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 CH |
44 | #include "ext4_jbd2.h" |
45 | #include "ext4_extents.h" | |
a86c6181 | 46 | |
0562e0ba JZ |
47 | #include <trace/events/ext4.h> |
48 | ||
d583fb87 AH |
49 | static int ext4_split_extent(handle_t *handle, |
50 | struct inode *inode, | |
51 | struct ext4_ext_path *path, | |
52 | struct ext4_map_blocks *map, | |
53 | int split_flag, | |
54 | int flags); | |
55 | ||
487caeef JK |
56 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
57 | struct inode *inode, | |
58 | int needed) | |
a86c6181 AT |
59 | { |
60 | int err; | |
61 | ||
0390131b FM |
62 | if (!ext4_handle_valid(handle)) |
63 | return 0; | |
a86c6181 | 64 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
65 | return 0; |
66 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 67 | if (err <= 0) |
9102e4fa | 68 | return err; |
487caeef | 69 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
70 | if (err == 0) |
71 | err = -EAGAIN; | |
487caeef JK |
72 | |
73 | return err; | |
a86c6181 AT |
74 | } |
75 | ||
76 | /* | |
77 | * could return: | |
78 | * - EROFS | |
79 | * - ENOMEM | |
80 | */ | |
81 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
82 | struct ext4_ext_path *path) | |
83 | { | |
84 | if (path->p_bh) { | |
85 | /* path points to block */ | |
86 | return ext4_journal_get_write_access(handle, path->p_bh); | |
87 | } | |
88 | /* path points to leaf/index in inode body */ | |
89 | /* we use in-core data, no need to protect them */ | |
90 | return 0; | |
91 | } | |
92 | ||
93 | /* | |
94 | * could return: | |
95 | * - EROFS | |
96 | * - ENOMEM | |
97 | * - EIO | |
98 | */ | |
9ea7a0df TT |
99 | #define ext4_ext_dirty(handle, inode, path) \ |
100 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
101 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
102 | handle_t *handle, struct inode *inode, | |
103 | struct ext4_ext_path *path) | |
a86c6181 AT |
104 | { |
105 | int err; | |
106 | if (path->p_bh) { | |
107 | /* path points to block */ | |
9ea7a0df TT |
108 | err = __ext4_handle_dirty_metadata(where, line, handle, |
109 | inode, path->p_bh); | |
a86c6181 AT |
110 | } else { |
111 | /* path points to leaf/index in inode body */ | |
112 | err = ext4_mark_inode_dirty(handle, inode); | |
113 | } | |
114 | return err; | |
115 | } | |
116 | ||
f65e6fba | 117 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 118 | struct ext4_ext_path *path, |
725d26d3 | 119 | ext4_lblk_t block) |
a86c6181 | 120 | { |
a86c6181 AT |
121 | int depth; |
122 | ||
123 | if (path) { | |
124 | struct ext4_extent *ex; | |
125 | depth = path->p_depth; | |
126 | ||
ad4fb9ca KM |
127 | /* |
128 | * Try to predict block placement assuming that we are | |
129 | * filling in a file which will eventually be | |
130 | * non-sparse --- i.e., in the case of libbfd writing | |
131 | * an ELF object sections out-of-order but in a way | |
132 | * the eventually results in a contiguous object or | |
133 | * executable file, or some database extending a table | |
134 | * space file. However, this is actually somewhat | |
135 | * non-ideal if we are writing a sparse file such as | |
136 | * qemu or KVM writing a raw image file that is going | |
137 | * to stay fairly sparse, since it will end up | |
138 | * fragmenting the file system's free space. Maybe we | |
139 | * should have some hueristics or some way to allow | |
140 | * userspace to pass a hint to file system, | |
b8d6568a | 141 | * especially if the latter case turns out to be |
ad4fb9ca KM |
142 | * common. |
143 | */ | |
7e028976 | 144 | ex = path[depth].p_ext; |
ad4fb9ca KM |
145 | if (ex) { |
146 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
147 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
148 | ||
149 | if (block > ext_block) | |
150 | return ext_pblk + (block - ext_block); | |
151 | else | |
152 | return ext_pblk - (ext_block - block); | |
153 | } | |
a86c6181 | 154 | |
d0d856e8 RD |
155 | /* it looks like index is empty; |
156 | * try to find starting block from index itself */ | |
a86c6181 AT |
157 | if (path[depth].p_bh) |
158 | return path[depth].p_bh->b_blocknr; | |
159 | } | |
160 | ||
161 | /* OK. use inode's group */ | |
f86186b4 | 162 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
163 | } |
164 | ||
654b4908 AK |
165 | /* |
166 | * Allocation for a meta data block | |
167 | */ | |
f65e6fba | 168 | static ext4_fsblk_t |
654b4908 | 169 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 170 | struct ext4_ext_path *path, |
55f020db | 171 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 172 | { |
f65e6fba | 173 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
174 | |
175 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
176 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
177 | NULL, err); | |
a86c6181 AT |
178 | return newblock; |
179 | } | |
180 | ||
55ad63bf | 181 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
182 | { |
183 | int size; | |
184 | ||
185 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
186 | / sizeof(struct ext4_extent); | |
55ad63bf | 187 | if (!check) { |
bbf2f9fb | 188 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
189 | if (size > 6) |
190 | size = 6; | |
a86c6181 | 191 | #endif |
55ad63bf | 192 | } |
a86c6181 AT |
193 | return size; |
194 | } | |
195 | ||
55ad63bf | 196 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
197 | { |
198 | int size; | |
199 | ||
200 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
201 | / sizeof(struct ext4_extent_idx); | |
55ad63bf | 202 | if (!check) { |
bbf2f9fb | 203 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
204 | if (size > 5) |
205 | size = 5; | |
a86c6181 | 206 | #endif |
55ad63bf | 207 | } |
a86c6181 AT |
208 | return size; |
209 | } | |
210 | ||
55ad63bf | 211 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
212 | { |
213 | int size; | |
214 | ||
215 | size = sizeof(EXT4_I(inode)->i_data); | |
216 | size -= sizeof(struct ext4_extent_header); | |
217 | size /= sizeof(struct ext4_extent); | |
55ad63bf | 218 | if (!check) { |
bbf2f9fb | 219 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
220 | if (size > 3) |
221 | size = 3; | |
a86c6181 | 222 | #endif |
55ad63bf | 223 | } |
a86c6181 AT |
224 | return size; |
225 | } | |
226 | ||
55ad63bf | 227 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
228 | { |
229 | int size; | |
230 | ||
231 | size = sizeof(EXT4_I(inode)->i_data); | |
232 | size -= sizeof(struct ext4_extent_header); | |
233 | size /= sizeof(struct ext4_extent_idx); | |
55ad63bf | 234 | if (!check) { |
bbf2f9fb | 235 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
236 | if (size > 4) |
237 | size = 4; | |
a86c6181 | 238 | #endif |
55ad63bf | 239 | } |
a86c6181 AT |
240 | return size; |
241 | } | |
242 | ||
d2a17637 MC |
243 | /* |
244 | * Calculate the number of metadata blocks needed | |
245 | * to allocate @blocks | |
246 | * Worse case is one block per extent | |
247 | */ | |
01f49d0b | 248 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 249 | { |
9d0be502 TT |
250 | struct ext4_inode_info *ei = EXT4_I(inode); |
251 | int idxs, num = 0; | |
d2a17637 | 252 | |
9d0be502 TT |
253 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
254 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
255 | |
256 | /* | |
9d0be502 TT |
257 | * If the new delayed allocation block is contiguous with the |
258 | * previous da block, it can share index blocks with the | |
259 | * previous block, so we only need to allocate a new index | |
260 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
261 | * an additional index block, and at ldxs**3 blocks, yet | |
262 | * another index blocks. | |
d2a17637 | 263 | */ |
9d0be502 TT |
264 | if (ei->i_da_metadata_calc_len && |
265 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
266 | if ((ei->i_da_metadata_calc_len % idxs) == 0) | |
267 | num++; | |
268 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
269 | num++; | |
270 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
271 | num++; | |
272 | ei->i_da_metadata_calc_len = 0; | |
273 | } else | |
274 | ei->i_da_metadata_calc_len++; | |
275 | ei->i_da_metadata_calc_last_lblock++; | |
276 | return num; | |
277 | } | |
d2a17637 | 278 | |
9d0be502 TT |
279 | /* |
280 | * In the worst case we need a new set of index blocks at | |
281 | * every level of the inode's extent tree. | |
282 | */ | |
283 | ei->i_da_metadata_calc_len = 1; | |
284 | ei->i_da_metadata_calc_last_lblock = lblock; | |
285 | return ext_depth(inode) + 1; | |
d2a17637 MC |
286 | } |
287 | ||
c29c0ae7 AT |
288 | static int |
289 | ext4_ext_max_entries(struct inode *inode, int depth) | |
290 | { | |
291 | int max; | |
292 | ||
293 | if (depth == ext_depth(inode)) { | |
294 | if (depth == 0) | |
55ad63bf | 295 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 296 | else |
55ad63bf | 297 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
298 | } else { |
299 | if (depth == 0) | |
55ad63bf | 300 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 301 | else |
55ad63bf | 302 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
303 | } |
304 | ||
305 | return max; | |
306 | } | |
307 | ||
56b19868 AK |
308 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
309 | { | |
bf89d16f | 310 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 311 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 312 | |
6fd058f7 | 313 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
314 | } |
315 | ||
316 | static int ext4_valid_extent_idx(struct inode *inode, | |
317 | struct ext4_extent_idx *ext_idx) | |
318 | { | |
bf89d16f | 319 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 320 | |
6fd058f7 | 321 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
322 | } |
323 | ||
324 | static int ext4_valid_extent_entries(struct inode *inode, | |
325 | struct ext4_extent_header *eh, | |
326 | int depth) | |
327 | { | |
328 | struct ext4_extent *ext; | |
329 | struct ext4_extent_idx *ext_idx; | |
330 | unsigned short entries; | |
331 | if (eh->eh_entries == 0) | |
332 | return 1; | |
333 | ||
334 | entries = le16_to_cpu(eh->eh_entries); | |
335 | ||
336 | if (depth == 0) { | |
337 | /* leaf entries */ | |
338 | ext = EXT_FIRST_EXTENT(eh); | |
339 | while (entries) { | |
340 | if (!ext4_valid_extent(inode, ext)) | |
341 | return 0; | |
342 | ext++; | |
343 | entries--; | |
344 | } | |
345 | } else { | |
346 | ext_idx = EXT_FIRST_INDEX(eh); | |
347 | while (entries) { | |
348 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
349 | return 0; | |
350 | ext_idx++; | |
351 | entries--; | |
352 | } | |
353 | } | |
354 | return 1; | |
355 | } | |
356 | ||
c398eda0 TT |
357 | static int __ext4_ext_check(const char *function, unsigned int line, |
358 | struct inode *inode, struct ext4_extent_header *eh, | |
359 | int depth) | |
c29c0ae7 AT |
360 | { |
361 | const char *error_msg; | |
362 | int max = 0; | |
363 | ||
364 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
365 | error_msg = "invalid magic"; | |
366 | goto corrupted; | |
367 | } | |
368 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
369 | error_msg = "unexpected eh_depth"; | |
370 | goto corrupted; | |
371 | } | |
372 | if (unlikely(eh->eh_max == 0)) { | |
373 | error_msg = "invalid eh_max"; | |
374 | goto corrupted; | |
375 | } | |
376 | max = ext4_ext_max_entries(inode, depth); | |
377 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
378 | error_msg = "too large eh_max"; | |
379 | goto corrupted; | |
380 | } | |
381 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
382 | error_msg = "invalid eh_entries"; | |
383 | goto corrupted; | |
384 | } | |
56b19868 AK |
385 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
386 | error_msg = "invalid extent entries"; | |
387 | goto corrupted; | |
388 | } | |
c29c0ae7 AT |
389 | return 0; |
390 | ||
391 | corrupted: | |
c398eda0 | 392 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 393 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 394 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 395 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
396 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
397 | max, le16_to_cpu(eh->eh_depth), depth); | |
398 | ||
399 | return -EIO; | |
400 | } | |
401 | ||
56b19868 | 402 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 403 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 404 | |
7a262f7c AK |
405 | int ext4_ext_check_inode(struct inode *inode) |
406 | { | |
407 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
408 | } | |
409 | ||
a86c6181 AT |
410 | #ifdef EXT_DEBUG |
411 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
412 | { | |
413 | int k, l = path->p_depth; | |
414 | ||
415 | ext_debug("path:"); | |
416 | for (k = 0; k <= l; k++, path++) { | |
417 | if (path->p_idx) { | |
2ae02107 | 418 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 419 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 420 | } else if (path->p_ext) { |
553f9008 | 421 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 422 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 423 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 424 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 425 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
426 | } else |
427 | ext_debug(" []"); | |
428 | } | |
429 | ext_debug("\n"); | |
430 | } | |
431 | ||
432 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
433 | { | |
434 | int depth = ext_depth(inode); | |
435 | struct ext4_extent_header *eh; | |
436 | struct ext4_extent *ex; | |
437 | int i; | |
438 | ||
439 | if (!path) | |
440 | return; | |
441 | ||
442 | eh = path[depth].p_hdr; | |
443 | ex = EXT_FIRST_EXTENT(eh); | |
444 | ||
553f9008 M |
445 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
446 | ||
a86c6181 | 447 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
448 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
449 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 450 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
451 | } |
452 | ext_debug("\n"); | |
453 | } | |
1b16da77 YY |
454 | |
455 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
456 | ext4_fsblk_t newblock, int level) | |
457 | { | |
458 | int depth = ext_depth(inode); | |
459 | struct ext4_extent *ex; | |
460 | ||
461 | if (depth != level) { | |
462 | struct ext4_extent_idx *idx; | |
463 | idx = path[level].p_idx; | |
464 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
465 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
466 | le32_to_cpu(idx->ei_block), | |
467 | ext4_idx_pblock(idx), | |
468 | newblock); | |
469 | idx++; | |
470 | } | |
471 | ||
472 | return; | |
473 | } | |
474 | ||
475 | ex = path[depth].p_ext; | |
476 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
477 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
478 | le32_to_cpu(ex->ee_block), | |
479 | ext4_ext_pblock(ex), | |
480 | ext4_ext_is_uninitialized(ex), | |
481 | ext4_ext_get_actual_len(ex), | |
482 | newblock); | |
483 | ex++; | |
484 | } | |
485 | } | |
486 | ||
a86c6181 | 487 | #else |
af5bc92d TT |
488 | #define ext4_ext_show_path(inode, path) |
489 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 490 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
491 | #endif |
492 | ||
b35905c1 | 493 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
494 | { |
495 | int depth = path->p_depth; | |
496 | int i; | |
497 | ||
498 | for (i = 0; i <= depth; i++, path++) | |
499 | if (path->p_bh) { | |
500 | brelse(path->p_bh); | |
501 | path->p_bh = NULL; | |
502 | } | |
503 | } | |
504 | ||
505 | /* | |
d0d856e8 RD |
506 | * ext4_ext_binsearch_idx: |
507 | * binary search for the closest index of the given block | |
c29c0ae7 | 508 | * the header must be checked before calling this |
a86c6181 AT |
509 | */ |
510 | static void | |
725d26d3 AK |
511 | ext4_ext_binsearch_idx(struct inode *inode, |
512 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
513 | { |
514 | struct ext4_extent_header *eh = path->p_hdr; | |
515 | struct ext4_extent_idx *r, *l, *m; | |
516 | ||
a86c6181 | 517 | |
bba90743 | 518 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
519 | |
520 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 521 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
522 | while (l <= r) { |
523 | m = l + (r - l) / 2; | |
524 | if (block < le32_to_cpu(m->ei_block)) | |
525 | r = m - 1; | |
526 | else | |
527 | l = m + 1; | |
26d535ed DM |
528 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
529 | m, le32_to_cpu(m->ei_block), | |
530 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
531 | } |
532 | ||
533 | path->p_idx = l - 1; | |
f65e6fba | 534 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 535 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
536 | |
537 | #ifdef CHECK_BINSEARCH | |
538 | { | |
539 | struct ext4_extent_idx *chix, *ix; | |
540 | int k; | |
541 | ||
542 | chix = ix = EXT_FIRST_INDEX(eh); | |
543 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
544 | if (k != 0 && | |
545 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
546 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
547 | "first=0x%p\n", k, | |
548 | ix, EXT_FIRST_INDEX(eh)); | |
549 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
550 | le32_to_cpu(ix->ei_block), |
551 | le32_to_cpu(ix[-1].ei_block)); | |
552 | } | |
553 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 554 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
555 | if (block < le32_to_cpu(ix->ei_block)) |
556 | break; | |
557 | chix = ix; | |
558 | } | |
559 | BUG_ON(chix != path->p_idx); | |
560 | } | |
561 | #endif | |
562 | ||
563 | } | |
564 | ||
565 | /* | |
d0d856e8 RD |
566 | * ext4_ext_binsearch: |
567 | * binary search for closest extent of the given block | |
c29c0ae7 | 568 | * the header must be checked before calling this |
a86c6181 AT |
569 | */ |
570 | static void | |
725d26d3 AK |
571 | ext4_ext_binsearch(struct inode *inode, |
572 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
573 | { |
574 | struct ext4_extent_header *eh = path->p_hdr; | |
575 | struct ext4_extent *r, *l, *m; | |
576 | ||
a86c6181 AT |
577 | if (eh->eh_entries == 0) { |
578 | /* | |
d0d856e8 RD |
579 | * this leaf is empty: |
580 | * we get such a leaf in split/add case | |
a86c6181 AT |
581 | */ |
582 | return; | |
583 | } | |
584 | ||
bba90743 | 585 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
586 | |
587 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 588 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
589 | |
590 | while (l <= r) { | |
591 | m = l + (r - l) / 2; | |
592 | if (block < le32_to_cpu(m->ee_block)) | |
593 | r = m - 1; | |
594 | else | |
595 | l = m + 1; | |
26d535ed DM |
596 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
597 | m, le32_to_cpu(m->ee_block), | |
598 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
599 | } |
600 | ||
601 | path->p_ext = l - 1; | |
553f9008 | 602 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 603 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 604 | ext4_ext_pblock(path->p_ext), |
553f9008 | 605 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 606 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
607 | |
608 | #ifdef CHECK_BINSEARCH | |
609 | { | |
610 | struct ext4_extent *chex, *ex; | |
611 | int k; | |
612 | ||
613 | chex = ex = EXT_FIRST_EXTENT(eh); | |
614 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
615 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 616 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
617 | if (block < le32_to_cpu(ex->ee_block)) |
618 | break; | |
619 | chex = ex; | |
620 | } | |
621 | BUG_ON(chex != path->p_ext); | |
622 | } | |
623 | #endif | |
624 | ||
625 | } | |
626 | ||
627 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
628 | { | |
629 | struct ext4_extent_header *eh; | |
630 | ||
631 | eh = ext_inode_hdr(inode); | |
632 | eh->eh_depth = 0; | |
633 | eh->eh_entries = 0; | |
634 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 635 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
636 | ext4_mark_inode_dirty(handle, inode); |
637 | ext4_ext_invalidate_cache(inode); | |
638 | return 0; | |
639 | } | |
640 | ||
641 | struct ext4_ext_path * | |
725d26d3 AK |
642 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
643 | struct ext4_ext_path *path) | |
a86c6181 AT |
644 | { |
645 | struct ext4_extent_header *eh; | |
646 | struct buffer_head *bh; | |
647 | short int depth, i, ppos = 0, alloc = 0; | |
648 | ||
649 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 650 | depth = ext_depth(inode); |
a86c6181 AT |
651 | |
652 | /* account possible depth increase */ | |
653 | if (!path) { | |
5d4958f9 | 654 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
655 | GFP_NOFS); |
656 | if (!path) | |
657 | return ERR_PTR(-ENOMEM); | |
658 | alloc = 1; | |
659 | } | |
a86c6181 | 660 | path[0].p_hdr = eh; |
