]>
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 | ||
487caeef JK |
49 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
50 | struct inode *inode, | |
51 | int needed) | |
a86c6181 AT |
52 | { |
53 | int err; | |
54 | ||
0390131b FM |
55 | if (!ext4_handle_valid(handle)) |
56 | return 0; | |
a86c6181 | 57 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
58 | return 0; |
59 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 60 | if (err <= 0) |
9102e4fa | 61 | return err; |
487caeef | 62 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
63 | if (err == 0) |
64 | err = -EAGAIN; | |
487caeef JK |
65 | |
66 | return err; | |
a86c6181 AT |
67 | } |
68 | ||
69 | /* | |
70 | * could return: | |
71 | * - EROFS | |
72 | * - ENOMEM | |
73 | */ | |
74 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
75 | struct ext4_ext_path *path) | |
76 | { | |
77 | if (path->p_bh) { | |
78 | /* path points to block */ | |
79 | return ext4_journal_get_write_access(handle, path->p_bh); | |
80 | } | |
81 | /* path points to leaf/index in inode body */ | |
82 | /* we use in-core data, no need to protect them */ | |
83 | return 0; | |
84 | } | |
85 | ||
86 | /* | |
87 | * could return: | |
88 | * - EROFS | |
89 | * - ENOMEM | |
90 | * - EIO | |
91 | */ | |
92 | static int ext4_ext_dirty(handle_t *handle, struct inode *inode, | |
93 | struct ext4_ext_path *path) | |
94 | { | |
95 | int err; | |
96 | if (path->p_bh) { | |
97 | /* path points to block */ | |
0390131b | 98 | err = ext4_handle_dirty_metadata(handle, inode, path->p_bh); |
a86c6181 AT |
99 | } else { |
100 | /* path points to leaf/index in inode body */ | |
101 | err = ext4_mark_inode_dirty(handle, inode); | |
102 | } | |
103 | return err; | |
104 | } | |
105 | ||
f65e6fba | 106 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 107 | struct ext4_ext_path *path, |
725d26d3 | 108 | ext4_lblk_t block) |
a86c6181 AT |
109 | { |
110 | struct ext4_inode_info *ei = EXT4_I(inode); | |
f65e6fba | 111 | ext4_fsblk_t bg_start; |
74d3487f | 112 | ext4_fsblk_t last_block; |
f65e6fba | 113 | ext4_grpblk_t colour; |
a4912123 TT |
114 | ext4_group_t block_group; |
115 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
a86c6181 AT |
116 | int depth; |
117 | ||
118 | if (path) { | |
119 | struct ext4_extent *ex; | |
120 | depth = path->p_depth; | |
121 | ||
ad4fb9ca KM |
122 | /* |
123 | * Try to predict block placement assuming that we are | |
124 | * filling in a file which will eventually be | |
125 | * non-sparse --- i.e., in the case of libbfd writing | |
126 | * an ELF object sections out-of-order but in a way | |
127 | * the eventually results in a contiguous object or | |
128 | * executable file, or some database extending a table | |
129 | * space file. However, this is actually somewhat | |
130 | * non-ideal if we are writing a sparse file such as | |
131 | * qemu or KVM writing a raw image file that is going | |
132 | * to stay fairly sparse, since it will end up | |
133 | * fragmenting the file system's free space. Maybe we | |
134 | * should have some hueristics or some way to allow | |
135 | * userspace to pass a hint to file system, | |
b8d6568a | 136 | * especially if the latter case turns out to be |
ad4fb9ca KM |
137 | * common. |
138 | */ | |
7e028976 | 139 | ex = path[depth].p_ext; |
ad4fb9ca KM |
140 | if (ex) { |
141 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
142 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
143 | ||
144 | if (block > ext_block) | |
145 | return ext_pblk + (block - ext_block); | |
146 | else | |
147 | return ext_pblk - (ext_block - block); | |
148 | } | |
a86c6181 | 149 | |
d0d856e8 RD |
150 | /* it looks like index is empty; |
151 | * try to find starting block from index itself */ | |
a86c6181 AT |
152 | if (path[depth].p_bh) |
153 | return path[depth].p_bh->b_blocknr; | |
154 | } | |
155 | ||
156 | /* OK. use inode's group */ | |
a4912123 TT |
157 | block_group = ei->i_block_group; |
158 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
159 | /* | |
160 | * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME | |
60e6679e TT |
161 | * block groups per flexgroup, reserve the first block |
162 | * group for directories and special files. Regular | |
a4912123 | 163 | * files will start at the second block group. This |
60e6679e | 164 | * tends to speed up directory access and improves |
a4912123 TT |
165 | * fsck times. |
166 | */ | |
167 | block_group &= ~(flex_size-1); | |
168 | if (S_ISREG(inode->i_mode)) | |
169 | block_group++; | |
170 | } | |
5661bd68 | 171 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); |
74d3487f VC |
172 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
173 | ||
a4912123 TT |
174 | /* |
175 | * If we are doing delayed allocation, we don't need take | |
176 | * colour into account. | |
177 | */ | |
178 | if (test_opt(inode->i_sb, DELALLOC)) | |
179 | return bg_start; | |
180 | ||
74d3487f VC |
181 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
182 | colour = (current->pid % 16) * | |
a86c6181 | 183 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
184 | else |
185 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
a86c6181 AT |
186 | return bg_start + colour + block; |
187 | } | |
188 | ||
654b4908 AK |
189 | /* |
190 | * Allocation for a meta data block | |
191 | */ | |
f65e6fba | 192 | static ext4_fsblk_t |
654b4908 | 193 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 194 | struct ext4_ext_path *path, |
55f020db | 195 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 196 | { |
f65e6fba | 197 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
198 | |
199 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
200 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
201 | NULL, err); | |
a86c6181 AT |
202 | return newblock; |
203 | } | |
204 | ||
55ad63bf | 205 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
206 | { |
207 | int size; | |
208 | ||
209 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
210 | / sizeof(struct ext4_extent); | |
55ad63bf | 211 | if (!check) { |
bbf2f9fb | 212 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
213 | if (size > 6) |
214 | size = 6; | |
a86c6181 | 215 | #endif |
55ad63bf | 216 | } |
a86c6181 AT |
217 | return size; |
218 | } | |
219 | ||
55ad63bf | 220 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
221 | { |
222 | int size; | |
223 | ||
224 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
225 | / sizeof(struct ext4_extent_idx); | |
55ad63bf | 226 | if (!check) { |
bbf2f9fb | 227 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
228 | if (size > 5) |
229 | size = 5; | |
a86c6181 | 230 | #endif |
55ad63bf | 231 | } |
a86c6181 AT |
232 | return size; |
233 | } | |
234 | ||
55ad63bf | 235 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
236 | { |
237 | int size; | |
238 | ||
239 | size = sizeof(EXT4_I(inode)->i_data); | |
240 | size -= sizeof(struct ext4_extent_header); | |
241 | size /= sizeof(struct ext4_extent); | |
55ad63bf | 242 | if (!check) { |
bbf2f9fb | 243 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
244 | if (size > 3) |
245 | size = 3; | |
a86c6181 | 246 | #endif |
55ad63bf | 247 | } |
a86c6181 AT |
248 | return size; |
249 | } | |
250 | ||
55ad63bf | 251 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
252 | { |
253 | int size; | |
254 | ||
255 | size = sizeof(EXT4_I(inode)->i_data); | |
256 | size -= sizeof(struct ext4_extent_header); | |
257 | size /= sizeof(struct ext4_extent_idx); | |
55ad63bf | 258 | if (!check) { |
bbf2f9fb | 259 | #ifdef AGGRESSIVE_TEST |
55ad63bf TT |
260 | if (size > 4) |
261 | size = 4; | |
a86c6181 | 262 | #endif |
55ad63bf | 263 | } |
a86c6181 AT |
264 | return size; |
265 | } | |
266 | ||
d2a17637 MC |
267 | /* |
268 | * Calculate the number of metadata blocks needed | |
269 | * to allocate @blocks | |
270 | * Worse case is one block per extent | |
271 | */ | |
01f49d0b | 272 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 273 | { |
9d0be502 TT |
274 | struct ext4_inode_info *ei = EXT4_I(inode); |
275 | int idxs, num = 0; | |
d2a17637 | 276 | |
9d0be502 TT |
277 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
278 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
279 | |
280 | /* | |
9d0be502 TT |
281 | * If the new delayed allocation block is contiguous with the |
282 | * previous da block, it can share index blocks with the | |
283 | * previous block, so we only need to allocate a new index | |
284 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
285 | * an additional index block, and at ldxs**3 blocks, yet | |
286 | * another index blocks. | |
d2a17637 | 287 | */ |
9d0be502 TT |
288 | if (ei->i_da_metadata_calc_len && |
289 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
290 | if ((ei->i_da_metadata_calc_len % idxs) == 0) | |
291 | num++; | |
292 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
293 | num++; | |
294 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
295 | num++; | |
296 | ei->i_da_metadata_calc_len = 0; | |
297 | } else | |
298 | ei->i_da_metadata_calc_len++; | |
299 | ei->i_da_metadata_calc_last_lblock++; | |
300 | return num; | |
301 | } | |
d2a17637 | 302 | |
9d0be502 TT |
303 | /* |
304 | * In the worst case we need a new set of index blocks at | |
305 | * every level of the inode's extent tree. | |
306 | */ | |
307 | ei->i_da_metadata_calc_len = 1; | |
308 | ei->i_da_metadata_calc_last_lblock = lblock; | |
309 | return ext_depth(inode) + 1; | |
d2a17637 MC |
310 | } |
311 | ||
c29c0ae7 AT |
312 | static int |
313 | ext4_ext_max_entries(struct inode *inode, int depth) | |
314 | { | |
315 | int max; | |
316 | ||
317 | if (depth == ext_depth(inode)) { | |
318 | if (depth == 0) | |
55ad63bf | 319 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 320 | else |
55ad63bf | 321 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
322 | } else { |
323 | if (depth == 0) | |
55ad63bf | 324 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 325 | else |
55ad63bf | 326 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
327 | } |
328 | ||
329 | return max; | |
330 | } | |
331 | ||
56b19868 AK |
332 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
333 | { | |
bf89d16f | 334 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 335 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 336 | |
6fd058f7 | 337 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
338 | } |
339 | ||
340 | static int ext4_valid_extent_idx(struct inode *inode, | |
341 | struct ext4_extent_idx *ext_idx) | |
342 | { | |
bf89d16f | 343 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 344 | |
6fd058f7 | 345 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
346 | } |
347 | ||
348 | static int ext4_valid_extent_entries(struct inode *inode, | |
349 | struct ext4_extent_header *eh, | |
350 | int depth) | |
351 | { | |
352 | struct ext4_extent *ext; | |
353 | struct ext4_extent_idx *ext_idx; | |
354 | unsigned short entries; | |
355 | if (eh->eh_entries == 0) | |
356 | return 1; | |
357 | ||
358 | entries = le16_to_cpu(eh->eh_entries); | |
359 | ||
360 | if (depth == 0) { | |
361 | /* leaf entries */ | |
362 | ext = EXT_FIRST_EXTENT(eh); | |
363 | while (entries) { | |
364 | if (!ext4_valid_extent(inode, ext)) | |
365 | return 0; | |
366 | ext++; | |
367 | entries--; | |
368 | } | |
369 | } else { | |
370 | ext_idx = EXT_FIRST_INDEX(eh); | |
371 | while (entries) { | |
372 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
373 | return 0; | |
374 | ext_idx++; | |
375 | entries--; | |
376 | } | |
377 | } | |
378 | return 1; | |
379 | } | |
380 | ||
c398eda0 TT |
381 | static int __ext4_ext_check(const char *function, unsigned int line, |
382 | struct inode *inode, struct ext4_extent_header *eh, | |
383 | int depth) | |
c29c0ae7 AT |
384 | { |
385 | const char *error_msg; | |
386 | int max = 0; | |
387 | ||
388 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
389 | error_msg = "invalid magic"; | |
390 | goto corrupted; | |
391 | } | |
392 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
393 | error_msg = "unexpected eh_depth"; | |
394 | goto corrupted; | |
395 | } | |
396 | if (unlikely(eh->eh_max == 0)) { | |
397 | error_msg = "invalid eh_max"; | |
398 | goto corrupted; | |
399 | } | |
400 | max = ext4_ext_max_entries(inode, depth); | |
401 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
402 | error_msg = "too large eh_max"; | |
403 | goto corrupted; | |
404 | } | |
405 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
406 | error_msg = "invalid eh_entries"; | |
407 | goto corrupted; | |
408 | } | |
56b19868 AK |
409 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
410 | error_msg = "invalid extent entries"; | |
411 | goto corrupted; | |
412 | } | |
c29c0ae7 AT |
413 | return 0; |
414 | ||
415 | corrupted: | |
c398eda0 | 416 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 417 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 418 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 419 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
420 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
421 | max, le16_to_cpu(eh->eh_depth), depth); | |
422 | ||
423 | return -EIO; | |
424 | } | |
425 | ||
56b19868 | 426 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 427 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 428 | |
7a262f7c AK |
429 | int ext4_ext_check_inode(struct inode *inode) |
430 | { | |
431 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
432 | } | |
433 | ||
a86c6181 AT |
434 | #ifdef EXT_DEBUG |
435 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
436 | { | |
437 | int k, l = path->p_depth; | |
438 | ||
439 | ext_debug("path:"); | |
440 | for (k = 0; k <= l; k++, path++) { | |
441 | if (path->p_idx) { | |
2ae02107 | 442 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 443 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 444 | } else if (path->p_ext) { |
553f9008 | 445 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 446 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 447 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 448 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 449 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
450 | } else |
451 | ext_debug(" []"); | |
452 | } | |
453 | ext_debug("\n"); | |
454 | } | |
455 | ||
456 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
457 | { | |
458 | int depth = ext_depth(inode); | |
459 | struct ext4_extent_header *eh; | |
460 | struct ext4_extent *ex; | |
461 | int i; | |
462 | ||
463 | if (!path) | |
464 | return; | |
465 | ||
466 | eh = path[depth].p_hdr; | |
467 | ex = EXT_FIRST_EXTENT(eh); | |
468 | ||
553f9008 M |
469 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
470 | ||
a86c6181 | 471 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
472 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
473 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 474 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
475 | } |
476 | ext_debug("\n"); | |
477 | } | |
478 | #else | |
af5bc92d TT |
479 | #define ext4_ext_show_path(inode, path) |
480 | #define ext4_ext_show_leaf(inode, path) | |
a86c6181 AT |
481 | #endif |
482 | ||
b35905c1 | 483 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
484 | { |
485 | int depth = path->p_depth; | |
486 | int i; | |
487 | ||
488 | for (i = 0; i <= depth; i++, path++) | |
489 | if (path->p_bh) { | |
490 | brelse(path->p_bh); | |
491 | path->p_bh = NULL; | |
492 | } | |
493 | } | |
494 | ||
495 | /* | |
d0d856e8 RD |
496 | * ext4_ext_binsearch_idx: |
497 | * binary search for the closest index of the given block | |
c29c0ae7 | 498 | * the header must be checked before calling this |
a86c6181 AT |
499 | */ |
500 | static void | |
725d26d3 AK |
501 | ext4_ext_binsearch_idx(struct inode *inode, |
502 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
503 | { |
504 | struct ext4_extent_header *eh = path->p_hdr; | |
505 | struct ext4_extent_idx *r, *l, *m; | |
506 | ||
a86c6181 | 507 | |
bba90743 | 508 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
509 | |
510 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 511 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
512 | while (l <= r) { |
513 | m = l + (r - l) / 2; | |
514 | if (block < le32_to_cpu(m->ei_block)) | |
515 | r = m - 1; | |
516 | else | |
517 | l = m + 1; | |
26d535ed DM |
518 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
519 | m, le32_to_cpu(m->ei_block), | |
520 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
521 | } |
522 | ||
523 | path->p_idx = l - 1; | |
f65e6fba | 524 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 525 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
526 | |
527 | #ifdef CHECK_BINSEARCH | |
528 | { | |
529 | struct ext4_extent_idx *chix, *ix; | |
530 | int k; | |
531 | ||
532 | chix = ix = EXT_FIRST_INDEX(eh); | |
533 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
534 | if (k != 0 && | |
535 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
536 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
537 | "first=0x%p\n", k, | |
538 | ix, EXT_FIRST_INDEX(eh)); | |
539 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
540 | le32_to_cpu(ix->ei_block), |
541 | le32_to_cpu(ix[-1].ei_block)); | |
542 | } | |
543 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 544 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
545 | if (block < le32_to_cpu(ix->ei_block)) |
546 | break; | |
547 | chix = ix; | |
548 | } | |
549 | BUG_ON(chix != path->p_idx); | |
550 | } | |
551 | #endif | |
552 | ||
553 | } | |
554 | ||
555 | /* | |
d0d856e8 RD |
556 | * ext4_ext_binsearch: |
557 | * binary search for closest extent of the given block | |
c29c0ae7 | 558 | * the header must be checked before calling this |
a86c6181 AT |
559 | */ |
560 | static void | |
725d26d3 AK |
561 | ext4_ext_binsearch(struct inode *inode, |
562 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
563 | { |
564 | struct ext4_extent_header *eh = path->p_hdr; | |
565 | struct ext4_extent *r, *l, *m; | |
566 | ||
a86c6181 AT |
567 | if (eh->eh_entries == 0) { |
568 | /* | |
d0d856e8 RD |
569 | * this leaf is empty: |
570 | * we get such a leaf in split/add case | |
a86c6181 AT |
