]>
Commit | Line | Data |
---|---|---|
a86c6181 AT |
1 | /* |
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 | ||
32 | #include <linux/module.h> | |
33 | #include <linux/fs.h> | |
34 | #include <linux/time.h> | |
cd02ff0b | 35 | #include <linux/jbd2.h> |
a86c6181 AT |
36 | #include <linux/highuid.h> |
37 | #include <linux/pagemap.h> | |
38 | #include <linux/quotaops.h> | |
39 | #include <linux/string.h> | |
40 | #include <linux/slab.h> | |
a2df2a63 | 41 | #include <linux/falloc.h> |
a86c6181 | 42 | #include <asm/uaccess.h> |
6873fa0d | 43 | #include <linux/fiemap.h> |
3dcf5451 | 44 | #include "ext4_jbd2.h" |
a86c6181 | 45 | |
0562e0ba JZ |
46 | #include <trace/events/ext4.h> |
47 | ||
d583fb87 AH |
48 | static int ext4_split_extent(handle_t *handle, |
49 | struct inode *inode, | |
50 | struct ext4_ext_path *path, | |
51 | struct ext4_map_blocks *map, | |
52 | int split_flag, | |
53 | int flags); | |
54 | ||
487caeef JK |
55 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
56 | struct inode *inode, | |
57 | int needed) | |
a86c6181 AT |
58 | { |
59 | int err; | |
60 | ||
0390131b FM |
61 | if (!ext4_handle_valid(handle)) |
62 | return 0; | |
a86c6181 | 63 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
64 | return 0; |
65 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 66 | if (err <= 0) |
9102e4fa | 67 | return err; |
487caeef | 68 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
69 | if (err == 0) |
70 | err = -EAGAIN; | |
487caeef JK |
71 | |
72 | return err; | |
a86c6181 AT |
73 | } |
74 | ||
75 | /* | |
76 | * could return: | |
77 | * - EROFS | |
78 | * - ENOMEM | |
79 | */ | |
80 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
81 | struct ext4_ext_path *path) | |
82 | { | |
83 | if (path->p_bh) { | |
84 | /* path points to block */ | |
85 | return ext4_journal_get_write_access(handle, path->p_bh); | |
86 | } | |
87 | /* path points to leaf/index in inode body */ | |
88 | /* we use in-core data, no need to protect them */ | |
89 | return 0; | |
90 | } | |
91 | ||
92 | /* | |
93 | * could return: | |
94 | * - EROFS | |
95 | * - ENOMEM | |
96 | * - EIO | |
97 | */ | |
9ea7a0df TT |
98 | #define ext4_ext_dirty(handle, inode, path) \ |
99 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
100 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
101 | handle_t *handle, struct inode *inode, | |
102 | struct ext4_ext_path *path) | |
a86c6181 AT |
103 | { |
104 | int err; | |
105 | if (path->p_bh) { | |
106 | /* path points to block */ | |
9ea7a0df TT |
107 | err = __ext4_handle_dirty_metadata(where, line, handle, |
108 | inode, path->p_bh); | |
a86c6181 AT |
109 | } else { |
110 | /* path points to leaf/index in inode body */ | |
111 | err = ext4_mark_inode_dirty(handle, inode); | |
112 | } | |
113 | return err; | |
114 | } | |
115 | ||
f65e6fba | 116 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 117 | struct ext4_ext_path *path, |
725d26d3 | 118 | ext4_lblk_t block) |
a86c6181 | 119 | { |
a86c6181 AT |
120 | int depth; |
121 | ||
122 | if (path) { | |
123 | struct ext4_extent *ex; | |
124 | depth = path->p_depth; | |
125 | ||
ad4fb9ca KM |
126 | /* |
127 | * Try to predict block placement assuming that we are | |
128 | * filling in a file which will eventually be | |
129 | * non-sparse --- i.e., in the case of libbfd writing | |
130 | * an ELF object sections out-of-order but in a way | |
131 | * the eventually results in a contiguous object or | |
132 | * executable file, or some database extending a table | |
133 | * space file. However, this is actually somewhat | |
134 | * non-ideal if we are writing a sparse file such as | |
135 | * qemu or KVM writing a raw image file that is going | |
136 | * to stay fairly sparse, since it will end up | |
137 | * fragmenting the file system's free space. Maybe we | |
138 | * should have some hueristics or some way to allow | |
139 | * userspace to pass a hint to file system, | |
b8d6568a | 140 | * especially if the latter case turns out to be |
ad4fb9ca KM |
141 | * common. |
142 | */ | |
7e028976 | 143 | ex = path[depth].p_ext; |
ad4fb9ca KM |
144 | if (ex) { |
145 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
146 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
147 | ||
148 | if (block > ext_block) | |
149 | return ext_pblk + (block - ext_block); | |
150 | else | |
151 | return ext_pblk - (ext_block - block); | |
152 | } | |
a86c6181 | 153 | |
d0d856e8 RD |
154 | /* it looks like index is empty; |
155 | * try to find starting block from index itself */ | |
a86c6181 AT |
156 | if (path[depth].p_bh) |
157 | return path[depth].p_bh->b_blocknr; | |
158 | } | |
159 | ||
160 | /* OK. use inode's group */ | |
f86186b4 | 161 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
162 | } |
163 | ||
654b4908 AK |
164 | /* |
165 | * Allocation for a meta data block | |
166 | */ | |
f65e6fba | 167 | static ext4_fsblk_t |
654b4908 | 168 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 169 | struct ext4_ext_path *path, |
55f020db | 170 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 171 | { |
f65e6fba | 172 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
173 | |
174 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
175 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
176 | NULL, err); | |
a86c6181 AT |
177 | return newblock; |
178 | } | |
179 | ||
55ad63bf | 180 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
181 | { |
182 | int size; | |
183 | ||
184 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
185 | / sizeof(struct ext4_extent); | |
55ad63bf | 186 | if (!check) { |
bbf2f9fb | 187 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
188 | if (size > 6) |
189 | size = 6; | |
a86c6181 | 190 | #endif |
55ad63bf | 191 | } |
a86c6181 AT |
192 | return size; |
193 | } | |
194 | ||
55ad63bf | 195 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
196 | { |
197 | int size; | |
198 | ||
199 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
200 | / sizeof(struct ext4_extent_idx); | |
55ad63bf | 201 | if (!check) { |
bbf2f9fb | 202 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
203 | if (size > 5) |
204 | size = 5; | |
a86c6181 | 205 | #endif |
55ad63bf | 206 | } |
a86c6181 AT |
207 | return size; |
208 | } | |
209 | ||
55ad63bf | 210 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
211 | { |
212 | int size; | |
213 | ||
214 | size = sizeof(EXT4_I(inode)->i_data); | |
215 | size -= sizeof(struct ext4_extent_header); | |
216 | size /= sizeof(struct ext4_extent); | |
55ad63bf | 217 | if (!check) { |
bbf2f9fb | 218 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
219 | if (size > 3) |
220 | size = 3; | |
a86c6181 | 221 | #endif |
55ad63bf | 222 | } |
a86c6181 AT |
223 | return size; |
224 | } | |
225 | ||
55ad63bf | 226 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
227 | { |
228 | int size; | |
229 | ||
230 | size = sizeof(EXT4_I(inode)->i_data); | |
231 | size -= sizeof(struct ext4_extent_header); | |
232 | size /= sizeof(struct ext4_extent_idx); | |
55ad63bf | 233 | if (!check) { |
bbf2f9fb | 234 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
235 | if (size > 4) |
236 | size = 4; | |
a86c6181 | 237 | #endif |
55ad63bf | 238 | } |
a86c6181 AT |
239 | return size; |
240 | } | |
241 | ||
d2a17637 MC |
242 | /* |
243 | * Calculate the number of metadata blocks needed | |
244 | * to allocate @blocks | |
245 | * Worse case is one block per extent | |
246 | */ | |
01f49d0b | 247 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 248 | { |
9d0be502 TT |
249 | struct ext4_inode_info *ei = EXT4_I(inode); |
250 | int idxs, num = 0; | |
d2a17637 | 251 | |
9d0be502 TT |
252 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
253 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
254 | |
255 | /* | |
9d0be502 TT |
256 | * If the new delayed allocation block is contiguous with the |
257 | * previous da block, it can share index blocks with the | |
258 | * previous block, so we only need to allocate a new index | |
259 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
260 | * an additional index block, and at ldxs**3 blocks, yet | |
261 | * another index blocks. | |
d2a17637 | 262 | */ |
9d0be502 TT |
263 | if (ei->i_da_metadata_calc_len && |
264 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
265 | if ((ei->i_da_metadata_calc_len % idxs) == 0) | |
266 | num++; | |
267 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
268 | num++; | |
269 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
270 | num++; | |
271 | ei->i_da_metadata_calc_len = 0; | |
272 | } else | |
273 | ei->i_da_metadata_calc_len++; | |
274 | ei->i_da_metadata_calc_last_lblock++; | |
275 | return num; | |
276 | } | |
d2a17637 | 277 | |
9d0be502 TT |
278 | /* |
279 | * In the worst case we need a new set of index blocks at | |
280 | * every level of the inode's extent tree. | |
281 | */ | |
282 | ei->i_da_metadata_calc_len = 1; | |
283 | ei->i_da_metadata_calc_last_lblock = lblock; | |
284 | return ext_depth(inode) + 1; | |
d2a17637 MC |
285 | } |
286 | ||
c29c0ae7 AT |
287 | static int |
288 | ext4_ext_max_entries(struct inode *inode, int depth) | |
289 | { | |
290 | int max; | |
291 | ||
292 | if (depth == ext_depth(inode)) { | |
293 | if (depth == 0) | |
55ad63bf | 294 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 295 | else |
55ad63bf | 296 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
297 | } else { |
298 | if (depth == 0) | |
55ad63bf | 299 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 300 | else |
55ad63bf | 301 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
302 | } |
303 | ||
304 | return max; | |
305 | } | |
306 | ||
56b19868 AK |
307 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
308 | { | |
bf89d16f | 309 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 310 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 311 | |
6fd058f7 | 312 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
313 | } |
314 | ||
315 | static int ext4_valid_extent_idx(struct inode *inode, | |
316 | struct ext4_extent_idx *ext_idx) | |
317 | { | |
bf89d16f | 318 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 319 | |
6fd058f7 | 320 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
321 | } |
322 | ||
323 | static int ext4_valid_extent_entries(struct inode *inode, | |
324 | struct ext4_extent_header *eh, | |
325 | int depth) | |
326 | { | |
327 | struct ext4_extent *ext; | |
328 | struct ext4_extent_idx *ext_idx; | |
329 | unsigned short entries; | |
330 | if (eh->eh_entries == 0) | |
331 | return 1; | |
332 | ||
333 | entries = le16_to_cpu(eh->eh_entries); | |
334 | ||
335 | if (depth == 0) { | |
336 | /* leaf entries */ | |
337 | ext = EXT_FIRST_EXTENT(eh); | |
338 | while (entries) { | |
339 | if (!ext4_valid_extent(inode, ext)) | |
340 | return 0; | |
341 | ext++; | |
342 | entries--; | |
343 | } | |
344 | } else { | |
345 | ext_idx = EXT_FIRST_INDEX(eh); | |
346 | while (entries) { | |
347 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
348 | return 0; | |
349 | ext_idx++; | |
350 | entries--; | |
351 | } | |
352 | } | |
353 | return 1; | |
354 | } | |
355 | ||
c398eda0 TT |
356 | static int __ext4_ext_check(const char *function, unsigned int line, |
357 | struct inode *inode, struct ext4_extent_header *eh, | |
358 | int depth) | |
c29c0ae7 AT |
359 | { |
360 | const char *error_msg; | |
361 | int max = 0; | |
362 | ||
363 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
364 | error_msg = "invalid magic"; | |
365 | goto corrupted; | |
366 | } | |
367 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
368 | error_msg = "unexpected eh_depth"; | |
369 | goto corrupted; | |
370 | } | |
371 | if (unlikely(eh->eh_max == 0)) { | |
372 | error_msg = "invalid eh_max"; | |
373 | goto corrupted; | |
374 | } | |
375 | max = ext4_ext_max_entries(inode, depth); | |
376 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
377 | error_msg = "too large eh_max"; | |
378 | goto corrupted; | |
379 | } | |
380 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
381 | error_msg = "invalid eh_entries"; | |
382 | goto corrupted; | |
383 | } | |
56b19868 AK |
384 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
385 | error_msg = "invalid extent entries"; | |
386 | goto corrupted; | |
387 | } | |
c29c0ae7 AT |
388 | return 0; |
389 | ||
390 | corrupted: | |
c398eda0 | 391 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 392 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 393 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 394 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
395 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
396 | max, le16_to_cpu(eh->eh_depth), depth); | |
397 | ||
398 | return -EIO; | |
399 | } | |
400 | ||
56b19868 | 401 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 402 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 403 | |
7a262f7c AK |
404 | int ext4_ext_check_inode(struct inode *inode) |
405 | { | |
406 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
407 | } | |
408 | ||
a86c6181 AT |
409 | #ifdef EXT_DEBUG |
410 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
411 | { | |
412 | int k, l = path->p_depth; | |
413 | ||
414 | ext_debug("path:"); | |
415 | for (k = 0; k <= l; k++, path++) { | |
416 | if (path->p_idx) { | |
2ae02107 | 417 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 418 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 419 | } else if (path->p_ext) { |
553f9008 | 420 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 421 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 422 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 423 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 424 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
425 | } else |
426 | ext_debug(" []"); | |
427 | } | |
428 | ext_debug("\n"); | |
429 | } | |
430 | ||
431 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
432 | { | |
433 | int depth = ext_depth(inode); | |
434 | struct ext4_extent_header *eh; | |
435 | struct ext4_extent *ex; | |
436 | int i; | |
437 | ||
438 | if (!path) | |
439 | return; | |
440 | ||
441 | eh = path[depth].p_hdr; | |
442 | ex = EXT_FIRST_EXTENT(eh); | |
443 | ||
553f9008 M |
444 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
445 | ||
a86c6181 | 446 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
447 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
448 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 449 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
450 | } |
451 | ext_debug("\n"); | |
452 | } | |
1b16da77 YY |
453 | |
454 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
455 | ext4_fsblk_t newblock, int level) | |
456 | { | |
457 | int depth = ext_depth(inode); | |
458 | struct ext4_extent *ex; | |
459 | ||
460 | if (depth != level) { | |
461 | struct ext4_extent_idx *idx; | |
462 | idx = path[level].p_idx; | |
463 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
464 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
465 | le32_to_cpu(idx->ei_block), | |
466 | ext4_idx_pblock(idx), | |
467 | newblock); | |
468 | idx++; | |
469 | } | |
470 | ||
471 | return; | |
472 | } | |
473 | ||
474 | ex = path[depth].p_ext; | |
475 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
476 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
477 | le32_to_cpu(ex->ee_block), | |
478 | ext4_ext_pblock(ex), | |
479 | ext4_ext_is_uninitialized(ex), | |
480 | ext4_ext_get_actual_len(ex), | |
481 | newblock); | |
482 | ex++; | |
483 | } | |
484 | } | |
485 | ||
a86c6181 | 486 | #else |
af5bc92d TT |
487 | #define ext4_ext_show_path(inode, path) |
488 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 489 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
490 | #endif |
491 | ||
b35905c1 | 492 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
493 | { |
494 | int depth = path->p_depth; | |
495 | int i; | |
496 | ||
497 | for (i = 0; i <= depth; i++, path++) | |
498 | if (path->p_bh) { | |
499 | brelse(path->p_bh); | |
500 | path->p_bh = NULL; | |
501 | } | |
502 | } | |
503 | ||
504 | /* | |
d0d856e8 RD |
505 | * ext4_ext_binsearch_idx: |
506 | * binary search for the closest index of the given block | |
c29c0ae7 | 507 | * the header must be checked before calling this |
a86c6181 AT |
508 | */ |
509 | static void | |
725d26d3 AK |
510 | ext4_ext_binsearch_idx(struct inode *inode, |
511 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
512 | { |
513 | struct ext4_extent_header *eh = path->p_hdr; | |
514 | struct ext4_extent_idx *r, *l, *m; | |
515 | ||
a86c6181 | 516 | |
bba90743 | 517 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
518 | |
519 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 520 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
521 | while (l <= r) { |
522 | m = l + (r - l) / 2; | |
523 | if (block < le32_to_cpu(m->ei_block)) | |
524 | r = m - 1; | |
525 | else | |
526 | l = m + 1; | |
26d535ed DM |
527 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
528 | m, le32_to_cpu(m->ei_block), | |
529 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
530 | } |
531 | ||
532 | path->p_idx = l - 1; | |
f65e6fba | 533 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 534 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
535 | |
536 | #ifdef CHECK_BINSEARCH | |
537 | { | |
538 | struct ext4_extent_idx *chix, *ix; | |
539 | int k; | |
540 | ||
541 | chix = ix = EXT_FIRST_INDEX(eh); | |
542 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
543 | if (k != 0 && | |
544 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
545 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
546 | "first=0x%p\n", k, | |
547 | ix, EXT_FIRST_INDEX(eh)); | |
548 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
549 | le32_to_cpu(ix->ei_block), |
550 | le32_to_cpu(ix[-1].ei_block)); | |
551 | } | |
552 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 553 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
554 | if (block < le32_to_cpu(ix->ei_block)) |
555 | break; | |
556 | chix = ix; | |
557 | } | |
558 | BUG_ON(chix != path->p_idx); | |
559 | } | |
560 | #endif | |
561 | ||
562 | } | |
563 | ||
564 | /* | |
d0d856e8 RD |
565 | * ext4_ext_binsearch: |
566 | * binary search for closest extent of the given block | |
c29c0ae7 | 567 | * the header must be checked before calling this |
a86c6181 AT |
568 | */ |
569 | static void | |
725d26d3 AK |
570 | ext4_ext_binsearch(struct inode *inode, |
571 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
572 | { |
573 | struct ext4_extent_header *eh = path->p_hdr; | |
574 | struct ext4_extent *r, *l, *m; | |
575 | ||
a86c6181 AT |
576 | if (eh->eh_entries == 0) { |
577 | /* | |
d0d856e8 RD |
578 | * this leaf is empty: |
579 | * we get such a leaf in split/add case | |
a86c6181 AT |
580 | */ |
581 | return; | |
582 | } | |
583 | ||
bba90743 | 584 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
585 | |
586 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 587 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
588 | |
589 | while (l <= r) { | |
590 | m = l + (r - l) / 2; | |
591 | if (block < le32_to_cpu(m->ee_block)) | |
592 | r = m - 1; | |
593 | else | |
594 | l = m + 1; | |
26d535ed DM |
595 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
596 | m, le32_to_cpu(m->ee_block), | |
597 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
598 | } |
599 | ||
600 | path->p_ext = l - 1; | |
553f9008 | 601 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 602 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 603 | ext4_ext_pblock(path->p_ext), |
553f9008 | 604 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 605 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
606 | |
607 | #ifdef CHECK_BINSEARCH | |
608 | { | |
609 | struct ext4_extent *chex, *ex; | |
610 | int k; | |
611 | ||
612 | chex = ex = EXT_FIRST_EXTENT(eh); | |
613 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
614 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 615 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
616 | if (block < le32_to_cpu(ex->ee_block)) |
617 | break; | |
618 | chex = ex; | |
619 | } | |
620 | BUG_ON(chex != path->p_ext); | |
621 | } | |
622 | #endif | |
623 | ||
624 | } | |
625 | ||
626 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
627 | { | |
628 | struct ext4_extent_header *eh; | |
629 | ||
630 | eh = ext_inode_hdr(inode); | |
631 | eh->eh_depth = 0; | |
632 | eh->eh_entries = 0; | |
633 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 634 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
635 | ext4_mark_inode_dirty(handle, inode); |
636 | ext4_ext_invalidate_cache(inode); | |
637 | return 0; | |
638 | } | |
639 | ||
640 | struct ext4_ext_path * | |
725d26d3 AK |
641 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
642 | struct ext4_ext_path *path) | |
a86c6181 AT |
643 | { |
644 | struct ext4_extent_header *eh; | |
645 | struct buffer_head *bh; | |
646 | short int depth, i, ppos = 0, alloc = 0; | |
647 | ||
648 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 649 | depth = ext_depth(inode); |
a86c6181 AT |
650 | |
651 | /* account possible depth increase */ | |
652 | if (!path) { | |
5d4958f9 | 653 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
654 | GFP_NOFS); |
655 | if (!path) | |
656 | return ERR_PTR(-ENOMEM); | |
657 | alloc = 1; | |
658 | } | |
a86c6181 | 659 | path[0].p_hdr = eh; |
1973adcb | 660 | path[0].p_bh = NULL; |
a86c6181 | 661 | |
c29c0ae7 | 662 | i = depth; |
a86c6181 AT |
663 | /* walk through the tree */ |
664 | while (i) { | |
7a262f7c AK |
665 | int need_to_validate = 0; |
666 | ||
a86c6181 AT |
667 | ext_debug("depth %d: num %d, max %d\n", |
668 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 669 | |
a86c6181 | 670 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 671 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
672 | path[ppos].p_depth = i; |
673 | path[ppos].p_ext = NULL; | |
674 | ||
7a262f7c AK |
675 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
676 | if (unlikely(!bh)) | |
a86c6181 | 677 | goto err; |
7a262f7c | 678 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
679 | trace_ext4_ext_load_extent(inode, block, |
680 | path[ppos].p_block); | |
7a262f7c AK |
681 | if (bh_submit_read(bh) < 0) { |
682 | put_bh(bh); | |
683 | goto err; | |
684 | } | |
685 | /* validate the extent entries */ | |
686 | need_to_validate = 1; | |
687 | } | |
a86c6181 AT |
688 | eh = ext_block_hdr(bh); |
689 | ppos++; | |
273df556 FM |
690 | if (unlikely(ppos > depth)) { |
691 | put_bh(bh); | |
692 | EXT4_ERROR_INODE(inode, | |
693 | "ppos %d > depth %d", ppos, depth); | |
694 | goto err; | |
695 | } | |
a86c6181 AT |
696 | path[ppos].p_bh = bh; |
697 | path[ppos].p_hdr = eh; | |
698 | i--; | |
699 | ||
7a262f7c | 700 | if (need_to_validate && ext4_ext_check(inode, eh, i)) |
a86c6181 AT |
701 | goto err; |
702 | } | |
703 | ||
704 | path[ppos].p_depth = i; | |
a86c6181 AT |
705 | path[ppos].p_ext = NULL; |
706 | path[ppos].p_idx = NULL; | |
707 | ||
a86c6181 AT |
708 | /* find extent */ |
709 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
710 | /* if not an empty leaf */ |