1973adcb | 661 | path[0].p_bh = NULL; |
a86c6181 | 662 | |
c29c0ae7 | 663 | i = depth; |
a86c6181 AT |
664 | /* walk through the tree */ |
665 | while (i) { | |
7a262f7c AK |
666 | int need_to_validate = 0; |
667 | ||
a86c6181 AT |
668 | ext_debug("depth %d: num %d, max %d\n", |
669 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 670 | |
a86c6181 | 671 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 672 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
673 | path[ppos].p_depth = i; |
674 | path[ppos].p_ext = NULL; | |
675 | ||
7a262f7c AK |
676 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
677 | if (unlikely(!bh)) | |
a86c6181 | 678 | goto err; |
7a262f7c | 679 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
680 | trace_ext4_ext_load_extent(inode, block, |
681 | path[ppos].p_block); | |
7a262f7c AK |
682 | if (bh_submit_read(bh) < 0) { |
683 | put_bh(bh); | |
684 | goto err; | |
685 | } | |
686 | /* validate the extent entries */ | |
687 | need_to_validate = 1; | |
688 | } | |
a86c6181 AT |
689 | eh = ext_block_hdr(bh); |
690 | ppos++; | |
273df556 FM |
691 | if (unlikely(ppos > depth)) { |
692 | put_bh(bh); | |
693 | EXT4_ERROR_INODE(inode, | |
694 | "ppos %d > depth %d", ppos, depth); | |
695 | goto err; | |
696 | } | |
a86c6181 AT |
697 | path[ppos].p_bh = bh; |
698 | path[ppos].p_hdr = eh; | |
699 | i--; | |
700 | ||
7a262f7c | 701 | if (need_to_validate && ext4_ext_check(inode, eh, i)) |
a86c6181 AT |
702 | goto err; |
703 | } | |
704 | ||
705 | path[ppos].p_depth = i; | |
a86c6181 AT |
706 | path[ppos].p_ext = NULL; |
707 | path[ppos].p_idx = NULL; | |
708 | ||
a86c6181 AT |
709 | /* find extent */ |
710 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
711 | /* if not an empty leaf */ |
712 | if (path[ppos].p_ext) | |
bf89d16f | 713 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
714 | |
715 | ext4_ext_show_path(inode, path); | |
716 | ||
717 | return path; | |
718 | ||
719 | err: | |
720 | ext4_ext_drop_refs(path); | |
721 | if (alloc) | |
722 | kfree(path); | |
723 | return ERR_PTR(-EIO); | |
724 | } | |
725 | ||
726 | /* | |
d0d856e8 RD |
727 | * ext4_ext_insert_index: |
728 | * insert new index [@logical;@ptr] into the block at @curp; | |
729 | * check where to insert: before @curp or after @curp | |
a86c6181 | 730 | */ |
1f109d5a TT |
731 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
732 | struct ext4_ext_path *curp, | |
733 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
734 | { |
735 | struct ext4_extent_idx *ix; | |
736 | int len, err; | |
737 | ||
7e028976 AM |
738 | err = ext4_ext_get_access(handle, inode, curp); |
739 | if (err) | |
a86c6181 AT |
740 | return err; |
741 | ||
273df556 FM |
742 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
743 | EXT4_ERROR_INODE(inode, | |
744 | "logical %d == ei_block %d!", | |
745 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
746 | return -EIO; | |
747 | } | |
d4620315 RD |
748 | |
749 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
750 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
751 | EXT4_ERROR_INODE(inode, | |
752 | "eh_entries %d >= eh_max %d!", | |
753 | le16_to_cpu(curp->p_hdr->eh_entries), | |
754 | le16_to_cpu(curp->p_hdr->eh_max)); | |
755 | return -EIO; | |
756 | } | |
757 | ||
a86c6181 AT |
758 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; |
759 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
760 | /* insert after */ | |
761 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
762 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
763 | len = len < 0 ? 0 : len; | |
26d535ed | 764 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
765 | "move %d from 0x%p to 0x%p\n", |
766 | logical, ptr, len, | |
767 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
768 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
769 | } | |
770 | ix = curp->p_idx + 1; | |
771 | } else { | |
772 | /* insert before */ | |
773 | len = len * sizeof(struct ext4_extent_idx); | |
774 | len = len < 0 ? 0 : len; | |
26d535ed | 775 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
776 | "move %d from 0x%p to 0x%p\n", |
777 | logical, ptr, len, | |
778 | curp->p_idx, (curp->p_idx + 1)); | |
779 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
780 | ix = curp->p_idx; | |
781 | } | |
782 | ||
783 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 784 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 785 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 786 | |
273df556 FM |
787 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
788 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
789 | return -EIO; | |
790 | } | |
a86c6181 AT |
791 | |
792 | err = ext4_ext_dirty(handle, inode, curp); | |
793 | ext4_std_error(inode->i_sb, err); | |
794 | ||
795 | return err; | |
796 | } | |
797 | ||
798 | /* | |
d0d856e8 RD |
799 | * ext4_ext_split: |
800 | * inserts new subtree into the path, using free index entry | |
801 | * at depth @at: | |
802 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
803 | * - makes decision where to split | |
804 | * - moves remaining extents and index entries (right to the split point) | |
805 | * into the newly allocated blocks | |
806 | * - initializes subtree | |
a86c6181 AT |
807 | */ |
808 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
809 | unsigned int flags, |
810 | struct ext4_ext_path *path, | |
811 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
812 | { |
813 | struct buffer_head *bh = NULL; | |
814 | int depth = ext_depth(inode); | |
815 | struct ext4_extent_header *neh; | |
816 | struct ext4_extent_idx *fidx; | |
a86c6181 | 817 | int i = at, k, m, a; |
f65e6fba | 818 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 819 | __le32 border; |
f65e6fba | 820 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
821 | int err = 0; |
822 | ||
823 | /* make decision: where to split? */ | |
d0d856e8 | 824 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 825 | |
d0d856e8 | 826 | /* if current leaf will be split, then we should use |
a86c6181 | 827 | * border from split point */ |
273df556 FM |
828 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
829 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
830 | return -EIO; | |
831 | } | |
a86c6181 AT |
832 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
833 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 834 | ext_debug("leaf will be split." |
a86c6181 | 835 | " next leaf starts at %d\n", |
8c55e204 | 836 | le32_to_cpu(border)); |
a86c6181 AT |
837 | } else { |
838 | border = newext->ee_block; | |
839 | ext_debug("leaf will be added." | |
840 | " next leaf starts at %d\n", | |
8c55e204 | 841 | le32_to_cpu(border)); |
a86c6181 AT |
842 | } |
843 | ||
844 | /* | |
d0d856e8 RD |
845 | * If error occurs, then we break processing |
846 | * and mark filesystem read-only. index won't | |
a86c6181 | 847 | * be inserted and tree will be in consistent |
d0d856e8 | 848 | * state. Next mount will repair buffers too. |
a86c6181 AT |
849 | */ |
850 | ||
851 | /* | |
d0d856e8 RD |
852 | * Get array to track all allocated blocks. |
853 | * We need this to handle errors and free blocks | |
854 | * upon them. | |
a86c6181 | 855 | */ |
5d4958f9 | 856 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
857 | if (!ablocks) |
858 | return -ENOMEM; | |
a86c6181 AT |
859 | |
860 | /* allocate all needed blocks */ | |
861 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
862 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 863 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 864 | newext, &err, flags); |
a86c6181 AT |
865 | if (newblock == 0) |
866 | goto cleanup; | |
867 | ablocks[a] = newblock; | |
868 | } | |
869 | ||
870 | /* initialize new leaf */ | |
871 | newblock = ablocks[--a]; | |
273df556 FM |
872 | if (unlikely(newblock == 0)) { |
873 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
874 | err = -EIO; | |
875 | goto cleanup; | |
876 | } | |
a86c6181 AT |
877 | bh = sb_getblk(inode->i_sb, newblock); |
878 | if (!bh) { | |
879 | err = -EIO; | |
880 | goto cleanup; | |
881 | } | |
882 | lock_buffer(bh); | |
883 | ||
7e028976 AM |
884 | err = ext4_journal_get_create_access(handle, bh); |
885 | if (err) | |
a86c6181 AT |
886 | goto cleanup; |
887 | ||
888 | neh = ext_block_hdr(bh); | |
889 | neh->eh_entries = 0; | |
55ad63bf | 890 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
891 | neh->eh_magic = EXT4_EXT_MAGIC; |
892 | neh->eh_depth = 0; | |
a86c6181 | 893 | |
d0d856e8 | 894 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
895 | if (unlikely(path[depth].p_hdr->eh_entries != |
896 | path[depth].p_hdr->eh_max)) { | |
897 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
898 | path[depth].p_hdr->eh_entries, | |
899 | path[depth].p_hdr->eh_max); | |
900 | err = -EIO; | |
901 | goto cleanup; | |
902 | } | |
a86c6181 | 903 | /* start copy from next extent */ |
1b16da77 YY |
904 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
905 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 906 | if (m) { |
1b16da77 YY |
907 | struct ext4_extent *ex; |
908 | ex = EXT_FIRST_EXTENT(neh); | |
909 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 910 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
911 | } |
912 | ||
913 | set_buffer_uptodate(bh); | |
914 | unlock_buffer(bh); | |
915 | ||
0390131b | 916 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 917 | if (err) |
a86c6181 AT |
918 | goto cleanup; |
919 | brelse(bh); | |
920 | bh = NULL; | |
921 | ||
922 | /* correct old leaf */ | |
923 | if (m) { | |
7e028976 AM |
924 | err = ext4_ext_get_access(handle, inode, path + depth); |
925 | if (err) | |
a86c6181 | 926 | goto cleanup; |
e8546d06 | 927 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
928 | err = ext4_ext_dirty(handle, inode, path + depth); |
929 | if (err) | |
a86c6181 AT |
930 | goto cleanup; |
931 | ||
932 | } | |
933 | ||
934 | /* create intermediate indexes */ | |
935 | k = depth - at - 1; | |
273df556 FM |
936 | if (unlikely(k < 0)) { |
937 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
938 | err = -EIO; | |
939 | goto cleanup; | |
940 | } | |
a86c6181 AT |
941 | if (k) |
942 | ext_debug("create %d intermediate indices\n", k); | |
943 | /* insert new index into current index block */ | |
944 | /* current depth stored in i var */ | |
945 | i = depth - 1; | |
946 | while (k--) { | |
947 | oldblock = newblock; | |
948 | newblock = ablocks[--a]; | |
bba90743 | 949 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
950 | if (!bh) { |
951 | err = -EIO; | |
952 | goto cleanup; | |
953 | } | |
954 | lock_buffer(bh); | |
955 | ||
7e028976 AM |
956 | err = ext4_journal_get_create_access(handle, bh); |
957 | if (err) | |
a86c6181 AT |
958 | goto cleanup; |
959 | ||
960 | neh = ext_block_hdr(bh); | |
961 | neh->eh_entries = cpu_to_le16(1); | |
962 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 963 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
964 | neh->eh_depth = cpu_to_le16(depth - i); |
965 | fidx = EXT_FIRST_INDEX(neh); | |
966 | fidx->ei_block = border; | |
f65e6fba | 967 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 968 | |
bba90743 ES |
969 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
970 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 971 | |
1b16da77 | 972 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
973 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
974 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
975 | EXT4_ERROR_INODE(inode, | |
976 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
977 | le32_to_cpu(path[i].p_ext->ee_block)); | |
978 | err = -EIO; | |
979 | goto cleanup; | |
980 | } | |
1b16da77 YY |
981 | /* start copy indexes */ |
982 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
983 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
984 | EXT_MAX_INDEX(path[i].p_hdr)); | |
985 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 986 | if (m) { |
1b16da77 | 987 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 988 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 989 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
990 | } |
991 | set_buffer_uptodate(bh); | |
992 | unlock_buffer(bh); | |
993 | ||
0390131b | 994 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 995 | if (err) |
a86c6181 AT |
996 | goto cleanup; |
997 | brelse(bh); | |
998 | bh = NULL; | |
999 | ||
1000 | /* correct old index */ | |
1001 | if (m) { | |
1002 | err = ext4_ext_get_access(handle, inode, path + i); | |
1003 | if (err) | |
1004 | goto cleanup; | |
e8546d06 | 1005 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1006 | err = ext4_ext_dirty(handle, inode, path + i); |
1007 | if (err) | |
1008 | goto cleanup; | |
1009 | } | |
1010 | ||
1011 | i--; | |
1012 | } | |
1013 | ||
1014 | /* insert new index */ | |
a86c6181 AT |
1015 | err = ext4_ext_insert_index(handle, inode, path + at, |
1016 | le32_to_cpu(border), newblock); | |
1017 | ||
1018 | cleanup: | |
1019 | if (bh) { | |
1020 | if (buffer_locked(bh)) | |
1021 | unlock_buffer(bh); | |
1022 | brelse(bh); | |
1023 | } | |
1024 | ||
1025 | if (err) { | |
1026 | /* free all allocated blocks in error case */ | |
1027 | for (i = 0; i < depth; i++) { | |
1028 | if (!ablocks[i]) | |
1029 | continue; | |
7dc57615 | 1030 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1031 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1032 | } |
1033 | } | |
1034 | kfree(ablocks); | |
1035 | ||
1036 | return err; | |
1037 | } | |
1038 | ||
1039 | /* | |
d0d856e8 RD |
1040 | * ext4_ext_grow_indepth: |
1041 | * implements tree growing procedure: | |
1042 | * - allocates new block | |
1043 | * - moves top-level data (index block or leaf) into the new block | |
1044 | * - initializes new top-level, creating index that points to the | |
1045 | * just created block | |
a86c6181 AT |
1046 | */ |
1047 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db AH |
1048 | unsigned int flags, |
1049 | struct ext4_ext_path *path, | |
1050 | struct ext4_extent *newext) | |
a86c6181 AT |
1051 | { |
1052 | struct ext4_ext_path *curp = path; | |
1053 | struct ext4_extent_header *neh; | |
a86c6181 | 1054 | struct buffer_head *bh; |
f65e6fba | 1055 | ext4_fsblk_t newblock; |
a86c6181 AT |
1056 | int err = 0; |
1057 | ||
55f020db AH |
1058 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
1059 | newext, &err, flags); | |
a86c6181 AT |
1060 | if (newblock == 0) |
1061 | return err; | |
1062 | ||
1063 | bh = sb_getblk(inode->i_sb, newblock); | |
1064 | if (!bh) { | |
1065 | err = -EIO; | |
1066 | ext4_std_error(inode->i_sb, err); | |
1067 | return err; | |
1068 | } | |
1069 | lock_buffer(bh); | |
1070 | ||
7e028976 AM |
1071 | err = ext4_journal_get_create_access(handle, bh); |
1072 | if (err) { | |
a86c6181 AT |
1073 | unlock_buffer(bh); |
1074 | goto out; | |
1075 | } | |
1076 | ||
1077 | /* move top-level index/leaf into new block */ | |
1078 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
1079 | ||
1080 | /* set size of new block */ | |
1081 | neh = ext_block_hdr(bh); | |
1082 | /* old root could have indexes or leaves | |
1083 | * so calculate e_max right way */ | |
1084 | if (ext_depth(inode)) | |
55ad63bf | 1085 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1086 | else |
55ad63bf | 1087 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1088 | neh->eh_magic = EXT4_EXT_MAGIC; |
1089 | set_buffer_uptodate(bh); | |
1090 | unlock_buffer(bh); | |
1091 | ||
0390131b | 1092 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1093 | if (err) |
a86c6181 AT |
1094 | goto out; |
1095 | ||
1096 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
1097 | err = ext4_ext_get_access(handle, inode, curp); |
1098 | if (err) | |
a86c6181 AT |
1099 | goto out; |
1100 | ||
1101 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1102 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); |
a86c6181 AT |
1103 | curp->p_hdr->eh_entries = cpu_to_le16(1); |
1104 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
1105 | |
1106 | if (path[0].p_hdr->eh_depth) | |
1107 | curp->p_idx->ei_block = | |
1108 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
1109 | else | |
1110 | curp->p_idx->ei_block = | |
1111 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 1112 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
1113 | |
1114 | neh = ext_inode_hdr(inode); | |
2ae02107 | 1115 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1116 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1117 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1118 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 AT |
1119 | |
1120 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
1121 | err = ext4_ext_dirty(handle, inode, curp); | |
1122 | out: | |
1123 | brelse(bh); | |
1124 | ||
1125 | return err; | |
1126 | } | |
1127 | ||
1128 | /* | |
d0d856e8 RD |
1129 | * ext4_ext_create_new_leaf: |
1130 | * finds empty index and adds new leaf. | |
1131 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1132 | */ |
1133 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1134 | unsigned int flags, |
1135 | struct ext4_ext_path *path, | |
1136 | struct ext4_extent *newext) | |
a86c6181 AT |
1137 | { |
1138 | struct ext4_ext_path *curp; | |
1139 | int depth, i, err = 0; | |
1140 | ||
1141 | repeat: | |
1142 | i = depth = ext_depth(inode); | |
1143 | ||
1144 | /* walk up to the tree and look for free index entry */ | |
1145 | curp = path + depth; | |
1146 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1147 | i--; | |
1148 | curp--; | |
1149 | } | |
1150 | ||
d0d856e8 RD |
1151 | /* we use already allocated block for index block, |
1152 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1153 | if (EXT_HAS_FREE_INDEX(curp)) { |
1154 | /* if we found index with free entry, then use that | |
1155 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1156 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1157 | if (err) |
1158 | goto out; | |
a86c6181 AT |
1159 | |
1160 | /* refill path */ | |
1161 | ext4_ext_drop_refs(path); | |
1162 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1163 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1164 | path); | |
a86c6181 AT |
1165 | if (IS_ERR(path)) |
1166 | err = PTR_ERR(path); | |
1167 | } else { | |
1168 | /* tree is full, time to grow in depth */ | |
55f020db AH |
1169 | err = ext4_ext_grow_indepth(handle, inode, flags, |
1170 | path, newext); | |
a86c6181 AT |
1171 | if (err) |
1172 | goto out; | |
1173 | ||
1174 | /* refill path */ | |
1175 | ext4_ext_drop_refs(path); | |
1176 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1177 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1178 | path); | |
a86c6181 AT |
1179 | if (IS_ERR(path)) { |
1180 | err = PTR_ERR(path); | |
1181 | goto out; | |
1182 | } | |
1183 | ||
1184 | /* | |
d0d856e8 RD |
1185 | * only first (depth 0 -> 1) produces free space; |
1186 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1187 | */ |
1188 | depth = ext_depth(inode); | |
1189 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1190 | /* now we need to split */ |
a86c6181 AT |
1191 | goto repeat; |
1192 | } | |
1193 | } | |
1194 | ||
1195 | out: | |
1196 | return err; | |
1197 | } | |
1198 | ||
1988b51e AT |
1199 | /* |
1200 | * search the closest allocated block to the left for *logical | |
1201 | * and returns it at @logical + it's physical address at @phys | |
1202 | * if *logical is the smallest allocated block, the function | |
1203 | * returns 0 at @phys | |
1204 | * return value contains 0 (success) or error code | |
1205 | */ | |
1f109d5a TT |
1206 | static int ext4_ext_search_left(struct inode *inode, |
1207 | struct ext4_ext_path *path, | |
1208 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1209 | { |
1210 | struct ext4_extent_idx *ix; | |
1211 | struct ext4_extent *ex; | |
b939e376 | 1212 | int depth, ee_len; |
1988b51e | 1213 | |
273df556 FM |
1214 | if (unlikely(path == NULL)) { |
1215 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1216 | return -EIO; | |
1217 | } | |
1988b51e AT |
1218 | depth = path->p_depth; |
1219 | *phys = 0; | |
1220 | ||
1221 | if (depth == 0 && path->p_ext == NULL) | |
1222 | return 0; | |
1223 | ||
1224 | /* usually extent in the path covers blocks smaller | |
1225 | * then *logical, but it can be that extent is the | |
1226 | * first one in the file */ | |
1227 | ||
1228 | ex = path[depth].p_ext; | |
b939e376 | 1229 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1230 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1231 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1232 | EXT4_ERROR_INODE(inode, | |
1233 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1234 | *logical, le32_to_cpu(ex->ee_block)); | |
1235 | return -EIO; | |
1236 | } | |
1988b51e AT |
1237 | while (--depth >= 0) { |
1238 | ix = path[depth].p_idx; | |
273df556 FM |
