 |= EXT4_EXT_DATA_VALID1;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003104 err = ext4_split_extent_at(handle, inode, path,
3105 map->m_lblk + map->m_len, split_flag1, flags1);
Yongqiang Yang93917412011-05-22 20:49:12 -04003106 if (err)
3107 goto out;
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003108 }
3109
3110 ext4_ext_drop_refs(path);
3111 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3112 if (IS_ERR(path))
3113 return PTR_ERR(path);
3114
3115 if (map->m_lblk >= ee_block) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003116 split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
3117 EXT4_EXT_DATA_VALID2);
Yongqiang Yang47ea3bb2011-05-03 12:23:07 -04003118 if (uninitialized)
3119 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3120 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3121 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3122 err = ext4_split_extent_at(handle, inode, path,
3123 map->m_lblk, split_flag1, flags);
3124 if (err)
3125 goto out;
3126 }
3127
3128 ext4_ext_show_leaf(inode, path);
3129out:
3130 return err ? err : map->m_len;
3131}
3132
Amit Arora56055d32007-07-17 21:42:38 -04003133/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003134 * This function is called by ext4_ext_map_blocks() if someone tries to write
Amit Arora56055d32007-07-17 21:42:38 -04003135 * to an uninitialized extent. It may result in splitting the uninitialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003136 * extent into multiple extents (up to three - one initialized and two
Amit Arora56055d32007-07-17 21:42:38 -04003137 * uninitialized).
3138 * There are three possibilities:
3139 * a> There is no split required: Entire extent should be initialized
3140 * b> Splits in two extents: Write is happening at either end of the extent
3141 * c> Splits in three extents: Somone is writing in middle of the extent
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003142 *
3143 * Pre-conditions:
3144 * - The extent pointed to by 'path' is uninitialized.
3145 * - The extent pointed to by 'path' contains a superset
3146 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3147 *
3148 * Post-conditions on success:
3149 * - the returned value is the number of blocks beyond map->l_lblk
3150 * that are allocated and initialized.
3151 * It is guaranteed to be >= map->m_len.
Amit Arora56055d32007-07-17 21:42:38 -04003152 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003153static int ext4_ext_convert_to_initialized(handle_t *handle,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003154 struct inode *inode,
3155 struct ext4_map_blocks *map,
3156 struct ext4_ext_path *path)
Amit Arora56055d32007-07-17 21:42:38 -04003157{
Zheng Liu67a5da52012-08-17 09:54:17 -04003158 struct ext4_sb_info *sbi;
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003159 struct ext4_extent_header *eh;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003160 struct ext4_map_blocks split_map;
3161 struct ext4_extent zero_ex;
3162 struct ext4_extent *ex;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003163 ext4_lblk_t ee_block, eof_block;
Dan Carpenterf85b2872011-10-26 03:42:36 -04003164 unsigned int ee_len, depth;
Zheng Liu67a5da52012-08-17 09:54:17 -04003165 int allocated, max_zeroout = 0;
Amit Arora56055d32007-07-17 21:42:38 -04003166 int err = 0;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003167 int split_flag = 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003168
3169 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3170 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003171 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003172
Zheng Liu67a5da52012-08-17 09:54:17 -04003173 sbi = EXT4_SB(inode->i_sb);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003174 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3175 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003176 if (eof_block < map->m_lblk + map->m_len)
3177 eof_block = map->m_lblk + map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003178
3179 depth = ext_depth(inode);
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003180 eh = path[depth].p_hdr;
Amit Arora56055d32007-07-17 21:42:38 -04003181 ex = path[depth].p_ext;
3182 ee_block = le32_to_cpu(ex->ee_block);
3183 ee_len = ext4_ext_get_actual_len(ex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003184 allocated = ee_len - (map->m_lblk - ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003185
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003186 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3187
3188 /* Pre-conditions */
3189 BUG_ON(!ext4_ext_is_uninitialized(ex));
3190 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
Eric Gouriou6f91bc52011-10-27 11:43:23 -04003191
3192 /*
3193 * Attempt to transfer newly initialized blocks from the currently
3194 * uninitialized extent to its left neighbor. This is much cheaper
3195 * than an insertion followed by a merge as those involve costly
3196 * memmove() calls. This is the common case in steady state for
3197 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3198 * writes.
3199 *
3200 * Limitations of the current logic:
3201 * - L1: we only deal with writes at the start of the extent.
3202 * The approach could be extended to writes at the end
3203 * of the extent but this scenario was deemed less common.
3204 * - L2: we do not deal with writes covering the whole extent.
3205 * This would require removing the extent if the transfer
3206 * is possible.
3207 * - L3: we only attempt to merge with an extent stored in the
3208 * same extent tree node.
3209 */
3210 if ((map->m_lblk == ee_block) && /*L1*/
3211 (map->m_len < ee_len) && /*L2*/
3212 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3213 struct ext4_extent *prev_ex;
3214 ext4_lblk_t prev_lblk;
3215 ext4_fsblk_t prev_pblk, ee_pblk;
3216 unsigned int prev_len, write_len;
3217
3218 prev_ex = ex - 1;
3219 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3220 prev_len = ext4_ext_get_actual_len(prev_ex);
3221 prev_pblk = ext4_ext_pblock(prev_ex);
3222 ee_pblk = ext4_ext_pblock(ex);
3223 write_len = map->m_len;
3224
3225 /*
3226 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3227 * upon those conditions:
3228 * - C1: prev_ex is initialized,
3229 * - C2: prev_ex is logically abutting ex,
3230 * - C3: prev_ex is physically abutting ex,
3231 * - C4: prev_ex can receive the additional blocks without
3232 * overflowing the (initialized) length limit.
3233 */
3234 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3235 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3236 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3237 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3238 err = ext4_ext_get_access(handle, inode, path + depth);
3239 if (err)
3240 goto out;
3241
3242 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3243 map, ex, prev_ex);
3244
3245 /* Shift the start of ex by 'write_len' blocks */
3246 ex->ee_block = cpu_to_le32(ee_block + write_len);
3247 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3248 ex->ee_len = cpu_to_le16(ee_len - write_len);
3249 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3250
3251 /* Extend prev_ex by 'write_len' blocks */
3252 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3253
3254 /* Mark the block containing both extents as dirty */
3255 ext4_ext_dirty(handle, inode, path + depth);
3256
3257 /* Update path to point to the right extent */
3258 path[depth].p_ext = prev_ex;
3259
3260 /* Result: number of initialized blocks past m_lblk */
3261 allocated = write_len;
3262 goto out;
3263 }
3264 }
3265
Yongqiang Yang667eff32011-05-03 12:25:07 -04003266 WARN_ON(map->m_lblk < ee_block);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003267 /*
3268 * It is safe to convert extent to initialized via explicit
3269 * zeroout only if extent is fully insde i_size or new_size.
3270 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003271 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003272
Zheng Liu67a5da52012-08-17 09:54:17 -04003273 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3274 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3275 inode->i_sb->s_blocksize_bits;
3276
3277 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3278 if (max_zeroout && (ee_len <= max_zeroout)) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003279 err = ext4_ext_zeroout(inode, ex);
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04003280 if (err)
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003281 goto out;
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04003282
3283 err = ext4_ext_get_access(handle, inode, path + depth);
3284 if (err)
3285 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003286 ext4_ext_mark_initialized(ex);
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003287 ext4_ext_try_to_merge(handle, inode, path, ex);
3288 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003289 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04003290 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003291
Amit Arora56055d32007-07-17 21:42:38 -04003292 /*
Yongqiang Yang667eff32011-05-03 12:25:07 -04003293 * four cases:
3294 * 1. split the extent into three extents.
3295 * 2. split the extent into two extents, zeroout the first half.
3296 * 3. split the extent into two extents, zeroout the second half.
3297 * 4. split the extent into two extents with out zeroout.
Amit Arora56055d32007-07-17 21:42:38 -04003298 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003299 split_map.m_lblk = map->m_lblk;
3300 split_map.m_len = map->m_len;
3301
Zheng Liu67a5da52012-08-17 09:54:17 -04003302 if (max_zeroout && (allocated > map->m_len)) {
3303 if (allocated <= max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003304 /* case 3 */
3305 zero_ex.ee_block =
Allison Henderson9b940f82011-05-16 10:11:09 -04003306 cpu_to_le32(map->m_lblk);
3307 zero_ex.ee_len = cpu_to_le16(allocated);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003308 ext4_ext_store_pblock(&zero_ex,
3309 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3310 err = ext4_ext_zeroout(inode, &zero_ex);
Amit Arora56055d32007-07-17 21:42:38 -04003311 if (err)
3312 goto out;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003313 split_map.m_lblk = map->m_lblk;
3314 split_map.m_len = allocated;
Zheng Liu67a5da52012-08-17 09:54:17 -04003315 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
Yongqiang Yang667eff32011-05-03 12:25:07 -04003316 /* case 2 */
3317 if (map->m_lblk != ee_block) {
3318 zero_ex.ee_block = ex->ee_block;
3319 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3320 ee_block);
3321 ext4_ext_store_pblock(&zero_ex,
3322 ext4_ext_pblock(ex));
3323 err = ext4_ext_zeroout(inode, &zero_ex);
3324 if (err)
3325 goto out;
3326 }
3327
Yongqiang Yang667eff32011-05-03 12:25:07 -04003328 split_map.m_lblk = ee_block;
Allison Henderson9b940f82011-05-16 10:11:09 -04003329 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3330 allocated = map->m_len;
Amit Arora56055d32007-07-17 21:42:38 -04003331 }
3332 }
Yongqiang Yang667eff32011-05-03 12:25:07 -04003333
3334 allocated = ext4_split_extent(handle, inode, path,
Zheng Liu67a5da52012-08-17 09:54:17 -04003335 &split_map, split_flag, 0);
Yongqiang Yang667eff32011-05-03 12:25:07 -04003336 if (allocated < 0)
3337 err = allocated;
3338
Amit Arora56055d32007-07-17 21:42:38 -04003339out:
3340 return err ? err : allocated;
3341}
3342
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003343/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003344 * This function is called by ext4_ext_map_blocks() from
Mingming Cao00314622009-09-28 15:49:08 -04003345 * ext4_get_blocks_dio_write() when DIO to write
3346 * to an uninitialized extent.