711 | if (path[ppos].p_ext) | |
bf89d16f | 712 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
713 | |
714 | ext4_ext_show_path(inode, path); | |
715 | ||
716 | return path; | |
717 | ||
718 | err: | |
719 | ext4_ext_drop_refs(path); | |
720 | if (alloc) | |
721 | kfree(path); | |
722 | return ERR_PTR(-EIO); | |
723 | } | |
724 | ||
725 | /* | |
d0d856e8 RD |
726 | * ext4_ext_insert_index: |
727 | * insert new index [@logical;@ptr] into the block at @curp; | |
728 | * check where to insert: before @curp or after @curp | |
a86c6181 | 729 | */ |
1f109d5a TT |
730 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
731 | struct ext4_ext_path *curp, | |
732 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
733 | { |
734 | struct ext4_extent_idx *ix; | |
735 | int len, err; | |
736 | ||
7e028976 AM |
737 | err = ext4_ext_get_access(handle, inode, curp); |
738 | if (err) | |
a86c6181 AT |
739 | return err; |
740 | ||
273df556 FM |
741 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
742 | EXT4_ERROR_INODE(inode, | |
743 | "logical %d == ei_block %d!", | |
744 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
745 | return -EIO; | |
746 | } | |
d4620315 RD |
747 | |
748 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
749 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
750 | EXT4_ERROR_INODE(inode, | |
751 | "eh_entries %d >= eh_max %d!", | |
752 | le16_to_cpu(curp->p_hdr->eh_entries), | |
753 | le16_to_cpu(curp->p_hdr->eh_max)); | |
754 | return -EIO; | |
755 | } | |
756 | ||
a86c6181 AT |
757 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; |
758 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
759 | /* insert after */ | |
760 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
761 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
762 | len = len < 0 ? 0 : len; | |
26d535ed | 763 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
764 | "move %d from 0x%p to 0x%p\n", |
765 | logical, ptr, len, | |
766 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
767 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
768 | } | |
769 | ix = curp->p_idx + 1; | |
770 | } else { | |
771 | /* insert before */ | |
772 | len = len * sizeof(struct ext4_extent_idx); | |
773 | len = len < 0 ? 0 : len; | |
26d535ed | 774 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
775 | "move %d from 0x%p to 0x%p\n", |
776 | logical, ptr, len, | |
777 | curp->p_idx, (curp->p_idx + 1)); | |
778 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
779 | ix = curp->p_idx; | |
780 | } | |
781 | ||
782 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 783 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 784 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 785 | |
273df556 FM |
786 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
787 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
788 | return -EIO; | |
789 | } | |
a86c6181 AT |
790 | |
791 | err = ext4_ext_dirty(handle, inode, curp); | |
792 | ext4_std_error(inode->i_sb, err); | |
793 | ||
794 | return err; | |
795 | } | |
796 | ||
797 | /* | |
d0d856e8 RD |
798 | * ext4_ext_split: |
799 | * inserts new subtree into the path, using free index entry | |
800 | * at depth @at: | |
801 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
802 | * - makes decision where to split | |
803 | * - moves remaining extents and index entries (right to the split point) | |
804 | * into the newly allocated blocks | |
805 | * - initializes subtree | |
a86c6181 AT |
806 | */ |
807 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
808 | unsigned int flags, |
809 | struct ext4_ext_path *path, | |
810 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
811 | { |
812 | struct buffer_head *bh = NULL; | |
813 | int depth = ext_depth(inode); | |
814 | struct ext4_extent_header *neh; | |
815 | struct ext4_extent_idx *fidx; | |
a86c6181 | 816 | int i = at, k, m, a; |
f65e6fba | 817 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 818 | __le32 border; |
f65e6fba | 819 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
820 | int err = 0; |
821 | ||
822 | /* make decision: where to split? */ | |
d0d856e8 | 823 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 824 | |
d0d856e8 | 825 | /* if current leaf will be split, then we should use |
a86c6181 | 826 | * border from split point */ |
273df556 FM |
827 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
828 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
829 | return -EIO; | |
830 | } | |
a86c6181 AT |
831 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
832 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 833 | ext_debug("leaf will be split." |
a86c6181 | 834 | " next leaf starts at %d\n", |
8c55e204 | 835 | le32_to_cpu(border)); |
a86c6181 AT |
836 | } else { |
837 | border = newext->ee_block; | |
838 | ext_debug("leaf will be added." | |
839 | " next leaf starts at %d\n", | |
8c55e204 | 840 | le32_to_cpu(border)); |
a86c6181 AT |
841 | } |
842 | ||
843 | /* | |
d0d856e8 RD |
844 | * If error occurs, then we break processing |
845 | * and mark filesystem read-only. index won't | |
a86c6181 | 846 | * be inserted and tree will be in consistent |
d0d856e8 | 847 | * state. Next mount will repair buffers too. |
a86c6181 AT |
848 | */ |
849 | ||
850 | /* | |
d0d856e8 RD |
851 | * Get array to track all allocated blocks. |
852 | * We need this to handle errors and free blocks | |
853 | * upon them. | |
a86c6181 | 854 | */ |
5d4958f9 | 855 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
856 | if (!ablocks) |
857 | return -ENOMEM; | |
a86c6181 AT |
858 | |
859 | /* allocate all needed blocks */ | |
860 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
861 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 862 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 863 | newext, &err, flags); |
a86c6181 AT |
864 | if (newblock == 0) |
865 | goto cleanup; | |
866 | ablocks[a] = newblock; | |
867 | } | |
868 | ||
869 | /* initialize new leaf */ | |
870 | newblock = ablocks[--a]; | |
273df556 FM |
871 | if (unlikely(newblock == 0)) { |
872 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
873 | err = -EIO; | |
874 | goto cleanup; | |
875 | } | |
a86c6181 AT |
876 | bh = sb_getblk(inode->i_sb, newblock); |
877 | if (!bh) { | |
878 | err = -EIO; | |
879 | goto cleanup; | |
880 | } | |
881 | lock_buffer(bh); | |
882 | ||
7e028976 AM |
883 | err = ext4_journal_get_create_access(handle, bh); |
884 | if (err) | |
a86c6181 AT |
885 | goto cleanup; |
886 | ||
887 | neh = ext_block_hdr(bh); | |
888 | neh->eh_entries = 0; | |
55ad63bf | 889 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
890 | neh->eh_magic = EXT4_EXT_MAGIC; |
891 | neh->eh_depth = 0; | |
a86c6181 | 892 | |
d0d856e8 | 893 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
894 | if (unlikely(path[depth].p_hdr->eh_entries != |
895 | path[depth].p_hdr->eh_max)) { | |
896 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
897 | path[depth].p_hdr->eh_entries, | |
898 | path[depth].p_hdr->eh_max); | |
899 | err = -EIO; | |
900 | goto cleanup; | |
901 | } | |
a86c6181 | 902 | /* start copy from next extent */ |
1b16da77 YY |
903 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
904 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 905 | if (m) { |
1b16da77 YY |
906 | struct ext4_extent *ex; |
907 | ex = EXT_FIRST_EXTENT(neh); | |
908 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 909 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
910 | } |
911 | ||
912 | set_buffer_uptodate(bh); | |
913 | unlock_buffer(bh); | |
914 | ||
0390131b | 915 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 916 | if (err) |
a86c6181 AT |
917 | goto cleanup; |
918 | brelse(bh); | |
919 | bh = NULL; | |
920 | ||
921 | /* correct old leaf */ | |
922 | if (m) { | |
7e028976 AM |
923 | err = ext4_ext_get_access(handle, inode, path + depth); |
924 | if (err) | |
a86c6181 | 925 | goto cleanup; |
e8546d06 | 926 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
927 | err = ext4_ext_dirty(handle, inode, path + depth); |
928 | if (err) | |
a86c6181 AT |
929 | goto cleanup; |
930 | ||
931 | } | |
932 | ||
933 | /* create intermediate indexes */ | |
934 | k = depth - at - 1; | |
273df556 FM |
935 | if (unlikely(k < 0)) { |
936 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
937 | err = -EIO; | |
938 | goto cleanup; | |
939 | } | |
a86c6181 AT |
940 | if (k) |
941 | ext_debug("create %d intermediate indices\n", k); | |
942 | /* insert new index into current index block */ | |
943 | /* current depth stored in i var */ | |
944 | i = depth - 1; | |
945 | while (k--) { | |
946 | oldblock = newblock; | |
947 | newblock = ablocks[--a]; | |
bba90743 | 948 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
949 | if (!bh) { |
950 | err = -EIO; | |
951 | goto cleanup; | |
952 | } | |
953 | lock_buffer(bh); | |
954 | ||
7e028976 AM |
955 | err = ext4_journal_get_create_access(handle, bh); |
956 | if (err) | |
a86c6181 AT |
957 | goto cleanup; |
958 | ||
959 | neh = ext_block_hdr(bh); | |
960 | neh->eh_entries = cpu_to_le16(1); | |
961 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 962 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
963 | neh->eh_depth = cpu_to_le16(depth - i); |
964 | fidx = EXT_FIRST_INDEX(neh); | |
965 | fidx->ei_block = border; | |
f65e6fba | 966 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 967 | |
bba90743 ES |
968 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
969 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 970 | |
1b16da77 | 971 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
972 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
973 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
974 | EXT4_ERROR_INODE(inode, | |
975 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
976 | le32_to_cpu(path[i].p_ext->ee_block)); | |
977 | err = -EIO; | |
978 | goto cleanup; | |
979 | } | |
1b16da77 YY |
980 | /* start copy indexes */ |
981 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
982 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
983 | EXT_MAX_INDEX(path[i].p_hdr)); | |
984 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 985 | if (m) { |
1b16da77 | 986 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 987 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 988 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
989 | } |
990 | set_buffer_uptodate(bh); | |
991 | unlock_buffer(bh); | |
992 | ||
0390131b | 993 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 994 | if (err) |
a86c6181 AT |
995 | goto cleanup; |
996 | brelse(bh); | |
997 | bh = NULL; | |
998 | ||
999 | /* correct old index */ | |
1000 | if (m) { | |
1001 | err = ext4_ext_get_access(handle, inode, path + i); | |
1002 | if (err) | |
1003 | goto cleanup; | |
e8546d06 | 1004 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1005 | err = ext4_ext_dirty(handle, inode, path + i); |
1006 | if (err) | |
1007 | goto cleanup; | |
1008 | } | |
1009 | ||
1010 | i--; | |
1011 | } | |
1012 | ||
1013 | /* insert new index */ | |
a86c6181 AT |
1014 | err = ext4_ext_insert_index(handle, inode, path + at, |
1015 | le32_to_cpu(border), newblock); | |
1016 | ||
1017 | cleanup: | |
1018 | if (bh) { | |
1019 | if (buffer_locked(bh)) | |
1020 | unlock_buffer(bh); | |
1021 | brelse(bh); | |
1022 | } | |
1023 | ||
1024 | if (err) { | |
1025 | /* free all allocated blocks in error case */ | |
1026 | for (i = 0; i < depth; i++) { | |
1027 | if (!ablocks[i]) | |
1028 | continue; | |
7dc57615 | 1029 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1030 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1031 | } |
1032 | } | |
1033 | kfree(ablocks); | |
1034 | ||
1035 | return err; | |
1036 | } | |
1037 | ||
1038 | /* | |
d0d856e8 RD |
1039 | * ext4_ext_grow_indepth: |
1040 | * implements tree growing procedure: | |
1041 | * - allocates new block | |
1042 | * - moves top-level data (index block or leaf) into the new block | |
1043 | * - initializes new top-level, creating index that points to the | |
1044 | * just created block | |
a86c6181 AT |
1045 | */ |
1046 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db AH |
1047 | unsigned int flags, |
1048 | struct ext4_ext_path *path, | |
1049 | struct ext4_extent *newext) | |
a86c6181 AT |
1050 | { |
1051 | struct ext4_ext_path *curp = path; | |
1052 | struct ext4_extent_header *neh; | |
a86c6181 | 1053 | struct buffer_head *bh; |
f65e6fba | 1054 | ext4_fsblk_t newblock; |
a86c6181 AT |
1055 | int err = 0; |
1056 | ||
55f020db AH |
1057 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
1058 | newext, &err, flags); | |
a86c6181 AT |
1059 | if (newblock == 0) |
1060 | return err; | |
1061 | ||
1062 | bh = sb_getblk(inode->i_sb, newblock); | |
1063 | if (!bh) { | |
1064 | err = -EIO; | |
1065 | ext4_std_error(inode->i_sb, err); | |
1066 | return err; | |
1067 | } | |
1068 | lock_buffer(bh); | |
1069 | ||
7e028976 AM |
1070 | err = ext4_journal_get_create_access(handle, bh); |
1071 | if (err) { | |
a86c6181 AT |
1072 | unlock_buffer(bh); |
1073 | goto out; | |
1074 | } | |
1075 | ||
1076 | /* move top-level index/leaf into new block */ | |
1077 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
1078 | ||
1079 | /* set size of new block */ | |
1080 | neh = ext_block_hdr(bh); | |
1081 | /* old root could have indexes or leaves | |
1082 | * so calculate e_max right way */ | |
1083 | if (ext_depth(inode)) | |
55ad63bf | 1084 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1085 | else |
55ad63bf | 1086 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1087 | neh->eh_magic = EXT4_EXT_MAGIC; |
1088 | set_buffer_uptodate(bh); | |
1089 | unlock_buffer(bh); | |
1090 | ||
0390131b | 1091 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1092 | if (err) |
a86c6181 AT |
1093 | goto out; |
1094 | ||
1095 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
1096 | err = ext4_ext_get_access(handle, inode, curp); |
1097 | if (err) | |
a86c6181 AT |
1098 | goto out; |
1099 | ||
1100 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1101 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); |
a86c6181 AT |
1102 | curp->p_hdr->eh_entries = cpu_to_le16(1); |
1103 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
1104 | |
1105 | if (path[0].p_hdr->eh_depth) | |
1106 | curp->p_idx->ei_block = | |
1107 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
1108 | else | |
1109 | curp->p_idx->ei_block = | |
1110 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 1111 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
1112 | |
1113 | neh = ext_inode_hdr(inode); | |
2ae02107 | 1114 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1115 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1116 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1117 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 AT |
1118 | |
1119 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
1120 | err = ext4_ext_dirty(handle, inode, curp); | |
1121 | out: | |
1122 | brelse(bh); | |
1123 | ||
1124 | return err; | |
1125 | } | |
1126 | ||
1127 | /* | |
d0d856e8 RD |
1128 | * ext4_ext_create_new_leaf: |
1129 | * finds empty index and adds new leaf. | |
1130 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1131 | */ |
1132 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1133 | unsigned int flags, |
1134 | struct ext4_ext_path *path, | |
1135 | struct ext4_extent *newext) | |
a86c6181 AT |
1136 | { |
1137 | struct ext4_ext_path *curp; | |
1138 | int depth, i, err = 0; | |
1139 | ||
1140 | repeat: | |
1141 | i = depth = ext_depth(inode); | |
1142 | ||
1143 | /* walk up to the tree and look for free index entry */ | |
1144 | curp = path + depth; | |
1145 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1146 | i--; | |
1147 | curp--; | |
1148 | } | |
1149 | ||
d0d856e8 RD |
1150 | /* we use already allocated block for index block, |
1151 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1152 | if (EXT_HAS_FREE_INDEX(curp)) { |
1153 | /* if we found index with free entry, then use that | |
1154 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1155 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1156 | if (err) |
1157 | goto out; | |
a86c6181 AT |
1158 | |
1159 | /* refill path */ | |
1160 | ext4_ext_drop_refs(path); | |
1161 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1162 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1163 | path); | |
a86c6181 AT |
1164 | if (IS_ERR(path)) |
1165 | err = PTR_ERR(path); | |
1166 | } else { | |
1167 | /* tree is full, time to grow in depth */ | |
55f020db AH |
1168 | err = ext4_ext_grow_indepth(handle, inode, flags, |
1169 | path, newext); | |
a86c6181 AT |
1170 | if (err) |
1171 | goto out; | |
1172 | ||
1173 | /* refill path */ | |
1174 | ext4_ext_drop_refs(path); | |
1175 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1176 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1177 | path); | |
a86c6181 AT |
1178 | if (IS_ERR(path)) { |
1179 | err = PTR_ERR(path); | |
1180 | goto out; | |
1181 | } | |
1182 | ||
1183 | /* | |
d0d856e8 RD |
1184 | * only first (depth 0 -> 1) produces free space; |
1185 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1186 | */ |
1187 | depth = ext_depth(inode); | |
1188 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1189 | /* now we need to split */ |
a86c6181 AT |
1190 | goto repeat; |
1191 | } | |
1192 | } | |
1193 | ||
1194 | out: | |
1195 | return err; | |
1196 | } | |
1197 | ||
1988b51e AT |
1198 | /* |
1199 | * search the closest allocated block to the left for *logical | |
1200 | * and returns it at @logical + it's physical address at @phys | |
1201 | * if *logical is the smallest allocated block, the function | |
1202 | * returns 0 at @phys | |
1203 | * return value contains 0 (success) or error code | |
1204 | */ | |
1f109d5a TT |
1205 | static int ext4_ext_search_left(struct inode *inode, |
1206 | struct ext4_ext_path *path, | |
1207 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1208 | { |
1209 | struct ext4_extent_idx *ix; | |
1210 | struct ext4_extent *ex; | |
b939e376 | 1211 | int depth, ee_len; |
1988b51e | 1212 | |
273df556 FM |
1213 | if (unlikely(path == NULL)) { |
1214 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1215 | return -EIO; | |
1216 | } | |
1988b51e AT |
1217 | depth = path->p_depth; |
1218 | *phys = 0; | |
1219 | ||
1220 | if (depth == 0 && path->p_ext == NULL) | |
1221 | return 0; | |
1222 | ||
1223 | /* usually extent in the path covers blocks smaller | |
1224 | * then *logical, but it can be that extent is the | |
1225 | * first one in the file */ | |
1226 | ||
1227 | ex = path[depth].p_ext; | |
b939e376 | 1228 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1229 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1230 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1231 | EXT4_ERROR_INODE(inode, | |
1232 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1233 | *logical, le32_to_cpu(ex->ee_block)); | |
1234 | return -EIO; | |
1235 | } | |
1988b51e AT |
1236 | while (--depth >= 0) { |
1237 | ix = path[depth].p_idx; | |
273df556 FM |
1238 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1239 | EXT4_ERROR_INODE(inode, | |
1240 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
1241 | ix != NULL ? ix->ei_block : 0, | |
1242 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? | |
1243 | EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0, | |
1244 | depth); | |
1245 | return -EIO; | |
1246 | } | |
1988b51e AT |
1247 | } |
1248 | return 0; | |
1249 | } | |
1250 | ||
273df556 FM |
1251 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1252 | EXT4_ERROR_INODE(inode, | |
1253 | "logical %d < ee_block %d + ee_len %d!", | |
1254 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1255 | return -EIO; | |
1256 | } | |
1988b51e | 1257 | |
b939e376 | 1258 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1259 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1260 | return 0; |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * search the closest allocated block to the right for *logical | |
1265 | * and returns it at @logical + it's physical address at @phys | |
1266 | * if *logical is the smallest allocated block, the function | |
1267 | * returns 0 at @phys | |
1268 | * return value contains 0 (success) or error code | |
1269 | */ | |
1f109d5a TT |
1270 | static int ext4_ext_search_right(struct inode *inode, |
1271 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1272 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1273 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1274 | { |
1275 | struct buffer_head *bh = NULL; | |
1276 | struct ext4_extent_header *eh; | |
1277 | struct ext4_extent_idx *ix; | |
1278 | struct ext4_extent *ex; | |
1279 | ext4_fsblk_t block; | |
395a87bf ES |
1280 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1281 | int ee_len; | |
1988b51e | 1282 | |
273df556 FM |
1283 | if (unlikely(path == NULL)) { |
1284 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1285 | return -EIO; | |
1286 | } | |
1988b51e AT |
1287 | depth = path->p_depth; |
1288 | *phys = 0; | |
1289 | ||
1290 | if (depth == 0 && path->p_ext == NULL) | |
1291 | return 0; | |
1292 | ||
1293 | /* usually extent in the path covers blocks smaller | |
1294 | * then *logical, but it can be that extent is the | |
1295 | * first one in the file */ | |
1296 | ||
1297 | ex = path[depth].p_ext; | |
b939e376 | 1298 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1299 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1300 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1301 | EXT4_ERROR_INODE(inode, | |
1302 | "first_extent(path[%d].p_hdr) != ex", | |
1303 | depth); | |
1304 | return -EIO; | |
1305 | } | |
1988b51e AT |
1306 | while (--depth >= 0) { |
1307 | ix = path[depth].p_idx; | |
273df556 FM |
1308 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1309 | EXT4_ERROR_INODE(inode, | |
1310 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1311 | *logical); | |
1312 | return -EIO; | |
1313 | } | |
1988b51e | 1314 | } |
4d33b1ef | 1315 | goto found_extent; |
1988b51e AT |
1316 | } |
1317 | ||
273df556 FM |
1318 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1319 | EXT4_ERROR_INODE(inode, | |
1320 | "logical %d < ee_block %d + ee_len %d!", | |
1321 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1322 | return -EIO; | |
1323 | } | |
1988b51e AT |
1324 | |
1325 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1326 | /* next allocated block in this leaf */ | |
1327 | ex++; | |
4d33b1ef | 1328 | goto found_extent; |
1988b51e AT |
1329 | } |
1330 | ||
1331 | /* go up and search for index to the right */ | |
1332 | while (--depth >= 0) { | |
1333 | ix = path[depth].p_idx; | |
1334 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1335 | goto got_index; |
1988b51e AT |
1336 | } |
1337 | ||
25f1ee3a WF |
1338 | /* we've gone up to the root and found no index to the right */ |
1339 | return 0; | |
1988b51e | 1340 | |
25f1ee3a | 1341 | got_index: |
1988b51e AT |
1342 | /* we've found index to the right, let's |
1343 | * follow it and find the closest allocated | |
1344 | * block to the right */ | |