1239 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1240 | EXT4_ERROR_INODE(inode, | |
1241 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
1242 | ix != NULL ? ix->ei_block : 0, | |
1243 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? | |
1244 | EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0, | |
1245 | depth); | |
1246 | return -EIO; | |
1247 | } | |
1988b51e AT |
1248 | } |
1249 | return 0; | |
1250 | } | |
1251 | ||
273df556 FM |
1252 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1253 | EXT4_ERROR_INODE(inode, | |
1254 | "logical %d < ee_block %d + ee_len %d!", | |
1255 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1256 | return -EIO; | |
1257 | } | |
1988b51e | 1258 | |
b939e376 | 1259 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1260 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1261 | return 0; |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * search the closest allocated block to the right for *logical | |
1266 | * and returns it at @logical + it's physical address at @phys | |
1267 | * if *logical is the smallest allocated block, the function | |
1268 | * returns 0 at @phys | |
1269 | * return value contains 0 (success) or error code | |
1270 | */ | |
1f109d5a TT |
1271 | static int ext4_ext_search_right(struct inode *inode, |
1272 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1273 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1274 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1275 | { |
1276 | struct buffer_head *bh = NULL; | |
1277 | struct ext4_extent_header *eh; | |
1278 | struct ext4_extent_idx *ix; | |
1279 | struct ext4_extent *ex; | |
1280 | ext4_fsblk_t block; | |
395a87bf ES |
1281 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1282 | int ee_len; | |
1988b51e | 1283 | |
273df556 FM |
1284 | if (unlikely(path == NULL)) { |
1285 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1286 | return -EIO; | |
1287 | } | |
1988b51e AT |
1288 | depth = path->p_depth; |
1289 | *phys = 0; | |
1290 | ||
1291 | if (depth == 0 && path->p_ext == NULL) | |
1292 | return 0; | |
1293 | ||
1294 | /* usually extent in the path covers blocks smaller | |
1295 | * then *logical, but it can be that extent is the | |
1296 | * first one in the file */ | |
1297 | ||
1298 | ex = path[depth].p_ext; | |
b939e376 | 1299 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1300 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1301 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1302 | EXT4_ERROR_INODE(inode, | |
1303 | "first_extent(path[%d].p_hdr) != ex", | |
1304 | depth); | |
1305 | return -EIO; | |
1306 | } | |
1988b51e AT |
1307 | while (--depth >= 0) { |
1308 | ix = path[depth].p_idx; | |
273df556 FM |
1309 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1310 | EXT4_ERROR_INODE(inode, | |
1311 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1312 | *logical); | |
1313 | return -EIO; | |
1314 | } | |
1988b51e | 1315 | } |
4d33b1ef | 1316 | goto found_extent; |
1988b51e AT |
1317 | } |
1318 | ||
273df556 FM |
1319 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1320 | EXT4_ERROR_INODE(inode, | |
1321 | "logical %d < ee_block %d + ee_len %d!", | |
1322 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1323 | return -EIO; | |
1324 | } | |
1988b51e AT |
1325 | |
1326 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1327 | /* next allocated block in this leaf */ | |
1328 | ex++; | |
4d33b1ef | 1329 | goto found_extent; |
1988b51e AT |
1330 | } |
1331 | ||
1332 | /* go up and search for index to the right */ | |
1333 | while (--depth >= 0) { | |
1334 | ix = path[depth].p_idx; | |
1335 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1336 | goto got_index; |
1988b51e AT |
1337 | } |
1338 | ||
25f1ee3a WF |
1339 | /* we've gone up to the root and found no index to the right */ |
1340 | return 0; | |
1988b51e | 1341 | |
25f1ee3a | 1342 | got_index: |
1988b51e AT |
1343 | /* we've found index to the right, let's |
1344 | * follow it and find the closest allocated | |
1345 | * block to the right */ | |
1346 | ix++; | |
bf89d16f | 1347 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1348 | while (++depth < path->p_depth) { |
1349 | bh = sb_bread(inode->i_sb, block); | |
1350 | if (bh == NULL) | |
1351 | return -EIO; | |
1352 | eh = ext_block_hdr(bh); | |
395a87bf | 1353 | /* subtract from p_depth to get proper eh_depth */ |
56b19868 | 1354 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1355 | put_bh(bh); |
1356 | return -EIO; | |
1357 | } | |
1358 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1359 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1360 | put_bh(bh); |
1361 | } | |
1362 | ||
1363 | bh = sb_bread(inode->i_sb, block); | |
1364 | if (bh == NULL) | |
1365 | return -EIO; | |
1366 | eh = ext_block_hdr(bh); | |
56b19868 | 1367 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1368 | put_bh(bh); |
1369 | return -EIO; | |
1370 | } | |
1371 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1372 | found_extent: |
1988b51e | 1373 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1374 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1375 | *ret_ex = ex; |
1376 | if (bh) | |
1377 | put_bh(bh); | |
1988b51e | 1378 | return 0; |
1988b51e AT |
1379 | } |
1380 | ||
a86c6181 | 1381 | /* |
d0d856e8 | 1382 | * ext4_ext_next_allocated_block: |
f17722f9 | 1383 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1384 | * NOTE: it considers block number from index entry as |
1385 | * allocated block. Thus, index entries have to be consistent | |
1386 | * with leaves. | |
a86c6181 | 1387 | */ |
725d26d3 | 1388 | static ext4_lblk_t |
a86c6181 AT |
1389 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1390 | { | |
1391 | int depth; | |
1392 | ||
1393 | BUG_ON(path == NULL); | |
1394 | depth = path->p_depth; | |
1395 | ||
1396 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1397 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1398 | |
1399 | while (depth >= 0) { | |
1400 | if (depth == path->p_depth) { | |
1401 | /* leaf */ | |
1402 | if (path[depth].p_ext != | |
1403 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1404 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1405 | } else { | |
1406 | /* index */ | |
1407 | if (path[depth].p_idx != | |
1408 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1409 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1410 | } | |
1411 | depth--; | |
1412 | } | |
1413 | ||
f17722f9 | 1414 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1415 | } |
1416 | ||
1417 | /* | |
d0d856e8 | 1418 | * ext4_ext_next_leaf_block: |
f17722f9 | 1419 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1420 | */ |
5718789d | 1421 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1422 | { |
1423 | int depth; | |
1424 | ||
1425 | BUG_ON(path == NULL); | |
1426 | depth = path->p_depth; | |
1427 | ||
1428 | /* zero-tree has no leaf blocks at all */ | |
1429 | if (depth == 0) | |
f17722f9 | 1430 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1431 | |
1432 | /* go to index block */ | |
1433 | depth--; | |
1434 | ||
1435 | while (depth >= 0) { | |
1436 | if (path[depth].p_idx != | |
1437 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1438 | return (ext4_lblk_t) |
1439 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1440 | depth--; |
1441 | } | |
1442 | ||
f17722f9 | 1443 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1444 | } |
1445 | ||
1446 | /* | |
d0d856e8 RD |
1447 | * ext4_ext_correct_indexes: |
1448 | * if leaf gets modified and modified extent is first in the leaf, | |
1449 | * then we have to correct all indexes above. | |
a86c6181 AT |
1450 | * TODO: do we need to correct tree in all cases? |
1451 | */ | |
1d03ec98 | 1452 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1453 | struct ext4_ext_path *path) |
1454 | { | |
1455 | struct ext4_extent_header *eh; | |
1456 | int depth = ext_depth(inode); | |
1457 | struct ext4_extent *ex; | |
1458 | __le32 border; | |
1459 | int k, err = 0; | |
1460 | ||
1461 | eh = path[depth].p_hdr; | |
1462 | ex = path[depth].p_ext; | |
273df556 FM |
1463 | |
1464 | if (unlikely(ex == NULL || eh == NULL)) { | |
1465 | EXT4_ERROR_INODE(inode, | |
1466 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1467 | return -EIO; | |
1468 | } | |
a86c6181 AT |
1469 | |
1470 | if (depth == 0) { | |
1471 | /* there is no tree at all */ | |
1472 | return 0; | |
1473 | } | |
1474 | ||
1475 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1476 | /* we correct tree if first leaf got modified only */ | |
1477 | return 0; | |
1478 | } | |
1479 | ||
1480 | /* | |
d0d856e8 | 1481 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1482 | */ |
1483 | k = depth - 1; | |
1484 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1485 | err = ext4_ext_get_access(handle, inode, path + k); |
1486 | if (err) | |
a86c6181 AT |
1487 | return err; |
1488 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1489 | err = ext4_ext_dirty(handle, inode, path + k); |
1490 | if (err) | |
a86c6181 AT |
1491 | return err; |
1492 | ||
1493 | while (k--) { | |
1494 | /* change all left-side indexes */ | |
1495 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1496 | break; | |
7e028976 AM |
1497 | err = ext4_ext_get_access(handle, inode, path + k); |
1498 | if (err) | |
a86c6181 AT |
1499 | break; |
1500 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1501 | err = ext4_ext_dirty(handle, inode, path + k); |
1502 | if (err) | |
a86c6181 AT |
1503 | break; |
1504 | } | |
1505 | ||
1506 | return err; | |
1507 | } | |
1508 | ||
748de673 | 1509 | int |
a86c6181 AT |
1510 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1511 | struct ext4_extent *ex2) | |
1512 | { | |
749269fa | 1513 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1514 | |
1515 | /* | |
1516 | * Make sure that either both extents are uninitialized, or | |
1517 | * both are _not_. | |
1518 | */ | |
1519 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1520 | return 0; | |
1521 | ||
749269fa AA |
1522 | if (ext4_ext_is_uninitialized(ex1)) |
1523 | max_len = EXT_UNINIT_MAX_LEN; | |
1524 | else | |
1525 | max_len = EXT_INIT_MAX_LEN; | |
1526 | ||
a2df2a63 AA |
1527 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1528 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1529 | ||
1530 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1531 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1532 | return 0; |
1533 | ||
471d4011 SB |
1534 | /* |
1535 | * To allow future support for preallocated extents to be added | |
1536 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1537 | * this can result in the top bit of ee_len being set. |
471d4011 | 1538 | */ |
749269fa | 1539 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1540 | return 0; |
bbf2f9fb | 1541 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1542 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1543 | return 0; |
1544 | #endif | |
1545 | ||
bf89d16f | 1546 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1547 | return 1; |
1548 | return 0; | |
1549 | } | |
1550 | ||
56055d3a AA |
1551 | /* |
1552 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1553 | * It always tries to merge towards right. If you want to merge towards | |
1554 | * left, pass "ex - 1" as argument instead of "ex". | |
1555 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1556 | * 1 if they got merged. | |
1557 | */ | |
197217a5 | 1558 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1559 | struct ext4_ext_path *path, |
1560 | struct ext4_extent *ex) | |
56055d3a AA |
1561 | { |
1562 | struct ext4_extent_header *eh; | |
1563 | unsigned int depth, len; | |
1564 | int merge_done = 0; | |
1565 | int uninitialized = 0; | |
1566 | ||
1567 | depth = ext_depth(inode); | |
1568 | BUG_ON(path[depth].p_hdr == NULL); | |
1569 | eh = path[depth].p_hdr; | |
1570 | ||
1571 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1572 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1573 | break; | |
1574 | /* merge with next extent! */ | |
1575 | if (ext4_ext_is_uninitialized(ex)) | |
1576 | uninitialized = 1; | |
1577 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1578 | + ext4_ext_get_actual_len(ex + 1)); | |
1579 | if (uninitialized) | |
1580 | ext4_ext_mark_uninitialized(ex); | |
1581 | ||
1582 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1583 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1584 | * sizeof(struct ext4_extent); | |
1585 | memmove(ex + 1, ex + 2, len); | |
1586 | } | |
e8546d06 | 1587 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1588 | merge_done = 1; |
1589 | WARN_ON(eh->eh_entries == 0); | |
1590 | if (!eh->eh_entries) | |
24676da4 | 1591 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1592 | } |
1593 | ||
1594 | return merge_done; | |
1595 | } | |
1596 | ||
197217a5 YY |
1597 | /* |
1598 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1599 | * return 1 if merge left else 0. | |
1600 | */ | |
1601 | static int ext4_ext_try_to_merge(struct inode *inode, | |
1602 | struct ext4_ext_path *path, | |
1603 | struct ext4_extent *ex) { | |
1604 | struct ext4_extent_header *eh; | |
1605 | unsigned int depth; | |
1606 | int merge_done = 0; | |
1607 | int ret = 0; | |
1608 | ||
1609 | depth = ext_depth(inode); | |
1610 | BUG_ON(path[depth].p_hdr == NULL); | |
1611 | eh = path[depth].p_hdr; | |
1612 | ||
1613 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1614 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1615 | ||
1616 | if (!merge_done) | |
1617 | ret = ext4_ext_try_to_merge_right(inode, path, ex); | |
1618 | ||
1619 | return ret; | |
1620 | } | |
1621 | ||
25d14f98 AA |
1622 | /* |
1623 | * check if a portion of the "newext" extent overlaps with an | |
1624 | * existing extent. | |
1625 | * | |
1626 | * If there is an overlap discovered, it updates the length of the newext | |
1627 | * such that there will be no overlap, and then returns 1. | |
1628 | * If there is no overlap found, it returns 0. | |
1629 | */ | |
4d33b1ef TT |
1630 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1631 | struct inode *inode, | |
1f109d5a TT |
1632 | struct ext4_extent *newext, |
1633 | struct ext4_ext_path *path) | |
25d14f98 | 1634 | { |
725d26d3 | 1635 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1636 | unsigned int depth, len1; |
1637 | unsigned int ret = 0; | |
1638 | ||
1639 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1640 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1641 | depth = ext_depth(inode); |
1642 | if (!path[depth].p_ext) | |
1643 | goto out; | |
1644 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1645 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1646 | |
1647 | /* | |
1648 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1649 | * is before the requested block(s) |
25d14f98 AA |
1650 | */ |
1651 | if (b2 < b1) { | |
1652 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1653 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1654 | goto out; |
4d33b1ef | 1655 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1656 | } |
1657 | ||
725d26d3 | 1658 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1659 | if (b1 + len1 < b1) { |
f17722f9 | 1660 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1661 | newext->ee_len = cpu_to_le16(len1); |
1662 | ret = 1; | |
1663 | } | |
1664 | ||
1665 | /* check for overlap */ | |
1666 | if (b1 + len1 > b2) { | |
1667 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1668 | ret = 1; | |
1669 | } | |
1670 | out: | |
1671 | return ret; | |
1672 | } | |
1673 | ||
a86c6181 | 1674 | /* |
d0d856e8 RD |
1675 | * ext4_ext_insert_extent: |
1676 | * tries to merge requsted extent into the existing extent or | |
1677 | * inserts requested extent as new one into the tree, | |
1678 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1679 | */ |
1680 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1681 | struct ext4_ext_path *path, | |
0031462b | 1682 | struct ext4_extent *newext, int flag) |
a86c6181 | 1683 | { |
af5bc92d | 1684 | struct ext4_extent_header *eh; |
a86c6181 AT |
1685 | struct ext4_extent *ex, *fex; |
1686 | struct ext4_extent *nearex; /* nearest extent */ | |
1687 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1688 | int depth, len, err; |
1689 | ext4_lblk_t next; | |
a2df2a63 | 1690 | unsigned uninitialized = 0; |
55f020db | 1691 | int flags = 0; |
a86c6181 | 1692 | |
273df556 FM |
1693 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1694 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1695 | return -EIO; | |
1696 | } | |
a86c6181 AT |
1697 | depth = ext_depth(inode); |
1698 | ex = path[depth].p_ext; | |
273df556 FM |
1699 | if (unlikely(path[depth].p_hdr == NULL)) { |
1700 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1701 | return -EIO; | |
1702 | } | |
a86c6181 AT |
1703 | |
1704 | /* try to insert block into found extent and return */ | |
744692dc | 1705 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1706 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
553f9008 | 1707 | ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", |
bf89d16f TT |
1708 | ext4_ext_is_uninitialized(newext), |
1709 | ext4_ext_get_actual_len(newext), | |
1710 | le32_to_cpu(ex->ee_block), | |
1711 | ext4_ext_is_uninitialized(ex), | |
1712 | ext4_ext_get_actual_len(ex), | |
1713 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1714 | err = ext4_ext_get_access(handle, inode, path + depth); |
1715 | if (err) | |
a86c6181 | 1716 | return err; |
a2df2a63 AA |
1717 | |
1718 | /* | |
1719 | * ext4_can_extents_be_merged should have checked that either | |
1720 | * both extents are uninitialized, or both aren't. Thus we | |
1721 | * need to check only one of them here. | |
1722 | */ | |
1723 | if (ext4_ext_is_uninitialized(ex)) | |
1724 | uninitialized = 1; | |
1725 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1726 | + ext4_ext_get_actual_len(newext)); | |
1727 | if (uninitialized) | |
1728 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1729 | eh = path[depth].p_hdr; |
1730 | nearex = ex; | |
1731 | goto merge; | |
1732 | } | |
1733 | ||
a86c6181 AT |
1734 | depth = ext_depth(inode); |
1735 | eh = path[depth].p_hdr; | |
1736 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1737 | goto has_space; | |
1738 | ||
1739 | /* probably next leaf has space for us? */ | |
1740 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1741 | next = EXT_MAX_BLOCKS; |
1742 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1743 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1744 | if (next != EXT_MAX_BLOCKS) { |
a86c6181 AT |
1745 | ext_debug("next leaf block - %d\n", next); |
1746 | BUG_ON(npath != NULL); | |
1747 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1748 | if (IS_ERR(npath)) | |
1749 | return PTR_ERR(npath); | |
1750 | BUG_ON(npath->p_depth != path->p_depth); | |
1751 | eh = npath[depth].p_hdr; | |
1752 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1753 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1754 | le16_to_cpu(eh->eh_entries)); |
1755 | path = npath; | |
ffb505ff | 1756 | goto has_space; |
a86c6181 AT |
1757 | } |
1758 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1759 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1760 | } | |
1761 | ||
1762 | /* | |
d0d856e8 RD |
1763 | * There is no free space in the found leaf. |
1764 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1765 | */ |
55f020db AH |
1766 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1767 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1768 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1769 | if (err) |
1770 | goto cleanup; | |
1771 | depth = ext_depth(inode); | |
1772 | eh = path[depth].p_hdr; | |
1773 | ||
1774 | has_space: | |
1775 | nearex = path[depth].p_ext; | |
1776 | ||
7e028976 AM |
1777 | err = ext4_ext_get_access(handle, inode, path + depth); |
1778 | if (err) | |
a86c6181 AT |
1779 | goto cleanup; |
1780 | ||
1781 | if (!nearex) { | |
1782 | /* there is no extent in this leaf, create first one */ | |
553f9008 | 1783 | ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n", |
8c55e204 | 1784 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1785 | ext4_ext_pblock(newext), |
553f9008 | 1786 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1787 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1788 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1789 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1790 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1791 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1792 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1793 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1794 | len = (len - 1) * sizeof(struct ext4_extent); | |
1795 | len = len < 0 ? 0 : len; | |
553f9008 | 1796 | ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " |
a86c6181 | 1797 | "move %d from 0x%p to 0x%p\n", |
