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 AT |
46 | |
47 | ||
d0d856e8 RD |
48 | /* |
49 | * ext_pblock: | |
50 | * combine low and high parts of physical block number into ext4_fsblk_t | |
51 | */ | |
748de673 | 52 | ext4_fsblk_t ext_pblock(struct ext4_extent *ex) |
f65e6fba AT |
53 | { |
54 | ext4_fsblk_t block; | |
55 | ||
b377611d | 56 | block = le32_to_cpu(ex->ee_start_lo); |
9b8f1f01 | 57 | block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1; |
f65e6fba AT |
58 | return block; |
59 | } | |
60 | ||
d0d856e8 RD |
61 | /* |
62 | * idx_pblock: | |
63 | * combine low and high parts of a leaf physical block number into ext4_fsblk_t | |
64 | */ | |
c14c6fd5 | 65 | ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix) |
f65e6fba AT |
66 | { |
67 | ext4_fsblk_t block; | |
68 | ||
d8dd0b45 | 69 | block = le32_to_cpu(ix->ei_leaf_lo); |
9b8f1f01 | 70 | block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1; |
f65e6fba AT |
71 | return block; |
72 | } | |
73 | ||
d0d856e8 RD |
74 | /* |
75 | * ext4_ext_store_pblock: | |
76 | * stores a large physical block number into an extent struct, | |
77 | * breaking it into parts | |
78 | */ | |
c14c6fd5 | 79 | void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb) |
f65e6fba | 80 | { |
b377611d | 81 | ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
9b8f1f01 | 82 | ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
f65e6fba AT |
83 | } |
84 | ||
d0d856e8 RD |
85 | /* |
86 | * ext4_idx_store_pblock: | |
87 | * stores a large physical block number into an index struct, | |
88 | * breaking it into parts | |
89 | */ | |
09b88252 | 90 | static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb) |
f65e6fba | 91 | { |
d8dd0b45 | 92 | ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
9b8f1f01 | 93 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
f65e6fba AT |
94 | } |
95 | ||
487caeef JK |
96 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
97 | struct inode *inode, | |
98 | int needed) | |
a86c6181 AT |
99 | { |
100 | int err; | |
101 | ||
0390131b FM |
102 | if (!ext4_handle_valid(handle)) |
103 | return 0; | |
a86c6181 | 104 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
105 | return 0; |
106 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 107 | if (err <= 0) |
9102e4fa | 108 | return err; |
487caeef JK |
109 | err = ext4_truncate_restart_trans(handle, inode, needed); |
110 | /* | |
111 | * We have dropped i_data_sem so someone might have cached again | |
112 | * an extent we are going to truncate. | |
113 | */ | |
114 | ext4_ext_invalidate_cache(inode); | |
115 | ||
116 | return err; | |
a86c6181 AT |
117 | } |
118 | ||
119 | /* | |
120 | * could return: | |
121 | * - EROFS | |
122 | * - ENOMEM | |
123 | */ | |
124 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
125 | struct ext4_ext_path *path) | |
126 | { | |
127 | if (path->p_bh) { | |
128 | /* path points to block */ | |
129 | return ext4_journal_get_write_access(handle, path->p_bh); | |
130 | } | |
131 | /* path points to leaf/index in inode body */ | |
132 | /* we use in-core data, no need to protect them */ | |
133 | return 0; | |
134 | } | |
135 | ||
136 | /* | |
137 | * could return: | |
138 | * - EROFS | |
139 | * - ENOMEM | |
140 | * - EIO | |
141 | */ | |
142 | static int ext4_ext_dirty(handle_t *handle, struct inode *inode, | |
143 | struct ext4_ext_path *path) | |
144 | { | |
145 | int err; | |
146 | if (path->p_bh) { | |
147 | /* path points to block */ | |
0390131b | 148 | err = ext4_handle_dirty_metadata(handle, inode, path->p_bh); |
a86c6181 AT |
149 | } else { |
150 | /* path points to leaf/index in inode body */ | |
151 | err = ext4_mark_inode_dirty(handle, inode); | |
152 | } | |
153 | return err; | |
154 | } | |
155 | ||
f65e6fba | 156 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 157 | struct ext4_ext_path *path, |
725d26d3 | 158 | ext4_lblk_t block) |
a86c6181 AT |
159 | { |
160 | struct ext4_inode_info *ei = EXT4_I(inode); | |
f65e6fba | 161 | ext4_fsblk_t bg_start; |
74d3487f | 162 | ext4_fsblk_t last_block; |
f65e6fba | 163 | ext4_grpblk_t colour; |
a4912123 TT |
164 | ext4_group_t block_group; |
165 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
a86c6181 AT |
166 | int depth; |
167 | ||
168 | if (path) { | |
169 | struct ext4_extent *ex; | |
170 | depth = path->p_depth; | |
171 | ||
172 | /* try to predict block placement */ | |
7e028976 AM |
173 | ex = path[depth].p_ext; |
174 | if (ex) | |
f65e6fba | 175 | return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block)); |
a86c6181 | 176 | |
d0d856e8 RD |
177 | /* it looks like index is empty; |
178 | * try to find starting block from index itself */ | |
a86c6181 AT |
179 | if (path[depth].p_bh) |
180 | return path[depth].p_bh->b_blocknr; | |
181 | } | |
182 | ||
183 | /* OK. use inode's group */ | |
a4912123 TT |
184 | block_group = ei->i_block_group; |
185 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
186 | /* | |
187 | * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME | |
188 | * block groups per flexgroup, reserve the first block | |
189 | * group for directories and special files. Regular | |
190 | * files will start at the second block group. This | |
191 | * tends to speed up directory access and improves | |
192 | * fsck times. | |
193 | */ | |
194 | block_group &= ~(flex_size-1); | |
195 | if (S_ISREG(inode->i_mode)) | |
196 | block_group++; | |
197 | } | |
198 | bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) + | |
a86c6181 | 199 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block); |
74d3487f VC |
200 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
201 | ||
a4912123 TT |
202 | /* |
203 | * If we are doing delayed allocation, we don't need take | |
204 | * colour into account. | |
205 | */ | |
206 | if (test_opt(inode->i_sb, DELALLOC)) | |
207 | return bg_start; | |
208 | ||
74d3487f VC |
209 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
210 | colour = (current->pid % 16) * | |
a86c6181 | 211 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
212 | else |
213 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
a86c6181 AT |
214 | return bg_start + colour + block; |
215 | } | |
216 | ||
654b4908 AK |
217 | /* |
218 | * Allocation for a meta data block | |
219 | */ | |
f65e6fba | 220 | static ext4_fsblk_t |
654b4908 | 221 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 AT |
222 | struct ext4_ext_path *path, |
223 | struct ext4_extent *ex, int *err) | |
224 | { | |
f65e6fba | 225 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
226 | |
227 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
97df5d15 | 228 | newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err); |
a86c6181 AT |
229 | return newblock; |
230 | } | |
231 | ||
09b88252 | 232 | static int ext4_ext_space_block(struct inode *inode) |
a86c6181 AT |
233 | { |
234 | int size; | |
235 | ||
236 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
237 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 238 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
239 | if (size > 6) |
240 | size = 6; | |
241 | #endif | |
242 | return size; | |
243 | } | |
244 | ||
09b88252 | 245 | static int ext4_ext_space_block_idx(struct inode *inode) |
a86c6181 AT |
246 | { |
247 | int size; | |
248 | ||
249 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
250 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 251 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
252 | if (size > 5) |
253 | size = 5; | |
254 | #endif | |
255 | return size; | |
256 | } | |
257 | ||
09b88252 | 258 | static int ext4_ext_space_root(struct inode *inode) |
a86c6181 AT |
259 | { |
260 | int size; | |
261 | ||
262 | size = sizeof(EXT4_I(inode)->i_data); | |
263 | size -= sizeof(struct ext4_extent_header); | |
264 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 265 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
266 | if (size > 3) |
267 | size = 3; | |
268 | #endif | |
269 | return size; | |
270 | } | |
271 | ||
09b88252 | 272 | static int ext4_ext_space_root_idx(struct inode *inode) |
a86c6181 AT |
273 | { |
274 | int size; | |
275 | ||
276 | size = sizeof(EXT4_I(inode)->i_data); | |
277 | size -= sizeof(struct ext4_extent_header); | |
278 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 279 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
280 | if (size > 4) |
281 | size = 4; | |
282 | #endif | |
283 | return size; | |
284 | } | |
285 | ||
d2a17637 MC |
286 | /* |
287 | * Calculate the number of metadata blocks needed | |
288 | * to allocate @blocks | |
289 | * Worse case is one block per extent | |
290 | */ | |
291 | int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks) | |
292 | { | |
293 | int lcap, icap, rcap, leafs, idxs, num; | |
294 | int newextents = blocks; | |
295 | ||
296 | rcap = ext4_ext_space_root_idx(inode); | |
297 | lcap = ext4_ext_space_block(inode); | |
298 | icap = ext4_ext_space_block_idx(inode); | |
299 | ||
300 | /* number of new leaf blocks needed */ | |
301 | num = leafs = (newextents + lcap - 1) / lcap; | |
302 | ||
303 | /* | |
304 | * Worse case, we need separate index block(s) | |
305 | * to link all new leaf blocks | |
306 | */ | |
307 | idxs = (leafs + icap - 1) / icap; | |
308 | do { | |
309 | num += idxs; | |
310 | idxs = (idxs + icap - 1) / icap; | |
311 | } while (idxs > rcap); | |
312 | ||
313 | return num; | |
314 | } | |
315 | ||
c29c0ae7 AT |
316 | static int |
317 | ext4_ext_max_entries(struct inode *inode, int depth) | |
318 | { | |
319 | int max; | |
320 | ||
321 | if (depth == ext_depth(inode)) { | |
322 | if (depth == 0) | |
323 | max = ext4_ext_space_root(inode); | |
324 | else | |
325 | max = ext4_ext_space_root_idx(inode); | |
326 | } else { | |
327 | if (depth == 0) | |
328 | max = ext4_ext_space_block(inode); | |
329 | else | |
330 | max = ext4_ext_space_block_idx(inode); | |
331 | } | |
332 | ||
333 | return max; | |
334 | } | |
335 | ||
56b19868 AK |
336 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
337 | { | |
6fd058f7 | 338 | ext4_fsblk_t block = ext_pblock(ext); |
56b19868 | 339 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 340 | |
6fd058f7 | 341 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
342 | } |
343 | ||
344 | static int ext4_valid_extent_idx(struct inode *inode, | |
345 | struct ext4_extent_idx *ext_idx) | |
346 | { | |
6fd058f7 | 347 | ext4_fsblk_t block = idx_pblock(ext_idx); |
e84a26ce | 348 | |
6fd058f7 | 349 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
350 | } |
351 | ||
352 | static int ext4_valid_extent_entries(struct inode *inode, | |
353 | struct ext4_extent_header *eh, | |
354 | int depth) | |
355 | { | |
356 | struct ext4_extent *ext; | |
357 | struct ext4_extent_idx *ext_idx; | |
358 | unsigned short entries; | |
359 | if (eh->eh_entries == 0) | |
360 | return 1; | |
361 | ||
362 | entries = le16_to_cpu(eh->eh_entries); | |
363 | ||
364 | if (depth == 0) { | |
365 | /* leaf entries */ | |
366 | ext = EXT_FIRST_EXTENT(eh); | |
367 | while (entries) { | |
368 | if (!ext4_valid_extent(inode, ext)) | |
369 | return 0; | |
370 | ext++; | |
371 | entries--; | |
372 | } | |
373 | } else { | |
374 | ext_idx = EXT_FIRST_INDEX(eh); | |
375 | while (entries) { | |
376 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
377 | return 0; | |
378 | ext_idx++; | |
379 | entries--; | |
380 | } | |
381 | } | |
382 | return 1; | |
383 | } | |
384 | ||
385 | static int __ext4_ext_check(const char *function, struct inode *inode, | |
c29c0ae7 AT |
386 | struct ext4_extent_header *eh, |
387 | int depth) | |
388 | { | |
389 | const char *error_msg; | |
390 | int max = 0; | |
391 | ||
392 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
393 | error_msg = "invalid magic"; | |
394 | goto corrupted; | |
395 | } | |
396 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
397 | error_msg = "unexpected eh_depth"; | |
398 | goto corrupted; | |
399 | } | |
400 | if (unlikely(eh->eh_max == 0)) { | |
401 | error_msg = "invalid eh_max"; | |
402 | goto corrupted; | |
403 | } | |
404 | max = ext4_ext_max_entries(inode, depth); | |
405 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
406 | error_msg = "too large eh_max"; | |
407 | goto corrupted; | |
408 | } | |
409 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
410 | error_msg = "invalid eh_entries"; | |
411 | goto corrupted; | |
412 | } | |
56b19868 AK |
413 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
414 | error_msg = "invalid extent entries"; | |
415 | goto corrupted; | |
416 | } | |
c29c0ae7 AT |
417 | return 0; |
418 | ||
419 | corrupted: | |
420 | ext4_error(inode->i_sb, function, | |
56b19868 | 421 | "bad header/extent in inode #%lu: %s - magic %x, " |
c29c0ae7 AT |
422 | "entries %u, max %u(%u), depth %u(%u)", |
423 | inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), | |
424 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
425 | max, le16_to_cpu(eh->eh_depth), depth); | |
426 | ||
427 | return -EIO; | |
428 | } | |
429 | ||
56b19868 AK |
430 | #define ext4_ext_check(inode, eh, depth) \ |
431 | __ext4_ext_check(__func__, inode, eh, depth) | |
c29c0ae7 | 432 | |
7a262f7c AK |
433 | int ext4_ext_check_inode(struct inode *inode) |
434 | { | |
435 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
436 | } | |
437 | ||
a86c6181 AT |
438 | #ifdef EXT_DEBUG |
439 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
440 | { | |
441 | int k, l = path->p_depth; | |
442 | ||
443 | ext_debug("path:"); | |
444 | for (k = 0; k <= l; k++, path++) { | |
445 | if (path->p_idx) { | |
2ae02107 | 446 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
f65e6fba | 447 | idx_pblock(path->p_idx)); |
a86c6181 | 448 | } else if (path->p_ext) { |
553f9008 | 449 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 450 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 451 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 452 | ext4_ext_get_actual_len(path->p_ext), |
f65e6fba | 453 | ext_pblock(path->p_ext)); |
a86c6181 AT |
454 | } else |
455 | ext_debug(" []"); | |
456 | } | |
457 | ext_debug("\n"); | |
458 | } | |
459 | ||
460 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
461 | { | |
462 | int depth = ext_depth(inode); | |
463 | struct ext4_extent_header *eh; | |
464 | struct ext4_extent *ex; | |
465 | int i; | |
466 | ||
467 | if (!path) | |
468 | return; | |
469 | ||
470 | eh = path[depth].p_hdr; | |
471 | ex = EXT_FIRST_EXTENT(eh); | |
472 | ||
553f9008 M |
473 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
474 | ||
a86c6181 | 475 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
476 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
477 | ext4_ext_is_uninitialized(ex), | |
a2df2a63 | 478 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
a86c6181 AT |
479 | } |
480 | ext_debug("\n"); | |
481 | } | |
482 | #else | |
af5bc92d TT |
483 | #define ext4_ext_show_path(inode, path) |
484 | #define ext4_ext_show_leaf(inode, path) | |
a86c6181 AT |
485 | #endif |
486 | ||
b35905c1 | 487 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
488 | { |
489 | int depth = path->p_depth; | |
490 | int i; | |