1345 | ix++; | |
bf89d16f | 1346 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1347 | while (++depth < path->p_depth) { |
1348 | bh = sb_bread(inode->i_sb, block); | |
1349 | if (bh == NULL) | |
1350 | return -EIO; | |
1351 | eh = ext_block_hdr(bh); | |
395a87bf | 1352 | /* subtract from p_depth to get proper eh_depth */ |
56b19868 | 1353 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1354 | put_bh(bh); |
1355 | return -EIO; | |
1356 | } | |
1357 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1358 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1359 | put_bh(bh); |
1360 | } | |
1361 | ||
1362 | bh = sb_bread(inode->i_sb, block); | |
1363 | if (bh == NULL) | |
1364 | return -EIO; | |
1365 | eh = ext_block_hdr(bh); | |
56b19868 | 1366 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1367 | put_bh(bh); |
1368 | return -EIO; | |
1369 | } | |
1370 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1371 | found_extent: |
1988b51e | 1372 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1373 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1374 | *ret_ex = ex; |
1375 | if (bh) | |
1376 | put_bh(bh); | |
1988b51e | 1377 | return 0; |
1988b51e AT |
1378 | } |
1379 | ||
a86c6181 | 1380 | /* |
d0d856e8 | 1381 | * ext4_ext_next_allocated_block: |
f17722f9 | 1382 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1383 | * NOTE: it considers block number from index entry as |
1384 | * allocated block. Thus, index entries have to be consistent | |
1385 | * with leaves. | |
a86c6181 | 1386 | */ |
725d26d3 | 1387 | static ext4_lblk_t |
a86c6181 AT |
1388 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1389 | { | |
1390 | int depth; | |
1391 | ||
1392 | BUG_ON(path == NULL); | |
1393 | depth = path->p_depth; | |
1394 | ||
1395 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1396 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1397 | |
1398 | while (depth >= 0) { | |
1399 | if (depth == path->p_depth) { | |
1400 | /* leaf */ | |
1401 | if (path[depth].p_ext != | |
1402 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1403 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1404 | } else { | |
1405 | /* index */ | |
1406 | if (path[depth].p_idx != | |
1407 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1408 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1409 | } | |
1410 | depth--; | |
1411 | } | |
1412 | ||
f17722f9 | 1413 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1414 | } |
1415 | ||
1416 | /* | |
d0d856e8 | 1417 | * ext4_ext_next_leaf_block: |
f17722f9 | 1418 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1419 | */ |
5718789d | 1420 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1421 | { |
1422 | int depth; | |
1423 | ||
1424 | BUG_ON(path == NULL); | |
1425 | depth = path->p_depth; | |
1426 | ||
1427 | /* zero-tree has no leaf blocks at all */ | |
1428 | if (depth == 0) | |
f17722f9 | 1429 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1430 | |
1431 | /* go to index block */ | |
1432 | depth--; | |
1433 | ||
1434 | while (depth >= 0) { | |
1435 | if (path[depth].p_idx != | |
1436 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1437 | return (ext4_lblk_t) |
1438 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1439 | depth--; |
1440 | } | |
1441 | ||
f17722f9 | 1442 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1443 | } |
1444 | ||
1445 | /* | |
d0d856e8 RD |
1446 | * ext4_ext_correct_indexes: |
1447 | * if leaf gets modified and modified extent is first in the leaf, | |
1448 | * then we have to correct all indexes above. | |
a86c6181 AT |
1449 | * TODO: do we need to correct tree in all cases? |
1450 | */ | |
1d03ec98 | 1451 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1452 | struct ext4_ext_path *path) |
1453 | { | |
1454 | struct ext4_extent_header *eh; | |
1455 | int depth = ext_depth(inode); | |
1456 | struct ext4_extent *ex; | |
1457 | __le32 border; | |
1458 | int k, err = 0; | |
1459 | ||
1460 | eh = path[depth].p_hdr; | |
1461 | ex = path[depth].p_ext; | |
273df556 FM |
1462 | |
1463 | if (unlikely(ex == NULL || eh == NULL)) { | |
1464 | EXT4_ERROR_INODE(inode, | |
1465 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1466 | return -EIO; | |
1467 | } | |
a86c6181 AT |
1468 | |
1469 | if (depth == 0) { | |
1470 | /* there is no tree at all */ | |
1471 | return 0; | |
1472 | } | |
1473 | ||
1474 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1475 | /* we correct tree if first leaf got modified only */ | |
1476 | return 0; | |
1477 | } | |
1478 | ||
1479 | /* | |
d0d856e8 | 1480 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1481 | */ |
1482 | k = depth - 1; | |
1483 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1484 | err = ext4_ext_get_access(handle, inode, path + k); |
1485 | if (err) | |
a86c6181 AT |
1486 | return err; |
1487 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1488 | err = ext4_ext_dirty(handle, inode, path + k); |
1489 | if (err) | |
a86c6181 AT |
1490 | return err; |
1491 | ||
1492 | while (k--) { | |
1493 | /* change all left-side indexes */ | |
1494 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1495 | break; | |
7e028976 AM |
1496 | err = ext4_ext_get_access(handle, inode, path + k); |
1497 | if (err) | |
a86c6181 AT |
1498 | break; |
1499 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1500 | err = ext4_ext_dirty(handle, inode, path + k); |
1501 | if (err) | |
a86c6181 AT |
1502 | break; |
1503 | } | |
1504 | ||
1505 | return err; | |
1506 | } | |
1507 | ||
748de673 | 1508 | int |
a86c6181 AT |
1509 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1510 | struct ext4_extent *ex2) | |
1511 | { | |
749269fa | 1512 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1513 | |
1514 | /* | |
1515 | * Make sure that either both extents are uninitialized, or | |
1516 | * both are _not_. | |
1517 | */ | |
1518 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1519 | return 0; | |
1520 | ||
749269fa AA |
1521 | if (ext4_ext_is_uninitialized(ex1)) |
1522 | max_len = EXT_UNINIT_MAX_LEN; | |
1523 | else | |
1524 | max_len = EXT_INIT_MAX_LEN; | |
1525 | ||
a2df2a63 AA |
1526 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1527 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1528 | ||
1529 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1530 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1531 | return 0; |
1532 | ||
471d4011 SB |
1533 | /* |
1534 | * To allow future support for preallocated extents to be added | |
1535 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1536 | * this can result in the top bit of ee_len being set. |
471d4011 | 1537 | */ |
749269fa | 1538 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1539 | return 0; |
bbf2f9fb | 1540 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1541 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1542 | return 0; |
1543 | #endif | |
1544 | ||
bf89d16f | 1545 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1546 | return 1; |
1547 | return 0; | |
1548 | } | |
1549 | ||
56055d3a AA |
1550 | /* |
1551 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1552 | * It always tries to merge towards right. If you want to merge towards | |
1553 | * left, pass "ex - 1" as argument instead of "ex". | |
1554 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1555 | * 1 if they got merged. | |
1556 | */ | |
197217a5 | 1557 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1558 | struct ext4_ext_path *path, |
1559 | struct ext4_extent *ex) | |
56055d3a AA |
1560 | { |
1561 | struct ext4_extent_header *eh; | |
1562 | unsigned int depth, len; | |
1563 | int merge_done = 0; | |
1564 | int uninitialized = 0; | |
1565 | ||
1566 | depth = ext_depth(inode); | |
1567 | BUG_ON(path[depth].p_hdr == NULL); | |
1568 | eh = path[depth].p_hdr; | |
1569 | ||
1570 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1571 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1572 | break; | |
1573 | /* merge with next extent! */ | |
1574 | if (ext4_ext_is_uninitialized(ex)) | |
1575 | uninitialized = 1; | |
1576 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1577 | + ext4_ext_get_actual_len(ex + 1)); | |
1578 | if (uninitialized) | |
1579 | ext4_ext_mark_uninitialized(ex); | |
1580 | ||
1581 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1582 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1583 | * sizeof(struct ext4_extent); | |
1584 | memmove(ex + 1, ex + 2, len); | |
1585 | } | |
e8546d06 | 1586 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1587 | merge_done = 1; |
1588 | WARN_ON(eh->eh_entries == 0); | |
1589 | if (!eh->eh_entries) | |
24676da4 | 1590 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1591 | } |
1592 | ||
1593 | return merge_done; | |
1594 | } | |
1595 | ||
197217a5 YY |
1596 | /* |
1597 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1598 | * return 1 if merge left else 0. | |
1599 | */ | |
1600 | static int ext4_ext_try_to_merge(struct inode *inode, | |
1601 | struct ext4_ext_path *path, | |
1602 | struct ext4_extent *ex) { | |
1603 | struct ext4_extent_header *eh; | |
1604 | unsigned int depth; | |
1605 | int merge_done = 0; | |
1606 | int ret = 0; | |
1607 | ||
1608 | depth = ext_depth(inode); | |
1609 | BUG_ON(path[depth].p_hdr == NULL); | |
1610 | eh = path[depth].p_hdr; | |
1611 | ||
1612 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1613 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1614 | ||
1615 | if (!merge_done) | |
1616 | ret = ext4_ext_try_to_merge_right(inode, path, ex); | |
1617 | ||
1618 | return ret; | |
1619 | } | |
1620 | ||
25d14f98 AA |
1621 | /* |
1622 | * check if a portion of the "newext" extent overlaps with an | |
1623 | * existing extent. | |
1624 | * | |
1625 | * If there is an overlap discovered, it updates the length of the newext | |
1626 | * such that there will be no overlap, and then returns 1. | |
1627 | * If there is no overlap found, it returns 0. | |
1628 | */ | |
4d33b1ef TT |
1629 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1630 | struct inode *inode, | |
1f109d5a TT |
1631 | struct ext4_extent *newext, |
1632 | struct ext4_ext_path *path) | |
25d14f98 | 1633 | { |
725d26d3 | 1634 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1635 | unsigned int depth, len1; |
1636 | unsigned int ret = 0; | |
1637 | ||
1638 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1639 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1640 | depth = ext_depth(inode); |
1641 | if (!path[depth].p_ext) | |
1642 | goto out; | |
1643 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1644 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1645 | |
1646 | /* | |
1647 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1648 | * is before the requested block(s) |
25d14f98 AA |
1649 | */ |
1650 | if (b2 < b1) { | |
1651 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1652 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1653 | goto out; |
4d33b1ef | 1654 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1655 | } |
1656 | ||
725d26d3 | 1657 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1658 | if (b1 + len1 < b1) { |
f17722f9 | 1659 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1660 | newext->ee_len = cpu_to_le16(len1); |
1661 | ret = 1; | |
1662 | } | |
1663 | ||
1664 | /* check for overlap */ | |
1665 | if (b1 + len1 > b2) { | |
1666 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1667 | ret = 1; | |
1668 | } | |
1669 | out: | |
1670 | return ret; | |
1671 | } | |
1672 | ||
a86c6181 | 1673 | /* |
d0d856e8 RD |
1674 | * ext4_ext_insert_extent: |
1675 | * tries to merge requsted extent into the existing extent or | |
1676 | * inserts requested extent as new one into the tree, | |
1677 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1678 | */ |
1679 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1680 | struct ext4_ext_path *path, | |
0031462b | 1681 | struct ext4_extent *newext, int flag) |
a86c6181 | 1682 | { |
af5bc92d | 1683 | struct ext4_extent_header *eh; |
a86c6181 AT |
1684 | struct ext4_extent *ex, *fex; |
1685 | struct ext4_extent *nearex; /* nearest extent */ | |
1686 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1687 | int depth, len, err; |
1688 | ext4_lblk_t next; | |
a2df2a63 | 1689 | unsigned uninitialized = 0; |
55f020db | 1690 | int flags = 0; |
a86c6181 | 1691 | |
273df556 FM |
1692 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1693 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1694 | return -EIO; | |
1695 | } | |
a86c6181 AT |
1696 | depth = ext_depth(inode); |
1697 | ex = path[depth].p_ext; | |
273df556 FM |
1698 | if (unlikely(path[depth].p_hdr == NULL)) { |
1699 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1700 | return -EIO; | |
1701 | } | |
a86c6181 AT |
1702 | |
1703 | /* try to insert block into found extent and return */ | |
744692dc | 1704 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1705 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
553f9008 | 1706 | ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", |
bf89d16f TT |
1707 | ext4_ext_is_uninitialized(newext), |
1708 | ext4_ext_get_actual_len(newext), | |
1709 | le32_to_cpu(ex->ee_block), | |
1710 | ext4_ext_is_uninitialized(ex), | |
1711 | ext4_ext_get_actual_len(ex), | |
1712 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1713 | err = ext4_ext_get_access(handle, inode, path + depth); |
1714 | if (err) | |
a86c6181 | 1715 | return err; |
a2df2a63 AA |
1716 | |
1717 | /* | |
1718 | * ext4_can_extents_be_merged should have checked that either | |
1719 | * both extents are uninitialized, or both aren't. Thus we | |
1720 | * need to check only one of them here. | |
1721 | */ | |
1722 | if (ext4_ext_is_uninitialized(ex)) | |
1723 | uninitialized = 1; | |
1724 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1725 | + ext4_ext_get_actual_len(newext)); | |
1726 | if (uninitialized) | |
1727 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1728 | eh = path[depth].p_hdr; |
1729 | nearex = ex; | |
1730 | goto merge; | |
1731 | } | |
1732 | ||
a86c6181 AT |
1733 | depth = ext_depth(inode); |
1734 | eh = path[depth].p_hdr; | |
1735 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1736 | goto has_space; | |
1737 | ||
1738 | /* probably next leaf has space for us? */ | |
1739 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1740 | next = EXT_MAX_BLOCKS; |
1741 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1742 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1743 | if (next != EXT_MAX_BLOCKS) { |
a86c6181 AT |
1744 | ext_debug("next leaf block - %d\n", next); |
1745 | BUG_ON(npath != NULL); | |
1746 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1747 | if (IS_ERR(npath)) | |
1748 | return PTR_ERR(npath); | |
1749 | BUG_ON(npath->p_depth != path->p_depth); | |
1750 | eh = npath[depth].p_hdr; | |
1751 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1752 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1753 | le16_to_cpu(eh->eh_entries)); |
1754 | path = npath; | |
ffb505ff | 1755 | goto has_space; |
a86c6181 AT |
1756 | } |
1757 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1758 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1759 | } | |
1760 | ||
1761 | /* | |
d0d856e8 RD |
1762 | * There is no free space in the found leaf. |
1763 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1764 | */ |
55f020db AH |
1765 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1766 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1767 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1768 | if (err) |
1769 | goto cleanup; | |
1770 | depth = ext_depth(inode); | |
1771 | eh = path[depth].p_hdr; | |
1772 | ||
1773 | has_space: | |
1774 | nearex = path[depth].p_ext; | |
1775 | ||
7e028976 AM |
1776 | err = ext4_ext_get_access(handle, inode, path + depth); |
1777 | if (err) | |
a86c6181 AT |
1778 | goto cleanup; |
1779 | ||
1780 | if (!nearex) { | |
1781 | /* there is no extent in this leaf, create first one */ | |
553f9008 | 1782 | ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n", |
8c55e204 | 1783 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1784 | ext4_ext_pblock(newext), |
553f9008 | 1785 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1786 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1787 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1788 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1789 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1790 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1791 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1792 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1793 | len = (len - 1) * sizeof(struct ext4_extent); | |
1794 | len = len < 0 ? 0 : len; | |
553f9008 | 1795 | ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " |
a86c6181 | 1796 | "move %d from 0x%p to 0x%p\n", |
8c55e204 | 1797 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1798 | ext4_ext_pblock(newext), |
553f9008 | 1799 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1800 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1801 | nearex, len, nearex + 1, nearex + 2); |
1802 | memmove(nearex + 2, nearex + 1, len); | |
1803 | } | |
1804 | path[depth].p_ext = nearex + 1; | |
1805 | } else { | |
1806 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1807 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1808 | len = len < 0 ? 0 : len; | |
553f9008 | 1809 | ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " |
a86c6181 AT |
1810 | "move %d from 0x%p to 0x%p\n", |
1811 | le32_to_cpu(newext->ee_block), | |
bf89d16f | 1812 | ext4_ext_pblock(newext), |
553f9008 | 1813 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1814 | ext4_ext_get_actual_len(newext), |
0737964b | 1815 | nearex, len, nearex, nearex + 1); |
a86c6181 AT |
1816 | memmove(nearex + 1, nearex, len); |
1817 | path[depth].p_ext = nearex; | |
1818 | } | |
1819 | ||
e8546d06 | 1820 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1821 | nearex = path[depth].p_ext; |
1822 | nearex->ee_block = newext->ee_block; | |
bf89d16f | 1823 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1824 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1825 | |
1826 | merge: | |
1827 | /* try to merge extents to the right */ | |
744692dc | 1828 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
0031462b | 1829 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1830 | |
1831 | /* try to merge extents to the left */ | |
1832 | ||
1833 | /* time to correct all indexes above */ | |
1834 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1835 | if (err) | |
1836 | goto cleanup; | |
1837 | ||
1838 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1839 | ||
1840 | cleanup: | |
1841 | if (npath) { | |
1842 | ext4_ext_drop_refs(npath); | |
1843 | kfree(npath); | |
1844 | } | |
a86c6181 AT |
1845 | ext4_ext_invalidate_cache(inode); |
1846 | return err; | |
1847 | } | |
1848 | ||
1f109d5a TT |
1849 | static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1850 | ext4_lblk_t num, ext_prepare_callback func, | |
1851 | void *cbdata) | |
6873fa0d ES |
1852 | { |
1853 | struct ext4_ext_path *path = NULL; | |
1854 | struct ext4_ext_cache cbex; | |
1855 | struct ext4_extent *ex; | |
1856 | ext4_lblk_t next, start = 0, end = 0; | |
1857 | ext4_lblk_t last = block + num; | |
1858 | int depth, exists, err = 0; | |
1859 | ||
1860 | BUG_ON(func == NULL); | |
1861 | BUG_ON(inode == NULL); | |
1862 | ||
f17722f9 | 1863 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1864 | num = last - block; |
1865 | /* find extent for this block */ | |
fab3a549 | 1866 | down_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d | 1867 | path = ext4_ext_find_extent(inode, block, path); |
fab3a549 | 1868 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1869 | if (IS_ERR(path)) { |
1870 | err = PTR_ERR(path); | |
1871 | path = NULL; | |
1872 | break; | |
1873 | } | |
1874 | ||
1875 | depth = ext_depth(inode); | |
273df556 FM |
1876 | if (unlikely(path[depth].p_hdr == NULL)) { |
1877 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1878 | err = -EIO; | |
1879 | break; | |
1880 | } | |
6873fa0d ES |
1881 | ex = path[depth].p_ext; |
1882 | next = ext4_ext_next_allocated_block(path); | |
1883 | ||
1884 | exists = 0; | |
1885 | if (!ex) { | |
1886 | /* there is no extent yet, so try to allocate | |
1887 | * all requested space */ | |
1888 | start = block; | |
1889 | end = block + num; | |
1890 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1891 | /* need to allocate space before found extent */ | |
1892 | start = block; | |
1893 | end = le32_to_cpu(ex->ee_block); | |
1894 | if (block + num < end) | |
1895 | end = block + num; | |
1896 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1897 | + ext4_ext_get_actual_len(ex)) { | |
1898 | /* need to allocate space after found extent */ | |
1899 | start = block; | |
1900 | end = block + num; | |
1901 | if (end >= next) | |
1902 | end = next; | |
1903 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1904 | /* | |
1905 | * some part of requested space is covered | |
1906 | * by found extent | |
1907 | */ | |
1908 | start = block; | |
1909 | end = le32_to_cpu(ex->ee_block) | |
1910 | + ext4_ext_get_actual_len(ex); | |
1911 | if (block + num < end) | |
1912 | end = block + num; | |
1913 | exists = 1; | |
1914 | } else { | |
1915 | BUG(); | |
1916 | } | |
1917 | BUG_ON(end <= start); | |
1918 | ||
1919 | if (!exists) { | |
1920 | cbex.ec_block = start; | |
1921 | cbex.ec_len = end - start; | |
1922 | cbex.ec_start = 0; | |
6873fa0d ES |
1923 | } else { |
1924 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1925 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
bf89d16f | 1926 | cbex.ec_start = ext4_ext_pblock(ex); |
6873fa0d ES |
1927 | } |
1928 | ||
273df556 FM |
1929 | if (unlikely(cbex.ec_len == 0)) { |
1930 | EXT4_ERROR_INODE(inode, "cbex.ec_len == 0"); | |
1931 | err = -EIO; | |
1932 | break; | |
1933 | } | |
c03f8aa9 | 1934 | err = func(inode, next, &cbex, ex, cbdata); |
6873fa0d ES |
1935 | ext4_ext_drop_refs(path); |
1936 | ||
1937 | if (err < 0) | |
1938 | break; | |
1939 | ||
1940 | if (err == EXT_REPEAT) | |
1941 | continue; | |
1942 | else if (err == EXT_BREAK) { | |
1943 | err = 0; | |
1944 | break; | |
1945 | } | |
1946 | ||
1947 | if (ext_depth(inode) != depth) { | |
1948 | /* depth was changed. we have to realloc path */ | |
1949 | kfree(path); | |
1950 | path = NULL; | |
1951 | } | |
1952 | ||
1953 | block = cbex.ec_block + cbex.ec_len; | |
1954 | } | |
1955 | ||
1956 | if (path) { | |
1957 | ext4_ext_drop_refs(path); | |
1958 | kfree(path); | |
1959 | } | |
1960 | ||
1961 | return err; | |
1962 | } | |
1963 | ||
09b88252 | 1964 | static void |
725d26d3 | 1965 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 1966 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
1967 | { |
1968 | struct ext4_ext_cache *cex; | |
1969 | BUG_ON(len == 0); | |
2ec0ae3a | 1970 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
d8990240 | 1971 | trace_ext4_ext_put_in_cache(inode, block, len, start); |
a86c6181 | 1972 | cex = &EXT4_I(inode)->i_cached_extent; |