8c55e204 | 1798 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1799 | ext4_ext_pblock(newext), |
553f9008 | 1800 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1801 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1802 | nearex, len, nearex + 1, nearex + 2); |
1803 | memmove(nearex + 2, nearex + 1, len); | |
1804 | } | |
1805 | path[depth].p_ext = nearex + 1; | |
1806 | } else { | |
1807 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1808 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1809 | len = len < 0 ? 0 : len; | |
553f9008 | 1810 | ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " |
a86c6181 AT |
1811 | "move %d from 0x%p to 0x%p\n", |
1812 | le32_to_cpu(newext->ee_block), | |
bf89d16f | 1813 | ext4_ext_pblock(newext), |
553f9008 | 1814 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1815 | ext4_ext_get_actual_len(newext), |
0737964b | 1816 | nearex, len, nearex, nearex + 1); |
a86c6181 AT |
1817 | memmove(nearex + 1, nearex, len); |
1818 | path[depth].p_ext = nearex; | |
1819 | } | |
1820 | ||
e8546d06 | 1821 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1822 | nearex = path[depth].p_ext; |
1823 | nearex->ee_block = newext->ee_block; | |
bf89d16f | 1824 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1825 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1826 | |
1827 | merge: | |
1828 | /* try to merge extents to the right */ | |
744692dc | 1829 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
0031462b | 1830 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1831 | |
1832 | /* try to merge extents to the left */ | |
1833 | ||
1834 | /* time to correct all indexes above */ | |
1835 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1836 | if (err) | |
1837 | goto cleanup; | |
1838 | ||
1839 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1840 | ||
1841 | cleanup: | |
1842 | if (npath) { | |
1843 | ext4_ext_drop_refs(npath); | |
1844 | kfree(npath); | |
1845 | } | |
a86c6181 AT |
1846 | ext4_ext_invalidate_cache(inode); |
1847 | return err; | |
1848 | } | |
1849 | ||
1f109d5a TT |
1850 | static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1851 | ext4_lblk_t num, ext_prepare_callback func, | |
1852 | void *cbdata) | |
6873fa0d ES |
1853 | { |
1854 | struct ext4_ext_path *path = NULL; | |
1855 | struct ext4_ext_cache cbex; | |
1856 | struct ext4_extent *ex; | |
1857 | ext4_lblk_t next, start = 0, end = 0; | |
1858 | ext4_lblk_t last = block + num; | |
1859 | int depth, exists, err = 0; | |
1860 | ||
1861 | BUG_ON(func == NULL); | |
1862 | BUG_ON(inode == NULL); | |
1863 | ||
f17722f9 | 1864 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1865 | num = last - block; |
1866 | /* find extent for this block */ | |
fab3a549 | 1867 | down_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d | 1868 | path = ext4_ext_find_extent(inode, block, path); |
fab3a549 | 1869 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1870 | if (IS_ERR(path)) { |
1871 | err = PTR_ERR(path); | |
1872 | path = NULL; | |
1873 | break; | |
1874 | } | |
1875 | ||
1876 | depth = ext_depth(inode); | |
273df556 FM |
1877 | if (unlikely(path[depth].p_hdr == NULL)) { |
1878 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1879 | err = -EIO; | |
1880 | break; | |
1881 | } | |
6873fa0d ES |
1882 | ex = path[depth].p_ext; |
1883 | next = ext4_ext_next_allocated_block(path); | |
1884 | ||
1885 | exists = 0; | |
1886 | if (!ex) { | |
1887 | /* there is no extent yet, so try to allocate | |
1888 | * all requested space */ | |
1889 | start = block; | |
1890 | end = block + num; | |
1891 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1892 | /* need to allocate space before found extent */ | |
1893 | start = block; | |
1894 | end = le32_to_cpu(ex->ee_block); | |
1895 | if (block + num < end) | |
1896 | end = block + num; | |
1897 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1898 | + ext4_ext_get_actual_len(ex)) { | |
1899 | /* need to allocate space after found extent */ | |
1900 | start = block; | |
1901 | end = block + num; | |
1902 | if (end >= next) | |
1903 | end = next; | |
1904 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1905 | /* | |
1906 | * some part of requested space is covered | |
1907 | * by found extent | |
1908 | */ | |
1909 | start = block; | |
1910 | end = le32_to_cpu(ex->ee_block) | |
1911 | + ext4_ext_get_actual_len(ex); | |
1912 | if (block + num < end) | |
1913 | end = block + num; | |
1914 | exists = 1; | |
1915 | } else { | |
1916 | BUG(); | |
1917 | } | |
1918 | BUG_ON(end <= start); | |
1919 | ||
1920 | if (!exists) { | |
1921 | cbex.ec_block = start; | |
1922 | cbex.ec_len = end - start; | |
1923 | cbex.ec_start = 0; | |
6873fa0d ES |
1924 | } else { |
1925 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1926 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
bf89d16f | 1927 | cbex.ec_start = ext4_ext_pblock(ex); |
6873fa0d ES |
1928 | } |
1929 | ||
273df556 FM |
1930 | if (unlikely(cbex.ec_len == 0)) { |
1931 | EXT4_ERROR_INODE(inode, "cbex.ec_len == 0"); | |
1932 | err = -EIO; | |
1933 | break; | |
1934 | } | |
c03f8aa9 | 1935 | err = func(inode, next, &cbex, ex, cbdata); |
6873fa0d ES |
1936 | ext4_ext_drop_refs(path); |
1937 | ||
1938 | if (err < 0) | |
1939 | break; | |
1940 | ||
1941 | if (err == EXT_REPEAT) | |
1942 | continue; | |
1943 | else if (err == EXT_BREAK) { | |
1944 | err = 0; | |
1945 | break; | |
1946 | } | |
1947 | ||
1948 | if (ext_depth(inode) != depth) { | |
1949 | /* depth was changed. we have to realloc path */ | |
1950 | kfree(path); | |
1951 | path = NULL; | |
1952 | } | |
1953 | ||
1954 | block = cbex.ec_block + cbex.ec_len; | |
1955 | } | |
1956 | ||
1957 | if (path) { | |
1958 | ext4_ext_drop_refs(path); | |
1959 | kfree(path); | |
1960 | } | |
1961 | ||
1962 | return err; | |
1963 | } | |
1964 | ||
09b88252 | 1965 | static void |
725d26d3 | 1966 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 1967 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
1968 | { |
1969 | struct ext4_ext_cache *cex; | |
1970 | BUG_ON(len == 0); | |
2ec0ae3a | 1971 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
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++; | |
2ec0ae3a TT |
2073 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2074 | return ret; | |
a86c6181 AT |
2075 | } |
2076 | ||
a4bb6b64 AH |
2077 | /* |
2078 | * ext4_ext_in_cache() | |
2079 | * Checks to see if the given block is in the cache. | |
2080 | * If it is, the cached extent is stored in the given | |
2081 | * extent pointer. | |
2082 | * | |
2083 | * @inode: The files inode | |
2084 | * @block: The block to look for in the cache | |
2085 | * @ex: Pointer where the cached extent will be stored | |
2086 | * if it contains block | |
2087 | * | |
2088 | * Return 0 if cache is invalid; 1 if the cache is valid | |
2089 | */ | |
2090 | static int | |
2091 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, | |
2092 | struct ext4_extent *ex) | |
2093 | { | |
2094 | struct ext4_ext_cache cex; | |
2095 | int ret = 0; | |
2096 | ||
2097 | if (ext4_ext_check_cache(inode, block, &cex)) { | |
2098 | ex->ee_block = cpu_to_le32(cex.ec_block); | |
2099 | ext4_ext_store_pblock(ex, cex.ec_start); | |
2100 | ex->ee_len = cpu_to_le16(cex.ec_len); | |
2101 | ret = 1; | |
2102 | } | |
2103 | ||
2104 | return ret; | |
2105 | } | |
2106 | ||
2107 | ||
a86c6181 | 2108 | /* |
d0d856e8 RD |
2109 | * ext4_ext_rm_idx: |
2110 | * removes index from the index block. | |
a86c6181 | 2111 | */ |
1d03ec98 | 2112 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
2113 | struct ext4_ext_path *path) |
2114 | { | |
a86c6181 | 2115 | int err; |
f65e6fba | 2116 | ext4_fsblk_t leaf; |
a86c6181 AT |
2117 | |
2118 | /* free index block */ | |
2119 | path--; | |
bf89d16f | 2120 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2121 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2122 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2123 | return -EIO; | |
2124 | } | |
7e028976 AM |
2125 | err = ext4_ext_get_access(handle, inode, path); |
2126 | if (err) | |
a86c6181 | 2127 | return err; |
0e1147b0 RD |
2128 | |
2129 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2130 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2131 | len *= sizeof(struct ext4_extent_idx); | |
2132 | memmove(path->p_idx, path->p_idx + 1, len); | |
2133 | } | |
2134 | ||
e8546d06 | 2135 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2136 | err = ext4_ext_dirty(handle, inode, path); |
2137 | if (err) | |
a86c6181 | 2138 | return err; |
2ae02107 | 2139 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
7dc57615 | 2140 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2141 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
a86c6181 AT |
2142 | return err; |
2143 | } | |
2144 | ||
2145 | /* | |
ee12b630 MC |
2146 | * ext4_ext_calc_credits_for_single_extent: |
2147 | * This routine returns max. credits that needed to insert an extent | |
2148 | * to the extent tree. | |
2149 | * When pass the actual path, the caller should calculate credits | |
2150 | * under i_data_sem. | |
a86c6181 | 2151 | */ |
525f4ed8 | 2152 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2153 | struct ext4_ext_path *path) |
2154 | { | |
a86c6181 | 2155 | if (path) { |
ee12b630 | 2156 | int depth = ext_depth(inode); |
f3bd1f3f | 2157 | int ret = 0; |
ee12b630 | 2158 | |
a86c6181 | 2159 | /* probably there is space in leaf? */ |
a86c6181 | 2160 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2161 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2162 | |
ee12b630 MC |
2163 | /* |
2164 | * There are some space in the leaf tree, no | |
2165 | * need to account for leaf block credit | |
2166 | * | |
2167 | * bitmaps and block group descriptor blocks | |
2168 | * and other metadat blocks still need to be | |
2169 | * accounted. | |
2170 | */ | |
525f4ed8 | 2171 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2172 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2173 | return ret; |
ee12b630 MC |
2174 | } |
2175 | } | |
a86c6181 | 2176 | |
525f4ed8 | 2177 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2178 | } |
a86c6181 | 2179 | |
ee12b630 MC |
2180 | /* |
2181 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2182 | * | |
2183 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2184 | * in the worse case, each tree level index/leaf need to be changed | |
2185 | * if the tree split due to insert a new extent, then the old tree | |
2186 | * index/leaf need to be updated too | |
2187 | * | |
2188 | * If the nrblocks are discontiguous, they could cause | |
2189 | * the whole tree split more than once, but this is really rare. | |
2190 | */ | |
525f4ed8 | 2191 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2192 | { |
2193 | int index; | |
2194 | int depth = ext_depth(inode); | |
a86c6181 | 2195 | |
ee12b630 MC |
2196 | if (chunk) |
2197 | index = depth * 2; | |
2198 | else | |
2199 | index = depth * 3; | |
a86c6181 | 2200 | |
ee12b630 | 2201 | return index; |
a86c6181 AT |
2202 | } |
2203 | ||
2204 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2205 | struct ext4_extent *ex, |
2206 | ext4_fsblk_t *partial_cluster, | |
2207 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2208 | { |
0aa06000 | 2209 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2210 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2211 | ext4_fsblk_t pblk; |
e6362609 | 2212 | int flags = EXT4_FREE_BLOCKS_FORGET; |
a86c6181 | 2213 | |
c9de560d | 2214 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
e6362609 | 2215 | flags |= EXT4_FREE_BLOCKS_METADATA; |
0aa06000 TT |
2216 | /* |
2217 | * For bigalloc file systems, we never free a partial cluster | |
2218 | * at the beginning of the extent. Instead, we make a note | |
2219 | * that we tried freeing the cluster, and check to see if we | |
2220 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2221 | * or at the end of the ext4_truncate() operation. | |
2222 | */ | |
2223 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2224 | ||
2225 | /* | |
2226 | * If we have a partial cluster, and it's different from the | |
2227 | * cluster of the last block, we need to explicitly free the | |
2228 | * partial cluster here. | |
2229 | */ | |
2230 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2231 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2232 | ext4_free_blocks(handle, inode, NULL, | |
2233 | EXT4_C2B(sbi, *partial_cluster), | |
2234 | sbi->s_cluster_ratio, flags); | |
2235 | *partial_cluster = 0; | |
2236 | } | |
2237 | ||
a86c6181 AT |
2238 | #ifdef EXTENTS_STATS |
2239 | { | |
2240 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2241 | spin_lock(&sbi->s_ext_stats_lock); |
2242 | sbi->s_ext_blocks += ee_len; | |
2243 | sbi->s_ext_extents++; | |
2244 | if (ee_len < sbi->s_ext_min) | |
2245 | sbi->s_ext_min = ee_len; | |
2246 | if (ee_len > sbi->s_ext_max) | |
2247 | sbi->s_ext_max = ee_len; | |
2248 | if (ext_depth(inode) > sbi->s_depth_max) | |
2249 | sbi->s_depth_max = ext_depth(inode); | |
2250 | spin_unlock(&sbi->s_ext_stats_lock); | |
2251 | } | |
2252 | #endif | |
2253 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2254 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2255 | /* tail removal */ |
725d26d3 | 2256 | ext4_lblk_t num; |
725d26d3 | 2257 | |
a2df2a63 | 2258 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2259 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2260 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2261 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2262 | /* | |
2263 | * If the block range to be freed didn't start at the | |
2264 | * beginning of a cluster, and we removed the entire | |
2265 | * extent, save the partial cluster here, since we | |
2266 | * might need to delete if we determine that the | |
2267 | * truncate operation has removed all of the blocks in | |
2268 | * the cluster. | |
2269 | */ | |
2270 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2271 | (ee_len == num)) | |
2272 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2273 | else | |
2274 | *partial_cluster = 0; | |
a86c6181 | 2275 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2276 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2277 | /* head removal */ |
2278 | ext4_lblk_t num; | |
2279 | ext4_fsblk_t start; | |
2280 | ||
2281 | num = to - from; | |
2282 | start = ext4_ext_pblock(ex); | |
2283 | ||
2284 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
2285 | ext4_free_blocks(handle, inode, 0, start, num, flags); | |
2286 | ||
a86c6181 | 2287 | } else { |
725d26d3 AK |
2288 | printk(KERN_INFO "strange request: removal(2) " |
2289 | "%u-%u from %u:%u\n", | |
2290 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2291 | } |
2292 | return 0; | |
2293 | } | |
2294 | ||
d583fb87 AH |
2295 | |
2296 | /* | |
2297 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2298 | * blocks appearing between "start" and "end", and splits the extents | |
2299 | * if "start" and "end" appear in the same extent | |
2300 | * | |
2301 | * @handle: The journal handle | |
2302 | * @inode: The files inode | |
2303 | * @path: The path to the leaf | |
2304 | * @start: The first block to remove | |
2305 | * @end: The last block to remove | |
2306 | */ | |
a86c6181 AT |
2307 | static int |
2308 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2309 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2310 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2311 | { |
0aa06000 | 2312 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2313 | int err = 0, correct_index = 0; |
2314 | int depth = ext_depth(inode), credits; | |
2315 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2316 | ext4_lblk_t a, b, block; |
2317 | unsigned num; | |
2318 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2319 | unsigned short ex_ee_len; |
a2df2a63 | 2320 | unsigned uninitialized = 0; |
a86c6181 | 2321 | struct ext4_extent *ex; |
d583fb87 | 2322 | struct ext4_map_blocks map; |
a86c6181 | 2323 | |
c29c0ae7 | 2324 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 2325 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
2326 | if (!path[depth].p_hdr) |
2327 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2328 | eh = path[depth].p_hdr; | |
273df556 FM |
2329 | if (unlikely(path[depth].p_hdr == NULL)) { |
2330 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2331 | return -EIO; | |
2332 | } | |
a86c6181 AT |
2333 | /* find where to start removing */ |
2334 | ex = EXT_LAST_EXTENT(eh); | |
2335 | ||
2336 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2337 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2338 | |
2339 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
2340 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2341 | |
2342 | if (ext4_ext_is_uninitialized(ex)) | |
2343 | uninitialized = 1; | |
2344 | else | |
2345 | uninitialized = 0; | |
2346 | ||
553f9008 M |
2347 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2348 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2349 | path[depth].p_ext = ex; |
2350 | ||
2351 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2352 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2353 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2354 | |
2355 | ext_debug(" border %u:%u\n", a, b); | |
2356 | ||
d583fb87 AH |
2357 | /* If this extent is beyond the end of the hole, skip it */ |
2358 | if (end <= ex_ee_block) { | |
2359 | ex--; | |
2360 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2361 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2362 | continue; | |
2363 | } else if (a != ex_ee_block && | |
2364 | b != ex_ee_block + ex_ee_len - 1) { | |
2365 | /* | |
2366 | * If this is a truncate, then this condition should | |
2367 | * never happen because at least one of the end points | |
2368 | * needs to be on the edge of the extent. | |
2369 | */ | |
f17722f9 | 2370 | if (end == EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2371 | ext_debug(" bad truncate %u:%u\n", |
2372 | start, end); | |
2373 | block = 0; | |
2374 | num = 0; | |
2375 | err = -EIO; | |
2376 | goto out; | |
2377 | } | |
2378 | /* | |
2379 | * else this is a hole punch, so the extent needs to | |
2380 | * be split since neither edge of the hole is on the | |
2381 | * extent edge | |
2382 | */ | |
2383 | else{ | |
2384 | map.m_pblk = ext4_ext_pblock(ex); | |
2385 | map.m_lblk = ex_ee_block; | |
2386 | map.m_len = b - ex_ee_block; | |
2387 | ||
2388 | err = ext4_split_extent(handle, | |
2389 | inode, path, &map, 0, | |
2390 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT | | |
2391 | EXT4_GET_BLOCKS_PRE_IO); | |
2392 | ||
2393 | if (err < 0) | |
2394 | goto out; | |
2395 | ||
2396 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2397 | ||
2398 | b = ex_ee_block+ex_ee_len - 1 < end ? | |
2399 | ex_ee_block+ex_ee_len - 1 : end; | |
2400 | ||
2401 | /* Then remove tail of this extent */ | |
2402 | block = ex_ee_block; | |
2403 | num = a - block; | |
2404 | } | |
a86c6181 AT |
2405 | } else if (a != ex_ee_block) { |
2406 | /* remove tail of the extent */ | |
2407 | block = ex_ee_block; | |
2408 | num = a - block; | |
2409 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
2410 | /* remove head of the extent */ | |
d583fb87 AH |
2411 | block = b; |
2412 | num = ex_ee_block + ex_ee_len - b; | |
2413 | ||
2414 | /* | |
2415 | * If this is a truncate, this condition | |
2416 | * should never happen | |
2417 | */ | |
f17722f9 | 2418 | if (end == EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2419 | ext_debug(" bad truncate %u:%u\n", |
2420 | start, end); | |
2421 | err = -EIO; | |
2422 | goto out; | |
2423 | } | |
a86c6181 AT |
2424 | } else { |
2425 | /* remove whole extent: excellent! */ | |
2426 | block = ex_ee_block; | |
2427 | num = 0; | |
d583fb87 AH |
2428 | if (a != ex_ee_block) { |
2429 | ext_debug(" bad truncate %u:%u\n", | |
2430 | start, end); | |
2431 | err = -EIO; | |
2432 | goto out; | |
2433 | } | |
2434 | ||
2435 | if (b != ex_ee_block + ex_ee_len - 1) { | |
2436 | ext_debug(" bad truncate %u:%u\n", | |
2437 | start, end); | |
2438 | err = -EIO; | |
2439 | goto out; | |
2440 | } | |
a86c6181 AT |
2441 | } |
2442 | ||
34071da7 TT |
2443 | /* |
2444 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2445 | * descriptor) for each block group; assume two block | |
2446 | * groups plus ex_ee_len/blocks_per_block_group for | |
2447 | * the worst case | |
2448 | */ | |
2449 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2450 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2451 | correct_index = 1; | |
2452 | credits += (ext_depth(inode)) + 1; | |
2453 | } | |
5aca07eb | 2454 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2455 | |
487caeef | 2456 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2457 | if (err) |
a86c6181 | 2458 | goto out; |
a86c6181 AT |
2459 | |