491 | ||
492 | for (i = 0; i <= depth; i++, path++) | |
493 | if (path->p_bh) { | |
494 | brelse(path->p_bh); | |
495 | path->p_bh = NULL; | |
496 | } | |
497 | } | |
498 | ||
499 | /* | |
d0d856e8 RD |
500 | * ext4_ext_binsearch_idx: |
501 | * binary search for the closest index of the given block | |
c29c0ae7 | 502 | * the header must be checked before calling this |
a86c6181 AT |
503 | */ |
504 | static void | |
725d26d3 AK |
505 | ext4_ext_binsearch_idx(struct inode *inode, |
506 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
507 | { |
508 | struct ext4_extent_header *eh = path->p_hdr; | |
509 | struct ext4_extent_idx *r, *l, *m; | |
510 | ||
a86c6181 | 511 | |
bba90743 | 512 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
513 | |
514 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 515 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
516 | while (l <= r) { |
517 | m = l + (r - l) / 2; | |
518 | if (block < le32_to_cpu(m->ei_block)) | |
519 | r = m - 1; | |
520 | else | |
521 | l = m + 1; | |
26d535ed DM |
522 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
523 | m, le32_to_cpu(m->ei_block), | |
524 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
525 | } |
526 | ||
527 | path->p_idx = l - 1; | |
f65e6fba | 528 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
26d535ed | 529 | idx_pblock(path->p_idx)); |
a86c6181 AT |
530 | |
531 | #ifdef CHECK_BINSEARCH | |
532 | { | |
533 | struct ext4_extent_idx *chix, *ix; | |
534 | int k; | |
535 | ||
536 | chix = ix = EXT_FIRST_INDEX(eh); | |
537 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
538 | if (k != 0 && | |
539 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
540 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
541 | "first=0x%p\n", k, | |
542 | ix, EXT_FIRST_INDEX(eh)); | |
543 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
544 | le32_to_cpu(ix->ei_block), |
545 | le32_to_cpu(ix[-1].ei_block)); | |
546 | } | |
547 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 548 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
549 | if (block < le32_to_cpu(ix->ei_block)) |
550 | break; | |
551 | chix = ix; | |
552 | } | |
553 | BUG_ON(chix != path->p_idx); | |
554 | } | |
555 | #endif | |
556 | ||
557 | } | |
558 | ||
559 | /* | |
d0d856e8 RD |
560 | * ext4_ext_binsearch: |
561 | * binary search for closest extent of the given block | |
c29c0ae7 | 562 | * the header must be checked before calling this |
a86c6181 AT |
563 | */ |
564 | static void | |
725d26d3 AK |
565 | ext4_ext_binsearch(struct inode *inode, |
566 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
567 | { |
568 | struct ext4_extent_header *eh = path->p_hdr; | |
569 | struct ext4_extent *r, *l, *m; | |
570 | ||
a86c6181 AT |
571 | if (eh->eh_entries == 0) { |
572 | /* | |
d0d856e8 RD |
573 | * this leaf is empty: |
574 | * we get such a leaf in split/add case | |
a86c6181 AT |
575 | */ |
576 | return; | |
577 | } | |
578 | ||
bba90743 | 579 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
580 | |
581 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 582 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
583 | |
584 | while (l <= r) { | |
585 | m = l + (r - l) / 2; | |
586 | if (block < le32_to_cpu(m->ee_block)) | |
587 | r = m - 1; | |
588 | else | |
589 | l = m + 1; | |
26d535ed DM |
590 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
591 | m, le32_to_cpu(m->ee_block), | |
592 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
593 | } |
594 | ||
595 | path->p_ext = l - 1; | |
553f9008 | 596 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 DK |
597 | le32_to_cpu(path->p_ext->ee_block), |
598 | ext_pblock(path->p_ext), | |
553f9008 | 599 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 600 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
601 | |
602 | #ifdef CHECK_BINSEARCH | |
603 | { | |
604 | struct ext4_extent *chex, *ex; | |
605 | int k; | |
606 | ||
607 | chex = ex = EXT_FIRST_EXTENT(eh); | |
608 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
609 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 610 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
611 | if (block < le32_to_cpu(ex->ee_block)) |
612 | break; | |
613 | chex = ex; | |
614 | } | |
615 | BUG_ON(chex != path->p_ext); | |
616 | } | |
617 | #endif | |
618 | ||
619 | } | |
620 | ||
621 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
622 | { | |
623 | struct ext4_extent_header *eh; | |
624 | ||
625 | eh = ext_inode_hdr(inode); | |
626 | eh->eh_depth = 0; | |
627 | eh->eh_entries = 0; | |
628 | eh->eh_magic = EXT4_EXT_MAGIC; | |
629 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode)); | |
630 | ext4_mark_inode_dirty(handle, inode); | |
631 | ext4_ext_invalidate_cache(inode); | |
632 | return 0; | |
633 | } | |
634 | ||
635 | struct ext4_ext_path * | |
725d26d3 AK |
636 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
637 | struct ext4_ext_path *path) | |
a86c6181 AT |
638 | { |
639 | struct ext4_extent_header *eh; | |
640 | struct buffer_head *bh; | |
641 | short int depth, i, ppos = 0, alloc = 0; | |
642 | ||
643 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 644 | depth = ext_depth(inode); |
a86c6181 AT |
645 | |
646 | /* account possible depth increase */ | |
647 | if (!path) { | |
5d4958f9 | 648 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
649 | GFP_NOFS); |
650 | if (!path) | |
651 | return ERR_PTR(-ENOMEM); | |
652 | alloc = 1; | |
653 | } | |
a86c6181 | 654 | path[0].p_hdr = eh; |
1973adcb | 655 | path[0].p_bh = NULL; |
a86c6181 | 656 | |
c29c0ae7 | 657 | i = depth; |
a86c6181 AT |
658 | /* walk through the tree */ |
659 | while (i) { | |
7a262f7c AK |
660 | int need_to_validate = 0; |
661 | ||
a86c6181 AT |
662 | ext_debug("depth %d: num %d, max %d\n", |
663 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 664 | |
a86c6181 | 665 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
f65e6fba | 666 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
667 | path[ppos].p_depth = i; |
668 | path[ppos].p_ext = NULL; | |
669 | ||
7a262f7c AK |
670 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
671 | if (unlikely(!bh)) | |
a86c6181 | 672 | goto err; |
7a262f7c AK |
673 | if (!bh_uptodate_or_lock(bh)) { |
674 | if (bh_submit_read(bh) < 0) { | |
675 | put_bh(bh); | |
676 | goto err; | |
677 | } | |
678 | /* validate the extent entries */ | |
679 | need_to_validate = 1; | |
680 | } | |
a86c6181 AT |
681 | eh = ext_block_hdr(bh); |
682 | ppos++; | |
683 | BUG_ON(ppos > depth); | |
684 | path[ppos].p_bh = bh; | |
685 | path[ppos].p_hdr = eh; | |
686 | i--; | |
687 | ||
7a262f7c | 688 | if (need_to_validate && ext4_ext_check(inode, eh, i)) |
a86c6181 AT |
689 | goto err; |
690 | } | |
691 | ||
692 | path[ppos].p_depth = i; | |
a86c6181 AT |
693 | path[ppos].p_ext = NULL; |
694 | path[ppos].p_idx = NULL; | |
695 | ||
a86c6181 AT |
696 | /* find extent */ |
697 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
698 | /* if not an empty leaf */ |
699 | if (path[ppos].p_ext) | |
700 | path[ppos].p_block = ext_pblock(path[ppos].p_ext); | |
a86c6181 AT |
701 | |
702 | ext4_ext_show_path(inode, path); | |
703 | ||
704 | return path; | |
705 | ||
706 | err: | |
707 | ext4_ext_drop_refs(path); | |
708 | if (alloc) | |
709 | kfree(path); | |
710 | return ERR_PTR(-EIO); | |
711 | } | |
712 | ||
713 | /* | |
d0d856e8 RD |
714 | * ext4_ext_insert_index: |
715 | * insert new index [@logical;@ptr] into the block at @curp; | |
716 | * check where to insert: before @curp or after @curp | |
a86c6181 AT |
717 | */ |
718 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |
719 | struct ext4_ext_path *curp, | |
f65e6fba | 720 | int logical, ext4_fsblk_t ptr) |
a86c6181 AT |
721 | { |
722 | struct ext4_extent_idx *ix; | |
723 | int len, err; | |
724 | ||
7e028976 AM |
725 | err = ext4_ext_get_access(handle, inode, curp); |
726 | if (err) | |
a86c6181 AT |
727 | return err; |
728 | ||
729 | BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block)); | |
730 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; | |
731 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
732 | /* insert after */ | |
733 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
734 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
735 | len = len < 0 ? 0 : len; | |
26d535ed | 736 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
737 | "move %d from 0x%p to 0x%p\n", |
738 | logical, ptr, len, | |
739 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
740 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
741 | } | |
742 | ix = curp->p_idx + 1; | |
743 | } else { | |
744 | /* insert before */ | |
745 | len = len * sizeof(struct ext4_extent_idx); | |
746 | len = len < 0 ? 0 : len; | |
26d535ed | 747 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
748 | "move %d from 0x%p to 0x%p\n", |
749 | logical, ptr, len, | |
750 | curp->p_idx, (curp->p_idx + 1)); | |
751 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
752 | ix = curp->p_idx; | |
753 | } | |
754 | ||
755 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 756 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 757 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 AT |
758 | |
759 | BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries) | |
8c55e204 | 760 | > le16_to_cpu(curp->p_hdr->eh_max)); |
a86c6181 AT |
761 | BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr)); |
762 | ||
763 | err = ext4_ext_dirty(handle, inode, curp); | |
764 | ext4_std_error(inode->i_sb, err); | |
765 | ||
766 | return err; | |
767 | } | |
768 | ||
769 | /* | |
d0d856e8 RD |
770 | * ext4_ext_split: |
771 | * inserts new subtree into the path, using free index entry | |
772 | * at depth @at: | |
773 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
774 | * - makes decision where to split | |
775 | * - moves remaining extents and index entries (right to the split point) | |
776 | * into the newly allocated blocks | |
777 | * - initializes subtree | |
a86c6181 AT |
778 | */ |
779 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
780 | struct ext4_ext_path *path, | |
781 | struct ext4_extent *newext, int at) | |
782 | { | |
783 | struct buffer_head *bh = NULL; | |
784 | int depth = ext_depth(inode); | |
785 | struct ext4_extent_header *neh; | |
786 | struct ext4_extent_idx *fidx; | |
787 | struct ext4_extent *ex; | |
788 | int i = at, k, m, a; | |
f65e6fba | 789 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 790 | __le32 border; |
f65e6fba | 791 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
792 | int err = 0; |
793 | ||
794 | /* make decision: where to split? */ | |
d0d856e8 | 795 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 796 | |
d0d856e8 | 797 | /* if current leaf will be split, then we should use |
a86c6181 AT |
798 | * border from split point */ |
799 | BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr)); | |
800 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
801 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 802 | ext_debug("leaf will be split." |
a86c6181 | 803 | " next leaf starts at %d\n", |
8c55e204 | 804 | le32_to_cpu(border)); |
a86c6181 AT |
805 | } else { |
806 | border = newext->ee_block; | |
807 | ext_debug("leaf will be added." | |
808 | " next leaf starts at %d\n", | |
8c55e204 | 809 | le32_to_cpu(border)); |
a86c6181 AT |
810 | } |
811 | ||
812 | /* | |
d0d856e8 RD |
813 | * If error occurs, then we break processing |
814 | * and mark filesystem read-only. index won't | |
a86c6181 | 815 | * be inserted and tree will be in consistent |
d0d856e8 | 816 | * state. Next mount will repair buffers too. |
a86c6181 AT |
817 | */ |
818 | ||
819 | /* | |
d0d856e8 RD |
820 | * Get array to track all allocated blocks. |
821 | * We need this to handle errors and free blocks | |
822 | * upon them. | |
a86c6181 | 823 | */ |
5d4958f9 | 824 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
825 | if (!ablocks) |
826 | return -ENOMEM; | |
a86c6181 AT |
827 | |
828 | /* allocate all needed blocks */ | |
829 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
830 | for (a = 0; a < depth - at; a++) { | |
654b4908 AK |
831 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
832 | newext, &err); | |
a86c6181 AT |
833 | if (newblock == 0) |
834 | goto cleanup; | |
835 | ablocks[a] = newblock; | |
836 | } | |
837 | ||
838 | /* initialize new leaf */ | |
839 | newblock = ablocks[--a]; | |
840 | BUG_ON(newblock == 0); | |
841 | bh = sb_getblk(inode->i_sb, newblock); | |
842 | if (!bh) { | |
843 | err = -EIO; | |
844 | goto cleanup; | |
845 | } | |
846 | lock_buffer(bh); | |
847 | ||
7e028976 AM |
848 | err = ext4_journal_get_create_access(handle, bh); |
849 | if (err) | |
a86c6181 AT |
850 | goto cleanup; |
851 | ||
852 | neh = ext_block_hdr(bh); | |
853 | neh->eh_entries = 0; | |
854 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
855 | neh->eh_magic = EXT4_EXT_MAGIC; | |
856 | neh->eh_depth = 0; | |
857 | ex = EXT_FIRST_EXTENT(neh); | |
858 | ||
d0d856e8 | 859 | /* move remainder of path[depth] to the new leaf */ |
a86c6181 AT |
860 | BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max); |
861 | /* start copy from next extent */ | |
862 | /* TODO: we could do it by single memmove */ | |
863 | m = 0; | |
864 | path[depth].p_ext++; | |
865 | while (path[depth].p_ext <= | |
866 | EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
553f9008 | 867 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", |
8c55e204 DK |
868 | le32_to_cpu(path[depth].p_ext->ee_block), |
869 | ext_pblock(path[depth].p_ext), | |
553f9008 | 870 | ext4_ext_is_uninitialized(path[depth].p_ext), |
a2df2a63 | 871 | ext4_ext_get_actual_len(path[depth].p_ext), |
a86c6181 AT |
872 | newblock); |
873 | /*memmove(ex++, path[depth].p_ext++, | |
874 | sizeof(struct ext4_extent)); | |
875 | neh->eh_entries++;*/ | |
876 | path[depth].p_ext++; | |
877 | m++; | |
878 | } | |
879 | if (m) { | |
880 | memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); | |
e8546d06 | 881 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
882 | } |
883 | ||
884 | set_buffer_uptodate(bh); | |
885 | unlock_buffer(bh); | |
886 | ||
0390131b | 887 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 888 | if (err) |
a86c6181 AT |
889 | goto cleanup; |
890 | brelse(bh); | |
891 | bh = NULL; | |
892 | ||
893 | /* correct old leaf */ | |
894 | if (m) { | |
7e028976 AM |
895 | err = ext4_ext_get_access(handle, inode, path + depth); |
896 | if (err) | |
a86c6181 | 897 | goto cleanup; |
e8546d06 | 898 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
899 | err = ext4_ext_dirty(handle, inode, path + depth); |
900 | if (err) | |
a86c6181 AT |
901 | goto cleanup; |
902 | ||
903 | } | |
904 | ||
905 | /* create intermediate indexes */ | |
906 | k = depth - at - 1; | |
907 | BUG_ON(k < 0); | |
908 | if (k) | |
909 | ext_debug("create %d intermediate indices\n", k); | |
910 | /* insert new index into current index block */ | |
911 | /* current depth stored in i var */ | |
912 | i = depth - 1; | |
913 | while (k--) { | |
914 | oldblock = newblock; | |
915 | newblock = ablocks[--a]; | |
