]>
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 | ||
a86c6181 AT |
32 | #include <linux/fs.h> |
33 | #include <linux/time.h> | |
cd02ff0b | 34 | #include <linux/jbd2.h> |
a86c6181 AT |
35 | #include <linux/highuid.h> |
36 | #include <linux/pagemap.h> | |
37 | #include <linux/quotaops.h> | |
38 | #include <linux/string.h> | |
39 | #include <linux/slab.h> | |
a2df2a63 | 40 | #include <linux/falloc.h> |
a86c6181 | 41 | #include <asm/uaccess.h> |
6873fa0d | 42 | #include <linux/fiemap.h> |
3dcf5451 | 43 | #include "ext4_jbd2.h" |
a86c6181 | 44 | |
0562e0ba JZ |
45 | #include <trace/events/ext4.h> |
46 | ||
5f95d21f LC |
47 | /* |
48 | * used by extent splitting. | |
49 | */ | |
50 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
51 | due to ENOSPC */ | |
52 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
53 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
54 | ||
7ac5990d DW |
55 | static __le32 ext4_extent_block_csum(struct inode *inode, |
56 | struct ext4_extent_header *eh) | |
57 | { | |
58 | struct ext4_inode_info *ei = EXT4_I(inode); | |
59 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
60 | __u32 csum; | |
61 | ||
62 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
63 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
64 | return cpu_to_le32(csum); | |
65 | } | |
66 | ||
67 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
68 | struct ext4_extent_header *eh) | |
69 | { | |
70 | struct ext4_extent_tail *et; | |
71 | ||
72 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
73 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
74 | return 1; | |
75 | ||
76 | et = find_ext4_extent_tail(eh); | |
77 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
78 | return 0; | |
79 | return 1; | |
80 | } | |
81 | ||
82 | static void ext4_extent_block_csum_set(struct inode *inode, | |
83 | struct ext4_extent_header *eh) | |
84 | { | |
85 | struct ext4_extent_tail *et; | |
86 | ||
87 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
88 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
89 | return; | |
90 | ||
91 | et = find_ext4_extent_tail(eh); | |
92 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
93 | } | |
94 | ||
d583fb87 AH |
95 | static int ext4_split_extent(handle_t *handle, |
96 | struct inode *inode, | |
97 | struct ext4_ext_path *path, | |
98 | struct ext4_map_blocks *map, | |
99 | int split_flag, | |
100 | int flags); | |
101 | ||
5f95d21f LC |
102 | static int ext4_split_extent_at(handle_t *handle, |
103 | struct inode *inode, | |
104 | struct ext4_ext_path *path, | |
105 | ext4_lblk_t split, | |
106 | int split_flag, | |
107 | int flags); | |
108 | ||
487caeef JK |
109 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
110 | struct inode *inode, | |
111 | int needed) | |
a86c6181 AT |
112 | { |
113 | int err; | |
114 | ||
0390131b FM |
115 | if (!ext4_handle_valid(handle)) |
116 | return 0; | |
a86c6181 | 117 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
118 | return 0; |
119 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 120 | if (err <= 0) |
9102e4fa | 121 | return err; |
487caeef | 122 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
123 | if (err == 0) |
124 | err = -EAGAIN; | |
487caeef JK |
125 | |
126 | return err; | |
a86c6181 AT |
127 | } |
128 | ||
129 | /* | |
130 | * could return: | |
131 | * - EROFS | |
132 | * - ENOMEM | |
133 | */ | |
134 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
135 | struct ext4_ext_path *path) | |
136 | { | |
137 | if (path->p_bh) { | |
138 | /* path points to block */ | |
139 | return ext4_journal_get_write_access(handle, path->p_bh); | |
140 | } | |
141 | /* path points to leaf/index in inode body */ | |
142 | /* we use in-core data, no need to protect them */ | |
143 | return 0; | |
144 | } | |
145 | ||
146 | /* | |
147 | * could return: | |
148 | * - EROFS | |
149 | * - ENOMEM | |
150 | * - EIO | |
151 | */ | |
9ea7a0df TT |
152 | #define ext4_ext_dirty(handle, inode, path) \ |
153 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
154 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
155 | handle_t *handle, struct inode *inode, | |
156 | struct ext4_ext_path *path) | |
a86c6181 AT |
157 | { |
158 | int err; | |
159 | if (path->p_bh) { | |
7ac5990d | 160 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 161 | /* path points to block */ |
9ea7a0df TT |
162 | err = __ext4_handle_dirty_metadata(where, line, handle, |
163 | inode, path->p_bh); | |
a86c6181 AT |
164 | } else { |
165 | /* path points to leaf/index in inode body */ | |
166 | err = ext4_mark_inode_dirty(handle, inode); | |
167 | } | |
168 | return err; | |
169 | } | |
170 | ||
f65e6fba | 171 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 172 | struct ext4_ext_path *path, |
725d26d3 | 173 | ext4_lblk_t block) |
a86c6181 | 174 | { |
a86c6181 | 175 | if (path) { |
81fdbb4a | 176 | int depth = path->p_depth; |
a86c6181 | 177 | struct ext4_extent *ex; |
a86c6181 | 178 | |
ad4fb9ca KM |
179 | /* |
180 | * Try to predict block placement assuming that we are | |
181 | * filling in a file which will eventually be | |
182 | * non-sparse --- i.e., in the case of libbfd writing | |
183 | * an ELF object sections out-of-order but in a way | |
184 | * the eventually results in a contiguous object or | |
185 | * executable file, or some database extending a table | |
186 | * space file. However, this is actually somewhat | |
187 | * non-ideal if we are writing a sparse file such as | |
188 | * qemu or KVM writing a raw image file that is going | |
189 | * to stay fairly sparse, since it will end up | |
190 | * fragmenting the file system's free space. Maybe we | |
191 | * should have some hueristics or some way to allow | |
192 | * userspace to pass a hint to file system, | |
b8d6568a | 193 | * especially if the latter case turns out to be |
ad4fb9ca KM |
194 | * common. |
195 | */ | |
7e028976 | 196 | ex = path[depth].p_ext; |
ad4fb9ca KM |
197 | if (ex) { |
198 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
199 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
200 | ||
201 | if (block > ext_block) | |
202 | return ext_pblk + (block - ext_block); | |
203 | else | |
204 | return ext_pblk - (ext_block - block); | |
205 | } | |
a86c6181 | 206 | |
d0d856e8 RD |
207 | /* it looks like index is empty; |
208 | * try to find starting block from index itself */ | |
a86c6181 AT |
209 | if (path[depth].p_bh) |
210 | return path[depth].p_bh->b_blocknr; | |
211 | } | |
212 | ||
213 | /* OK. use inode's group */ | |
f86186b4 | 214 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
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 | 222 | struct ext4_ext_path *path, |
55f020db | 223 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 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)); | |
55f020db AH |
228 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
229 | NULL, err); | |
a86c6181 AT |
230 | return newblock; |
231 | } | |
232 | ||
55ad63bf | 233 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
234 | { |
235 | int size; | |
236 | ||
237 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
238 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 239 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
240 | if (!check && size > 6) |
241 | size = 6; | |
a86c6181 AT |
242 | #endif |
243 | return size; | |
244 | } | |
245 | ||
55ad63bf | 246 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
247 | { |
248 | int size; | |
249 | ||
250 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
251 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 252 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
253 | if (!check && size > 5) |
254 | size = 5; | |
a86c6181 AT |
255 | #endif |
256 | return size; | |
257 | } | |
258 | ||
55ad63bf | 259 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
260 | { |
261 | int size; | |
262 | ||
263 | size = sizeof(EXT4_I(inode)->i_data); | |