2460 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2461 | if (err) | |
2462 | goto out; | |
2463 | ||
0aa06000 TT |
2464 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2465 | a, b); | |
a86c6181 AT |
2466 | if (err) |
2467 | goto out; | |
2468 | ||
2469 | if (num == 0) { | |
d0d856e8 | 2470 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2471 | ext4_ext_store_pblock(ex, 0); |
d583fb87 AH |
2472 | } else if (block != ex_ee_block) { |
2473 | /* | |
2474 | * If this was a head removal, then we need to update | |
2475 | * the physical block since it is now at a different | |
2476 | * location | |
2477 | */ | |
2478 | ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a)); | |
a86c6181 AT |
2479 | } |
2480 | ||
2481 | ex->ee_block = cpu_to_le32(block); | |
2482 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
2483 | /* |
2484 | * Do not mark uninitialized if all the blocks in the | |
2485 | * extent have been removed. | |
2486 | */ | |
2487 | if (uninitialized && num) | |
a2df2a63 | 2488 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
2489 | |
2490 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2491 | if (err) | |
2492 | goto out; | |
2493 | ||
d583fb87 AH |
2494 | /* |
2495 | * If the extent was completely released, | |
2496 | * we need to remove it from the leaf | |
2497 | */ | |
2498 | if (num == 0) { | |
f17722f9 | 2499 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2500 | /* |
2501 | * For hole punching, we need to scoot all the | |
2502 | * extents up when an extent is removed so that | |
2503 | * we dont have blank extents in the middle | |
2504 | */ | |
2505 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2506 | sizeof(struct ext4_extent)); | |
2507 | ||
2508 | /* Now get rid of the one at the end */ | |
2509 | memset(EXT_LAST_EXTENT(eh), 0, | |
2510 | sizeof(struct ext4_extent)); | |
2511 | } | |
2512 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2513 | } else |
2514 | *partial_cluster = 0; | |
d583fb87 | 2515 | |
2ae02107 | 2516 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
bf89d16f | 2517 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2518 | ex--; |
2519 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2520 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2521 | } |
2522 | ||
2523 | if (correct_index && eh->eh_entries) | |
2524 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2525 | ||
0aa06000 TT |
2526 | /* |
2527 | * If there is still a entry in the leaf node, check to see if | |
2528 | * it references the partial cluster. This is the only place | |
2529 | * where it could; if it doesn't, we can free the cluster. | |
2530 | */ | |
2531 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2532 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2533 | *partial_cluster)) { | |
2534 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2535 | ||
2536 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2537 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2538 | ||
2539 | ext4_free_blocks(handle, inode, NULL, | |
2540 | EXT4_C2B(sbi, *partial_cluster), | |
2541 | sbi->s_cluster_ratio, flags); | |
2542 | *partial_cluster = 0; | |
2543 | } | |
2544 | ||
a86c6181 AT |
2545 | /* if this leaf is free, then we should |
2546 | * remove it from index block above */ | |
2547 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2548 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2549 | ||
2550 | out: | |
2551 | return err; | |
2552 | } | |
2553 | ||
2554 | /* | |
d0d856e8 RD |
2555 | * ext4_ext_more_to_rm: |
2556 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2557 | */ |
09b88252 | 2558 | static int |
a86c6181 AT |
2559 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2560 | { | |
2561 | BUG_ON(path->p_idx == NULL); | |
2562 | ||
2563 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2564 | return 0; | |
2565 | ||
2566 | /* | |
d0d856e8 | 2567 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2568 | * so we have to consider current index for truncation |
2569 | */ | |
2570 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2571 | return 0; | |
2572 | return 1; | |
2573 | } | |
2574 | ||
c6a0371c | 2575 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
2576 | { |
2577 | struct super_block *sb = inode->i_sb; | |
2578 | int depth = ext_depth(inode); | |
2579 | struct ext4_ext_path *path; | |
0aa06000 | 2580 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2581 | handle_t *handle; |
0617b83f | 2582 | int i, err; |
a86c6181 | 2583 | |
725d26d3 | 2584 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
2585 | |
2586 | /* probably first extent we're gonna free will be last in block */ | |
2587 | handle = ext4_journal_start(inode, depth + 1); | |
2588 | if (IS_ERR(handle)) | |
2589 | return PTR_ERR(handle); | |
2590 | ||
0617b83f | 2591 | again: |
a86c6181 AT |
2592 | ext4_ext_invalidate_cache(inode); |
2593 | ||
2594 | /* | |
d0d856e8 RD |
2595 | * We start scanning from right side, freeing all the blocks |
2596 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2597 | */ |
0617b83f | 2598 | depth = ext_depth(inode); |
216553c4 | 2599 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
2600 | if (path == NULL) { |
2601 | ext4_journal_stop(handle); | |
2602 | return -ENOMEM; | |
2603 | } | |
0617b83f | 2604 | path[0].p_depth = depth; |
a86c6181 | 2605 | path[0].p_hdr = ext_inode_hdr(inode); |
56b19868 | 2606 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
2607 | err = -EIO; |
2608 | goto out; | |
2609 | } | |
0617b83f | 2610 | i = err = 0; |
a86c6181 AT |
2611 | |
2612 | while (i >= 0 && err == 0) { | |
2613 | if (i == depth) { | |
2614 | /* this is leaf block */ | |
d583fb87 | 2615 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 TT |
2616 | &partial_cluster, start, |
2617 | EXT_MAX_BLOCKS - 1); | |
d0d856e8 | 2618 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2619 | brelse(path[i].p_bh); |
2620 | path[i].p_bh = NULL; | |
2621 | i--; | |
2622 | continue; | |
2623 | } | |
2624 | ||
2625 | /* this is index block */ | |
2626 | if (!path[i].p_hdr) { | |
2627 | ext_debug("initialize header\n"); | |
2628 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2629 | } |
2630 | ||
a86c6181 | 2631 | if (!path[i].p_idx) { |
d0d856e8 | 2632 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2633 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2634 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2635 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2636 | path[i].p_hdr, | |
2637 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2638 | } else { | |
d0d856e8 | 2639 | /* we were already here, see at next index */ |
a86c6181 AT |
2640 | path[i].p_idx--; |
2641 | } | |
2642 | ||
2643 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2644 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2645 | path[i].p_idx); | |
2646 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2647 | struct buffer_head *bh; |
a86c6181 | 2648 | /* go to the next level */ |
2ae02107 | 2649 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2650 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2651 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2652 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2653 | if (!bh) { |
a86c6181 AT |
2654 | /* should we reset i_size? */ |
2655 | err = -EIO; | |
2656 | break; | |
2657 | } | |
c29c0ae7 AT |
2658 | if (WARN_ON(i + 1 > depth)) { |
2659 | err = -EIO; | |
2660 | break; | |
2661 | } | |
56b19868 | 2662 | if (ext4_ext_check(inode, ext_block_hdr(bh), |
c29c0ae7 AT |
2663 | depth - i - 1)) { |
2664 | err = -EIO; | |
2665 | break; | |
2666 | } | |
2667 | path[i + 1].p_bh = bh; | |
a86c6181 | 2668 | |
d0d856e8 RD |
2669 | /* save actual number of indexes since this |
2670 | * number is changed at the next iteration */ | |
a86c6181 AT |
2671 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2672 | i++; | |
2673 | } else { | |
d0d856e8 | 2674 | /* we finished processing this index, go up */ |
a86c6181 | 2675 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2676 | /* index is empty, remove it; |
a86c6181 AT |
2677 | * handle must be already prepared by the |
2678 | * truncatei_leaf() */ | |
2679 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2680 | } | |
d0d856e8 | 2681 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2682 | brelse(path[i].p_bh); |
2683 | path[i].p_bh = NULL; | |
2684 | i--; | |
2685 | ext_debug("return to level %d\n", i); | |
2686 | } | |
2687 | } | |
2688 | ||
2689 | /* TODO: flexible tree reduction should be here */ | |
2690 | if (path->p_hdr->eh_entries == 0) { | |
2691 | /* | |
d0d856e8 RD |
2692 | * truncate to zero freed all the tree, |
2693 | * so we need to correct eh_depth | |
a86c6181 AT |
2694 | */ |
2695 | err = ext4_ext_get_access(handle, inode, path); | |
2696 | if (err == 0) { | |
2697 | ext_inode_hdr(inode)->eh_depth = 0; | |
2698 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2699 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2700 | err = ext4_ext_dirty(handle, inode, path); |
2701 | } | |
2702 | } | |
2703 | out: | |
a86c6181 AT |
2704 | ext4_ext_drop_refs(path); |
2705 | kfree(path); | |
0617b83f DM |
2706 | if (err == -EAGAIN) |
2707 | goto again; | |
a86c6181 AT |
2708 | ext4_journal_stop(handle); |
2709 | ||
2710 | return err; | |
2711 | } | |
2712 | ||
2713 | /* | |
2714 | * called at mount time | |
2715 | */ | |
2716 | void ext4_ext_init(struct super_block *sb) | |
2717 | { | |
2718 | /* | |
2719 | * possible initialization would be here | |
2720 | */ | |
2721 | ||
83982b6f | 2722 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2723 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
4776004f | 2724 | printk(KERN_INFO "EXT4-fs: file extents enabled"); |
bbf2f9fb RD |
2725 | #ifdef AGGRESSIVE_TEST |
2726 | printk(", aggressive tests"); | |
a86c6181 AT |
2727 | #endif |
2728 | #ifdef CHECK_BINSEARCH | |
2729 | printk(", check binsearch"); | |
2730 | #endif | |
2731 | #ifdef EXTENTS_STATS | |
2732 | printk(", stats"); | |
2733 | #endif | |
2734 | printk("\n"); | |
90576c0b | 2735 | #endif |
a86c6181 AT |
2736 | #ifdef EXTENTS_STATS |
2737 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2738 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2739 | EXT4_SB(sb)->s_ext_max = 0; | |
2740 | #endif | |
2741 | } | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * called at umount time | |
2746 | */ | |
2747 | void ext4_ext_release(struct super_block *sb) | |
2748 | { | |
83982b6f | 2749 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2750 | return; |
2751 | ||
2752 | #ifdef EXTENTS_STATS | |
2753 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2754 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2755 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2756 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2757 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2758 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2759 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2760 | } | |
2761 | #endif | |
2762 | } | |
2763 | ||
093a088b AK |
2764 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2765 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2766 | { | |
2407518d LC |
2767 | ext4_fsblk_t ee_pblock; |
2768 | unsigned int ee_len; | |
b720303d | 2769 | int ret; |
093a088b | 2770 | |
093a088b | 2771 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2772 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2773 | |
a107e5a3 | 2774 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2775 | if (ret > 0) |
2776 | ret = 0; | |
093a088b | 2777 | |
2407518d | 2778 | return ret; |
093a088b AK |
2779 | } |
2780 | ||
47ea3bb5 YY |
2781 | /* |
2782 | * used by extent splitting. | |
2783 | */ | |
2784 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
2785 | due to ENOSPC */ | |
2786 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
2787 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
2788 | ||
2789 | /* | |
2790 | * ext4_split_extent_at() splits an extent at given block. | |
2791 | * | |
2792 | * @handle: the journal handle | |
2793 | * @inode: the file inode | |
2794 | * @path: the path to the extent | |
2795 | * @split: the logical block where the extent is splitted. | |
2796 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2797 | * the states(init or uninit) of new extents. | |
2798 | * @flags: flags used to insert new extent to extent tree. | |
2799 | * | |
2800 | * | |
2801 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2802 | * of which are deterimined by split_flag. | |
2803 | * | |
2804 | * There are two cases: | |
2805 | * a> the extent are splitted into two extent. | |
2806 | * b> split is not needed, and just mark the extent. | |
2807 | * | |
2808 | * return 0 on success. | |
2809 | */ | |
2810 | static int ext4_split_extent_at(handle_t *handle, | |
2811 | struct inode *inode, | |
2812 | struct ext4_ext_path *path, | |
2813 | ext4_lblk_t split, | |
2814 | int split_flag, | |
2815 | int flags) | |
2816 | { | |
2817 | ext4_fsblk_t newblock; | |
2818 | ext4_lblk_t ee_block; | |
2819 | struct ext4_extent *ex, newex, orig_ex; | |
2820 | struct ext4_extent *ex2 = NULL; | |
2821 | unsigned int ee_len, depth; | |
2822 | int err = 0; | |
2823 | ||
2824 | ext_debug("ext4_split_extents_at: inode %lu, logical" | |
2825 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2826 | ||
2827 | ext4_ext_show_leaf(inode, path); | |
2828 | ||
2829 | depth = ext_depth(inode); | |
2830 | ex = path[depth].p_ext; | |
2831 | ee_block = le32_to_cpu(ex->ee_block); | |
2832 | ee_len = ext4_ext_get_actual_len(ex); | |
2833 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2834 | ||
2835 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
2836 | ||
2837 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2838 | if (err) | |
2839 | goto out; | |
2840 | ||
2841 | if (split == ee_block) { | |
2842 | /* | |
2843 | * case b: block @split is the block that the extent begins with | |
2844 | * then we just change the state of the extent, and splitting | |
2845 | * is not needed. | |
2846 | */ | |
2847 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2848 | ext4_ext_mark_uninitialized(ex); | |
2849 | else | |
2850 | ext4_ext_mark_initialized(ex); | |
2851 | ||
2852 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
2853 | ext4_ext_try_to_merge(inode, path, ex); | |
2854 | ||
2855 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2856 | goto out; | |
2857 | } | |
2858 | ||
2859 | /* case a */ | |
2860 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
2861 | ex->ee_len = cpu_to_le16(split - ee_block); | |
2862 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
2863 | ext4_ext_mark_uninitialized(ex); | |
2864 | ||
2865 | /* | |
2866 | * path may lead to new leaf, not to original leaf any more | |
2867 | * after ext4_ext_insert_extent() returns, | |
2868 | */ | |
2869 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2870 | if (err) | |
2871 | goto fix_extent_len; | |
2872 | ||
2873 | ex2 = &newex; | |
2874 | ex2->ee_block = cpu_to_le32(split); | |
2875 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
2876 | ext4_ext_store_pblock(ex2, newblock); | |
2877 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2878 | ext4_ext_mark_uninitialized(ex2); | |
2879 | ||
2880 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
2881 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2882 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2883 | if (err) | |
2884 | goto fix_extent_len; | |
2885 | /* update the extent length and mark as initialized */ | |
2886 | ex->ee_len = cpu_to_le32(ee_len); | |
2887 | ext4_ext_try_to_merge(inode, path, ex); | |
2888 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2889 | goto out; | |
2890 | } else if (err) | |
2891 | goto fix_extent_len; | |
2892 | ||
2893 | out: | |
2894 | ext4_ext_show_leaf(inode, path); | |
2895 | return err; | |
2896 | ||
2897 | fix_extent_len: | |
2898 | ex->ee_len = orig_ex.ee_len; | |
2899 | ext4_ext_dirty(handle, inode, path + depth); | |
2900 | return err; | |
2901 | } | |
2902 | ||
2903 | /* | |
2904 | * ext4_split_extents() splits an extent and mark extent which is covered | |
2905 | * by @map as split_flags indicates | |
2906 | * | |
2907 | * It may result in splitting the extent into multiple extents (upto three) | |
2908 | * There are three possibilities: | |
2909 | * a> There is no split required | |
2910 | * b> Splits in two extents: Split is happening at either end of the extent | |
2911 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
2912 | * | |
2913 | */ | |
2914 | static int ext4_split_extent(handle_t *handle, | |
2915 | struct inode *inode, | |
2916 | struct ext4_ext_path *path, | |
2917 | struct ext4_map_blocks *map, | |
2918 | int split_flag, | |
2919 | int flags) | |
2920 | { | |
2921 | ext4_lblk_t ee_block; | |
2922 | struct ext4_extent *ex; | |
2923 | unsigned int ee_len, depth; | |
2924 | int err = 0; | |
2925 | int uninitialized; | |
2926 | int split_flag1, flags1; | |
2927 | ||
2928 | depth = ext_depth(inode); | |
2929 | ex = path[depth].p_ext; | |
2930 | ee_block = le32_to_cpu(ex->ee_block); | |
2931 | ee_len = ext4_ext_get_actual_len(ex); | |
2932 | uninitialized = ext4_ext_is_uninitialized(ex); | |
2933 | ||
2934 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
2935 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
2936 | EXT4_EXT_MAY_ZEROOUT : 0; | |
2937 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; | |
2938 | if (uninitialized) | |
2939 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
2940 | EXT4_EXT_MARK_UNINIT2; | |
2941 | err = ext4_split_extent_at(handle, inode, path, | |
2942 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
2943 | if (err) |
2944 | goto out; | |
47ea3bb5 YY |
2945 | } |
2946 | ||
2947 | ext4_ext_drop_refs(path); | |
2948 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
2949 | if (IS_ERR(path)) | |
2950 | return PTR_ERR(path); | |
2951 | ||
2952 | if (map->m_lblk >= ee_block) { | |
2953 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
2954 | EXT4_EXT_MAY_ZEROOUT : 0; | |
2955 | if (uninitialized) | |
2956 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
2957 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2958 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
2959 | err = ext4_split_extent_at(handle, inode, path, | |
2960 | map->m_lblk, split_flag1, flags); | |
2961 | if (err) | |
2962 | goto out; | |
2963 | } | |
2964 | ||
2965 | ext4_ext_show_leaf(inode, path); | |
2966 | out: | |
2967 | return err ? err : map->m_len; | |
2968 | } | |
2969 | ||
3977c965 | 2970 | #define EXT4_EXT_ZERO_LEN 7 |
56055d3a | 2971 | /* |
e35fd660 | 2972 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 2973 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 2974 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
2975 | * uninitialized). |
2976 | * There are three possibilities: | |
2977 | * a> There is no split required: Entire extent should be initialized | |
2978 | * b> Splits in two extents: Write is happening at either end of the extent | |
2979 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2980 | */ | |
725d26d3 | 2981 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
2982 | struct inode *inode, |
2983 | struct ext4_map_blocks *map, | |
2984 | struct ext4_ext_path *path) | |
56055d3a | 2985 | { |
667eff35 YY |
2986 | struct ext4_map_blocks split_map; |
2987 | struct ext4_extent zero_ex; | |
2988 | struct ext4_extent *ex; | |
21ca087a | 2989 | ext4_lblk_t ee_block, eof_block; |
725d26d3 | 2990 | unsigned int allocated, ee_len, depth; |
56055d3a | 2991 | int err = 0; |
667eff35 | 2992 | int split_flag = 0; |
21ca087a DM |
2993 | |
2994 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
2995 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 2996 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
2997 | |
2998 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
2999 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3000 | if (eof_block < map->m_lblk + map->m_len) |
3001 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3002 | |
3003 | depth = ext_depth(inode); | |
56055d3a AA |
3004 | ex = path[depth].p_ext; |
3005 | ee_block = le32_to_cpu(ex->ee_block); | |
3006 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3007 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 3008 | |
667eff35 | 3009 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3010 | /* |
3011 | * It is safe to convert extent to initialized via explicit | |
3012 | * zeroout only if extent is fully insde i_size or new_size. | |
3013 | */ | |
667eff35 | 3014 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3015 | |
3977c965 | 3016 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
667eff35 YY |
3017 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN && |
3018 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3019 | err = ext4_ext_zeroout(inode, ex); | |