571 | */ |
572 | return; | |
573 | } | |
574 | ||
bba90743 | 575 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
576 | |
577 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 578 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
579 | |
580 | while (l <= r) { | |
581 | m = l + (r - l) / 2; | |
582 | if (block < le32_to_cpu(m->ee_block)) | |
583 | r = m - 1; | |
584 | else | |
585 | l = m + 1; | |
26d535ed DM |
586 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
587 | m, le32_to_cpu(m->ee_block), | |
588 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
589 | } |
590 | ||
591 | path->p_ext = l - 1; | |
553f9008 | 592 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 593 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 594 | ext4_ext_pblock(path->p_ext), |
553f9008 | 595 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 596 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
597 | |
598 | #ifdef CHECK_BINSEARCH | |
599 | { | |
600 | struct ext4_extent *chex, *ex; | |
601 | int k; | |
602 | ||
603 | chex = ex = EXT_FIRST_EXTENT(eh); | |
604 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
605 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 606 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
607 | if (block < le32_to_cpu(ex->ee_block)) |
608 | break; | |
609 | chex = ex; | |
610 | } | |
611 | BUG_ON(chex != path->p_ext); | |
612 | } | |
613 | #endif | |
614 | ||
615 | } | |
616 | ||
617 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
618 | { | |
619 | struct ext4_extent_header *eh; | |
620 | ||
621 | eh = ext_inode_hdr(inode); | |
622 | eh->eh_depth = 0; | |
623 | eh->eh_entries = 0; | |
624 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 625 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
626 | ext4_mark_inode_dirty(handle, inode); |
627 | ext4_ext_invalidate_cache(inode); | |
628 | return 0; | |
629 | } | |
630 | ||
631 | struct ext4_ext_path * | |
725d26d3 AK |
632 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
633 | struct ext4_ext_path *path) | |
a86c6181 AT |
634 | { |
635 | struct ext4_extent_header *eh; | |
636 | struct buffer_head *bh; | |
637 | short int depth, i, ppos = 0, alloc = 0; | |
638 | ||
639 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 640 | depth = ext_depth(inode); |
a86c6181 AT |
641 | |
642 | /* account possible depth increase */ | |
643 | if (!path) { | |
5d4958f9 | 644 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
645 | GFP_NOFS); |
646 | if (!path) | |
647 | return ERR_PTR(-ENOMEM); | |
648 | alloc = 1; | |
649 | } | |
a86c6181 | 650 | path[0].p_hdr = eh; |
1973adcb | 651 | path[0].p_bh = NULL; |
a86c6181 | 652 | |
c29c0ae7 | 653 | i = depth; |
a86c6181 AT |
654 | /* walk through the tree */ |
655 | while (i) { | |
7a262f7c AK |
656 | int need_to_validate = 0; |
657 | ||
a86c6181 AT |
658 | ext_debug("depth %d: num %d, max %d\n", |
659 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 660 | |
a86c6181 | 661 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 662 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
663 | path[ppos].p_depth = i; |
664 | path[ppos].p_ext = NULL; | |
665 | ||
7a262f7c AK |
666 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
667 | if (unlikely(!bh)) | |
a86c6181 | 668 | goto err; |
7a262f7c | 669 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
670 | trace_ext4_ext_load_extent(inode, block, |
671 | path[ppos].p_block); | |
7a262f7c AK |
672 | if (bh_submit_read(bh) < 0) { |
673 | put_bh(bh); | |
674 | goto err; | |
675 | } | |
676 | /* validate the extent entries */ | |
677 | need_to_validate = 1; | |
678 | } | |
a86c6181 AT |
679 | eh = ext_block_hdr(bh); |
680 | ppos++; | |
273df556 FM |
681 | if (unlikely(ppos > depth)) { |
682 | put_bh(bh); | |
683 | EXT4_ERROR_INODE(inode, | |
684 | "ppos %d > depth %d", ppos, depth); | |
685 | goto err; | |
686 | } | |
a86c6181 AT |
687 | path[ppos].p_bh = bh; |
688 | path[ppos].p_hdr = eh; | |
689 | i--; | |
690 | ||
7a262f7c | 691 | if (need_to_validate && ext4_ext_check(inode, eh, i)) |
a86c6181 AT |
692 | goto err; |
693 | } | |
694 | ||
695 | path[ppos].p_depth = i; | |
a86c6181 AT |
696 | path[ppos].p_ext = NULL; |
697 | path[ppos].p_idx = NULL; | |
698 | ||
a86c6181 AT |
699 | /* find extent */ |
700 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
701 | /* if not an empty leaf */ |
702 | if (path[ppos].p_ext) | |
bf89d16f | 703 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
704 | |
705 | ext4_ext_show_path(inode, path); | |
706 | ||
707 | return path; | |
708 | ||
709 | err: | |
710 | ext4_ext_drop_refs(path); | |
711 | if (alloc) | |
712 | kfree(path); | |
713 | return ERR_PTR(-EIO); | |
714 | } | |
715 | ||
716 | /* | |
d0d856e8 RD |
717 | * ext4_ext_insert_index: |
718 | * insert new index [@logical;@ptr] into the block at @curp; | |
719 | * check where to insert: before @curp or after @curp | |
a86c6181 | 720 | */ |
1f109d5a TT |
721 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
722 | struct ext4_ext_path *curp, | |
723 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
724 | { |
725 | struct ext4_extent_idx *ix; | |
726 | int len, err; | |
727 | ||
7e028976 AM |
728 | err = ext4_ext_get_access(handle, inode, curp); |
729 | if (err) | |
a86c6181 AT |
730 | return err; |
731 | ||
273df556 FM |
732 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
733 | EXT4_ERROR_INODE(inode, | |
734 | "logical %d == ei_block %d!", | |
735 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
736 | return -EIO; | |
737 | } | |
a86c6181 AT |
738 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; |
739 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
740 | /* insert after */ | |
741 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
742 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
743 | len = len < 0 ? 0 : len; | |
26d535ed | 744 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
745 | "move %d from 0x%p to 0x%p\n", |
746 | logical, ptr, len, | |
747 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
748 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
749 | } | |
750 | ix = curp->p_idx + 1; | |
751 | } else { | |
752 | /* insert before */ | |
753 | len = len * sizeof(struct ext4_extent_idx); | |
754 | len = len < 0 ? 0 : len; | |
26d535ed | 755 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
756 | "move %d from 0x%p to 0x%p\n", |
757 | logical, ptr, len, | |
758 | curp->p_idx, (curp->p_idx + 1)); | |
759 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
760 | ix = curp->p_idx; | |
761 | } | |
762 | ||
763 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 764 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 765 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 766 | |
273df556 FM |
767 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) |
768 | > le16_to_cpu(curp->p_hdr->eh_max))) { | |
769 | EXT4_ERROR_INODE(inode, | |
770 | "logical %d == ei_block %d!", | |
771 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
772 | return -EIO; | |
773 | } | |
774 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { | |
775 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
776 | return -EIO; | |
777 | } | |
a86c6181 AT |
778 | |
779 | err = ext4_ext_dirty(handle, inode, curp); | |
780 | ext4_std_error(inode->i_sb, err); | |
781 | ||
782 | return err; | |
783 | } | |
784 | ||
785 | /* | |
d0d856e8 RD |
786 | * ext4_ext_split: |
787 | * inserts new subtree into the path, using free index entry | |
788 | * at depth @at: | |
789 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
790 | * - makes decision where to split | |
791 | * - moves remaining extents and index entries (right to the split point) | |
792 | * into the newly allocated blocks | |
793 | * - initializes subtree | |
a86c6181 AT |
794 | */ |
795 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
796 | unsigned int flags, |
797 | struct ext4_ext_path *path, | |
798 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
799 | { |
800 | struct buffer_head *bh = NULL; | |
801 | int depth = ext_depth(inode); | |
802 | struct ext4_extent_header *neh; | |
803 | struct ext4_extent_idx *fidx; | |
804 | struct ext4_extent *ex; | |
805 | int i = at, k, m, a; | |
f65e6fba | 806 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 807 | __le32 border; |
f65e6fba | 808 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
809 | int err = 0; |
810 | ||
811 | /* make decision: where to split? */ | |
d0d856e8 | 812 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 813 | |
d0d856e8 | 814 | /* if current leaf will be split, then we should use |
a86c6181 | 815 | * border from split point */ |
273df556 FM |
816 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
817 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
818 | return -EIO; | |
819 | } | |
a86c6181 AT |
820 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
821 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 822 | ext_debug("leaf will be split." |
a86c6181 | 823 | " next leaf starts at %d\n", |
8c55e204 | 824 | le32_to_cpu(border)); |
a86c6181 AT |
825 | } else { |
826 | border = newext->ee_block; | |
827 | ext_debug("leaf will be added." | |
828 | " next leaf starts at %d\n", | |
8c55e204 | 829 | le32_to_cpu(border)); |
a86c6181 AT |
830 | } |
831 | ||
832 | /* | |
d0d856e8 RD |
833 | * If error occurs, then we break processing |
834 | * and mark filesystem read-only. index won't | |
a86c6181 | 835 | * be inserted and tree will be in consistent |
d0d856e8 | 836 | * state. Next mount will repair buffers too. |
a86c6181 AT |
837 | */ |
838 | ||
839 | /* | |
d0d856e8 RD |
840 | * Get array to track all allocated blocks. |
841 | * We need this to handle errors and free blocks | |
842 | * upon them. | |
a86c6181 | 843 | */ |
5d4958f9 | 844 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
845 | if (!ablocks) |
846 | return -ENOMEM; | |
a86c6181 AT |
847 | |
848 | /* allocate all needed blocks */ | |
849 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
850 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 851 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 852 | newext, &err, flags); |
a86c6181 AT |
853 | if (newblock == 0) |
854 | goto cleanup; | |
855 | ablocks[a] = newblock; | |
856 | } | |
857 | ||
858 | /* initialize new leaf */ | |
859 | newblock = ablocks[--a]; | |
273df556 FM |
860 | if (unlikely(newblock == 0)) { |
861 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
862 | err = -EIO; | |
863 | goto cleanup; | |
864 | } | |
a86c6181 AT |
865 | bh = sb_getblk(inode->i_sb, newblock); |
866 | if (!bh) { | |
867 | err = -EIO; | |
868 | goto cleanup; | |
869 | } | |
870 | lock_buffer(bh); | |
871 | ||
7e028976 AM |
872 | err = ext4_journal_get_create_access(handle, bh); |
873 | if (err) | |
a86c6181 AT |
874 | goto cleanup; |
875 | ||
876 | neh = ext_block_hdr(bh); | |
877 | neh->eh_entries = 0; | |
55ad63bf | 878 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
879 | neh->eh_magic = EXT4_EXT_MAGIC; |
880 | neh->eh_depth = 0; | |
881 | ex = EXT_FIRST_EXTENT(neh); | |
882 | ||
d0d856e8 | 883 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
884 | if (unlikely(path[depth].p_hdr->eh_entries != |
885 | path[depth].p_hdr->eh_max)) { | |
886 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
887 | path[depth].p_hdr->eh_entries, | |
888 | path[depth].p_hdr->eh_max); | |
889 | err = -EIO; | |
890 | goto cleanup; | |
891 | } | |
a86c6181 AT |
892 | /* start copy from next extent */ |
893 | /* TODO: we could do it by single memmove */ | |
894 | m = 0; | |
895 | path[depth].p_ext++; | |
896 | while (path[depth].p_ext <= | |
897 | EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
553f9008 | 898 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", |
8c55e204 | 899 | le32_to_cpu(path[depth].p_ext->ee_block), |
bf89d16f | 900 | ext4_ext_pblock(path[depth].p_ext), |
553f9008 | 901 | ext4_ext_is_uninitialized(path[depth].p_ext), |
a2df2a63 | 902 | ext4_ext_get_actual_len(path[depth].p_ext), |
a86c6181 AT |
903 | newblock); |
904 | /*memmove(ex++, path[depth].p_ext++, | |
905 | sizeof(struct ext4_extent)); | |
906 | neh->eh_entries++;*/ | |
907 | path[depth].p_ext++; | |
908 | m++; | |
909 | } | |
910 | if (m) { | |
911 | memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); | |
e8546d06 | 912 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
913 | } |
914 | ||
915 | set_buffer_uptodate(bh); | |
916 | unlock_buffer(bh); | |
917 | ||
0390131b | 918 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 919 | if (err) |
a86c6181 AT |
920 | goto cleanup; |
921 | brelse(bh); | |
922 | bh = NULL; | |
923 | ||
924 | /* correct old leaf */ | |
925 | if (m) { | |
7e028976 AM |
926 | err = ext4_ext_get_access(handle, inode, path + depth); |
927 | if (err) | |
a86c6181 | 928 | goto cleanup; |
e8546d06 | 929 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
930 | err = ext4_ext_dirty(handle, inode, path + depth); |
931 | if (err) | |
a86c6181 AT |
932 | goto cleanup; |
933 | ||
934 | } | |
935 | ||
936 | /* create intermediate indexes */ | |
937 | k = depth - at - 1; | |
273df556 FM |
938 | if (unlikely(k < 0)) { |
939 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
940 | err = -EIO; | |
941 | goto cleanup; | |
942 | } | |
a86c6181 AT |
943 | if (k) |
944 | ext_debug("create %d intermediate indices\n", k); | |
945 | /* insert new index into current index block */ | |
946 | /* current depth stored in i var */ | |
947 | i = depth - 1; | |
948 | while (k--) { | |
949 | oldblock = newblock; | |
950 | newblock = ablocks[--a]; | |
bba90743 | 951 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
952 | if (!bh) { |
953 | err = -EIO; | |
954 | goto cleanup; | |
955 | } | |
956 | lock_buffer(bh); | |
957 | ||
7e028976 AM |
958 | err = ext4_journal_get_create_access(handle, bh); |
959 | if (err) | |
a86c6181 AT |
960 | goto cleanup; |
961 | ||
962 | neh = ext_block_hdr(bh); | |
963 | neh->eh_entries = cpu_to_le16(1); | |
964 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 965 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
966 | neh->eh_depth = cpu_to_le16(depth - i); |
967 | fidx = EXT_FIRST_INDEX(neh); | |
968 | fidx->ei_block = border; | |
f65e6fba | 969 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 970 | |
bba90743 ES |
971 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
972 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 AT |
973 | /* copy indexes */ |
974 | m = 0; | |
975 | path[i].p_idx++; | |
976 | ||
977 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
978 | EXT_MAX_INDEX(path[i].p_hdr)); | |
273df556 FM |
979 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
980 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
981 | EXT4_ERROR_INODE(inode, | |
982 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
983 | le32_to_cpu(path[i].p_ext->ee_block)); | |
984 | err = -EIO; | |
985 | goto cleanup; | |
986 | } | |
a86c6181 | 987 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { |
26d535ed | 988 | ext_debug("%d: move %d:%llu in new index %llu\n", i, |
8c55e204 | 989 | le32_to_cpu(path[i].p_idx->ei_block), |
bf89d16f | 990 | ext4_idx_pblock(path[i].p_idx), |
8c55e204 | 991 | newblock); |
a86c6181 AT |
992 | /*memmove(++fidx, path[i].p_idx++, |
993 | sizeof(struct ext4_extent_idx)); | |
994 | neh->eh_entries++; | |
995 | BUG_ON(neh->eh_entries > neh->eh_max);*/ | |
996 | path[i].p_idx++; | |
997 | m++; | |
998 | } | |
999 | if (m) { | |
1000 | memmove(++fidx, path[i].p_idx - m, | |
1001 | sizeof(struct ext4_extent_idx) * m); | |
e8546d06 | 1002 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1003 | } |
1004 | set_buffer_uptodate(bh); | |
1005 | unlock_buffer(bh); | |
1006 | ||
0390131b | 1007 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1008 | if (err) |
a86c6181 AT |
1009 | goto cleanup; |
1010 | brelse(bh); | |
1011 | bh = NULL; | |
1012 | ||
1013 | /* correct old index */ | |
1014 | if (m) { | |
1015 | err = ext4_ext_get_access(handle, inode, path + i); | |
1016 | if (err) | |
1017 | goto cleanup; | |
e8546d06 | 1018 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1019 | err = ext4_ext_dirty(handle, inode, path + i); |
1020 | if (err) | |
1021 | goto cleanup; | |
1022 | } | |
1023 | ||
1024 | i--; | |
1025 | } | |
1026 | ||
1027 | /* insert new index */ | |
a86c6181 AT |
1028 | err = ext4_ext_insert_index(handle, inode, path + at, |
1029 | le32_to_cpu(border), newblock); | |
1030 | ||
1031 | cleanup: | |
1032 | if (bh) { | |
1033 | if (buffer_locked(bh)) | |
1034 | unlock_buffer(bh); | |
1035 | brelse(bh); | |
1036 | } | |
1037 | ||
1038 | if (err) { | |
1039 | /* free all allocated blocks in error case */ | |
1040 | for (i = 0; i < depth; i++) { | |
1041 | if (!ablocks[i]) | |
1042 | continue; | |
7dc57615 | 1043 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1044 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1045 | } |
1046 | } | |
1047 | kfree(ablocks); | |
1048 | ||
1049 | return err; | |
1050 | } | |
1051 | ||
1052 | /* | |
d0d856e8 RD |
1053 | * ext4_ext_grow_indepth: |
1054 | * implements tree growing procedure: | |
1055 | * - allocates new block | |
1056 | * - moves top-level data (index block or leaf) into the new block | |
1057 | * - initializes new top-level, creating index that points to the | |
1058 | * just created block | |
a86c6181 AT |
1059 | */ |
1060 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db AH |
1061 | unsigned int flags, |
1062 | struct ext4_ext_path *path, | |
1063 | struct ext4_extent *newext) | |
a86c6181 AT |
1064 | { |
1065 | struct ext4_ext_path *curp = path; | |
1066 | struct ext4_extent_header *neh; | |
a86c6181 | 1067 | struct buffer_head *bh; |
f65e6fba | 1068 | ext4_fsblk_t newblock; |
a86c6181 AT |
1069 | int err = 0; |
1070 | ||
55f020db AH |
1071 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
1072 | newext, &err, flags); | |
a86c6181 AT |
1073 | if (newblock == 0) |
1074 | return err; | |
1075 | ||
1076 | bh = sb_getblk(inode->i_sb, newblock); | |
1077 | if (!bh) { | |
1078 | err = -EIO; | |
1079 | ext4_std_error(inode->i_sb, err); | |
1080 | return err; | |
1081 | } | |
1082 | lock_buffer(bh); | |
1083 | ||
7e028976 AM |