264 | size -= sizeof(struct ext4_extent_header); | |
265 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 266 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
267 | if (!check && size > 3) |
268 | size = 3; | |
a86c6181 AT |
269 | #endif |
270 | return size; | |
271 | } | |
272 | ||
55ad63bf | 273 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
274 | { |
275 | int size; | |
276 | ||
277 | size = sizeof(EXT4_I(inode)->i_data); | |
278 | size -= sizeof(struct ext4_extent_header); | |
279 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 280 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
281 | if (!check && size > 4) |
282 | size = 4; | |
a86c6181 AT |
283 | #endif |
284 | return size; | |
285 | } | |
286 | ||
d2a17637 MC |
287 | /* |
288 | * Calculate the number of metadata blocks needed | |
289 | * to allocate @blocks | |
290 | * Worse case is one block per extent | |
291 | */ | |
01f49d0b | 292 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 293 | { |
9d0be502 | 294 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 295 | int idxs; |
d2a17637 | 296 | |
9d0be502 TT |
297 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
298 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
299 | |
300 | /* | |
9d0be502 TT |
301 | * If the new delayed allocation block is contiguous with the |
302 | * previous da block, it can share index blocks with the | |
303 | * previous block, so we only need to allocate a new index | |
304 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
305 | * an additional index block, and at ldxs**3 blocks, yet | |
306 | * another index blocks. | |
d2a17637 | 307 | */ |
9d0be502 TT |
308 | if (ei->i_da_metadata_calc_len && |
309 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
310 | int num = 0; |
311 | ||
9d0be502 TT |
312 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
313 | num++; | |
314 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
315 | num++; | |
316 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
317 | num++; | |
318 | ei->i_da_metadata_calc_len = 0; | |
319 | } else | |
320 | ei->i_da_metadata_calc_len++; | |
321 | ei->i_da_metadata_calc_last_lblock++; | |
322 | return num; | |
323 | } | |
d2a17637 | 324 | |
9d0be502 TT |
325 | /* |
326 | * In the worst case we need a new set of index blocks at | |
327 | * every level of the inode's extent tree. | |
328 | */ | |
329 | ei->i_da_metadata_calc_len = 1; | |
330 | ei->i_da_metadata_calc_last_lblock = lblock; | |
331 | return ext_depth(inode) + 1; | |
d2a17637 MC |
332 | } |
333 | ||
c29c0ae7 AT |
334 | static int |
335 | ext4_ext_max_entries(struct inode *inode, int depth) | |
336 | { | |
337 | int max; | |
338 | ||
339 | if (depth == ext_depth(inode)) { | |
340 | if (depth == 0) | |
55ad63bf | 341 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 342 | else |
55ad63bf | 343 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
344 | } else { |
345 | if (depth == 0) | |
55ad63bf | 346 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 347 | else |
55ad63bf | 348 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
349 | } |
350 | ||
351 | return max; | |
352 | } | |
353 | ||
56b19868 AK |
354 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
355 | { | |
bf89d16f | 356 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 357 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 358 | |
31d4f3a2 TT |
359 | if (len == 0) |
360 | return 0; | |
6fd058f7 | 361 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
362 | } |
363 | ||
364 | static int ext4_valid_extent_idx(struct inode *inode, | |
365 | struct ext4_extent_idx *ext_idx) | |
366 | { | |
bf89d16f | 367 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 368 | |
6fd058f7 | 369 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
370 | } |
371 | ||
372 | static int ext4_valid_extent_entries(struct inode *inode, | |
373 | struct ext4_extent_header *eh, | |
374 | int depth) | |
375 | { | |
56b19868 AK |
376 | unsigned short entries; |
377 | if (eh->eh_entries == 0) | |
378 | return 1; | |
379 | ||
380 | entries = le16_to_cpu(eh->eh_entries); | |
381 | ||
382 | if (depth == 0) { | |
383 | /* leaf entries */ | |
81fdbb4a | 384 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
56b19868 AK |
385 | while (entries) { |
386 | if (!ext4_valid_extent(inode, ext)) | |
387 | return 0; | |
388 | ext++; | |
389 | entries--; | |
390 | } | |
391 | } else { | |
81fdbb4a | 392 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
393 | while (entries) { |
394 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
395 | return 0; | |
396 | ext_idx++; | |
397 | entries--; | |
398 | } | |
399 | } | |
400 | return 1; | |
401 | } | |
402 | ||
c398eda0 TT |
403 | static int __ext4_ext_check(const char *function, unsigned int line, |
404 | struct inode *inode, struct ext4_extent_header *eh, | |
405 | int depth) | |
c29c0ae7 AT |
406 | { |
407 | const char *error_msg; | |
408 | int max = 0; | |
409 | ||
410 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
411 | error_msg = "invalid magic"; | |
412 | goto corrupted; | |
413 | } | |
414 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
415 | error_msg = "unexpected eh_depth"; | |
416 | goto corrupted; | |
417 | } | |
418 | if (unlikely(eh->eh_max == 0)) { | |
419 | error_msg = "invalid eh_max"; | |
420 | goto corrupted; | |
421 | } | |
422 | max = ext4_ext_max_entries(inode, depth); | |
423 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
424 | error_msg = "too large eh_max"; | |
425 | goto corrupted; | |
426 | } | |
427 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
428 | error_msg = "invalid eh_entries"; | |
429 | goto corrupted; | |
430 | } | |
56b19868 AK |
431 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
432 | error_msg = "invalid extent entries"; | |
433 | goto corrupted; | |
434 | } | |
7ac5990d DW |
435 | /* Verify checksum on non-root extent tree nodes */ |
436 | if (ext_depth(inode) != depth && | |
437 | !ext4_extent_block_csum_verify(inode, eh)) { | |
438 | error_msg = "extent tree corrupted"; | |
439 | goto corrupted; | |
440 | } | |
c29c0ae7 AT |
441 | return 0; |
442 | ||
443 | corrupted: | |
c398eda0 | 444 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 445 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 446 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 447 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
448 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
449 | max, le16_to_cpu(eh->eh_depth), depth); | |
450 | ||
451 | return -EIO; | |
452 | } | |
453 | ||
56b19868 | 454 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 455 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 456 | |
7a262f7c AK |
457 | int ext4_ext_check_inode(struct inode *inode) |
458 | { | |
459 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
460 | } | |
461 | ||
f8489128 DW |
462 | static int __ext4_ext_check_block(const char *function, unsigned int line, |
463 | struct inode *inode, | |
464 | struct ext4_extent_header *eh, | |
465 | int depth, | |
466 | struct buffer_head *bh) | |
467 | { | |
468 | int ret; | |
469 | ||
470 | if (buffer_verified(bh)) | |
471 | return 0; | |
472 | ret = ext4_ext_check(inode, eh, depth); | |
473 | if (ret) | |
474 | return ret; | |
475 | set_buffer_verified(bh); | |
476 | return ret; | |
477 | } | |
478 | ||
479 | #define ext4_ext_check_block(inode, eh, depth, bh) \ | |
480 | __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh) | |
481 | ||
a86c6181 AT |
482 | #ifdef EXT_DEBUG |
483 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
484 | { | |
485 | int k, l = path->p_depth; | |
486 | ||
487 | ext_debug("path:"); | |
488 | for (k = 0; k <= l; k++, path++) { | |
489 | if (path->p_idx) { | |
2ae02107 | 490 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 491 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 492 | } else if (path->p_ext) { |