a86c6181 AT |
1973 | cex->ec_block = block; |
1974 | cex->ec_len = len; | |
1975 | cex->ec_start = start; | |
2ec0ae3a | 1976 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
1977 | } |
1978 | ||
1979 | /* | |
d0d856e8 RD |
1980 | * ext4_ext_put_gap_in_cache: |
1981 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
1982 | * and cache this gap |
1983 | */ | |
09b88252 | 1984 | static void |
a86c6181 | 1985 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 1986 | ext4_lblk_t block) |
a86c6181 AT |
1987 | { |
1988 | int depth = ext_depth(inode); | |
725d26d3 AK |
1989 | unsigned long len; |
1990 | ext4_lblk_t lblock; | |
a86c6181 AT |
1991 | struct ext4_extent *ex; |
1992 | ||
1993 | ex = path[depth].p_ext; | |
1994 | if (ex == NULL) { | |
1995 | /* there is no extent yet, so gap is [0;-] */ | |
1996 | lblock = 0; | |
f17722f9 | 1997 | len = EXT_MAX_BLOCKS; |
a86c6181 AT |
1998 | ext_debug("cache gap(whole file):"); |
1999 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2000 | lblock = block; | |
2001 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2002 | ext_debug("cache gap(before): %u [%u:%u]", |
2003 | block, | |
2004 | le32_to_cpu(ex->ee_block), | |
2005 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 2006 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2007 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2008 | ext4_lblk_t next; |
8c55e204 | 2009 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2010 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2011 | |
2012 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2013 | ext_debug("cache gap(after): [%u:%u] %u", |
2014 | le32_to_cpu(ex->ee_block), | |
2015 | ext4_ext_get_actual_len(ex), | |
2016 | block); | |
725d26d3 AK |
2017 | BUG_ON(next == lblock); |
2018 | len = next - lblock; | |
a86c6181 AT |
2019 | } else { |
2020 | lblock = len = 0; | |
2021 | BUG(); | |
2022 | } | |
2023 | ||
bba90743 | 2024 | ext_debug(" -> %u:%lu\n", lblock, len); |
b05e6ae5 | 2025 | ext4_ext_put_in_cache(inode, lblock, len, 0); |
a86c6181 AT |
2026 | } |
2027 | ||
b05e6ae5 | 2028 | /* |
b7ca1e8e | 2029 | * ext4_ext_check_cache() |
a4bb6b64 AH |
2030 | * Checks to see if the given block is in the cache. |
2031 | * If it is, the cached extent is stored in the given | |
2032 | * cache extent pointer. If the cached extent is a hole, | |
2033 | * this routine should be used instead of | |
2034 | * ext4_ext_in_cache if the calling function needs to | |
2035 | * know the size of the hole. | |
2036 | * | |
2037 | * @inode: The files inode | |
2038 | * @block: The block to look for in the cache | |
2039 | * @ex: Pointer where the cached extent will be stored | |
2040 | * if it contains block | |
2041 | * | |
b05e6ae5 TT |
2042 | * Return 0 if cache is invalid; 1 if the cache is valid |
2043 | */ | |
a4bb6b64 AH |
2044 | static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block, |
2045 | struct ext4_ext_cache *ex){ | |
a86c6181 | 2046 | struct ext4_ext_cache *cex; |
77f4135f | 2047 | struct ext4_sb_info *sbi; |
b05e6ae5 | 2048 | int ret = 0; |
a86c6181 | 2049 | |
60e6679e | 2050 | /* |
2ec0ae3a TT |
2051 | * We borrow i_block_reservation_lock to protect i_cached_extent |
2052 | */ | |
2053 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
a86c6181 | 2054 | cex = &EXT4_I(inode)->i_cached_extent; |
77f4135f | 2055 | sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2056 | |
2057 | /* has cache valid data? */ | |
b05e6ae5 | 2058 | if (cex->ec_len == 0) |
2ec0ae3a | 2059 | goto errout; |
a86c6181 | 2060 | |
731eb1a0 | 2061 | if (in_range(block, cex->ec_block, cex->ec_len)) { |
a4bb6b64 | 2062 | memcpy(ex, cex, sizeof(struct ext4_ext_cache)); |
bba90743 ES |
2063 | ext_debug("%u cached by %u:%u:%llu\n", |
2064 | block, | |
2065 | cex->ec_block, cex->ec_len, cex->ec_start); | |
b05e6ae5 | 2066 | ret = 1; |
a86c6181 | 2067 | } |
2ec0ae3a | 2068 | errout: |
77f4135f VH |
2069 | if (!ret) |
2070 | sbi->extent_cache_misses++; | |
2071 | else | |
2072 | sbi->extent_cache_hits++; | |
d8990240 | 2073 | trace_ext4_ext_in_cache(inode, block, ret); |
2ec0ae3a TT |
2074 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2075 | return ret; | |
a86c6181 AT |
2076 | } |
2077 | ||
a4bb6b64 AH |
2078 | /* |
2079 | * ext4_ext_in_cache() | |
2080 | * Checks to see if the given block is in the cache. | |
2081 | * If it is, the cached extent is stored in the given | |
2082 | * extent pointer. | |
2083 | * | |
2084 | * @inode: The files inode | |
2085 | * @block: The block to look for in the cache | |
2086 | * @ex: Pointer where the cached extent will be stored | |
2087 | * if it contains block | |
2088 | * | |
2089 | * Return 0 if cache is invalid; 1 if the cache is valid | |
2090 | */ | |
2091 | static int | |
2092 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, | |
2093 | struct ext4_extent *ex) | |
2094 | { | |
2095 | struct ext4_ext_cache cex; | |
2096 | int ret = 0; | |
2097 | ||
2098 | if (ext4_ext_check_cache(inode, block, &cex)) { | |
2099 | ex->ee_block = cpu_to_le32(cex.ec_block); | |
2100 | ext4_ext_store_pblock(ex, cex.ec_start); | |
2101 | ex->ee_len = cpu_to_le16(cex.ec_len); | |
2102 | ret = 1; | |
2103 | } | |
2104 | ||
2105 | return ret; | |
2106 | } | |
2107 | ||
2108 | ||
a86c6181 | 2109 | /* |
d0d856e8 RD |
2110 | * ext4_ext_rm_idx: |
2111 | * removes index from the index block. | |
a86c6181 | 2112 | */ |
1d03ec98 | 2113 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
2114 | struct ext4_ext_path *path) |
2115 | { | |
a86c6181 | 2116 | int err; |
f65e6fba | 2117 | ext4_fsblk_t leaf; |
a86c6181 AT |
2118 | |
2119 | /* free index block */ | |
2120 | path--; | |
bf89d16f | 2121 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2122 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2123 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2124 | return -EIO; | |
2125 | } | |
7e028976 AM |
2126 | err = ext4_ext_get_access(handle, inode, path); |
2127 | if (err) | |
a86c6181 | 2128 | return err; |
0e1147b0 RD |
2129 | |
2130 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2131 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2132 | len *= sizeof(struct ext4_extent_idx); | |
2133 | memmove(path->p_idx, path->p_idx + 1, len); | |
2134 | } | |
2135 | ||
e8546d06 | 2136 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2137 | err = ext4_ext_dirty(handle, inode, path); |
2138 | if (err) | |
a86c6181 | 2139 | return err; |
2ae02107 | 2140 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2141 | trace_ext4_ext_rm_idx(inode, leaf); |
2142 | ||
7dc57615 | 2143 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2144 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
a86c6181 AT |
2145 | return err; |
2146 | } | |
2147 | ||
2148 | /* | |
ee12b630 MC |
2149 | * ext4_ext_calc_credits_for_single_extent: |
2150 | * This routine returns max. credits that needed to insert an extent | |
2151 | * to the extent tree. | |
2152 | * When pass the actual path, the caller should calculate credits | |
2153 | * under i_data_sem. | |
a86c6181 | 2154 | */ |
525f4ed8 | 2155 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2156 | struct ext4_ext_path *path) |
2157 | { | |
a86c6181 | 2158 | if (path) { |
ee12b630 | 2159 | int depth = ext_depth(inode); |
f3bd1f3f | 2160 | int ret = 0; |
ee12b630 | 2161 | |
a86c6181 | 2162 | /* probably there is space in leaf? */ |
a86c6181 | 2163 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2164 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2165 | |
ee12b630 MC |
2166 | /* |
2167 | * There are some space in the leaf tree, no | |
2168 | * need to account for leaf block credit | |
2169 | * | |
2170 | * bitmaps and block group descriptor blocks | |
2171 | * and other metadat blocks still need to be | |
2172 | * accounted. | |
2173 | */ | |
525f4ed8 | 2174 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2175 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2176 | return ret; |
ee12b630 MC |
2177 | } |
2178 | } | |
a86c6181 | 2179 | |
525f4ed8 | 2180 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2181 | } |
a86c6181 | 2182 | |
ee12b630 MC |
2183 | /* |
2184 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2185 | * | |
2186 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2187 | * in the worse case, each tree level index/leaf need to be changed | |
2188 | * if the tree split due to insert a new extent, then the old tree | |
2189 | * index/leaf need to be updated too | |
2190 | * | |
2191 | * If the nrblocks are discontiguous, they could cause | |
2192 | * the whole tree split more than once, but this is really rare. | |
2193 | */ | |
525f4ed8 | 2194 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2195 | { |
2196 | int index; | |
2197 | int depth = ext_depth(inode); | |
a86c6181 | 2198 | |
ee12b630 MC |
2199 | if (chunk) |
2200 | index = depth * 2; | |
2201 | else | |
2202 | index = depth * 3; | |
a86c6181 | 2203 | |
ee12b630 | 2204 | return index; |
a86c6181 AT |
2205 | } |
2206 | ||
2207 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2208 | struct ext4_extent *ex, |
2209 | ext4_fsblk_t *partial_cluster, | |
2210 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2211 | { |
0aa06000 | 2212 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2213 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2214 | ext4_fsblk_t pblk; |
e6362609 | 2215 | int flags = EXT4_FREE_BLOCKS_FORGET; |
a86c6181 | 2216 | |
c9de560d | 2217 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
e6362609 | 2218 | flags |= EXT4_FREE_BLOCKS_METADATA; |
0aa06000 TT |
2219 | /* |
2220 | * For bigalloc file systems, we never free a partial cluster | |
2221 | * at the beginning of the extent. Instead, we make a note | |
2222 | * that we tried freeing the cluster, and check to see if we | |
2223 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2224 | * or at the end of the ext4_truncate() operation. | |
2225 | */ | |
2226 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2227 | ||
d8990240 | 2228 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2229 | /* |
2230 | * If we have a partial cluster, and it's different from the | |
2231 | * cluster of the last block, we need to explicitly free the | |
2232 | * partial cluster here. | |
2233 | */ | |
2234 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2235 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2236 | ext4_free_blocks(handle, inode, NULL, | |
2237 | EXT4_C2B(sbi, *partial_cluster), | |
2238 | sbi->s_cluster_ratio, flags); | |
2239 | *partial_cluster = 0; | |
2240 | } | |
2241 | ||
a86c6181 AT |
2242 | #ifdef EXTENTS_STATS |
2243 | { | |
2244 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2245 | spin_lock(&sbi->s_ext_stats_lock); |
2246 | sbi->s_ext_blocks += ee_len; | |
2247 | sbi->s_ext_extents++; | |
2248 | if (ee_len < sbi->s_ext_min) | |
2249 | sbi->s_ext_min = ee_len; | |
2250 | if (ee_len > sbi->s_ext_max) | |
2251 | sbi->s_ext_max = ee_len; | |
2252 | if (ext_depth(inode) > sbi->s_depth_max) | |
2253 | sbi->s_depth_max = ext_depth(inode); | |
2254 | spin_unlock(&sbi->s_ext_stats_lock); | |
2255 | } | |
2256 | #endif | |
2257 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2258 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2259 | /* tail removal */ |
725d26d3 | 2260 | ext4_lblk_t num; |
725d26d3 | 2261 | |
a2df2a63 | 2262 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2263 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2264 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2265 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2266 | /* | |
2267 | * If the block range to be freed didn't start at the | |
2268 | * beginning of a cluster, and we removed the entire | |
2269 | * extent, save the partial cluster here, since we | |
2270 | * might need to delete if we determine that the | |
2271 | * truncate operation has removed all of the blocks in | |
2272 | * the cluster. | |
2273 | */ | |
2274 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2275 | (ee_len == num)) | |
2276 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2277 | else | |
2278 | *partial_cluster = 0; | |
a86c6181 | 2279 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2280 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2281 | /* head removal */ |
2282 | ext4_lblk_t num; | |
2283 | ext4_fsblk_t start; | |
2284 | ||
2285 | num = to - from; | |
2286 | start = ext4_ext_pblock(ex); | |
2287 | ||
2288 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
2289 | ext4_free_blocks(handle, inode, 0, start, num, flags); | |
2290 | ||
a86c6181 | 2291 | } else { |
725d26d3 AK |
2292 | printk(KERN_INFO "strange request: removal(2) " |
2293 | "%u-%u from %u:%u\n", | |
2294 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2295 | } |
2296 | return 0; | |
2297 | } | |
2298 | ||
d583fb87 AH |
2299 | |
2300 | /* | |
2301 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2302 | * blocks appearing between "start" and "end", and splits the extents | |
2303 | * if "start" and "end" appear in the same extent | |
2304 | * | |
2305 | * @handle: The journal handle | |
2306 | * @inode: The files inode | |
2307 | * @path: The path to the leaf | |
2308 | * @start: The first block to remove | |
2309 | * @end: The last block to remove | |
2310 | */ | |
a86c6181 AT |
2311 | static int |
2312 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2313 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2314 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2315 | { |
0aa06000 | 2316 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2317 | int err = 0, correct_index = 0; |
2318 | int depth = ext_depth(inode), credits; | |
2319 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2320 | ext4_lblk_t a, b, block; |
2321 | unsigned num; | |
2322 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2323 | unsigned short ex_ee_len; |
a2df2a63 | 2324 | unsigned uninitialized = 0; |
a86c6181 | 2325 | struct ext4_extent *ex; |
d583fb87 | 2326 | struct ext4_map_blocks map; |
a86c6181 | 2327 | |
c29c0ae7 | 2328 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 2329 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
2330 | if (!path[depth].p_hdr) |
2331 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2332 | eh = path[depth].p_hdr; | |
273df556 FM |
2333 | if (unlikely(path[depth].p_hdr == NULL)) { |
2334 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2335 | return -EIO; | |
2336 | } | |
a86c6181 AT |
2337 | /* find where to start removing */ |
2338 | ex = EXT_LAST_EXTENT(eh); | |
2339 | ||
2340 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2341 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2342 | |
d8990240 AK |
2343 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2344 | ||
a86c6181 AT |
2345 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2346 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2347 | |
2348 | if (ext4_ext_is_uninitialized(ex)) | |
2349 | uninitialized = 1; | |
2350 | else | |
2351 | uninitialized = 0; | |
2352 | ||
553f9008 M |
2353 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2354 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2355 | path[depth].p_ext = ex; |
2356 | ||
2357 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2358 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2359 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2360 | |
2361 | ext_debug(" border %u:%u\n", a, b); | |
2362 | ||
d583fb87 AH |
2363 | /* If this extent is beyond the end of the hole, skip it */ |
2364 | if (end <= ex_ee_block) { | |
2365 | ex--; | |
2366 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2367 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2368 | continue; | |
2369 | } else if (a != ex_ee_block && | |
2370 | b != ex_ee_block + ex_ee_len - 1) { | |
2371 | /* | |
2372 | * If this is a truncate, then this condition should | |
2373 | * never happen because at least one of the end points | |
2374 | * needs to be on the edge of the extent. | |
2375 | */ | |
f17722f9 | 2376 | if (end == EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2377 | ext_debug(" bad truncate %u:%u\n", |
2378 | start, end); | |
2379 | block = 0; | |
2380 | num = 0; | |
2381 | err = -EIO; | |
2382 | goto out; | |
2383 | } | |
2384 | /* | |
2385 | * else this is a hole punch, so the extent needs to | |
2386 | * be split since neither edge of the hole is on the | |
2387 | * extent edge | |
2388 | */ | |
2389 | else{ | |
2390 | map.m_pblk = ext4_ext_pblock(ex); | |
2391 | map.m_lblk = ex_ee_block; | |
2392 | map.m_len = b - ex_ee_block; | |
2393 | ||
2394 | err = ext4_split_extent(handle, | |
2395 | inode, path, &map, 0, | |
2396 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT | | |
2397 | EXT4_GET_BLOCKS_PRE_IO); | |
2398 | ||
2399 | if (err < 0) | |
2400 | goto out; | |
2401 | ||
2402 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2403 | ||
2404 | b = ex_ee_block+ex_ee_len - 1 < end ? | |
2405 | ex_ee_block+ex_ee_len - 1 : end; | |
2406 | ||
2407 | /* Then remove tail of this extent */ | |
2408 | block = ex_ee_block; | |
2409 | num = a - block; | |
2410 | } | |
a86c6181 AT |
2411 | } else if (a != ex_ee_block) { |
2412 | /* remove tail of the extent */ | |
2413 | block = ex_ee_block; | |
2414 | num = a - block; | |
2415 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
2416 | /* remove head of the extent */ | |
d583fb87 AH |
2417 | block = b; |
2418 | num = ex_ee_block + ex_ee_len - b; | |
2419 | ||
2420 | /* | |
2421 | * If this is a truncate, this condition | |
2422 | * should never happen | |
2423 | */ | |
f17722f9 | 2424 | if (end == EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2425 | ext_debug(" bad truncate %u:%u\n", |
2426 | start, end); | |
2427 | err = -EIO; | |
2428 | goto out; | |
2429 | } | |
a86c6181 AT |
2430 | } else { |
2431 | /* remove whole extent: excellent! */ | |
2432 | block = ex_ee_block; | |
2433 | num = 0; | |
d583fb87 AH |
2434 | if (a != ex_ee_block) { |
2435 | ext_debug(" bad truncate %u:%u\n", | |
2436 | start, end); | |
2437 | err = -EIO; | |
2438 | goto out; | |
2439 | } | |
2440 | ||
2441 | if (b != ex_ee_block + ex_ee_len - 1) { | |
2442 | ext_debug(" bad truncate %u:%u\n", | |
2443 | start, end); | |
2444 | err = -EIO; | |
2445 | goto out; | |
2446 | } | |
a86c6181 AT |
2447 | } |
2448 | ||
34071da7 TT |
2449 | /* |
2450 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2451 | * descriptor) for each block group; assume two block | |
2452 | * groups plus ex_ee_len/blocks_per_block_group for | |
2453 | * the worst case | |
2454 | */ | |
2455 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2456 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2457 | correct_index = 1; | |
2458 | credits += (ext_depth(inode)) + 1; | |
2459 | } | |
5aca07eb | 2460 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2461 | |
487caeef | 2462 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2463 | if (err) |
a86c6181 | 2464 | goto out; |
a86c6181 AT |
2465 | |
2466 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2467 | if (err) | |
2468 | goto out; | |
2469 | ||
0aa06000 TT |
2470 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2471 | a, b); | |
a86c6181 AT |
2472 | if (err) |
2473 | goto out; | |
2474 | ||
2475 | if (num == 0) { | |
d0d856e8 | 2476 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2477 | ext4_ext_store_pblock(ex, 0); |
d583fb87 AH |
2478 | } else if (block != ex_ee_block) { |
2479 | /* | |
2480 | * If this was a head removal, then we need to update | |
2481 | * the physical block since it is now at a different | |
2482 | * location | |
2483 | */ | |
2484 | ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a)); | |
a86c6181 AT |
2485 | } |
2486 | ||
2487 | ex->ee_block = cpu_to_le32(block); | |
2488 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
2489 | /* |
2490 | * Do not mark uninitialized if all the blocks in the | |
2491 | * extent have been removed. | |
2492 | */ | |
2493 | if (uninitialized && num) | |
a2df2a63 | 2494 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
2495 | |
2496 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2497 | if (err) | |
2498 | goto out; | |
2499 | ||
d583fb87 AH |
2500 | /* |
2501 | * If the extent was completely released, | |
2502 | * we need to remove it from the leaf | |
2503 | */ | |
2504 | if (num == 0) { | |
f17722f9 | 2505 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2506 | /* |
2507 | * For hole punching, we need to scoot all the | |
2508 | * extents up when an extent is removed so that | |
2509 | * we dont have blank extents in the middle | |
2510 | */ | |
2511 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2512 | sizeof(struct ext4_extent)); | |
2513 | ||
2514 | /* Now get rid of the one at the end */ | |
2515 | memset(EXT_LAST_EXTENT(eh), 0, | |
2516 | sizeof(struct ext4_extent)); | |
2517 | } | |
2518 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2519 | } else |
2520 | *partial_cluster = 0; | |
d583fb87 | 2521 | |
2ae02107 | 2522 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
bf89d16f | 2523 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2524 | ex--; |
2525 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2526 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2527 | } |
2528 | ||
2529 | if (correct_index && eh->eh_entries) | |
2530 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2531 | ||
0aa06000 TT |
2532 | /* |
2533 | * If there is still a entry in the leaf node, check to see if | |
2534 | * it references the partial cluster. This is the only place | |
2535 | * where it could; if it doesn't, we can free the cluster. | |
2536 | */ | |
2537 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2538 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2539 | *partial_cluster)) { | |
2540 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2541 | ||
2542 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2543 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2544 | ||
2545 | ext4_free_blocks(handle, inode, NULL, | |
2546 | EXT4_C2B(sbi, *partial_cluster), | |
2547 | sbi->s_cluster_ratio, flags); | |
2548 | *partial_cluster = 0; | |
2549 | } | |
2550 | ||
a86c6181 AT |
2551 | /* if this leaf is free, then we should |
2552 | * remove it from index block above */ | |
2553 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2554 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2555 | ||
2556 | out: | |
2557 | return err; | |
2558 | } | |
2559 | ||
2560 | /* | |
d0d856e8 RD |
2561 | * ext4_ext_more_to_rm: |
2562 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2563 | */ |
09b88252 | 2564 | static int |
a86c6181 AT |
2565 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2566 | { | |
2567 | BUG_ON(path->p_idx == NULL); | |
2568 | ||
2569 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2570 | return 0; | |
2571 | ||
2572 | /* | |
d0d856e8 | 2573 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2574 | * so we have to consider current index for truncation |
2575 | */ | |
2576 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2577 | return 0; | |
2578 | return 1; | |
2579 | } | |
2580 | ||