bba90743 | 916 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
917 | if (!bh) { |
918 | err = -EIO; | |
919 | goto cleanup; | |
920 | } | |
921 | lock_buffer(bh); | |
922 | ||
7e028976 AM |
923 | err = ext4_journal_get_create_access(handle, bh); |
924 | if (err) | |
a86c6181 AT |
925 | goto cleanup; |
926 | ||
927 | neh = ext_block_hdr(bh); | |
928 | neh->eh_entries = cpu_to_le16(1); | |
929 | neh->eh_magic = EXT4_EXT_MAGIC; | |
930 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
931 | neh->eh_depth = cpu_to_le16(depth - i); | |
932 | fidx = EXT_FIRST_INDEX(neh); | |
933 | fidx->ei_block = border; | |
f65e6fba | 934 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 935 | |
bba90743 ES |
936 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
937 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 AT |
938 | /* copy indexes */ |
939 | m = 0; | |
940 | path[i].p_idx++; | |
941 | ||
942 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
943 | EXT_MAX_INDEX(path[i].p_hdr)); | |
944 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != | |
945 | EXT_LAST_INDEX(path[i].p_hdr)); | |
946 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { | |
26d535ed | 947 | ext_debug("%d: move %d:%llu in new index %llu\n", i, |
8c55e204 DK |
948 | le32_to_cpu(path[i].p_idx->ei_block), |
949 | idx_pblock(path[i].p_idx), | |
950 | newblock); | |
a86c6181 AT |
951 | /*memmove(++fidx, path[i].p_idx++, |
952 | sizeof(struct ext4_extent_idx)); | |
953 | neh->eh_entries++; | |
954 | BUG_ON(neh->eh_entries > neh->eh_max);*/ | |
955 | path[i].p_idx++; | |
956 | m++; | |
957 | } | |
958 | if (m) { | |
959 | memmove(++fidx, path[i].p_idx - m, | |
960 | sizeof(struct ext4_extent_idx) * m); | |
e8546d06 | 961 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
962 | } |
963 | set_buffer_uptodate(bh); | |
964 | unlock_buffer(bh); | |
965 | ||
0390131b | 966 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 967 | if (err) |
a86c6181 AT |
968 | goto cleanup; |
969 | brelse(bh); | |
970 | bh = NULL; | |
971 | ||
972 | /* correct old index */ | |
973 | if (m) { | |
974 | err = ext4_ext_get_access(handle, inode, path + i); | |
975 | if (err) | |
976 | goto cleanup; | |
e8546d06 | 977 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
978 | err = ext4_ext_dirty(handle, inode, path + i); |
979 | if (err) | |
980 | goto cleanup; | |
981 | } | |
982 | ||
983 | i--; | |
984 | } | |
985 | ||
986 | /* insert new index */ | |
a86c6181 AT |
987 | err = ext4_ext_insert_index(handle, inode, path + at, |
988 | le32_to_cpu(border), newblock); | |
989 | ||
990 | cleanup: | |
991 | if (bh) { | |
992 | if (buffer_locked(bh)) | |
993 | unlock_buffer(bh); | |
994 | brelse(bh); | |
995 | } | |
996 | ||
997 | if (err) { | |
998 | /* free all allocated blocks in error case */ | |
999 | for (i = 0; i < depth; i++) { | |
1000 | if (!ablocks[i]) | |
1001 | continue; | |
c9de560d | 1002 | ext4_free_blocks(handle, inode, ablocks[i], 1, 1); |
a86c6181 AT |
1003 | } |
1004 | } | |
1005 | kfree(ablocks); | |
1006 | ||
1007 | return err; | |
1008 | } | |
1009 | ||
1010 | /* | |
d0d856e8 RD |
1011 | * ext4_ext_grow_indepth: |
1012 | * implements tree growing procedure: | |
1013 | * - allocates new block | |
1014 | * - moves top-level data (index block or leaf) into the new block | |
1015 | * - initializes new top-level, creating index that points to the | |
1016 | * just created block | |
a86c6181 AT |
1017 | */ |
1018 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
1019 | struct ext4_ext_path *path, | |
1020 | struct ext4_extent *newext) | |
1021 | { | |
1022 | struct ext4_ext_path *curp = path; | |
1023 | struct ext4_extent_header *neh; | |
1024 | struct ext4_extent_idx *fidx; | |
1025 | struct buffer_head *bh; | |
f65e6fba | 1026 | ext4_fsblk_t newblock; |
a86c6181 AT |
1027 | int err = 0; |
1028 | ||
654b4908 | 1029 | newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err); |
a86c6181 AT |
1030 | if (newblock == 0) |
1031 | return err; | |
1032 | ||
1033 | bh = sb_getblk(inode->i_sb, newblock); | |
1034 | if (!bh) { | |
1035 | err = -EIO; | |
1036 | ext4_std_error(inode->i_sb, err); | |
1037 | return err; | |
1038 | } | |
1039 | lock_buffer(bh); | |
1040 | ||
7e028976 AM |
1041 | err = ext4_journal_get_create_access(handle, bh); |
1042 | if (err) { | |
a86c6181 AT |
1043 | unlock_buffer(bh); |
1044 | goto out; | |
1045 | } | |
1046 | ||
1047 | /* move top-level index/leaf into new block */ | |
1048 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
1049 | ||
1050 | /* set size of new block */ | |
1051 | neh = ext_block_hdr(bh); | |
1052 | /* old root could have indexes or leaves | |
1053 | * so calculate e_max right way */ | |
1054 | if (ext_depth(inode)) | |
1055 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
1056 | else | |
1057 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
1058 | neh->eh_magic = EXT4_EXT_MAGIC; | |
1059 | set_buffer_uptodate(bh); | |
1060 | unlock_buffer(bh); | |
1061 | ||
0390131b | 1062 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1063 | if (err) |
a86c6181 AT |
1064 | goto out; |
1065 | ||
1066 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
1067 | err = ext4_ext_get_access(handle, inode, curp); |
1068 | if (err) | |
a86c6181 AT |
1069 | goto out; |
1070 | ||
1071 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
1072 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); | |
1073 | curp->p_hdr->eh_entries = cpu_to_le16(1); | |
1074 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
1075 | |
1076 | if (path[0].p_hdr->eh_depth) | |
1077 | curp->p_idx->ei_block = | |
1078 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
1079 | else | |
1080 | curp->p_idx->ei_block = | |
1081 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 1082 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
1083 | |
1084 | neh = ext_inode_hdr(inode); | |
1085 | fidx = EXT_FIRST_INDEX(neh); | |
2ae02107 | 1086 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1087 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
f65e6fba | 1088 | le32_to_cpu(fidx->ei_block), idx_pblock(fidx)); |
a86c6181 AT |
1089 | |
1090 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
1091 | err = ext4_ext_dirty(handle, inode, curp); | |
1092 | out: | |
1093 | brelse(bh); | |
1094 | ||
1095 | return err; | |
1096 | } | |
1097 | ||
1098 | /* | |
d0d856e8 RD |
1099 | * ext4_ext_create_new_leaf: |
1100 | * finds empty index and adds new leaf. | |
1101 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1102 | */ |
1103 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
1104 | struct ext4_ext_path *path, | |
1105 | struct ext4_extent *newext) | |
1106 | { | |
1107 | struct ext4_ext_path *curp; | |
1108 | int depth, i, err = 0; | |
1109 | ||
1110 | repeat: | |
1111 | i = depth = ext_depth(inode); | |
1112 | ||
1113 | /* walk up to the tree and look for free index entry */ | |
1114 | curp = path + depth; | |
1115 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1116 | i--; | |
1117 | curp--; | |
1118 | } | |
1119 | ||
d0d856e8 RD |
1120 | /* we use already allocated block for index block, |
1121 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1122 | if (EXT_HAS_FREE_INDEX(curp)) { |
1123 | /* if we found index with free entry, then use that | |
1124 | * entry: create all needed subtree and add new leaf */ | |
1125 | err = ext4_ext_split(handle, inode, path, newext, i); | |
787e0981 SF |
1126 | if (err) |
1127 | goto out; | |
a86c6181 AT |
1128 | |
1129 | /* refill path */ | |
1130 | ext4_ext_drop_refs(path); | |
1131 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1132 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1133 | path); | |
a86c6181 AT |
1134 | if (IS_ERR(path)) |
1135 | err = PTR_ERR(path); | |
1136 | } else { | |
1137 | /* tree is full, time to grow in depth */ | |
1138 | err = ext4_ext_grow_indepth(handle, inode, path, newext); | |
1139 | if (err) | |
1140 | goto out; | |
1141 | ||
1142 | /* refill path */ | |
1143 | ext4_ext_drop_refs(path); | |
1144 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1145 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1146 | path); | |
a86c6181 AT |
1147 | if (IS_ERR(path)) { |
1148 | err = PTR_ERR(path); | |
1149 | goto out; | |
1150 | } | |
1151 | ||
1152 | /* | |
d0d856e8 RD |
1153 | * only first (depth 0 -> 1) produces free space; |
1154 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1155 | */ |
1156 | depth = ext_depth(inode); | |
1157 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1158 | /* now we need to split */ |
a86c6181 AT |
1159 | goto repeat; |
1160 | } | |
1161 | } | |
1162 | ||
1163 | out: | |
1164 | return err; | |
1165 | } | |
1166 | ||
1988b51e AT |
1167 | /* |
1168 | * search the closest allocated block to the left for *logical | |
1169 | * and returns it at @logical + it's physical address at @phys | |
1170 | * if *logical is the smallest allocated block, the function | |
1171 | * returns 0 at @phys | |
1172 | * return value contains 0 (success) or error code | |
1173 | */ | |
1174 | int | |
1175 | ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path, | |
1176 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1177 | { | |
1178 | struct ext4_extent_idx *ix; | |
1179 | struct ext4_extent *ex; | |
b939e376 | 1180 | int depth, ee_len; |
1988b51e AT |
1181 | |
1182 | BUG_ON(path == NULL); | |
1183 | depth = path->p_depth; | |
1184 | *phys = 0; | |
1185 | ||
1186 | if (depth == 0 && path->p_ext == NULL) | |
1187 | return 0; | |
1188 | ||
1189 | /* usually extent in the path covers blocks smaller | |
1190 | * then *logical, but it can be that extent is the | |
1191 | * first one in the file */ | |
1192 | ||
1193 | ex = path[depth].p_ext; | |
b939e376 | 1194 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1195 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1196 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1197 | while (--depth >= 0) { | |
1198 | ix = path[depth].p_idx; | |
1199 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1200 | } | |
1201 | return 0; | |
1202 | } | |
1203 | ||
b939e376 | 1204 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e | 1205 | |
b939e376 AK |
1206 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
1207 | *phys = ext_pblock(ex) + ee_len - 1; | |
1988b51e AT |
1208 | return 0; |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * search the closest allocated block to the right for *logical | |
1213 | * and returns it at @logical + it's physical address at @phys | |
1214 | * if *logical is the smallest allocated block, the function | |
1215 | * returns 0 at @phys | |
1216 | * return value contains 0 (success) or error code | |
1217 | */ | |
1218 | int | |
1219 | ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path, | |
1220 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1221 | { | |
1222 | struct buffer_head *bh = NULL; | |
1223 | struct ext4_extent_header *eh; | |
1224 | struct ext4_extent_idx *ix; | |
1225 | struct ext4_extent *ex; | |
1226 | ext4_fsblk_t block; | |
395a87bf ES |
1227 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1228 | int ee_len; | |
1988b51e AT |
1229 | |
1230 | BUG_ON(path == NULL); | |
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 AT |
1243 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1244 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1245 | while (--depth >= 0) { | |
1246 | ix = path[depth].p_idx; | |
1247 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1248 | } | |
1249 | *logical = le32_to_cpu(ex->ee_block); | |
1250 | *phys = ext_pblock(ex); | |
1251 | return 0; | |
1252 | } | |
1253 | ||
b939e376 | 1254 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e AT |
1255 | |
1256 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1257 | /* next allocated block in this leaf */ | |
1258 | ex++; | |
1259 | *logical = le32_to_cpu(ex->ee_block); | |
1260 | *phys = ext_pblock(ex); | |
1261 | return 0; | |
1262 | } | |
1263 | ||
1264 | /* go up and search for index to the right */ | |
1265 | while (--depth >= 0) { | |
1266 | ix = path[depth].p_idx; | |
1267 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1268 | goto got_index; |
1988b51e AT |
1269 | } |
1270 | ||
25f1ee3a WF |
1271 | /* we've gone up to the root and found no index to the right */ |
1272 | return 0; | |
1988b51e | 1273 | |
25f1ee3a | 1274 | got_index: |
1988b51e AT |
1275 | /* we've found index to the right, let's |
1276 | * follow it and find the closest allocated | |
1277 | * block to the right */ | |
1278 | ix++; | |
1279 | block = idx_pblock(ix); | |
1280 | while (++depth < path->p_depth) { | |
1281 | bh = sb_bread(inode->i_sb, block); | |
1282 | if (bh == NULL) | |
1283 | return -EIO; | |
1284 | eh = ext_block_hdr(bh); | |
395a87bf | 1285 | /* subtract from p_depth to get proper eh_depth */ |
56b19868 | 1286 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1287 | put_bh(bh); |
1288 | return -EIO; | |
1289 | } | |
1290 | ix = EXT_FIRST_INDEX(eh); | |
1291 | block = idx_pblock(ix); | |
1292 | put_bh(bh); | |
1293 | } | |
1294 | ||
1295 | bh = sb_bread(inode->i_sb, block); | |
1296 | if (bh == NULL) | |
1297 | return -EIO; | |
1298 | eh = ext_block_hdr(bh); | |
56b19868 | 1299 | if (ext4_ext_check(inode, eh, path->p_depth - depth)) { |
1988b51e AT |
1300 | put_bh(bh); |
1301 | return -EIO; | |
1302 | } | |
1303 | ex = EXT_FIRST_EXTENT(eh); | |
1304 | *logical = le32_to_cpu(ex->ee_block); | |
1305 | *phys = ext_pblock(ex); | |
1306 | put_bh(bh); | |
1307 | return 0; | |
1988b51e AT |
1308 | } |
1309 | ||
a86c6181 | 1310 | /* |
d0d856e8 RD |
1311 | * ext4_ext_next_allocated_block: |
1312 | * returns allocated block in subsequent extent or EXT_MAX_BLOCK. | |
1313 | * NOTE: it considers block number from index entry as | |
1314 | * allocated block. Thus, index entries have to be consistent | |
1315 | * with leaves. | |
a86c6181 | 1316 | */ |
725d26d3 | 1317 | static ext4_lblk_t |
a86c6181 AT |
1318 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1319 | { | |
1320 | int depth; | |
1321 | ||
1322 | BUG_ON(path == NULL); | |
1323 | depth = path->p_depth; | |
1324 | ||
1325 | if (depth == 0 && path->p_ext == NULL) | |
1326 | return EXT_MAX_BLOCK; | |
1327 | ||
1328 | while (depth >= 0) { | |
1329 | if (depth == path->p_depth) { | |
1330 | /* leaf */ | |
1331 | if (path[depth].p_ext != | |
1332 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1333 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1334 | } else { | |
1335 | /* index */ | |
1336 | if (path[depth].p_idx != | |
1337 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1338 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1339 | } | |
1340 | depth--; | |
1341 | } | |
1342 | ||
1343 | return EXT_MAX_BLOCK; | |
1344 | } | |
1345 | ||
1346 | /* | |
d0d856e8 | 1347 | * ext4_ext_next_leaf_block: |
a86c6181 AT |
1348 | * returns first allocated block from next leaf or EXT_MAX_BLOCK |
1349 | */ | |
725d26d3 | 1350 | static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode, |
63f57933 | 1351 | struct ext4_ext_path *path) |
a86c6181 AT |
1352 | { |
1353 | int depth; | |
1354 | ||
1355 | BUG_ON(path == NULL); | |