1084 | err = ext4_journal_get_create_access(handle, bh); |
1085 | if (err) { | |
a86c6181 AT |
1086 | unlock_buffer(bh); |
1087 | goto out; | |
1088 | } | |
1089 | ||
1090 | /* move top-level index/leaf into new block */ | |
1091 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
1092 | ||
1093 | /* set size of new block */ | |
1094 | neh = ext_block_hdr(bh); | |
1095 | /* old root could have indexes or leaves | |
1096 | * so calculate e_max right way */ | |
1097 | if (ext_depth(inode)) | |
55ad63bf | 1098 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1099 | else |
55ad63bf | 1100 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1101 | neh->eh_magic = EXT4_EXT_MAGIC; |
1102 | set_buffer_uptodate(bh); | |
1103 | unlock_buffer(bh); | |
1104 | ||
0390131b | 1105 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1106 | if (err) |
a86c6181 AT |
1107 | goto out; |
1108 | ||
1109 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
1110 | err = ext4_ext_get_access(handle, inode, curp); |
1111 | if (err) | |
a86c6181 AT |
1112 | goto out; |
1113 | ||
1114 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1115 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); |
a86c6181 AT |
1116 | curp->p_hdr->eh_entries = cpu_to_le16(1); |
1117 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
1118 | |
1119 | if (path[0].p_hdr->eh_depth) | |
1120 | curp->p_idx->ei_block = | |
1121 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
1122 | else | |
1123 | curp->p_idx->ei_block = | |
1124 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 1125 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
1126 | |
1127 | neh = ext_inode_hdr(inode); | |
2ae02107 | 1128 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1129 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1130 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1131 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 AT |
1132 | |
1133 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
1134 | err = ext4_ext_dirty(handle, inode, curp); | |
1135 | out: | |
1136 | brelse(bh); | |
1137 | ||
1138 | return err; | |
1139 | } | |
1140 | ||
1141 | /* | |
d0d856e8 RD |
1142 | * ext4_ext_create_new_leaf: |
1143 | * finds empty index and adds new leaf. | |
1144 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1145 | */ |
1146 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1147 | unsigned int flags, |
1148 | struct ext4_ext_path *path, | |
1149 | struct ext4_extent *newext) | |
a86c6181 AT |
1150 | { |
1151 | struct ext4_ext_path *curp; | |
1152 | int depth, i, err = 0; | |
1153 | ||
1154 | repeat: | |
1155 | i = depth = ext_depth(inode); | |
1156 | ||
1157 | /* walk up to the tree and look for free index entry */ | |
1158 | curp = path + depth; | |
1159 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1160 | i--; | |
1161 | curp--; | |
1162 | } | |
1163 | ||
d0d856e8 RD |
1164 | /* we use already allocated block for index block, |
1165 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1166 | if (EXT_HAS_FREE_INDEX(curp)) { |
1167 | /* if we found index with free entry, then use that | |
1168 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1169 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1170 | if (err) |
1171 | goto out; | |
a86c6181 AT |
1172 | |
1173 | /* refill path */ | |
1174 | ext4_ext_drop_refs(path); | |
1175 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1176 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1177 | path); | |
a86c6181 AT |
1178 | if (IS_ERR(path)) |
1179 | err = PTR_ERR(path); | |
1180 | } else { | |
1181 | /* tree is full, time to grow in depth */ | |
55f020db AH |
1182 | err = ext4_ext_grow_indepth(handle, inode, flags, |
1183 | path, newext); | |
a86c6181 AT |
1184 | if (err) |
1185 | goto out; | |
1186 | ||
1187 | /* refill path */ | |
1188 | ext4_ext_drop_refs(path); | |
1189 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1190 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1191 | path); | |
a86c6181 AT |
1192 | if (IS_ERR(path)) { |
1193 | err = PTR_ERR(path); | |
1194 | goto out; | |
1195 | } | |
1196 | ||
1197 | /* | |
d0d856e8 RD |
1198 | * only first (depth 0 -> 1) produces free space; |
1199 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1200 | */ |
1201 | depth = ext_depth(inode); | |
1202 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1203 | /* now we need to split */ |
a86c6181 AT |
1204 | goto repeat; |
1205 | } | |
1206 | } | |
1207 | ||
1208 | out: | |
1209 | return err; | |
1210 | } | |
1211 | ||
1988b51e AT |
1212 | /* |
1213 | * search the closest allocated block to the left for *logical | |
1214 | * and returns it at @logical + it's physical address at @phys | |
1215 | * if *logical is the smallest allocated block, the function | |
1216 | * returns 0 at @phys | |
1217 | * return value contains 0 (success) or error code | |
1218 | */ | |
1f109d5a TT |
1219 | static int ext4_ext_search_left(struct inode *inode, |
1220 | struct ext4_ext_path *path, | |
1221 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1222 | { |
1223 | struct ext4_extent_idx *ix; | |
1224 | struct ext4_extent *ex; | |
b939e376 | 1225 | int depth, ee_len; |
1988b51e | 1226 | |
273df556 FM |
1227 | if (unlikely(path == NULL)) { |
1228 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1229 | return -EIO; | |
1230 | } | |
1988b51e AT |
1231 | depth = path->p_depth; |
1232 | *phys = 0; | |
1233 | ||
1234 | if (depth == 0 && path->p_ext == NULL) | |
1235 | return 0; | |
1236 | ||
1237 | /* usually extent in the path covers blocks smaller | |
1238 | * then *logical, but it can be that extent is the | |
1239 | * first one in the file */ | |
1240 | ||
1241 | ex = path[depth].p_ext; | |
b939e376 | 1242 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1243 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1244 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1245 | EXT4_ERROR_INODE(inode, | |
1246 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1247 | *logical, le32_to_cpu(ex->ee_block)); | |
1248 | return -EIO; | |
1249 | } | |
1988b51e AT |
1250 | while (--depth >= 0) { |
1251 | ix = path[depth].p_idx; | |
273df556 FM |
1252 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1253 | EXT4_ERROR_INODE(inode, | |
1254 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
1255 | ix != NULL ? ix->ei_block : 0, | |
1256 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? | |
1257 | EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0, | |
1258 | depth); | |
1259 | return -EIO; | |
1260 | } | |
1988b51e AT |
1261 | } |
1262 | return 0; | |
1263 | } | |
1264 | ||
273df556 FM |
1265 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1266 | EXT4_ERROR_INODE(inode, | |
1267 | "logical %d < ee_block %d + ee_len %d!", | |
1268 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1269 | return -EIO; | |
1270 | } | |
1988b51e | 1271 | |
b939e376 | 1272 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1273 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1274 | return 0; |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * search the closest allocated block to the right for *logical | |
1279 | * and returns it at @logical + it's physical address at @phys | |
1280 | * if *logical is the smallest allocated block, the function | |
1281 | * returns 0 at @phys | |
1282 | * return value contains 0 (success) or error code | |
1283 | */ | |
1f109d5a TT |
1284 | static int ext4_ext_search_right(struct inode *inode, |
1285 | struct ext4_ext_path *path, | |
1286 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1287 | { |
1288 | struct buffer_head *bh = NULL; | |
1289 | struct ext4_extent_header *eh; | |
1290 | struct ext4_extent_idx *ix; | |
1291 | struct ext4_extent *ex; | |
1292 | ext4_fsblk_t block; | |
395a87bf ES |
1293 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1294 | int ee_len; | |
1988b51e | 1295 | |
273df556 FM |
1296 | if (unlikely(path == NULL)) { |
1297 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1298 | return -EIO; | |
1299 | } | |
1988b51e AT |
1300 | depth = path->p_depth; |
1301 | *phys = 0; | |
1302 | ||
1303 | if (depth == 0 && path->p_ext == NULL) | |
1304 | return 0; | |
1305 | ||
1306 | /* usually extent in the path covers blocks smaller | |
1307 | * then *logical, but it can be that extent is the | |
1308 | * first one in the file */ | |
1309 | ||
1310 | ex = path[depth].p_ext; | |
b939e376 | 1311 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1312 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1313 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1314 | EXT4_ERROR_INODE(inode, | |
1315 | "first_extent(path[%d].p_hdr) != ex", | |
1316 | depth); | |
1317 | return -EIO; | |
1318 | } | |
1988b51e AT |
1319 | while (--depth >= 0) { |
1320 | ix = path[depth].p_idx; | |
273df556 FM |
1321 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1322 | EXT4_ERROR_INODE(inode, | |
1323 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1324 | *logical); | |
1325 | return -EIO; | |
1326 | } | |
1988b51e AT |
1327 | } |
1328 | *logical = le32_to_cpu(ex->ee_block); | |
bf89d16f | 1329 | *phys = ext4_ext_pblock(ex); |
1988b51e AT |
1330 | return 0; |
1331 | } | |
1332 | ||
273df556 FM |
1333 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1334 | EXT4_ERROR_INODE(inode, | |
1335 | "logical %d < ee_block %d + ee_len %d!", | |
1336 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1337 | return -EIO; | |
1338 | } | |
1988b51e AT |
1339 | |
1340 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1341 | /* next allocated block in this leaf */ | |
1342 | ex++; | |
1343 | *logical = le32_to_cpu(ex->ee_block); | |
bf89d16f | 1344 | *phys = ext4_ext_pblock(ex); |
1988b51e AT |
1345 | return 0; |
1346 | } | |
1347 | ||
1348 | /* go up and search for index to the right */ | |
1349 | while (--depth >= 0) { | |
1350 | ix = path[depth].p_idx; | |
1351 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1352 | goto got_index; |
1988b51e AT |
1353 | } |
1354 | ||
25f1ee3a WF |
1355 | /* we've gone up to the root and found no index to the right */ |
1356 | return 0; | |
1988b51e | 1357 | |
25f1ee3a | 1358 | got_index: |
1988b51e AT |
1359 | /* we've found index to the right, let's |
1360 | * follow it and find the closest allocated | |
1361 | * block to the right */ | |
1362 | ix++; | |
bf89d16f | 1363 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1364 | while (++depth < path->p_depth) { |
1365 | bh = sb_bread(inode->i_sb, block); | |
1366 | if (bh == NULL) | |
1367 | return -EIO; | |
1368 | eh = ext_block_hdr(bh); | |
395a87bf | 1369 | /* subtract from p_depth to get proper eh_depth */ |
56b19868 | 1370 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1371 | put_bh(bh); |
1372 | return -EIO; | |
1373 | } | |
1374 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1375 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1376 | put_bh(bh); |
1377 | } | |
1378 | ||
1379 | bh = sb_bread(inode->i_sb, block); | |
1380 | if (bh == NULL) | |
1381 | return -EIO; | |
1382 | eh = ext_block_hdr(bh); | |
56b19868 | 1383 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1384 | put_bh(bh); |
1385 | return -EIO; | |
1386 | } | |
1387 | ex = EXT_FIRST_EXTENT(eh); | |
1388 | *logical = le32_to_cpu(ex->ee_block); | |
bf89d16f | 1389 | *phys = ext4_ext_pblock(ex); |
1988b51e AT |
1390 | put_bh(bh); |
1391 | return 0; | |
1988b51e AT |
1392 | } |
1393 | ||
a86c6181 | 1394 | /* |
d0d856e8 RD |
1395 | * ext4_ext_next_allocated_block: |
1396 | * returns allocated block in subsequent extent or EXT_MAX_BLOCK. | |
1397 | * NOTE: it considers block number from index entry as | |
1398 | * allocated block. Thus, index entries have to be consistent | |
1399 | * with leaves. | |
a86c6181 | 1400 | */ |
725d26d3 | 1401 | static ext4_lblk_t |
a86c6181 AT |
1402 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1403 | { | |
1404 | int depth; | |
1405 | ||
1406 | BUG_ON(path == NULL); | |
1407 | depth = path->p_depth; | |
1408 | ||
1409 | if (depth == 0 && path->p_ext == NULL) | |
1410 | return EXT_MAX_BLOCK; | |
1411 | ||
1412 | while (depth >= 0) { | |
1413 | if (depth == path->p_depth) { | |
1414 | /* leaf */ | |
1415 | if (path[depth].p_ext != | |
1416 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1417 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1418 | } else { | |
1419 | /* index */ | |
1420 | if (path[depth].p_idx != | |
1421 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1422 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1423 | } | |
1424 | depth--; | |
1425 | } | |
1426 | ||
1427 | return EXT_MAX_BLOCK; | |
1428 | } | |
1429 | ||
1430 | /* | |
d0d856e8 | 1431 | * ext4_ext_next_leaf_block: |
a86c6181 AT |
1432 | * returns first allocated block from next leaf or EXT_MAX_BLOCK |
1433 | */ | |
725d26d3 | 1434 | static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode, |
63f57933 | 1435 | struct ext4_ext_path *path) |
a86c6181 AT |
1436 | { |
1437 | int depth; | |
1438 | ||
1439 | BUG_ON(path == NULL); | |
1440 | depth = path->p_depth; | |
1441 | ||
1442 | /* zero-tree has no leaf blocks at all */ | |
1443 | if (depth == 0) | |
1444 | return EXT_MAX_BLOCK; | |
1445 | ||
1446 | /* go to index block */ | |
1447 | depth--; | |
1448 | ||
1449 | while (depth >= 0) { | |
1450 | if (path[depth].p_idx != | |
1451 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1452 | return (ext4_lblk_t) |
1453 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1454 | depth--; |
1455 | } | |
1456 | ||
1457 | return EXT_MAX_BLOCK; | |
1458 | } | |
1459 | ||
1460 | /* | |
d0d856e8 RD |
1461 | * ext4_ext_correct_indexes: |
1462 | * if leaf gets modified and modified extent is first in the leaf, | |
1463 | * then we have to correct all indexes above. | |
a86c6181 AT |
1464 | * TODO: do we need to correct tree in all cases? |
1465 | */ | |
1d03ec98 | 1466 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1467 | struct ext4_ext_path *path) |
1468 | { | |
1469 | struct ext4_extent_header *eh; | |
1470 | int depth = ext_depth(inode); | |
1471 | struct ext4_extent *ex; | |
1472 | __le32 border; | |
1473 | int k, err = 0; | |
1474 | ||
1475 | eh = path[depth].p_hdr; | |
1476 | ex = path[depth].p_ext; | |
273df556 FM |
1477 | |
1478 | if (unlikely(ex == NULL || eh == NULL)) { | |
1479 | EXT4_ERROR_INODE(inode, | |
1480 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1481 | return -EIO; | |
1482 | } | |
a86c6181 AT |
1483 | |
1484 | if (depth == 0) { | |
1485 | /* there is no tree at all */ | |
1486 | return 0; | |
1487 | } | |
1488 | ||
1489 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1490 | /* we correct tree if first leaf got modified only */ | |
1491 | return 0; | |
1492 | } | |
1493 | ||
1494 | /* | |
d0d856e8 | 1495 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1496 | */ |
1497 | k = depth - 1; | |
1498 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1499 | err = ext4_ext_get_access(handle, inode, path + k); |
1500 | if (err) | |
a86c6181 AT |
1501 | return err; |
1502 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1503 | err = ext4_ext_dirty(handle, inode, path + k); |
1504 | if (err) | |
a86c6181 AT |
1505 | return err; |
1506 | ||
1507 | while (k--) { | |
1508 | /* change all left-side indexes */ | |
1509 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1510 | break; | |
7e028976 AM |
1511 | err = ext4_ext_get_access(handle, inode, path + k); |
1512 | if (err) | |
a86c6181 AT |
1513 | break; |
1514 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1515 | err = ext4_ext_dirty(handle, inode, path + k); |
1516 | if (err) | |
a86c6181 AT |
1517 | break; |
1518 | } | |
1519 | ||
1520 | return err; | |
1521 | } | |
1522 | ||
748de673 | 1523 | int |
a86c6181 AT |
1524 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1525 | struct ext4_extent *ex2) | |
1526 | { | |
749269fa | 1527 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1528 | |
1529 | /* | |
1530 | * Make sure that either both extents are uninitialized, or | |
1531 | * both are _not_. | |
1532 | */ | |
1533 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1534 | return 0; | |
1535 | ||
749269fa AA |
1536 | if (ext4_ext_is_uninitialized(ex1)) |
1537 | max_len = EXT_UNINIT_MAX_LEN; | |
1538 | else | |
1539 | max_len = EXT_INIT_MAX_LEN; | |
1540 | ||
a2df2a63 AA |
1541 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1542 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1543 | ||
1544 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1545 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1546 | return 0; |
1547 | ||
471d4011 SB |
1548 | /* |
1549 | * To allow future support for preallocated extents to be added | |
1550 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1551 | * this can result in the top bit of ee_len being set. |
471d4011 | 1552 | */ |
749269fa | 1553 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1554 | return 0; |
bbf2f9fb | 1555 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1556 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1557 | return 0; |
1558 | #endif | |
1559 | ||
bf89d16f | 1560 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1561 | return 1; |
1562 | return 0; | |
1563 | } | |
1564 | ||
56055d3a AA |
1565 | /* |
1566 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1567 | * It always tries to merge towards right. If you want to merge towards | |
1568 | * left, pass "ex - 1" as argument instead of "ex". | |
1569 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1570 | * 1 if they got merged. | |
1571 | */ | |
197217a5 | 1572 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1573 | struct ext4_ext_path *path, |
1574 | struct ext4_extent *ex) | |
56055d3a AA |
1575 | { |
1576 | struct ext4_extent_header *eh; | |
1577 | unsigned int depth, len; | |
1578 | int merge_done = 0; | |
1579 | int uninitialized = 0; | |
1580 | ||
1581 | depth = ext_depth(inode); | |
1582 | BUG_ON(path[depth].p_hdr == NULL); | |
1583 | eh = path[depth].p_hdr; | |
1584 | ||
1585 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1586 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1587 | break; | |
1588 | /* merge with next extent! */ | |
1589 | if (ext4_ext_is_uninitialized(ex)) | |