553f9008 | 493 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 494 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 495 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 496 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 497 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
498 | } else |
499 | ext_debug(" []"); | |
500 | } | |
501 | ext_debug("\n"); | |
502 | } | |
503 | ||
504 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
505 | { | |
506 | int depth = ext_depth(inode); | |
507 | struct ext4_extent_header *eh; | |
508 | struct ext4_extent *ex; | |
509 | int i; | |
510 | ||
511 | if (!path) | |
512 | return; | |
513 | ||
514 | eh = path[depth].p_hdr; | |
515 | ex = EXT_FIRST_EXTENT(eh); | |
516 | ||
553f9008 M |
517 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
518 | ||
a86c6181 | 519 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
520 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
521 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 522 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
523 | } |
524 | ext_debug("\n"); | |
525 | } | |
1b16da77 YY |
526 | |
527 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
528 | ext4_fsblk_t newblock, int level) | |
529 | { | |
530 | int depth = ext_depth(inode); | |
531 | struct ext4_extent *ex; | |
532 | ||
533 | if (depth != level) { | |
534 | struct ext4_extent_idx *idx; | |
535 | idx = path[level].p_idx; | |
536 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
537 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
538 | le32_to_cpu(idx->ei_block), | |
539 | ext4_idx_pblock(idx), | |
540 | newblock); | |
541 | idx++; | |
542 | } | |
543 | ||
544 | return; | |
545 | } | |
546 | ||
547 | ex = path[depth].p_ext; | |
548 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
549 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
550 | le32_to_cpu(ex->ee_block), | |
551 | ext4_ext_pblock(ex), | |
552 | ext4_ext_is_uninitialized(ex), | |
553 | ext4_ext_get_actual_len(ex), | |
554 | newblock); | |
555 | ex++; | |
556 | } | |
557 | } | |
558 | ||
a86c6181 | 559 | #else |
af5bc92d TT |
560 | #define ext4_ext_show_path(inode, path) |
561 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 562 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
563 | #endif |
564 | ||
b35905c1 | 565 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
566 | { |
567 | int depth = path->p_depth; | |
568 | int i; | |
569 | ||
570 | for (i = 0; i <= depth; i++, path++) | |
571 | if (path->p_bh) { | |
572 | brelse(path->p_bh); | |
573 | path->p_bh = NULL; | |
574 | } | |
575 | } | |
576 | ||
577 | /* | |
d0d856e8 RD |
578 | * ext4_ext_binsearch_idx: |
579 | * binary search for the closest index of the given block | |
c29c0ae7 | 580 | * the header must be checked before calling this |
a86c6181 AT |
581 | */ |
582 | static void | |
725d26d3 AK |
583 | ext4_ext_binsearch_idx(struct inode *inode, |
584 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
585 | { |
586 | struct ext4_extent_header *eh = path->p_hdr; | |
587 | struct ext4_extent_idx *r, *l, *m; | |
588 | ||
a86c6181 | 589 | |
bba90743 | 590 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
591 | |
592 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 593 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
594 | while (l <= r) { |
595 | m = l + (r - l) / 2; | |
596 | if (block < le32_to_cpu(m->ei_block)) | |
597 | r = m - 1; | |
598 | else | |
599 | l = m + 1; | |
26d535ed DM |
600 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
601 | m, le32_to_cpu(m->ei_block), | |
602 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
603 | } |
604 | ||
605 | path->p_idx = l - 1; | |
4a3c3a51 | 606 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 607 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
608 | |
609 | #ifdef CHECK_BINSEARCH | |
610 | { | |
611 | struct ext4_extent_idx *chix, *ix; | |
612 | int k; | |
613 | ||
614 | chix = ix = EXT_FIRST_INDEX(eh); | |
615 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
616 | if (k != 0 && | |
617 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
618 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
619 | "first=0x%p\n", k, | |
620 | ix, EXT_FIRST_INDEX(eh)); | |
621 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
622 | le32_to_cpu(ix->ei_block), |
623 | le32_to_cpu(ix[-1].ei_block)); | |
624 | } | |
625 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 626 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
627 | if (block < le32_to_cpu(ix->ei_block)) |
628 | break; | |
629 | chix = ix; | |
630 | } | |
631 | BUG_ON(chix != path->p_idx); | |
632 | } | |
633 | #endif | |
634 | ||
635 | } | |
636 | ||
637 | /* | |
d0d856e8 RD |
638 | * ext4_ext_binsearch: |
639 | * binary search for closest extent of the given block | |
c29c0ae7 | 640 | * the header must be checked before calling this |
a86c6181 AT |
641 | */ |
642 | static void | |
725d26d3 AK |
643 | ext4_ext_binsearch(struct inode *inode, |
644 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
645 | { |
646 | struct ext4_extent_header *eh = path->p_hdr; | |
647 | struct ext4_extent *r, *l, *m; | |
648 | ||
a86c6181 AT |
649 | if (eh->eh_entries == 0) { |
650 | /* | |
d0d856e8 RD |
651 | * this leaf is empty: |
652 | * we get such a leaf in split/add case | |
a86c6181 AT |
653 | */ |
654 | return; | |
655 | } | |
656 | ||
bba90743 | 657 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
658 | |
659 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 660 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
661 | |
662 | while (l <= r) { | |
663 | m = l + (r - l) / 2; | |
664 | if (block < le32_to_cpu(m->ee_block)) | |
665 | r = m - 1; | |
666 | else | |
667 | l = m + 1; | |
26d535ed DM |
668 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
669 | m, le32_to_cpu(m->ee_block), | |
670 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
671 | } |
672 | ||
673 | path->p_ext = l - 1; | |
553f9008 | 674 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 675 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 676 | ext4_ext_pblock(path->p_ext), |
553f9008 | 677 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 678 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
679 | |
680 | #ifdef CHECK_BINSEARCH | |
681 | { | |
682 | struct ext4_extent *chex, *ex; | |
683 | int k; | |
684 | ||
685 | chex = ex = EXT_FIRST_EXTENT(eh); | |
686 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
687 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 688 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
689 | if (block < le32_to_cpu(ex->ee_block)) |
690 | break; | |
691 | chex = ex; | |
692 | } | |
693 | BUG_ON(chex != path->p_ext); | |
694 | } | |
695 | #endif | |
696 | ||
697 | } | |
698 | ||
699 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
700 | { | |
701 | struct ext4_extent_header *eh; | |
702 | ||
703 | eh = ext_inode_hdr(inode); | |
704 | eh->eh_depth = 0; | |
705 | eh->eh_entries = 0; | |
706 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 707 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
708 | ext4_mark_inode_dirty(handle, inode); |
709 | ext4_ext_invalidate_cache(inode); | |
710 | return 0; | |
711 | } | |
712 | ||
713 | struct ext4_ext_path * | |
725d26d3 AK |
714 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
715 | struct ext4_ext_path *path) | |
a86c6181 AT |
716 | { |
717 | struct ext4_extent_header *eh; | |
718 | struct buffer_head *bh; | |
719 | short int depth, i, ppos = 0, alloc = 0; | |
720 | ||
721 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 722 | depth = ext_depth(inode); |
a86c6181 AT |
723 | |
724 | /* account possible depth increase */ | |
725 | if (!path) { | |
5d4958f9 | 726 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
727 | GFP_NOFS); |
728 | if (!path) | |
729 | return ERR_PTR(-ENOMEM); | |
730 | alloc = 1; | |
731 | } | |
a86c6181 | 732 | path[0].p_hdr = eh; |
1973adcb | 733 | path[0].p_bh = NULL; |
a86c6181 | 734 | |