c6a0371c | 2581 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
2582 | { |
2583 | struct super_block *sb = inode->i_sb; | |
2584 | int depth = ext_depth(inode); | |
2585 | struct ext4_ext_path *path; | |
0aa06000 | 2586 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2587 | handle_t *handle; |
0617b83f | 2588 | int i, err; |
a86c6181 | 2589 | |
725d26d3 | 2590 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
2591 | |
2592 | /* probably first extent we're gonna free will be last in block */ | |
2593 | handle = ext4_journal_start(inode, depth + 1); | |
2594 | if (IS_ERR(handle)) | |
2595 | return PTR_ERR(handle); | |
2596 | ||
0617b83f | 2597 | again: |
a86c6181 AT |
2598 | ext4_ext_invalidate_cache(inode); |
2599 | ||
d8990240 AK |
2600 | trace_ext4_ext_remove_space(inode, start, depth); |
2601 | ||
a86c6181 | 2602 | /* |
d0d856e8 RD |
2603 | * We start scanning from right side, freeing all the blocks |
2604 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2605 | */ |
0617b83f | 2606 | depth = ext_depth(inode); |
216553c4 | 2607 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
2608 | if (path == NULL) { |
2609 | ext4_journal_stop(handle); | |
2610 | return -ENOMEM; | |
2611 | } | |
0617b83f | 2612 | path[0].p_depth = depth; |
a86c6181 | 2613 | path[0].p_hdr = ext_inode_hdr(inode); |
56b19868 | 2614 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
2615 | err = -EIO; |
2616 | goto out; | |
2617 | } | |
0617b83f | 2618 | i = err = 0; |
a86c6181 AT |
2619 | |
2620 | while (i >= 0 && err == 0) { | |
2621 | if (i == depth) { | |
2622 | /* this is leaf block */ | |
d583fb87 | 2623 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 TT |
2624 | &partial_cluster, start, |
2625 | EXT_MAX_BLOCKS - 1); | |
d0d856e8 | 2626 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2627 | brelse(path[i].p_bh); |
2628 | path[i].p_bh = NULL; | |
2629 | i--; | |
2630 | continue; | |
2631 | } | |
2632 | ||
2633 | /* this is index block */ | |
2634 | if (!path[i].p_hdr) { | |
2635 | ext_debug("initialize header\n"); | |
2636 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2637 | } |
2638 | ||
a86c6181 | 2639 | if (!path[i].p_idx) { |
d0d856e8 | 2640 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2641 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2642 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2643 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2644 | path[i].p_hdr, | |
2645 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2646 | } else { | |
d0d856e8 | 2647 | /* we were already here, see at next index */ |
a86c6181 AT |
2648 | path[i].p_idx--; |
2649 | } | |
2650 | ||
2651 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2652 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2653 | path[i].p_idx); | |
2654 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2655 | struct buffer_head *bh; |
a86c6181 | 2656 | /* go to the next level */ |
2ae02107 | 2657 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2658 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2659 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2660 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2661 | if (!bh) { |
a86c6181 AT |
2662 | /* should we reset i_size? */ |
2663 | err = -EIO; | |
2664 | break; | |
2665 | } | |
c29c0ae7 AT |
2666 | if (WARN_ON(i + 1 > depth)) { |
2667 | err = -EIO; | |
2668 | break; | |
2669 | } | |
56b19868 | 2670 | if (ext4_ext_check(inode, ext_block_hdr(bh), |
c29c0ae7 AT |
2671 | depth - i - 1)) { |
2672 | err = -EIO; | |
2673 | break; | |
2674 | } | |
2675 | path[i + 1].p_bh = bh; | |
a86c6181 | 2676 | |
d0d856e8 RD |
2677 | /* save actual number of indexes since this |
2678 | * number is changed at the next iteration */ | |
a86c6181 AT |
2679 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2680 | i++; | |
2681 | } else { | |
d0d856e8 | 2682 | /* we finished processing this index, go up */ |
a86c6181 | 2683 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2684 | /* index is empty, remove it; |
a86c6181 AT |
2685 | * handle must be already prepared by the |
2686 | * truncatei_leaf() */ | |
2687 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2688 | } | |
d0d856e8 | 2689 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2690 | brelse(path[i].p_bh); |
2691 | path[i].p_bh = NULL; | |
2692 | i--; | |
2693 | ext_debug("return to level %d\n", i); | |
2694 | } | |
2695 | } | |
2696 | ||
d8990240 AK |
2697 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2698 | path->p_hdr->eh_entries); | |
2699 | ||
7b415bf6 AK |
2700 | /* If we still have something in the partial cluster and we have removed |
2701 | * even the first extent, then we should free the blocks in the partial | |
2702 | * cluster as well. */ | |
2703 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2704 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2705 | ||
2706 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2707 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2708 | ||
2709 | ext4_free_blocks(handle, inode, NULL, | |
2710 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2711 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2712 | partial_cluster = 0; | |
2713 | } | |
2714 | ||
a86c6181 AT |
2715 | /* TODO: flexible tree reduction should be here */ |
2716 | if (path->p_hdr->eh_entries == 0) { | |
2717 | /* | |
d0d856e8 RD |
2718 | * truncate to zero freed all the tree, |
2719 | * so we need to correct eh_depth | |
a86c6181 AT |
2720 | */ |
2721 | err = ext4_ext_get_access(handle, inode, path); | |
2722 | if (err == 0) { | |
2723 | ext_inode_hdr(inode)->eh_depth = 0; | |
2724 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2725 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2726 | err = ext4_ext_dirty(handle, inode, path); |
2727 | } | |
2728 | } | |
2729 | out: | |
a86c6181 AT |
2730 | ext4_ext_drop_refs(path); |
2731 | kfree(path); | |
0617b83f DM |
2732 | if (err == -EAGAIN) |
2733 | goto again; | |
a86c6181 AT |
2734 | ext4_journal_stop(handle); |
2735 | ||
2736 | return err; | |
2737 | } | |
2738 | ||
2739 | /* | |
2740 | * called at mount time | |
2741 | */ | |
2742 | void ext4_ext_init(struct super_block *sb) | |
2743 | { | |
2744 | /* | |
2745 | * possible initialization would be here | |
2746 | */ | |
2747 | ||
83982b6f | 2748 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2749 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
4776004f | 2750 | printk(KERN_INFO "EXT4-fs: file extents enabled"); |
bbf2f9fb RD |
2751 | #ifdef AGGRESSIVE_TEST |
2752 | printk(", aggressive tests"); | |
a86c6181 AT |
2753 | #endif |
2754 | #ifdef CHECK_BINSEARCH | |
2755 | printk(", check binsearch"); | |
2756 | #endif | |
2757 | #ifdef EXTENTS_STATS | |
2758 | printk(", stats"); | |
2759 | #endif | |
2760 | printk("\n"); | |
90576c0b | 2761 | #endif |
a86c6181 AT |
2762 | #ifdef EXTENTS_STATS |
2763 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2764 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2765 | EXT4_SB(sb)->s_ext_max = 0; | |
2766 | #endif | |
2767 | } | |
2768 | } | |
2769 | ||
2770 | /* | |
2771 | * called at umount time | |
2772 | */ | |
2773 | void ext4_ext_release(struct super_block *sb) | |
2774 | { | |
83982b6f | 2775 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2776 | return; |
2777 | ||
2778 | #ifdef EXTENTS_STATS | |
2779 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2780 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2781 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2782 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2783 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2784 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2785 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2786 | } | |
2787 | #endif | |
2788 | } | |
2789 | ||
093a088b AK |
2790 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2791 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2792 | { | |
2407518d LC |
2793 | ext4_fsblk_t ee_pblock; |
2794 | unsigned int ee_len; | |
b720303d | 2795 | int ret; |
093a088b | 2796 | |
093a088b | 2797 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2798 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2799 | |
a107e5a3 | 2800 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2801 | if (ret > 0) |
2802 | ret = 0; | |
093a088b | 2803 | |
2407518d | 2804 | return ret; |
093a088b AK |
2805 | } |
2806 | ||
47ea3bb5 YY |
2807 | /* |
2808 | * used by extent splitting. | |
2809 | */ | |
2810 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
2811 | due to ENOSPC */ | |
2812 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
2813 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
2814 | ||
2815 | /* | |
2816 | * ext4_split_extent_at() splits an extent at given block. | |
2817 | * | |
2818 | * @handle: the journal handle | |
2819 | * @inode: the file inode | |
2820 | * @path: the path to the extent | |
2821 | * @split: the logical block where the extent is splitted. | |
2822 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2823 | * the states(init or uninit) of new extents. | |
2824 | * @flags: flags used to insert new extent to extent tree. | |
2825 | * | |
2826 | * | |
2827 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2828 | * of which are deterimined by split_flag. | |
2829 | * | |
2830 | * There are two cases: | |
2831 | * a> the extent are splitted into two extent. | |
2832 | * b> split is not needed, and just mark the extent. | |
2833 | * | |
2834 | * return 0 on success. | |
2835 | */ | |
2836 | static int ext4_split_extent_at(handle_t *handle, | |
2837 | struct inode *inode, | |
2838 | struct ext4_ext_path *path, | |
2839 | ext4_lblk_t split, | |
2840 | int split_flag, | |
2841 | int flags) | |
2842 | { | |
2843 | ext4_fsblk_t newblock; | |
2844 | ext4_lblk_t ee_block; | |
2845 | struct ext4_extent *ex, newex, orig_ex; | |
2846 | struct ext4_extent *ex2 = NULL; | |
2847 | unsigned int ee_len, depth; | |
2848 | int err = 0; | |
2849 | ||
2850 | ext_debug("ext4_split_extents_at: inode %lu, logical" | |
2851 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2852 | ||
2853 | ext4_ext_show_leaf(inode, path); | |
2854 | ||
2855 | depth = ext_depth(inode); | |
2856 | ex = path[depth].p_ext; | |
2857 | ee_block = le32_to_cpu(ex->ee_block); | |
2858 | ee_len = ext4_ext_get_actual_len(ex); | |
2859 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2860 | ||
2861 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
2862 | ||
2863 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2864 | if (err) | |
2865 | goto out; | |
2866 | ||
2867 | if (split == ee_block) { | |
2868 | /* | |
2869 | * case b: block @split is the block that the extent begins with | |
2870 | * then we just change the state of the extent, and splitting | |
2871 | * is not needed. | |
2872 | */ | |
2873 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2874 | ext4_ext_mark_uninitialized(ex); | |
2875 | else | |
2876 | ext4_ext_mark_initialized(ex); | |
2877 | ||
2878 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
2879 | ext4_ext_try_to_merge(inode, path, ex); | |
2880 | ||
2881 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2882 | goto out; | |
2883 | } | |
2884 | ||
2885 | /* case a */ | |
2886 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
2887 | ex->ee_len = cpu_to_le16(split - ee_block); | |
2888 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
2889 | ext4_ext_mark_uninitialized(ex); | |
2890 | ||
2891 | /* | |
2892 | * path may lead to new leaf, not to original leaf any more | |
2893 | * after ext4_ext_insert_extent() returns, | |
2894 | */ | |
2895 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2896 | if (err) | |
2897 | goto fix_extent_len; | |
2898 | ||
2899 | ex2 = &newex; | |
2900 | ex2->ee_block = cpu_to_le32(split); | |
2901 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
2902 | ext4_ext_store_pblock(ex2, newblock); | |
2903 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2904 | ext4_ext_mark_uninitialized(ex2); | |
2905 | ||
2906 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
2907 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2908 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2909 | if (err) | |
2910 | goto fix_extent_len; | |
2911 | /* update the extent length and mark as initialized */ | |
2912 | ex->ee_len = cpu_to_le32(ee_len); | |
2913 | ext4_ext_try_to_merge(inode, path, ex); | |
2914 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2915 | goto out; | |
2916 | } else if (err) | |
2917 | goto fix_extent_len; | |
2918 | ||
2919 | out: | |
2920 | ext4_ext_show_leaf(inode, path); | |
2921 | return err; | |
2922 | ||
2923 | fix_extent_len: | |
2924 | ex->ee_len = orig_ex.ee_len; | |
2925 | ext4_ext_dirty(handle, inode, path + depth); | |
2926 | return err; | |
2927 | } | |
2928 | ||
2929 | /* | |
2930 | * ext4_split_extents() splits an extent and mark extent which is covered | |
2931 | * by @map as split_flags indicates | |
2932 | * | |
2933 | * It may result in splitting the extent into multiple extents (upto three) | |
2934 | * There are three possibilities: | |
2935 | * a> There is no split required | |
2936 | * b> Splits in two extents: Split is happening at either end of the extent | |
2937 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
2938 | * | |
2939 | */ | |
2940 | static int ext4_split_extent(handle_t *handle, | |
2941 | struct inode *inode, | |
2942 | struct ext4_ext_path *path, | |
2943 | struct ext4_map_blocks *map, | |
2944 | int split_flag, | |
2945 | int flags) | |
2946 | { | |
2947 | ext4_lblk_t ee_block; | |
2948 | struct ext4_extent *ex; | |
2949 | unsigned int ee_len, depth; | |
2950 | int err = 0; | |
2951 | int uninitialized; | |
2952 | int split_flag1, flags1; | |
2953 | ||
2954 | depth = ext_depth(inode); | |
2955 | ex = path[depth].p_ext; | |
2956 | ee_block = le32_to_cpu(ex->ee_block); | |
2957 | ee_len = ext4_ext_get_actual_len(ex); | |
2958 | uninitialized = ext4_ext_is_uninitialized(ex); | |
2959 | ||
2960 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
2961 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
2962 | EXT4_EXT_MAY_ZEROOUT : 0; | |
2963 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; | |
2964 | if (uninitialized) | |
2965 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
2966 | EXT4_EXT_MARK_UNINIT2; | |
2967 | err = ext4_split_extent_at(handle, inode, path, | |
2968 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
2969 | if (err) |
2970 | goto out; | |
47ea3bb5 YY |
2971 | } |
2972 | ||
2973 | ext4_ext_drop_refs(path); | |
2974 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
2975 | if (IS_ERR(path)) | |
2976 | return PTR_ERR(path); | |
2977 | ||
2978 | if (map->m_lblk >= ee_block) { | |
2979 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
2980 | EXT4_EXT_MAY_ZEROOUT : 0; | |
2981 | if (uninitialized) | |
2982 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
2983 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2984 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
2985 | err = ext4_split_extent_at(handle, inode, path, | |
2986 | map->m_lblk, split_flag1, flags); | |
2987 | if (err) | |
2988 | goto out; | |
2989 | } | |
2990 | ||
2991 | ext4_ext_show_leaf(inode, path); | |
2992 | out: | |
2993 | return err ? err : map->m_len; | |
2994 | } | |
2995 | ||
3977c965 | 2996 | #define EXT4_EXT_ZERO_LEN 7 |
56055d3a | 2997 | /* |
e35fd660 | 2998 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 2999 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3000 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3001 | * uninitialized). |
3002 | * There are three possibilities: | |
3003 | * a> There is no split required: Entire extent should be initialized | |
3004 | * b> Splits in two extents: Write is happening at either end of the extent | |
3005 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3006 | */ | |
725d26d3 | 3007 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3008 | struct inode *inode, |
3009 | struct ext4_map_blocks *map, | |
3010 | struct ext4_ext_path *path) | |
56055d3a | 3011 | { |
667eff35 YY |
3012 | struct ext4_map_blocks split_map; |
3013 | struct ext4_extent zero_ex; | |
3014 | struct ext4_extent *ex; | |
21ca087a | 3015 | ext4_lblk_t ee_block, eof_block; |
725d26d3 | 3016 | unsigned int allocated, ee_len, depth; |
56055d3a | 3017 | int err = 0; |
667eff35 | 3018 | int split_flag = 0; |
21ca087a DM |
3019 | |
3020 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3021 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3022 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3023 | |
3024 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3025 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3026 | if (eof_block < map->m_lblk + map->m_len) |
3027 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3028 | |
3029 | depth = ext_depth(inode); | |
56055d3a AA |
3030 | ex = path[depth].p_ext; |
3031 | ee_block = le32_to_cpu(ex->ee_block); | |
3032 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3033 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 3034 | |
667eff35 | 3035 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3036 | /* |
3037 | * It is safe to convert extent to initialized via explicit | |
3038 | * zeroout only if extent is fully insde i_size or new_size. | |
3039 | */ | |
667eff35 | 3040 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3041 | |
3977c965 | 3042 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
667eff35 YY |
3043 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN && |
3044 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3045 | err = ext4_ext_zeroout(inode, ex); | |
3977c965 | 3046 | if (err) |
d03856bd | 3047 | goto out; |
d03856bd AK |
3048 | |
3049 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3050 | if (err) | |
3051 | goto out; | |
667eff35 YY |
3052 | ext4_ext_mark_initialized(ex); |
3053 | ext4_ext_try_to_merge(inode, path, ex); | |
3054 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3055 | goto out; | |
56055d3a | 3056 | } |
667eff35 | 3057 | |
56055d3a | 3058 | /* |
667eff35 YY |
3059 | * four cases: |
3060 | * 1. split the extent into three extents. | |
3061 | * 2. split the extent into two extents, zeroout the first half. | |
3062 | * 3. split the extent into two extents, zeroout the second half. | |
3063 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3064 | */ |
667eff35 YY |
3065 | split_map.m_lblk = map->m_lblk; |
3066 | split_map.m_len = map->m_len; | |
3067 | ||
3068 | if (allocated > map->m_len) { | |
3069 | if (allocated <= EXT4_EXT_ZERO_LEN && | |
3070 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3071 | /* case 3 */ | |
3072 | zero_ex.ee_block = | |
9b940f8e AH |
3073 | cpu_to_le32(map->m_lblk); |
3074 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3075 | ext4_ext_store_pblock(&zero_ex, |
3076 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3077 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3078 | if (err) |
3079 | goto out; | |
667eff35 YY |
3080 | split_map.m_lblk = map->m_lblk; |
3081 | split_map.m_len = allocated; | |
3082 | } else if ((map->m_lblk - ee_block + map->m_len < | |
3083 | EXT4_EXT_ZERO_LEN) && | |
3084 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3085 | /* case 2 */ | |
3086 | if (map->m_lblk != ee_block) { | |
3087 | zero_ex.ee_block = ex->ee_block; | |
3088 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3089 | ee_block); | |
3090 | ext4_ext_store_pblock(&zero_ex, | |
3091 | ext4_ext_pblock(ex)); | |
3092 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3093 | if (err) | |
3094 | goto out; | |
3095 | } | |
3096 | ||
667eff35 | 3097 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3098 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3099 | allocated = map->m_len; | |
56055d3a AA |
3100 | } |
3101 | } | |
667eff35 YY |
3102 | |
3103 | allocated = ext4_split_extent(handle, inode, path, | |
3104 | &split_map, split_flag, 0); | |
3105 | if (allocated < 0) | |
3106 | err = allocated; | |
3107 | ||
56055d3a AA |
3108 | out: |
3109 | return err ? err : allocated; | |
3110 | } | |
3111 | ||
0031462b | 3112 | /* |
e35fd660 | 3113 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3114 | * ext4_get_blocks_dio_write() when DIO to write |
3115 | * to an uninitialized extent. | |
3116 | * | |
fd018fe8 | 3117 | * Writing to an uninitialized extent may result in splitting the uninitialized |
b595076a | 3118 | * extent into multiple /initialized uninitialized extents (up to three) |
0031462b MC |
3119 | * There are three possibilities: |
3120 | * a> There is no split required: Entire extent should be uninitialized | |
3121 | * b> Splits in two extents: Write is happening at either end of the extent | |
3122 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3123 | * | |
3124 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3125 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3126 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3127 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3128 | * into three uninitialized extent(at most). After IO complete, the part |
3129 | * being filled will be convert to initialized by the end_io callback function | |
3130 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3131 | * |
3132 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3133 | */ |
3134 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3135 | struct inode *inode, | |
e35fd660 | 3136 | struct ext4_map_blocks *map, |
0031462b | 3137 | struct ext4_ext_path *path, |
0031462b MC |
3138 | int flags) |
3139 | { | |
667eff35 YY |
3140 | ext4_lblk_t eof_block; |
3141 | ext4_lblk_t ee_block; | |
3142 | struct ext4_extent *ex; | |
3143 | unsigned int ee_len; | |
3144 | int split_flag = 0, depth; | |
21ca087a DM |
3145 | |
3146 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3147 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3148 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3149 | |
3150 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3151 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3152 | if (eof_block < map->m_lblk + map->m_len) |
3153 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3154 | /* |
3155 | * It is safe to convert extent to initialized via explicit | |
3156 | * zeroout only if extent is fully insde i_size or new_size. | |
3157 | */ | |
667eff35 YY |
3158 | depth = ext_depth(inode); |
3159 | ex = path[depth].p_ext; | |
3160 | ee_block = le32_to_cpu(ex->ee_block); | |
3161 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3162 | |
667eff35 YY |
3163 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3164 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
0031462b | 3165 | |
667eff35 YY |
3166 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3167 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3168 | } |
197217a5 | 3169 | |
c7064ef1 | 3170 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
0031462b MC |
3171 | struct inode *inode, |
3172 | struct ext4_ext_path *path) | |
3173 | { | |
3174 | struct ext4_extent *ex; | |
0031462b MC |
3175 | int depth; |
3176 | int err = 0; | |
0031462b MC |
3177 | |