1590 | uninitialized = 1; | |
1591 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1592 | + ext4_ext_get_actual_len(ex + 1)); | |
1593 | if (uninitialized) | |
1594 | ext4_ext_mark_uninitialized(ex); | |
1595 | ||
1596 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1597 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1598 | * sizeof(struct ext4_extent); | |
1599 | memmove(ex + 1, ex + 2, len); | |
1600 | } | |
e8546d06 | 1601 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1602 | merge_done = 1; |
1603 | WARN_ON(eh->eh_entries == 0); | |
1604 | if (!eh->eh_entries) | |
24676da4 | 1605 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1606 | } |
1607 | ||
1608 | return merge_done; | |
1609 | } | |
1610 | ||
197217a5 YY |
1611 | /* |
1612 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1613 | * return 1 if merge left else 0. | |
1614 | */ | |
1615 | static int ext4_ext_try_to_merge(struct inode *inode, | |
1616 | struct ext4_ext_path *path, | |
1617 | struct ext4_extent *ex) { | |
1618 | struct ext4_extent_header *eh; | |
1619 | unsigned int depth; | |
1620 | int merge_done = 0; | |
1621 | int ret = 0; | |
1622 | ||
1623 | depth = ext_depth(inode); | |
1624 | BUG_ON(path[depth].p_hdr == NULL); | |
1625 | eh = path[depth].p_hdr; | |
1626 | ||
1627 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1628 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1629 | ||
1630 | if (!merge_done) | |
1631 | ret = ext4_ext_try_to_merge_right(inode, path, ex); | |
1632 | ||
1633 | return ret; | |
1634 | } | |
1635 | ||
25d14f98 AA |
1636 | /* |
1637 | * check if a portion of the "newext" extent overlaps with an | |
1638 | * existing extent. | |
1639 | * | |
1640 | * If there is an overlap discovered, it updates the length of the newext | |
1641 | * such that there will be no overlap, and then returns 1. | |
1642 | * If there is no overlap found, it returns 0. | |
1643 | */ | |
1f109d5a TT |
1644 | static unsigned int ext4_ext_check_overlap(struct inode *inode, |
1645 | struct ext4_extent *newext, | |
1646 | struct ext4_ext_path *path) | |
25d14f98 | 1647 | { |
725d26d3 | 1648 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1649 | unsigned int depth, len1; |
1650 | unsigned int ret = 0; | |
1651 | ||
1652 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1653 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1654 | depth = ext_depth(inode); |
1655 | if (!path[depth].p_ext) | |
1656 | goto out; | |
1657 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
1658 | ||
1659 | /* | |
1660 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1661 | * is before the requested block(s) |
25d14f98 AA |
1662 | */ |
1663 | if (b2 < b1) { | |
1664 | b2 = ext4_ext_next_allocated_block(path); | |
1665 | if (b2 == EXT_MAX_BLOCK) | |
1666 | goto out; | |
1667 | } | |
1668 | ||
725d26d3 | 1669 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 AA |
1670 | if (b1 + len1 < b1) { |
1671 | len1 = EXT_MAX_BLOCK - b1; | |
1672 | newext->ee_len = cpu_to_le16(len1); | |
1673 | ret = 1; | |
1674 | } | |
1675 | ||
1676 | /* check for overlap */ | |
1677 | if (b1 + len1 > b2) { | |
1678 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1679 | ret = 1; | |
1680 | } | |
1681 | out: | |
1682 | return ret; | |
1683 | } | |
1684 | ||
a86c6181 | 1685 | /* |
d0d856e8 RD |
1686 | * ext4_ext_insert_extent: |
1687 | * tries to merge requsted extent into the existing extent or | |
1688 | * inserts requested extent as new one into the tree, | |
1689 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1690 | */ |
1691 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1692 | struct ext4_ext_path *path, | |
0031462b | 1693 | struct ext4_extent *newext, int flag) |
a86c6181 | 1694 | { |
af5bc92d | 1695 | struct ext4_extent_header *eh; |
a86c6181 AT |
1696 | struct ext4_extent *ex, *fex; |
1697 | struct ext4_extent *nearex; /* nearest extent */ | |
1698 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1699 | int depth, len, err; |
1700 | ext4_lblk_t next; | |
a2df2a63 | 1701 | unsigned uninitialized = 0; |
55f020db | 1702 | int flags = 0; |
a86c6181 | 1703 | |
273df556 FM |
1704 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1705 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1706 | return -EIO; | |
1707 | } | |
a86c6181 AT |
1708 | depth = ext_depth(inode); |
1709 | ex = path[depth].p_ext; | |
273df556 FM |
1710 | if (unlikely(path[depth].p_hdr == NULL)) { |
1711 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1712 | return -EIO; | |
1713 | } | |
a86c6181 AT |
1714 | |
1715 | /* try to insert block into found extent and return */ | |
744692dc | 1716 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1717 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
553f9008 | 1718 | ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", |
bf89d16f TT |
1719 | ext4_ext_is_uninitialized(newext), |
1720 | ext4_ext_get_actual_len(newext), | |
1721 | le32_to_cpu(ex->ee_block), | |
1722 | ext4_ext_is_uninitialized(ex), | |
1723 | ext4_ext_get_actual_len(ex), | |
1724 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1725 | err = ext4_ext_get_access(handle, inode, path + depth); |
1726 | if (err) | |
a86c6181 | 1727 | return err; |
a2df2a63 AA |
1728 | |
1729 | /* | |
1730 | * ext4_can_extents_be_merged should have checked that either | |
1731 | * both extents are uninitialized, or both aren't. Thus we | |
1732 | * need to check only one of them here. | |
1733 | */ | |
1734 | if (ext4_ext_is_uninitialized(ex)) | |
1735 | uninitialized = 1; | |
1736 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1737 | + ext4_ext_get_actual_len(newext)); | |
1738 | if (uninitialized) | |
1739 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1740 | eh = path[depth].p_hdr; |
1741 | nearex = ex; | |
1742 | goto merge; | |
1743 | } | |
1744 | ||
1745 | repeat: | |
1746 | depth = ext_depth(inode); | |
1747 | eh = path[depth].p_hdr; | |
1748 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1749 | goto has_space; | |
1750 | ||
1751 | /* probably next leaf has space for us? */ | |
1752 | fex = EXT_LAST_EXTENT(eh); | |
1753 | next = ext4_ext_next_leaf_block(inode, path); | |
1754 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) | |
1755 | && next != EXT_MAX_BLOCK) { | |
1756 | ext_debug("next leaf block - %d\n", next); | |
1757 | BUG_ON(npath != NULL); | |
1758 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1759 | if (IS_ERR(npath)) | |
1760 | return PTR_ERR(npath); | |
1761 | BUG_ON(npath->p_depth != path->p_depth); | |
1762 | eh = npath[depth].p_hdr; | |
1763 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1764 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1765 | le16_to_cpu(eh->eh_entries)); |
1766 | path = npath; | |
1767 | goto repeat; | |
1768 | } | |
1769 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1770 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1771 | } | |
1772 | ||
1773 | /* | |
d0d856e8 RD |
1774 | * There is no free space in the found leaf. |
1775 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1776 | */ |
55f020db AH |
1777 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1778 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1779 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1780 | if (err) |
1781 | goto cleanup; | |
1782 | depth = ext_depth(inode); | |
1783 | eh = path[depth].p_hdr; | |
1784 | ||
1785 | has_space: | |
1786 | nearex = path[depth].p_ext; | |
1787 | ||
7e028976 AM |
1788 | err = ext4_ext_get_access(handle, inode, path + depth); |
1789 | if (err) | |
a86c6181 AT |
1790 | goto cleanup; |
1791 | ||
1792 | if (!nearex) { | |
1793 | /* there is no extent in this leaf, create first one */ | |
553f9008 | 1794 | ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n", |
8c55e204 | 1795 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1796 | ext4_ext_pblock(newext), |
553f9008 | 1797 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1798 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1799 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1800 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1801 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1802 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1803 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1804 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1805 | len = (len - 1) * sizeof(struct ext4_extent); | |
1806 | len = len < 0 ? 0 : len; | |
553f9008 | 1807 | ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " |
a86c6181 | 1808 | "move %d from 0x%p to 0x%p\n", |
8c55e204 | 1809 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1810 | ext4_ext_pblock(newext), |
553f9008 | 1811 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1812 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1813 | nearex, len, nearex + 1, nearex + 2); |
1814 | memmove(nearex + 2, nearex + 1, len); | |
1815 | } | |
1816 | path[depth].p_ext = nearex + 1; | |
1817 | } else { | |
1818 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1819 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1820 | len = len < 0 ? 0 : len; | |
553f9008 | 1821 | ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " |
a86c6181 AT |
1822 | "move %d from 0x%p to 0x%p\n", |
1823 | le32_to_cpu(newext->ee_block), | |
bf89d16f | 1824 | ext4_ext_pblock(newext), |
553f9008 | 1825 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1826 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1827 | nearex, len, nearex + 1, nearex + 2); |
1828 | memmove(nearex + 1, nearex, len); | |
1829 | path[depth].p_ext = nearex; | |
1830 | } | |
1831 | ||
e8546d06 | 1832 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1833 | nearex = path[depth].p_ext; |
1834 | nearex->ee_block = newext->ee_block; | |
bf89d16f | 1835 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1836 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1837 | |
1838 | merge: | |
1839 | /* try to merge extents to the right */ | |
744692dc | 1840 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
0031462b | 1841 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1842 | |
1843 | /* try to merge extents to the left */ | |
1844 | ||
1845 | /* time to correct all indexes above */ | |
1846 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1847 | if (err) | |
1848 | goto cleanup; | |
1849 | ||
1850 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1851 | ||
1852 | cleanup: | |
1853 | if (npath) { | |
1854 | ext4_ext_drop_refs(npath); | |
1855 | kfree(npath); | |
1856 | } | |
a86c6181 AT |
1857 | ext4_ext_invalidate_cache(inode); |
1858 | return err; | |
1859 | } | |
1860 | ||
1f109d5a TT |
1861 | static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1862 | ext4_lblk_t num, ext_prepare_callback func, | |
1863 | void *cbdata) | |
6873fa0d ES |
1864 | { |
1865 | struct ext4_ext_path *path = NULL; | |
1866 | struct ext4_ext_cache cbex; | |
1867 | struct ext4_extent *ex; | |
1868 | ext4_lblk_t next, start = 0, end = 0; | |
1869 | ext4_lblk_t last = block + num; | |
1870 | int depth, exists, err = 0; | |
1871 | ||
1872 | BUG_ON(func == NULL); | |
1873 | BUG_ON(inode == NULL); | |
1874 | ||
1875 | while (block < last && block != EXT_MAX_BLOCK) { | |
1876 | num = last - block; | |
1877 | /* find extent for this block */ | |
fab3a549 | 1878 | down_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d | 1879 | path = ext4_ext_find_extent(inode, block, path); |
fab3a549 | 1880 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1881 | if (IS_ERR(path)) { |
1882 | err = PTR_ERR(path); | |
1883 | path = NULL; | |
1884 | break; | |
1885 | } | |
1886 | ||
1887 | depth = ext_depth(inode); | |
273df556 FM |
1888 | if (unlikely(path[depth].p_hdr == NULL)) { |
1889 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1890 | err = -EIO; | |
1891 | break; | |
1892 | } | |
6873fa0d ES |
1893 | ex = path[depth].p_ext; |
1894 | next = ext4_ext_next_allocated_block(path); | |
1895 | ||
1896 | exists = 0; | |
1897 | if (!ex) { | |
1898 | /* there is no extent yet, so try to allocate | |
1899 | * all requested space */ | |
1900 | start = block; | |
1901 | end = block + num; | |
1902 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1903 | /* need to allocate space before found extent */ | |
1904 | start = block; | |
1905 | end = le32_to_cpu(ex->ee_block); | |
1906 | if (block + num < end) | |
1907 | end = block + num; | |
1908 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1909 | + ext4_ext_get_actual_len(ex)) { | |
1910 | /* need to allocate space after found extent */ | |
1911 | start = block; | |
1912 | end = block + num; | |
1913 | if (end >= next) | |
1914 | end = next; | |
1915 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1916 | /* | |
1917 | * some part of requested space is covered | |
1918 | * by found extent | |
1919 | */ | |
1920 | start = block; | |
1921 | end = le32_to_cpu(ex->ee_block) | |
1922 | + ext4_ext_get_actual_len(ex); | |
1923 | if (block + num < end) | |
1924 | end = block + num; | |
1925 | exists = 1; | |
1926 | } else { | |
1927 | BUG(); | |
1928 | } | |
1929 | BUG_ON(end <= start); | |
1930 | ||
1931 | if (!exists) { | |
1932 | cbex.ec_block = start; | |
1933 | cbex.ec_len = end - start; | |
1934 | cbex.ec_start = 0; | |
6873fa0d ES |
1935 | } else { |
1936 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1937 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
bf89d16f | 1938 | cbex.ec_start = ext4_ext_pblock(ex); |
6873fa0d ES |
1939 | } |
1940 | ||
273df556 FM |
1941 | if (unlikely(cbex.ec_len == 0)) { |
1942 | EXT4_ERROR_INODE(inode, "cbex.ec_len == 0"); | |
1943 | err = -EIO; | |
1944 | break; | |
1945 | } | |
6873fa0d ES |
1946 | err = func(inode, path, &cbex, ex, cbdata); |
1947 | ext4_ext_drop_refs(path); | |
1948 | ||
1949 | if (err < 0) | |
1950 | break; | |
1951 | ||
1952 | if (err == EXT_REPEAT) | |
1953 | continue; | |
1954 | else if (err == EXT_BREAK) { | |
1955 | err = 0; | |
1956 | break; | |
1957 | } | |
1958 | ||
1959 | if (ext_depth(inode) != depth) { | |
1960 | /* depth was changed. we have to realloc path */ | |
1961 | kfree(path); | |
1962 | path = NULL; | |
1963 | } | |
1964 | ||
1965 | block = cbex.ec_block + cbex.ec_len; | |
1966 | } | |
1967 | ||
1968 | if (path) { | |
1969 | ext4_ext_drop_refs(path); | |
1970 | kfree(path); | |
1971 | } | |
1972 | ||
1973 | return err; | |
1974 | } | |
1975 | ||
09b88252 | 1976 | static void |
725d26d3 | 1977 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 1978 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
1979 | { |
1980 | struct ext4_ext_cache *cex; | |
1981 | BUG_ON(len == 0); | |
2ec0ae3a | 1982 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 | 1983 | cex = &EXT4_I(inode)->i_cached_extent; |
a86c6181 AT |
1984 | cex->ec_block = block; |
1985 | cex->ec_len = len; | |
1986 | cex->ec_start = start; | |
2ec0ae3a | 1987 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
1988 | } |
1989 | ||
1990 | /* | |
d0d856e8 RD |
1991 | * ext4_ext_put_gap_in_cache: |
1992 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
1993 | * and cache this gap |
1994 | */ | |
09b88252 | 1995 | static void |
a86c6181 | 1996 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 1997 | ext4_lblk_t block) |
a86c6181 AT |
1998 | { |
1999 | int depth = ext_depth(inode); | |
725d26d3 AK |
2000 | unsigned long len; |
2001 | ext4_lblk_t lblock; | |
a86c6181 AT |
2002 | struct ext4_extent *ex; |
2003 | ||
2004 | ex = path[depth].p_ext; | |
2005 | if (ex == NULL) { | |
2006 | /* there is no extent yet, so gap is [0;-] */ | |
2007 | lblock = 0; | |
2008 | len = EXT_MAX_BLOCK; | |
2009 | ext_debug("cache gap(whole file):"); | |
2010 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2011 | lblock = block; | |
2012 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2013 | ext_debug("cache gap(before): %u [%u:%u]", |
2014 | block, | |
2015 | le32_to_cpu(ex->ee_block), | |
2016 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 2017 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2018 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2019 | ext4_lblk_t next; |
8c55e204 | 2020 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2021 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2022 | |
2023 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2024 | ext_debug("cache gap(after): [%u:%u] %u", |
2025 | le32_to_cpu(ex->ee_block), | |
2026 | ext4_ext_get_actual_len(ex), | |
2027 | block); | |
725d26d3 AK |
2028 | BUG_ON(next == lblock); |
2029 | len = next - lblock; | |
a86c6181 AT |
2030 | } else { |
2031 | lblock = len = 0; | |
2032 | BUG(); | |
2033 | } | |
2034 | ||
bba90743 | 2035 | ext_debug(" -> %u:%lu\n", lblock, len); |
b05e6ae5 | 2036 | ext4_ext_put_in_cache(inode, lblock, len, 0); |
a86c6181 AT |
2037 | } |
2038 | ||
b05e6ae5 TT |
2039 | /* |
2040 | * Return 0 if cache is invalid; 1 if the cache is valid | |
2041 | */ | |
09b88252 | 2042 | static int |
725d26d3 | 2043 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, |
a86c6181 AT |
2044 | struct ext4_extent *ex) |
2045 | { | |
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)) { |
8c55e204 | 2062 | ex->ee_block = cpu_to_le32(cex->ec_block); |
f65e6fba | 2063 | ext4_ext_store_pblock(ex, cex->ec_start); |
8c55e204 | 2064 | ex->ee_len = cpu_to_le16(cex->ec_len); |
bba90743 ES |
2065 | ext_debug("%u cached by %u:%u:%llu\n", |
2066 | block, | |
2067 | cex->ec_block, cex->ec_len, cex->ec_start); | |
b05e6ae5 | 2068 | ret = 1; |
a86c6181 | 2069 | } |
2ec0ae3a | 2070 | errout: |
77f4135f VH |
2071 | if (!ret) |
2072 | sbi->extent_cache_misses++; | |
2073 | else | |
2074 | sbi->extent_cache_hits++; | |
2ec0ae3a TT |
2075 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2076 | return ret; | |
a86c6181 AT |
2077 | } |
2078 | ||
2079 | /* | |
d0d856e8 RD |
2080 | * ext4_ext_rm_idx: |
2081 | * removes index from the index block. | |
2082 | * It's used in truncate case only, thus all requests are for | |
2083 | * last index in the block only. | |
a86c6181 | 2084 | */ |
1d03ec98 | 2085 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
2086 | struct ext4_ext_path *path) |
2087 | { | |
a86c6181 | 2088 | int err; |
f65e6fba | 2089 | ext4_fsblk_t leaf; |