1356 | depth = path->p_depth; | |
1357 | ||
1358 | /* zero-tree has no leaf blocks at all */ | |
1359 | if (depth == 0) | |
1360 | return EXT_MAX_BLOCK; | |
1361 | ||
1362 | /* go to index block */ | |
1363 | depth--; | |
1364 | ||
1365 | while (depth >= 0) { | |
1366 | if (path[depth].p_idx != | |
1367 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1368 | return (ext4_lblk_t) |
1369 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1370 | depth--; |
1371 | } | |
1372 | ||
1373 | return EXT_MAX_BLOCK; | |
1374 | } | |
1375 | ||
1376 | /* | |
d0d856e8 RD |
1377 | * ext4_ext_correct_indexes: |
1378 | * if leaf gets modified and modified extent is first in the leaf, | |
1379 | * then we have to correct all indexes above. | |
a86c6181 AT |
1380 | * TODO: do we need to correct tree in all cases? |
1381 | */ | |
1d03ec98 | 1382 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1383 | struct ext4_ext_path *path) |
1384 | { | |
1385 | struct ext4_extent_header *eh; | |
1386 | int depth = ext_depth(inode); | |
1387 | struct ext4_extent *ex; | |
1388 | __le32 border; | |
1389 | int k, err = 0; | |
1390 | ||
1391 | eh = path[depth].p_hdr; | |
1392 | ex = path[depth].p_ext; | |
1393 | BUG_ON(ex == NULL); | |
1394 | BUG_ON(eh == NULL); | |
1395 | ||
1396 | if (depth == 0) { | |
1397 | /* there is no tree at all */ | |
1398 | return 0; | |
1399 | } | |
1400 | ||
1401 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1402 | /* we correct tree if first leaf got modified only */ | |
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | /* | |
d0d856e8 | 1407 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1408 | */ |
1409 | k = depth - 1; | |
1410 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1411 | err = ext4_ext_get_access(handle, inode, path + k); |
1412 | if (err) | |
a86c6181 AT |
1413 | return err; |
1414 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1415 | err = ext4_ext_dirty(handle, inode, path + k); |
1416 | if (err) | |
a86c6181 AT |
1417 | return err; |
1418 | ||
1419 | while (k--) { | |
1420 | /* change all left-side indexes */ | |
1421 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1422 | break; | |
7e028976 AM |
1423 | err = ext4_ext_get_access(handle, inode, path + k); |
1424 | if (err) | |
a86c6181 AT |
1425 | break; |
1426 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1427 | err = ext4_ext_dirty(handle, inode, path + k); |
1428 | if (err) | |
a86c6181 AT |
1429 | break; |
1430 | } | |
1431 | ||
1432 | return err; | |
1433 | } | |
1434 | ||
748de673 | 1435 | int |
a86c6181 AT |
1436 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1437 | struct ext4_extent *ex2) | |
1438 | { | |
749269fa | 1439 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1440 | |
1441 | /* | |
1442 | * Make sure that either both extents are uninitialized, or | |
1443 | * both are _not_. | |
1444 | */ | |
1445 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1446 | return 0; | |
1447 | ||
749269fa AA |
1448 | if (ext4_ext_is_uninitialized(ex1)) |
1449 | max_len = EXT_UNINIT_MAX_LEN; | |
1450 | else | |
1451 | max_len = EXT_INIT_MAX_LEN; | |
1452 | ||
a2df2a63 AA |
1453 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1454 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1455 | ||
1456 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1457 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1458 | return 0; |
1459 | ||
471d4011 SB |
1460 | /* |
1461 | * To allow future support for preallocated extents to be added | |
1462 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1463 | * this can result in the top bit of ee_len being set. |
471d4011 | 1464 | */ |
749269fa | 1465 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1466 | return 0; |
bbf2f9fb | 1467 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1468 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1469 | return 0; |
1470 | #endif | |
1471 | ||
a2df2a63 | 1472 | if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) |
a86c6181 AT |
1473 | return 1; |
1474 | return 0; | |
1475 | } | |
1476 | ||
56055d3a AA |
1477 | /* |
1478 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1479 | * It always tries to merge towards right. If you want to merge towards | |
1480 | * left, pass "ex - 1" as argument instead of "ex". | |
1481 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1482 | * 1 if they got merged. | |
1483 | */ | |
1484 | int ext4_ext_try_to_merge(struct inode *inode, | |
1485 | struct ext4_ext_path *path, | |
1486 | struct ext4_extent *ex) | |
1487 | { | |
1488 | struct ext4_extent_header *eh; | |
1489 | unsigned int depth, len; | |
1490 | int merge_done = 0; | |
1491 | int uninitialized = 0; | |
1492 | ||
1493 | depth = ext_depth(inode); | |
1494 | BUG_ON(path[depth].p_hdr == NULL); | |
1495 | eh = path[depth].p_hdr; | |
1496 | ||
1497 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1498 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1499 | break; | |
1500 | /* merge with next extent! */ | |
1501 | if (ext4_ext_is_uninitialized(ex)) | |
1502 | uninitialized = 1; | |
1503 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1504 | + ext4_ext_get_actual_len(ex + 1)); | |
1505 | if (uninitialized) | |
1506 | ext4_ext_mark_uninitialized(ex); | |
1507 | ||
1508 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1509 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1510 | * sizeof(struct ext4_extent); | |
1511 | memmove(ex + 1, ex + 2, len); | |
1512 | } | |
e8546d06 | 1513 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1514 | merge_done = 1; |
1515 | WARN_ON(eh->eh_entries == 0); | |
1516 | if (!eh->eh_entries) | |
1517 | ext4_error(inode->i_sb, "ext4_ext_try_to_merge", | |
1518 | "inode#%lu, eh->eh_entries = 0!", inode->i_ino); | |
1519 | } | |
1520 | ||
1521 | return merge_done; | |
1522 | } | |
1523 | ||
25d14f98 AA |
1524 | /* |
1525 | * check if a portion of the "newext" extent overlaps with an | |
1526 | * existing extent. | |
1527 | * | |
1528 | * If there is an overlap discovered, it updates the length of the newext | |
1529 | * such that there will be no overlap, and then returns 1. | |
1530 | * If there is no overlap found, it returns 0. | |
1531 | */ | |
1532 | unsigned int ext4_ext_check_overlap(struct inode *inode, | |
1533 | struct ext4_extent *newext, | |
1534 | struct ext4_ext_path *path) | |
1535 | { | |
725d26d3 | 1536 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1537 | unsigned int depth, len1; |
1538 | unsigned int ret = 0; | |
1539 | ||
1540 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1541 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1542 | depth = ext_depth(inode); |
1543 | if (!path[depth].p_ext) | |
1544 | goto out; | |
1545 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
1546 | ||
1547 | /* | |
1548 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1549 | * is before the requested block(s) |
25d14f98 AA |
1550 | */ |
1551 | if (b2 < b1) { | |
1552 | b2 = ext4_ext_next_allocated_block(path); | |
1553 | if (b2 == EXT_MAX_BLOCK) | |
1554 | goto out; | |
1555 | } | |
1556 | ||
725d26d3 | 1557 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 AA |
1558 | if (b1 + len1 < b1) { |
1559 | len1 = EXT_MAX_BLOCK - b1; | |
1560 | newext->ee_len = cpu_to_le16(len1); | |
1561 | ret = 1; | |
1562 | } | |
1563 | ||
1564 | /* check for overlap */ | |
1565 | if (b1 + len1 > b2) { | |
1566 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1567 | ret = 1; | |
1568 | } | |
1569 | out: | |
1570 | return ret; | |
1571 | } | |
1572 | ||
a86c6181 | 1573 | /* |
d0d856e8 RD |
1574 | * ext4_ext_insert_extent: |
1575 | * tries to merge requsted extent into the existing extent or | |
1576 | * inserts requested extent as new one into the tree, | |
1577 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1578 | */ |
1579 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1580 | struct ext4_ext_path *path, | |
1581 | struct ext4_extent *newext) | |
1582 | { | |
af5bc92d | 1583 | struct ext4_extent_header *eh; |
a86c6181 AT |
1584 | struct ext4_extent *ex, *fex; |
1585 | struct ext4_extent *nearex; /* nearest extent */ | |
1586 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1587 | int depth, len, err; |
1588 | ext4_lblk_t next; | |
a2df2a63 | 1589 | unsigned uninitialized = 0; |
a86c6181 | 1590 | |
a2df2a63 | 1591 | BUG_ON(ext4_ext_get_actual_len(newext) == 0); |
a86c6181 AT |
1592 | depth = ext_depth(inode); |
1593 | ex = path[depth].p_ext; | |
1594 | BUG_ON(path[depth].p_hdr == NULL); | |
1595 | ||
1596 | /* try to insert block into found extent and return */ | |
1597 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | |
553f9008 M |
1598 | ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", |
1599 | ext4_ext_is_uninitialized(newext), | |
a2df2a63 | 1600 | ext4_ext_get_actual_len(newext), |
a86c6181 | 1601 | le32_to_cpu(ex->ee_block), |
553f9008 | 1602 | ext4_ext_is_uninitialized(ex), |
a2df2a63 | 1603 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
7e028976 AM |
1604 | err = ext4_ext_get_access(handle, inode, path + depth); |
1605 | if (err) | |
a86c6181 | 1606 | return err; |
a2df2a63 AA |
1607 | |
1608 | /* | |
1609 | * ext4_can_extents_be_merged should have checked that either | |
1610 | * both extents are uninitialized, or both aren't. Thus we | |
1611 | * need to check only one of them here. | |
1612 | */ | |
1613 | if (ext4_ext_is_uninitialized(ex)) | |
1614 | uninitialized = 1; | |
1615 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1616 | + ext4_ext_get_actual_len(newext)); | |
1617 | if (uninitialized) | |
1618 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1619 | eh = path[depth].p_hdr; |
1620 | nearex = ex; | |
1621 | goto merge; | |
1622 | } | |
1623 | ||
1624 | repeat: | |
1625 | depth = ext_depth(inode); | |
1626 | eh = path[depth].p_hdr; | |
1627 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1628 | goto has_space; | |
1629 | ||
1630 | /* probably next leaf has space for us? */ | |
1631 | fex = EXT_LAST_EXTENT(eh); | |
1632 | next = ext4_ext_next_leaf_block(inode, path); | |
1633 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) | |
1634 | && next != EXT_MAX_BLOCK) { | |
1635 | ext_debug("next leaf block - %d\n", next); | |
1636 | BUG_ON(npath != NULL); | |
1637 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1638 | if (IS_ERR(npath)) | |
1639 | return PTR_ERR(npath); | |
1640 | BUG_ON(npath->p_depth != path->p_depth); | |
1641 | eh = npath[depth].p_hdr; | |
1642 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
1643 | ext_debug("next leaf isnt full(%d)\n", | |
1644 | le16_to_cpu(eh->eh_entries)); | |
1645 | path = npath; | |
1646 | goto repeat; | |
1647 | } | |
1648 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1649 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1650 | } | |
1651 | ||
1652 | /* | |
d0d856e8 RD |
1653 | * There is no free space in the found leaf. |
1654 | * We're gonna add a new leaf in the tree. | |
a86c6181 AT |
1655 | */ |
1656 | err = ext4_ext_create_new_leaf(handle, inode, path, newext); | |
1657 | if (err) | |
1658 | goto cleanup; | |
1659 | depth = ext_depth(inode); | |
1660 | eh = path[depth].p_hdr; | |
1661 | ||
1662 | has_space: | |
1663 | nearex = path[depth].p_ext; | |
1664 | ||
7e028976 AM |
1665 | err = ext4_ext_get_access(handle, inode, path + depth); |
1666 | if (err) | |
a86c6181 AT |
1667 | goto cleanup; |
1668 | ||
1669 | if (!nearex) { | |
1670 | /* there is no extent in this leaf, create first one */ | |
553f9008 | 1671 | ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n", |
8c55e204 DK |
1672 | le32_to_cpu(newext->ee_block), |
1673 | ext_pblock(newext), | |
553f9008 | 1674 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1675 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1676 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1677 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1678 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1679 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1680 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1681 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1682 | len = (len - 1) * sizeof(struct ext4_extent); | |
1683 | len = len < 0 ? 0 : len; | |
553f9008 | 1684 | ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " |
a86c6181 | 1685 | "move %d from 0x%p to 0x%p\n", |
8c55e204 DK |
1686 | le32_to_cpu(newext->ee_block), |
1687 | ext_pblock(newext), | |
553f9008 | 1688 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1689 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1690 | nearex, len, nearex + 1, nearex + 2); |
1691 | memmove(nearex + 2, nearex + 1, len); | |
1692 | } | |
1693 | path[depth].p_ext = nearex + 1; | |
1694 | } else { | |
1695 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1696 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1697 | len = len < 0 ? 0 : len; | |
553f9008 | 1698 | ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " |
a86c6181 AT |
1699 | "move %d from 0x%p to 0x%p\n", |
1700 | le32_to_cpu(newext->ee_block), | |
f65e6fba | 1701 | ext_pblock(newext), |
553f9008 | 1702 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1703 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1704 | nearex, len, nearex + 1, nearex + 2); |
1705 | memmove(nearex + 1, nearex, len); | |
1706 | path[depth].p_ext = nearex; | |
1707 | } | |
1708 | ||
e8546d06 | 1709 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1710 | nearex = path[depth].p_ext; |
1711 | nearex->ee_block = newext->ee_block; | |
b377611d | 1712 | ext4_ext_store_pblock(nearex, ext_pblock(newext)); |
a86c6181 | 1713 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1714 | |
1715 | merge: | |
1716 | /* try to merge extents to the right */ | |
56055d3a | 1717 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1718 | |
1719 | /* try to merge extents to the left */ | |
1720 | ||
1721 | /* time to correct all indexes above */ | |
1722 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1723 | if (err) | |
1724 | goto cleanup; | |
1725 | ||
1726 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1727 | ||
1728 | cleanup: | |
1729 | if (npath) { | |
1730 | ext4_ext_drop_refs(npath); | |
1731 | kfree(npath); | |
1732 | } | |
a86c6181 AT |
1733 | ext4_ext_invalidate_cache(inode); |
1734 | return err; | |
1735 | } | |
1736 | ||
6873fa0d ES |
1737 | int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1738 | ext4_lblk_t num, ext_prepare_callback func, | |
1739 | void *cbdata) | |
1740 | { | |
1741 | struct ext4_ext_path *path = NULL; | |
1742 | struct ext4_ext_cache cbex; | |
1743 | struct ext4_extent *ex; | |
1744 | ext4_lblk_t next, start = 0, end = 0; | |
1745 | ext4_lblk_t last = block + num; | |
1746 | int depth, exists, err = 0; | |
1747 | ||
1748 | BUG_ON(func == NULL); | |
1749 | BUG_ON(inode == NULL); | |
1750 | ||
1751 | while (block < last && block != EXT_MAX_BLOCK) { | |
1752 | num = last - block; | |
1753 | /* find extent for this block */ | |
1754 | path = ext4_ext_find_extent(inode, block, path); | |
1755 | if (IS_ERR(path)) { | |
1756 | err = PTR_ERR(path); | |