3347 *
Paul Bollefd018fe2011-02-15 00:05:43 +01003348 * Writing to an uninitialized extent may result in splitting the uninitialized
Wang Sheng-Hui30cb27d2012-08-18 22:38:07 -04003349 * extent into multiple initialized/uninitialized extents (up to three)
Mingming Cao00314622009-09-28 15:49:08 -04003350 * There are three possibilities:
3351 * a> There is no split required: Entire extent should be uninitialized
3352 * b> Splits in two extents: Write is happening at either end of the extent
3353 * c> Splits in three extents: Somone is writing in middle of the extent
3354 *
3355 * One of more index blocks maybe needed if the extent tree grow after
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003356 * the uninitialized extent split. To prevent ENOSPC occur at the IO
Mingming Cao00314622009-09-28 15:49:08 -04003357 * complete, we need to split the uninitialized extent before DIO submit
Uwe Kleine-König421f91d2010-06-11 12:17:00 +02003358 * the IO. The uninitialized extent called at this time will be split
Mingming Cao00314622009-09-28 15:49:08 -04003359 * into three uninitialized extent(at most). After IO complete, the part
3360 * being filled will be convert to initialized by the end_io callback function
3361 * via ext4_convert_unwritten_extents().
Mingmingba230c32009-11-06 04:01:23 -05003362 *
3363 * Returns the size of uninitialized extent to be written on success.
Mingming Cao00314622009-09-28 15:49:08 -04003364 */
3365static int ext4_split_unwritten_extents(handle_t *handle,
3366 struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003367 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003368 struct ext4_ext_path *path,
Mingming Cao00314622009-09-28 15:49:08 -04003369 int flags)
3370{
Yongqiang Yang667eff32011-05-03 12:25:07 -04003371 ext4_lblk_t eof_block;
3372 ext4_lblk_t ee_block;
3373 struct ext4_extent *ex;
3374 unsigned int ee_len;
3375 int split_flag = 0, depth;
Mingming Cao00314622009-09-28 15:49:08 -04003376
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003377 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3378 "block %llu, max_blocks %u\n", inode->i_ino,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003379 (unsigned long long)map->m_lblk, map->m_len);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003380
3381 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3382 inode->i_sb->s_blocksize_bits;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003383 if (eof_block < map->m_lblk + map->m_len)
3384 eof_block = map->m_lblk + map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003385 /*
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003386 * It is safe to convert extent to initialized via explicit
3387 * zeroout only if extent is fully insde i_size or new_size.
3388 */
Yongqiang Yang667eff32011-05-03 12:25:07 -04003389 depth = ext_depth(inode);
3390 ex = path[depth].p_ext;
3391 ee_block = le32_to_cpu(ex->ee_block);
3392 ee_len = ext4_ext_get_actual_len(ex);
Dmitry Monakhov21ca0872010-05-16 06:00:00 -04003393
Yongqiang Yang667eff32011-05-03 12:25:07 -04003394 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3395 split_flag |= EXT4_EXT_MARK_UNINIT2;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003396 if (flags & EXT4_GET_BLOCKS_CONVERT)
3397 split_flag |= EXT4_EXT_DATA_VALID2;
Yongqiang Yang667eff32011-05-03 12:25:07 -04003398 flags |= EXT4_GET_BLOCKS_PRE_IO;
3399 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
Mingming Cao00314622009-09-28 15:49:08 -04003400}
Yongqiang Yang197217a2011-05-03 11:45:29 -04003401
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003402static int ext4_convert_unwritten_extents_endio(handle_t *handle,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003403 struct inode *inode,
3404 struct ext4_map_blocks *map,
3405 struct ext4_ext_path *path)
Mingming Cao00314622009-09-28 15:49:08 -04003406{
3407 struct ext4_extent *ex;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003408 ext4_lblk_t ee_block;
3409 unsigned int ee_len;
Mingming Cao00314622009-09-28 15:49:08 -04003410 int depth;
3411 int err = 0;
Mingming Cao00314622009-09-28 15:49:08 -04003412
3413 depth = ext_depth(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003414 ex = path[depth].p_ext;
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003415 ee_block = le32_to_cpu(ex->ee_block);
3416 ee_len = ext4_ext_get_actual_len(ex);
Mingming Cao00314622009-09-28 15:49:08 -04003417
Yongqiang Yang197217a2011-05-03 11:45:29 -04003418 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3419 "block %llu, max_blocks %u\n", inode->i_ino,
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003420 (unsigned long long)ee_block, ee_len);
3421
3422 /* If extent is larger than requested then split is required */
3423 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3424 err = ext4_split_unwritten_extents(handle, inode, map, path,
3425 EXT4_GET_BLOCKS_CONVERT);
3426 if (err < 0)
3427 goto out;
3428 ext4_ext_drop_refs(path);
3429 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3430 if (IS_ERR(path)) {
3431 err = PTR_ERR(path);
3432 goto out;
3433 }
3434 depth = ext_depth(inode);
3435 ex = path[depth].p_ext;
3436 }
Yongqiang Yang197217a2011-05-03 11:45:29 -04003437
Mingming Cao00314622009-09-28 15:49:08 -04003438 err = ext4_ext_get_access(handle, inode, path + depth);
3439 if (err)
3440 goto out;
3441 /* first mark the extent as initialized */
3442 ext4_ext_mark_initialized(ex);
3443
Yongqiang Yang197217a2011-05-03 11:45:29 -04003444 /* note: ext4_ext_correct_indexes() isn't needed here because
3445 * borders are not changed
Mingming Cao00314622009-09-28 15:49:08 -04003446 */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003447 ext4_ext_try_to_merge(handle, inode, path, ex);
Yongqiang Yang197217a2011-05-03 11:45:29 -04003448
Mingming Cao00314622009-09-28 15:49:08 -04003449 /* Mark modified extent as dirty */
Theodore Ts'oecb94f52012-08-17 09:44:17 -04003450 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
Mingming Cao00314622009-09-28 15:49:08 -04003451out:
3452 ext4_ext_show_leaf(inode, path);
3453 return err;
3454}
3455
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003456static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3457 sector_t block, int count)
3458{
3459 int i;
3460 for (i = 0; i < count; i++)
3461 unmap_underlying_metadata(bdev, block + i);
3462}
3463
Theodore Ts'o58590b02010-10-27 21:23:12 -04003464/*
3465 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3466 */
3467static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
Eric Sandeend002ebf2011-01-10 13:03:35 -05003468 ext4_lblk_t lblk,
Theodore Ts'o58590b02010-10-27 21:23:12 -04003469 struct ext4_ext_path *path,
3470 unsigned int len)
3471{
3472 int i, depth;
3473 struct ext4_extent_header *eh;
Sergey Senozhatsky65922cb2011-03-23 14:08:27 -04003474 struct ext4_extent *last_ex;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003475
3476 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3477 return 0;
3478
3479 depth = ext_depth(inode);
3480 eh = path[depth].p_hdr;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003481
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003482 /*
3483 * We're going to remove EOFBLOCKS_FL entirely in future so we
3484 * do not care for this case anymore. Simply remove the flag
3485 * if there are no extents.
3486 */
3487 if (unlikely(!eh->eh_entries))
3488 goto out;
Theodore Ts'o58590b02010-10-27 21:23:12 -04003489 last_ex = EXT_LAST_EXTENT(eh);
3490 /*
3491 * We should clear the EOFBLOCKS_FL flag if we are writing the
3492 * last block in the last extent in the file. We test this by
3493 * first checking to see if the caller to
3494 * ext4_ext_get_blocks() was interested in the last block (or
3495 * a block beyond the last block) in the current extent. If
3496 * this turns out to be false, we can bail out from this
3497 * function immediately.
3498 */
Eric Sandeend002ebf2011-01-10 13:03:35 -05003499 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
Theodore Ts'o58590b02010-10-27 21:23:12 -04003500 ext4_ext_get_actual_len(last_ex))
3501 return 0;
3502 /*
3503 * If the caller does appear to be planning to write at or
3504 * beyond the end of the current extent, we then test to see
3505 * if the current extent is the last extent in the file, by
3506 * checking to make sure it was reached via the rightmost node
3507 * at each level of the tree.