c29c0ae7 | 735 | i = depth; |
a86c6181 AT |
736 | /* walk through the tree */ |
737 | while (i) { | |
738 | ext_debug("depth %d: num %d, max %d\n", | |
739 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 740 | |
a86c6181 | 741 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 742 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
743 | path[ppos].p_depth = i; |
744 | path[ppos].p_ext = NULL; | |
745 | ||
7a262f7c AK |
746 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
747 | if (unlikely(!bh)) | |
a86c6181 | 748 | goto err; |
7a262f7c | 749 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
750 | trace_ext4_ext_load_extent(inode, block, |
751 | path[ppos].p_block); | |
7a262f7c AK |
752 | if (bh_submit_read(bh) < 0) { |
753 | put_bh(bh); | |
754 | goto err; | |
755 | } | |
7a262f7c | 756 | } |
a86c6181 AT |
757 | eh = ext_block_hdr(bh); |
758 | ppos++; | |
273df556 FM |
759 | if (unlikely(ppos > depth)) { |
760 | put_bh(bh); | |
761 | EXT4_ERROR_INODE(inode, | |
762 | "ppos %d > depth %d", ppos, depth); | |
763 | goto err; | |
764 | } | |
a86c6181 AT |
765 | path[ppos].p_bh = bh; |
766 | path[ppos].p_hdr = eh; | |
767 | i--; | |
768 | ||
f8489128 | 769 | if (ext4_ext_check_block(inode, eh, i, bh)) |
a86c6181 AT |
770 | goto err; |
771 | } | |
772 | ||
773 | path[ppos].p_depth = i; | |
a86c6181 AT |
774 | path[ppos].p_ext = NULL; |
775 | path[ppos].p_idx = NULL; | |
776 | ||
a86c6181 AT |
777 | /* find extent */ |
778 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
779 | /* if not an empty leaf */ |
780 | if (path[ppos].p_ext) | |
bf89d16f | 781 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
782 | |
783 | ext4_ext_show_path(inode, path); | |
784 | ||
785 | return path; | |
786 | ||
787 | err: | |
788 | ext4_ext_drop_refs(path); | |
789 | if (alloc) | |
790 | kfree(path); | |
791 | return ERR_PTR(-EIO); | |
792 | } | |
793 | ||
794 | /* | |
d0d856e8 RD |
795 | * ext4_ext_insert_index: |
796 | * insert new index [@logical;@ptr] into the block at @curp; | |
797 | * check where to insert: before @curp or after @curp | |
a86c6181 | 798 | */ |
1f109d5a TT |
799 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
800 | struct ext4_ext_path *curp, | |
801 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
802 | { |
803 | struct ext4_extent_idx *ix; | |
804 | int len, err; | |
805 | ||
7e028976 AM |
806 | err = ext4_ext_get_access(handle, inode, curp); |
807 | if (err) | |
a86c6181 AT |
808 | return err; |
809 | ||
273df556 FM |
810 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
811 | EXT4_ERROR_INODE(inode, | |
812 | "logical %d == ei_block %d!", | |
813 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
814 | return -EIO; | |
815 | } | |
d4620315 RD |
816 | |
817 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
818 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
819 | EXT4_ERROR_INODE(inode, | |
820 | "eh_entries %d >= eh_max %d!", | |
821 | le16_to_cpu(curp->p_hdr->eh_entries), | |
822 | le16_to_cpu(curp->p_hdr->eh_max)); | |
823 | return -EIO; | |
824 | } | |
825 | ||
a86c6181 AT |
826 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
827 | /* insert after */ | |
80e675f9 | 828 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
829 | ix = curp->p_idx + 1; |
830 | } else { | |
831 | /* insert before */ | |
80e675f9 | 832 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
833 | ix = curp->p_idx; |
834 | } | |
835 | ||
80e675f9 EG |
836 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
837 | BUG_ON(len < 0); | |
838 | if (len > 0) { | |
839 | ext_debug("insert new index %d: " | |
840 | "move %d indices from 0x%p to 0x%p\n", | |
841 | logical, len, ix, ix + 1); | |
842 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
843 | } | |
844 | ||
f472e026 TM |
845 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
846 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
847 | return -EIO; | |
848 | } | |
849 | ||
a86c6181 | 850 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 851 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 852 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 853 | |
273df556 FM |
854 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
855 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
856 | return -EIO; | |
857 | } | |
a86c6181 AT |
858 | |
859 | err = ext4_ext_dirty(handle, inode, curp); | |
860 | ext4_std_error(inode->i_sb, err); | |
861 | ||
862 | return err; | |
863 | } | |
864 | ||
865 | /* | |
d0d856e8 RD |
866 | * ext4_ext_split: |
867 | * inserts new subtree into the path, using free index entry | |
868 | * at depth @at: | |
869 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
870 | * - makes decision where to split | |
871 | * - moves remaining extents and index entries (right to the split point) | |
872 | * into the newly allocated blocks | |
873 | * - initializes subtree | |
a86c6181 AT |
874 | */ |
875 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
876 | unsigned int flags, |
877 | struct ext4_ext_path *path, | |
878 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
879 | { |
880 | struct buffer_head *bh = NULL; | |
881 | int depth = ext_depth(inode); | |
882 | struct ext4_extent_header *neh; | |
883 | struct ext4_extent_idx *fidx; | |
a86c6181 | 884 | int i = at, k, m, a; |
f65e6fba | 885 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 886 | __le32 border; |
f65e6fba | 887 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
888 | int err = 0; |
889 | ||
890 | /* make decision: where to split? */ | |
d0d856e8 | 891 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 892 | |
d0d856e8 | 893 | /* if current leaf will be split, then we should use |
a86c6181 | 894 | * border from split point */ |
273df556 FM |
895 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
896 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
897 | return -EIO; | |
898 | } | |
a86c6181 AT |
899 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
900 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 901 | ext_debug("leaf will be split." |
a86c6181 | 902 | " next leaf starts at %d\n", |
8c55e204 | 903 | le32_to_cpu(border)); |
a86c6181 AT |
904 | } else { |
905 | border = newext->ee_block; | |
906 | ext_debug("leaf will be added." | |
907 | " next leaf starts at %d\n", | |
8c55e204 | 908 | le32_to_cpu(border)); |
a86c6181 AT |
909 | } |
910 | ||
911 | /* | |
d0d856e8 RD |
912 | * If error occurs, then we break processing |
913 | * and mark filesystem read-only. index won't | |
a86c6181 | 914 | * be inserted and tree will be in consistent |
d0d856e8 | 915 | * state. Next mount will repair buffers too. |
a86c6181 AT |
916 | */ |
917 | ||
918 | /* | |
d0d856e8 RD |
919 | * Get array to track all allocated blocks. |
920 | * We need this to handle errors and free blocks | |
921 | * upon them. | |
a86c6181 | 922 | */ |
5d4958f9 | 923 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
924 | if (!ablocks) |
925 | return -ENOMEM; | |
a86c6181 AT |
926 | |
927 | /* allocate all needed blocks */ | |
928 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
929 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 930 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 931 | newext, &err, flags); |
a86c6181 AT |
932 | if (newblock == 0) |
933 | goto cleanup; | |
934 | ablocks[a] = newblock; | |
935 | } | |
936 | ||
937 | /* initialize new leaf */ | |
938 | newblock = ablocks[--a]; | |
273df556 FM |
939 | if (unlikely(newblock == 0)) { |
940 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
941 | err = -EIO; | |
942 | goto cleanup; | |
943 | } | |
a86c6181 AT |
944 | bh = sb_getblk(inode->i_sb, newblock); |
945 | if (!bh) { | |
946 | err = -EIO; | |
947 | goto cleanup; | |
948 | } | |
949 | lock_buffer(bh); | |
950 | ||
7e028976 AM |
951 | err = ext4_journal_get_create_access(handle, bh); |
952 | if (err) | |
a86c6181 AT |
953 | goto cleanup; |
954 | ||
955 | neh = ext_block_hdr(bh); | |