1757 | path = NULL; | |
1758 | break; | |
1759 | } | |
1760 | ||
1761 | depth = ext_depth(inode); | |
1762 | BUG_ON(path[depth].p_hdr == NULL); | |
1763 | ex = path[depth].p_ext; | |
1764 | next = ext4_ext_next_allocated_block(path); | |
1765 | ||
1766 | exists = 0; | |
1767 | if (!ex) { | |
1768 | /* there is no extent yet, so try to allocate | |
1769 | * all requested space */ | |
1770 | start = block; | |
1771 | end = block + num; | |
1772 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1773 | /* need to allocate space before found extent */ | |
1774 | start = block; | |
1775 | end = le32_to_cpu(ex->ee_block); | |
1776 | if (block + num < end) | |
1777 | end = block + num; | |
1778 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1779 | + ext4_ext_get_actual_len(ex)) { | |
1780 | /* need to allocate space after found extent */ | |
1781 | start = block; | |
1782 | end = block + num; | |
1783 | if (end >= next) | |
1784 | end = next; | |
1785 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1786 | /* | |
1787 | * some part of requested space is covered | |
1788 | * by found extent | |
1789 | */ | |
1790 | start = block; | |
1791 | end = le32_to_cpu(ex->ee_block) | |
1792 | + ext4_ext_get_actual_len(ex); | |
1793 | if (block + num < end) | |
1794 | end = block + num; | |
1795 | exists = 1; | |
1796 | } else { | |
1797 | BUG(); | |
1798 | } | |
1799 | BUG_ON(end <= start); | |
1800 | ||
1801 | if (!exists) { | |
1802 | cbex.ec_block = start; | |
1803 | cbex.ec_len = end - start; | |
1804 | cbex.ec_start = 0; | |
1805 | cbex.ec_type = EXT4_EXT_CACHE_GAP; | |
1806 | } else { | |
1807 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1808 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
1809 | cbex.ec_start = ext_pblock(ex); | |
1810 | cbex.ec_type = EXT4_EXT_CACHE_EXTENT; | |
1811 | } | |
1812 | ||
1813 | BUG_ON(cbex.ec_len == 0); | |
1814 | err = func(inode, path, &cbex, ex, cbdata); | |
1815 | ext4_ext_drop_refs(path); | |
1816 | ||
1817 | if (err < 0) | |
1818 | break; | |
1819 | ||
1820 | if (err == EXT_REPEAT) | |
1821 | continue; | |
1822 | else if (err == EXT_BREAK) { | |
1823 | err = 0; | |
1824 | break; | |
1825 | } | |
1826 | ||
1827 | if (ext_depth(inode) != depth) { | |
1828 | /* depth was changed. we have to realloc path */ | |
1829 | kfree(path); | |
1830 | path = NULL; | |
1831 | } | |
1832 | ||
1833 | block = cbex.ec_block + cbex.ec_len; | |
1834 | } | |
1835 | ||
1836 | if (path) { | |
1837 | ext4_ext_drop_refs(path); | |
1838 | kfree(path); | |
1839 | } | |
1840 | ||
1841 | return err; | |
1842 | } | |
1843 | ||
09b88252 | 1844 | static void |
725d26d3 | 1845 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
dd54567a | 1846 | __u32 len, ext4_fsblk_t start, int type) |
a86c6181 AT |
1847 | { |
1848 | struct ext4_ext_cache *cex; | |
1849 | BUG_ON(len == 0); | |
2ec0ae3a | 1850 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
1851 | cex = &EXT4_I(inode)->i_cached_extent; |
1852 | cex->ec_type = type; | |
1853 | cex->ec_block = block; | |
1854 | cex->ec_len = len; | |
1855 | cex->ec_start = start; | |
2ec0ae3a | 1856 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
1857 | } |
1858 | ||
1859 | /* | |
d0d856e8 RD |
1860 | * ext4_ext_put_gap_in_cache: |
1861 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
1862 | * and cache this gap |
1863 | */ | |
09b88252 | 1864 | static void |
a86c6181 | 1865 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 1866 | ext4_lblk_t block) |
a86c6181 AT |
1867 | { |
1868 | int depth = ext_depth(inode); | |
725d26d3 AK |
1869 | unsigned long len; |
1870 | ext4_lblk_t lblock; | |
a86c6181 AT |
1871 | struct ext4_extent *ex; |
1872 | ||
1873 | ex = path[depth].p_ext; | |
1874 | if (ex == NULL) { | |
1875 | /* there is no extent yet, so gap is [0;-] */ | |
1876 | lblock = 0; | |
1877 | len = EXT_MAX_BLOCK; | |
1878 | ext_debug("cache gap(whole file):"); | |
1879 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
1880 | lblock = block; | |
1881 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
1882 | ext_debug("cache gap(before): %u [%u:%u]", |
1883 | block, | |
1884 | le32_to_cpu(ex->ee_block), | |
1885 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 1886 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 1887 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 1888 | ext4_lblk_t next; |
8c55e204 | 1889 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 1890 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
1891 | |
1892 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
1893 | ext_debug("cache gap(after): [%u:%u] %u", |
1894 | le32_to_cpu(ex->ee_block), | |
1895 | ext4_ext_get_actual_len(ex), | |
1896 | block); | |
725d26d3 AK |
1897 | BUG_ON(next == lblock); |
1898 | len = next - lblock; | |
a86c6181 AT |
1899 | } else { |
1900 | lblock = len = 0; | |
1901 | BUG(); | |
1902 | } | |
1903 | ||
bba90743 | 1904 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
1905 | ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP); |
1906 | } | |
1907 | ||
09b88252 | 1908 | static int |
725d26d3 | 1909 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, |
a86c6181 AT |
1910 | struct ext4_extent *ex) |
1911 | { | |
1912 | struct ext4_ext_cache *cex; | |
2ec0ae3a | 1913 | int ret = EXT4_EXT_CACHE_NO; |
a86c6181 | 1914 | |
2ec0ae3a TT |
1915 | /* |
1916 | * We borrow i_block_reservation_lock to protect i_cached_extent | |
1917 | */ | |
1918 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
a86c6181 AT |
1919 | cex = &EXT4_I(inode)->i_cached_extent; |
1920 | ||
1921 | /* has cache valid data? */ | |
1922 | if (cex->ec_type == EXT4_EXT_CACHE_NO) | |
2ec0ae3a | 1923 | goto errout; |
a86c6181 AT |
1924 | |
1925 | BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP && | |
1926 | cex->ec_type != EXT4_EXT_CACHE_EXTENT); | |
1927 | if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) { | |
8c55e204 | 1928 | ex->ee_block = cpu_to_le32(cex->ec_block); |
f65e6fba | 1929 | ext4_ext_store_pblock(ex, cex->ec_start); |
8c55e204 | 1930 | ex->ee_len = cpu_to_le16(cex->ec_len); |
bba90743 ES |
1931 | ext_debug("%u cached by %u:%u:%llu\n", |
1932 | block, | |
1933 | cex->ec_block, cex->ec_len, cex->ec_start); | |
2ec0ae3a | 1934 | ret = cex->ec_type; |
a86c6181 | 1935 | } |
2ec0ae3a TT |
1936 | errout: |
1937 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1938 | return ret; | |
a86c6181 AT |
1939 | } |
1940 | ||
1941 | /* | |
d0d856e8 RD |
1942 | * ext4_ext_rm_idx: |
1943 | * removes index from the index block. | |
1944 | * It's used in truncate case only, thus all requests are for | |
1945 | * last index in the block only. | |
a86c6181 | 1946 | */ |
1d03ec98 | 1947 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1948 | struct ext4_ext_path *path) |
1949 | { | |
1950 | struct buffer_head *bh; | |
1951 | int err; | |
f65e6fba | 1952 | ext4_fsblk_t leaf; |
a86c6181 AT |
1953 | |
1954 | /* free index block */ | |
1955 | path--; | |
f65e6fba | 1956 | leaf = idx_pblock(path->p_idx); |
a86c6181 | 1957 | BUG_ON(path->p_hdr->eh_entries == 0); |
7e028976 AM |
1958 | err = ext4_ext_get_access(handle, inode, path); |
1959 | if (err) | |
a86c6181 | 1960 | return err; |
e8546d06 | 1961 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
1962 | err = ext4_ext_dirty(handle, inode, path); |
1963 | if (err) | |
a86c6181 | 1964 | return err; |
2ae02107 | 1965 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
a86c6181 AT |
1966 | bh = sb_find_get_block(inode->i_sb, leaf); |
1967 | ext4_forget(handle, 1, inode, bh, leaf); | |
c9de560d | 1968 | ext4_free_blocks(handle, inode, leaf, 1, 1); |
a86c6181 AT |
1969 | return err; |
1970 | } | |
1971 | ||
1972 | /* | |
ee12b630 MC |
1973 | * ext4_ext_calc_credits_for_single_extent: |
1974 | * This routine returns max. credits that needed to insert an extent | |
1975 | * to the extent tree. | |
1976 | * When pass the actual path, the caller should calculate credits | |
1977 | * under i_data_sem. | |
a86c6181 | 1978 | */ |
525f4ed8 | 1979 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
1980 | struct ext4_ext_path *path) |
1981 | { | |
a86c6181 | 1982 | if (path) { |
ee12b630 | 1983 | int depth = ext_depth(inode); |
f3bd1f3f | 1984 | int ret = 0; |
ee12b630 | 1985 | |
a86c6181 | 1986 | /* probably there is space in leaf? */ |
a86c6181 | 1987 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 1988 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 1989 | |
ee12b630 MC |
1990 | /* |
1991 | * There are some space in the leaf tree, no | |
1992 | * need to account for leaf block credit | |
1993 | * | |
1994 | * bitmaps and block group descriptor blocks | |
1995 | * and other metadat blocks still need to be | |
1996 | * accounted. | |
1997 | */ | |
525f4ed8 | 1998 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 1999 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2000 | return ret; |
ee12b630 MC |
2001 | } |
2002 | } | |
a86c6181 | 2003 | |
525f4ed8 | 2004 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2005 | } |
a86c6181 | 2006 | |
ee12b630 MC |
2007 | /* |
2008 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2009 | * | |
2010 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2011 | * in the worse case, each tree level index/leaf need to be changed | |
2012 | * if the tree split due to insert a new extent, then the old tree | |
2013 | * index/leaf need to be updated too | |
2014 | * | |
2015 | * If the nrblocks are discontiguous, they could cause | |
2016 | * the whole tree split more than once, but this is really rare. | |
2017 | */ | |
525f4ed8 | 2018 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2019 | { |
2020 | int index; | |
2021 | int depth = ext_depth(inode); | |
a86c6181 | 2022 | |
ee12b630 MC |
2023 | if (chunk) |
2024 | index = depth * 2; | |
2025 | else | |
2026 | index = depth * 3; | |
a86c6181 | 2027 | |
ee12b630 | 2028 | return index; |
a86c6181 AT |
2029 | } |
2030 | ||
2031 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
2032 | struct ext4_extent *ex, | |
725d26d3 | 2033 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 AT |
2034 | { |
2035 | struct buffer_head *bh; | |
a2df2a63 | 2036 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
c9de560d | 2037 | int i, metadata = 0; |
a86c6181 | 2038 | |
c9de560d AT |
2039 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
2040 | metadata = 1; | |
a86c6181 AT |
2041 | #ifdef EXTENTS_STATS |
2042 | { | |
2043 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2044 | spin_lock(&sbi->s_ext_stats_lock); |
2045 | sbi->s_ext_blocks += ee_len; | |
2046 | sbi->s_ext_extents++; | |
2047 | if (ee_len < sbi->s_ext_min) | |
2048 | sbi->s_ext_min = ee_len; | |
2049 | if (ee_len > sbi->s_ext_max) | |
2050 | sbi->s_ext_max = ee_len; | |
2051 | if (ext_depth(inode) > sbi->s_depth_max) | |
2052 | sbi->s_depth_max = ext_depth(inode); | |
2053 | spin_unlock(&sbi->s_ext_stats_lock); | |
2054 | } | |
2055 | #endif | |
2056 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2057 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2058 | /* tail removal */ |
725d26d3 | 2059 | ext4_lblk_t num; |
f65e6fba | 2060 | ext4_fsblk_t start; |
725d26d3 | 2061 | |
a2df2a63 AA |
2062 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
2063 | start = ext_pblock(ex) + ee_len - num; | |
725d26d3 | 2064 | ext_debug("free last %u blocks starting %llu\n", num, start); |
a86c6181 AT |
2065 | for (i = 0; i < num; i++) { |
2066 | bh = sb_find_get_block(inode->i_sb, start + i); | |
2067 | ext4_forget(handle, 0, inode, bh, start + i); | |
2068 | } | |
c9de560d | 2069 | ext4_free_blocks(handle, inode, start, num, metadata); |
a86c6181 | 2070 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2071 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
725d26d3 | 2072 | printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n", |
a2df2a63 | 2073 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
a86c6181 | 2074 | } else { |
725d26d3 AK |
2075 | printk(KERN_INFO "strange request: removal(2) " |
2076 | "%u-%u from %u:%u\n", | |
2077 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2078 | } |
2079 | return 0; | |
2080 | } | |
2081 | ||
2082 | static int | |
2083 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
725d26d3 | 2084 | struct ext4_ext_path *path, ext4_lblk_t start) |
a86c6181 AT |
2085 | { |
2086 | int err = 0, correct_index = 0; | |
2087 | int depth = ext_depth(inode), credits; | |
2088 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2089 | ext4_lblk_t a, b, block; |
2090 | unsigned num; | |
2091 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2092 | unsigned short ex_ee_len; |
a2df2a63 | 2093 | unsigned uninitialized = 0; |
a86c6181 AT |
2094 | struct ext4_extent *ex; |
2095 | ||
c29c0ae7 | 2096 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 2097 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
2098 | if (!path[depth].p_hdr) |
2099 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2100 | eh = path[depth].p_hdr; | |
2101 | BUG_ON(eh == NULL); | |
a86c6181 AT |
2102 | |
2103 | /* find where to start removing */ | |
2104 | ex = EXT_LAST_EXTENT(eh); | |
2105 | ||
2106 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2107 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2108 | |
2109 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
2110 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2111 | |
2112 | if (ext4_ext_is_uninitialized(ex)) | |
2113 | uninitialized = 1; | |
2114 | else | |
2115 | uninitialized = 0; | |
2116 | ||
553f9008 M |
2117 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2118 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2119 | path[depth].p_ext = ex; |
2120 | ||
2121 | a = ex_ee_block > start ? ex_ee_block : start; | |
2122 | b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? | |
2123 | ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; | |
2124 | ||
2125 | ext_debug(" border %u:%u\n", a, b); | |
2126 | ||
2127 | if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { | |
2128 | block = 0; | |
2129 | num = 0; | |
2130 | BUG(); | |
2131 | } else if (a != ex_ee_block) { | |
2132 | /* remove tail of the extent */ | |
2133 | block = ex_ee_block; | |
2134 | num = a - block; | |
2135 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
2136 | /* remove head of the extent */ | |
2137 | block = a; | |
2138 | num = b - a; | |
2139 | /* there is no "make a hole" API yet */ | |
2140 | BUG(); | |
2141 | } else { | |
2142 | /* remove whole extent: excellent! */ | |
2143 | block = ex_ee_block; | |
2144 | num = 0; | |
2145 | BUG_ON(a != ex_ee_block); | |
2146 | BUG_ON(b != ex_ee_block + ex_ee_len - 1); | |
2147 | } | |
2148 | ||
34071da7 TT |
2149 | /* |
2150 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2151 | * descriptor) for each block group; assume two block | |
2152 | * groups plus ex_ee_len/blocks_per_block_group for | |
2153 | * the worst case | |
2154 | */ | |
2155 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2156 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2157 | correct_index = 1; | |
2158 | credits += (ext_depth(inode)) + 1; | |
2159 | } | |