3508 */
3509 for (i = depth-1; i >= 0; i--)
3510 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3511 return 0;
Lukas Czernerafcff5d2012-03-21 21:47:55 -04003512out:
Theodore Ts'o58590b02010-10-27 21:23:12 -04003513 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3514 return ext4_mark_inode_dirty(handle, inode);
3515}
3516
Aditya Kali7b415bf2011-09-09 19:04:51 -04003517/**
3518 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3519 *
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003520 * Return 1 if there is a delalloc block in the range, otherwise 0.
Aditya Kali7b415bf2011-09-09 19:04:51 -04003521 */
3522static int ext4_find_delalloc_range(struct inode *inode,
3523 ext4_lblk_t lblk_start,
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003524 ext4_lblk_t lblk_end)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003525{
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003526 struct extent_status es;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003527
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003528 es.start = lblk_start;
3529 ext4_es_find_extent(inode, &es);
3530 if (es.len == 0)
3531 return 0; /* there is no delay extent in this tree */
3532 else if (es.start <= lblk_start && lblk_start < es.start + es.len)
3533 return 1;
3534 else if (lblk_start <= es.start && es.start <= lblk_end)
3535 return 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003536 else
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003537 return 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04003538}
3539
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003540int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
Aditya Kali7b415bf2011-09-09 19:04:51 -04003541{
3542 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3543 ext4_lblk_t lblk_start, lblk_end;
3544 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3545 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3546
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003547 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
Aditya Kali7b415bf2011-09-09 19:04:51 -04003548}
3549
3550/**
3551 * Determines how many complete clusters (out of those specified by the 'map')
3552 * are under delalloc and were reserved quota for.
3553 * This function is called when we are writing out the blocks that were
3554 * originally written with their allocation delayed, but then the space was
3555 * allocated using fallocate() before the delayed allocation could be resolved.
3556 * The cases to look for are:
3557 * ('=' indicated delayed allocated blocks
3558 * '-' indicates non-delayed allocated blocks)
3559 * (a) partial clusters towards beginning and/or end outside of allocated range
3560 * are not delalloc'ed.
3561 * Ex:
3562 * |----c---=|====c====|====c====|===-c----|
3563 * |++++++ allocated ++++++|
3564 * ==> 4 complete clusters in above example
3565 *
3566 * (b) partial cluster (outside of allocated range) towards either end is
3567 * marked for delayed allocation. In this case, we will exclude that
3568 * cluster.
3569 * Ex:
3570 * |----====c========|========c========|
3571 * |++++++ allocated ++++++|
3572 * ==> 1 complete clusters in above example
3573 *
3574 * Ex:
3575 * |================c================|
3576 * |++++++ allocated ++++++|
3577 * ==> 0 complete clusters in above example
3578 *
3579 * The ext4_da_update_reserve_space will be called only if we
3580 * determine here that there were some "entire" clusters that span
3581 * this 'allocated' range.
3582 * In the non-bigalloc case, this function will just end up returning num_blks
3583 * without ever calling ext4_find_delalloc_range.
3584 */
3585static unsigned int
3586get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3587 unsigned int num_blks)
3588{
3589 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3590 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3591 ext4_lblk_t lblk_from, lblk_to, c_offset;
3592 unsigned int allocated_clusters = 0;
3593
3594 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3595 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3596
3597 /* max possible clusters for this allocation */
3598 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3599
Aditya Kalid8990242011-09-09 19:18:51 -04003600 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3601
Aditya Kali7b415bf2011-09-09 19:04:51 -04003602 /* Check towards left side */
3603 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3604 if (c_offset) {
3605 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3606 lblk_to = lblk_from + c_offset - 1;
3607
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003608 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003609 allocated_clusters--;
3610 }
3611
3612 /* Now check towards right. */
3613 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3614 if (allocated_clusters && c_offset) {
3615 lblk_from = lblk_start + num_blks;
3616 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3617
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003618 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003619 allocated_clusters--;
3620 }
3621
3622 return allocated_clusters;
3623}
3624
Mingming Cao00314622009-09-28 15:49:08 -04003625static int
3626ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003627 struct ext4_map_blocks *map,
Mingming Cao00314622009-09-28 15:49:08 -04003628 struct ext4_ext_path *path, int flags,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003629 unsigned int allocated, ext4_fsblk_t newblock)
Mingming Cao00314622009-09-28 15:49:08 -04003630{
3631 int ret = 0;
3632 int err = 0;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003633 ext4_io_end_t *io = ext4_inode_aio(inode);
Mingming Cao00314622009-09-28 15:49:08 -04003634
Zheng Liu88635ca2011-12-28 19:00:25 -05003635 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3636 "block %llu, max_blocks %u, flags %x, allocated %u\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003637 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
Mingming Cao00314622009-09-28 15:49:08 -04003638 flags, allocated);
3639 ext4_ext_show_leaf(inode, path);
3640
Zheng Liub5645532012-11-08 14:33:43 -05003641 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3642 allocated, newblock);
Aditya Kalid8990242011-09-09 19:18:51 -04003643
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003644 /* get_block() before submit the IO, split the extent */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003645 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003646 ret = ext4_split_unwritten_extents(handle, inode, map,
3647 path, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04003648 if (ret <= 0)
3649 goto out;
Mingming5f524952009-11-10 10:48:04 -05003650 /*
3651 * Flag the inode(non aio case) or end_io struct (aio case)
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003652 * that this IO needs to conversion to written when IO is
Mingming5f524952009-11-10 10:48:04 -05003653 * completed
3654 */
Tao Ma0edeb712011-10-31 17:30:44 -04003655 if (io)
3656 ext4_set_io_unwritten_flag(inode, io);
3657 else
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003658 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003659 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003660 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao00314622009-09-28 15:49:08 -04003661 goto out;
3662 }
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003663 /* IO end_io complete, convert the filled extent to written */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003664 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
Dmitry Monakhovdee1f972012-10-10 01:04:58 -04003665 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
Mingming Cao00314622009-09-28 15:49:08 -04003666 path);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003667 if (ret >= 0) {
Jan Karab436b9b2009-12-08 23:51:10 -05003668 ext4_update_inode_fsync_trans(handle, inode, 1);
Eric Sandeend002ebf2011-01-10 13:03:35 -05003669 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3670 path, map->m_len);
Theodore Ts'o58590b02010-10-27 21:23:12 -04003671 } else
3672 err = ret;
Mingming Cao00314622009-09-28 15:49:08 -04003673 goto out2;
3674 }
3675 /* buffered IO case */
3676 /*
3677 * repeat fallocate creation request
3678 * we already have an unwritten extent
3679 */
3680 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3681 goto map_out;
3682
3683 /* buffered READ or buffered write_begin() lookup */
3684 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3685 /*
3686 * We have blocks reserved already. We
3687 * return allocated blocks so that delalloc
3688 * won't do block reservation for us. But
3689 * the buffer head will be unmapped so that
3690 * a read from the block returns 0s.
3691 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003692 map->m_flags |= EXT4_MAP_UNWRITTEN;
Mingming Cao00314622009-09-28 15:49:08 -04003693 goto out1;
3694 }
3695
3696 /* buffered write, writepage time, convert*/
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003697 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003698 if (ret >= 0)
Jan Karab436b9b2009-12-08 23:51:10 -05003699 ext4_update_inode_fsync_trans(handle, inode, 1);
Mingming Cao00314622009-09-28 15:49:08 -04003700out:
3701 if (ret <= 0) {
3702 err = ret;
3703 goto out2;
3704 } else
3705 allocated = ret;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003706 map->m_flags |= EXT4_MAP_NEW;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003707 /*
3708 * if we allocated more blocks than requested
3709 * we need to make sure we unmap the extra block
3710 * allocated. The actual needed block will get
3711 * unmapped later when we find the buffer_head marked
3712 * new.
3713 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003714 if (allocated > map->m_len) {
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003715 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003716 newblock + map->m_len,
3717 allocated - map->m_len);
3718 allocated = map->m_len;
Aneesh Kumar K.V515f41c2009-12-29 23:39:06 -05003719 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003720
3721 /*
3722 * If we have done fallocate with the offset that is already
3723 * delayed allocated, we would have block reservation
3724 * and quota reservation done in the delayed write path.
3725 * But fallocate would have already updated quota and block
3726 * count for this offset. So cancel these reservation
3727 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003728 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3729 unsigned int reserved_clusters;
3730 reserved_clusters = get_reserved_cluster_alloc(inode,
3731 map->m_lblk, map->m_len);
3732 if (reserved_clusters)
3733 ext4_da_update_reserve_space(inode,
3734 reserved_clusters,
3735 0);
3736 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05003737
Mingming Cao00314622009-09-28 15:49:08 -04003738map_out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003739 map->m_flags |= EXT4_MAP_MAPPED;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04003740 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3741 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3742 map->m_len);
3743 if (err < 0)
3744 goto out2;
3745 }
Mingming Cao00314622009-09-28 15:49:08 -04003746out1:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003747 if (allocated > map->m_len)
3748 allocated = map->m_len;
Mingming Cao00314622009-09-28 15:49:08 -04003749 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003750 map->m_pblk = newblock;
3751 map->m_len = allocated;
Mingming Cao00314622009-09-28 15:49:08 -04003752out2:
3753 if (path) {
3754 ext4_ext_drop_refs(path);
3755 kfree(path);
3756 }
3757 return err ? err : allocated;
3758}
Theodore Ts'o58590b02010-10-27 21:23:12 -04003759
Mingming Cao00314622009-09-28 15:49:08 -04003760/*
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003761 * get_implied_cluster_alloc - check to see if the requested
3762 * allocation (in the map structure) overlaps with a cluster already
3763 * allocated in an extent.