956 | neh->eh_entries = 0; | |
55ad63bf | 957 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
958 | neh->eh_magic = EXT4_EXT_MAGIC; |
959 | neh->eh_depth = 0; | |
a86c6181 | 960 | |
d0d856e8 | 961 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
962 | if (unlikely(path[depth].p_hdr->eh_entries != |
963 | path[depth].p_hdr->eh_max)) { | |
964 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
965 | path[depth].p_hdr->eh_entries, | |
966 | path[depth].p_hdr->eh_max); | |
967 | err = -EIO; | |
968 | goto cleanup; | |
969 | } | |
a86c6181 | 970 | /* start copy from next extent */ |
1b16da77 YY |
971 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
972 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 973 | if (m) { |
1b16da77 YY |
974 | struct ext4_extent *ex; |
975 | ex = EXT_FIRST_EXTENT(neh); | |
976 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 977 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
978 | } |
979 | ||
7ac5990d | 980 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
981 | set_buffer_uptodate(bh); |
982 | unlock_buffer(bh); | |
983 | ||
0390131b | 984 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 985 | if (err) |
a86c6181 AT |
986 | goto cleanup; |
987 | brelse(bh); | |
988 | bh = NULL; | |
989 | ||
990 | /* correct old leaf */ | |
991 | if (m) { | |
7e028976 AM |
992 | err = ext4_ext_get_access(handle, inode, path + depth); |
993 | if (err) | |
a86c6181 | 994 | goto cleanup; |
e8546d06 | 995 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
996 | err = ext4_ext_dirty(handle, inode, path + depth); |
997 | if (err) | |
a86c6181 AT |
998 | goto cleanup; |
999 | ||
1000 | } | |
1001 | ||
1002 | /* create intermediate indexes */ | |
1003 | k = depth - at - 1; | |
273df556 FM |
1004 | if (unlikely(k < 0)) { |
1005 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1006 | err = -EIO; | |
1007 | goto cleanup; | |
1008 | } | |
a86c6181 AT |
1009 | if (k) |
1010 | ext_debug("create %d intermediate indices\n", k); | |
1011 | /* insert new index into current index block */ | |
1012 | /* current depth stored in i var */ | |
1013 | i = depth - 1; | |
1014 | while (k--) { | |
1015 | oldblock = newblock; | |
1016 | newblock = ablocks[--a]; | |
bba90743 | 1017 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
1018 | if (!bh) { |
1019 | err = -EIO; | |
1020 | goto cleanup; | |
1021 | } | |
1022 | lock_buffer(bh); | |
1023 | ||
7e028976 AM |
1024 | err = ext4_journal_get_create_access(handle, bh); |
1025 | if (err) | |
a86c6181 AT |
1026 | goto cleanup; |
1027 | ||
1028 | neh = ext_block_hdr(bh); | |
1029 | neh->eh_entries = cpu_to_le16(1); | |
1030 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1031 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1032 | neh->eh_depth = cpu_to_le16(depth - i); |
1033 | fidx = EXT_FIRST_INDEX(neh); | |
1034 | fidx->ei_block = border; | |
f65e6fba | 1035 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1036 | |
bba90743 ES |
1037 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1038 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1039 | |
1b16da77 | 1040 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1041 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1042 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1043 | EXT4_ERROR_INODE(inode, | |
1044 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1045 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1046 | err = -EIO; | |
1047 | goto cleanup; | |
1048 | } | |
1b16da77 YY |
1049 | /* start copy indexes */ |
1050 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1051 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1052 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1053 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1054 | if (m) { |
1b16da77 | 1055 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1056 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1057 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1058 | } |
7ac5990d | 1059 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1060 | set_buffer_uptodate(bh); |
1061 | unlock_buffer(bh); | |
1062 | ||
0390131b | 1063 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1064 | if (err) |
a86c6181 AT |
1065 | goto cleanup; |
1066 | brelse(bh); | |
1067 | bh = NULL; | |
1068 | ||
1069 | /* correct old index */ | |
1070 | if (m) { | |
1071 | err = ext4_ext_get_access(handle, inode, path + i); | |
1072 | if (err) | |
1073 | goto cleanup; | |
e8546d06 | 1074 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1075 | err = ext4_ext_dirty(handle, inode, path + i); |
1076 | if (err) | |
1077 | goto cleanup; | |
1078 | } | |
1079 | ||
1080 | i--; | |
1081 | } | |
1082 | ||
1083 | /* insert new index */ | |
a86c6181 AT |
1084 | err = ext4_ext_insert_index(handle, inode, path + at, |
1085 | le32_to_cpu(border), newblock); | |
1086 | ||
1087 | cleanup: | |
1088 | if (bh) { | |
1089 | if (buffer_locked(bh)) | |
1090 | unlock_buffer(bh); | |
1091 | brelse(bh); | |
1092 | } | |
1093 | ||
1094 | if (err) { | |
1095 | /* free all allocated blocks in error case */ | |
1096 | for (i = 0; i < depth; i++) { | |
1097 | if (!ablocks[i]) | |
1098 | continue; | |
7dc57615 | 1099 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1100 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1101 | } |
1102 | } | |
1103 | kfree(ablocks); | |
1104 | ||
1105 | return err; | |
1106 | } | |
1107 | ||
1108 | /* | |
d0d856e8 RD |
1109 | * ext4_ext_grow_indepth: |
1110 | * implements tree growing procedure: | |
1111 | * - allocates new block | |
1112 | * - moves top-level data (index block or leaf) into the new block | |
1113 | * - initializes new top-level, creating index that points to the | |
1114 | * just created block | |
a86c6181 AT |
1115 | */ |
1116 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1117 | unsigned int flags, |
55f020db | 1118 | struct ext4_extent *newext) |
a86c6181 | 1119 | { |
a86c6181 | 1120 | struct ext4_extent_header *neh; |
a86c6181 | 1121 | struct buffer_head *bh; |
f65e6fba | 1122 | ext4_fsblk_t newblock; |
a86c6181 AT |
1123 | int err = 0; |
1124 | ||
1939dd84 | 1125 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1126 | newext, &err, flags); |
a86c6181 AT |
1127 | if (newblock == 0) |
1128 | return err; | |
1129 | ||
1130 | bh = sb_getblk(inode->i_sb, newblock); | |
1131 | if (!bh) { | |
1132 | err = -EIO; | |
1133 | ext4_std_error(inode->i_sb, err); | |
1134 | return err; | |
1135 | } | |
1136 | lock_buffer(bh); | |
1137 | ||
7e028976 AM |
1138 | err = ext4_journal_get_create_access(handle, bh); |
1139 | if (err) { | |
a86c6181 AT |
1140 | unlock_buffer(bh); |
1141 | goto out; | |
1142 | } | |
1143 | ||
1144 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1145 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1146 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1147 | |
1148 | /* set size of new block */ | |
1149 | neh = ext_block_hdr(bh); | |
1150 | /* old root could have indexes or leaves | |
1151 | * so calculate e_max right way */ | |
1152 | if (ext_depth(inode)) | |
55ad63bf | 1153 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1154 | else |
55ad63bf | 1155 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1156 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1157 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1158 | set_buffer_uptodate(bh); |
1159 | unlock_buffer(bh); | |
1160 | ||
0390131b | 1161 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1162 | if (err) |
a86c6181 AT |
1163 | goto out; |
1164 | ||
1939dd84 | 1165 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1166 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1167 | neh->eh_entries = cpu_to_le16(1); |
1168 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1169 | if (neh->eh_depth == 0) { | |
1170 | /* Root extent block becomes index block */ | |
1171 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1172 | EXT_FIRST_INDEX(neh)->ei_block = | |