3977c965 | 3020 | if (err) |
d03856bd | 3021 | goto out; |
d03856bd AK |
3022 | |
3023 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3024 | if (err) | |
3025 | goto out; | |
667eff35 YY |
3026 | ext4_ext_mark_initialized(ex); |
3027 | ext4_ext_try_to_merge(inode, path, ex); | |
3028 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3029 | goto out; | |
56055d3a | 3030 | } |
667eff35 | 3031 | |
56055d3a | 3032 | /* |
667eff35 YY |
3033 | * four cases: |
3034 | * 1. split the extent into three extents. | |
3035 | * 2. split the extent into two extents, zeroout the first half. | |
3036 | * 3. split the extent into two extents, zeroout the second half. | |
3037 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3038 | */ |
667eff35 YY |
3039 | split_map.m_lblk = map->m_lblk; |
3040 | split_map.m_len = map->m_len; | |
3041 | ||
3042 | if (allocated > map->m_len) { | |
3043 | if (allocated <= EXT4_EXT_ZERO_LEN && | |
3044 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3045 | /* case 3 */ | |
3046 | zero_ex.ee_block = | |
9b940f8e AH |
3047 | cpu_to_le32(map->m_lblk); |
3048 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3049 | ext4_ext_store_pblock(&zero_ex, |
3050 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3051 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3052 | if (err) |
3053 | goto out; | |
667eff35 YY |
3054 | split_map.m_lblk = map->m_lblk; |
3055 | split_map.m_len = allocated; | |
3056 | } else if ((map->m_lblk - ee_block + map->m_len < | |
3057 | EXT4_EXT_ZERO_LEN) && | |
3058 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3059 | /* case 2 */ | |
3060 | if (map->m_lblk != ee_block) { | |
3061 | zero_ex.ee_block = ex->ee_block; | |
3062 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3063 | ee_block); | |
3064 | ext4_ext_store_pblock(&zero_ex, | |
3065 | ext4_ext_pblock(ex)); | |
3066 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3067 | if (err) | |
3068 | goto out; | |
3069 | } | |
3070 | ||
667eff35 | 3071 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3072 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3073 | allocated = map->m_len; | |
56055d3a AA |
3074 | } |
3075 | } | |
667eff35 YY |
3076 | |
3077 | allocated = ext4_split_extent(handle, inode, path, | |
3078 | &split_map, split_flag, 0); | |
3079 | if (allocated < 0) | |
3080 | err = allocated; | |
3081 | ||
56055d3a AA |
3082 | out: |
3083 | return err ? err : allocated; | |
3084 | } | |
3085 | ||
0031462b | 3086 | /* |
e35fd660 | 3087 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3088 | * ext4_get_blocks_dio_write() when DIO to write |
3089 | * to an uninitialized extent. | |
3090 | * | |
fd018fe8 | 3091 | * Writing to an uninitialized extent may result in splitting the uninitialized |
b595076a | 3092 | * extent into multiple /initialized uninitialized extents (up to three) |
0031462b MC |
3093 | * There are three possibilities: |
3094 | * a> There is no split required: Entire extent should be uninitialized | |
3095 | * b> Splits in two extents: Write is happening at either end of the extent | |
3096 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3097 | * | |
3098 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3099 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3100 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3101 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3102 | * into three uninitialized extent(at most). After IO complete, the part |
3103 | * being filled will be convert to initialized by the end_io callback function | |
3104 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3105 | * |
3106 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3107 | */ |
3108 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3109 | struct inode *inode, | |
e35fd660 | 3110 | struct ext4_map_blocks *map, |
0031462b | 3111 | struct ext4_ext_path *path, |
0031462b MC |
3112 | int flags) |
3113 | { | |
667eff35 YY |
3114 | ext4_lblk_t eof_block; |
3115 | ext4_lblk_t ee_block; | |
3116 | struct ext4_extent *ex; | |
3117 | unsigned int ee_len; | |
3118 | int split_flag = 0, depth; | |
21ca087a DM |
3119 | |
3120 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3121 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3122 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3123 | |
3124 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3125 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3126 | if (eof_block < map->m_lblk + map->m_len) |
3127 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3128 | /* |
3129 | * It is safe to convert extent to initialized via explicit | |
3130 | * zeroout only if extent is fully insde i_size or new_size. | |
3131 | */ | |
667eff35 YY |
3132 | depth = ext_depth(inode); |
3133 | ex = path[depth].p_ext; | |
3134 | ee_block = le32_to_cpu(ex->ee_block); | |
3135 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3136 | |
667eff35 YY |
3137 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3138 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
0031462b | 3139 | |
667eff35 YY |
3140 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3141 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3142 | } |
197217a5 | 3143 | |
c7064ef1 | 3144 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
0031462b MC |
3145 | struct inode *inode, |
3146 | struct ext4_ext_path *path) | |
3147 | { | |
3148 | struct ext4_extent *ex; | |
0031462b MC |
3149 | int depth; |
3150 | int err = 0; | |
0031462b MC |
3151 | |
3152 | depth = ext_depth(inode); | |
0031462b MC |
3153 | ex = path[depth].p_ext; |
3154 | ||
197217a5 YY |
3155 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3156 | "block %llu, max_blocks %u\n", inode->i_ino, | |
3157 | (unsigned long long)le32_to_cpu(ex->ee_block), | |
3158 | ext4_ext_get_actual_len(ex)); | |
3159 | ||
0031462b MC |
3160 | err = ext4_ext_get_access(handle, inode, path + depth); |
3161 | if (err) | |
3162 | goto out; | |
3163 | /* first mark the extent as initialized */ | |
3164 | ext4_ext_mark_initialized(ex); | |
3165 | ||
197217a5 YY |
3166 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3167 | * borders are not changed | |
0031462b | 3168 | */ |
197217a5 YY |
3169 | ext4_ext_try_to_merge(inode, path, ex); |
3170 | ||
0031462b MC |
3171 | /* Mark modified extent as dirty */ |
3172 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3173 | out: | |
3174 | ext4_ext_show_leaf(inode, path); | |
3175 | return err; | |
3176 | } | |
3177 | ||
515f41c3 AK |
3178 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3179 | sector_t block, int count) | |
3180 | { | |
3181 | int i; | |
3182 | for (i = 0; i < count; i++) | |
3183 | unmap_underlying_metadata(bdev, block + i); | |
3184 | } | |
3185 | ||
58590b06 TT |
3186 | /* |
3187 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3188 | */ | |
3189 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3190 | ext4_lblk_t lblk, |
58590b06 TT |
3191 | struct ext4_ext_path *path, |
3192 | unsigned int len) | |
3193 | { | |
3194 | int i, depth; | |
3195 | struct ext4_extent_header *eh; | |
65922cb5 | 3196 | struct ext4_extent *last_ex; |
58590b06 TT |
3197 | |
3198 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3199 | return 0; | |
3200 | ||
3201 | depth = ext_depth(inode); | |
3202 | eh = path[depth].p_hdr; | |
58590b06 TT |
3203 | |
3204 | if (unlikely(!eh->eh_entries)) { | |
3205 | EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and " | |
3206 | "EOFBLOCKS_FL set"); | |
3207 | return -EIO; | |
3208 | } | |
3209 | last_ex = EXT_LAST_EXTENT(eh); | |
3210 | /* | |
3211 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3212 | * last block in the last extent in the file. We test this by | |
3213 | * first checking to see if the caller to | |
3214 | * ext4_ext_get_blocks() was interested in the last block (or | |
3215 | * a block beyond the last block) in the current extent. If | |
3216 | * this turns out to be false, we can bail out from this | |
3217 | * function immediately. | |
3218 | */ | |
d002ebf1 | 3219 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3220 | ext4_ext_get_actual_len(last_ex)) |
3221 | return 0; | |
3222 | /* | |
3223 | * If the caller does appear to be planning to write at or | |
3224 | * beyond the end of the current extent, we then test to see | |
3225 | * if the current extent is the last extent in the file, by | |
3226 | * checking to make sure it was reached via the rightmost node | |
3227 | * at each level of the tree. | |
3228 | */ | |
3229 | for (i = depth-1; i >= 0; i--) | |
3230 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3231 | return 0; | |
3232 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); | |
3233 | return ext4_mark_inode_dirty(handle, inode); | |
3234 | } | |
3235 | ||
0031462b MC |
3236 | static int |
3237 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3238 | struct ext4_map_blocks *map, |
0031462b | 3239 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3240 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3241 | { |
3242 | int ret = 0; | |
3243 | int err = 0; | |
8d5d02e6 | 3244 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
0031462b MC |
3245 | |
3246 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical" | |
3247 | "block %llu, max_blocks %u, flags %d, allocated %u", | |
e35fd660 | 3248 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3249 | flags, allocated); |
3250 | ext4_ext_show_leaf(inode, path); | |
3251 | ||
c7064ef1 | 3252 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3253 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3254 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3255 | path, flags); | |
5f524950 M |
3256 | /* |
3257 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3258 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3259 | * completed |
3260 | */ | |
e9e3bcec | 3261 | if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) { |
bd2d0210 | 3262 | io->flag = EXT4_IO_END_UNWRITTEN; |
e9e3bcec ES |
3263 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); |
3264 | } else | |
19f5fb7a | 3265 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
744692dc | 3266 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3267 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3268 | goto out; |
3269 | } | |
c7064ef1 | 3270 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3271 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
c7064ef1 | 3272 | ret = ext4_convert_unwritten_extents_endio(handle, inode, |
0031462b | 3273 | path); |
58590b06 | 3274 | if (ret >= 0) { |
b436b9be | 3275 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3276 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3277 | path, map->m_len); | |
58590b06 TT |
3278 | } else |
3279 | err = ret; | |
0031462b MC |
3280 | goto out2; |
3281 | } | |
3282 | /* buffered IO case */ | |
3283 | /* | |
3284 | * repeat fallocate creation request | |
3285 | * we already have an unwritten extent | |
3286 | */ | |
3287 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | |
3288 | goto map_out; | |
3289 | ||
3290 | /* buffered READ or buffered write_begin() lookup */ | |
3291 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3292 | /* | |
3293 | * We have blocks reserved already. We | |
3294 | * return allocated blocks so that delalloc | |
3295 | * won't do block reservation for us. But | |
3296 | * the buffer head will be unmapped so that | |
3297 | * a read from the block returns 0s. | |
3298 | */ | |
e35fd660 | 3299 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3300 | goto out1; |
3301 | } | |
3302 | ||
3303 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3304 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
58590b06 | 3305 | if (ret >= 0) { |
b436b9be | 3306 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3307 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, |
3308 | map->m_len); | |
58590b06 TT |
3309 | if (err < 0) |
3310 | goto out2; | |
3311 | } | |
3312 | ||
0031462b MC |
3313 | out: |
3314 | if (ret <= 0) { | |
3315 | err = ret; | |
3316 | goto out2; | |
3317 | } else | |
3318 | allocated = ret; | |
e35fd660 | 3319 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3320 | /* |
3321 | * if we allocated more blocks than requested | |
3322 | * we need to make sure we unmap the extra block | |
3323 | * allocated. The actual needed block will get | |
3324 | * unmapped later when we find the buffer_head marked | |
3325 | * new. | |
3326 | */ | |
e35fd660 | 3327 | if (allocated > map->m_len) { |
515f41c3 | 3328 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3329 | newblock + map->m_len, |
3330 | allocated - map->m_len); | |
3331 | allocated = map->m_len; | |
515f41c3 | 3332 | } |
5f634d06 AK |
3333 | |
3334 | /* | |
3335 | * If we have done fallocate with the offset that is already | |
3336 | * delayed allocated, we would have block reservation | |
3337 | * and quota reservation done in the delayed write path. | |
3338 | * But fallocate would have already updated quota and block | |
3339 | * count for this offset. So cancel these reservation | |
3340 | */ | |
1296cc85 | 3341 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
5f634d06 AK |
3342 | ext4_da_update_reserve_space(inode, allocated, 0); |
3343 | ||
0031462b | 3344 | map_out: |
e35fd660 | 3345 | map->m_flags |= EXT4_MAP_MAPPED; |
0031462b | 3346 | out1: |
e35fd660 TT |
3347 | if (allocated > map->m_len) |
3348 | allocated = map->m_len; | |
0031462b | 3349 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3350 | map->m_pblk = newblock; |
3351 | map->m_len = allocated; | |
0031462b MC |
3352 | out2: |
3353 | if (path) { | |
3354 | ext4_ext_drop_refs(path); | |
3355 | kfree(path); | |
3356 | } | |
3357 | return err ? err : allocated; | |
3358 | } | |
58590b06 | 3359 | |
4d33b1ef TT |
3360 | /* |
3361 | * get_implied_cluster_alloc - check to see if the requested | |
3362 | * allocation (in the map structure) overlaps with a cluster already | |
3363 | * allocated in an extent. | |
3364 | * @sbi The ext4-specific superblock structure | |
3365 | * @map The requested lblk->pblk mapping | |
3366 | * @ex The extent structure which might contain an implied | |
3367 | * cluster allocation | |
3368 | * | |
3369 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3370 | * find blocks that were already in the inode's extent tree. Hence, | |
3371 | * we know that the beginning of the requested region cannot overlap | |
3372 | * the extent from the inode's extent tree. There are three cases we | |
3373 | * want to catch. The first is this case: | |
3374 | * | |
3375 | * |--- cluster # N--| | |
3376 | * |--- extent ---| |---- requested region ---| | |
3377 | * |==========| | |
3378 | * | |
3379 | * The second case that we need to test for is this one: | |
3380 | * | |
3381 | * |--------- cluster # N ----------------| | |
3382 | * |--- requested region --| |------- extent ----| | |
3383 | * |=======================| | |
3384 | * | |
3385 | * The third case is when the requested region lies between two extents | |
3386 | * within the same cluster: | |
3387 | * |------------- cluster # N-------------| | |
3388 | * |----- ex -----| |---- ex_right ----| | |
3389 | * |------ requested region ------| | |
3390 | * |================| | |
3391 | * | |
3392 | * In each of the above cases, we need to set the map->m_pblk and | |
3393 | * map->m_len so it corresponds to the return the extent labelled as | |
3394 | * "|====|" from cluster #N, since it is already in use for data in | |
3395 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3396 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3397 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3398 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3399 | * by calling ext4_mb_new_blocks(). | |
3400 | */ | |
3401 | static int get_implied_cluster_alloc(struct ext4_sb_info *sbi, | |
3402 | struct ext4_map_blocks *map, | |
3403 | struct ext4_extent *ex, | |
3404 | struct ext4_ext_path *path) | |
3405 | { | |
3406 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
3407 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
3408 | ext4_lblk_t rr_cluster_start, rr_cluster_end; | |
3409 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); | |
3410 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3411 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3412 | ||
3413 | /* The extent passed in that we are trying to match */ | |
3414 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3415 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3416 | ||
3417 | /* The requested region passed into ext4_map_blocks() */ | |
3418 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
3419 | rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1); | |
3420 | ||
3421 | if ((rr_cluster_start == ex_cluster_end) || | |
3422 | (rr_cluster_start == ex_cluster_start)) { | |
3423 | if (rr_cluster_start == ex_cluster_end) | |
3424 | ee_start += ee_len - 1; | |
3425 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3426 | c_offset; | |
3427 | map->m_len = min(map->m_len, | |
3428 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3429 | /* | |
3430 | * Check for and handle this case: | |
3431 | * | |
3432 | * |--------- cluster # N-------------| | |
3433 | * |------- extent ----| | |
3434 | * |--- requested region ---| | |
3435 | * |===========| | |
3436 | */ | |
3437 | ||
3438 | if (map->m_lblk < ee_block) | |
3439 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3440 | ||
3441 | /* | |
3442 | * Check for the case where there is already another allocated | |
3443 | * block to the right of 'ex' but before the end of the cluster. | |
3444 | * | |
3445 | * |------------- cluster # N-------------| | |
3446 | * |----- ex -----| |---- ex_right ----| | |
3447 | * |------ requested region ------| | |
3448 | * |================| | |
3449 | */ | |
3450 | if (map->m_lblk > ee_block) { | |
3451 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3452 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3453 | } | |
3454 | return 1; | |
3455 | } | |
3456 | return 0; | |
3457 | } | |
3458 | ||
3459 | ||
c278bfec | 3460 | /* |
f5ab0d1f MC |
3461 | * Block allocation/map/preallocation routine for extents based files |
3462 | * | |
3463 | * | |
c278bfec | 3464 | * Need to be called with |
0e855ac8 AK |
3465 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3466 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3467 | * |
3468 | * return > 0, number of of blocks already mapped/allocated | |
3469 | * if create == 0 and these are pre-allocated blocks | |
3470 | * buffer head is unmapped | |
3471 | * otherwise blocks are mapped | |
3472 | * | |
3473 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3474 | * buffer head is unmapped | |
3475 | * | |
3476 | * return < 0, error case. | |
c278bfec | 3477 | */ |
e35fd660 TT |
3478 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3479 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3480 | { |
3481 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3482 | struct ext4_extent newex, *ex, *ex2; |
3483 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3484 | ext4_fsblk_t newblock = 0; |
4d33b1ef TT |
3485 | int free_on_err = 0, err = 0, depth, ret; |
3486 | unsigned int allocated = 0, offset = 0; | |
e861304b AH |
3487 | unsigned int punched_out = 0; |
3488 | unsigned int result = 0; | |
c9de560d | 3489 | struct ext4_allocation_request ar; |
8d5d02e6 | 3490 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
4d33b1ef | 3491 | ext4_lblk_t cluster_offset; |
e861304b | 3492 | struct ext4_map_blocks punch_map; |
a86c6181 | 3493 | |
84fe3bef | 3494 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3495 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3496 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3497 | |
3498 | /* check in cache */ | |
015861ba RD |
3499 | if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) && |
3500 | ext4_ext_in_cache(inode, map->m_lblk, &newex)) { | |
b05e6ae5 | 3501 | if (!newex.ee_start_lo && !newex.ee_start_hi) { |
c2177057 | 3502 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3503 | /* |
3504 | * block isn't allocated yet and | |
3505 | * user doesn't want to allocate it | |
3506 | */ | |
a86c6181 AT |
3507 | goto out2; |
3508 | } | |
3509 | /* we should allocate requested block */ | |
b05e6ae5 | 3510 | } else { |
a86c6181 | 3511 | /* block is already allocated */ |
e35fd660 | 3512 | newblock = map->m_lblk |
8c55e204 | 3513 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3514 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3515 | /* number of remaining blocks in the extent */ |
b939e376 | 3516 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3517 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3518 | goto out; |
a86c6181 AT |
3519 | } |
3520 | } | |
3521 | ||
3522 | /* find extent for this block */ | |