Aditya Kalid8990242011-09-09 19:18:51 -04003764 * @sb The filesystem superblock structure
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003765 * @map The requested lblk->pblk mapping
3766 * @ex The extent structure which might contain an implied
3767 * cluster allocation
3768 *
3769 * This function is called by ext4_ext_map_blocks() after we failed to
3770 * find blocks that were already in the inode's extent tree. Hence,
3771 * we know that the beginning of the requested region cannot overlap
3772 * the extent from the inode's extent tree. There are three cases we
3773 * want to catch. The first is this case:
3774 *
3775 * |--- cluster # N--|
3776 * |--- extent ---| |---- requested region ---|
3777 * |==========|
3778 *
3779 * The second case that we need to test for is this one:
3780 *
3781 * |--------- cluster # N ----------------|
3782 * |--- requested region --| |------- extent ----|
3783 * |=======================|
3784 *
3785 * The third case is when the requested region lies between two extents
3786 * within the same cluster:
3787 * |------------- cluster # N-------------|
3788 * |----- ex -----| |---- ex_right ----|
3789 * |------ requested region ------|
3790 * |================|
3791 *
3792 * In each of the above cases, we need to set the map->m_pblk and
3793 * map->m_len so it corresponds to the return the extent labelled as
3794 * "|====|" from cluster #N, since it is already in use for data in
3795 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3796 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3797 * as a new "allocated" block region. Otherwise, we will return 0 and
3798 * ext4_ext_map_blocks() will then allocate one or more new clusters
3799 * by calling ext4_mb_new_blocks().
3800 */
Aditya Kalid8990242011-09-09 19:18:51 -04003801static int get_implied_cluster_alloc(struct super_block *sb,
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003802 struct ext4_map_blocks *map,
3803 struct ext4_extent *ex,
3804 struct ext4_ext_path *path)
3805{
Aditya Kalid8990242011-09-09 19:18:51 -04003806 struct ext4_sb_info *sbi = EXT4_SB(sb);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003807 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3808 ext4_lblk_t ex_cluster_start, ex_cluster_end;
Curt Wohlgemuth14d7f3e2011-12-18 17:39:02 -05003809 ext4_lblk_t rr_cluster_start;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003810 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3811 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3812 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3813
3814 /* The extent passed in that we are trying to match */
3815 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3816 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3817
3818 /* The requested region passed into ext4_map_blocks() */
3819 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003820
3821 if ((rr_cluster_start == ex_cluster_end) ||
3822 (rr_cluster_start == ex_cluster_start)) {
3823 if (rr_cluster_start == ex_cluster_end)
3824 ee_start += ee_len - 1;
3825 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3826 c_offset;
3827 map->m_len = min(map->m_len,
3828 (unsigned) sbi->s_cluster_ratio - c_offset);
3829 /*
3830 * Check for and handle this case:
3831 *
3832 * |--------- cluster # N-------------|
3833 * |------- extent ----|
3834 * |--- requested region ---|
3835 * |===========|
3836 */
3837
3838 if (map->m_lblk < ee_block)
3839 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3840
3841 /*
3842 * Check for the case where there is already another allocated
3843 * block to the right of 'ex' but before the end of the cluster.
3844 *
3845 * |------------- cluster # N-------------|
3846 * |----- ex -----| |---- ex_right ----|
3847 * |------ requested region ------|
3848 * |================|
3849 */
3850 if (map->m_lblk > ee_block) {
3851 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3852 map->m_len = min(map->m_len, next - map->m_lblk);
3853 }
Aditya Kalid8990242011-09-09 19:18:51 -04003854
3855 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003856 return 1;
3857 }
Aditya Kalid8990242011-09-09 19:18:51 -04003858
3859 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003860 return 0;
3861}
3862
3863
3864/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05003865 * Block allocation/map/preallocation routine for extents based files
3866 *
3867 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003868 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05003869 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3870 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05003871 *
3872 * return > 0, number of of blocks already mapped/allocated
3873 * if create == 0 and these are pre-allocated blocks
3874 * buffer head is unmapped
3875 * otherwise blocks are mapped
3876 *
3877 * return = 0, if plain look up failed (blocks have not been allocated)
3878 * buffer head is unmapped
3879 *
3880 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05003881 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003882int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3883 struct ext4_map_blocks *map, int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07003884{
3885 struct ext4_ext_path *path = NULL;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003886 struct ext4_extent newex, *ex, *ex2;
3887 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003888 ext4_fsblk_t newblock = 0;
Zheng Liu37794732012-11-08 14:47:52 -05003889 int free_on_err = 0, err = 0, depth;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003890 unsigned int allocated = 0, offset = 0;
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04003891 unsigned int allocated_clusters = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05003892 struct ext4_allocation_request ar;
Dmitry Monakhovf45ee3a2012-09-28 23:21:09 -04003893 ext4_io_end_t *io = ext4_inode_aio(inode);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04003894 ext4_lblk_t cluster_offset;
Dmitry Monakhov82e54222012-09-28 23:36:25 -04003895 int set_unwritten = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07003896
Mingming84fe3be2009-09-01 08:44:37 -04003897 ext_debug("blocks %u/%u requested for inode %lu\n",
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003898 map->m_lblk, map->m_len, inode->i_ino);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003899 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07003900
3901 /* check in cache */
Lukas Czerner78771912012-03-19 23:05:43 -04003902 if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003903 if (!newex.ee_start_lo && !newex.ee_start_hi) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04003904 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003905 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003906 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3907
Theodore Ts'oc2177052009-05-14 00:58:52 -04003908 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04003909 /*
3910 * block isn't allocated yet and
3911 * user doesn't want to allocate it
3912 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003913 goto out2;
3914 }
3915 /* we should allocate requested block */
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05003916 } else {
Alex Tomasa86c6182006-10-11 01:21:03 -07003917 /* block is already allocated */
Aditya Kali7b415bf2011-09-09 19:04:51 -04003918 if (sbi->s_cluster_ratio > 1)
3919 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003920 newblock = map->m_lblk
Dave Kleikamp8c55e202007-05-24 13:04:54 -04003921 - le32_to_cpu(newex.ee_block)
Theodore Ts'obf89d162010-10-27 21:30:14 -04003922 + ext4_ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003923 /* number of remaining blocks in the extent */
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05003924 allocated = ext4_ext_get_actual_len(&newex) -
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003925 (map->m_lblk - le32_to_cpu(newex.ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -07003926 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07003927 }
3928 }
3929
3930 /* find extent for this block */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003931 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07003932 if (IS_ERR(path)) {
3933 err = PTR_ERR(path);
3934 path = NULL;
3935 goto out2;
3936 }
3937
3938 depth = ext_depth(inode);
3939
3940 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003941 * consistent leaf must not be empty;
3942 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07003943 * this is why assert can't be put in ext4_ext_find_extent()
3944 */
Frank Mayhar273df552010-03-02 11:46:09 -05003945 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3946 EXT4_ERROR_INODE(inode, "bad extent address "
Theodore Ts'of70f3622010-05-16 23:00:00 -04003947 "lblock: %lu, depth: %d pblock %lld",
3948 (unsigned long) map->m_lblk, depth,
3949 path[depth].p_block);
Surbhi Palande034fb4c2009-12-14 09:53:52 -05003950 err = -EIO;
3951 goto out2;
3952 }
Alex Tomasa86c6182006-10-11 01:21:03 -07003953
Avantika Mathur7e028972006-12-06 20:41:33 -08003954 ex = path[depth].p_ext;
3955 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003956 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Theodore Ts'obf89d162010-10-27 21:30:14 -04003957 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04003958 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003959
3960 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003961 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04003962 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07003963 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04003964 ee_len = ext4_ext_get_actual_len(ex);
Aditya Kalid8990242011-09-09 19:18:51 -04003965
3966 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3967
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003968 /* if found extent covers block, simply return it */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003969 if (in_range(map->m_lblk, ee_block, ee_len)) {
3970 newblock = map->m_lblk - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003971 /* number of remaining blocks in the extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04003972 allocated = ee_len - (map->m_lblk - ee_block);
3973 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3974 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04003975
Allison Hendersone8613042011-05-25 07:41:46 -04003976 /*
Lukas Czerner78771912012-03-19 23:05:43 -04003977 * Do not put uninitialized extent
3978 * in the cache
Allison Hendersone8613042011-05-25 07:41:46 -04003979 */
Lukas Czerner78771912012-03-19 23:05:43 -04003980 if (!ext4_ext_is_uninitialized(ex)) {
3981 ext4_ext_put_in_cache(inode, ee_block,
3982 ee_len, ee_start);
3983 goto out;
Allison Hendersone8613042011-05-25 07:41:46 -04003984 }
Zheng Liu37794732012-11-08 14:47:52 -05003985 allocated = ext4_ext_handle_uninitialized_extents(
Lukas Czerner78771912012-03-19 23:05:43 -04003986 handle, inode, map, path, flags,
3987 allocated, newblock);
Zheng Liu37794732012-11-08 14:47:52 -05003988 goto out3;
Alex Tomasa86c6182006-10-11 01:21:03 -07003989 }
3990 }
3991
Aditya Kali7b415bf2011-09-09 19:04:51 -04003992 if ((sbi->s_cluster_ratio > 1) &&
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05003993 ext4_find_delalloc_cluster(inode, map->m_lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04003994 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3995
Alex Tomasa86c6182006-10-11 01:21:03 -07003996 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003997 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07003998 * we couldn't try to create block if create flag is zero
3999 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04004000 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04004001 /*
4002 * put just found gap into cache to speed up
4003 * subsequent requests
4004 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004005 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
Alex Tomasa86c6182006-10-11 01:21:03 -07004006 goto out2;
4007 }
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004008
Alex Tomasa86c6182006-10-11 01:21:03 -07004009 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004010 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07004011 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004012 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004013 newex.ee_block = cpu_to_le32(map->m_lblk);
4014 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4015
4016 /*
4017 * If we are doing bigalloc, check to see if the extent returned
4018 * by ext4_ext_find_extent() implies a cluster we can use.