3178 | depth = ext_depth(inode); | |
0031462b MC |
3179 | ex = path[depth].p_ext; |
3180 | ||
197217a5 YY |
3181 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3182 | "block %llu, max_blocks %u\n", inode->i_ino, | |
3183 | (unsigned long long)le32_to_cpu(ex->ee_block), | |
3184 | ext4_ext_get_actual_len(ex)); | |
3185 | ||
0031462b MC |
3186 | err = ext4_ext_get_access(handle, inode, path + depth); |
3187 | if (err) | |
3188 | goto out; | |
3189 | /* first mark the extent as initialized */ | |
3190 | ext4_ext_mark_initialized(ex); | |
3191 | ||
197217a5 YY |
3192 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3193 | * borders are not changed | |
0031462b | 3194 | */ |
197217a5 YY |
3195 | ext4_ext_try_to_merge(inode, path, ex); |
3196 | ||
0031462b MC |
3197 | /* Mark modified extent as dirty */ |
3198 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3199 | out: | |
3200 | ext4_ext_show_leaf(inode, path); | |
3201 | return err; | |
3202 | } | |
3203 | ||
515f41c3 AK |
3204 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3205 | sector_t block, int count) | |
3206 | { | |
3207 | int i; | |
3208 | for (i = 0; i < count; i++) | |
3209 | unmap_underlying_metadata(bdev, block + i); | |
3210 | } | |
3211 | ||
58590b06 TT |
3212 | /* |
3213 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3214 | */ | |
3215 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3216 | ext4_lblk_t lblk, |
58590b06 TT |
3217 | struct ext4_ext_path *path, |
3218 | unsigned int len) | |
3219 | { | |
3220 | int i, depth; | |
3221 | struct ext4_extent_header *eh; | |
65922cb5 | 3222 | struct ext4_extent *last_ex; |
58590b06 TT |
3223 | |
3224 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3225 | return 0; | |
3226 | ||
3227 | depth = ext_depth(inode); | |
3228 | eh = path[depth].p_hdr; | |
58590b06 TT |
3229 | |
3230 | if (unlikely(!eh->eh_entries)) { | |
3231 | EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and " | |
3232 | "EOFBLOCKS_FL set"); | |
3233 | return -EIO; | |
3234 | } | |
3235 | last_ex = EXT_LAST_EXTENT(eh); | |
3236 | /* | |
3237 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3238 | * last block in the last extent in the file. We test this by | |
3239 | * first checking to see if the caller to | |
3240 | * ext4_ext_get_blocks() was interested in the last block (or | |
3241 | * a block beyond the last block) in the current extent. If | |
3242 | * this turns out to be false, we can bail out from this | |
3243 | * function immediately. | |
3244 | */ | |
d002ebf1 | 3245 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3246 | ext4_ext_get_actual_len(last_ex)) |
3247 | return 0; | |
3248 | /* | |
3249 | * If the caller does appear to be planning to write at or | |
3250 | * beyond the end of the current extent, we then test to see | |
3251 | * if the current extent is the last extent in the file, by | |
3252 | * checking to make sure it was reached via the rightmost node | |
3253 | * at each level of the tree. | |
3254 | */ | |
3255 | for (i = depth-1; i >= 0; i--) | |
3256 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3257 | return 0; | |
3258 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); | |
3259 | return ext4_mark_inode_dirty(handle, inode); | |
3260 | } | |
3261 | ||
7b415bf6 AK |
3262 | /** |
3263 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3264 | * | |
3265 | * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns | |
3266 | * whether there are any buffers marked for delayed allocation. It returns '1' | |
3267 | * on the first delalloc'ed buffer head found. If no buffer head in the given | |
3268 | * range is marked for delalloc, it returns 0. | |
3269 | * lblk_start should always be <= lblk_end. | |
3270 | * search_hint_reverse is to indicate that searching in reverse from lblk_end to | |
3271 | * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed | |
3272 | * block sooner). This is useful when blocks are truncated sequentially from | |
3273 | * lblk_start towards lblk_end. | |
3274 | */ | |
3275 | static int ext4_find_delalloc_range(struct inode *inode, | |
3276 | ext4_lblk_t lblk_start, | |
3277 | ext4_lblk_t lblk_end, | |
3278 | int search_hint_reverse) | |
3279 | { | |
3280 | struct address_space *mapping = inode->i_mapping; | |
3281 | struct buffer_head *head, *bh = NULL; | |
3282 | struct page *page; | |
3283 | ext4_lblk_t i, pg_lblk; | |
3284 | pgoff_t index; | |
3285 | ||
3286 | /* reverse search wont work if fs block size is less than page size */ | |
3287 | if (inode->i_blkbits < PAGE_CACHE_SHIFT) | |
3288 | search_hint_reverse = 0; | |
3289 | ||
3290 | if (search_hint_reverse) | |
3291 | i = lblk_end; | |
3292 | else | |
3293 | i = lblk_start; | |
3294 | ||
3295 | index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
3296 | ||
3297 | while ((i >= lblk_start) && (i <= lblk_end)) { | |
3298 | page = find_get_page(mapping, index); | |
3299 | if (!page || !PageDirty(page)) | |
3300 | goto nextpage; | |
3301 | ||
3302 | if (PageWriteback(page)) { | |
3303 | /* | |
3304 | * This might be a race with allocation and writeout. In | |
3305 | * this case we just assume that the rest of the range | |
3306 | * will eventually be written and there wont be any | |
3307 | * delalloc blocks left. | |
3308 | * TODO: the above assumption is troublesome, but might | |
3309 | * work better in practice. other option could be note | |
3310 | * somewhere that the cluster is getting written out and | |
3311 | * detect that here. | |
3312 | */ | |
3313 | page_cache_release(page); | |
d8990240 AK |
3314 | trace_ext4_find_delalloc_range(inode, |
3315 | lblk_start, lblk_end, | |
3316 | search_hint_reverse, | |
3317 | 0, i); | |
7b415bf6 AK |
3318 | return 0; |
3319 | } | |
3320 | ||
3321 | if (!page_has_buffers(page)) | |
3322 | goto nextpage; | |
3323 | ||
3324 | head = page_buffers(page); | |
3325 | if (!head) | |
3326 | goto nextpage; | |
3327 | ||
3328 | bh = head; | |
3329 | pg_lblk = index << (PAGE_CACHE_SHIFT - | |
3330 | inode->i_blkbits); | |
3331 | do { | |
3332 | if (unlikely(pg_lblk < lblk_start)) { | |
3333 | /* | |
3334 | * This is possible when fs block size is less | |
3335 | * than page size and our cluster starts/ends in | |
3336 | * middle of the page. So we need to skip the | |
3337 | * initial few blocks till we reach the 'lblk' | |
3338 | */ | |
3339 | pg_lblk++; | |
3340 | continue; | |
3341 | } | |
3342 | ||
3343 | if (buffer_delay(bh)) { | |
3344 | page_cache_release(page); | |
d8990240 AK |
3345 | trace_ext4_find_delalloc_range(inode, |
3346 | lblk_start, lblk_end, | |
3347 | search_hint_reverse, | |
3348 | 1, i); | |
7b415bf6 AK |
3349 | return 1; |
3350 | } | |
3351 | if (search_hint_reverse) | |
3352 | i--; | |
3353 | else | |
3354 | i++; | |
3355 | } while ((i >= lblk_start) && (i <= lblk_end) && | |
3356 | ((bh = bh->b_this_page) != head)); | |
3357 | nextpage: | |
3358 | if (page) | |
3359 | page_cache_release(page); | |
3360 | /* | |
3361 | * Move to next page. 'i' will be the first lblk in the next | |
3362 | * page. | |
3363 | */ | |
3364 | if (search_hint_reverse) | |
3365 | index--; | |
3366 | else | |
3367 | index++; | |
3368 | i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
3369 | } | |
3370 | ||
d8990240 AK |
3371 | trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end, |
3372 | search_hint_reverse, 0, 0); | |
7b415bf6 AK |
3373 | return 0; |
3374 | } | |
3375 | ||
3376 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk, | |
3377 | int search_hint_reverse) | |
3378 | { | |
3379 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3380 | ext4_lblk_t lblk_start, lblk_end; | |
3381 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3382 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3383 | ||
3384 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end, | |
3385 | search_hint_reverse); | |
3386 | } | |
3387 | ||
3388 | /** | |
3389 | * Determines how many complete clusters (out of those specified by the 'map') | |
3390 | * are under delalloc and were reserved quota for. | |
3391 | * This function is called when we are writing out the blocks that were | |
3392 | * originally written with their allocation delayed, but then the space was | |
3393 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3394 | * The cases to look for are: | |
3395 | * ('=' indicated delayed allocated blocks | |
3396 | * '-' indicates non-delayed allocated blocks) | |
3397 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3398 | * are not delalloc'ed. | |
3399 | * Ex: | |
3400 | * |----c---=|====c====|====c====|===-c----| | |
3401 | * |++++++ allocated ++++++| | |
3402 | * ==> 4 complete clusters in above example | |
3403 | * | |
3404 | * (b) partial cluster (outside of allocated range) towards either end is | |
3405 | * marked for delayed allocation. In this case, we will exclude that | |
3406 | * cluster. | |
3407 | * Ex: | |
3408 | * |----====c========|========c========| | |
3409 | * |++++++ allocated ++++++| | |
3410 | * ==> 1 complete clusters in above example | |
3411 | * | |
3412 | * Ex: | |
3413 | * |================c================| | |
3414 | * |++++++ allocated ++++++| | |
3415 | * ==> 0 complete clusters in above example | |
3416 | * | |
3417 | * The ext4_da_update_reserve_space will be called only if we | |
3418 | * determine here that there were some "entire" clusters that span | |
3419 | * this 'allocated' range. | |
3420 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3421 | * without ever calling ext4_find_delalloc_range. | |
3422 | */ | |
3423 | static unsigned int | |
3424 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3425 | unsigned int num_blks) | |
3426 | { | |
3427 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3428 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3429 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3430 | unsigned int allocated_clusters = 0; | |
3431 | ||
3432 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3433 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3434 | ||
3435 | /* max possible clusters for this allocation */ | |
3436 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3437 | ||
d8990240 AK |
3438 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3439 | ||
7b415bf6 AK |
3440 | /* Check towards left side */ |
3441 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3442 | if (c_offset) { | |
3443 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3444 | lblk_to = lblk_from + c_offset - 1; | |
3445 | ||
3446 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0)) | |
3447 | allocated_clusters--; | |
3448 | } | |
3449 | ||
3450 | /* Now check towards right. */ | |
3451 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3452 | if (allocated_clusters && c_offset) { | |
3453 | lblk_from = lblk_start + num_blks; | |
3454 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3455 | ||
3456 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0)) | |
3457 | allocated_clusters--; | |
3458 | } | |
3459 | ||
3460 | return allocated_clusters; | |
3461 | } | |
3462 | ||
0031462b MC |
3463 | static int |
3464 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3465 | struct ext4_map_blocks *map, |
0031462b | 3466 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3467 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3468 | { |
3469 | int ret = 0; | |
3470 | int err = 0; | |
8d5d02e6 | 3471 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
0031462b MC |
3472 | |
3473 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical" | |
3474 | "block %llu, max_blocks %u, flags %d, allocated %u", | |
e35fd660 | 3475 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3476 | flags, allocated); |
3477 | ext4_ext_show_leaf(inode, path); | |
3478 | ||
d8990240 AK |
3479 | trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated, |
3480 | newblock); | |
3481 | ||
c7064ef1 | 3482 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3483 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3484 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3485 | path, flags); | |
5f524950 M |
3486 | /* |
3487 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3488 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3489 | * completed |
3490 | */ | |
e9e3bcec | 3491 | if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) { |
bd2d0210 | 3492 | io->flag = EXT4_IO_END_UNWRITTEN; |
e9e3bcec ES |
3493 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); |
3494 | } else | |
19f5fb7a | 3495 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
744692dc | 3496 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3497 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3498 | goto out; |
3499 | } | |
c7064ef1 | 3500 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3501 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
c7064ef1 | 3502 | ret = ext4_convert_unwritten_extents_endio(handle, inode, |
0031462b | 3503 | path); |
58590b06 | 3504 | if (ret >= 0) { |
b436b9be | 3505 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3506 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3507 | path, map->m_len); | |
58590b06 TT |
3508 | } else |
3509 | err = ret; | |
0031462b MC |
3510 | goto out2; |
3511 | } | |
3512 | /* buffered IO case */ | |
3513 | /* | |
3514 | * repeat fallocate creation request | |
3515 | * we already have an unwritten extent | |
3516 | */ | |
3517 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | |
3518 | goto map_out; | |
3519 | ||
3520 | /* buffered READ or buffered write_begin() lookup */ | |
3521 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3522 | /* | |
3523 | * We have blocks reserved already. We | |
3524 | * return allocated blocks so that delalloc | |
3525 | * won't do block reservation for us. But | |
3526 | * the buffer head will be unmapped so that | |
3527 | * a read from the block returns 0s. | |
3528 | */ | |
e35fd660 | 3529 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3530 | goto out1; |
3531 | } | |
3532 | ||
3533 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3534 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
58590b06 | 3535 | if (ret >= 0) { |
b436b9be | 3536 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3537 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, |
3538 | map->m_len); | |
58590b06 TT |
3539 | if (err < 0) |
3540 | goto out2; | |
3541 | } | |
3542 | ||
0031462b MC |
3543 | out: |
3544 | if (ret <= 0) { | |
3545 | err = ret; | |
3546 | goto out2; | |
3547 | } else | |
3548 | allocated = ret; | |
e35fd660 | 3549 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3550 | /* |
3551 | * if we allocated more blocks than requested | |
3552 | * we need to make sure we unmap the extra block | |
3553 | * allocated. The actual needed block will get | |
3554 | * unmapped later when we find the buffer_head marked | |
3555 | * new. | |
3556 | */ | |
e35fd660 | 3557 | if (allocated > map->m_len) { |
515f41c3 | 3558 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3559 | newblock + map->m_len, |
3560 | allocated - map->m_len); | |
3561 | allocated = map->m_len; | |
515f41c3 | 3562 | } |
5f634d06 AK |
3563 | |
3564 | /* | |
3565 | * If we have done fallocate with the offset that is already | |
3566 | * delayed allocated, we would have block reservation | |
3567 | * and quota reservation done in the delayed write path. | |
3568 | * But fallocate would have already updated quota and block | |
3569 | * count for this offset. So cancel these reservation | |
3570 | */ | |
7b415bf6 AK |
3571 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3572 | unsigned int reserved_clusters; | |
3573 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3574 | map->m_lblk, map->m_len); | |
3575 | if (reserved_clusters) | |
3576 | ext4_da_update_reserve_space(inode, | |
3577 | reserved_clusters, | |
3578 | 0); | |
3579 | } | |
5f634d06 | 3580 | |
0031462b | 3581 | map_out: |
e35fd660 | 3582 | map->m_flags |= EXT4_MAP_MAPPED; |
0031462b | 3583 | out1: |
e35fd660 TT |
3584 | if (allocated > map->m_len) |
3585 | allocated = map->m_len; | |
0031462b | 3586 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3587 | map->m_pblk = newblock; |
3588 | map->m_len = allocated; | |
0031462b MC |
3589 | out2: |
3590 | if (path) { | |
3591 | ext4_ext_drop_refs(path); | |
3592 | kfree(path); | |
3593 | } | |
3594 | return err ? err : allocated; | |
3595 | } | |
58590b06 | 3596 | |
4d33b1ef TT |
3597 | /* |
3598 | * get_implied_cluster_alloc - check to see if the requested | |
3599 | * allocation (in the map structure) overlaps with a cluster already | |
3600 | * allocated in an extent. | |
d8990240 | 3601 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3602 | * @map The requested lblk->pblk mapping |
3603 | * @ex The extent structure which might contain an implied | |
3604 | * cluster allocation | |
3605 | * | |
3606 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3607 | * find blocks that were already in the inode's extent tree. Hence, | |
3608 | * we know that the beginning of the requested region cannot overlap | |
3609 | * the extent from the inode's extent tree. There are three cases we | |
3610 | * want to catch. The first is this case: | |
3611 | * | |
3612 | * |--- cluster # N--| | |
3613 | * |--- extent ---| |---- requested region ---| | |
3614 | * |==========| | |
3615 | * | |
3616 | * The second case that we need to test for is this one: | |
3617 | * | |
3618 | * |--------- cluster # N ----------------| | |
3619 | * |--- requested region --| |------- extent ----| | |
3620 | * |=======================| | |
3621 | * | |
3622 | * The third case is when the requested region lies between two extents | |
3623 | * within the same cluster: | |
3624 | * |------------- cluster # N-------------| | |
3625 | * |----- ex -----| |---- ex_right ----| | |
3626 | * |------ requested region ------| | |
3627 | * |================| | |
3628 | * | |
3629 | * In each of the above cases, we need to set the map->m_pblk and | |
3630 | * map->m_len so it corresponds to the return the extent labelled as | |
3631 | * "|====|" from cluster #N, since it is already in use for data in | |
3632 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3633 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3634 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3635 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3636 | * by calling ext4_mb_new_blocks(). | |
3637 | */ | |
d8990240 | 3638 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3639 | struct ext4_map_blocks *map, |
3640 | struct ext4_extent *ex, | |
3641 | struct ext4_ext_path *path) | |
3642 | { | |
d8990240 | 3643 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3644 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3645 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
3646 | ext4_lblk_t rr_cluster_start, rr_cluster_end; | |
3647 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); | |
3648 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3649 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3650 | ||
3651 | /* The extent passed in that we are trying to match */ | |
3652 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3653 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3654 | ||
3655 | /* The requested region passed into ext4_map_blocks() */ | |
3656 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
3657 | rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1); | |
3658 | ||
3659 | if ((rr_cluster_start == ex_cluster_end) || | |
3660 | (rr_cluster_start == ex_cluster_start)) { | |
3661 | if (rr_cluster_start == ex_cluster_end) | |
3662 | ee_start += ee_len - 1; | |
3663 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3664 | c_offset; | |
3665 | map->m_len = min(map->m_len, | |
3666 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3667 | /* | |
3668 | * Check for and handle this case: | |
3669 | * | |
3670 | * |--------- cluster # N-------------| | |
3671 | * |------- extent ----| | |
3672 | * |--- requested region ---| | |
3673 | * |===========| | |
3674 | */ | |
3675 | ||
3676 | if (map->m_lblk < ee_block) | |
3677 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3678 | ||
3679 | /* | |
3680 | * Check for the case where there is already another allocated | |
3681 | * block to the right of 'ex' but before the end of the cluster. | |
3682 | * | |
3683 | * |------------- cluster # N-------------| | |
3684 | * |----- ex -----| |---- ex_right ----| | |
3685 | * |------ requested region ------| | |
3686 | * |================| | |
3687 | */ | |
3688 | if (map->m_lblk > ee_block) { | |
3689 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3690 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3691 | } | |
d8990240 AK |
3692 | |
3693 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3694 | return 1; |
3695 | } | |
d8990240 AK |
3696 | |
3697 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3698 | return 0; |
3699 | } | |
3700 | ||
3701 | ||
c278bfec | 3702 | /* |
f5ab0d1f MC |
3703 | * Block allocation/map/preallocation routine for extents based files |
3704 | * | |
3705 | * | |
c278bfec | 3706 | * Need to be called with |
0e855ac8 AK |
3707 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3708 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3709 | * |
3710 | * return > 0, number of of blocks already mapped/allocated | |
3711 | * if create == 0 and these are pre-allocated blocks | |
3712 | * buffer head is unmapped | |
3713 | * otherwise blocks are mapped | |
3714 | * | |
3715 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3716 | * buffer head is unmapped | |
3717 | * | |
3718 | * return < 0, error case. | |
c278bfec | 3719 | */ |
e35fd660 TT |
3720 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3721 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3722 | { |
3723 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3724 | struct ext4_extent newex, *ex, *ex2; |
3725 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3726 | ext4_fsblk_t newblock = 0; |
4d33b1ef TT |
3727 | int free_on_err = 0, err = 0, depth, ret; |
3728 | unsigned int allocated = 0, offset = 0; | |
7b415bf6 | 3729 | unsigned int allocated_clusters = 0, reserved_clusters = 0; |
e861304b AH |
3730 | unsigned int punched_out = 0; |
3731 | unsigned int result = 0; | |
c9de560d | 3732 | struct ext4_allocation_request ar; |
8d5d02e6 | 3733 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
4d33b1ef | 3734 | ext4_lblk_t cluster_offset; |
e861304b | 3735 | struct ext4_map_blocks punch_map; |
a86c6181 | 3736 | |
84fe3bef | 3737 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3738 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3739 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3740 | |
3741 | /* check in cache */ | |
015861ba RD |
3742 | if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) && |
3743 | ext4_ext_in_cache(inode, map->m_lblk, &newex)) { | |
b05e6ae5 | 3744 | if (!newex.ee_start_lo && !newex.ee_start_hi) { |
7b415bf6 AK |
3745 | if ((sbi->s_cluster_ratio > 1) && |
3746 | ext4_find_delalloc_cluster(inode, map->m_lblk, 0)) | |
3747 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
3748 | ||