e35fd660 | 3523 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3524 | if (IS_ERR(path)) { |
3525 | err = PTR_ERR(path); | |
3526 | path = NULL; | |
3527 | goto out2; | |
3528 | } | |
3529 | ||
3530 | depth = ext_depth(inode); | |
3531 | ||
3532 | /* | |
d0d856e8 RD |
3533 | * consistent leaf must not be empty; |
3534 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3535 | * this is why assert can't be put in ext4_ext_find_extent() |
3536 | */ | |
273df556 FM |
3537 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3538 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3539 | "lblock: %lu, depth: %d pblock %lld", |
3540 | (unsigned long) map->m_lblk, depth, | |
3541 | path[depth].p_block); | |
034fb4c9 SP |
3542 | err = -EIO; |
3543 | goto out2; | |
3544 | } | |
a86c6181 | 3545 | |
7e028976 AM |
3546 | ex = path[depth].p_ext; |
3547 | if (ex) { | |
725d26d3 | 3548 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3549 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3550 | unsigned short ee_len; |
471d4011 SB |
3551 | |
3552 | /* | |
471d4011 | 3553 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3554 | * we split out initialized portions during a write. |
471d4011 | 3555 | */ |
a2df2a63 | 3556 | ee_len = ext4_ext_get_actual_len(ex); |
d0d856e8 | 3557 | /* if found extent covers block, simply return it */ |
e35fd660 | 3558 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
0aa06000 TT |
3559 | ext4_fsblk_t partial_cluster = 0; |
3560 | ||
e35fd660 | 3561 | newblock = map->m_lblk - ee_block + ee_start; |
d0d856e8 | 3562 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3563 | allocated = ee_len - (map->m_lblk - ee_block); |
3564 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3565 | ee_block, ee_len, newblock); | |
56055d3a | 3566 | |
e861304b AH |
3567 | if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) { |
3568 | /* | |
3569 | * Do not put uninitialized extent | |
3570 | * in the cache | |
3571 | */ | |
3572 | if (!ext4_ext_is_uninitialized(ex)) { | |
3573 | ext4_ext_put_in_cache(inode, ee_block, | |
3574 | ee_len, ee_start); | |
3575 | goto out; | |
3576 | } | |
3577 | ret = ext4_ext_handle_uninitialized_extents( | |
3578 | handle, inode, map, path, flags, | |
3579 | allocated, newblock); | |
3580 | return ret; | |
56055d3a | 3581 | } |
e861304b AH |
3582 | |
3583 | /* | |
3584 | * Punch out the map length, but only to the | |
3585 | * end of the extent | |
3586 | */ | |
3587 | punched_out = allocated < map->m_len ? | |
3588 | allocated : map->m_len; | |
3589 | ||
3590 | /* | |
3591 | * Sense extents need to be converted to | |
3592 | * uninitialized, they must fit in an | |
3593 | * uninitialized extent | |
3594 | */ | |
3595 | if (punched_out > EXT_UNINIT_MAX_LEN) | |
3596 | punched_out = EXT_UNINIT_MAX_LEN; | |
3597 | ||
3598 | punch_map.m_lblk = map->m_lblk; | |
3599 | punch_map.m_pblk = newblock; | |
3600 | punch_map.m_len = punched_out; | |
3601 | punch_map.m_flags = 0; | |
3602 | ||
3603 | /* Check to see if the extent needs to be split */ | |
3604 | if (punch_map.m_len != ee_len || | |
3605 | punch_map.m_lblk != ee_block) { | |
3606 | ||
3607 | ret = ext4_split_extent(handle, inode, | |
3608 | path, &punch_map, 0, | |
3609 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT | | |
3610 | EXT4_GET_BLOCKS_PRE_IO); | |
3611 | ||
3612 | if (ret < 0) { | |
3613 | err = ret; | |
3614 | goto out2; | |
3615 | } | |
3616 | /* | |
3617 | * find extent for the block at | |
3618 | * the start of the hole | |
3619 | */ | |
3620 | ext4_ext_drop_refs(path); | |
3621 | kfree(path); | |
3622 | ||
3623 | path = ext4_ext_find_extent(inode, | |
3624 | map->m_lblk, NULL); | |
3625 | if (IS_ERR(path)) { | |
3626 | err = PTR_ERR(path); | |
3627 | path = NULL; | |
3628 | goto out2; | |
3629 | } | |
3630 | ||
3631 | depth = ext_depth(inode); | |
3632 | ex = path[depth].p_ext; | |
3633 | ee_len = ext4_ext_get_actual_len(ex); | |
3634 | ee_block = le32_to_cpu(ex->ee_block); | |
3635 | ee_start = ext4_ext_pblock(ex); | |
3636 | ||
3637 | } | |
3638 | ||
3639 | ext4_ext_mark_uninitialized(ex); | |
3640 | ||
f7d0d379 AH |
3641 | ext4_ext_invalidate_cache(inode); |
3642 | ||
3643 | err = ext4_ext_rm_leaf(handle, inode, path, | |
0aa06000 TT |
3644 | &partial_cluster, map->m_lblk, |
3645 | map->m_lblk + punched_out); | |
f7d0d379 AH |
3646 | |
3647 | if (!err && path->p_hdr->eh_entries == 0) { | |
3648 | /* | |
3649 | * Punch hole freed all of this sub tree, | |
3650 | * so we need to correct eh_depth | |
3651 | */ | |
3652 | err = ext4_ext_get_access(handle, inode, path); | |
3653 | if (err == 0) { | |
3654 | ext_inode_hdr(inode)->eh_depth = 0; | |
3655 | ext_inode_hdr(inode)->eh_max = | |
3656 | cpu_to_le16(ext4_ext_space_root( | |
3657 | inode, 0)); | |
3658 | ||
3659 | err = ext4_ext_dirty( | |
3660 | handle, inode, path); | |
3661 | } | |
3662 | } | |
e861304b AH |
3663 | |
3664 | goto out2; | |
a86c6181 AT |
3665 | } |
3666 | } | |
3667 | ||
3668 | /* | |
d0d856e8 | 3669 | * requested block isn't allocated yet; |
a86c6181 AT |
3670 | * we couldn't try to create block if create flag is zero |
3671 | */ | |
c2177057 | 3672 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3673 | /* |
3674 | * put just found gap into cache to speed up | |
3675 | * subsequent requests | |
3676 | */ | |
e35fd660 | 3677 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
3678 | goto out2; |
3679 | } | |
4d33b1ef | 3680 | |
a86c6181 | 3681 | /* |
c2ea3fde | 3682 | * Okay, we need to do block allocation. |
63f57933 | 3683 | */ |
4d33b1ef TT |
3684 | newex.ee_block = cpu_to_le32(map->m_lblk); |
3685 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
3686 | ||
3687 | /* | |
3688 | * If we are doing bigalloc, check to see if the extent returned | |
3689 | * by ext4_ext_find_extent() implies a cluster we can use. | |
3690 | */ | |
3691 | if (cluster_offset && ex && | |
3692 | get_implied_cluster_alloc(sbi, map, ex, path)) { | |
3693 | ar.len = allocated = map->m_len; | |
3694 | newblock = map->m_pblk; | |
3695 | goto got_allocated_blocks; | |
3696 | } | |
a86c6181 | 3697 | |
c9de560d | 3698 | /* find neighbour allocated blocks */ |
e35fd660 | 3699 | ar.lleft = map->m_lblk; |
c9de560d AT |
3700 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
3701 | if (err) | |
3702 | goto out2; | |
e35fd660 | 3703 | ar.lright = map->m_lblk; |
4d33b1ef TT |
3704 | ex2 = NULL; |
3705 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
3706 | if (err) |
3707 | goto out2; | |
25d14f98 | 3708 | |
4d33b1ef TT |
3709 | /* Check if the extent after searching to the right implies a |
3710 | * cluster we can use. */ | |
3711 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
3712 | get_implied_cluster_alloc(sbi, map, ex2, path)) { | |
3713 | ar.len = allocated = map->m_len; | |
3714 | newblock = map->m_pblk; | |
3715 | goto got_allocated_blocks; | |
3716 | } | |
3717 | ||
749269fa AA |
3718 | /* |
3719 | * See if request is beyond maximum number of blocks we can have in | |
3720 | * a single extent. For an initialized extent this limit is | |
3721 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
3722 | * EXT_UNINIT_MAX_LEN. | |
3723 | */ | |
e35fd660 | 3724 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 3725 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
3726 | map->m_len = EXT_INIT_MAX_LEN; |
3727 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 3728 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 3729 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 3730 | |
e35fd660 | 3731 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 3732 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 3733 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 3734 | if (err) |
b939e376 | 3735 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 3736 | else |
e35fd660 | 3737 | allocated = map->m_len; |
c9de560d AT |
3738 | |
3739 | /* allocate new block */ | |
3740 | ar.inode = inode; | |
e35fd660 TT |
3741 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
3742 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
3743 | /* |
3744 | * We calculate the offset from the beginning of the cluster | |
3745 | * for the logical block number, since when we allocate a | |
3746 | * physical cluster, the physical block should start at the | |
3747 | * same offset from the beginning of the cluster. This is | |
3748 | * needed so that future calls to get_implied_cluster_alloc() | |
3749 | * work correctly. | |
3750 | */ | |
3751 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
3752 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
3753 | ar.goal -= offset; | |
3754 | ar.logical -= offset; | |
c9de560d AT |
3755 | if (S_ISREG(inode->i_mode)) |
3756 | ar.flags = EXT4_MB_HINT_DATA; | |
3757 | else | |
3758 | /* disable in-core preallocation for non-regular files */ | |
3759 | ar.flags = 0; | |
556b27ab VH |
3760 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
3761 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 3762 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
3763 | if (!newblock) |
3764 | goto out2; | |
84fe3bef | 3765 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 3766 | ar.goal, newblock, allocated); |
4d33b1ef TT |
3767 | free_on_err = 1; |
3768 | ar.len = EXT4_C2B(sbi, ar.len) - offset; | |
3769 | if (ar.len > allocated) | |
3770 | ar.len = allocated; | |
a86c6181 | 3771 | |
4d33b1ef | 3772 | got_allocated_blocks: |
a86c6181 | 3773 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 3774 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 3775 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
3776 | /* Mark uninitialized */ |
3777 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 3778 | ext4_ext_mark_uninitialized(&newex); |
8d5d02e6 | 3779 | /* |
744692dc | 3780 | * io_end structure was created for every IO write to an |
25985edc | 3781 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 3782 | * here we flag the IO that really needs the conversion. |
5f524950 | 3783 | * For non asycn direct IO case, flag the inode state |
25985edc | 3784 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 3785 | */ |
744692dc | 3786 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e9e3bcec | 3787 | if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) { |
bd2d0210 | 3788 | io->flag = EXT4_IO_END_UNWRITTEN; |
e9e3bcec ES |
3789 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); |
3790 | } else | |
19f5fb7a TT |
3791 | ext4_set_inode_state(inode, |
3792 | EXT4_STATE_DIO_UNWRITTEN); | |
5f524950 | 3793 | } |
744692dc | 3794 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3795 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 3796 | } |
c8d46e41 | 3797 | |
d002ebf1 | 3798 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len); |
575a1d4b JZ |
3799 | if (!err) |
3800 | err = ext4_ext_insert_extent(handle, inode, path, | |
3801 | &newex, flags); | |
4d33b1ef | 3802 | if (err && free_on_err) { |
7132de74 MP |
3803 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
3804 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 3805 | /* free data blocks we just allocated */ |
c9de560d AT |
3806 | /* not a good idea to call discard here directly, |
3807 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 3808 | ext4_discard_preallocations(inode); |
7dc57615 | 3809 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 3810 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 3811 | goto out2; |
315054f0 | 3812 | } |
a86c6181 | 3813 | |
a86c6181 | 3814 | /* previous routine could use block we allocated */ |
bf89d16f | 3815 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 3816 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
3817 | if (allocated > map->m_len) |
3818 | allocated = map->m_len; | |
3819 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 3820 | |
5f634d06 AK |
3821 | /* |
3822 | * Update reserved blocks/metadata blocks after successful | |
3823 | * block allocation which had been deferred till now. | |
3824 | */ | |
1296cc85 | 3825 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
5f634d06 AK |
3826 | ext4_da_update_reserve_space(inode, allocated, 1); |
3827 | ||
b436b9be JK |
3828 | /* |
3829 | * Cache the extent and update transaction to commit on fdatasync only | |
3830 | * when it is _not_ an uninitialized extent. | |
3831 | */ | |
3832 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 3833 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
3834 | ext4_update_inode_fsync_trans(handle, inode, 1); |
3835 | } else | |
3836 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 3837 | out: |
e35fd660 TT |
3838 | if (allocated > map->m_len) |
3839 | allocated = map->m_len; | |
a86c6181 | 3840 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3841 | map->m_flags |= EXT4_MAP_MAPPED; |
3842 | map->m_pblk = newblock; | |
3843 | map->m_len = allocated; | |
a86c6181 AT |
3844 | out2: |
3845 | if (path) { | |
3846 | ext4_ext_drop_refs(path); | |
3847 | kfree(path); | |
3848 | } | |
0562e0ba JZ |
3849 | trace_ext4_ext_map_blocks_exit(inode, map->m_lblk, |
3850 | newblock, map->m_len, err ? err : allocated); | |
e861304b AH |
3851 | |
3852 | result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ? | |
3853 | punched_out : allocated; | |
3854 | ||
3855 | return err ? err : result; | |
a86c6181 AT |
3856 | } |
3857 | ||
cf108bca | 3858 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
3859 | { |
3860 | struct address_space *mapping = inode->i_mapping; | |
3861 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 3862 | ext4_lblk_t last_block; |
a86c6181 | 3863 | handle_t *handle; |
189e868f | 3864 | loff_t page_len; |
a86c6181 AT |
3865 | int err = 0; |
3866 | ||
3889fd57 JZ |
3867 | /* |
3868 | * finish any pending end_io work so we won't run the risk of | |
3869 | * converting any truncated blocks to initialized later | |
3870 | */ | |
3871 | ext4_flush_completed_IO(inode); | |
3872 | ||
a86c6181 AT |
3873 | /* |
3874 | * probably first extent we're gonna free will be last in block | |
3875 | */ | |
f3bd1f3f | 3876 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 3877 | handle = ext4_journal_start(inode, err); |
cf108bca | 3878 | if (IS_ERR(handle)) |
a86c6181 | 3879 | return; |
a86c6181 | 3880 | |
189e868f AH |
3881 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
3882 | page_len = PAGE_CACHE_SIZE - | |
3883 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3884 | ||
3885 | err = ext4_discard_partial_page_buffers(handle, | |
3886 | mapping, inode->i_size, page_len, 0); | |
3887 | ||
3888 | if (err) | |
3889 | goto out_stop; | |
3890 | } | |
a86c6181 | 3891 | |
9ddfc3dc JK |
3892 | if (ext4_orphan_add(handle, inode)) |
3893 | goto out_stop; | |
3894 | ||
0e855ac8 | 3895 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
3896 | ext4_ext_invalidate_cache(inode); |
3897 | ||
c2ea3fde | 3898 | ext4_discard_preallocations(inode); |
c9de560d | 3899 | |
a86c6181 | 3900 | /* |
d0d856e8 RD |
3901 | * TODO: optimization is possible here. |
3902 | * Probably we need not scan at all, | |
3903 | * because page truncation is enough. | |
a86c6181 | 3904 | */ |
a86c6181 AT |
3905 | |
3906 | /* we have to know where to truncate from in crash case */ | |
3907 | EXT4_I(inode)->i_disksize = inode->i_size; | |
3908 | ext4_mark_inode_dirty(handle, inode); | |
3909 | ||
3910 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
3911 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
c6a0371c | 3912 | err = ext4_ext_remove_space(inode, last_block); |
a86c6181 AT |
3913 | |
3914 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
3915 | * transaction synchronous. |
3916 | */ | |
a86c6181 | 3917 | if (IS_SYNC(inode)) |
0390131b | 3918 | ext4_handle_sync(handle); |
a86c6181 | 3919 | |
9ddfc3dc | 3920 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
3921 | |
3922 | out_stop: | |
a86c6181 | 3923 | /* |
d0d856e8 | 3924 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
3925 | * then we need to clear up the orphan record which we created above. |
3926 | * However, if this was a real unlink then we were called by | |
3927 | * ext4_delete_inode(), and we allow that function to clean up the | |
3928 | * orphan info for us. | |
3929 | */ | |
3930 | if (inode->i_nlink) | |
3931 | ext4_orphan_del(handle, inode); | |
3932 | ||
ef737728 SR |
3933 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3934 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
3935 | ext4_journal_stop(handle); |
3936 | } | |
3937 | ||
fd28784a AK |
3938 | static void ext4_falloc_update_inode(struct inode *inode, |
3939 | int mode, loff_t new_size, int update_ctime) | |
3940 | { | |
3941 | struct timespec now; | |
3942 | ||
3943 | if (update_ctime) { | |
3944 | now = current_fs_time(inode->i_sb); | |
3945 | if (!timespec_equal(&inode->i_ctime, &now)) | |
3946 | inode->i_ctime = now; | |
3947 | } | |
3948 | /* | |
3949 | * Update only when preallocation was requested beyond | |
3950 | * the file size. | |
3951 | */ | |
cf17fea6 AK |
3952 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
3953 | if (new_size > i_size_read(inode)) | |
3954 | i_size_write(inode, new_size); | |
3955 | if (new_size > EXT4_I(inode)->i_disksize) | |
3956 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
3957 | } else { |
3958 | /* | |
3959 | * Mark that we allocate beyond EOF so the subsequent truncate | |
3960 | * can proceed even if the new size is the same as i_size. | |
3961 | */ | |
3962 | if (new_size > i_size_read(inode)) | |
12e9b892 | 3963 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
3964 | } |
3965 | ||
3966 | } | |
3967 | ||
a2df2a63 | 3968 | /* |
2fe17c10 | 3969 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
3970 | * operation, which gets called from sys_fallocate system call. |
3971 | * For block-mapped files, posix_fallocate should fall back to the method | |
3972 | * of writing zeroes to the required new blocks (the same behavior which is | |
3973 | * expected for file systems which do not support fallocate() system call). | |
3974 | */ | |
2fe17c10 | 3975 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 3976 | { |
2fe17c10 | 3977 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 3978 | handle_t *handle; |
fd28784a | 3979 | loff_t new_size; |
498e5f24 | 3980 | unsigned int max_blocks; |
a2df2a63 AA |
3981 | int ret = 0; |
3982 | int ret2 = 0; | |
3983 | int retries = 0; | |
2ed88685 | 3984 | struct ext4_map_blocks map; |
a2df2a63 AA |
3985 | unsigned int credits, blkbits = inode->i_blkbits; |
3986 | ||
3987 | /* | |
3988 | * currently supporting (pre)allocate mode for extent-based | |
3989 | * files _only_ | |
3990 | */ | |
12e9b892 | 3991 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
a2df2a63 AA |
3992 | return -EOPNOTSUPP; |
3993 | ||
a4bb6b64 AH |
3994 | /* Return error if mode is not supported */ |
3995 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
3996 | return -EOPNOTSUPP; | |
3997 | ||
3998 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
3999 | return ext4_punch_hole(file, offset, len); | |
4000 | ||
0562e0ba | 4001 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4002 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4003 | /* |
4004 | * We can't just convert len to max_blocks because | |
4005 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4006 | */ | |
a2df2a63 | 4007 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4008 | - map.m_lblk; |
a2df2a63 | 4009 | /* |
f3bd1f3f | 4010 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4011 | */ |
f3bd1f3f | 4012 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4013 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4014 | ret = inode_newsize_ok(inode, (len + offset)); |
4015 | if (ret) { | |
4016 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4017 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4018 | return ret; |
4019 | } | |
a2df2a63 AA |
4020 | retry: |
4021 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4022 | map.m_lblk = map.m_lblk + ret; |
4023 | map.m_len = max_blocks = max_blocks - ret; | |
a2df2a63 AA |
4024 | handle = ext4_journal_start(inode, credits); |
4025 | if (IS_ERR(handle)) { | |
4026 | ret = PTR_ERR(handle); | |
4027 | break; | |
4028 | } | |
2ed88685 | 4029 | ret = ext4_map_blocks(handle, inode, &map, |
556b27ab VH |
4030 | EXT4_GET_BLOCKS_CREATE_UNINIT_EXT | |
4031 | EXT4_GET_BLOCKS_NO_NORMALIZE); | |
221879c9 | 4032 | if (ret <= 0) { |
2c98615d AK |
4033 | #ifdef EXT4FS_DEBUG |
4034 | WARN_ON(ret <= 0); | |
e35fd660 | 4035 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
2c98615d | 4036 | "returned error inode#%lu, block=%u, " |
9fd9784c | 4037 | "max_blocks=%u", __func__, |
a6371b63 | 4038 | inode->i_ino, map.m_lblk, max_blocks); |
2c98615d | 4039 | #endif |
a2df2a63 AA |
4040 | ext4_mark_inode_dirty(handle, inode); |
4041 | ret2 = ext4_journal_stop(handle); | |
4042 | break; | |
4043 | } | |
2ed88685 | 4044 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4045 | blkbits) >> blkbits)) |