a86c6181 | 2160 | credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2161 | |
487caeef | 2162 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2163 | if (err) |
a86c6181 | 2164 | goto out; |
a86c6181 AT |
2165 | |
2166 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2167 | if (err) | |
2168 | goto out; | |
2169 | ||
2170 | err = ext4_remove_blocks(handle, inode, ex, a, b); | |
2171 | if (err) | |
2172 | goto out; | |
2173 | ||
2174 | if (num == 0) { | |
d0d856e8 | 2175 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2176 | ext4_ext_store_pblock(ex, 0); |
e8546d06 | 2177 | le16_add_cpu(&eh->eh_entries, -1); |
a86c6181 AT |
2178 | } |
2179 | ||
2180 | ex->ee_block = cpu_to_le32(block); | |
2181 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
2182 | /* |
2183 | * Do not mark uninitialized if all the blocks in the | |
2184 | * extent have been removed. | |
2185 | */ | |
2186 | if (uninitialized && num) | |
a2df2a63 | 2187 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
2188 | |
2189 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2190 | if (err) | |
2191 | goto out; | |
2192 | ||
2ae02107 | 2193 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
f65e6fba | 2194 | ext_pblock(ex)); |
a86c6181 AT |
2195 | ex--; |
2196 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2197 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2198 | } |
2199 | ||
2200 | if (correct_index && eh->eh_entries) | |
2201 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2202 | ||
2203 | /* if this leaf is free, then we should | |
2204 | * remove it from index block above */ | |
2205 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2206 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2207 | ||
2208 | out: | |
2209 | return err; | |
2210 | } | |
2211 | ||
2212 | /* | |
d0d856e8 RD |
2213 | * ext4_ext_more_to_rm: |
2214 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2215 | */ |
09b88252 | 2216 | static int |
a86c6181 AT |
2217 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2218 | { | |
2219 | BUG_ON(path->p_idx == NULL); | |
2220 | ||
2221 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2222 | return 0; | |
2223 | ||
2224 | /* | |
d0d856e8 | 2225 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2226 | * so we have to consider current index for truncation |
2227 | */ | |
2228 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2229 | return 0; | |
2230 | return 1; | |
2231 | } | |
2232 | ||
1d03ec98 | 2233 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
2234 | { |
2235 | struct super_block *sb = inode->i_sb; | |
2236 | int depth = ext_depth(inode); | |
2237 | struct ext4_ext_path *path; | |
2238 | handle_t *handle; | |
2239 | int i = 0, err = 0; | |
2240 | ||
725d26d3 | 2241 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
2242 | |
2243 | /* probably first extent we're gonna free will be last in block */ | |
2244 | handle = ext4_journal_start(inode, depth + 1); | |
2245 | if (IS_ERR(handle)) | |
2246 | return PTR_ERR(handle); | |
2247 | ||
2248 | ext4_ext_invalidate_cache(inode); | |
2249 | ||
2250 | /* | |
d0d856e8 RD |
2251 | * We start scanning from right side, freeing all the blocks |
2252 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2253 | */ |
216553c4 | 2254 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
2255 | if (path == NULL) { |
2256 | ext4_journal_stop(handle); | |
2257 | return -ENOMEM; | |
2258 | } | |
a86c6181 | 2259 | path[0].p_hdr = ext_inode_hdr(inode); |
56b19868 | 2260 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
2261 | err = -EIO; |
2262 | goto out; | |
2263 | } | |
2264 | path[0].p_depth = depth; | |
2265 | ||
2266 | while (i >= 0 && err == 0) { | |
2267 | if (i == depth) { | |
2268 | /* this is leaf block */ | |
2269 | err = ext4_ext_rm_leaf(handle, inode, path, start); | |
d0d856e8 | 2270 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2271 | brelse(path[i].p_bh); |
2272 | path[i].p_bh = NULL; | |
2273 | i--; | |
2274 | continue; | |
2275 | } | |
2276 | ||
2277 | /* this is index block */ | |
2278 | if (!path[i].p_hdr) { | |
2279 | ext_debug("initialize header\n"); | |
2280 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2281 | } |
2282 | ||
a86c6181 | 2283 | if (!path[i].p_idx) { |
d0d856e8 | 2284 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2285 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2286 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2287 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2288 | path[i].p_hdr, | |
2289 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2290 | } else { | |
d0d856e8 | 2291 | /* we were already here, see at next index */ |
a86c6181 AT |
2292 | path[i].p_idx--; |
2293 | } | |
2294 | ||
2295 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2296 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2297 | path[i].p_idx); | |
2298 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2299 | struct buffer_head *bh; |
a86c6181 | 2300 | /* go to the next level */ |
2ae02107 | 2301 | ext_debug("move to level %d (block %llu)\n", |
f65e6fba | 2302 | i + 1, idx_pblock(path[i].p_idx)); |
a86c6181 | 2303 | memset(path + i + 1, 0, sizeof(*path)); |
c29c0ae7 AT |
2304 | bh = sb_bread(sb, idx_pblock(path[i].p_idx)); |
2305 | if (!bh) { | |
a86c6181 AT |
2306 | /* should we reset i_size? */ |
2307 | err = -EIO; | |
2308 | break; | |
2309 | } | |
c29c0ae7 AT |
2310 | if (WARN_ON(i + 1 > depth)) { |
2311 | err = -EIO; | |
2312 | break; | |
2313 | } | |
56b19868 | 2314 | if (ext4_ext_check(inode, ext_block_hdr(bh), |
c29c0ae7 AT |
2315 | depth - i - 1)) { |
2316 | err = -EIO; | |
2317 | break; | |
2318 | } | |
2319 | path[i + 1].p_bh = bh; | |
a86c6181 | 2320 | |
d0d856e8 RD |
2321 | /* save actual number of indexes since this |
2322 | * number is changed at the next iteration */ | |
a86c6181 AT |
2323 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2324 | i++; | |
2325 | } else { | |
d0d856e8 | 2326 | /* we finished processing this index, go up */ |
a86c6181 | 2327 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2328 | /* index is empty, remove it; |
a86c6181 AT |
2329 | * handle must be already prepared by the |
2330 | * truncatei_leaf() */ | |
2331 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2332 | } | |
d0d856e8 | 2333 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2334 | brelse(path[i].p_bh); |
2335 | path[i].p_bh = NULL; | |
2336 | i--; | |
2337 | ext_debug("return to level %d\n", i); | |
2338 | } | |
2339 | } | |
2340 | ||
2341 | /* TODO: flexible tree reduction should be here */ | |
2342 | if (path->p_hdr->eh_entries == 0) { | |
2343 | /* | |
d0d856e8 RD |
2344 | * truncate to zero freed all the tree, |
2345 | * so we need to correct eh_depth | |
a86c6181 AT |
2346 | */ |
2347 | err = ext4_ext_get_access(handle, inode, path); | |
2348 | if (err == 0) { | |
2349 | ext_inode_hdr(inode)->eh_depth = 0; | |
2350 | ext_inode_hdr(inode)->eh_max = | |
2351 | cpu_to_le16(ext4_ext_space_root(inode)); | |
2352 | err = ext4_ext_dirty(handle, inode, path); | |
2353 | } | |
2354 | } | |
2355 | out: | |
a86c6181 AT |
2356 | ext4_ext_drop_refs(path); |
2357 | kfree(path); | |
2358 | ext4_journal_stop(handle); | |
2359 | ||
2360 | return err; | |
2361 | } | |
2362 | ||
2363 | /* | |
2364 | * called at mount time | |
2365 | */ | |
2366 | void ext4_ext_init(struct super_block *sb) | |
2367 | { | |
2368 | /* | |
2369 | * possible initialization would be here | |
2370 | */ | |
2371 | ||
83982b6f | 2372 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
4776004f | 2373 | printk(KERN_INFO "EXT4-fs: file extents enabled"); |
bbf2f9fb RD |
2374 | #ifdef AGGRESSIVE_TEST |
2375 | printk(", aggressive tests"); | |
a86c6181 AT |
2376 | #endif |
2377 | #ifdef CHECK_BINSEARCH | |
2378 | printk(", check binsearch"); | |
2379 | #endif | |
2380 | #ifdef EXTENTS_STATS | |
2381 | printk(", stats"); | |
2382 | #endif | |
2383 | printk("\n"); | |
2384 | #ifdef EXTENTS_STATS | |
2385 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2386 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2387 | EXT4_SB(sb)->s_ext_max = 0; | |
2388 | #endif | |
2389 | } | |
2390 | } | |
2391 | ||
2392 | /* | |
2393 | * called at umount time | |
2394 | */ | |
2395 | void ext4_ext_release(struct super_block *sb) | |
2396 | { | |
83982b6f | 2397 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2398 | return; |
2399 | ||
2400 | #ifdef EXTENTS_STATS | |
2401 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2402 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2403 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2404 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2405 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2406 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2407 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2408 | } | |
2409 | #endif | |
2410 | } | |
2411 | ||
093a088b AK |
2412 | static void bi_complete(struct bio *bio, int error) |
2413 | { | |
2414 | complete((struct completion *)bio->bi_private); | |
2415 | } | |
2416 | ||
2417 | /* FIXME!! we need to try to merge to left or right after zero-out */ | |
2418 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2419 | { | |
2420 | int ret = -EIO; | |
2421 | struct bio *bio; | |
2422 | int blkbits, blocksize; | |
2423 | sector_t ee_pblock; | |
2424 | struct completion event; | |
2425 | unsigned int ee_len, len, done, offset; | |
2426 | ||
2427 | ||
2428 | blkbits = inode->i_blkbits; | |
2429 | blocksize = inode->i_sb->s_blocksize; | |
2430 | ee_len = ext4_ext_get_actual_len(ex); | |
2431 | ee_pblock = ext_pblock(ex); | |
2432 | ||
2433 | /* convert ee_pblock to 512 byte sectors */ | |
2434 | ee_pblock = ee_pblock << (blkbits - 9); | |
2435 | ||
2436 | while (ee_len > 0) { | |
2437 | ||
2438 | if (ee_len > BIO_MAX_PAGES) | |
2439 | len = BIO_MAX_PAGES; | |
2440 | else | |
2441 | len = ee_len; | |
2442 | ||
2443 | bio = bio_alloc(GFP_NOIO, len); | |
093a088b AK |
2444 | bio->bi_sector = ee_pblock; |
2445 | bio->bi_bdev = inode->i_sb->s_bdev; | |
2446 | ||
2447 | done = 0; | |
2448 | offset = 0; | |
2449 | while (done < len) { | |
2450 | ret = bio_add_page(bio, ZERO_PAGE(0), | |
2451 | blocksize, offset); | |
2452 | if (ret != blocksize) { | |
2453 | /* | |
2454 | * We can't add any more pages because of | |
2455 | * hardware limitations. Start a new bio. | |
2456 | */ | |
2457 | break; | |
2458 | } | |
2459 | done++; | |
2460 | offset += blocksize; | |
2461 | if (offset >= PAGE_CACHE_SIZE) | |
2462 | offset = 0; | |
2463 | } | |
2464 | ||
2465 | init_completion(&event); | |
2466 | bio->bi_private = &event; | |
2467 | bio->bi_end_io = bi_complete; | |
2468 | submit_bio(WRITE, bio); | |
2469 | wait_for_completion(&event); | |
2470 | ||
2471 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2472 | ret = 0; | |
2473 | else { | |
2474 | ret = -EIO; | |
2475 | break; | |
2476 | } | |
2477 | bio_put(bio); | |
2478 | ee_len -= done; | |
2479 | ee_pblock += done << (blkbits - 9); | |
2480 | } | |
2481 | return ret; | |
2482 | } | |
2483 | ||
3977c965 AK |
2484 | #define EXT4_EXT_ZERO_LEN 7 |
2485 | ||
56055d3a AA |
2486 | /* |
2487 | * This function is called by ext4_ext_get_blocks() if someone tries to write | |
2488 | * to an uninitialized extent. It may result in splitting the uninitialized | |
2489 | * extent into multiple extents (upto three - one initialized and two | |
2490 | * uninitialized). | |
2491 | * There are three possibilities: | |
2492 | * a> There is no split required: Entire extent should be initialized | |
2493 | * b> Splits in two extents: Write is happening at either end of the extent | |
2494 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2495 | */ | |
725d26d3 AK |
2496 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
2497 | struct inode *inode, | |
2498 | struct ext4_ext_path *path, | |
2499 | ext4_lblk_t iblock, | |
498e5f24 | 2500 | unsigned int max_blocks) |
56055d3a | 2501 | { |
95c3889c | 2502 | struct ext4_extent *ex, newex, orig_ex; |
56055d3a AA |
2503 | struct ext4_extent *ex1 = NULL; |
2504 | struct ext4_extent *ex2 = NULL; | |
2505 | struct ext4_extent *ex3 = NULL; | |
2506 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2507 | ext4_lblk_t ee_block; |
2508 | unsigned int allocated, ee_len, depth; | |
56055d3a AA |
2509 | ext4_fsblk_t newblock; |
2510 | int err = 0; | |
2511 | int ret = 0; | |
2512 | ||
2513 | depth = ext_depth(inode); | |
2514 | eh = path[depth].p_hdr; | |
2515 | ex = path[depth].p_ext; | |
2516 | ee_block = le32_to_cpu(ex->ee_block); | |
2517 | ee_len = ext4_ext_get_actual_len(ex); | |
2518 | allocated = ee_len - (iblock - ee_block); | |
2519 | newblock = iblock - ee_block + ext_pblock(ex); | |
2520 | ex2 = ex; | |
95c3889c AK |
2521 | orig_ex.ee_block = ex->ee_block; |
2522 | orig_ex.ee_len = cpu_to_le16(ee_len); | |
2523 | ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); | |
56055d3a | 2524 | |
9df5643a AK |
2525 | err = ext4_ext_get_access(handle, inode, path + depth); |
2526 | if (err) | |
2527 | goto out; | |
3977c965 AK |
2528 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
2529 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN) { | |
2530 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2531 | if (err) | |
2532 | goto fix_extent_len; | |
2533 | /* update the extent length and mark as initialized */ | |
2534 | ex->ee_block = orig_ex.ee_block; | |
2535 | ex->ee_len = orig_ex.ee_len; | |
2536 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2537 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2538 | /* zeroed the full extent */ |
2539 | return allocated; | |
3977c965 | 2540 | } |
9df5643a | 2541 | |
56055d3a AA |
2542 | /* ex1: ee_block to iblock - 1 : uninitialized */ |
2543 | if (iblock > ee_block) { | |
2544 | ex1 = ex; | |
2545 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2546 | ext4_ext_mark_uninitialized(ex1); | |
2547 | ex2 = &newex; | |
2548 | } | |
2549 | /* | |
2550 | * for sanity, update the length of the ex2 extent before | |
2551 | * we insert ex3, if ex1 is NULL. This is to avoid temporary | |
2552 | * overlap of blocks. | |
2553 | */ | |
2554 | if (!ex1 && allocated > max_blocks) | |
2555 | ex2->ee_len = cpu_to_le16(max_blocks); | |
2556 | /* ex3: to ee_block + ee_len : uninitialised */ | |
2557 | if (allocated > max_blocks) { | |
2558 | unsigned int newdepth; | |
3977c965 AK |
2559 | /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */ |
2560 | if (allocated <= EXT4_EXT_ZERO_LEN) { | |
d03856bd AK |
2561 | /* |
2562 | * iblock == ee_block is handled by the zerouout | |
2563 | * at the beginning. | |
2564 | * Mark first half uninitialized. | |
3977c965 AK |
2565 | * Mark second half initialized and zero out the |
2566 | * initialized extent | |
2567 | */ | |
2568 | ex->ee_block = orig_ex.ee_block; | |
2569 | ex->ee_len = cpu_to_le16(ee_len - allocated); | |
2570 | ext4_ext_mark_uninitialized(ex); | |
2571 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2572 | ext4_ext_dirty(handle, inode, path + depth); | |
2573 | ||
2574 | ex3 = &newex; | |
2575 | ex3->ee_block = cpu_to_le32(iblock); | |