1173 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1174 | } | |
2ae02107 | 1175 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1176 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1177 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1178 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1179 | |
b4611abf | 1180 | neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1); |
1939dd84 | 1181 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1182 | out: |
1183 | brelse(bh); | |
1184 | ||
1185 | return err; | |
1186 | } | |
1187 | ||
1188 | /* | |
d0d856e8 RD |
1189 | * ext4_ext_create_new_leaf: |
1190 | * finds empty index and adds new leaf. | |
1191 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1192 | */ |
1193 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1194 | unsigned int flags, |
1195 | struct ext4_ext_path *path, | |
1196 | struct ext4_extent *newext) | |
a86c6181 AT |
1197 | { |
1198 | struct ext4_ext_path *curp; | |
1199 | int depth, i, err = 0; | |
1200 | ||
1201 | repeat: | |
1202 | i = depth = ext_depth(inode); | |
1203 | ||
1204 | /* walk up to the tree and look for free index entry */ | |
1205 | curp = path + depth; | |
1206 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1207 | i--; | |
1208 | curp--; | |
1209 | } | |
1210 | ||
d0d856e8 RD |
1211 | /* we use already allocated block for index block, |
1212 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1213 | if (EXT_HAS_FREE_INDEX(curp)) { |
1214 | /* if we found index with free entry, then use that | |
1215 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1216 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1217 | if (err) |
1218 | goto out; | |
a86c6181 AT |
1219 | |
1220 | /* refill path */ | |
1221 | ext4_ext_drop_refs(path); | |
1222 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1223 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1224 | path); | |
a86c6181 AT |
1225 | if (IS_ERR(path)) |
1226 | err = PTR_ERR(path); | |
1227 | } else { | |
1228 | /* tree is full, time to grow in depth */ | |
1939dd84 | 1229 | err = ext4_ext_grow_indepth(handle, inode, flags, newext); |
a86c6181 AT |
1230 | if (err) |
1231 | goto out; | |
1232 | ||
1233 | /* refill path */ | |
1234 | ext4_ext_drop_refs(path); | |
1235 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1236 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1237 | path); | |
a86c6181 AT |
1238 | if (IS_ERR(path)) { |
1239 | err = PTR_ERR(path); | |
1240 | goto out; | |
1241 | } | |
1242 | ||
1243 | /* | |
d0d856e8 RD |
1244 | * only first (depth 0 -> 1) produces free space; |
1245 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1246 | */ |
1247 | depth = ext_depth(inode); | |
1248 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1249 | /* now we need to split */ |
a86c6181 AT |
1250 | goto repeat; |
1251 | } | |
1252 | } | |
1253 | ||
1254 | out: | |
1255 | return err; | |
1256 | } | |
1257 | ||
1988b51e AT |
1258 | /* |
1259 | * search the closest allocated block to the left for *logical | |
1260 | * and returns it at @logical + it's physical address at @phys | |
1261 | * if *logical is the smallest allocated block, the function | |
1262 | * returns 0 at @phys | |
1263 | * return value contains 0 (success) or error code | |
1264 | */ | |
1f109d5a TT |
1265 | static int ext4_ext_search_left(struct inode *inode, |
1266 | struct ext4_ext_path *path, | |
1267 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1268 | { |
1269 | struct ext4_extent_idx *ix; | |
1270 | struct ext4_extent *ex; | |
b939e376 | 1271 | int depth, ee_len; |
1988b51e | 1272 | |
273df556 FM |
1273 | if (unlikely(path == NULL)) { |
1274 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1275 | return -EIO; | |
1276 | } | |
1988b51e AT |
1277 | depth = path->p_depth; |
1278 | *phys = 0; | |
1279 | ||
1280 | if (depth == 0 && path->p_ext == NULL) | |
1281 | return 0; | |
1282 | ||
1283 | /* usually extent in the path covers blocks smaller | |
1284 | * then *logical, but it can be that extent is the | |
1285 | * first one in the file */ | |
1286 | ||
1287 | ex = path[depth].p_ext; | |
b939e376 | 1288 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1289 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1290 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1291 | EXT4_ERROR_INODE(inode, | |
1292 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1293 | *logical, le32_to_cpu(ex->ee_block)); | |
1294 | return -EIO; | |
1295 | } | |
1988b51e AT |
1296 | while (--depth >= 0) { |
1297 | ix = path[depth].p_idx; | |
273df556 FM |
1298 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1299 | EXT4_ERROR_INODE(inode, | |
1300 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1301 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1302 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1303 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1304 | depth); |
1305 | return -EIO; | |
1306 | } | |
1988b51e AT |
1307 | } |
1308 | return 0; | |
1309 | } | |
1310 | ||
273df556 FM |
1311 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1312 | EXT4_ERROR_INODE(inode, | |
1313 | "logical %d < ee_block %d + ee_len %d!", | |
1314 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1315 | return -EIO; | |
1316 | } | |
1988b51e | 1317 | |
b939e376 | 1318 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1319 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1320 | return 0; |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * search the closest allocated block to the right for *logical | |
1325 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1326 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1327 | * returns 0 at @phys |
1328 | * return value contains 0 (success) or error code | |
1329 | */ | |
1f109d5a TT |
1330 | static int ext4_ext_search_right(struct inode *inode, |
1331 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1332 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1333 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1334 | { |
1335 | struct buffer_head *bh = NULL; | |
1336 | struct ext4_extent_header *eh; | |
1337 | struct ext4_extent_idx *ix; | |
1338 | struct ext4_extent *ex; | |
1339 | ext4_fsblk_t block; | |
395a87bf ES |
1340 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1341 | int ee_len; | |
1988b51e | 1342 | |
273df556 FM |
1343 | if (unlikely(path == NULL)) { |
1344 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1345 | return -EIO; | |
1346 | } | |
1988b51e AT |
1347 | depth = path->p_depth; |
1348 | *phys = 0; | |
1349 | ||
1350 | if (depth == 0 && path->p_ext == NULL) | |
1351 | return 0; | |
1352 | ||
1353 | /* usually extent in the path covers blocks smaller | |
1354 | * then *logical, but it can be that extent is the | |
1355 | * first one in the file */ | |
1356 | ||
1357 | ex = path[depth].p_ext; | |
b939e376 | 1358 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1359 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1360 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1361 | EXT4_ERROR_INODE(inode, | |
1362 | "first_extent(path[%d].p_hdr) != ex", | |
1363 | depth); | |
1364 | return -EIO; | |
1365 | } | |
1988b51e AT |
1366 | while (--depth >= 0) { |
1367 | ix = path[depth].p_idx; | |
273df556 FM |
1368 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1369 | EXT4_ERROR_INODE(inode, | |
1370 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1371 | *logical); | |
1372 | return -EIO; | |
1373 | } | |
1988b51e | 1374 | } |
4d33b1ef | 1375 | goto found_extent; |
1988b51e AT |
1376 | } |
1377 | ||
273df556 FM |
1378 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1379 | EXT4_ERROR_INODE(inode, | |
1380 | "logical %d < ee_block %d + ee_len %d!", | |
1381 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1382 | return -EIO; | |
1383 | } | |
1988b51e AT |
1384 | |
1385 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1386 | /* next allocated block in this leaf */ | |