4046 | new_size = offset + len; | |
4047 | else | |
29ae07b7 | 4048 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4049 | |
fd28784a | 4050 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4051 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 AA |
4052 | ext4_mark_inode_dirty(handle, inode); |
4053 | ret2 = ext4_journal_stop(handle); | |
4054 | if (ret2) | |
4055 | break; | |
4056 | } | |
fd28784a AK |
4057 | if (ret == -ENOSPC && |
4058 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4059 | ret = 0; | |
a2df2a63 | 4060 | goto retry; |
a2df2a63 | 4061 | } |
55bd725a | 4062 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4063 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4064 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4065 | return ret > 0 ? ret2 : ret; |
4066 | } | |
6873fa0d | 4067 | |
0031462b MC |
4068 | /* |
4069 | * This function convert a range of blocks to written extents | |
4070 | * The caller of this function will pass the start offset and the size. | |
4071 | * all unwritten extents within this range will be converted to | |
4072 | * written extents. | |
4073 | * | |
4074 | * This function is called from the direct IO end io call back | |
4075 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4076 | * Returns 0 on success. |
0031462b MC |
4077 | */ |
4078 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4079 | ssize_t len) |
0031462b MC |
4080 | { |
4081 | handle_t *handle; | |
0031462b MC |
4082 | unsigned int max_blocks; |
4083 | int ret = 0; | |
4084 | int ret2 = 0; | |
2ed88685 | 4085 | struct ext4_map_blocks map; |
0031462b MC |
4086 | unsigned int credits, blkbits = inode->i_blkbits; |
4087 | ||
2ed88685 | 4088 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4089 | /* |
4090 | * We can't just convert len to max_blocks because | |
4091 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4092 | */ | |
2ed88685 TT |
4093 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4094 | map.m_lblk); | |
0031462b MC |
4095 | /* |
4096 | * credits to insert 1 extent into extent tree | |
4097 | */ | |
4098 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4099 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4100 | map.m_lblk += ret; |
4101 | map.m_len = (max_blocks -= ret); | |
0031462b MC |
4102 | handle = ext4_journal_start(inode, credits); |
4103 | if (IS_ERR(handle)) { | |
4104 | ret = PTR_ERR(handle); | |
4105 | break; | |
4106 | } | |
2ed88685 | 4107 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4108 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
0031462b MC |
4109 | if (ret <= 0) { |
4110 | WARN_ON(ret <= 0); | |
e35fd660 | 4111 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
0031462b MC |
4112 | "returned error inode#%lu, block=%u, " |
4113 | "max_blocks=%u", __func__, | |
2ed88685 | 4114 | inode->i_ino, map.m_lblk, map.m_len); |
0031462b MC |
4115 | } |
4116 | ext4_mark_inode_dirty(handle, inode); | |
4117 | ret2 = ext4_journal_stop(handle); | |
4118 | if (ret <= 0 || ret2 ) | |
4119 | break; | |
4120 | } | |
4121 | return ret > 0 ? ret2 : ret; | |
4122 | } | |
6d9c85eb | 4123 | |
6873fa0d ES |
4124 | /* |
4125 | * Callback function called for each extent to gather FIEMAP information. | |
4126 | */ | |
c03f8aa9 | 4127 | static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next, |
6873fa0d ES |
4128 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
4129 | void *data) | |
4130 | { | |
6873fa0d ES |
4131 | __u64 logical; |
4132 | __u64 physical; | |
4133 | __u64 length; | |
4134 | __u32 flags = 0; | |
6d9c85eb YY |
4135 | int ret = 0; |
4136 | struct fiemap_extent_info *fieinfo = data; | |
4137 | unsigned char blksize_bits; | |
6873fa0d | 4138 | |
6d9c85eb YY |
4139 | blksize_bits = inode->i_sb->s_blocksize_bits; |
4140 | logical = (__u64)newex->ec_block << blksize_bits; | |
6873fa0d | 4141 | |
b05e6ae5 | 4142 | if (newex->ec_start == 0) { |
6d9c85eb YY |
4143 | /* |
4144 | * No extent in extent-tree contains block @newex->ec_start, | |
4145 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
4146 | * | |
4147 | * Holes or delayed-extents are processed as follows. | |
4148 | * 1. lookup dirty pages with specified range in pagecache. | |
4149 | * If no page is got, then there is no delayed-extent and | |
4150 | * return with EXT_CONTINUE. | |
4151 | * 2. find the 1st mapped buffer, | |
4152 | * 3. check if the mapped buffer is both in the request range | |
4153 | * and a delayed buffer. If not, there is no delayed-extent, | |
4154 | * then return. | |
4155 | * 4. a delayed-extent is found, the extent will be collected. | |
4156 | */ | |
4157 | ext4_lblk_t end = 0; | |
4158 | pgoff_t last_offset; | |
4159 | pgoff_t offset; | |
4160 | pgoff_t index; | |
b221349f | 4161 | pgoff_t start_index = 0; |
6d9c85eb | 4162 | struct page **pages = NULL; |
6873fa0d | 4163 | struct buffer_head *bh = NULL; |
6d9c85eb YY |
4164 | struct buffer_head *head = NULL; |
4165 | unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *); | |
4166 | ||
4167 | pages = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4168 | if (pages == NULL) | |
4169 | return -ENOMEM; | |
6873fa0d ES |
4170 | |
4171 | offset = logical >> PAGE_SHIFT; | |
6d9c85eb YY |
4172 | repeat: |
4173 | last_offset = offset; | |
4174 | head = NULL; | |
4175 | ret = find_get_pages_tag(inode->i_mapping, &offset, | |
4176 | PAGECACHE_TAG_DIRTY, nr_pages, pages); | |
4177 | ||
4178 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4179 | /* First time, try to find a mapped buffer. */ | |
4180 | if (ret == 0) { | |
4181 | out: | |
4182 | for (index = 0; index < ret; index++) | |
4183 | page_cache_release(pages[index]); | |
4184 | /* just a hole. */ | |
4185 | kfree(pages); | |
4186 | return EXT_CONTINUE; | |
4187 | } | |
b221349f | 4188 | index = 0; |
6873fa0d | 4189 | |
b221349f | 4190 | next_page: |
6d9c85eb | 4191 | /* Try to find the 1st mapped buffer. */ |
b221349f | 4192 | end = ((__u64)pages[index]->index << PAGE_SHIFT) >> |
6d9c85eb | 4193 | blksize_bits; |
b221349f | 4194 | if (!page_has_buffers(pages[index])) |
6d9c85eb | 4195 | goto out; |
b221349f | 4196 | head = page_buffers(pages[index]); |
6d9c85eb YY |
4197 | if (!head) |
4198 | goto out; | |
6873fa0d | 4199 | |
b221349f | 4200 | index++; |
6d9c85eb YY |
4201 | bh = head; |
4202 | do { | |
b221349f YY |
4203 | if (end >= newex->ec_block + |
4204 | newex->ec_len) | |
4205 | /* The buffer is out of | |
4206 | * the request range. | |
4207 | */ | |
4208 | goto out; | |
4209 | ||
4210 | if (buffer_mapped(bh) && | |
4211 | end >= newex->ec_block) { | |
4212 | start_index = index - 1; | |
6d9c85eb | 4213 | /* get the 1st mapped buffer. */ |
6d9c85eb YY |
4214 | goto found_mapped_buffer; |
4215 | } | |
b221349f | 4216 | |
6d9c85eb YY |
4217 | bh = bh->b_this_page; |
4218 | end++; | |
4219 | } while (bh != head); | |
6873fa0d | 4220 | |
b221349f YY |
4221 | /* No mapped buffer in the range found in this page, |
4222 | * We need to look up next page. | |
4223 | */ | |
4224 | if (index >= ret) { | |
4225 | /* There is no page left, but we need to limit | |
4226 | * newex->ec_len. | |
4227 | */ | |
4228 | newex->ec_len = end - newex->ec_block; | |
4229 | goto out; | |
4230 | } | |
4231 | goto next_page; | |
6873fa0d | 4232 | } else { |
6d9c85eb YY |
4233 | /*Find contiguous delayed buffers. */ |
4234 | if (ret > 0 && pages[0]->index == last_offset) | |
4235 | head = page_buffers(pages[0]); | |
4236 | bh = head; | |
b221349f YY |
4237 | index = 1; |
4238 | start_index = 0; | |
6873fa0d | 4239 | } |
6d9c85eb YY |
4240 | |
4241 | found_mapped_buffer: | |
4242 | if (bh != NULL && buffer_delay(bh)) { | |
4243 | /* 1st or contiguous delayed buffer found. */ | |
4244 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4245 | /* | |
4246 | * 1st delayed buffer found, record | |
4247 | * the start of extent. | |
4248 | */ | |
4249 | flags |= FIEMAP_EXTENT_DELALLOC; | |
4250 | newex->ec_block = end; | |
4251 | logical = (__u64)end << blksize_bits; | |
4252 | } | |
4253 | /* Find contiguous delayed buffers. */ | |
4254 | do { | |
4255 | if (!buffer_delay(bh)) | |
4256 | goto found_delayed_extent; | |
4257 | bh = bh->b_this_page; | |
4258 | end++; | |
4259 | } while (bh != head); | |
4260 | ||
b221349f | 4261 | for (; index < ret; index++) { |
6d9c85eb YY |
4262 | if (!page_has_buffers(pages[index])) { |
4263 | bh = NULL; | |
4264 | break; | |
4265 | } | |
4266 | head = page_buffers(pages[index]); | |
4267 | if (!head) { | |
4268 | bh = NULL; | |
4269 | break; | |
4270 | } | |
b221349f | 4271 | |
6d9c85eb | 4272 | if (pages[index]->index != |
b221349f YY |
4273 | pages[start_index]->index + index |
4274 | - start_index) { | |
6d9c85eb YY |
4275 | /* Blocks are not contiguous. */ |
4276 | bh = NULL; | |
4277 | break; | |
4278 | } | |
4279 | bh = head; | |
4280 | do { | |
4281 | if (!buffer_delay(bh)) | |
4282 | /* Delayed-extent ends. */ | |
4283 | goto found_delayed_extent; | |
4284 | bh = bh->b_this_page; | |
4285 | end++; | |
4286 | } while (bh != head); | |
4287 | } | |
4288 | } else if (!(flags & FIEMAP_EXTENT_DELALLOC)) | |
4289 | /* a hole found. */ | |
4290 | goto out; | |
4291 | ||
4292 | found_delayed_extent: | |
4293 | newex->ec_len = min(end - newex->ec_block, | |
4294 | (ext4_lblk_t)EXT_INIT_MAX_LEN); | |
4295 | if (ret == nr_pages && bh != NULL && | |
4296 | newex->ec_len < EXT_INIT_MAX_LEN && | |
4297 | buffer_delay(bh)) { | |
4298 | /* Have not collected an extent and continue. */ | |
4299 | for (index = 0; index < ret; index++) | |
4300 | page_cache_release(pages[index]); | |
4301 | goto repeat; | |
6873fa0d | 4302 | } |
6d9c85eb YY |
4303 | |
4304 | for (index = 0; index < ret; index++) | |
4305 | page_cache_release(pages[index]); | |
4306 | kfree(pages); | |
6873fa0d ES |
4307 | } |
4308 | ||
4309 | physical = (__u64)newex->ec_start << blksize_bits; | |
4310 | length = (__u64)newex->ec_len << blksize_bits; | |
4311 | ||
4312 | if (ex && ext4_ext_is_uninitialized(ex)) | |
4313 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
4314 | ||
c03f8aa9 | 4315 | if (next == EXT_MAX_BLOCKS) |
6873fa0d ES |
4316 | flags |= FIEMAP_EXTENT_LAST; |
4317 | ||
6d9c85eb | 4318 | ret = fiemap_fill_next_extent(fieinfo, logical, physical, |
6873fa0d | 4319 | length, flags); |
6d9c85eb YY |
4320 | if (ret < 0) |
4321 | return ret; | |
4322 | if (ret == 1) | |
6873fa0d | 4323 | return EXT_BREAK; |
6873fa0d ES |
4324 | return EXT_CONTINUE; |
4325 | } | |
6873fa0d ES |
4326 | /* fiemap flags we can handle specified here */ |
4327 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4328 | ||
3a06d778 AK |
4329 | static int ext4_xattr_fiemap(struct inode *inode, |
4330 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4331 | { |
4332 | __u64 physical = 0; | |
4333 | __u64 length; | |
4334 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4335 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4336 | int error = 0; | |
4337 | ||
4338 | /* in-inode? */ | |
19f5fb7a | 4339 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4340 | struct ext4_iloc iloc; |
4341 | int offset; /* offset of xattr in inode */ | |
4342 | ||
4343 | error = ext4_get_inode_loc(inode, &iloc); | |
4344 | if (error) | |
4345 | return error; | |
4346 | physical = iloc.bh->b_blocknr << blockbits; | |
4347 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4348 | EXT4_I(inode)->i_extra_isize; | |
4349 | physical += offset; | |
4350 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4351 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4352 | brelse(iloc.bh); |
6873fa0d ES |
4353 | } else { /* external block */ |
4354 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4355 | length = inode->i_sb->s_blocksize; | |
4356 | } | |
4357 | ||
4358 | if (physical) | |
4359 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4360 | length, flags); | |
4361 | return (error < 0 ? error : 0); | |
4362 | } | |
4363 | ||
a4bb6b64 AH |
4364 | /* |
4365 | * ext4_ext_punch_hole | |
4366 | * | |
4367 | * Punches a hole of "length" bytes in a file starting | |
4368 | * at byte "offset" | |
4369 | * | |
4370 | * @inode: The inode of the file to punch a hole in | |
4371 | * @offset: The starting byte offset of the hole | |
4372 | * @length: The length of the hole | |
4373 | * | |
4374 | * Returns the number of blocks removed or negative on err | |
4375 | */ | |
4376 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4377 | { | |
4378 | struct inode *inode = file->f_path.dentry->d_inode; | |
4379 | struct super_block *sb = inode->i_sb; | |
4380 | struct ext4_ext_cache cache_ex; | |
4381 | ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks; | |
4382 | struct address_space *mapping = inode->i_mapping; | |
4383 | struct ext4_map_blocks map; | |
4384 | handle_t *handle; | |
ba06208a AH |
4385 | loff_t first_page, last_page, page_len; |
4386 | loff_t first_page_offset, last_page_offset; | |
a4bb6b64 AH |
4387 | int ret, credits, blocks_released, err = 0; |
4388 | ||
2be4751b AH |
4389 | /* No need to punch hole beyond i_size */ |
4390 | if (offset >= inode->i_size) | |
4391 | return 0; | |
4392 | ||
4393 | /* | |
4394 | * If the hole extends beyond i_size, set the hole | |
4395 | * to end after the page that contains i_size | |
4396 | */ | |
4397 | if (offset + length > inode->i_size) { | |
4398 | length = inode->i_size + | |
4399 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4400 | offset; | |
4401 | } | |
4402 | ||
a4bb6b64 AH |
4403 | first_block = (offset + sb->s_blocksize - 1) >> |
4404 | EXT4_BLOCK_SIZE_BITS(sb); | |
4405 | last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4406 | ||
a4bb6b64 AH |
4407 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4408 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4409 | ||
4410 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4411 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4412 | ||
4413 | /* | |
4414 | * Write out all dirty pages to avoid race conditions | |
4415 | * Then release them. | |
4416 | */ | |
4417 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4418 | err = filemap_write_and_wait_range(mapping, | |
2be4751b | 4419 | offset, offset + length - 1); |
a4bb6b64 | 4420 | |
2be4751b AH |
4421 | if (err) |
4422 | return err; | |
a4bb6b64 AH |
4423 | } |
4424 | ||
4425 | /* Now release the pages */ | |
4426 | if (last_page_offset > first_page_offset) { | |
4427 | truncate_inode_pages_range(mapping, first_page_offset, | |
4428 | last_page_offset-1); | |
4429 | } | |
4430 | ||
4431 | /* finish any pending end_io work */ | |
4432 | ext4_flush_completed_IO(inode); | |
4433 | ||
4434 | credits = ext4_writepage_trans_blocks(inode); | |
4435 | handle = ext4_journal_start(inode, credits); | |
4436 | if (IS_ERR(handle)) | |
4437 | return PTR_ERR(handle); | |
4438 | ||
4439 | err = ext4_orphan_add(handle, inode); | |
4440 | if (err) | |
4441 | goto out; | |
4442 | ||
4443 | /* | |
ba06208a AH |
4444 | * Now we need to zero out the non-page-aligned data in the |
4445 | * pages at the start and tail of the hole, and unmap the buffer | |
4446 | * heads for the block aligned regions of the page that were | |
4447 | * completely zeroed. | |
a4bb6b64 | 4448 | */ |
ba06208a AH |
4449 | if (first_page > last_page) { |
4450 | /* | |
4451 | * If the file space being truncated is contained within a page | |
4452 | * just zero out and unmap the middle of that page | |
4453 | */ | |
4454 | err = ext4_discard_partial_page_buffers(handle, | |
4455 | mapping, offset, length, 0); | |
4456 | ||
4457 | if (err) | |
4458 | goto out; | |
4459 | } else { | |
4460 | /* | |
4461 | * zero out and unmap the partial page that contains | |
4462 | * the start of the hole | |
4463 | */ | |
4464 | page_len = first_page_offset - offset; | |
4465 | if (page_len > 0) { | |
4466 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4467 | offset, page_len, 0); | |
4468 | if (err) | |
4469 | goto out; | |
4470 | } | |
4471 | ||
4472 | /* | |
4473 | * zero out and unmap the partial page that contains | |
4474 | * the end of the hole | |
4475 | */ | |
4476 | page_len = offset + length - last_page_offset; | |
4477 | if (page_len > 0) { | |
4478 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4479 | last_page_offset, page_len, 0); | |
4480 | if (err) | |
4481 | goto out; | |
a4bb6b64 AH |
4482 | } |
4483 | } | |
4484 | ||
2be4751b AH |
4485 | |
4486 | /* | |
4487 | * If i_size is contained in the last page, we need to | |
4488 | * unmap and zero the partial page after i_size | |
4489 | */ | |
4490 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4491 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4492 | ||
4493 | page_len = PAGE_CACHE_SIZE - | |
4494 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4495 | ||
4496 | if (page_len > 0) { | |
4497 | err = ext4_discard_partial_page_buffers(handle, | |
4498 | mapping, inode->i_size, page_len, 0); | |
4499 | ||
4500 | if (err) | |
4501 | goto out; | |
4502 | } | |
4503 | } | |
4504 | ||
a4bb6b64 AH |
4505 | /* If there are no blocks to remove, return now */ |
4506 | if (first_block >= last_block) | |
4507 | goto out; | |
4508 | ||
4509 | down_write(&EXT4_I(inode)->i_data_sem); | |
4510 | ext4_ext_invalidate_cache(inode); | |
4511 | ext4_discard_preallocations(inode); | |
4512 | ||
4513 | /* | |
4514 | * Loop over all the blocks and identify blocks | |
4515 | * that need to be punched out | |
4516 | */ | |
4517 | iblock = first_block; | |
4518 | blocks_released = 0; | |
4519 | while (iblock < last_block) { | |
4520 | max_blocks = last_block - iblock; | |
4521 | num_blocks = 1; | |
4522 | memset(&map, 0, sizeof(map)); | |
4523 | map.m_lblk = iblock; | |
4524 | map.m_len = max_blocks; | |
4525 | ret = ext4_ext_map_blocks(handle, inode, &map, | |
4526 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
4527 | ||
4528 | if (ret > 0) { | |
4529 | blocks_released += ret; | |
4530 | num_blocks = ret; | |
4531 | } else if (ret == 0) { | |
4532 | /* | |
4533 | * If map blocks could not find the block, | |
4534 | * then it is in a hole. If the hole was | |
4535 | * not already cached, then map blocks should | |
4536 | * put it in the cache. So we can get the hole | |
4537 | * out of the cache | |
4538 | */ | |
4539 | memset(&cache_ex, 0, sizeof(cache_ex)); | |
4540 | if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) && | |
4541 | !cache_ex.ec_start) { | |
4542 | ||
4543 | /* The hole is cached */ | |
4544 | num_blocks = cache_ex.ec_block + | |
4545 | cache_ex.ec_len - iblock; | |
4546 | ||
4547 | } else { | |
4548 | /* The block could not be identified */ | |
4549 | err = -EIO; | |
4550 | break; | |
4551 | } | |
4552 | } else { | |
4553 | /* Map blocks error */ | |
4554 | err = ret; | |
4555 | break; | |
4556 | } | |
4557 | ||
4558 | if (num_blocks == 0) { | |
4559 | /* This condition should never happen */ | |
4560 | ext_debug("Block lookup failed"); | |
4561 | err = -EIO; | |
4562 | break; | |
4563 | } | |
4564 | ||
4565 | iblock += num_blocks; | |
4566 | } | |
4567 | ||
4568 | if (blocks_released > 0) { | |
4569 | ext4_ext_invalidate_cache(inode); | |
4570 | ext4_discard_preallocations(inode); | |
4571 | } | |
4572 | ||
4573 | if (IS_SYNC(inode)) | |
4574 | ext4_handle_sync(handle); | |
4575 | ||
4576 | up_write(&EXT4_I(inode)->i_data_sem); | |
4577 | ||
4578 | out: | |
4579 | ext4_orphan_del(handle, inode); | |
4580 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
4581 | ext4_mark_inode_dirty(handle, inode); | |
4582 | ext4_journal_stop(handle); | |
4583 | return err; | |
4584 | } | |
6873fa0d ES |
4585 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4586 | __u64 start, __u64 len) | |
4587 | { | |
4588 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4589 | int error = 0; |
4590 | ||
4591 | /* fallback to generic here if not in extents fmt */ | |
12e9b892 | 4592 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4593 | return generic_block_fiemap(inode, fieinfo, start, len, |
4594 | ext4_get_block); | |
4595 | ||
4596 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4597 | return -EBADR; | |
4598 | ||
4599 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4600 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4601 | } else { | |
aca92ff6 LM |
4602 | ext4_lblk_t len_blks; |
4603 | __u64 last_blk; | |
4604 | ||
6873fa0d | 4605 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4606 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4607 | if (last_blk >= EXT_MAX_BLOCKS) |
4608 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4609 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4610 | |
4611 | /* | |
4612 | * Walk the extent tree gathering extent information. | |
4613 | * ext4_ext_fiemap_cb will push extents back to user. | |
4614 | */ | |
6873fa0d ES |
4615 | error = ext4_ext_walk_space(inode, start_blk, len_blks, |
4616 | ext4_ext_fiemap_cb, fieinfo); | |
6873fa0d ES |
4617 | } |
4618 | ||
4619 | return error; | |
4620 | } |