2576 | ext4_ext_store_pblock(ex3, newblock); | |
2577 | ex3->ee_len = cpu_to_le16(allocated); | |
2578 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
2579 | if (err == -ENOSPC) { | |
2580 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2581 | if (err) | |
2582 | goto fix_extent_len; | |
2583 | ex->ee_block = orig_ex.ee_block; | |
2584 | ex->ee_len = orig_ex.ee_len; | |
2585 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2586 | ext4_ext_dirty(handle, inode, path + depth); | |
d03856bd | 2587 | /* blocks available from iblock */ |
161e7b7c | 2588 | return allocated; |
3977c965 AK |
2589 | |
2590 | } else if (err) | |
2591 | goto fix_extent_len; | |
2592 | ||
161e7b7c AK |
2593 | /* |
2594 | * We need to zero out the second half because | |
2595 | * an fallocate request can update file size and | |
2596 | * converting the second half to initialized extent | |
2597 | * implies that we can leak some junk data to user | |
2598 | * space. | |
2599 | */ | |
2600 | err = ext4_ext_zeroout(inode, ex3); | |
2601 | if (err) { | |
2602 | /* | |
2603 | * We should actually mark the | |
2604 | * second half as uninit and return error | |
2605 | * Insert would have changed the extent | |
2606 | */ | |
2607 | depth = ext_depth(inode); | |
2608 | ext4_ext_drop_refs(path); | |
2609 | path = ext4_ext_find_extent(inode, | |
2610 | iblock, path); | |
2611 | if (IS_ERR(path)) { | |
2612 | err = PTR_ERR(path); | |
2613 | return err; | |
2614 | } | |
d03856bd | 2615 | /* get the second half extent details */ |
161e7b7c AK |
2616 | ex = path[depth].p_ext; |
2617 | err = ext4_ext_get_access(handle, inode, | |
2618 | path + depth); | |
2619 | if (err) | |
2620 | return err; | |
2621 | ext4_ext_mark_uninitialized(ex); | |
2622 | ext4_ext_dirty(handle, inode, path + depth); | |
2623 | return err; | |
2624 | } | |
2625 | ||
2626 | /* zeroed the second half */ | |
3977c965 AK |
2627 | return allocated; |
2628 | } | |
56055d3a AA |
2629 | ex3 = &newex; |
2630 | ex3->ee_block = cpu_to_le32(iblock + max_blocks); | |
2631 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | |
2632 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | |
2633 | ext4_ext_mark_uninitialized(ex3); | |
2634 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
093a088b AK |
2635 | if (err == -ENOSPC) { |
2636 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2637 | if (err) | |
2638 | goto fix_extent_len; | |
2639 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2640 | ex->ee_block = orig_ex.ee_block; |
2641 | ex->ee_len = orig_ex.ee_len; | |
2642 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2643 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c | 2644 | /* zeroed the full extent */ |
d03856bd | 2645 | /* blocks available from iblock */ |
161e7b7c | 2646 | return allocated; |
093a088b AK |
2647 | |
2648 | } else if (err) | |
2649 | goto fix_extent_len; | |
56055d3a AA |
2650 | /* |
2651 | * The depth, and hence eh & ex might change | |
2652 | * as part of the insert above. | |
2653 | */ | |
2654 | newdepth = ext_depth(inode); | |
95c3889c | 2655 | /* |
73ac36ea | 2656 | * update the extent length after successful insert of the |
95c3889c AK |
2657 | * split extent |
2658 | */ | |
2659 | orig_ex.ee_len = cpu_to_le16(ee_len - | |
2660 | ext4_ext_get_actual_len(ex3)); | |
d03856bd AK |
2661 | depth = newdepth; |
2662 | ext4_ext_drop_refs(path); | |
2663 | path = ext4_ext_find_extent(inode, iblock, path); | |
2664 | if (IS_ERR(path)) { | |
2665 | err = PTR_ERR(path); | |
2666 | goto out; | |
56055d3a | 2667 | } |
d03856bd AK |
2668 | eh = path[depth].p_hdr; |
2669 | ex = path[depth].p_ext; | |
2670 | if (ex2 != &newex) | |
2671 | ex2 = ex; | |
2672 | ||
2673 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2674 | if (err) | |
2675 | goto out; | |
2676 | ||
56055d3a | 2677 | allocated = max_blocks; |
3977c965 AK |
2678 | |
2679 | /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying | |
2680 | * to insert a extent in the middle zerout directly | |
2681 | * otherwise give the extent a chance to merge to left | |
2682 | */ | |
2683 | if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN && | |
2684 | iblock != ee_block) { | |
2685 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2686 | if (err) | |
2687 | goto fix_extent_len; | |
2688 | /* update the extent length and mark as initialized */ | |
2689 | ex->ee_block = orig_ex.ee_block; | |
2690 | ex->ee_len = orig_ex.ee_len; | |
2691 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2692 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c | 2693 | /* zero out the first half */ |
d03856bd | 2694 | /* blocks available from iblock */ |
161e7b7c | 2695 | return allocated; |
3977c965 | 2696 | } |
56055d3a AA |
2697 | } |
2698 | /* | |
2699 | * If there was a change of depth as part of the | |
2700 | * insertion of ex3 above, we need to update the length | |
2701 | * of the ex1 extent again here | |
2702 | */ | |
2703 | if (ex1 && ex1 != ex) { | |
2704 | ex1 = ex; | |
2705 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2706 | ext4_ext_mark_uninitialized(ex1); | |
2707 | ex2 = &newex; | |
2708 | } | |
2709 | /* ex2: iblock to iblock + maxblocks-1 : initialised */ | |
2710 | ex2->ee_block = cpu_to_le32(iblock); | |
56055d3a AA |
2711 | ext4_ext_store_pblock(ex2, newblock); |
2712 | ex2->ee_len = cpu_to_le16(allocated); | |
2713 | if (ex2 != ex) | |
2714 | goto insert; | |
56055d3a AA |
2715 | /* |
2716 | * New (initialized) extent starts from the first block | |
2717 | * in the current extent. i.e., ex2 == ex | |
2718 | * We have to see if it can be merged with the extent | |
2719 | * on the left. | |
2720 | */ | |
2721 | if (ex2 > EXT_FIRST_EXTENT(eh)) { | |
2722 | /* | |
2723 | * To merge left, pass "ex2 - 1" to try_to_merge(), | |
2724 | * since it merges towards right _only_. | |
2725 | */ | |
2726 | ret = ext4_ext_try_to_merge(inode, path, ex2 - 1); | |
2727 | if (ret) { | |
2728 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2729 | if (err) | |
2730 | goto out; | |
2731 | depth = ext_depth(inode); | |
2732 | ex2--; | |
2733 | } | |
2734 | } | |
2735 | /* | |
2736 | * Try to Merge towards right. This might be required | |
2737 | * only when the whole extent is being written to. | |
2738 | * i.e. ex2 == ex and ex3 == NULL. | |
2739 | */ | |
2740 | if (!ex3) { | |
2741 | ret = ext4_ext_try_to_merge(inode, path, ex2); | |
2742 | if (ret) { | |
2743 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2744 | if (err) | |
2745 | goto out; | |
2746 | } | |
2747 | } | |
2748 | /* Mark modified extent as dirty */ | |
2749 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2750 | goto out; | |
2751 | insert: | |
2752 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | |
093a088b AK |
2753 | if (err == -ENOSPC) { |
2754 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2755 | if (err) | |
2756 | goto fix_extent_len; | |
2757 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2758 | ex->ee_block = orig_ex.ee_block; |
2759 | ex->ee_len = orig_ex.ee_len; | |
2760 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2761 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c AK |
2762 | /* zero out the first half */ |
2763 | return allocated; | |
093a088b AK |
2764 | } else if (err) |
2765 | goto fix_extent_len; | |
56055d3a | 2766 | out: |
553f9008 | 2767 | ext4_ext_show_leaf(inode, path); |
56055d3a | 2768 | return err ? err : allocated; |
093a088b AK |
2769 | |
2770 | fix_extent_len: | |
2771 | ex->ee_block = orig_ex.ee_block; | |
2772 | ex->ee_len = orig_ex.ee_len; | |
2773 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2774 | ext4_ext_mark_uninitialized(ex); | |
2775 | ext4_ext_dirty(handle, inode, path + depth); | |
2776 | return err; | |
56055d3a AA |
2777 | } |
2778 | ||
c278bfec | 2779 | /* |
f5ab0d1f MC |
2780 | * Block allocation/map/preallocation routine for extents based files |
2781 | * | |
2782 | * | |
c278bfec | 2783 | * Need to be called with |
0e855ac8 AK |
2784 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
2785 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
2786 | * |
2787 | * return > 0, number of of blocks already mapped/allocated | |
2788 | * if create == 0 and these are pre-allocated blocks | |
2789 | * buffer head is unmapped | |
2790 | * otherwise blocks are mapped | |
2791 | * | |
2792 | * return = 0, if plain look up failed (blocks have not been allocated) | |
2793 | * buffer head is unmapped | |
2794 | * | |
2795 | * return < 0, error case. | |
c278bfec | 2796 | */ |
f65e6fba | 2797 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, |
725d26d3 | 2798 | ext4_lblk_t iblock, |
498e5f24 | 2799 | unsigned int max_blocks, struct buffer_head *bh_result, |
c2177057 | 2800 | int flags) |
a86c6181 AT |
2801 | { |
2802 | struct ext4_ext_path *path = NULL; | |
56055d3a | 2803 | struct ext4_extent_header *eh; |
a86c6181 | 2804 | struct ext4_extent newex, *ex; |
498e5f24 TT |
2805 | ext4_fsblk_t newblock; |
2806 | int err = 0, depth, ret, cache_type; | |
2807 | unsigned int allocated = 0; | |
c9de560d | 2808 | struct ext4_allocation_request ar; |
a86c6181 AT |
2809 | |
2810 | __clear_bit(BH_New, &bh_result->b_state); | |
84fe3bef | 2811 | ext_debug("blocks %u/%u requested for inode %lu\n", |
bba90743 | 2812 | iblock, max_blocks, inode->i_ino); |
a86c6181 AT |
2813 | |
2814 | /* check in cache */ | |
498e5f24 TT |
2815 | cache_type = ext4_ext_in_cache(inode, iblock, &newex); |
2816 | if (cache_type) { | |
2817 | if (cache_type == EXT4_EXT_CACHE_GAP) { | |
c2177057 | 2818 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
2819 | /* |
2820 | * block isn't allocated yet and | |
2821 | * user doesn't want to allocate it | |
2822 | */ | |
a86c6181 AT |
2823 | goto out2; |
2824 | } | |
2825 | /* we should allocate requested block */ | |
498e5f24 | 2826 | } else if (cache_type == EXT4_EXT_CACHE_EXTENT) { |
a86c6181 | 2827 | /* block is already allocated */ |
8c55e204 DK |
2828 | newblock = iblock |
2829 | - le32_to_cpu(newex.ee_block) | |
2830 | + ext_pblock(&newex); | |
d0d856e8 | 2831 | /* number of remaining blocks in the extent */ |
b939e376 | 2832 | allocated = ext4_ext_get_actual_len(&newex) - |
a86c6181 AT |
2833 | (iblock - le32_to_cpu(newex.ee_block)); |
2834 | goto out; | |
2835 | } else { | |
2836 | BUG(); | |
2837 | } | |
2838 | } | |
2839 | ||
2840 | /* find extent for this block */ | |
2841 | path = ext4_ext_find_extent(inode, iblock, NULL); | |
2842 | if (IS_ERR(path)) { | |
2843 | err = PTR_ERR(path); | |
2844 | path = NULL; | |
2845 | goto out2; | |
2846 | } | |
2847 | ||
2848 | depth = ext_depth(inode); | |
2849 | ||
2850 | /* | |
d0d856e8 RD |
2851 | * consistent leaf must not be empty; |
2852 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
2853 | * this is why assert can't be put in ext4_ext_find_extent() |
2854 | */ | |
2855 | BUG_ON(path[depth].p_ext == NULL && depth != 0); | |
56055d3a | 2856 | eh = path[depth].p_hdr; |
a86c6181 | 2857 | |
7e028976 AM |
2858 | ex = path[depth].p_ext; |
2859 | if (ex) { | |
725d26d3 | 2860 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
f65e6fba | 2861 | ext4_fsblk_t ee_start = ext_pblock(ex); |
a2df2a63 | 2862 | unsigned short ee_len; |
471d4011 SB |
2863 | |
2864 | /* | |
471d4011 | 2865 | * Uninitialized extents are treated as holes, except that |
56055d3a | 2866 | * we split out initialized portions during a write. |
471d4011 | 2867 | */ |
a2df2a63 | 2868 | ee_len = ext4_ext_get_actual_len(ex); |
d0d856e8 | 2869 | /* if found extent covers block, simply return it */ |
8c55e204 | 2870 | if (iblock >= ee_block && iblock < ee_block + ee_len) { |
a86c6181 | 2871 | newblock = iblock - ee_block + ee_start; |
d0d856e8 | 2872 | /* number of remaining blocks in the extent */ |
a86c6181 | 2873 | allocated = ee_len - (iblock - ee_block); |
84fe3bef | 2874 | ext_debug("%u fit into %u:%d -> %llu\n", iblock, |
a86c6181 | 2875 | ee_block, ee_len, newblock); |
56055d3a | 2876 | |
a2df2a63 | 2877 | /* Do not put uninitialized extent in the cache */ |
56055d3a | 2878 | if (!ext4_ext_is_uninitialized(ex)) { |
a2df2a63 AA |
2879 | ext4_ext_put_in_cache(inode, ee_block, |
2880 | ee_len, ee_start, | |
2881 | EXT4_EXT_CACHE_EXTENT); | |
56055d3a AA |
2882 | goto out; |
2883 | } | |
c2177057 | 2884 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) |
56055d3a | 2885 | goto out; |
c2177057 | 2886 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
29fa89d0 AK |
2887 | if (allocated > max_blocks) |
2888 | allocated = max_blocks; | |
e067ba00 AK |
2889 | /* |
2890 | * We have blocks reserved already. We | |
2891 | * return allocated blocks so that delalloc | |
2892 | * won't do block reservation for us. But | |
2893 | * the buffer head will be unmapped so that | |
2894 | * a read from the block returns 0s. | |
2895 | */ | |
953e622b | 2896 | set_buffer_unwritten(bh_result); |
9c1ee184 AK |
2897 | bh_result->b_bdev = inode->i_sb->s_bdev; |
2898 | bh_result->b_blocknr = newblock; | |
56055d3a | 2899 | goto out2; |
e067ba00 | 2900 | } |
56055d3a AA |
2901 | |
2902 | ret = ext4_ext_convert_to_initialized(handle, inode, | |
2903 | path, iblock, | |
2904 | max_blocks); | |
dbf9d7da DM |
2905 | if (ret <= 0) { |
2906 | err = ret; | |
56055d3a | 2907 | goto out2; |
dbf9d7da | 2908 | } else |
56055d3a AA |
2909 | allocated = ret; |
2910 | goto outnew; | |
a86c6181 AT |
2911 | } |
2912 | } | |
2913 | ||
2914 | /* | |
d0d856e8 | 2915 | * requested block isn't allocated yet; |
a86c6181 AT |
2916 | * we couldn't try to create block if create flag is zero |
2917 | */ | |
c2177057 | 2918 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
2919 | /* |
2920 | * put just found gap into cache to speed up | |
2921 | * subsequent requests | |
2922 | */ | |
a86c6181 AT |
2923 | ext4_ext_put_gap_in_cache(inode, path, iblock); |
2924 | goto out2; | |
2925 | } | |
2926 | /* | |
c2ea3fde | 2927 | * Okay, we need to do block allocation. |
63f57933 | 2928 | */ |
a86c6181 | 2929 | |
c9de560d AT |
2930 | /* find neighbour allocated blocks */ |
2931 | ar.lleft = iblock; | |
2932 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); | |
2933 | if (err) | |
2934 | goto out2; | |
2935 | ar.lright = iblock; | |
2936 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright); | |
2937 | if (err) | |
2938 | goto out2; | |
25d14f98 | 2939 | |
749269fa AA |
2940 | /* |
2941 | * See if request is beyond maximum number of blocks we can have in | |
2942 | * a single extent. For an initialized extent this limit is | |
2943 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
2944 | * EXT_UNINIT_MAX_LEN. | |
2945 | */ | |
2946 | if (max_blocks > EXT_INIT_MAX_LEN && | |
c2177057 | 2947 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
749269fa AA |
2948 | max_blocks = EXT_INIT_MAX_LEN; |
2949 | else if (max_blocks > EXT_UNINIT_MAX_LEN && | |
c2177057 | 2950 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
749269fa AA |
2951 | max_blocks = EXT_UNINIT_MAX_LEN; |
2952 | ||
25d14f98 AA |
2953 | /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ |
2954 | newex.ee_block = cpu_to_le32(iblock); | |
2955 | newex.ee_len = cpu_to_le16(max_blocks); | |