4019 */
4020 if (cluster_offset && ex &&
Aditya Kalid8990242011-09-09 19:18:51 -04004021 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004022 ar.len = allocated = map->m_len;
4023 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004024 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004025 goto got_allocated_blocks;
4026 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004027
Alex Tomasc9de5602008-01-29 00:19:52 -05004028 /* find neighbour allocated blocks */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004029 ar.lleft = map->m_lblk;
Alex Tomasc9de5602008-01-29 00:19:52 -05004030 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4031 if (err)
4032 goto out2;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004033 ar.lright = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004034 ex2 = NULL;
4035 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
Alex Tomasc9de5602008-01-29 00:19:52 -05004036 if (err)
4037 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04004038
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004039 /* Check if the extent after searching to the right implies a
4040 * cluster we can use. */
4041 if ((sbi->s_cluster_ratio > 1) && ex2 &&
Aditya Kalid8990242011-09-09 19:18:51 -04004042 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004043 ar.len = allocated = map->m_len;
4044 newblock = map->m_pblk;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004045 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004046 goto got_allocated_blocks;
4047 }
4048
Amit Arora749269f2007-07-18 09:02:56 -04004049 /*
4050 * See if request is beyond maximum number of blocks we can have in
4051 * a single extent. For an initialized extent this limit is
4052 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4053 * EXT_UNINIT_MAX_LEN.
4054 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004055 if (map->m_len > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004056 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004057 map->m_len = EXT_INIT_MAX_LEN;
4058 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04004059 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004060 map->m_len = EXT_UNINIT_MAX_LEN;
Amit Arora749269f2007-07-18 09:02:56 -04004061
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004062 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004063 newex.ee_len = cpu_to_le16(map->m_len);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004064 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
Amit Arora25d14f92007-05-24 13:04:13 -04004065 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004066 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04004067 else
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004068 allocated = map->m_len;
Alex Tomasc9de5602008-01-29 00:19:52 -05004069
4070 /* allocate new block */
4071 ar.inode = inode;
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004072 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4073 ar.logical = map->m_lblk;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004074 /*
4075 * We calculate the offset from the beginning of the cluster
4076 * for the logical block number, since when we allocate a
4077 * physical cluster, the physical block should start at the
4078 * same offset from the beginning of the cluster. This is
4079 * needed so that future calls to get_implied_cluster_alloc()
4080 * work correctly.
4081 */
4082 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4083 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4084 ar.goal -= offset;
4085 ar.logical -= offset;
Alex Tomasc9de5602008-01-29 00:19:52 -05004086 if (S_ISREG(inode->i_mode))
4087 ar.flags = EXT4_MB_HINT_DATA;
4088 else
4089 /* disable in-core preallocation for non-regular files */
4090 ar.flags = 0;
Vivek Haldar556b27a2011-05-25 07:41:54 -04004091 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4092 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
Alex Tomasc9de5602008-01-29 00:19:52 -05004093 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07004094 if (!newblock)
4095 goto out2;
Mingming84fe3be2009-09-01 08:44:37 -04004096 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004097 ar.goal, newblock, allocated);
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004098 free_on_err = 1;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004099 allocated_clusters = ar.len;
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004100 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4101 if (ar.len > allocated)
4102 ar.len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004103
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004104got_allocated_blocks:
Alex Tomasa86c6182006-10-11 01:21:03 -07004105 /* try to insert new extent into found leaf and return */
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004106 ext4_ext_store_pblock(&newex, newblock + offset);
Alex Tomasc9de5602008-01-29 00:19:52 -05004107 newex.ee_len = cpu_to_le16(ar.len);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004108 /* Mark uninitialized */
4109 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
Amit Aroraa2df2a62007-07-17 21:42:41 -04004110 ext4_ext_mark_uninitialized(&newex);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004111 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05004112 * io_end structure was created for every IO write to an
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004113 * uninitialized extent. To avoid unnecessary conversion,
Jiaying Zhang744692d2010-03-04 16:14:02 -05004114 * here we flag the IO that really needs the conversion.
Mingming5f524952009-11-10 10:48:04 -05004115 * For non asycn direct IO case, flag the inode state
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004116 * that we need to perform conversion when IO is done.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004117 */
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004118 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4119 set_unwritten = 1;
Jiaying Zhang744692d2010-03-04 16:14:02 -05004120 if (ext4_should_dioread_nolock(inode))
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004121 map->m_flags |= EXT4_MAP_UNINIT;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04004122 }
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004123
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004124 err = 0;
4125 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4126 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4127 path, ar.len);
Jiaying Zhang575a1d42011-07-10 20:07:25 -04004128 if (!err)
4129 err = ext4_ext_insert_extent(handle, inode, path,
4130 &newex, flags);
Dmitry Monakhov82e54222012-09-28 23:36:25 -04004131
4132 if (!err && set_unwritten) {
4133 if (io)
4134 ext4_set_io_unwritten_flag(inode, io);
4135 else
4136 ext4_set_inode_state(inode,
4137 EXT4_STATE_DIO_UNWRITTEN);
4138 }
4139
Theodore Ts'o4d33b1e2011-09-09 18:52:51 -04004140 if (err && free_on_err) {
Maxim Patlasov7132de72011-07-10 19:37:48 -04004141 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4142 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
Alex Tomas315054f2007-05-24 13:04:25 -04004143 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05004144 /* not a good idea to call discard here directly,
4145 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004146 ext4_discard_preallocations(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004147 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
Maxim Patlasov7132de72011-07-10 19:37:48 -04004148 ext4_ext_get_actual_len(&newex), fb_flags);
Alex Tomasa86c6182006-10-11 01:21:03 -07004149 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04004150 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004151
Alex Tomasa86c6182006-10-11 01:21:03 -07004152 /* previous routine could use block we allocated */
Theodore Ts'obf89d162010-10-27 21:30:14 -04004153 newblock = ext4_ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05004154 allocated = ext4_ext_get_actual_len(&newex);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004155 if (allocated > map->m_len)
4156 allocated = map->m_len;
4157 map->m_flags |= EXT4_MAP_NEW;
Alex Tomasa86c6182006-10-11 01:21:03 -07004158
Jan Karab436b9b2009-12-08 23:51:10 -05004159 /*
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004160 * Update reserved blocks/metadata blocks after successful
4161 * block allocation which had been deferred till now.
4162 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04004163 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004164 unsigned int reserved_clusters;
Aditya Kali7b415bf2011-09-09 19:04:51 -04004165 /*
Yongqiang Yang81fdbb42011-10-29 09:23:38 -04004166 * Check how many clusters we had reserved this allocated range
Aditya Kali7b415bf2011-09-09 19:04:51 -04004167 */
4168 reserved_clusters = get_reserved_cluster_alloc(inode,
4169 map->m_lblk, allocated);
4170 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4171 if (reserved_clusters) {
4172 /*
4173 * We have clusters reserved for this range.
4174 * But since we are not doing actual allocation
4175 * and are simply using blocks from previously
4176 * allocated cluster, we should release the
4177 * reservation and not claim quota.
4178 */
4179 ext4_da_update_reserve_space(inode,
4180 reserved_clusters, 0);
4181 }
4182 } else {
4183 BUG_ON(allocated_clusters < reserved_clusters);
4184 /* We will claim quota for all newly allocated blocks.*/
4185 ext4_da_update_reserve_space(inode, allocated_clusters,
4186 1);
4187 if (reserved_clusters < allocated_clusters) {
Aditya Kali5356f262011-09-09 19:20:51 -04004188 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004189 int reservation = allocated_clusters -
4190 reserved_clusters;
4191 /*
4192 * It seems we claimed few clusters outside of
4193 * the range of this allocation. We should give
4194 * it back to the reservation pool. This can
4195 * happen in the following case:
4196 *
4197 * * Suppose s_cluster_ratio is 4 (i.e., each
4198 * cluster has 4 blocks. Thus, the clusters
4199 * are [0-3],[4-7],[8-11]...
4200 * * First comes delayed allocation write for
4201 * logical blocks 10 & 11. Since there were no
4202 * previous delayed allocated blocks in the
4203 * range [8-11], we would reserve 1 cluster
4204 * for this write.