c2177057 | 3749 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3750 | /* |
3751 | * block isn't allocated yet and | |
3752 | * user doesn't want to allocate it | |
3753 | */ | |
a86c6181 AT |
3754 | goto out2; |
3755 | } | |
3756 | /* we should allocate requested block */ | |
b05e6ae5 | 3757 | } else { |
a86c6181 | 3758 | /* block is already allocated */ |
7b415bf6 AK |
3759 | if (sbi->s_cluster_ratio > 1) |
3760 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
e35fd660 | 3761 | newblock = map->m_lblk |
8c55e204 | 3762 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3763 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3764 | /* number of remaining blocks in the extent */ |
b939e376 | 3765 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3766 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3767 | goto out; |
a86c6181 AT |
3768 | } |
3769 | } | |
3770 | ||
3771 | /* find extent for this block */ | |
e35fd660 | 3772 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3773 | if (IS_ERR(path)) { |
3774 | err = PTR_ERR(path); | |
3775 | path = NULL; | |
3776 | goto out2; | |
3777 | } | |
3778 | ||
3779 | depth = ext_depth(inode); | |
3780 | ||
3781 | /* | |
d0d856e8 RD |
3782 | * consistent leaf must not be empty; |
3783 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3784 | * this is why assert can't be put in ext4_ext_find_extent() |
3785 | */ | |
273df556 FM |
3786 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3787 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3788 | "lblock: %lu, depth: %d pblock %lld", |
3789 | (unsigned long) map->m_lblk, depth, | |
3790 | path[depth].p_block); | |
034fb4c9 SP |
3791 | err = -EIO; |
3792 | goto out2; | |
3793 | } | |
a86c6181 | 3794 | |
7e028976 AM |
3795 | ex = path[depth].p_ext; |
3796 | if (ex) { | |
725d26d3 | 3797 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3798 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3799 | unsigned short ee_len; |
471d4011 SB |
3800 | |
3801 | /* | |
471d4011 | 3802 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3803 | * we split out initialized portions during a write. |
471d4011 | 3804 | */ |
a2df2a63 | 3805 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
3806 | |
3807 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
3808 | ||
d0d856e8 | 3809 | /* if found extent covers block, simply return it */ |
e35fd660 | 3810 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
0aa06000 TT |
3811 | ext4_fsblk_t partial_cluster = 0; |
3812 | ||
e35fd660 | 3813 | newblock = map->m_lblk - ee_block + ee_start; |
d0d856e8 | 3814 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3815 | allocated = ee_len - (map->m_lblk - ee_block); |
3816 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3817 | ee_block, ee_len, newblock); | |
56055d3a | 3818 | |
e861304b AH |
3819 | if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) { |
3820 | /* | |
3821 | * Do not put uninitialized extent | |
3822 | * in the cache | |
3823 | */ | |
3824 | if (!ext4_ext_is_uninitialized(ex)) { | |
3825 | ext4_ext_put_in_cache(inode, ee_block, | |
3826 | ee_len, ee_start); | |
3827 | goto out; | |
3828 | } | |
3829 | ret = ext4_ext_handle_uninitialized_extents( | |
3830 | handle, inode, map, path, flags, | |
3831 | allocated, newblock); | |
3832 | return ret; | |
56055d3a | 3833 | } |
e861304b AH |
3834 | |
3835 | /* | |
3836 | * Punch out the map length, but only to the | |
3837 | * end of the extent | |
3838 | */ | |
3839 | punched_out = allocated < map->m_len ? | |
3840 | allocated : map->m_len; | |
3841 | ||
3842 | /* | |
3843 | * Sense extents need to be converted to | |
3844 | * uninitialized, they must fit in an | |
3845 | * uninitialized extent | |
3846 | */ | |
3847 | if (punched_out > EXT_UNINIT_MAX_LEN) | |
3848 | punched_out = EXT_UNINIT_MAX_LEN; | |
3849 | ||
3850 | punch_map.m_lblk = map->m_lblk; | |
3851 | punch_map.m_pblk = newblock; | |
3852 | punch_map.m_len = punched_out; | |
3853 | punch_map.m_flags = 0; | |
3854 | ||
3855 | /* Check to see if the extent needs to be split */ | |
3856 | if (punch_map.m_len != ee_len || | |
3857 | punch_map.m_lblk != ee_block) { | |
3858 | ||
3859 | ret = ext4_split_extent(handle, inode, | |
3860 | path, &punch_map, 0, | |
3861 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT | | |
3862 | EXT4_GET_BLOCKS_PRE_IO); | |
3863 | ||
3864 | if (ret < 0) { | |
3865 | err = ret; | |
3866 | goto out2; | |
3867 | } | |
3868 | /* | |
3869 | * find extent for the block at | |
3870 | * the start of the hole | |
3871 | */ | |
3872 | ext4_ext_drop_refs(path); | |
3873 | kfree(path); | |
3874 | ||
3875 | path = ext4_ext_find_extent(inode, | |
3876 | map->m_lblk, NULL); | |
3877 | if (IS_ERR(path)) { | |
3878 | err = PTR_ERR(path); | |
3879 | path = NULL; | |
3880 | goto out2; | |
3881 | } | |
3882 | ||
3883 | depth = ext_depth(inode); | |
3884 | ex = path[depth].p_ext; | |
3885 | ee_len = ext4_ext_get_actual_len(ex); | |
3886 | ee_block = le32_to_cpu(ex->ee_block); | |
3887 | ee_start = ext4_ext_pblock(ex); | |
3888 | ||
3889 | } | |
3890 | ||
3891 | ext4_ext_mark_uninitialized(ex); | |
3892 | ||
f7d0d379 AH |
3893 | ext4_ext_invalidate_cache(inode); |
3894 | ||
3895 | err = ext4_ext_rm_leaf(handle, inode, path, | |
0aa06000 TT |
3896 | &partial_cluster, map->m_lblk, |
3897 | map->m_lblk + punched_out); | |
f7d0d379 AH |
3898 | |
3899 | if (!err && path->p_hdr->eh_entries == 0) { | |
3900 | /* | |
3901 | * Punch hole freed all of this sub tree, | |
3902 | * so we need to correct eh_depth | |
3903 | */ | |
3904 | err = ext4_ext_get_access(handle, inode, path); | |
3905 | if (err == 0) { | |
3906 | ext_inode_hdr(inode)->eh_depth = 0; | |
3907 | ext_inode_hdr(inode)->eh_max = | |
3908 | cpu_to_le16(ext4_ext_space_root( | |
3909 | inode, 0)); | |
3910 | ||
3911 | err = ext4_ext_dirty( | |
3912 | handle, inode, path); | |
3913 | } | |
3914 | } | |
e861304b AH |
3915 | |
3916 | goto out2; | |
a86c6181 AT |
3917 | } |
3918 | } | |
3919 | ||
7b415bf6 AK |
3920 | if ((sbi->s_cluster_ratio > 1) && |
3921 | ext4_find_delalloc_cluster(inode, map->m_lblk, 0)) | |
3922 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
3923 | ||
a86c6181 | 3924 | /* |
d0d856e8 | 3925 | * requested block isn't allocated yet; |
a86c6181 AT |
3926 | * we couldn't try to create block if create flag is zero |
3927 | */ | |
c2177057 | 3928 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3929 | /* |
3930 | * put just found gap into cache to speed up | |
3931 | * subsequent requests | |
3932 | */ | |
e35fd660 | 3933 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
3934 | goto out2; |
3935 | } | |
4d33b1ef | 3936 | |
a86c6181 | 3937 | /* |
c2ea3fde | 3938 | * Okay, we need to do block allocation. |
63f57933 | 3939 | */ |
7b415bf6 | 3940 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
3941 | newex.ee_block = cpu_to_le32(map->m_lblk); |
3942 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
3943 | ||
3944 | /* | |
3945 | * If we are doing bigalloc, check to see if the extent returned | |
3946 | * by ext4_ext_find_extent() implies a cluster we can use. | |
3947 | */ | |
3948 | if (cluster_offset && ex && | |
d8990240 | 3949 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
3950 | ar.len = allocated = map->m_len; |
3951 | newblock = map->m_pblk; | |
7b415bf6 | 3952 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
3953 | goto got_allocated_blocks; |
3954 | } | |
a86c6181 | 3955 | |
c9de560d | 3956 | /* find neighbour allocated blocks */ |
e35fd660 | 3957 | ar.lleft = map->m_lblk; |
c9de560d AT |
3958 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
3959 | if (err) | |
3960 | goto out2; | |
e35fd660 | 3961 | ar.lright = map->m_lblk; |
4d33b1ef TT |
3962 | ex2 = NULL; |
3963 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
3964 | if (err) |
3965 | goto out2; | |
25d14f98 | 3966 | |
4d33b1ef TT |
3967 | /* Check if the extent after searching to the right implies a |
3968 | * cluster we can use. */ | |
3969 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 3970 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
3971 | ar.len = allocated = map->m_len; |
3972 | newblock = map->m_pblk; | |
7b415bf6 | 3973 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
3974 | goto got_allocated_blocks; |
3975 | } | |
3976 | ||
749269fa AA |
3977 | /* |
3978 | * See if request is beyond maximum number of blocks we can have in | |
3979 | * a single extent. For an initialized extent this limit is | |
3980 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
3981 | * EXT_UNINIT_MAX_LEN. | |
3982 | */ | |
e35fd660 | 3983 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 3984 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
3985 | map->m_len = EXT_INIT_MAX_LEN; |
3986 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 3987 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 3988 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 3989 | |
e35fd660 | 3990 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 3991 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 3992 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 3993 | if (err) |
b939e376 | 3994 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 3995 | else |
e35fd660 | 3996 | allocated = map->m_len; |
c9de560d AT |
3997 | |
3998 | /* allocate new block */ | |
3999 | ar.inode = inode; | |
e35fd660 TT |
4000 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4001 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4002 | /* |
4003 | * We calculate the offset from the beginning of the cluster | |
4004 | * for the logical block number, since when we allocate a | |
4005 | * physical cluster, the physical block should start at the | |
4006 | * same offset from the beginning of the cluster. This is | |
4007 | * needed so that future calls to get_implied_cluster_alloc() | |
4008 | * work correctly. | |
4009 | */ | |
4010 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4011 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4012 | ar.goal -= offset; | |
4013 | ar.logical -= offset; | |
c9de560d AT |
4014 | if (S_ISREG(inode->i_mode)) |
4015 | ar.flags = EXT4_MB_HINT_DATA; | |
4016 | else | |
4017 | /* disable in-core preallocation for non-regular files */ | |
4018 | ar.flags = 0; | |
556b27ab VH |
4019 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4020 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4021 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4022 | if (!newblock) |
4023 | goto out2; | |
84fe3bef | 4024 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4025 | ar.goal, newblock, allocated); |
4d33b1ef | 4026 | free_on_err = 1; |
7b415bf6 | 4027 | allocated_clusters = ar.len; |
4d33b1ef TT |
4028 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4029 | if (ar.len > allocated) | |
4030 | ar.len = allocated; | |
a86c6181 | 4031 | |
4d33b1ef | 4032 | got_allocated_blocks: |
a86c6181 | 4033 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4034 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4035 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4036 | /* Mark uninitialized */ |
4037 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4038 | ext4_ext_mark_uninitialized(&newex); |
8d5d02e6 | 4039 | /* |
744692dc | 4040 | * io_end structure was created for every IO write to an |
25985edc | 4041 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4042 | * here we flag the IO that really needs the conversion. |
5f524950 | 4043 | * For non asycn direct IO case, flag the inode state |
25985edc | 4044 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4045 | */ |
744692dc | 4046 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e9e3bcec | 4047 | if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) { |
bd2d0210 | 4048 | io->flag = EXT4_IO_END_UNWRITTEN; |
e9e3bcec ES |
4049 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); |
4050 | } else | |
19f5fb7a TT |
4051 | ext4_set_inode_state(inode, |
4052 | EXT4_STATE_DIO_UNWRITTEN); | |
5f524950 | 4053 | } |
744692dc | 4054 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4055 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4056 | } |
c8d46e41 | 4057 | |
d002ebf1 | 4058 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len); |
575a1d4b JZ |
4059 | if (!err) |
4060 | err = ext4_ext_insert_extent(handle, inode, path, | |
4061 | &newex, flags); | |
4d33b1ef | 4062 | if (err && free_on_err) { |
7132de74 MP |
4063 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4064 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4065 | /* free data blocks we just allocated */ |
c9de560d AT |
4066 | /* not a good idea to call discard here directly, |
4067 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4068 | ext4_discard_preallocations(inode); |
7dc57615 | 4069 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4070 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4071 | goto out2; |
315054f0 | 4072 | } |
a86c6181 | 4073 | |
a86c6181 | 4074 | /* previous routine could use block we allocated */ |
bf89d16f | 4075 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4076 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4077 | if (allocated > map->m_len) |
4078 | allocated = map->m_len; | |
4079 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4080 | |
5f634d06 AK |
4081 | /* |
4082 | * Update reserved blocks/metadata blocks after successful | |
4083 | * block allocation which had been deferred till now. | |
4084 | */ | |
7b415bf6 AK |
4085 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
4086 | /* | |
4087 | * Check how many clusters we had reserved this allocted range. | |
4088 | */ | |
4089 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4090 | map->m_lblk, allocated); | |
4091 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4092 | if (reserved_clusters) { | |
4093 | /* | |
4094 | * We have clusters reserved for this range. | |
4095 | * But since we are not doing actual allocation | |
4096 | * and are simply using blocks from previously | |
4097 | * allocated cluster, we should release the | |
4098 | * reservation and not claim quota. | |
4099 | */ | |
4100 | ext4_da_update_reserve_space(inode, | |
4101 | reserved_clusters, 0); | |
4102 | } | |
4103 | } else { | |
4104 | BUG_ON(allocated_clusters < reserved_clusters); | |
4105 | /* We will claim quota for all newly allocated blocks.*/ | |
4106 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4107 | 1); | |
4108 | if (reserved_clusters < allocated_clusters) { | |
4109 | int reservation = allocated_clusters - | |
4110 | reserved_clusters; | |
4111 | /* | |
4112 | * It seems we claimed few clusters outside of | |
4113 | * the range of this allocation. We should give | |
4114 | * it back to the reservation pool. This can | |
4115 | * happen in the following case: | |
4116 | * | |
4117 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4118 | * cluster has 4 blocks. Thus, the clusters | |
4119 | * are [0-3],[4-7],[8-11]... | |
4120 | * * First comes delayed allocation write for | |
4121 | * logical blocks 10 & 11. Since there were no | |
4122 | * previous delayed allocated blocks in the | |
4123 | * range [8-11], we would reserve 1 cluster | |
4124 | * for this write. | |
4125 | * * Next comes write for logical blocks 3 to 8. | |
4126 | * In this case, we will reserve 2 clusters | |
4127 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4128 | * that range has a delayed allocated blocks. | |
4129 | * Thus total reserved clusters now becomes 3. | |
4130 | * * Now, during the delayed allocation writeout | |
4131 | * time, we will first write blocks [3-8] and | |
4132 | * allocate 3 clusters for writing these | |
4133 | * blocks. Also, we would claim all these | |
4134 | * three clusters above. | |
4135 | * * Now when we come here to writeout the | |
4136 | * blocks [10-11], we would expect to claim | |
4137 | * the reservation of 1 cluster we had made | |
4138 | * (and we would claim it since there are no | |
4139 | * more delayed allocated blocks in the range | |
4140 | * [8-11]. But our reserved cluster count had | |
4141 | * already gone to 0. | |
4142 | * | |
4143 | * Thus, at the step 4 above when we determine | |
4144 | * that there are still some unwritten delayed | |
4145 | * allocated blocks outside of our current | |
4146 | * block range, we should increment the | |
4147 | * reserved clusters count so that when the | |
4148 | * remaining blocks finally gets written, we | |
4149 | * could claim them. | |
4150 | */ | |
4151 | while (reservation) { | |
4152 | ext4_da_reserve_space(inode, | |
4153 | map->m_lblk); | |
4154 | reservation--; | |
4155 | } | |
4156 | } | |
4157 | } | |
4158 | } | |
5f634d06 | 4159 | |
b436b9be JK |
4160 | /* |
4161 | * Cache the extent and update transaction to commit on fdatasync only | |
4162 | * when it is _not_ an uninitialized extent. | |
4163 | */ | |
4164 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 4165 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
4166 | ext4_update_inode_fsync_trans(handle, inode, 1); |
4167 | } else | |
4168 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 4169 | out: |
e35fd660 TT |
4170 | if (allocated > map->m_len) |
4171 | allocated = map->m_len; | |
a86c6181 | 4172 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4173 | map->m_flags |= EXT4_MAP_MAPPED; |
4174 | map->m_pblk = newblock; | |
4175 | map->m_len = allocated; | |
a86c6181 AT |
4176 | out2: |
4177 | if (path) { | |
4178 | ext4_ext_drop_refs(path); | |
4179 | kfree(path); | |
4180 | } | |
0562e0ba JZ |
4181 | trace_ext4_ext_map_blocks_exit(inode, map->m_lblk, |
4182 | newblock, map->m_len, err ? err : allocated); | |
e861304b AH |
4183 | |
4184 | result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ? | |
4185 | punched_out : allocated; | |
4186 | ||
4187 | return err ? err : result; | |
a86c6181 AT |
4188 | } |
4189 | ||
cf108bca | 4190 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
4191 | { |
4192 | struct address_space *mapping = inode->i_mapping; | |
4193 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 4194 | ext4_lblk_t last_block; |
a86c6181 | 4195 | handle_t *handle; |
189e868f | 4196 | loff_t page_len; |
a86c6181 AT |
4197 | int err = 0; |
4198 | ||
3889fd57 JZ |
4199 | /* |
4200 | * finish any pending end_io work so we won't run the risk of | |
4201 | * converting any truncated blocks to initialized later | |
4202 | */ | |
4203 | ext4_flush_completed_IO(inode); | |
4204 | ||
a86c6181 AT |
4205 | /* |
4206 | * probably first extent we're gonna free will be last in block | |
4207 | */ | |
f3bd1f3f | 4208 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 4209 | handle = ext4_journal_start(inode, err); |
cf108bca | 4210 | if (IS_ERR(handle)) |
a86c6181 | 4211 | return; |
a86c6181 | 4212 | |
189e868f AH |
4213 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
4214 | page_len = PAGE_CACHE_SIZE - | |
4215 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4216 | ||
4217 | err = ext4_discard_partial_page_buffers(handle, | |
4218 | mapping, inode->i_size, page_len, 0); | |
4219 | ||
4220 | if (err) | |
4221 | goto out_stop; | |
4222 | } | |
a86c6181 | 4223 | |
9ddfc3dc JK |
4224 | if (ext4_orphan_add(handle, inode)) |
4225 | goto out_stop; | |
4226 | ||
0e855ac8 | 4227 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
4228 | ext4_ext_invalidate_cache(inode); |
4229 | ||
c2ea3fde | 4230 | ext4_discard_preallocations(inode); |
c9de560d | 4231 | |
a86c6181 | 4232 | /* |
d0d856e8 RD |
4233 | * TODO: optimization is possible here. |
4234 | * Probably we need not scan at all, | |
4235 | * because page truncation is enough. | |
a86c6181 | 4236 | */ |
a86c6181 AT |
4237 | |
4238 | /* we have to know where to truncate from in crash case */ | |
4239 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4240 | ext4_mark_inode_dirty(handle, inode); | |
4241 | ||
4242 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4243 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
c6a0371c | 4244 | err = ext4_ext_remove_space(inode, last_block); |
a86c6181 AT |
4245 | |
4246 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
4247 | * transaction synchronous. |
4248 | */ | |
a86c6181 | 4249 | if (IS_SYNC(inode)) |
0390131b | 4250 | ext4_handle_sync(handle); |
a86c6181 | 4251 | |
9ddfc3dc | 4252 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
4253 | |
4254 | out_stop: | |
a86c6181 | 4255 | /* |
d0d856e8 | 4256 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
4257 | * then we need to clear up the orphan record which we created above. |
4258 | * However, if this was a real unlink then we were called by | |
4259 | * ext4_delete_inode(), and we allow that function to clean up the | |
4260 | * orphan info for us. | |
4261 | */ | |
4262 | if (inode->i_nlink) | |
4263 | ext4_orphan_del(handle, inode); | |
4264 | ||
ef737728 SR |
4265 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4266 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
4267 | ext4_journal_stop(handle); |
4268 | } | |
4269 | ||
fd28784a AK |
4270 | static void ext4_falloc_update_inode(struct inode *inode, |
4271 | int mode, loff_t new_size, int update_ctime) | |
4272 | { | |
4273 | struct timespec now; | |
4274 | ||
4275 | if (update_ctime) { | |
4276 | now = current_fs_time(inode->i_sb); | |
4277 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4278 | inode->i_ctime = now; | |
4279 | } | |
4280 | /* | |
4281 | * Update only when preallocation was requested beyond | |
4282 | * the file size. | |
4283 | */ | |
cf17fea6 AK |
4284 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4285 | if (new_size > i_size_read(inode)) | |
4286 | i_size_write(inode, new_size); | |
4287 | if (new_size > EXT4_I(inode)->i_disksize) | |
4288 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4289 | } else { |
4290 | /* | |
4291 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4292 | * can proceed even if the new size is the same as i_size. | |
4293 | */ | |
4294 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4295 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4296 | } |
4297 | ||
4298 | } | |
4299 | ||
a2df2a63 | 4300 | /* |
2fe17c10 | 4301 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4302 | * operation, which gets called from sys_fallocate system call. |
4303 | * For block-mapped files, posix_fallocate should fall back to the method | |
4304 | * of writing zeroes to the required new blocks (the same behavior which is | |
4305 | * expected for file systems which do not support fallocate() system call). | |
4306 | */ | |
2fe17c10 | 4307 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4308 | { |
2fe17c10 | 4309 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 4310 | handle_t *handle; |
fd28784a | 4311 | loff_t new_size; |
498e5f24 | 4312 | unsigned int max_blocks; |
a2df2a63 AA |
4313 | int ret = 0; |
4314 | int ret2 = 0; | |
4315 | int retries = 0; | |
2ed88685 | 4316 | struct ext4_map_blocks map; |
a2df2a63 AA |
4317 | unsigned int credits, blkbits = inode->i_blkbits; |
4318 | ||
4319 | /* | |
4320 | * currently supporting (pre)allocate mode for extent-based | |
4321 | * files _only_ | |
4322 | */ | |
12e9b892 | 4323 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