a86c6181 AT |
2090 | |
2091 | /* free index block */ | |
2092 | path--; | |
bf89d16f | 2093 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2094 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2095 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2096 | return -EIO; | |
2097 | } | |
7e028976 AM |
2098 | err = ext4_ext_get_access(handle, inode, path); |
2099 | if (err) | |
a86c6181 | 2100 | return err; |
e8546d06 | 2101 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2102 | err = ext4_ext_dirty(handle, inode, path); |
2103 | if (err) | |
a86c6181 | 2104 | return err; |
2ae02107 | 2105 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
7dc57615 | 2106 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2107 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
a86c6181 AT |
2108 | return err; |
2109 | } | |
2110 | ||
2111 | /* | |
ee12b630 MC |
2112 | * ext4_ext_calc_credits_for_single_extent: |
2113 | * This routine returns max. credits that needed to insert an extent | |
2114 | * to the extent tree. | |
2115 | * When pass the actual path, the caller should calculate credits | |
2116 | * under i_data_sem. | |
a86c6181 | 2117 | */ |
525f4ed8 | 2118 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2119 | struct ext4_ext_path *path) |
2120 | { | |
a86c6181 | 2121 | if (path) { |
ee12b630 | 2122 | int depth = ext_depth(inode); |
f3bd1f3f | 2123 | int ret = 0; |
ee12b630 | 2124 | |
a86c6181 | 2125 | /* probably there is space in leaf? */ |
a86c6181 | 2126 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2127 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2128 | |
ee12b630 MC |
2129 | /* |
2130 | * There are some space in the leaf tree, no | |
2131 | * need to account for leaf block credit | |
2132 | * | |
2133 | * bitmaps and block group descriptor blocks | |
2134 | * and other metadat blocks still need to be | |
2135 | * accounted. | |
2136 | */ | |
525f4ed8 | 2137 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2138 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2139 | return ret; |
ee12b630 MC |
2140 | } |
2141 | } | |
a86c6181 | 2142 | |
525f4ed8 | 2143 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2144 | } |
a86c6181 | 2145 | |
ee12b630 MC |
2146 | /* |
2147 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2148 | * | |
2149 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2150 | * in the worse case, each tree level index/leaf need to be changed | |
2151 | * if the tree split due to insert a new extent, then the old tree | |
2152 | * index/leaf need to be updated too | |
2153 | * | |
2154 | * If the nrblocks are discontiguous, they could cause | |
2155 | * the whole tree split more than once, but this is really rare. | |
2156 | */ | |
525f4ed8 | 2157 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2158 | { |
2159 | int index; | |
2160 | int depth = ext_depth(inode); | |
a86c6181 | 2161 | |
ee12b630 MC |
2162 | if (chunk) |
2163 | index = depth * 2; | |
2164 | else | |
2165 | index = depth * 3; | |
a86c6181 | 2166 | |
ee12b630 | 2167 | return index; |
a86c6181 AT |
2168 | } |
2169 | ||
2170 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
2171 | struct ext4_extent *ex, | |
725d26d3 | 2172 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2173 | { |
a2df2a63 | 2174 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
e6362609 | 2175 | int flags = EXT4_FREE_BLOCKS_FORGET; |
a86c6181 | 2176 | |
c9de560d | 2177 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
e6362609 | 2178 | flags |= EXT4_FREE_BLOCKS_METADATA; |
a86c6181 AT |
2179 | #ifdef EXTENTS_STATS |
2180 | { | |
2181 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2182 | spin_lock(&sbi->s_ext_stats_lock); |
2183 | sbi->s_ext_blocks += ee_len; | |
2184 | sbi->s_ext_extents++; | |
2185 | if (ee_len < sbi->s_ext_min) | |
2186 | sbi->s_ext_min = ee_len; | |
2187 | if (ee_len > sbi->s_ext_max) | |
2188 | sbi->s_ext_max = ee_len; | |
2189 | if (ext_depth(inode) > sbi->s_depth_max) | |
2190 | sbi->s_depth_max = ext_depth(inode); | |
2191 | spin_unlock(&sbi->s_ext_stats_lock); | |
2192 | } | |
2193 | #endif | |
2194 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2195 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2196 | /* tail removal */ |
725d26d3 | 2197 | ext4_lblk_t num; |
f65e6fba | 2198 | ext4_fsblk_t start; |
725d26d3 | 2199 | |
a2df2a63 | 2200 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
bf89d16f | 2201 | start = ext4_ext_pblock(ex) + ee_len - num; |
725d26d3 | 2202 | ext_debug("free last %u blocks starting %llu\n", num, start); |
7dc57615 | 2203 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
a86c6181 | 2204 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2205 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
725d26d3 | 2206 | printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n", |
a2df2a63 | 2207 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
a86c6181 | 2208 | } else { |
725d26d3 AK |
2209 | printk(KERN_INFO "strange request: removal(2) " |
2210 | "%u-%u from %u:%u\n", | |
2211 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2212 | } |
2213 | return 0; | |
2214 | } | |
2215 | ||
2216 | static int | |
2217 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
725d26d3 | 2218 | struct ext4_ext_path *path, ext4_lblk_t start) |
a86c6181 AT |
2219 | { |
2220 | int err = 0, correct_index = 0; | |
2221 | int depth = ext_depth(inode), credits; | |
2222 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2223 | ext4_lblk_t a, b, block; |
2224 | unsigned num; | |
2225 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2226 | unsigned short ex_ee_len; |
a2df2a63 | 2227 | unsigned uninitialized = 0; |
a86c6181 AT |
2228 | struct ext4_extent *ex; |
2229 | ||
c29c0ae7 | 2230 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 2231 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
2232 | if (!path[depth].p_hdr) |
2233 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2234 | eh = path[depth].p_hdr; | |
273df556 FM |
2235 | if (unlikely(path[depth].p_hdr == NULL)) { |
2236 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2237 | return -EIO; | |
2238 | } | |
a86c6181 AT |
2239 | /* find where to start removing */ |
2240 | ex = EXT_LAST_EXTENT(eh); | |
2241 | ||
2242 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2243 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2244 | |
2245 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
2246 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2247 | |
2248 | if (ext4_ext_is_uninitialized(ex)) | |
2249 | uninitialized = 1; | |
2250 | else | |
2251 | uninitialized = 0; | |
2252 | ||
553f9008 M |
2253 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2254 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2255 | path[depth].p_ext = ex; |
2256 | ||
2257 | a = ex_ee_block > start ? ex_ee_block : start; | |
2258 | b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? | |
2259 | ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; | |
2260 | ||
2261 | ext_debug(" border %u:%u\n", a, b); | |
2262 | ||
2263 | if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { | |
2264 | block = 0; | |
2265 | num = 0; | |
2266 | BUG(); | |
2267 | } else if (a != ex_ee_block) { | |
2268 | /* remove tail of the extent */ | |
2269 | block = ex_ee_block; | |
2270 | num = a - block; | |
2271 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
2272 | /* remove head of the extent */ | |
2273 | block = a; | |
2274 | num = b - a; | |
2275 | /* there is no "make a hole" API yet */ | |
2276 | BUG(); | |
2277 | } else { | |
2278 | /* remove whole extent: excellent! */ | |
2279 | block = ex_ee_block; | |
2280 | num = 0; | |
2281 | BUG_ON(a != ex_ee_block); | |
2282 | BUG_ON(b != ex_ee_block + ex_ee_len - 1); | |
2283 | } | |
2284 | ||
34071da7 TT |
2285 | /* |
2286 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2287 | * descriptor) for each block group; assume two block | |
2288 | * groups plus ex_ee_len/blocks_per_block_group for | |
2289 | * the worst case | |
2290 | */ | |
2291 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2292 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2293 | correct_index = 1; | |
2294 | credits += (ext_depth(inode)) + 1; | |
2295 | } | |
5aca07eb | 2296 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2297 | |
487caeef | 2298 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2299 | if (err) |
a86c6181 | 2300 | goto out; |
a86c6181 AT |
2301 | |
2302 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2303 | if (err) | |
2304 | goto out; | |
2305 | ||
2306 | err = ext4_remove_blocks(handle, inode, ex, a, b); | |
2307 | if (err) | |
2308 | goto out; | |
2309 | ||
2310 | if (num == 0) { | |
d0d856e8 | 2311 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2312 | ext4_ext_store_pblock(ex, 0); |
e8546d06 | 2313 | le16_add_cpu(&eh->eh_entries, -1); |
a86c6181 AT |
2314 | } |
2315 | ||
2316 | ex->ee_block = cpu_to_le32(block); | |
2317 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
2318 | /* |
2319 | * Do not mark uninitialized if all the blocks in the | |
2320 | * extent have been removed. | |
2321 | */ | |
2322 | if (uninitialized && num) | |
a2df2a63 | 2323 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
2324 | |
2325 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2326 | if (err) | |
2327 | goto out; | |
2328 | ||
2ae02107 | 2329 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
bf89d16f | 2330 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2331 | ex--; |
2332 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2333 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2334 | } |
2335 | ||
2336 | if (correct_index && eh->eh_entries) | |
2337 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2338 | ||
2339 | /* if this leaf is free, then we should | |
2340 | * remove it from index block above */ | |
2341 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2342 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2343 | ||
2344 | out: | |
2345 | return err; | |
2346 | } | |
2347 | ||
2348 | /* | |
d0d856e8 RD |
2349 | * ext4_ext_more_to_rm: |
2350 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2351 | */ |
09b88252 | 2352 | static int |
a86c6181 AT |
2353 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2354 | { | |
2355 | BUG_ON(path->p_idx == NULL); | |
2356 | ||
2357 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2358 | return 0; | |
2359 | ||
2360 | /* | |
d0d856e8 | 2361 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2362 | * so we have to consider current index for truncation |
2363 | */ | |
2364 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2365 | return 0; | |
2366 | return 1; | |
2367 | } | |
2368 | ||
1d03ec98 | 2369 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
2370 | { |
2371 | struct super_block *sb = inode->i_sb; | |
2372 | int depth = ext_depth(inode); | |
2373 | struct ext4_ext_path *path; | |
2374 | handle_t *handle; | |
0617b83f | 2375 | int i, err; |
a86c6181 | 2376 | |
725d26d3 | 2377 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
2378 | |
2379 | /* probably first extent we're gonna free will be last in block */ | |
2380 | handle = ext4_journal_start(inode, depth + 1); | |
2381 | if (IS_ERR(handle)) | |
2382 | return PTR_ERR(handle); | |
2383 | ||
0617b83f | 2384 | again: |
a86c6181 AT |
2385 | ext4_ext_invalidate_cache(inode); |
2386 | ||
2387 | /* | |
d0d856e8 RD |
2388 | * We start scanning from right side, freeing all the blocks |
2389 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2390 | */ |
0617b83f | 2391 | depth = ext_depth(inode); |
216553c4 | 2392 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
2393 | if (path == NULL) { |
2394 | ext4_journal_stop(handle); | |
2395 | return -ENOMEM; | |
2396 | } | |
0617b83f | 2397 | path[0].p_depth = depth; |
a86c6181 | 2398 | path[0].p_hdr = ext_inode_hdr(inode); |
56b19868 | 2399 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
2400 | err = -EIO; |
2401 | goto out; | |
2402 | } | |
0617b83f | 2403 | i = err = 0; |
a86c6181 AT |
2404 | |
2405 | while (i >= 0 && err == 0) { | |
2406 | if (i == depth) { | |
2407 | /* this is leaf block */ | |
2408 | err = ext4_ext_rm_leaf(handle, inode, path, start); | |
d0d856e8 | 2409 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2410 | brelse(path[i].p_bh); |
2411 | path[i].p_bh = NULL; | |
2412 | i--; | |
2413 | continue; | |
2414 | } | |
2415 | ||
2416 | /* this is index block */ | |
2417 | if (!path[i].p_hdr) { | |
2418 | ext_debug("initialize header\n"); | |
2419 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2420 | } |
2421 | ||
a86c6181 | 2422 | if (!path[i].p_idx) { |
d0d856e8 | 2423 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2424 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2425 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2426 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2427 | path[i].p_hdr, | |
2428 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2429 | } else { | |
d0d856e8 | 2430 | /* we were already here, see at next index */ |
a86c6181 AT |
2431 | path[i].p_idx--; |
2432 | } | |
2433 | ||
2434 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2435 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2436 | path[i].p_idx); | |
2437 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2438 | struct buffer_head *bh; |
a86c6181 | 2439 | /* go to the next level */ |
2ae02107 | 2440 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2441 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2442 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2443 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2444 | if (!bh) { |
a86c6181 AT |
2445 | /* should we reset i_size? */ |
2446 | err = -EIO; | |
2447 | break; | |
2448 | } | |
c29c0ae7 AT |
2449 | if (WARN_ON(i + 1 > depth)) { |
2450 | err = -EIO; | |
2451 | break; | |
2452 | } | |
56b19868 | 2453 | if (ext4_ext_check(inode, ext_block_hdr(bh), |
c29c0ae7 AT |
2454 | depth - i - 1)) { |
2455 | err = -EIO; | |
2456 | break; | |
2457 | } | |
2458 | path[i + 1].p_bh = bh; | |
a86c6181 | 2459 | |
d0d856e8 RD |
2460 | /* save actual number of indexes since this |
2461 | * number is changed at the next iteration */ | |
a86c6181 AT |
2462 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2463 | i++; | |
2464 | } else { | |
d0d856e8 | 2465 | /* we finished processing this index, go up */ |
a86c6181 | 2466 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2467 | /* index is empty, remove it; |
a86c6181 AT |
2468 | * handle must be already prepared by the |
2469 | * truncatei_leaf() */ | |
2470 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2471 | } | |
d0d856e8 | 2472 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2473 | brelse(path[i].p_bh); |
2474 | path[i].p_bh = NULL; | |
2475 | i--; | |
2476 | ext_debug("return to level %d\n", i); | |
2477 | } | |
2478 | } | |
2479 | ||
2480 | /* TODO: flexible tree reduction should be here */ | |
2481 | if (path->p_hdr->eh_entries == 0) { | |
2482 | /* | |
d0d856e8 RD |
2483 | * truncate to zero freed all the tree, |
2484 | * so we need to correct eh_depth | |
a86c6181 AT |
2485 | */ |
2486 | err = ext4_ext_get_access(handle, inode, path); | |
2487 | if (err == 0) { | |
2488 | ext_inode_hdr(inode)->eh_depth = 0; | |
2489 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2490 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2491 | err = ext4_ext_dirty(handle, inode, path); |
2492 | } | |
2493 | } | |
2494 | out: | |
a86c6181 AT |
2495 | ext4_ext_drop_refs(path); |
2496 | kfree(path); | |
0617b83f DM |
2497 | if (err == -EAGAIN) |
2498 | goto again; | |
a86c6181 AT |
2499 | ext4_journal_stop(handle); |
2500 | ||
2501 | return err; | |
2502 | } | |
2503 | ||
2504 | /* | |
2505 | * called at mount time | |
2506 | */ | |
2507 | void ext4_ext_init(struct super_block *sb) | |
2508 | { | |
2509 | /* | |
2510 | * possible initialization would be here | |
2511 | */ | |
2512 | ||
83982b6f | 2513 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2514 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
4776004f | 2515 | printk(KERN_INFO "EXT4-fs: file extents enabled"); |
bbf2f9fb RD |
2516 | #ifdef AGGRESSIVE_TEST |
2517 | printk(", aggressive tests"); | |
a86c6181 AT |
2518 | #endif |
2519 | #ifdef CHECK_BINSEARCH | |
2520 | printk(", check binsearch"); | |
2521 | #endif | |
2522 | #ifdef EXTENTS_STATS | |
2523 | printk(", stats"); | |
2524 | #endif | |
2525 | printk("\n"); | |
90576c0b | 2526 | #endif |
a86c6181 AT |
2527 | #ifdef EXTENTS_STATS |
2528 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2529 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2530 | EXT4_SB(sb)->s_ext_max = 0; | |
2531 | #endif | |
2532 | } | |
2533 | } | |
2534 | ||
2535 | /* | |
2536 | * called at umount time | |
2537 | */ | |
2538 | void ext4_ext_release(struct super_block *sb) | |
2539 | { | |
83982b6f | 2540 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2541 | return; |
2542 | ||
2543 | #ifdef EXTENTS_STATS | |
2544 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2545 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2546 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2547 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2548 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2549 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2550 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2551 | } | |
2552 | #endif | |
2553 | } | |
2554 | ||
093a088b AK |
2555 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2556 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2557 | { | |
2407518d LC |
2558 | ext4_fsblk_t ee_pblock; |
2559 | unsigned int ee_len; | |
b720303d | 2560 | int ret; |
093a088b | 2561 | |
093a088b | 2562 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2563 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2564 | |
a107e5a3 | 2565 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2566 | if (ret > 0) |
2567 | ret = 0; | |
093a088b | 2568 | |
2407518d | 2569 | return ret; |
093a088b AK |
2570 | } |
2571 | ||
47ea3bb5 YY |
2572 | /* |
2573 | * used by extent splitting. | |
2574 | */ | |
2575 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