4205 * * Next comes write for logical blocks 3 to 8.
4206 * In this case, we will reserve 2 clusters
4207 * (for [0-3] and [4-7]; and not for [8-11] as
4208 * that range has a delayed allocated blocks.
4209 * Thus total reserved clusters now becomes 3.
4210 * * Now, during the delayed allocation writeout
4211 * time, we will first write blocks [3-8] and
4212 * allocate 3 clusters for writing these
4213 * blocks. Also, we would claim all these
4214 * three clusters above.
4215 * * Now when we come here to writeout the
4216 * blocks [10-11], we would expect to claim
4217 * the reservation of 1 cluster we had made
4218 * (and we would claim it since there are no
4219 * more delayed allocated blocks in the range
4220 * [8-11]. But our reserved cluster count had
4221 * already gone to 0.
4222 *
4223 * Thus, at the step 4 above when we determine
4224 * that there are still some unwritten delayed
4225 * allocated blocks outside of our current
4226 * block range, we should increment the
4227 * reserved clusters count so that when the
4228 * remaining blocks finally gets written, we
4229 * could claim them.
4230 */
Aditya Kali5356f262011-09-09 19:20:51 -04004231 dquot_reserve_block(inode,
4232 EXT4_C2B(sbi, reservation));
4233 spin_lock(&ei->i_block_reservation_lock);
4234 ei->i_reserved_data_blocks += reservation;
4235 spin_unlock(&ei->i_block_reservation_lock);
Aditya Kali7b415bf2011-09-09 19:04:51 -04004236 }
4237 }
4238 }
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -05004239
4240 /*
Jan Karab436b9b2009-12-08 23:51:10 -05004241 * Cache the extent and update transaction to commit on fdatasync only
4242 * when it is _not_ an uninitialized extent.
4243 */
4244 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05004245 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
Jan Karab436b9b2009-12-08 23:51:10 -05004246 ext4_update_inode_fsync_trans(handle, inode, 1);
4247 } else
4248 ext4_update_inode_fsync_trans(handle, inode, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07004249out:
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004250 if (allocated > map->m_len)
4251 allocated = map->m_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004252 ext4_ext_show_leaf(inode, path);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004253 map->m_flags |= EXT4_MAP_MAPPED;
4254 map->m_pblk = newblock;
4255 map->m_len = allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004256out2:
4257 if (path) {
4258 ext4_ext_drop_refs(path);
4259 kfree(path);
4260 }
Allison Hendersone8613042011-05-25 07:41:46 -04004261
Zheng Liu37794732012-11-08 14:47:52 -05004262out3:
Zheng Liu19b303d2012-11-08 14:34:04 -05004263 trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated);
Yongqiang Yange7b319e2011-10-29 09:39:51 -04004264
Lukas Czerner78771912012-03-19 23:05:43 -04004265 return err ? err : allocated;
Alex Tomasa86c6182006-10-11 01:21:03 -07004266}
4267
Jan Karacf108bc2008-07-11 19:27:31 -04004268void ext4_ext_truncate(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07004269{
4270 struct address_space *mapping = inode->i_mapping;
4271 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05004272 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07004273 handle_t *handle;
Allison Henderson189e8682011-09-06 21:49:44 -04004274 loff_t page_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07004275 int err = 0;
4276
4277 /*
Jiaying Zhang3889fd52011-01-10 12:47:05 -05004278 * finish any pending end_io work so we won't run the risk of
4279 * converting any truncated blocks to initialized later
4280 */
Dmitry Monakhovc2785312012-10-05 11:31:55 -04004281 ext4_flush_unwritten_io(inode);
Jiaying Zhang3889fd52011-01-10 12:47:05 -05004282
4283 /*
Alex Tomasa86c6182006-10-11 01:21:03 -07004284 * probably first extent we're gonna free will be last in block
4285 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004286 err = ext4_writepage_trans_blocks(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07004287 handle = ext4_journal_start(inode, err);
Jan Karacf108bc2008-07-11 19:27:31 -04004288 if (IS_ERR(handle))
Alex Tomasa86c6182006-10-11 01:21:03 -07004289 return;
Alex Tomasa86c6182006-10-11 01:21:03 -07004290
Allison Henderson189e8682011-09-06 21:49:44 -04004291 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4292 page_len = PAGE_CACHE_SIZE -
4293 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4294
4295 err = ext4_discard_partial_page_buffers(handle,
4296 mapping, inode->i_size, page_len, 0);
4297
4298 if (err)
4299 goto out_stop;
4300 }
Alex Tomasa86c6182006-10-11 01:21:03 -07004301
Jan Kara9ddfc3d2008-07-11 19:27:31 -04004302 if (ext4_orphan_add(handle, inode))
4303 goto out_stop;
4304
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05004305 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07004306 ext4_ext_invalidate_cache(inode);
4307
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04004308 ext4_discard_preallocations(inode);
Alex Tomasc9de5602008-01-29 00:19:52 -05004309
Alex Tomasa86c6182006-10-11 01:21:03 -07004310 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004311 * TODO: optimization is possible here.
4312 * Probably we need not scan at all,
4313 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07004314 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004315
4316 /* we have to know where to truncate from in crash case */
4317 EXT4_I(inode)->i_disksize = inode->i_size;
4318 ext4_mark_inode_dirty(handle, inode);
4319
4320 last_block = (inode->i_size + sb->s_blocksize - 1)
4321 >> EXT4_BLOCK_SIZE_BITS(sb);
Zheng Liu51865fd2012-11-08 21:57:32 -05004322 err = ext4_es_remove_extent(inode, last_block,
4323 EXT_MAX_BLOCKS - last_block);
Lukas Czerner5f95d212012-03-19 23:03:19 -04004324 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07004325
4326 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04004327 * transaction synchronous.
4328 */
Alex Tomasa86c6182006-10-11 01:21:03 -07004329 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05004330 ext4_handle_sync(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07004331
Jan Kara9ddfc3d2008-07-11 19:27:31 -04004332 up_write(&EXT4_I(inode)->i_data_sem);
Eric Gouriouf6d2f6b2011-05-22 21:33:00 -04004333
4334out_stop:
Alex Tomasa86c6182006-10-11 01:21:03 -07004335 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07004336 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07004337 * then we need to clear up the orphan record which we created above.
4338 * However, if this was a real unlink then we were called by
4339 * ext4_delete_inode(), and we allow that function to clean up the
4340 * orphan info for us.
4341 */
4342 if (inode->i_nlink)
4343 ext4_orphan_del(handle, inode);
4344
Solofo Ramangalahyef737722008-04-29 22:00:41 -04004345 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4346 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07004347 ext4_journal_stop(handle);
4348}
4349
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004350static void ext4_falloc_update_inode(struct inode *inode,
4351 int mode, loff_t new_size, int update_ctime)
4352{
4353 struct timespec now;
4354
4355 if (update_ctime) {
4356 now = current_fs_time(inode->i_sb);
4357 if (!timespec_equal(&inode->i_ctime, &now))
4358 inode->i_ctime = now;
4359 }
4360 /*
4361 * Update only when preallocation was requested beyond
4362 * the file size.
4363 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04004364 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4365 if (new_size > i_size_read(inode))
4366 i_size_write(inode, new_size);
4367 if (new_size > EXT4_I(inode)->i_disksize)
4368 ext4_update_i_disksize(inode, new_size);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004369 } else {
4370 /*
4371 * Mark that we allocate beyond EOF so the subsequent truncate
4372 * can proceed even if the new size is the same as i_size.
4373 */
4374 if (new_size > i_size_read(inode))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004375 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004376 }
4377
4378}
4379
Amit Aroraa2df2a62007-07-17 21:42:41 -04004380/*
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004381 * preallocate space for a file. This implements ext4's fallocate file
Amit Aroraa2df2a62007-07-17 21:42:41 -04004382 * operation, which gets called from sys_fallocate system call.
4383 * For block-mapped files, posix_fallocate should fall back to the method
4384 * of writing zeroes to the required new blocks (the same behavior which is
4385 * expected for file systems which do not support fallocate() system call).
4386 */
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004387long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
Amit Aroraa2df2a62007-07-17 21:42:41 -04004388{
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01004389 struct inode *inode = file->f_path.dentry->d_inode;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004390 handle_t *handle;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004391 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05004392 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004393 int ret = 0;
4394 int ret2 = 0;
4395 int retries = 0;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004396 int flags;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004397 struct ext4_map_blocks map;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004398 unsigned int credits, blkbits = inode->i_blkbits;
4399
4400 /*
4401 * currently supporting (pre)allocate mode for extent-based
4402 * files _only_
4403 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004404 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Amit Aroraa2df2a62007-07-17 21:42:41 -04004405 return -EOPNOTSUPP;
4406
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004407 /* Return error if mode is not supported */
4408 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4409 return -EOPNOTSUPP;
4410
4411 if (mode & FALLOC_FL_PUNCH_HOLE)
4412 return ext4_punch_hole(file, offset, len);
4413
Tao Ma0c8d4142012-12-10 14:06:03 -05004414 ret = ext4_convert_inline_data(inode);
4415 if (ret)
4416 return ret;
4417
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004418 trace_ext4_fallocate_enter(inode, offset, len, mode);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004419 map.m_lblk = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004420 /*
4421 * We can't just convert len to max_blocks because
4422 * If blocksize = 4096 offset = 3072 and len = 2048
4423 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04004424 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004425 - map.m_lblk;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004426 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04004427 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04004428 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04004429 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004430 mutex_lock(&inode->i_mutex);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004431 ret = inode_newsize_ok(inode, (len + offset));
4432 if (ret) {
4433 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004434 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
Nikanth Karthikesan6d19c422010-05-16 14:00:00 -04004435 return ret;
4436 }
Greg Harm3c6fe772011-10-31 18:41:47 -04004437 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004438 if (mode & FALLOC_FL_KEEP_SIZE)
4439 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
Greg Harm3c6fe772011-10-31 18:41:47 -04004440 /*
4441 * Don't normalize the request if it can fit in one extent so
4442 * that it doesn't get unnecessarily split into multiple
4443 * extents.