1387 | ex++; | |
4d33b1ef | 1388 | goto found_extent; |
1988b51e AT |
1389 | } |
1390 | ||
1391 | /* go up and search for index to the right */ | |
1392 | while (--depth >= 0) { | |
1393 | ix = path[depth].p_idx; | |
1394 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1395 | goto got_index; |
1988b51e AT |
1396 | } |
1397 | ||
25f1ee3a WF |
1398 | /* we've gone up to the root and found no index to the right */ |
1399 | return 0; | |
1988b51e | 1400 | |
25f1ee3a | 1401 | got_index: |
1988b51e AT |
1402 | /* we've found index to the right, let's |
1403 | * follow it and find the closest allocated | |
1404 | * block to the right */ | |
1405 | ix++; | |
bf89d16f | 1406 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1407 | while (++depth < path->p_depth) { |
1408 | bh = sb_bread(inode->i_sb, block); | |
1409 | if (bh == NULL) | |
1410 | return -EIO; | |
1411 | eh = ext_block_hdr(bh); | |
395a87bf | 1412 | /* subtract from p_depth to get proper eh_depth */ |
f8489128 DW |
1413 | if (ext4_ext_check_block(inode, eh, |
1414 | path->p_depth - depth, bh)) { | |
1988b51e AT |
1415 | put_bh(bh); |
1416 | return -EIO; | |
1417 | } | |
1418 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1419 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1420 | put_bh(bh); |
1421 | } | |
1422 | ||
1423 | bh = sb_bread(inode->i_sb, block); | |
1424 | if (bh == NULL) | |
1425 | return -EIO; | |
1426 | eh = ext_block_hdr(bh); | |
f8489128 | 1427 | if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) { |
1988b51e AT |
1428 | put_bh(bh); |
1429 | return -EIO; | |
1430 | } | |
1431 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1432 | found_extent: |
1988b51e | 1433 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1434 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1435 | *ret_ex = ex; |
1436 | if (bh) | |
1437 | put_bh(bh); | |
1988b51e | 1438 | return 0; |
1988b51e AT |
1439 | } |
1440 | ||
a86c6181 | 1441 | /* |
d0d856e8 | 1442 | * ext4_ext_next_allocated_block: |
f17722f9 | 1443 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1444 | * NOTE: it considers block number from index entry as |
1445 | * allocated block. Thus, index entries have to be consistent | |
1446 | * with leaves. | |
a86c6181 | 1447 | */ |
725d26d3 | 1448 | static ext4_lblk_t |
a86c6181 AT |
1449 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1450 | { | |
1451 | int depth; | |
1452 | ||
1453 | BUG_ON(path == NULL); | |
1454 | depth = path->p_depth; | |
1455 | ||
1456 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1457 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1458 | |
1459 | while (depth >= 0) { | |
1460 | if (depth == path->p_depth) { | |
1461 | /* leaf */ | |
6f8ff537 CW |
1462 | if (path[depth].p_ext && |
1463 | path[depth].p_ext != | |
a86c6181 AT |
1464 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1465 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1466 | } else { | |
1467 | /* index */ | |
1468 | if (path[depth].p_idx != | |
1469 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1470 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1471 | } | |
1472 | depth--; | |
1473 | } | |
1474 | ||
f17722f9 | 1475 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1476 | } |
1477 | ||
1478 | /* | |
d0d856e8 | 1479 | * ext4_ext_next_leaf_block: |
f17722f9 | 1480 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1481 | */ |
5718789d | 1482 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1483 | { |
1484 | int depth; | |
1485 | ||
1486 | BUG_ON(path == NULL); | |
1487 | depth = path->p_depth; | |
1488 | ||
1489 | /* zero-tree has no leaf blocks at all */ | |
1490 | if (depth == 0) | |
f17722f9 | 1491 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1492 | |
1493 | /* go to index block */ | |
1494 | depth--; | |
1495 | ||
1496 | while (depth >= 0) { | |
1497 | if (path[depth].p_idx != | |
1498 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1499 | return (ext4_lblk_t) |
1500 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1501 | depth--; |
1502 | } | |
1503 | ||
f17722f9 | 1504 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1505 | } |
1506 | ||
1507 | /* | |
d0d856e8 RD |
1508 | * ext4_ext_correct_indexes: |
1509 | * if leaf gets modified and modified extent is first in the leaf, | |
1510 | * then we have to correct all indexes above. | |
a86c6181 AT |
1511 | * TODO: do we need to correct tree in all cases? |
1512 | */ | |
1d03ec98 | 1513 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1514 | struct ext4_ext_path *path) |
1515 | { | |
1516 | struct ext4_extent_header *eh; | |
1517 | int depth = ext_depth(inode); | |
1518 | struct ext4_extent *ex; | |
1519 | __le32 border; | |
1520 | int k, err = 0; | |
1521 | ||
1522 | eh = path[depth].p_hdr; | |
1523 | ex = path[depth].p_ext; | |
273df556 FM |
1524 | |
1525 | if (unlikely(ex == NULL || eh == NULL)) { | |
1526 | EXT4_ERROR_INODE(inode, | |
1527 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1528 | return -EIO; | |
1529 | } | |
a86c6181 AT |
1530 | |
1531 | if (depth == 0) { | |
1532 | /* there is no tree at all */ | |
1533 | return 0; | |
1534 | } | |
1535 | ||
1536 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1537 | /* we correct tree if first leaf got modified only */ | |
1538 | return 0; | |
1539 | } | |
1540 | ||
1541 | /* | |
d0d856e8 | 1542 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1543 | */ |
1544 | k = depth - 1; | |
1545 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1546 | err = ext4_ext_get_access(handle, inode, path + k); |
1547 | if (err) | |
a86c6181 AT |
1548 | return err; |
1549 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1550 | err = ext4_ext_dirty(handle, inode, path + k); |
1551 | if (err) | |
a86c6181 AT |
1552 | return err; |
1553 | ||
1554 | while (k--) { | |
1555 | /* change all left-side indexes */ | |
1556 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1557 | break; | |
7e028976 AM |
1558 | err = ext4_ext_get_access(handle, inode, path + k); |
1559 | if (err) | |
a86c6181 AT |
1560 | break; |
1561 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1562 | err = ext4_ext_dirty(handle, inode, path + k); |
1563 | if (err) | |
a86c6181 AT |
1564 | break; |
1565 | } | |
1566 | ||
1567 | return err; | |
1568 | } | |
1569 | ||
748de673 | 1570 | int |
a86c6181 AT |
1571 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1572 | struct ext4_extent *ex2) | |
1573 | { | |
749269fa | 1574 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1575 | |
1576 | /* | |
1577 | * Make sure that either both extents are uninitialized, or | |
1578 | * both are _not_. | |
1579 | */ | |
1580 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1581 | return 0; | |
1582 | ||
749269fa AA |
1583 | if (ext4_ext_is_uninitialized(ex1)) |
1584 | max_len = EXT_UNINIT_MAX_LEN; | |
1585 | else | |
1586 | max_len = EXT_INIT_MAX_LEN; | |
1587 | ||
a2df2a63 AA |
1588 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1589 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1590 | ||
1591 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1592 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1593 | return 0; |
1594 | ||
471d4011 SB |
1595 | /* |
1596 | * To allow future support for preallocated extents to be added | |
1597 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1598 | * this can result in the top bit of ee_len being set. |
471d4011 | 1599 | */ |
749269fa | 1600 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1601 | return 0; |
bbf2f9fb | 1602 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1603 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1604 | return 0; |
1605 | #endif | |
1606 | ||
bf89d16f | 1607 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1608 | return 1; |
1609 | return 0; | |
1610 | } | |
1611 | ||
56055d3a AA |
1612 | /* |
1613 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1614 | * It always tries to merge towards right. If you want to merge towards | |