2576 | due to ENOSPC */ | |
2577 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
2578 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
2579 | ||
2580 | /* | |
2581 | * ext4_split_extent_at() splits an extent at given block. | |
2582 | * | |
2583 | * @handle: the journal handle | |
2584 | * @inode: the file inode | |
2585 | * @path: the path to the extent | |
2586 | * @split: the logical block where the extent is splitted. | |
2587 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2588 | * the states(init or uninit) of new extents. | |
2589 | * @flags: flags used to insert new extent to extent tree. | |
2590 | * | |
2591 | * | |
2592 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2593 | * of which are deterimined by split_flag. | |
2594 | * | |
2595 | * There are two cases: | |
2596 | * a> the extent are splitted into two extent. | |
2597 | * b> split is not needed, and just mark the extent. | |
2598 | * | |
2599 | * return 0 on success. | |
2600 | */ | |
2601 | static int ext4_split_extent_at(handle_t *handle, | |
2602 | struct inode *inode, | |
2603 | struct ext4_ext_path *path, | |
2604 | ext4_lblk_t split, | |
2605 | int split_flag, | |
2606 | int flags) | |
2607 | { | |
2608 | ext4_fsblk_t newblock; | |
2609 | ext4_lblk_t ee_block; | |
2610 | struct ext4_extent *ex, newex, orig_ex; | |
2611 | struct ext4_extent *ex2 = NULL; | |
2612 | unsigned int ee_len, depth; | |
2613 | int err = 0; | |
2614 | ||
2615 | ext_debug("ext4_split_extents_at: inode %lu, logical" | |
2616 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2617 | ||
2618 | ext4_ext_show_leaf(inode, path); | |
2619 | ||
2620 | depth = ext_depth(inode); | |
2621 | ex = path[depth].p_ext; | |
2622 | ee_block = le32_to_cpu(ex->ee_block); | |
2623 | ee_len = ext4_ext_get_actual_len(ex); | |
2624 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2625 | ||
2626 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
2627 | ||
2628 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2629 | if (err) | |
2630 | goto out; | |
2631 | ||
2632 | if (split == ee_block) { | |
2633 | /* | |
2634 | * case b: block @split is the block that the extent begins with | |
2635 | * then we just change the state of the extent, and splitting | |
2636 | * is not needed. | |
2637 | */ | |
2638 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2639 | ext4_ext_mark_uninitialized(ex); | |
2640 | else | |
2641 | ext4_ext_mark_initialized(ex); | |
2642 | ||
2643 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
2644 | ext4_ext_try_to_merge(inode, path, ex); | |
2645 | ||
2646 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2647 | goto out; | |
2648 | } | |
2649 | ||
2650 | /* case a */ | |
2651 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
2652 | ex->ee_len = cpu_to_le16(split - ee_block); | |
2653 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
2654 | ext4_ext_mark_uninitialized(ex); | |
2655 | ||
2656 | /* | |
2657 | * path may lead to new leaf, not to original leaf any more | |
2658 | * after ext4_ext_insert_extent() returns, | |
2659 | */ | |
2660 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2661 | if (err) | |
2662 | goto fix_extent_len; | |
2663 | ||
2664 | ex2 = &newex; | |
2665 | ex2->ee_block = cpu_to_le32(split); | |
2666 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
2667 | ext4_ext_store_pblock(ex2, newblock); | |
2668 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2669 | ext4_ext_mark_uninitialized(ex2); | |
2670 | ||
2671 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
2672 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2673 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2674 | if (err) | |
2675 | goto fix_extent_len; | |
2676 | /* update the extent length and mark as initialized */ | |
2677 | ex->ee_len = cpu_to_le32(ee_len); | |
2678 | ext4_ext_try_to_merge(inode, path, ex); | |
2679 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2680 | goto out; | |
2681 | } else if (err) | |
2682 | goto fix_extent_len; | |
2683 | ||
2684 | out: | |
2685 | ext4_ext_show_leaf(inode, path); | |
2686 | return err; | |
2687 | ||
2688 | fix_extent_len: | |
2689 | ex->ee_len = orig_ex.ee_len; | |
2690 | ext4_ext_dirty(handle, inode, path + depth); | |
2691 | return err; | |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * ext4_split_extents() splits an extent and mark extent which is covered | |
2696 | * by @map as split_flags indicates | |
2697 | * | |
2698 | * It may result in splitting the extent into multiple extents (upto three) | |
2699 | * There are three possibilities: | |
2700 | * a> There is no split required | |
2701 | * b> Splits in two extents: Split is happening at either end of the extent | |
2702 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
2703 | * | |
2704 | */ | |
2705 | static int ext4_split_extent(handle_t *handle, | |
2706 | struct inode *inode, | |
2707 | struct ext4_ext_path *path, | |
2708 | struct ext4_map_blocks *map, | |
2709 | int split_flag, | |
2710 | int flags) | |
2711 | { | |
2712 | ext4_lblk_t ee_block; | |
2713 | struct ext4_extent *ex; | |
2714 | unsigned int ee_len, depth; | |
2715 | int err = 0; | |
2716 | int uninitialized; | |
2717 | int split_flag1, flags1; | |
2718 | ||
2719 | depth = ext_depth(inode); | |
2720 | ex = path[depth].p_ext; | |
2721 | ee_block = le32_to_cpu(ex->ee_block); | |
2722 | ee_len = ext4_ext_get_actual_len(ex); | |
2723 | uninitialized = ext4_ext_is_uninitialized(ex); | |
2724 | ||
2725 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
2726 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
2727 | EXT4_EXT_MAY_ZEROOUT : 0; | |
2728 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; | |
2729 | if (uninitialized) | |
2730 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
2731 | EXT4_EXT_MARK_UNINIT2; | |
2732 | err = ext4_split_extent_at(handle, inode, path, | |
2733 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
2734 | if (err) |
2735 | goto out; | |
47ea3bb5 YY |
2736 | } |
2737 | ||
2738 | ext4_ext_drop_refs(path); | |
2739 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
2740 | if (IS_ERR(path)) | |
2741 | return PTR_ERR(path); | |
2742 | ||
2743 | if (map->m_lblk >= ee_block) { | |
2744 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
2745 | EXT4_EXT_MAY_ZEROOUT : 0; | |
2746 | if (uninitialized) | |
2747 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
2748 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2749 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
2750 | err = ext4_split_extent_at(handle, inode, path, | |
2751 | map->m_lblk, split_flag1, flags); | |
2752 | if (err) | |
2753 | goto out; | |
2754 | } | |
2755 | ||
2756 | ext4_ext_show_leaf(inode, path); | |
2757 | out: | |
2758 | return err ? err : map->m_len; | |
2759 | } | |
2760 | ||
3977c965 | 2761 | #define EXT4_EXT_ZERO_LEN 7 |
56055d3a | 2762 | /* |
e35fd660 | 2763 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 2764 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 2765 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
2766 | * uninitialized). |
2767 | * There are three possibilities: | |
2768 | * a> There is no split required: Entire extent should be initialized | |
2769 | * b> Splits in two extents: Write is happening at either end of the extent | |
2770 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2771 | */ | |
725d26d3 | 2772 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
2773 | struct inode *inode, |
2774 | struct ext4_map_blocks *map, | |
2775 | struct ext4_ext_path *path) | |
56055d3a | 2776 | { |
667eff35 YY |
2777 | struct ext4_map_blocks split_map; |
2778 | struct ext4_extent zero_ex; | |
2779 | struct ext4_extent *ex; | |
21ca087a | 2780 | ext4_lblk_t ee_block, eof_block; |
725d26d3 | 2781 | unsigned int allocated, ee_len, depth; |
56055d3a | 2782 | int err = 0; |
667eff35 | 2783 | int split_flag = 0; |
21ca087a DM |
2784 | |
2785 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
2786 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 2787 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
2788 | |
2789 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
2790 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
2791 | if (eof_block < map->m_lblk + map->m_len) |
2792 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
2793 | |
2794 | depth = ext_depth(inode); | |
56055d3a AA |
2795 | ex = path[depth].p_ext; |
2796 | ee_block = le32_to_cpu(ex->ee_block); | |
2797 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 2798 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 2799 | |
667eff35 | 2800 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
2801 | /* |
2802 | * It is safe to convert extent to initialized via explicit | |
2803 | * zeroout only if extent is fully insde i_size or new_size. | |
2804 | */ | |
667eff35 | 2805 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 2806 | |
3977c965 | 2807 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
667eff35 YY |
2808 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN && |
2809 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2810 | err = ext4_ext_zeroout(inode, ex); | |
3977c965 | 2811 | if (err) |
d03856bd | 2812 | goto out; |
d03856bd AK |
2813 | |
2814 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2815 | if (err) | |
2816 | goto out; | |
667eff35 YY |
2817 | ext4_ext_mark_initialized(ex); |
2818 | ext4_ext_try_to_merge(inode, path, ex); | |
2819 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2820 | goto out; | |
56055d3a | 2821 | } |
667eff35 | 2822 | |
56055d3a | 2823 | /* |
667eff35 YY |
2824 | * four cases: |
2825 | * 1. split the extent into three extents. | |
2826 | * 2. split the extent into two extents, zeroout the first half. | |
2827 | * 3. split the extent into two extents, zeroout the second half. | |
2828 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 2829 | */ |
667eff35 YY |
2830 | split_map.m_lblk = map->m_lblk; |
2831 | split_map.m_len = map->m_len; | |
2832 | ||
2833 | if (allocated > map->m_len) { | |
2834 | if (allocated <= EXT4_EXT_ZERO_LEN && | |
2835 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2836 | /* case 3 */ | |
2837 | zero_ex.ee_block = | |
9b940f8e AH |
2838 | cpu_to_le32(map->m_lblk); |
2839 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
2840 | ext4_ext_store_pblock(&zero_ex, |
2841 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
2842 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
2843 | if (err) |
2844 | goto out; | |
667eff35 YY |
2845 | split_map.m_lblk = map->m_lblk; |
2846 | split_map.m_len = allocated; | |
2847 | } else if ((map->m_lblk - ee_block + map->m_len < | |
2848 | EXT4_EXT_ZERO_LEN) && | |
2849 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2850 | /* case 2 */ | |
2851 | if (map->m_lblk != ee_block) { | |
2852 | zero_ex.ee_block = ex->ee_block; | |
2853 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
2854 | ee_block); | |
2855 | ext4_ext_store_pblock(&zero_ex, | |
2856 | ext4_ext_pblock(ex)); | |
2857 | err = ext4_ext_zeroout(inode, &zero_ex); | |
2858 | if (err) | |
2859 | goto out; | |
2860 | } | |
2861 | ||
667eff35 | 2862 | split_map.m_lblk = ee_block; |
9b940f8e AH |
2863 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
2864 | allocated = map->m_len; | |
56055d3a AA |
2865 | } |
2866 | } | |
667eff35 YY |
2867 | |
2868 | allocated = ext4_split_extent(handle, inode, path, | |
2869 | &split_map, split_flag, 0); | |
2870 | if (allocated < 0) | |
2871 | err = allocated; | |
2872 | ||
56055d3a AA |
2873 | out: |
2874 | return err ? err : allocated; | |
2875 | } | |
2876 | ||
0031462b | 2877 | /* |
e35fd660 | 2878 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
2879 | * ext4_get_blocks_dio_write() when DIO to write |
2880 | * to an uninitialized extent. | |
2881 | * | |
fd018fe8 | 2882 | * Writing to an uninitialized extent may result in splitting the uninitialized |
b595076a | 2883 | * extent into multiple /initialized uninitialized extents (up to three) |
0031462b MC |
2884 | * There are three possibilities: |
2885 | * a> There is no split required: Entire extent should be uninitialized | |
2886 | * b> Splits in two extents: Write is happening at either end of the extent | |
2887 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2888 | * | |
2889 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 2890 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 2891 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 2892 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
2893 | * into three uninitialized extent(at most). After IO complete, the part |
2894 | * being filled will be convert to initialized by the end_io callback function | |
2895 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
2896 | * |
2897 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
2898 | */ |
2899 | static int ext4_split_unwritten_extents(handle_t *handle, | |
2900 | struct inode *inode, | |
e35fd660 | 2901 | struct ext4_map_blocks *map, |
0031462b | 2902 | struct ext4_ext_path *path, |
0031462b MC |
2903 | int flags) |
2904 | { | |
667eff35 YY |
2905 | ext4_lblk_t eof_block; |
2906 | ext4_lblk_t ee_block; | |
2907 | struct ext4_extent *ex; | |
2908 | unsigned int ee_len; | |
2909 | int split_flag = 0, depth; | |
21ca087a DM |
2910 | |
2911 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
2912 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 2913 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
2914 | |
2915 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
2916 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
2917 | if (eof_block < map->m_lblk + map->m_len) |
2918 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
2919 | /* |
2920 | * It is safe to convert extent to initialized via explicit | |
2921 | * zeroout only if extent is fully insde i_size or new_size. | |
2922 | */ | |
667eff35 YY |
2923 | depth = ext_depth(inode); |
2924 | ex = path[depth].p_ext; | |
2925 | ee_block = le32_to_cpu(ex->ee_block); | |
2926 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 2927 | |
667eff35 YY |
2928 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
2929 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
0031462b | 2930 | |
667eff35 YY |
2931 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
2932 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 2933 | } |
197217a5 | 2934 | |
c7064ef1 | 2935 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
0031462b MC |
2936 | struct inode *inode, |
2937 | struct ext4_ext_path *path) | |
2938 | { | |
2939 | struct ext4_extent *ex; | |
2940 | struct ext4_extent_header *eh; | |
2941 | int depth; | |
2942 | int err = 0; | |
0031462b MC |
2943 | |
2944 | depth = ext_depth(inode); | |
2945 | eh = path[depth].p_hdr; | |
2946 | ex = path[depth].p_ext; | |
2947 | ||
197217a5 YY |
2948 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
2949 | "block %llu, max_blocks %u\n", inode->i_ino, | |
2950 | (unsigned long long)le32_to_cpu(ex->ee_block), | |
2951 | ext4_ext_get_actual_len(ex)); | |
2952 | ||
0031462b MC |
2953 | err = ext4_ext_get_access(handle, inode, path + depth); |
2954 | if (err) | |
2955 | goto out; | |
2956 | /* first mark the extent as initialized */ | |
2957 | ext4_ext_mark_initialized(ex); | |
2958 | ||
197217a5 YY |
2959 | /* note: ext4_ext_correct_indexes() isn't needed here because |
2960 | * borders are not changed | |
0031462b | 2961 | */ |
197217a5 YY |
2962 | ext4_ext_try_to_merge(inode, path, ex); |
2963 | ||
0031462b MC |
2964 | /* Mark modified extent as dirty */ |
2965 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2966 | out: | |
2967 | ext4_ext_show_leaf(inode, path); | |
2968 | return err; | |
2969 | } | |
2970 | ||
515f41c3 AK |
2971 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
2972 | sector_t block, int count) | |
2973 | { | |
2974 | int i; | |
2975 | for (i = 0; i < count; i++) | |
2976 | unmap_underlying_metadata(bdev, block + i); | |
2977 | } | |
2978 | ||
58590b06 TT |
2979 | /* |
2980 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
2981 | */ | |
2982 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 2983 | ext4_lblk_t lblk, |
58590b06 TT |
2984 | struct ext4_ext_path *path, |
2985 | unsigned int len) | |
2986 | { | |
2987 | int i, depth; | |
2988 | struct ext4_extent_header *eh; | |
65922cb5 | 2989 | struct ext4_extent *last_ex; |
58590b06 TT |
2990 | |
2991 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
2992 | return 0; | |
2993 | ||
2994 | depth = ext_depth(inode); | |
2995 | eh = path[depth].p_hdr; | |
58590b06 TT |
2996 | |
2997 | if (unlikely(!eh->eh_entries)) { | |
2998 | EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and " | |
2999 | "EOFBLOCKS_FL set"); | |
3000 | return -EIO; | |
3001 | } | |
3002 | last_ex = EXT_LAST_EXTENT(eh); | |
3003 | /* | |
3004 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3005 | * last block in the last extent in the file. We test this by | |
3006 | * first checking to see if the caller to | |
3007 | * ext4_ext_get_blocks() was interested in the last block (or | |
3008 | * a block beyond the last block) in the current extent. If | |
3009 | * this turns out to be false, we can bail out from this | |
3010 | * function immediately. | |
3011 | */ | |
d002ebf1 | 3012 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3013 | ext4_ext_get_actual_len(last_ex)) |
3014 | return 0; | |
3015 | /* | |
3016 | * If the caller does appear to be planning to write at or | |
3017 | * beyond the end of the current extent, we then test to see | |
3018 | * if the current extent is the last extent in the file, by | |
3019 | * checking to make sure it was reached via the rightmost node | |
3020 | * at each level of the tree. | |
3021 | */ | |
3022 | for (i = depth-1; i >= 0; i--) | |
3023 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3024 | return 0; | |
3025 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); | |
3026 | return ext4_mark_inode_dirty(handle, inode); | |
3027 | } | |
3028 | ||
0031462b MC |
3029 | static int |
3030 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3031 | struct ext4_map_blocks *map, |