a2df2a63 AA |
4324 | return -EOPNOTSUPP; |
4325 | ||
a4bb6b64 AH |
4326 | /* Return error if mode is not supported */ |
4327 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4328 | return -EOPNOTSUPP; | |
4329 | ||
4330 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4331 | return ext4_punch_hole(file, offset, len); | |
4332 | ||
0562e0ba | 4333 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4334 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4335 | /* |
4336 | * We can't just convert len to max_blocks because | |
4337 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4338 | */ | |
a2df2a63 | 4339 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4340 | - map.m_lblk; |
a2df2a63 | 4341 | /* |
f3bd1f3f | 4342 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4343 | */ |
f3bd1f3f | 4344 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4345 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4346 | ret = inode_newsize_ok(inode, (len + offset)); |
4347 | if (ret) { | |
4348 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4349 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4350 | return ret; |
4351 | } | |
a2df2a63 AA |
4352 | retry: |
4353 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4354 | map.m_lblk = map.m_lblk + ret; |
4355 | map.m_len = max_blocks = max_blocks - ret; | |
a2df2a63 AA |
4356 | handle = ext4_journal_start(inode, credits); |
4357 | if (IS_ERR(handle)) { | |
4358 | ret = PTR_ERR(handle); | |
4359 | break; | |
4360 | } | |
2ed88685 | 4361 | ret = ext4_map_blocks(handle, inode, &map, |
556b27ab VH |
4362 | EXT4_GET_BLOCKS_CREATE_UNINIT_EXT | |
4363 | EXT4_GET_BLOCKS_NO_NORMALIZE); | |
221879c9 | 4364 | if (ret <= 0) { |
2c98615d AK |
4365 | #ifdef EXT4FS_DEBUG |
4366 | WARN_ON(ret <= 0); | |
e35fd660 | 4367 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
2c98615d | 4368 | "returned error inode#%lu, block=%u, " |
9fd9784c | 4369 | "max_blocks=%u", __func__, |
a6371b63 | 4370 | inode->i_ino, map.m_lblk, max_blocks); |
2c98615d | 4371 | #endif |
a2df2a63 AA |
4372 | ext4_mark_inode_dirty(handle, inode); |
4373 | ret2 = ext4_journal_stop(handle); | |
4374 | break; | |
4375 | } | |
2ed88685 | 4376 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4377 | blkbits) >> blkbits)) |
4378 | new_size = offset + len; | |
4379 | else | |
29ae07b7 | 4380 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4381 | |
fd28784a | 4382 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4383 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 AA |
4384 | ext4_mark_inode_dirty(handle, inode); |
4385 | ret2 = ext4_journal_stop(handle); | |
4386 | if (ret2) | |
4387 | break; | |
4388 | } | |
fd28784a AK |
4389 | if (ret == -ENOSPC && |
4390 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4391 | ret = 0; | |
a2df2a63 | 4392 | goto retry; |
a2df2a63 | 4393 | } |
55bd725a | 4394 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4395 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4396 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4397 | return ret > 0 ? ret2 : ret; |
4398 | } | |
6873fa0d | 4399 | |
0031462b MC |
4400 | /* |
4401 | * This function convert a range of blocks to written extents | |
4402 | * The caller of this function will pass the start offset and the size. | |
4403 | * all unwritten extents within this range will be converted to | |
4404 | * written extents. | |
4405 | * | |
4406 | * This function is called from the direct IO end io call back | |
4407 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4408 | * Returns 0 on success. |
0031462b MC |
4409 | */ |
4410 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4411 | ssize_t len) |
0031462b MC |
4412 | { |
4413 | handle_t *handle; | |
0031462b MC |
4414 | unsigned int max_blocks; |
4415 | int ret = 0; | |
4416 | int ret2 = 0; | |
2ed88685 | 4417 | struct ext4_map_blocks map; |
0031462b MC |
4418 | unsigned int credits, blkbits = inode->i_blkbits; |
4419 | ||
2ed88685 | 4420 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4421 | /* |
4422 | * We can't just convert len to max_blocks because | |
4423 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4424 | */ | |
2ed88685 TT |
4425 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4426 | map.m_lblk); | |
0031462b MC |
4427 | /* |
4428 | * credits to insert 1 extent into extent tree | |
4429 | */ | |
4430 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4431 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4432 | map.m_lblk += ret; |
4433 | map.m_len = (max_blocks -= ret); | |
0031462b MC |
4434 | handle = ext4_journal_start(inode, credits); |
4435 | if (IS_ERR(handle)) { | |
4436 | ret = PTR_ERR(handle); | |
4437 | break; | |
4438 | } | |
2ed88685 | 4439 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4440 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
0031462b MC |
4441 | if (ret <= 0) { |
4442 | WARN_ON(ret <= 0); | |
e35fd660 | 4443 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
0031462b MC |
4444 | "returned error inode#%lu, block=%u, " |
4445 | "max_blocks=%u", __func__, | |
2ed88685 | 4446 | inode->i_ino, map.m_lblk, map.m_len); |
0031462b MC |
4447 | } |
4448 | ext4_mark_inode_dirty(handle, inode); | |
4449 | ret2 = ext4_journal_stop(handle); | |
4450 | if (ret <= 0 || ret2 ) | |
4451 | break; | |
4452 | } | |
4453 | return ret > 0 ? ret2 : ret; | |
4454 | } | |
6d9c85eb | 4455 | |
6873fa0d ES |
4456 | /* |
4457 | * Callback function called for each extent to gather FIEMAP information. | |
4458 | */ | |
c03f8aa9 | 4459 | static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next, |
6873fa0d ES |
4460 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
4461 | void *data) | |
4462 | { | |
6873fa0d ES |
4463 | __u64 logical; |
4464 | __u64 physical; | |
4465 | __u64 length; | |
4466 | __u32 flags = 0; | |
6d9c85eb YY |
4467 | int ret = 0; |
4468 | struct fiemap_extent_info *fieinfo = data; | |
4469 | unsigned char blksize_bits; | |
6873fa0d | 4470 | |
6d9c85eb YY |
4471 | blksize_bits = inode->i_sb->s_blocksize_bits; |
4472 | logical = (__u64)newex->ec_block << blksize_bits; | |
6873fa0d | 4473 | |
b05e6ae5 | 4474 | if (newex->ec_start == 0) { |
6d9c85eb YY |
4475 | /* |
4476 | * No extent in extent-tree contains block @newex->ec_start, | |
4477 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
4478 | * | |
4479 | * Holes or delayed-extents are processed as follows. | |
4480 | * 1. lookup dirty pages with specified range in pagecache. | |
4481 | * If no page is got, then there is no delayed-extent and | |
4482 | * return with EXT_CONTINUE. | |
4483 | * 2. find the 1st mapped buffer, | |
4484 | * 3. check if the mapped buffer is both in the request range | |
4485 | * and a delayed buffer. If not, there is no delayed-extent, | |
4486 | * then return. | |
4487 | * 4. a delayed-extent is found, the extent will be collected. | |
4488 | */ | |
4489 | ext4_lblk_t end = 0; | |
4490 | pgoff_t last_offset; | |
4491 | pgoff_t offset; | |
4492 | pgoff_t index; | |
b221349f | 4493 | pgoff_t start_index = 0; |
6d9c85eb | 4494 | struct page **pages = NULL; |
6873fa0d | 4495 | struct buffer_head *bh = NULL; |
6d9c85eb YY |
4496 | struct buffer_head *head = NULL; |
4497 | unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *); | |
4498 | ||
4499 | pages = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4500 | if (pages == NULL) | |
4501 | return -ENOMEM; | |
6873fa0d ES |
4502 | |
4503 | offset = logical >> PAGE_SHIFT; | |
6d9c85eb YY |
4504 | repeat: |
4505 | last_offset = offset; | |
4506 | head = NULL; | |
4507 | ret = find_get_pages_tag(inode->i_mapping, &offset, | |
4508 | PAGECACHE_TAG_DIRTY, nr_pages, pages); | |
4509 | ||
4510 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4511 | /* First time, try to find a mapped buffer. */ | |
4512 | if (ret == 0) { | |
4513 | out: | |
4514 | for (index = 0; index < ret; index++) | |
4515 | page_cache_release(pages[index]); | |
4516 | /* just a hole. */ | |
4517 | kfree(pages); | |
4518 | return EXT_CONTINUE; | |
4519 | } | |
b221349f | 4520 | index = 0; |
6873fa0d | 4521 | |
b221349f | 4522 | next_page: |
6d9c85eb | 4523 | /* Try to find the 1st mapped buffer. */ |
b221349f | 4524 | end = ((__u64)pages[index]->index << PAGE_SHIFT) >> |
6d9c85eb | 4525 | blksize_bits; |
b221349f | 4526 | if (!page_has_buffers(pages[index])) |
6d9c85eb | 4527 | goto out; |
b221349f | 4528 | head = page_buffers(pages[index]); |
6d9c85eb YY |
4529 | if (!head) |
4530 | goto out; | |
6873fa0d | 4531 | |
b221349f | 4532 | index++; |
6d9c85eb YY |
4533 | bh = head; |
4534 | do { | |
b221349f YY |
4535 | if (end >= newex->ec_block + |
4536 | newex->ec_len) | |
4537 | /* The buffer is out of | |
4538 | * the request range. | |
4539 | */ | |
4540 | goto out; | |
4541 | ||
4542 | if (buffer_mapped(bh) && | |
4543 | end >= newex->ec_block) { | |
4544 | start_index = index - 1; | |
6d9c85eb | 4545 | /* get the 1st mapped buffer. */ |
6d9c85eb YY |
4546 | goto found_mapped_buffer; |
4547 | } | |
b221349f | 4548 | |
6d9c85eb YY |
4549 | bh = bh->b_this_page; |
4550 | end++; | |
4551 | } while (bh != head); | |
6873fa0d | 4552 | |
b221349f YY |
4553 | /* No mapped buffer in the range found in this page, |
4554 | * We need to look up next page. | |
4555 | */ | |
4556 | if (index >= ret) { | |
4557 | /* There is no page left, but we need to limit | |
4558 | * newex->ec_len. | |
4559 | */ | |
4560 | newex->ec_len = end - newex->ec_block; | |
4561 | goto out; | |
4562 | } | |
4563 | goto next_page; | |
6873fa0d | 4564 | } else { |
6d9c85eb YY |
4565 | /*Find contiguous delayed buffers. */ |
4566 | if (ret > 0 && pages[0]->index == last_offset) | |
4567 | head = page_buffers(pages[0]); | |
4568 | bh = head; | |
b221349f YY |
4569 | index = 1; |
4570 | start_index = 0; | |
6873fa0d | 4571 | } |
6d9c85eb YY |
4572 | |
4573 | found_mapped_buffer: | |
4574 | if (bh != NULL && buffer_delay(bh)) { | |
4575 | /* 1st or contiguous delayed buffer found. */ | |
4576 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4577 | /* | |
4578 | * 1st delayed buffer found, record | |
4579 | * the start of extent. | |
4580 | */ | |
4581 | flags |= FIEMAP_EXTENT_DELALLOC; | |
4582 | newex->ec_block = end; | |
4583 | logical = (__u64)end << blksize_bits; | |
4584 | } | |
4585 | /* Find contiguous delayed buffers. */ | |
4586 | do { | |
4587 | if (!buffer_delay(bh)) | |
4588 | goto found_delayed_extent; | |
4589 | bh = bh->b_this_page; | |
4590 | end++; | |
4591 | } while (bh != head); | |
4592 | ||
b221349f | 4593 | for (; index < ret; index++) { |
6d9c85eb YY |
4594 | if (!page_has_buffers(pages[index])) { |
4595 | bh = NULL; | |
4596 | break; | |
4597 | } | |
4598 | head = page_buffers(pages[index]); | |
4599 | if (!head) { | |
4600 | bh = NULL; | |
4601 | break; | |
4602 | } | |
b221349f | 4603 | |
6d9c85eb | 4604 | if (pages[index]->index != |
b221349f YY |
4605 | pages[start_index]->index + index |
4606 | - start_index) { | |
6d9c85eb YY |
4607 | /* Blocks are not contiguous. */ |
4608 | bh = NULL; | |
4609 | break; | |
4610 | } | |
4611 | bh = head; | |
4612 | do { | |
4613 | if (!buffer_delay(bh)) | |
4614 | /* Delayed-extent ends. */ | |
4615 | goto found_delayed_extent; | |
4616 | bh = bh->b_this_page; | |
4617 | end++; | |
4618 | } while (bh != head); | |
4619 | } | |
4620 | } else if (!(flags & FIEMAP_EXTENT_DELALLOC)) | |
4621 | /* a hole found. */ | |
4622 | goto out; | |
4623 | ||
4624 | found_delayed_extent: | |
4625 | newex->ec_len = min(end - newex->ec_block, | |
4626 | (ext4_lblk_t)EXT_INIT_MAX_LEN); | |
4627 | if (ret == nr_pages && bh != NULL && | |
4628 | newex->ec_len < EXT_INIT_MAX_LEN && | |
4629 | buffer_delay(bh)) { | |
4630 | /* Have not collected an extent and continue. */ | |
4631 | for (index = 0; index < ret; index++) | |
4632 | page_cache_release(pages[index]); | |
4633 | goto repeat; | |
6873fa0d | 4634 | } |
6d9c85eb YY |
4635 | |
4636 | for (index = 0; index < ret; index++) | |
4637 | page_cache_release(pages[index]); | |
4638 | kfree(pages); | |
6873fa0d ES |
4639 | } |
4640 | ||
4641 | physical = (__u64)newex->ec_start << blksize_bits; | |
4642 | length = (__u64)newex->ec_len << blksize_bits; | |
4643 | ||
4644 | if (ex && ext4_ext_is_uninitialized(ex)) | |
4645 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
4646 | ||
c03f8aa9 | 4647 | if (next == EXT_MAX_BLOCKS) |
6873fa0d ES |
4648 | flags |= FIEMAP_EXTENT_LAST; |
4649 | ||
6d9c85eb | 4650 | ret = fiemap_fill_next_extent(fieinfo, logical, physical, |
6873fa0d | 4651 | length, flags); |
6d9c85eb YY |
4652 | if (ret < 0) |
4653 | return ret; | |
4654 | if (ret == 1) | |
6873fa0d | 4655 | return EXT_BREAK; |
6873fa0d ES |
4656 | return EXT_CONTINUE; |
4657 | } | |
6873fa0d ES |
4658 | /* fiemap flags we can handle specified here */ |
4659 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4660 | ||
3a06d778 AK |
4661 | static int ext4_xattr_fiemap(struct inode *inode, |
4662 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4663 | { |
4664 | __u64 physical = 0; | |
4665 | __u64 length; | |
4666 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4667 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4668 | int error = 0; | |
4669 | ||
4670 | /* in-inode? */ | |
19f5fb7a | 4671 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4672 | struct ext4_iloc iloc; |
4673 | int offset; /* offset of xattr in inode */ | |
4674 | ||
4675 | error = ext4_get_inode_loc(inode, &iloc); | |
4676 | if (error) | |
4677 | return error; | |
4678 | physical = iloc.bh->b_blocknr << blockbits; | |
4679 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4680 | EXT4_I(inode)->i_extra_isize; | |
4681 | physical += offset; | |
4682 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4683 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4684 | brelse(iloc.bh); |
6873fa0d ES |
4685 | } else { /* external block */ |
4686 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4687 | length = inode->i_sb->s_blocksize; | |
4688 | } | |
4689 | ||
4690 | if (physical) | |
4691 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4692 | length, flags); | |
4693 | return (error < 0 ? error : 0); | |
4694 | } | |
4695 | ||
a4bb6b64 AH |
4696 | /* |
4697 | * ext4_ext_punch_hole | |
4698 | * | |
4699 | * Punches a hole of "length" bytes in a file starting | |
4700 | * at byte "offset" | |
4701 | * | |
4702 | * @inode: The inode of the file to punch a hole in | |
4703 | * @offset: The starting byte offset of the hole | |
4704 | * @length: The length of the hole | |
4705 | * | |
4706 | * Returns the number of blocks removed or negative on err | |
4707 | */ | |
4708 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4709 | { | |
4710 | struct inode *inode = file->f_path.dentry->d_inode; | |
4711 | struct super_block *sb = inode->i_sb; | |
4712 | struct ext4_ext_cache cache_ex; | |
4713 | ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks; | |
4714 | struct address_space *mapping = inode->i_mapping; | |
4715 | struct ext4_map_blocks map; | |
4716 | handle_t *handle; | |
ba06208a AH |
4717 | loff_t first_page, last_page, page_len; |
4718 | loff_t first_page_offset, last_page_offset; | |
a4bb6b64 AH |
4719 | int ret, credits, blocks_released, err = 0; |
4720 | ||
2be4751b AH |
4721 | /* No need to punch hole beyond i_size */ |
4722 | if (offset >= inode->i_size) | |
4723 | return 0; | |
4724 | ||
4725 | /* | |
4726 | * If the hole extends beyond i_size, set the hole | |
4727 | * to end after the page that contains i_size | |
4728 | */ | |
4729 | if (offset + length > inode->i_size) { | |
4730 | length = inode->i_size + | |
4731 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4732 | offset; | |
4733 | } | |
4734 | ||
a4bb6b64 AH |
4735 | first_block = (offset + sb->s_blocksize - 1) >> |
4736 | EXT4_BLOCK_SIZE_BITS(sb); | |
4737 | last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4738 | ||
a4bb6b64 AH |
4739 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4740 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4741 | ||
4742 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4743 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4744 | ||
4745 | /* | |
4746 | * Write out all dirty pages to avoid race conditions | |
4747 | * Then release them. | |
4748 | */ | |
4749 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4750 | err = filemap_write_and_wait_range(mapping, | |
2be4751b | 4751 | offset, offset + length - 1); |
a4bb6b64 | 4752 | |
2be4751b AH |
4753 | if (err) |
4754 | return err; | |
a4bb6b64 AH |
4755 | } |
4756 | ||
4757 | /* Now release the pages */ | |
4758 | if (last_page_offset > first_page_offset) { | |
4759 | truncate_inode_pages_range(mapping, first_page_offset, | |
4760 | last_page_offset-1); | |
4761 | } | |
4762 | ||
4763 | /* finish any pending end_io work */ | |
4764 | ext4_flush_completed_IO(inode); | |
4765 | ||
4766 | credits = ext4_writepage_trans_blocks(inode); | |
4767 | handle = ext4_journal_start(inode, credits); | |
4768 | if (IS_ERR(handle)) | |
4769 | return PTR_ERR(handle); | |
4770 | ||
4771 | err = ext4_orphan_add(handle, inode); | |
4772 | if (err) | |
4773 | goto out; | |
4774 | ||
4775 | /* | |
ba06208a AH |
4776 | * Now we need to zero out the non-page-aligned data in the |
4777 | * pages at the start and tail of the hole, and unmap the buffer | |
4778 | * heads for the block aligned regions of the page that were | |
4779 | * completely zeroed. | |
a4bb6b64 | 4780 | */ |
ba06208a AH |
4781 | if (first_page > last_page) { |
4782 | /* | |
4783 | * If the file space being truncated is contained within a page | |
4784 | * just zero out and unmap the middle of that page | |
4785 | */ | |
4786 | err = ext4_discard_partial_page_buffers(handle, | |
4787 | mapping, offset, length, 0); | |
4788 | ||
4789 | if (err) | |
4790 | goto out; | |
4791 | } else { | |
4792 | /* | |
4793 | * zero out and unmap the partial page that contains | |
4794 | * the start of the hole | |
4795 | */ | |
4796 | page_len = first_page_offset - offset; | |
4797 | if (page_len > 0) { | |
4798 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4799 | offset, page_len, 0); | |
4800 | if (err) | |
4801 | goto out; | |
4802 | } | |
4803 | ||
4804 | /* | |
4805 | * zero out and unmap the partial page that contains | |
4806 | * the end of the hole | |
4807 | */ | |
4808 | page_len = offset + length - last_page_offset; | |
4809 | if (page_len > 0) { | |
4810 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4811 | last_page_offset, page_len, 0); | |
4812 | if (err) | |
4813 | goto out; | |
a4bb6b64 AH |
4814 | } |
4815 | } | |
4816 | ||
2be4751b AH |
4817 | |
4818 | /* | |
4819 | * If i_size is contained in the last page, we need to | |
4820 | * unmap and zero the partial page after i_size | |
4821 | */ | |
4822 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4823 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4824 | ||
4825 | page_len = PAGE_CACHE_SIZE - | |
4826 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4827 | ||
4828 | if (page_len > 0) { | |
4829 | err = ext4_discard_partial_page_buffers(handle, | |
4830 | mapping, inode->i_size, page_len, 0); | |
4831 | ||
4832 | if (err) | |
4833 | goto out; | |
4834 | } | |
4835 | } | |
4836 | ||
a4bb6b64 AH |
4837 | /* If there are no blocks to remove, return now */ |
4838 | if (first_block >= last_block) | |
4839 | goto out; | |
4840 | ||
4841 | down_write(&EXT4_I(inode)->i_data_sem); | |
4842 | ext4_ext_invalidate_cache(inode); | |
4843 | ext4_discard_preallocations(inode); | |
4844 | ||
4845 | /* | |
4846 | * Loop over all the blocks and identify blocks | |
4847 | * that need to be punched out | |
4848 | */ | |
4849 | iblock = first_block; | |
4850 | blocks_released = 0; | |
4851 | while (iblock < last_block) { | |
4852 | max_blocks = last_block - iblock; | |
4853 | num_blocks = 1; | |
4854 | memset(&map, 0, sizeof(map)); | |
4855 | map.m_lblk = iblock; | |
4856 | map.m_len = max_blocks; | |
4857 | ret = ext4_ext_map_blocks(handle, inode, &map, | |
4858 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
4859 | ||
4860 | if (ret > 0) { | |
4861 | blocks_released += ret; | |
4862 | num_blocks = ret; | |
4863 | } else if (ret == 0) { | |
4864 | /* | |
4865 | * If map blocks could not find the block, | |
4866 | * then it is in a hole. If the hole was | |
4867 | * not already cached, then map blocks should | |
4868 | * put it in the cache. So we can get the hole | |
4869 | * out of the cache | |
4870 | */ | |
4871 | memset(&cache_ex, 0, sizeof(cache_ex)); | |
4872 | if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) && | |
4873 | !cache_ex.ec_start) { | |
4874 | ||
4875 | /* The hole is cached */ | |
4876 | num_blocks = cache_ex.ec_block + | |
4877 | cache_ex.ec_len - iblock; | |
4878 | ||
4879 | } else { | |
4880 | /* The block could not be identified */ | |
4881 | err = -EIO; | |
4882 | break; | |
4883 | } | |
4884 | } else { | |
4885 | /* Map blocks error */ | |
4886 | err = ret; | |
4887 | break; | |
4888 | } | |
4889 | ||
4890 | if (num_blocks == 0) { | |
4891 | /* This condition should never happen */ | |
4892 | ext_debug("Block lookup failed"); | |
4893 | err = -EIO; | |
4894 | break; | |
4895 | } | |
4896 | ||
4897 | iblock += num_blocks; | |
4898 | } | |
4899 | ||
4900 | if (blocks_released > 0) { | |
4901 | ext4_ext_invalidate_cache(inode); | |
4902 | ext4_discard_preallocations(inode); | |
4903 | } | |
4904 | ||
4905 | if (IS_SYNC(inode)) | |
4906 | ext4_handle_sync(handle); | |
4907 | ||
4908 | up_write(&EXT4_I(inode)->i_data_sem); | |
4909 | ||
4910 | out: | |
4911 | ext4_orphan_del(handle, inode); | |
4912 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
4913 | ext4_mark_inode_dirty(handle, inode); | |
4914 | ext4_journal_stop(handle); | |
4915 | return err; | |
4916 | } | |
6873fa0d ES |
4917 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4918 | __u64 start, __u64 len) | |
4919 | { | |
4920 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4921 | int error = 0; |
4922 | ||
4923 | /* fallback to generic here if not in extents fmt */ | |
12e9b892 | 4924 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4925 | return generic_block_fiemap(inode, fieinfo, start, len, |
4926 | ext4_get_block); | |
4927 | ||
4928 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4929 | return -EBADR; | |
4930 | ||
4931 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4932 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4933 | } else { | |
aca92ff6 LM |
4934 | ext4_lblk_t len_blks; |
4935 | __u64 last_blk; | |
4936 | ||
6873fa0d | 4937 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4938 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4939 | if (last_blk >= EXT_MAX_BLOCKS) |
4940 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4941 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4942 | |
4943 | /* | |
4944 | * Walk the extent tree gathering extent information. | |
4945 | * ext4_ext_fiemap_cb will push extents back to user. | |
4946 | */ | |
6873fa0d ES |
4947 | error = ext4_ext_walk_space(inode, start_blk, len_blks, |
4948 | ext4_ext_fiemap_cb, fieinfo); | |
6873fa0d ES |
4949 | } |
4950 | ||
4951 | return error; | |
4952 | } |