4444 */
4445 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4446 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
Dmitry Monakhov60d46162012-10-05 11:32:02 -04004447
4448 /* Prevent race condition between unwritten */
4449 ext4_flush_unwritten_io(inode);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004450retry:
4451 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004452 map.m_lblk = map.m_lblk + ret;
4453 map.m_len = max_blocks = max_blocks - ret;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004454 handle = ext4_journal_start(inode, credits);
4455 if (IS_ERR(handle)) {
4456 ret = PTR_ERR(handle);
4457 break;
4458 }
Dmitry Monakhova4e5d882011-10-25 08:15:12 -04004459 ret = ext4_map_blocks(handle, inode, &map, flags);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05004460 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004461#ifdef EXT4FS_DEBUG
4462 WARN_ON(ret <= 0);
Theodore Ts'oe35fd662010-05-16 19:00:00 -04004463 printk(KERN_ERR "%s: ext4_ext_map_blocks "
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004464 "returned error inode#%lu, block=%u, "
Thadeu Lima de Souza Cascardo9fd97842009-01-26 19:26:26 -05004465 "max_blocks=%u", __func__,
Kazuya Mioa6371b62010-10-27 21:30:15 -04004466 inode->i_ino, map.m_lblk, max_blocks);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05004467#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04004468 ext4_mark_inode_dirty(handle, inode);
4469 ret2 = ext4_journal_stop(handle);
4470 break;
4471 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004472 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004473 blkbits) >> blkbits))
4474 new_size = offset + len;
4475 else
Utako Kusaka29ae07b2011-07-27 22:11:20 -04004476 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004477
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004478 ext4_falloc_update_inode(inode, mode, new_size,
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004479 (map.m_flags & EXT4_MAP_NEW));
Amit Aroraa2df2a62007-07-17 21:42:41 -04004480 ext4_mark_inode_dirty(handle, inode);
Zheng Liuf4e95b32012-06-30 19:12:57 -04004481 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4482 ext4_handle_sync(handle);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004483 ret2 = ext4_journal_stop(handle);
4484 if (ret2)
4485 break;
4486 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04004487 if (ret == -ENOSPC &&
4488 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4489 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004490 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04004491 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05004492 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04004493 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4494 ret > 0 ? ret2 : ret);
Amit Aroraa2df2a62007-07-17 21:42:41 -04004495 return ret > 0 ? ret2 : ret;
4496}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004497
4498/*
Mingming Cao00314622009-09-28 15:49:08 -04004499 * This function convert a range of blocks to written extents
4500 * The caller of this function will pass the start offset and the size.
4501 * all unwritten extents within this range will be converted to
4502 * written extents.
4503 *
4504 * This function is called from the direct IO end io call back
4505 * function, to convert the fallocated extents after IO is completed.
Mingming109f5562009-11-10 10:48:08 -05004506 * Returns 0 on success.
Mingming Cao00314622009-09-28 15:49:08 -04004507 */
4508int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
Eric Sandeena1de02d2010-02-04 23:58:38 -05004509 ssize_t len)
Mingming Cao00314622009-09-28 15:49:08 -04004510{
4511 handle_t *handle;
Mingming Cao00314622009-09-28 15:49:08 -04004512 unsigned int max_blocks;
4513 int ret = 0;
4514 int ret2 = 0;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004515 struct ext4_map_blocks map;
Mingming Cao00314622009-09-28 15:49:08 -04004516 unsigned int credits, blkbits = inode->i_blkbits;
4517
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004518 map.m_lblk = offset >> blkbits;
Mingming Cao00314622009-09-28 15:49:08 -04004519 /*
4520 * We can't just convert len to max_blocks because
4521 * If blocksize = 4096 offset = 3072 and len = 2048
4522 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004523 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4524 map.m_lblk);
Mingming Cao00314622009-09-28 15:49:08 -04004525 /*
4526 * credits to insert 1 extent into extent tree
4527 */
4528 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4529 while (ret >= 0 && ret < max_blocks) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004530 map.m_lblk += ret;
4531 map.m_len = (max_blocks -= ret);
Mingming Cao00314622009-09-28 15:49:08 -04004532 handle = ext4_journal_start(inode, credits);
4533 if (IS_ERR(handle)) {
4534 ret = PTR_ERR(handle);
4535 break;
4536 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -04004537 ret = ext4_map_blocks(handle, inode, &map,
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05004538 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
Mingming Cao00314622009-09-28 15:49:08 -04004539 if (ret <= 0) {
4540 WARN_ON(ret <= 0);
Theodore Ts'o92b97812012-03-19 23:41:49 -04004541 ext4_msg(inode->i_sb, KERN_ERR,
4542 "%s:%d: inode #%lu: block %u: len %u: "
4543 "ext4_ext_map_blocks returned %d",
4544 __func__, __LINE__, inode->i_ino, map.m_lblk,
4545 map.m_len, ret);
Mingming Cao00314622009-09-28 15:49:08 -04004546 }
4547 ext4_mark_inode_dirty(handle, inode);
4548 ret2 = ext4_journal_stop(handle);
4549 if (ret <= 0 || ret2 )
4550 break;
4551 }
4552 return ret > 0 ? ret2 : ret;
4553}
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004554
Mingming Cao00314622009-09-28 15:49:08 -04004555/*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004556 * If newex is not existing extent (newex->ec_start equals zero) find
4557 * delayed extent at start of newex and update newex accordingly and
4558 * return start of the next delayed extent.
4559 *
4560 * If newex is existing extent (newex->ec_start is not equal zero)
4561 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4562 * extent found. Leave newex unmodified.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004563 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004564static int ext4_find_delayed_extent(struct inode *inode,
4565 struct ext4_ext_cache *newex)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004566{
Zheng Liub3aff3e2012-11-08 21:57:37 -05004567 struct extent_status es;
Zheng Liub3aff3e2012-11-08 21:57:37 -05004568 ext4_lblk_t next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004569
Zheng Liub3aff3e2012-11-08 21:57:37 -05004570 es.start = newex->ec_block;
4571 next_del = ext4_es_find_extent(inode, &es);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004572
Theodore Ts'ob05e6ae2011-01-10 12:13:26 -05004573 if (newex->ec_start == 0) {
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004574 /*
4575 * No extent in extent-tree contains block @newex->ec_start,
4576 * then the block may stay in 1)a hole or 2)delayed-extent.
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004577 */
Zheng Liub3aff3e2012-11-08 21:57:37 -05004578 if (es.len == 0)
4579 /* A hole found. */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004580 return 0;
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004581
Zheng Liub3aff3e2012-11-08 21:57:37 -05004582 if (es.start > newex->ec_block) {
4583 /* A hole found. */
4584 newex->ec_len = min(es.start - newex->ec_block,
4585 newex->ec_len);
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004586 return 0;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004587 }
Yongqiang Yang6d9c85e2011-02-27 17:25:47 -05004588
Zheng Liub3aff3e2012-11-08 21:57:37 -05004589 newex->ec_len = es.start + es.len - newex->ec_block;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004590 }
4591
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004592 return next_del;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004593}
Eric Sandeen6873fa02008-10-07 00:46:36 -04004594/* fiemap flags we can handle specified here */
4595#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4596
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05004597static int ext4_xattr_fiemap(struct inode *inode,
4598 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04004599{
4600 __u64 physical = 0;
4601 __u64 length;
4602 __u32 flags = FIEMAP_EXTENT_LAST;
4603 int blockbits = inode->i_sb->s_blocksize_bits;
4604 int error = 0;
4605
4606 /* in-inode? */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004607 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
Eric Sandeen6873fa02008-10-07 00:46:36 -04004608 struct ext4_iloc iloc;
4609 int offset; /* offset of xattr in inode */
4610
4611 error = ext4_get_inode_loc(inode, &iloc);
4612 if (error)
4613 return error;
4614 physical = iloc.bh->b_blocknr << blockbits;
4615 offset = EXT4_GOOD_OLD_INODE_SIZE +
4616 EXT4_I(inode)->i_extra_isize;
4617 physical += offset;
4618 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4619 flags |= FIEMAP_EXTENT_DATA_INLINE;
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004620 brelse(iloc.bh);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004621 } else { /* external block */
4622 physical = EXT4_I(inode)->i_file_acl << blockbits;
4623 length = inode->i_sb->s_blocksize;
4624 }
4625
4626 if (physical)
4627 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4628 length, flags);
4629 return (error < 0 ? error : 0);
4630}
4631
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004632/*
4633 * ext4_ext_punch_hole
4634 *
4635 * Punches a hole of "length" bytes in a file starting
4636 * at byte "offset"
4637 *
4638 * @inode: The inode of the file to punch a hole in
4639 * @offset: The starting byte offset of the hole
4640 * @length: The length of the hole
4641 *
4642 * Returns the number of blocks removed or negative on err
4643 */
4644int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4645{
4646 struct inode *inode = file->f_path.dentry->d_inode;
4647 struct super_block *sb = inode->i_sb;
Lukas Czerner5f95d212012-03-19 23:03:19 -04004648 ext4_lblk_t first_block, stop_block;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004649 struct address_space *mapping = inode->i_mapping;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004650 handle_t *handle;
Allison Hendersonba062082011-09-03 11:55:59 -04004651 loff_t first_page, last_page, page_len;
4652 loff_t first_page_offset, last_page_offset;
Lukas Czerner5f95d212012-03-19 23:03:19 -04004653 int credits, err = 0;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004654
Dmitry Monakhov02d262d2012-09-30 23:03:42 -04004655 /*
4656 * Write out all dirty pages to avoid race conditions
4657 * Then release them.