0031462b | 3032 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3033 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3034 | { |
3035 | int ret = 0; | |
3036 | int err = 0; | |
8d5d02e6 | 3037 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
0031462b MC |
3038 | |
3039 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical" | |
3040 | "block %llu, max_blocks %u, flags %d, allocated %u", | |
e35fd660 | 3041 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3042 | flags, allocated); |
3043 | ext4_ext_show_leaf(inode, path); | |
3044 | ||
c7064ef1 | 3045 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3046 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3047 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3048 | path, flags); | |
5f524950 M |
3049 | /* |
3050 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3051 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3052 | * completed |
3053 | */ | |
e9e3bcec | 3054 | if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) { |
bd2d0210 | 3055 | io->flag = EXT4_IO_END_UNWRITTEN; |
e9e3bcec ES |
3056 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); |
3057 | } else | |
19f5fb7a | 3058 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
744692dc | 3059 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3060 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3061 | goto out; |
3062 | } | |
c7064ef1 | 3063 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3064 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
c7064ef1 | 3065 | ret = ext4_convert_unwritten_extents_endio(handle, inode, |
0031462b | 3066 | path); |
58590b06 | 3067 | if (ret >= 0) { |
b436b9be | 3068 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3069 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3070 | path, map->m_len); | |
58590b06 TT |
3071 | } else |
3072 | err = ret; | |
0031462b MC |
3073 | goto out2; |
3074 | } | |
3075 | /* buffered IO case */ | |
3076 | /* | |
3077 | * repeat fallocate creation request | |
3078 | * we already have an unwritten extent | |
3079 | */ | |
3080 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | |
3081 | goto map_out; | |
3082 | ||
3083 | /* buffered READ or buffered write_begin() lookup */ | |
3084 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3085 | /* | |
3086 | * We have blocks reserved already. We | |
3087 | * return allocated blocks so that delalloc | |
3088 | * won't do block reservation for us. But | |
3089 | * the buffer head will be unmapped so that | |
3090 | * a read from the block returns 0s. | |
3091 | */ | |
e35fd660 | 3092 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3093 | goto out1; |
3094 | } | |
3095 | ||
3096 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3097 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
58590b06 | 3098 | if (ret >= 0) { |
b436b9be | 3099 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3100 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, |
3101 | map->m_len); | |
58590b06 TT |
3102 | if (err < 0) |
3103 | goto out2; | |
3104 | } | |
3105 | ||
0031462b MC |
3106 | out: |
3107 | if (ret <= 0) { | |
3108 | err = ret; | |
3109 | goto out2; | |
3110 | } else | |
3111 | allocated = ret; | |
e35fd660 | 3112 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3113 | /* |
3114 | * if we allocated more blocks than requested | |
3115 | * we need to make sure we unmap the extra block | |
3116 | * allocated. The actual needed block will get | |
3117 | * unmapped later when we find the buffer_head marked | |
3118 | * new. | |
3119 | */ | |
e35fd660 | 3120 | if (allocated > map->m_len) { |
515f41c3 | 3121 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3122 | newblock + map->m_len, |
3123 | allocated - map->m_len); | |
3124 | allocated = map->m_len; | |
515f41c3 | 3125 | } |
5f634d06 AK |
3126 | |
3127 | /* | |
3128 | * If we have done fallocate with the offset that is already | |
3129 | * delayed allocated, we would have block reservation | |
3130 | * and quota reservation done in the delayed write path. | |
3131 | * But fallocate would have already updated quota and block | |
3132 | * count for this offset. So cancel these reservation | |
3133 | */ | |
1296cc85 | 3134 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
5f634d06 AK |
3135 | ext4_da_update_reserve_space(inode, allocated, 0); |
3136 | ||
0031462b | 3137 | map_out: |
e35fd660 | 3138 | map->m_flags |= EXT4_MAP_MAPPED; |
0031462b | 3139 | out1: |
e35fd660 TT |
3140 | if (allocated > map->m_len) |
3141 | allocated = map->m_len; | |
0031462b | 3142 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3143 | map->m_pblk = newblock; |
3144 | map->m_len = allocated; | |
0031462b MC |
3145 | out2: |
3146 | if (path) { | |
3147 | ext4_ext_drop_refs(path); | |
3148 | kfree(path); | |
3149 | } | |
3150 | return err ? err : allocated; | |
3151 | } | |
58590b06 | 3152 | |
c278bfec | 3153 | /* |
f5ab0d1f MC |
3154 | * Block allocation/map/preallocation routine for extents based files |
3155 | * | |
3156 | * | |
c278bfec | 3157 | * Need to be called with |
0e855ac8 AK |
3158 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3159 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3160 | * |
3161 | * return > 0, number of of blocks already mapped/allocated | |
3162 | * if create == 0 and these are pre-allocated blocks | |
3163 | * buffer head is unmapped | |
3164 | * otherwise blocks are mapped | |
3165 | * | |
3166 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3167 | * buffer head is unmapped | |
3168 | * | |
3169 | * return < 0, error case. | |
c278bfec | 3170 | */ |
e35fd660 TT |
3171 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3172 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3173 | { |
3174 | struct ext4_ext_path *path = NULL; | |
58590b06 | 3175 | struct ext4_extent newex, *ex; |
0562e0ba | 3176 | ext4_fsblk_t newblock = 0; |
b05e6ae5 | 3177 | int err = 0, depth, ret; |
498e5f24 | 3178 | unsigned int allocated = 0; |
c9de560d | 3179 | struct ext4_allocation_request ar; |
8d5d02e6 | 3180 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
a86c6181 | 3181 | |
84fe3bef | 3182 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3183 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3184 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3185 | |
3186 | /* check in cache */ | |
b05e6ae5 TT |
3187 | if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) { |
3188 | if (!newex.ee_start_lo && !newex.ee_start_hi) { | |
c2177057 | 3189 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3190 | /* |
3191 | * block isn't allocated yet and | |
3192 | * user doesn't want to allocate it | |
3193 | */ | |
a86c6181 AT |
3194 | goto out2; |
3195 | } | |
3196 | /* we should allocate requested block */ | |
b05e6ae5 | 3197 | } else { |
a86c6181 | 3198 | /* block is already allocated */ |
e35fd660 | 3199 | newblock = map->m_lblk |
8c55e204 | 3200 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3201 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3202 | /* number of remaining blocks in the extent */ |
b939e376 | 3203 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3204 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3205 | goto out; |
a86c6181 AT |
3206 | } |
3207 | } | |
3208 | ||
3209 | /* find extent for this block */ | |
e35fd660 | 3210 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3211 | if (IS_ERR(path)) { |
3212 | err = PTR_ERR(path); | |
3213 | path = NULL; | |
3214 | goto out2; | |
3215 | } | |
3216 | ||
3217 | depth = ext_depth(inode); | |
3218 | ||
3219 | /* | |
d0d856e8 RD |
3220 | * consistent leaf must not be empty; |
3221 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3222 | * this is why assert can't be put in ext4_ext_find_extent() |
3223 | */ | |
273df556 FM |
3224 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3225 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3226 | "lblock: %lu, depth: %d pblock %lld", |
3227 | (unsigned long) map->m_lblk, depth, | |
3228 | path[depth].p_block); | |
034fb4c9 SP |
3229 | err = -EIO; |
3230 | goto out2; | |
3231 | } | |
a86c6181 | 3232 | |
7e028976 AM |
3233 | ex = path[depth].p_ext; |
3234 | if (ex) { | |
725d26d3 | 3235 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3236 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3237 | unsigned short ee_len; |
471d4011 SB |
3238 | |
3239 | /* | |
471d4011 | 3240 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3241 | * we split out initialized portions during a write. |
471d4011 | 3242 | */ |
a2df2a63 | 3243 | ee_len = ext4_ext_get_actual_len(ex); |
d0d856e8 | 3244 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
3245 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
3246 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 3247 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3248 | allocated = ee_len - (map->m_lblk - ee_block); |
3249 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3250 | ee_block, ee_len, newblock); | |
56055d3a | 3251 | |
a2df2a63 | 3252 | /* Do not put uninitialized extent in the cache */ |
56055d3a | 3253 | if (!ext4_ext_is_uninitialized(ex)) { |
a2df2a63 | 3254 | ext4_ext_put_in_cache(inode, ee_block, |
b05e6ae5 | 3255 | ee_len, ee_start); |
56055d3a AA |
3256 | goto out; |
3257 | } | |
0031462b | 3258 | ret = ext4_ext_handle_uninitialized_extents(handle, |
e35fd660 TT |
3259 | inode, map, path, flags, allocated, |
3260 | newblock); | |
0031462b | 3261 | return ret; |
a86c6181 AT |
3262 | } |
3263 | } | |
3264 | ||
3265 | /* | |
d0d856e8 | 3266 | * requested block isn't allocated yet; |
a86c6181 AT |
3267 | * we couldn't try to create block if create flag is zero |
3268 | */ | |
c2177057 | 3269 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3270 | /* |
3271 | * put just found gap into cache to speed up | |
3272 | * subsequent requests | |
3273 | */ | |
e35fd660 | 3274 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
3275 | goto out2; |
3276 | } | |
3277 | /* | |
c2ea3fde | 3278 | * Okay, we need to do block allocation. |
63f57933 | 3279 | */ |
a86c6181 | 3280 | |
c9de560d | 3281 | /* find neighbour allocated blocks */ |
e35fd660 | 3282 | ar.lleft = map->m_lblk; |
c9de560d AT |
3283 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
3284 | if (err) | |
3285 | goto out2; | |
e35fd660 | 3286 | ar.lright = map->m_lblk; |
c9de560d AT |
3287 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright); |
3288 | if (err) | |
3289 | goto out2; | |
25d14f98 | 3290 | |
749269fa AA |
3291 | /* |
3292 | * See if request is beyond maximum number of blocks we can have in | |
3293 | * a single extent. For an initialized extent this limit is | |
3294 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
3295 | * EXT_UNINIT_MAX_LEN. | |
3296 | */ | |
e35fd660 | 3297 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 3298 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
3299 | map->m_len = EXT_INIT_MAX_LEN; |
3300 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 3301 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 3302 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 3303 | |
e35fd660 TT |
3304 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
3305 | newex.ee_block = cpu_to_le32(map->m_lblk); | |
3306 | newex.ee_len = cpu_to_le16(map->m_len); | |
25d14f98 AA |
3307 | err = ext4_ext_check_overlap(inode, &newex, path); |
3308 | if (err) | |
b939e376 | 3309 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 3310 | else |
e35fd660 | 3311 | allocated = map->m_len; |
c9de560d AT |
3312 | |
3313 | /* allocate new block */ | |
3314 | ar.inode = inode; | |
e35fd660 TT |
3315 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
3316 | ar.logical = map->m_lblk; | |
c9de560d AT |
3317 | ar.len = allocated; |
3318 | if (S_ISREG(inode->i_mode)) | |
3319 | ar.flags = EXT4_MB_HINT_DATA; | |
3320 | else | |
3321 | /* disable in-core preallocation for non-regular files */ | |
3322 | ar.flags = 0; | |
3323 | newblock = ext4_mb_new_blocks(handle, &ar, &err); | |
a86c6181 AT |
3324 | if (!newblock) |
3325 | goto out2; | |
84fe3bef | 3326 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 3327 | ar.goal, newblock, allocated); |
a86c6181 AT |
3328 | |
3329 | /* try to insert new extent into found leaf and return */ | |
f65e6fba | 3330 | ext4_ext_store_pblock(&newex, newblock); |
c9de560d | 3331 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
3332 | /* Mark uninitialized */ |
3333 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 3334 | ext4_ext_mark_uninitialized(&newex); |
8d5d02e6 | 3335 | /* |
744692dc | 3336 | * io_end structure was created for every IO write to an |
25985edc | 3337 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 3338 | * here we flag the IO that really needs the conversion. |
5f524950 | 3339 | * For non asycn direct IO case, flag the inode state |
25985edc | 3340 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 3341 | */ |
744692dc | 3342 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e9e3bcec | 3343 | if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) { |
bd2d0210 | 3344 | io->flag = EXT4_IO_END_UNWRITTEN; |
e9e3bcec ES |
3345 | atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten); |
3346 | } else | |
19f5fb7a TT |
3347 | ext4_set_inode_state(inode, |
3348 | EXT4_STATE_DIO_UNWRITTEN); | |
5f524950 | 3349 | } |
744692dc | 3350 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3351 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 3352 | } |
c8d46e41 | 3353 | |
d002ebf1 | 3354 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len); |
58590b06 TT |
3355 | if (err) |
3356 | goto out2; | |
3357 | ||
0031462b | 3358 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); |
315054f0 AT |
3359 | if (err) { |
3360 | /* free data blocks we just allocated */ | |
c9de560d AT |
3361 | /* not a good idea to call discard here directly, |
3362 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 3363 | ext4_discard_preallocations(inode); |
7dc57615 | 3364 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
e6362609 | 3365 | ext4_ext_get_actual_len(&newex), 0); |
a86c6181 | 3366 | goto out2; |
315054f0 | 3367 | } |
a86c6181 | 3368 | |
a86c6181 | 3369 | /* previous routine could use block we allocated */ |
bf89d16f | 3370 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 3371 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
3372 | if (allocated > map->m_len) |
3373 | allocated = map->m_len; | |
3374 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 3375 | |
5f634d06 AK |
3376 | /* |
3377 | * Update reserved blocks/metadata blocks after successful | |
3378 | * block allocation which had been deferred till now. | |
3379 | */ | |
1296cc85 | 3380 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
5f634d06 AK |
3381 | ext4_da_update_reserve_space(inode, allocated, 1); |
3382 | ||
b436b9be JK |
3383 | /* |
3384 | * Cache the extent and update transaction to commit on fdatasync only | |
3385 | * when it is _not_ an uninitialized extent. | |
3386 | */ | |
3387 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 3388 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
3389 | ext4_update_inode_fsync_trans(handle, inode, 1); |
3390 | } else | |
3391 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 3392 | out: |
e35fd660 TT |
3393 | if (allocated > map->m_len) |
3394 | allocated = map->m_len; | |
a86c6181 | 3395 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3396 | map->m_flags |= EXT4_MAP_MAPPED; |
3397 | map->m_pblk = newblock; | |
3398 | map->m_len = allocated; | |
a86c6181 AT |
3399 | out2: |
3400 | if (path) { | |
3401 | ext4_ext_drop_refs(path); | |
3402 | kfree(path); | |
3403 | } | |
0562e0ba JZ |
3404 | trace_ext4_ext_map_blocks_exit(inode, map->m_lblk, |
3405 | newblock, map->m_len, err ? err : allocated); | |
a86c6181 AT |
3406 | return err ? err : allocated; |
3407 | } | |
3408 | ||
cf108bca | 3409 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
3410 | { |
3411 | struct address_space *mapping = inode->i_mapping; | |
3412 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 3413 | ext4_lblk_t last_block; |
a86c6181 AT |
3414 | handle_t *handle; |
3415 | int err = 0; | |
3416 | ||
3889fd57 JZ |
3417 | /* |
3418 | * finish any pending end_io work so we won't run the risk of | |
3419 | * converting any truncated blocks to initialized later | |
3420 | */ | |
3421 | ext4_flush_completed_IO(inode); | |
3422 | ||
a86c6181 AT |
3423 | /* |
3424 | * probably first extent we're gonna free will be last in block | |
3425 | */ | |
f3bd1f3f | 3426 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 3427 | handle = ext4_journal_start(inode, err); |
cf108bca | 3428 | if (IS_ERR(handle)) |
a86c6181 | 3429 | return; |
a86c6181 | 3430 | |
cf108bca JK |
3431 | if (inode->i_size & (sb->s_blocksize - 1)) |
3432 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
a86c6181 | 3433 | |
9ddfc3dc JK |
3434 | if (ext4_orphan_add(handle, inode)) |
3435 | goto out_stop; | |
3436 | ||
0e855ac8 | 3437 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
3438 | ext4_ext_invalidate_cache(inode); |
3439 | ||
c2ea3fde | 3440 | ext4_discard_preallocations(inode); |
c9de560d | 3441 | |
a86c6181 | 3442 | /* |
d0d856e8 RD |
3443 | * TODO: optimization is possible here. |
3444 | * Probably we need not scan at all, | |
3445 | * because page truncation is enough. | |
a86c6181 | 3446 | */ |
a86c6181 AT |
3447 | |
3448 | /* we have to know where to truncate from in crash case */ | |
3449 | EXT4_I(inode)->i_disksize = inode->i_size; | |
3450 | ext4_mark_inode_dirty(handle, inode); | |
3451 | ||
3452 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
3453 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
3454 | err = ext4_ext_remove_space(inode, last_block); | |
3455 | ||
3456 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
3457 | * transaction synchronous. |
3458 | */ | |
a86c6181 | 3459 | if (IS_SYNC(inode)) |
0390131b | 3460 | ext4_handle_sync(handle); |
a86c6181 | 3461 | |
9ddfc3dc | 3462 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
3463 | |
3464 | out_stop: | |
a86c6181 | 3465 | /* |
d0d856e8 | 3466 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
3467 | * then we need to clear up the orphan record which we created above. |