2956 | err = ext4_ext_check_overlap(inode, &newex, path); | |
2957 | if (err) | |
b939e376 | 2958 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 AA |
2959 | else |
2960 | allocated = max_blocks; | |
c9de560d AT |
2961 | |
2962 | /* allocate new block */ | |
2963 | ar.inode = inode; | |
2964 | ar.goal = ext4_ext_find_goal(inode, path, iblock); | |
2965 | ar.logical = iblock; | |
2966 | ar.len = allocated; | |
2967 | if (S_ISREG(inode->i_mode)) | |
2968 | ar.flags = EXT4_MB_HINT_DATA; | |
2969 | else | |
2970 | /* disable in-core preallocation for non-regular files */ | |
2971 | ar.flags = 0; | |
2972 | newblock = ext4_mb_new_blocks(handle, &ar, &err); | |
a86c6181 AT |
2973 | if (!newblock) |
2974 | goto out2; | |
84fe3bef | 2975 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 2976 | ar.goal, newblock, allocated); |
a86c6181 AT |
2977 | |
2978 | /* try to insert new extent into found leaf and return */ | |
f65e6fba | 2979 | ext4_ext_store_pblock(&newex, newblock); |
c9de560d | 2980 | newex.ee_len = cpu_to_le16(ar.len); |
c2177057 | 2981 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) /* Mark uninitialized */ |
a2df2a63 | 2982 | ext4_ext_mark_uninitialized(&newex); |
a86c6181 | 2983 | err = ext4_ext_insert_extent(handle, inode, path, &newex); |
315054f0 AT |
2984 | if (err) { |
2985 | /* free data blocks we just allocated */ | |
c9de560d AT |
2986 | /* not a good idea to call discard here directly, |
2987 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 2988 | ext4_discard_preallocations(inode); |
315054f0 | 2989 | ext4_free_blocks(handle, inode, ext_pblock(&newex), |
b939e376 | 2990 | ext4_ext_get_actual_len(&newex), 0); |
a86c6181 | 2991 | goto out2; |
315054f0 | 2992 | } |
a86c6181 | 2993 | |
a86c6181 | 2994 | /* previous routine could use block we allocated */ |
f65e6fba | 2995 | newblock = ext_pblock(&newex); |
b939e376 | 2996 | allocated = ext4_ext_get_actual_len(&newex); |
56055d3a | 2997 | outnew: |
953e622b | 2998 | set_buffer_new(bh_result); |
a86c6181 | 2999 | |
a2df2a63 | 3000 | /* Cache only when it is _not_ an uninitialized extent */ |
c2177057 | 3001 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) |
a2df2a63 AA |
3002 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, |
3003 | EXT4_EXT_CACHE_EXTENT); | |
a86c6181 AT |
3004 | out: |
3005 | if (allocated > max_blocks) | |
3006 | allocated = max_blocks; | |
3007 | ext4_ext_show_leaf(inode, path); | |
953e622b | 3008 | set_buffer_mapped(bh_result); |
a86c6181 AT |
3009 | bh_result->b_bdev = inode->i_sb->s_bdev; |
3010 | bh_result->b_blocknr = newblock; | |
3011 | out2: | |
3012 | if (path) { | |
3013 | ext4_ext_drop_refs(path); | |
3014 | kfree(path); | |
3015 | } | |
a86c6181 AT |
3016 | return err ? err : allocated; |
3017 | } | |
3018 | ||
cf108bca | 3019 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
3020 | { |
3021 | struct address_space *mapping = inode->i_mapping; | |
3022 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 3023 | ext4_lblk_t last_block; |
a86c6181 AT |
3024 | handle_t *handle; |
3025 | int err = 0; | |
3026 | ||
3027 | /* | |
3028 | * probably first extent we're gonna free will be last in block | |
3029 | */ | |
f3bd1f3f | 3030 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 3031 | handle = ext4_journal_start(inode, err); |
cf108bca | 3032 | if (IS_ERR(handle)) |
a86c6181 | 3033 | return; |
a86c6181 | 3034 | |
cf108bca JK |
3035 | if (inode->i_size & (sb->s_blocksize - 1)) |
3036 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
a86c6181 | 3037 | |
9ddfc3dc JK |
3038 | if (ext4_orphan_add(handle, inode)) |
3039 | goto out_stop; | |
3040 | ||
0e855ac8 | 3041 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
3042 | ext4_ext_invalidate_cache(inode); |
3043 | ||
c2ea3fde | 3044 | ext4_discard_preallocations(inode); |
c9de560d | 3045 | |
a86c6181 | 3046 | /* |
d0d856e8 RD |
3047 | * TODO: optimization is possible here. |
3048 | * Probably we need not scan at all, | |
3049 | * because page truncation is enough. | |
a86c6181 | 3050 | */ |
a86c6181 AT |
3051 | |
3052 | /* we have to know where to truncate from in crash case */ | |
3053 | EXT4_I(inode)->i_disksize = inode->i_size; | |
3054 | ext4_mark_inode_dirty(handle, inode); | |
3055 | ||
3056 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
3057 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
3058 | err = ext4_ext_remove_space(inode, last_block); | |
3059 | ||
3060 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
3061 | * transaction synchronous. |
3062 | */ | |
a86c6181 | 3063 | if (IS_SYNC(inode)) |
0390131b | 3064 | ext4_handle_sync(handle); |
a86c6181 AT |
3065 | |
3066 | out_stop: | |
9ddfc3dc | 3067 | up_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 | 3068 | /* |
d0d856e8 | 3069 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
3070 | * then we need to clear up the orphan record which we created above. |
3071 | * However, if this was a real unlink then we were called by | |
3072 | * ext4_delete_inode(), and we allow that function to clean up the | |
3073 | * orphan info for us. | |
3074 | */ | |
3075 | if (inode->i_nlink) | |
3076 | ext4_orphan_del(handle, inode); | |
3077 | ||
ef737728 SR |
3078 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3079 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
3080 | ext4_journal_stop(handle); |
3081 | } | |
3082 | ||
fd28784a AK |
3083 | static void ext4_falloc_update_inode(struct inode *inode, |
3084 | int mode, loff_t new_size, int update_ctime) | |
3085 | { | |
3086 | struct timespec now; | |
3087 | ||
3088 | if (update_ctime) { | |
3089 | now = current_fs_time(inode->i_sb); | |
3090 | if (!timespec_equal(&inode->i_ctime, &now)) | |
3091 | inode->i_ctime = now; | |
3092 | } | |
3093 | /* | |
3094 | * Update only when preallocation was requested beyond | |
3095 | * the file size. | |
3096 | */ | |
cf17fea6 AK |
3097 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
3098 | if (new_size > i_size_read(inode)) | |
3099 | i_size_write(inode, new_size); | |
3100 | if (new_size > EXT4_I(inode)->i_disksize) | |
3101 | ext4_update_i_disksize(inode, new_size); | |
fd28784a AK |
3102 | } |
3103 | ||
3104 | } | |
3105 | ||
a2df2a63 AA |
3106 | /* |
3107 | * preallocate space for a file. This implements ext4's fallocate inode | |
3108 | * operation, which gets called from sys_fallocate system call. | |
3109 | * For block-mapped files, posix_fallocate should fall back to the method | |
3110 | * of writing zeroes to the required new blocks (the same behavior which is | |
3111 | * expected for file systems which do not support fallocate() system call). | |
3112 | */ | |
3113 | long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) | |
3114 | { | |
3115 | handle_t *handle; | |
725d26d3 | 3116 | ext4_lblk_t block; |
fd28784a | 3117 | loff_t new_size; |
498e5f24 | 3118 | unsigned int max_blocks; |
a2df2a63 AA |
3119 | int ret = 0; |
3120 | int ret2 = 0; | |
3121 | int retries = 0; | |
3122 | struct buffer_head map_bh; | |
3123 | unsigned int credits, blkbits = inode->i_blkbits; | |
3124 | ||
3125 | /* | |
3126 | * currently supporting (pre)allocate mode for extent-based | |
3127 | * files _only_ | |
3128 | */ | |
3129 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
3130 | return -EOPNOTSUPP; | |
3131 | ||
3132 | /* preallocation to directories is currently not supported */ | |
3133 | if (S_ISDIR(inode->i_mode)) | |
3134 | return -ENODEV; | |
3135 | ||
3136 | block = offset >> blkbits; | |
fd28784a AK |
3137 | /* |
3138 | * We can't just convert len to max_blocks because | |
3139 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
3140 | */ | |
a2df2a63 | 3141 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
fd28784a | 3142 | - block; |
a2df2a63 | 3143 | /* |
f3bd1f3f | 3144 | * credits to insert 1 extent into extent tree |
a2df2a63 | 3145 | */ |
f3bd1f3f | 3146 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 3147 | mutex_lock(&inode->i_mutex); |
a2df2a63 AA |
3148 | retry: |
3149 | while (ret >= 0 && ret < max_blocks) { | |
3150 | block = block + ret; | |
3151 | max_blocks = max_blocks - ret; | |
3152 | handle = ext4_journal_start(inode, credits); | |
3153 | if (IS_ERR(handle)) { | |
3154 | ret = PTR_ERR(handle); | |
3155 | break; | |
3156 | } | |
79ffab34 | 3157 | map_bh.b_state = 0; |
12b7ac17 TT |
3158 | ret = ext4_get_blocks(handle, inode, block, |
3159 | max_blocks, &map_bh, | |
c2177057 | 3160 | EXT4_GET_BLOCKS_CREATE_UNINIT_EXT); |
221879c9 | 3161 | if (ret <= 0) { |
2c98615d AK |
3162 | #ifdef EXT4FS_DEBUG |
3163 | WARN_ON(ret <= 0); | |
3164 | printk(KERN_ERR "%s: ext4_ext_get_blocks " | |
3165 | "returned error inode#%lu, block=%u, " | |
9fd9784c | 3166 | "max_blocks=%u", __func__, |
221879c9 | 3167 | inode->i_ino, block, max_blocks); |
2c98615d | 3168 | #endif |
a2df2a63 AA |
3169 | ext4_mark_inode_dirty(handle, inode); |
3170 | ret2 = ext4_journal_stop(handle); | |
3171 | break; | |
3172 | } | |
fd28784a AK |
3173 | if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
3174 | blkbits) >> blkbits)) | |
3175 | new_size = offset + len; | |
3176 | else | |
3177 | new_size = (block + ret) << blkbits; | |
a2df2a63 | 3178 | |
fd28784a AK |
3179 | ext4_falloc_update_inode(inode, mode, new_size, |
3180 | buffer_new(&map_bh)); | |
a2df2a63 AA |
3181 | ext4_mark_inode_dirty(handle, inode); |
3182 | ret2 = ext4_journal_stop(handle); | |
3183 | if (ret2) | |
3184 | break; | |
3185 | } | |
fd28784a AK |
3186 | if (ret == -ENOSPC && |
3187 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
3188 | ret = 0; | |
a2df2a63 | 3189 | goto retry; |
a2df2a63 | 3190 | } |
55bd725a | 3191 | mutex_unlock(&inode->i_mutex); |
a2df2a63 AA |
3192 | return ret > 0 ? ret2 : ret; |
3193 | } | |
6873fa0d ES |
3194 | |
3195 | /* | |
3196 | * Callback function called for each extent to gather FIEMAP information. | |
3197 | */ | |
3a06d778 | 3198 | static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path, |
6873fa0d ES |
3199 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
3200 | void *data) | |
3201 | { | |
3202 | struct fiemap_extent_info *fieinfo = data; | |
c9877b20 | 3203 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; |
6873fa0d ES |
3204 | __u64 logical; |
3205 | __u64 physical; | |
3206 | __u64 length; | |
3207 | __u32 flags = 0; | |
3208 | int error; | |
3209 | ||
3210 | logical = (__u64)newex->ec_block << blksize_bits; | |
3211 | ||
3212 | if (newex->ec_type == EXT4_EXT_CACHE_GAP) { | |
3213 | pgoff_t offset; | |
3214 | struct page *page; | |
3215 | struct buffer_head *bh = NULL; | |
3216 | ||
3217 | offset = logical >> PAGE_SHIFT; | |
3218 | page = find_get_page(inode->i_mapping, offset); | |
3219 | if (!page || !page_has_buffers(page)) | |
3220 | return EXT_CONTINUE; | |
3221 | ||
3222 | bh = page_buffers(page); | |
3223 | ||
3224 | if (!bh) | |
3225 | return EXT_CONTINUE; | |
3226 | ||
3227 | if (buffer_delay(bh)) { | |
3228 | flags |= FIEMAP_EXTENT_DELALLOC; | |
3229 | page_cache_release(page); | |
3230 | } else { | |
3231 | page_cache_release(page); | |
3232 | return EXT_CONTINUE; | |
3233 | } | |
3234 | } | |
3235 | ||
3236 | physical = (__u64)newex->ec_start << blksize_bits; | |
3237 | length = (__u64)newex->ec_len << blksize_bits; | |
3238 | ||
3239 | if (ex && ext4_ext_is_uninitialized(ex)) | |
3240 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
3241 | ||
3242 | /* | |
3243 | * If this extent reaches EXT_MAX_BLOCK, it must be last. | |
3244 | * | |
3245 | * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK, | |
3246 | * this also indicates no more allocated blocks. | |
3247 | * | |
3248 | * XXX this might miss a single-block extent at EXT_MAX_BLOCK | |
3249 | */ | |
c9877b20 | 3250 | if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK || |
eefd7f03 TT |
3251 | newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) { |
3252 | loff_t size = i_size_read(inode); | |
3253 | loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb); | |
3254 | ||
6873fa0d | 3255 | flags |= FIEMAP_EXTENT_LAST; |
eefd7f03 TT |
3256 | if ((flags & FIEMAP_EXTENT_DELALLOC) && |
3257 | logical+length > size) | |
3258 | length = (size - logical + bs - 1) & ~(bs-1); | |
3259 | } | |
6873fa0d ES |
3260 | |
3261 | error = fiemap_fill_next_extent(fieinfo, logical, physical, | |
3262 | length, flags); | |
3263 | if (error < 0) | |
3264 | return error; | |
3265 | if (error == 1) | |
3266 | return EXT_BREAK; | |
3267 | ||
3268 | return EXT_CONTINUE; | |
3269 | } | |
3270 | ||
3271 | /* fiemap flags we can handle specified here */ | |
3272 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
3273 | ||
3a06d778 AK |
3274 | static int ext4_xattr_fiemap(struct inode *inode, |
3275 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
3276 | { |
3277 | __u64 physical = 0; | |
3278 | __u64 length; | |
3279 | __u32 flags = FIEMAP_EXTENT_LAST; | |
3280 | int blockbits = inode->i_sb->s_blocksize_bits; | |
3281 | int error = 0; | |
3282 | ||
3283 | /* in-inode? */ | |
3284 | if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) { | |
3285 | struct ext4_iloc iloc; | |
3286 | int offset; /* offset of xattr in inode */ | |
3287 | ||
3288 | error = ext4_get_inode_loc(inode, &iloc); | |
3289 | if (error) | |
3290 | return error; | |
3291 | physical = iloc.bh->b_blocknr << blockbits; | |
3292 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
3293 | EXT4_I(inode)->i_extra_isize; | |
3294 | physical += offset; | |
3295 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
3296 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
3297 | } else { /* external block */ | |
3298 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
3299 | length = inode->i_sb->s_blocksize; | |
3300 | } | |
3301 | ||
3302 | if (physical) | |
3303 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
3304 | length, flags); | |
3305 | return (error < 0 ? error : 0); | |
3306 | } | |
3307 | ||
3308 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
3309 | __u64 start, __u64 len) | |
3310 | { | |
3311 | ext4_lblk_t start_blk; | |
3312 | ext4_lblk_t len_blks; | |
3313 | int error = 0; | |
3314 | ||
3315 | /* fallback to generic here if not in extents fmt */ | |
3316 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
3317 | return generic_block_fiemap(inode, fieinfo, start, len, | |
3318 | ext4_get_block); | |
3319 | ||
3320 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
3321 | return -EBADR; | |
3322 | ||
3323 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
3324 | error = ext4_xattr_fiemap(inode, fieinfo); | |
3325 | } else { | |
3326 | start_blk = start >> inode->i_sb->s_blocksize_bits; | |
3327 | len_blks = len >> inode->i_sb->s_blocksize_bits; | |
3328 | ||
3329 | /* | |
3330 | * Walk the extent tree gathering extent information. | |
3331 | * ext4_ext_fiemap_cb will push extents back to user. | |
3332 | */ | |
0568c518 | 3333 | down_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
3334 | error = ext4_ext_walk_space(inode, start_blk, len_blks, |
3335 | ext4_ext_fiemap_cb, fieinfo); | |
0568c518 | 3336 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
3337 | } |
3338 | ||
3339 | return error; | |
3340 | } | |
3341 |