4658 */
4659 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4660 err = filemap_write_and_wait_range(mapping,
4661 offset, offset + length - 1);
4662
4663 if (err)
4664 return err;
4665 }
4666
4667 mutex_lock(&inode->i_mutex);
4668 /* It's not possible punch hole on append only file */
4669 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4670 err = -EPERM;
4671 goto out_mutex;
4672 }
4673 if (IS_SWAPFILE(inode)) {
4674 err = -ETXTBSY;
4675 goto out_mutex;
4676 }
4677
Allison Henderson2be47512011-09-03 11:56:52 -04004678 /* No need to punch hole beyond i_size */
4679 if (offset >= inode->i_size)
Dmitry Monakhov02d262d2012-09-30 23:03:42 -04004680 goto out_mutex;
Allison Henderson2be47512011-09-03 11:56:52 -04004681
4682 /*
4683 * If the hole extends beyond i_size, set the hole
4684 * to end after the page that contains i_size
4685 */
4686 if (offset + length > inode->i_size) {
4687 length = inode->i_size +
4688 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4689 offset;
4690 }
4691
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004692 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4693 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4694
4695 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4696 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4697
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004698 /* Now release the pages */
4699 if (last_page_offset > first_page_offset) {
Hugh Dickins5e44f8c2012-06-01 00:15:28 -04004700 truncate_pagecache_range(inode, first_page_offset,
4701 last_page_offset - 1);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004702 }
4703
Dmitry Monakhov02d262d2012-09-30 23:03:42 -04004704 /* Wait all existing dio workers, newcomers will block on i_mutex */
4705 ext4_inode_block_unlocked_dio(inode);
Dmitry Monakhovc2785312012-10-05 11:31:55 -04004706 err = ext4_flush_unwritten_io(inode);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -04004707 if (err)
Dmitry Monakhov02d262d2012-09-30 23:03:42 -04004708 goto out_dio;
Dmitry Monakhovc2785312012-10-05 11:31:55 -04004709 inode_dio_wait(inode);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004710
4711 credits = ext4_writepage_trans_blocks(inode);
4712 handle = ext4_journal_start(inode, credits);
Dmitry Monakhov02d262d2012-09-30 23:03:42 -04004713 if (IS_ERR(handle)) {
4714 err = PTR_ERR(handle);
4715 goto out_dio;
4716 }
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004717
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004718
4719 /*
Allison Hendersonba062082011-09-03 11:55:59 -04004720 * Now we need to zero out the non-page-aligned data in the
4721 * pages at the start and tail of the hole, and unmap the buffer
4722 * heads for the block aligned regions of the page that were
4723 * completely zeroed.
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004724 */
Allison Hendersonba062082011-09-03 11:55:59 -04004725 if (first_page > last_page) {
4726 /*
4727 * If the file space being truncated is contained within a page
4728 * just zero out and unmap the middle of that page
4729 */
4730 err = ext4_discard_partial_page_buffers(handle,
4731 mapping, offset, length, 0);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004732
Allison Hendersonba062082011-09-03 11:55:59 -04004733 if (err)
4734 goto out;
4735 } else {
4736 /*
4737 * zero out and unmap the partial page that contains
4738 * the start of the hole
4739 */
4740 page_len = first_page_offset - offset;
4741 if (page_len > 0) {
4742 err = ext4_discard_partial_page_buffers(handle, mapping,
4743 offset, page_len, 0);
4744 if (err)
4745 goto out;
4746 }
4747
4748 /*
4749 * zero out and unmap the partial page that contains
4750 * the end of the hole
4751 */
4752 page_len = offset + length - last_page_offset;
4753 if (page_len > 0) {
4754 err = ext4_discard_partial_page_buffers(handle, mapping,
4755 last_page_offset, page_len, 0);
4756 if (err)
4757 goto out;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004758 }
4759 }
4760
Allison Henderson2be47512011-09-03 11:56:52 -04004761 /*
4762 * If i_size is contained in the last page, we need to
4763 * unmap and zero the partial page after i_size
4764 */
4765 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4766 inode->i_size % PAGE_CACHE_SIZE != 0) {
4767
4768 page_len = PAGE_CACHE_SIZE -
4769 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4770
4771 if (page_len > 0) {
4772 err = ext4_discard_partial_page_buffers(handle,
4773 mapping, inode->i_size, page_len, 0);
4774
4775 if (err)
4776 goto out;
4777 }
4778 }
4779
Lukas Czerner5f95d212012-03-19 23:03:19 -04004780 first_block = (offset + sb->s_blocksize - 1) >>
4781 EXT4_BLOCK_SIZE_BITS(sb);
4782 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4783
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004784 /* If there are no blocks to remove, return now */
Lukas Czerner5f95d212012-03-19 23:03:19 -04004785 if (first_block >= stop_block)
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004786 goto out;
4787
4788 down_write(&EXT4_I(inode)->i_data_sem);
4789 ext4_ext_invalidate_cache(inode);
4790 ext4_discard_preallocations(inode);
4791
Zheng Liu51865fd2012-11-08 21:57:32 -05004792 err = ext4_es_remove_extent(inode, first_block,
4793 stop_block - first_block);
Lukas Czerner5f95d212012-03-19 23:03:19 -04004794 err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004795
Lukas Czerner5f95d212012-03-19 23:03:19 -04004796 ext4_ext_invalidate_cache(inode);
4797 ext4_discard_preallocations(inode);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004798
4799 if (IS_SYNC(inode))
4800 ext4_handle_sync(handle);
4801
4802 up_write(&EXT4_I(inode)->i_data_sem);
4803
4804out:
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004805 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4806 ext4_mark_inode_dirty(handle, inode);
4807 ext4_journal_stop(handle);
Dmitry Monakhov02d262d2012-09-30 23:03:42 -04004808out_dio:
4809 ext4_inode_resume_unlocked_dio(inode);
4810out_mutex:
4811 mutex_unlock(&inode->i_mutex);
Allison Hendersona4bb6b62011-05-25 07:41:50 -04004812 return err;
4813}
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004814
Eric Sandeen6873fa02008-10-07 00:46:36 -04004815int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4816 __u64 start, __u64 len)
4817{
4818 ext4_lblk_t start_blk;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004819 int error = 0;
4820
Tao Ma94191982012-12-10 14:06:02 -05004821 if (ext4_has_inline_data(inode)) {
4822 int has_inline = 1;
4823
4824 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4825
4826 if (has_inline)
4827 return error;
4828 }
4829
Eric Sandeen6873fa02008-10-07 00:46:36 -04004830 /* fallback to generic here if not in extents fmt */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004831 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Eric Sandeen6873fa02008-10-07 00:46:36 -04004832 return generic_block_fiemap(inode, fieinfo, start, len,
4833 ext4_get_block);
4834
4835 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4836 return -EBADR;
4837
4838 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4839 error = ext4_xattr_fiemap(inode, fieinfo);
4840 } else {
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004841 ext4_lblk_t len_blks;
4842 __u64 last_blk;
4843
Eric Sandeen6873fa02008-10-07 00:46:36 -04004844 start_blk = start >> inode->i_sb->s_blocksize_bits;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004845 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
Lukas Czernerf17722f2011-06-06 00:05:17 -04004846 if (last_blk >= EXT_MAX_BLOCKS)
4847 last_blk = EXT_MAX_BLOCKS-1;
Leonard Michlmayraca92ff2010-03-04 17:07:28 -05004848 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
Eric Sandeen6873fa02008-10-07 00:46:36 -04004849
4850 /*
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004851 * Walk the extent tree gathering extent information
4852 * and pushing extents back to the user.
Eric Sandeen6873fa02008-10-07 00:46:36 -04004853 */
Lukas Czerner91dd8c12012-11-28 12:32:26 -05004854 error = ext4_fill_fiemap_extents(inode, start_blk,
4855 len_blks, fieinfo);
Eric Sandeen6873fa02008-10-07 00:46:36 -04004856 }
4857
4858 return error;
4859}