3468 | * However, if this was a real unlink then we were called by | |
3469 | * ext4_delete_inode(), and we allow that function to clean up the | |
3470 | * orphan info for us. | |
3471 | */ | |
3472 | if (inode->i_nlink) | |
3473 | ext4_orphan_del(handle, inode); | |
3474 | ||
ef737728 SR |
3475 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3476 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
3477 | ext4_journal_stop(handle); |
3478 | } | |
3479 | ||
fd28784a AK |
3480 | static void ext4_falloc_update_inode(struct inode *inode, |
3481 | int mode, loff_t new_size, int update_ctime) | |
3482 | { | |
3483 | struct timespec now; | |
3484 | ||
3485 | if (update_ctime) { | |
3486 | now = current_fs_time(inode->i_sb); | |
3487 | if (!timespec_equal(&inode->i_ctime, &now)) | |
3488 | inode->i_ctime = now; | |
3489 | } | |
3490 | /* | |
3491 | * Update only when preallocation was requested beyond | |
3492 | * the file size. | |
3493 | */ | |
cf17fea6 AK |
3494 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
3495 | if (new_size > i_size_read(inode)) | |
3496 | i_size_write(inode, new_size); | |
3497 | if (new_size > EXT4_I(inode)->i_disksize) | |
3498 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
3499 | } else { |
3500 | /* | |
3501 | * Mark that we allocate beyond EOF so the subsequent truncate | |
3502 | * can proceed even if the new size is the same as i_size. | |
3503 | */ | |
3504 | if (new_size > i_size_read(inode)) | |
12e9b892 | 3505 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
3506 | } |
3507 | ||
3508 | } | |
3509 | ||
a2df2a63 | 3510 | /* |
2fe17c10 | 3511 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
3512 | * operation, which gets called from sys_fallocate system call. |
3513 | * For block-mapped files, posix_fallocate should fall back to the method | |
3514 | * of writing zeroes to the required new blocks (the same behavior which is | |
3515 | * expected for file systems which do not support fallocate() system call). | |
3516 | */ | |
2fe17c10 | 3517 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 3518 | { |
2fe17c10 | 3519 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 3520 | handle_t *handle; |
fd28784a | 3521 | loff_t new_size; |
498e5f24 | 3522 | unsigned int max_blocks; |
a2df2a63 AA |
3523 | int ret = 0; |
3524 | int ret2 = 0; | |
3525 | int retries = 0; | |
2ed88685 | 3526 | struct ext4_map_blocks map; |
a2df2a63 AA |
3527 | unsigned int credits, blkbits = inode->i_blkbits; |
3528 | ||
d6dc8462 | 3529 | /* We only support the FALLOC_FL_KEEP_SIZE mode */ |
64c23e86 | 3530 | if (mode & ~FALLOC_FL_KEEP_SIZE) |
d6dc8462 JB |
3531 | return -EOPNOTSUPP; |
3532 | ||
a2df2a63 AA |
3533 | /* |
3534 | * currently supporting (pre)allocate mode for extent-based | |
3535 | * files _only_ | |
3536 | */ | |
12e9b892 | 3537 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
a2df2a63 AA |
3538 | return -EOPNOTSUPP; |
3539 | ||
0562e0ba | 3540 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 3541 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
3542 | /* |
3543 | * We can't just convert len to max_blocks because | |
3544 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
3545 | */ | |
a2df2a63 | 3546 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 3547 | - map.m_lblk; |
a2df2a63 | 3548 | /* |
f3bd1f3f | 3549 | * credits to insert 1 extent into extent tree |
a2df2a63 | 3550 | */ |
f3bd1f3f | 3551 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 3552 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
3553 | ret = inode_newsize_ok(inode, (len + offset)); |
3554 | if (ret) { | |
3555 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 3556 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
3557 | return ret; |
3558 | } | |
a2df2a63 AA |
3559 | retry: |
3560 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
3561 | map.m_lblk = map.m_lblk + ret; |
3562 | map.m_len = max_blocks = max_blocks - ret; | |
a2df2a63 AA |
3563 | handle = ext4_journal_start(inode, credits); |
3564 | if (IS_ERR(handle)) { | |
3565 | ret = PTR_ERR(handle); | |
3566 | break; | |
3567 | } | |
2ed88685 | 3568 | ret = ext4_map_blocks(handle, inode, &map, |
c2177057 | 3569 | EXT4_GET_BLOCKS_CREATE_UNINIT_EXT); |
221879c9 | 3570 | if (ret <= 0) { |
2c98615d AK |
3571 | #ifdef EXT4FS_DEBUG |
3572 | WARN_ON(ret <= 0); | |
e35fd660 | 3573 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
2c98615d | 3574 | "returned error inode#%lu, block=%u, " |
9fd9784c | 3575 | "max_blocks=%u", __func__, |
a6371b63 | 3576 | inode->i_ino, map.m_lblk, max_blocks); |
2c98615d | 3577 | #endif |
a2df2a63 AA |
3578 | ext4_mark_inode_dirty(handle, inode); |
3579 | ret2 = ext4_journal_stop(handle); | |
3580 | break; | |
3581 | } | |
2ed88685 | 3582 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
3583 | blkbits) >> blkbits)) |
3584 | new_size = offset + len; | |
3585 | else | |
2ed88685 | 3586 | new_size = (map.m_lblk + ret) << blkbits; |
a2df2a63 | 3587 | |
fd28784a | 3588 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 3589 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 AA |
3590 | ext4_mark_inode_dirty(handle, inode); |
3591 | ret2 = ext4_journal_stop(handle); | |
3592 | if (ret2) | |
3593 | break; | |
3594 | } | |
fd28784a AK |
3595 | if (ret == -ENOSPC && |
3596 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
3597 | ret = 0; | |
a2df2a63 | 3598 | goto retry; |
a2df2a63 | 3599 | } |
55bd725a | 3600 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
3601 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
3602 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
3603 | return ret > 0 ? ret2 : ret; |
3604 | } | |
6873fa0d | 3605 | |
0031462b MC |
3606 | /* |
3607 | * This function convert a range of blocks to written extents | |
3608 | * The caller of this function will pass the start offset and the size. | |
3609 | * all unwritten extents within this range will be converted to | |
3610 | * written extents. | |
3611 | * | |
3612 | * This function is called from the direct IO end io call back | |
3613 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 3614 | * Returns 0 on success. |
0031462b MC |
3615 | */ |
3616 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 3617 | ssize_t len) |
0031462b MC |
3618 | { |
3619 | handle_t *handle; | |
0031462b MC |
3620 | unsigned int max_blocks; |
3621 | int ret = 0; | |
3622 | int ret2 = 0; | |
2ed88685 | 3623 | struct ext4_map_blocks map; |
0031462b MC |
3624 | unsigned int credits, blkbits = inode->i_blkbits; |
3625 | ||
2ed88685 | 3626 | map.m_lblk = offset >> blkbits; |
0031462b MC |
3627 | /* |
3628 | * We can't just convert len to max_blocks because | |
3629 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
3630 | */ | |
2ed88685 TT |
3631 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
3632 | map.m_lblk); | |
0031462b MC |
3633 | /* |
3634 | * credits to insert 1 extent into extent tree | |
3635 | */ | |
3636 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
3637 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
3638 | map.m_lblk += ret; |
3639 | map.m_len = (max_blocks -= ret); | |
0031462b MC |
3640 | handle = ext4_journal_start(inode, credits); |
3641 | if (IS_ERR(handle)) { | |
3642 | ret = PTR_ERR(handle); | |
3643 | break; | |
3644 | } | |
2ed88685 | 3645 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 3646 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
0031462b MC |
3647 | if (ret <= 0) { |
3648 | WARN_ON(ret <= 0); | |
e35fd660 | 3649 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
0031462b MC |
3650 | "returned error inode#%lu, block=%u, " |
3651 | "max_blocks=%u", __func__, | |
2ed88685 | 3652 | inode->i_ino, map.m_lblk, map.m_len); |
0031462b MC |
3653 | } |
3654 | ext4_mark_inode_dirty(handle, inode); | |
3655 | ret2 = ext4_journal_stop(handle); | |
3656 | if (ret <= 0 || ret2 ) | |
3657 | break; | |
3658 | } | |
3659 | return ret > 0 ? ret2 : ret; | |
3660 | } | |
6d9c85eb | 3661 | |
6873fa0d ES |
3662 | /* |
3663 | * Callback function called for each extent to gather FIEMAP information. | |
3664 | */ | |
3a06d778 | 3665 | static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path, |
6873fa0d ES |
3666 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
3667 | void *data) | |
3668 | { | |
6873fa0d ES |
3669 | __u64 logical; |
3670 | __u64 physical; | |
3671 | __u64 length; | |
6d9c85eb | 3672 | loff_t size; |
6873fa0d | 3673 | __u32 flags = 0; |
6d9c85eb YY |
3674 | int ret = 0; |
3675 | struct fiemap_extent_info *fieinfo = data; | |
3676 | unsigned char blksize_bits; | |
6873fa0d | 3677 | |
6d9c85eb YY |
3678 | blksize_bits = inode->i_sb->s_blocksize_bits; |
3679 | logical = (__u64)newex->ec_block << blksize_bits; | |
6873fa0d | 3680 | |
b05e6ae5 | 3681 | if (newex->ec_start == 0) { |
6d9c85eb YY |
3682 | /* |
3683 | * No extent in extent-tree contains block @newex->ec_start, | |
3684 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
3685 | * | |
3686 | * Holes or delayed-extents are processed as follows. | |
3687 | * 1. lookup dirty pages with specified range in pagecache. | |
3688 | * If no page is got, then there is no delayed-extent and | |
3689 | * return with EXT_CONTINUE. | |
3690 | * 2. find the 1st mapped buffer, | |
3691 | * 3. check if the mapped buffer is both in the request range | |
3692 | * and a delayed buffer. If not, there is no delayed-extent, | |
3693 | * then return. | |
3694 | * 4. a delayed-extent is found, the extent will be collected. | |
3695 | */ | |
3696 | ext4_lblk_t end = 0; | |
3697 | pgoff_t last_offset; | |
3698 | pgoff_t offset; | |
3699 | pgoff_t index; | |
b221349f | 3700 | pgoff_t start_index = 0; |
6d9c85eb | 3701 | struct page **pages = NULL; |
6873fa0d | 3702 | struct buffer_head *bh = NULL; |
6d9c85eb YY |
3703 | struct buffer_head *head = NULL; |
3704 | unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *); | |
3705 | ||
3706 | pages = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
3707 | if (pages == NULL) | |
3708 | return -ENOMEM; | |
6873fa0d ES |
3709 | |
3710 | offset = logical >> PAGE_SHIFT; | |
6d9c85eb YY |
3711 | repeat: |
3712 | last_offset = offset; | |
3713 | head = NULL; | |
3714 | ret = find_get_pages_tag(inode->i_mapping, &offset, | |
3715 | PAGECACHE_TAG_DIRTY, nr_pages, pages); | |
3716 | ||
3717 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
3718 | /* First time, try to find a mapped buffer. */ | |
3719 | if (ret == 0) { | |
3720 | out: | |
3721 | for (index = 0; index < ret; index++) | |
3722 | page_cache_release(pages[index]); | |
3723 | /* just a hole. */ | |
3724 | kfree(pages); | |
3725 | return EXT_CONTINUE; | |
3726 | } | |
b221349f | 3727 | index = 0; |
6873fa0d | 3728 | |
b221349f | 3729 | next_page: |
6d9c85eb | 3730 | /* Try to find the 1st mapped buffer. */ |
b221349f | 3731 | end = ((__u64)pages[index]->index << PAGE_SHIFT) >> |
6d9c85eb | 3732 | blksize_bits; |
b221349f | 3733 | if (!page_has_buffers(pages[index])) |
6d9c85eb | 3734 | goto out; |
b221349f | 3735 | head = page_buffers(pages[index]); |
6d9c85eb YY |
3736 | if (!head) |
3737 | goto out; | |
6873fa0d | 3738 | |
b221349f | 3739 | index++; |
6d9c85eb YY |
3740 | bh = head; |
3741 | do { | |
b221349f YY |
3742 | if (end >= newex->ec_block + |
3743 | newex->ec_len) | |
3744 | /* The buffer is out of | |
3745 | * the request range. | |
3746 | */ | |
3747 | goto out; | |
3748 | ||
3749 | if (buffer_mapped(bh) && | |
3750 | end >= newex->ec_block) { | |
3751 | start_index = index - 1; | |
6d9c85eb | 3752 | /* get the 1st mapped buffer. */ |
6d9c85eb YY |
3753 | goto found_mapped_buffer; |
3754 | } | |
b221349f | 3755 | |
6d9c85eb YY |
3756 | bh = bh->b_this_page; |
3757 | end++; | |
3758 | } while (bh != head); | |
6873fa0d | 3759 | |
b221349f YY |
3760 | /* No mapped buffer in the range found in this page, |
3761 | * We need to look up next page. | |
3762 | */ | |
3763 | if (index >= ret) { | |
3764 | /* There is no page left, but we need to limit | |
3765 | * newex->ec_len. | |
3766 | */ | |
3767 | newex->ec_len = end - newex->ec_block; | |
3768 | goto out; | |
3769 | } | |
3770 | goto next_page; | |
6873fa0d | 3771 | } else { |
6d9c85eb YY |
3772 | /*Find contiguous delayed buffers. */ |
3773 | if (ret > 0 && pages[0]->index == last_offset) | |
3774 | head = page_buffers(pages[0]); | |
3775 | bh = head; | |
b221349f YY |
3776 | index = 1; |
3777 | start_index = 0; | |
6873fa0d | 3778 | } |
6d9c85eb YY |
3779 | |
3780 | found_mapped_buffer: | |
3781 | if (bh != NULL && buffer_delay(bh)) { | |
3782 | /* 1st or contiguous delayed buffer found. */ | |
3783 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
3784 | /* | |
3785 | * 1st delayed buffer found, record | |
3786 | * the start of extent. | |
3787 | */ | |
3788 | flags |= FIEMAP_EXTENT_DELALLOC; | |
3789 | newex->ec_block = end; | |
3790 | logical = (__u64)end << blksize_bits; | |
3791 | } | |
3792 | /* Find contiguous delayed buffers. */ | |
3793 | do { | |
3794 | if (!buffer_delay(bh)) | |
3795 | goto found_delayed_extent; | |
3796 | bh = bh->b_this_page; | |
3797 | end++; | |
3798 | } while (bh != head); | |
3799 | ||
b221349f | 3800 | for (; index < ret; index++) { |
6d9c85eb YY |
3801 | if (!page_has_buffers(pages[index])) { |
3802 | bh = NULL; | |
3803 | break; | |
3804 | } | |
3805 | head = page_buffers(pages[index]); | |
3806 | if (!head) { | |
3807 | bh = NULL; | |
3808 | break; | |
3809 | } | |
b221349f | 3810 | |
6d9c85eb | 3811 | if (pages[index]->index != |
b221349f YY |
3812 | pages[start_index]->index + index |
3813 | - start_index) { | |
6d9c85eb YY |
3814 | /* Blocks are not contiguous. */ |
3815 | bh = NULL; | |
3816 | break; | |
3817 | } | |
3818 | bh = head; | |
3819 | do { | |
3820 | if (!buffer_delay(bh)) | |
3821 | /* Delayed-extent ends. */ | |
3822 | goto found_delayed_extent; | |
3823 | bh = bh->b_this_page; | |
3824 | end++; | |
3825 | } while (bh != head); | |
3826 | } | |
3827 | } else if (!(flags & FIEMAP_EXTENT_DELALLOC)) | |
3828 | /* a hole found. */ | |
3829 | goto out; | |
3830 | ||
3831 | found_delayed_extent: | |
3832 | newex->ec_len = min(end - newex->ec_block, | |
3833 | (ext4_lblk_t)EXT_INIT_MAX_LEN); | |
3834 | if (ret == nr_pages && bh != NULL && | |
3835 | newex->ec_len < EXT_INIT_MAX_LEN && | |
3836 | buffer_delay(bh)) { | |
3837 | /* Have not collected an extent and continue. */ | |
3838 | for (index = 0; index < ret; index++) | |
3839 | page_cache_release(pages[index]); | |
3840 | goto repeat; | |
6873fa0d | 3841 | } |
6d9c85eb YY |
3842 | |
3843 | for (index = 0; index < ret; index++) | |
3844 | page_cache_release(pages[index]); | |
3845 | kfree(pages); | |
6873fa0d ES |
3846 | } |
3847 | ||
3848 | physical = (__u64)newex->ec_start << blksize_bits; | |
3849 | length = (__u64)newex->ec_len << blksize_bits; | |
3850 | ||
3851 | if (ex && ext4_ext_is_uninitialized(ex)) | |
3852 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
3853 | ||
6d9c85eb YY |
3854 | size = i_size_read(inode); |
3855 | if (logical + length >= size) | |
6873fa0d ES |
3856 | flags |= FIEMAP_EXTENT_LAST; |
3857 | ||
6d9c85eb | 3858 | ret = fiemap_fill_next_extent(fieinfo, logical, physical, |
6873fa0d | 3859 | length, flags); |
6d9c85eb YY |
3860 | if (ret < 0) |
3861 | return ret; | |
3862 | if (ret == 1) | |
6873fa0d | 3863 | return EXT_BREAK; |
6873fa0d ES |
3864 | return EXT_CONTINUE; |
3865 | } | |
3866 | ||
3867 | /* fiemap flags we can handle specified here */ | |
3868 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
3869 | ||
3a06d778 AK |
3870 | static int ext4_xattr_fiemap(struct inode *inode, |
3871 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
3872 | { |
3873 | __u64 physical = 0; | |
3874 | __u64 length; | |
3875 | __u32 flags = FIEMAP_EXTENT_LAST; | |
3876 | int blockbits = inode->i_sb->s_blocksize_bits; | |
3877 | int error = 0; | |
3878 | ||
3879 | /* in-inode? */ | |
19f5fb7a | 3880 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
3881 | struct ext4_iloc iloc; |
3882 | int offset; /* offset of xattr in inode */ | |
3883 | ||
3884 | error = ext4_get_inode_loc(inode, &iloc); | |
3885 | if (error) | |
3886 | return error; | |
3887 | physical = iloc.bh->b_blocknr << blockbits; | |
3888 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
3889 | EXT4_I(inode)->i_extra_isize; | |
3890 | physical += offset; | |
3891 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
3892 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 3893 | brelse(iloc.bh); |
6873fa0d ES |
3894 | } else { /* external block */ |
3895 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
3896 | length = inode->i_sb->s_blocksize; | |
3897 | } | |
3898 | ||
3899 | if (physical) | |
3900 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
3901 | length, flags); | |
3902 | return (error < 0 ? error : 0); | |
3903 | } | |
3904 | ||
3905 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
3906 | __u64 start, __u64 len) | |
3907 | { | |
3908 | ext4_lblk_t start_blk; | |
6873fa0d ES |
3909 | int error = 0; |
3910 | ||
3911 | /* fallback to generic here if not in extents fmt */ | |
12e9b892 | 3912 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
3913 | return generic_block_fiemap(inode, fieinfo, start, len, |
3914 | ext4_get_block); | |
3915 | ||
3916 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
3917 | return -EBADR; | |
3918 | ||
3919 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
3920 | error = ext4_xattr_fiemap(inode, fieinfo); | |
3921 | } else { | |
aca92ff6 LM |
3922 | ext4_lblk_t len_blks; |
3923 | __u64 last_blk; | |
3924 | ||
6873fa0d | 3925 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 LM |
3926 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
3927 | if (last_blk >= EXT_MAX_BLOCK) | |
3928 | last_blk = EXT_MAX_BLOCK-1; | |
3929 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; | |
6873fa0d ES |
3930 | |
3931 | /* | |
3932 | * Walk the extent tree gathering extent information. | |
3933 | * ext4_ext_fiemap_cb will push extents back to user. | |
3934 | */ | |
6873fa0d ES |
3935 | error = ext4_ext_walk_space(inode, start_blk, len_blks, |
3936 | ext4_ext_fiemap_cb, fieinfo); | |
6873fa0d ES |
3937 | } |
3938 | ||
3939 | return error; | |
3940 | } |