1615 | * left, pass "ex - 1" as argument instead of "ex". | |
1616 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1617 | * 1 if they got merged. | |
1618 | */ | |
197217a5 | 1619 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1620 | struct ext4_ext_path *path, |
1621 | struct ext4_extent *ex) | |
56055d3a AA |
1622 | { |
1623 | struct ext4_extent_header *eh; | |
1624 | unsigned int depth, len; | |
1625 | int merge_done = 0; | |
1626 | int uninitialized = 0; | |
1627 | ||
1628 | depth = ext_depth(inode); | |
1629 | BUG_ON(path[depth].p_hdr == NULL); | |
1630 | eh = path[depth].p_hdr; | |
1631 | ||
1632 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1633 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1634 | break; | |
1635 | /* merge with next extent! */ | |
1636 | if (ext4_ext_is_uninitialized(ex)) | |
1637 | uninitialized = 1; | |
1638 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1639 | + ext4_ext_get_actual_len(ex + 1)); | |
1640 | if (uninitialized) | |
1641 | ext4_ext_mark_uninitialized(ex); | |
1642 | ||
1643 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1644 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1645 | * sizeof(struct ext4_extent); | |
1646 | memmove(ex + 1, ex + 2, len); | |
1647 | } | |
e8546d06 | 1648 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1649 | merge_done = 1; |
1650 | WARN_ON(eh->eh_entries == 0); | |
1651 | if (!eh->eh_entries) | |
24676da4 | 1652 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1653 | } |
1654 | ||
1655 | return merge_done; | |
1656 | } | |
1657 | ||
197217a5 YY |
1658 | /* |
1659 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1660 | * return 1 if merge left else 0. | |
1661 | */ | |
1662 | static int ext4_ext_try_to_merge(struct inode *inode, | |
1663 | struct ext4_ext_path *path, | |
1664 | struct ext4_extent *ex) { | |
1665 | struct ext4_extent_header *eh; | |
1666 | unsigned int depth; | |
1667 | int merge_done = 0; | |
1668 | int ret = 0; | |
1669 | ||
1670 | depth = ext_depth(inode); | |
1671 | BUG_ON(path[depth].p_hdr == NULL); | |
1672 | eh = path[depth].p_hdr; | |
1673 | ||
1674 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1675 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1676 | ||
1677 | if (!merge_done) | |
1678 | ret = ext4_ext_try_to_merge_right(inode, path, ex); | |
1679 | ||
1680 | return ret; | |
1681 | } | |
1682 | ||
25d14f98 AA |
1683 | /* |
1684 | * check if a portion of the "newext" extent overlaps with an | |
1685 | * existing extent. | |
1686 | * | |
1687 | * If there is an overlap discovered, it updates the length of the newext | |
1688 | * such that there will be no overlap, and then returns 1. | |
1689 | * If there is no overlap found, it returns 0. | |
1690 | */ | |
4d33b1ef TT |
1691 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1692 | struct inode *inode, | |
1f109d5a TT |
1693 | struct ext4_extent *newext, |
1694 | struct ext4_ext_path *path) | |
25d14f98 | 1695 | { |
725d26d3 | 1696 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1697 | unsigned int depth, len1; |
1698 | unsigned int ret = 0; | |
1699 | ||
1700 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1701 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1702 | depth = ext_depth(inode); |
1703 | if (!path[depth].p_ext) | |
1704 | goto out; | |
1705 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1706 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1707 | |
1708 | /* | |
1709 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1710 | * is before the requested block(s) |
25d14f98 AA |
1711 | */ |
1712 | if (b2 < b1) { | |
1713 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1714 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1715 | goto out; |
4d33b1ef | 1716 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1717 | } |
1718 | ||
725d26d3 | 1719 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1720 | if (b1 + len1 < b1) { |
f17722f9 | 1721 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1722 | newext->ee_len = cpu_to_le16(len1); |
1723 | ret = 1; | |
1724 | } | |
1725 | ||
1726 | /* check for overlap */ | |
1727 | if (b1 + len1 > b2) { | |
1728 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1729 | ret = 1; | |
1730 | } | |
1731 | out: | |
1732 | return ret; | |
1733 | } | |
1734 | ||
a86c6181 | 1735 | /* |
d0d856e8 RD |
1736 | * ext4_ext_insert_extent: |
1737 | * tries to merge requsted extent into the existing extent or | |
1738 | * inserts requested extent as new one into the tree, | |
1739 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1740 | */ |
1741 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1742 | struct ext4_ext_path *path, | |
0031462b | 1743 | struct ext4_extent *newext, int flag) |
a86c6181 | 1744 | { |
af5bc92d | 1745 | struct ext4_extent_header *eh; |
a86c6181 AT |
1746 | struct ext4_extent *ex, *fex; |
1747 | struct ext4_extent *nearex; /* nearest extent */ | |
1748 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1749 | int depth, len, err; |
1750 | ext4_lblk_t next; | |
a2df2a63 | 1751 | unsigned uninitialized = 0; |
55f020db | 1752 | int flags = 0; |
a86c6181 | 1753 | |
273df556 FM |
1754 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1755 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1756 | return -EIO; | |
1757 | } | |
a86c6181 AT |
1758 | depth = ext_depth(inode); |
1759 | ex = path[depth].p_ext; | |
273df556 FM |
1760 | if (unlikely(path[depth].p_hdr == NULL)) { |
1761 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1762 | return -EIO; | |
1763 | } | |
a86c6181 AT |
1764 | |
1765 | /* try to insert block into found extent and return */ | |
744692dc | 1766 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1767 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
32de6756 | 1768 | ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n", |
bf89d16f TT |
1769 | ext4_ext_is_uninitialized(newext), |
1770 | ext4_ext_get_actual_len(newext), | |
1771 | le32_to_cpu(ex->ee_block), | |
1772 | ext4_ext_is_uninitialized(ex), | |
1773 | ext4_ext_get_actual_len(ex), | |
1774 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1775 | err = ext4_ext_get_access(handle, inode, path + depth); |
1776 | if (err) | |
a86c6181 | 1777 | return err; |
a2df2a63 AA |
1778 | |
1779 | /* | |
1780 | * ext4_can_extents_be_merged should have checked that either | |
1781 | * both extents are uninitialized, or both aren't. Thus we | |
1782 | * need to check only one of them here. | |
1783 | */ | |
1784 | if (ext4_ext_is_uninitialized(ex)) | |
1785 | uninitialized = 1; | |
1786 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1787 | + ext4_ext_get_actual_len(newext)); | |
1788 | if (uninitialized) | |
1789 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1790 | eh = path[depth].p_hdr; |
1791 | nearex = ex; | |
1792 | goto merge; | |
1793 | } | |
1794 | ||
a86c6181 AT |
1795 | depth = ext_depth(inode); |
1796 | eh = path[depth].p_hdr; | |
1797 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1798 | goto has_space; | |
1799 | ||
1800 | /* probably next leaf has space for us? */ | |
1801 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1802 | next = EXT_MAX_BLOCKS; |
1803 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1804 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1805 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1806 | ext_debug("next leaf block - %u\n", next); |
a86c6181 AT |
1807 | BUG_ON(npath != NULL); |
1808 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1809 | if (IS_ERR(npath)) | |
1810 | return PTR_ERR(npath); | |
1811 | BUG_ON(npath->p_depth != path->p_depth); | |
1812 | eh = npath[depth].p_hdr; | |
1813 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1814 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1815 | le16_to_cpu(eh->eh_entries)); |
1816 | path = npath; | |
ffb505ff | 1817 | goto has_space; |
a86c6181 AT |
1818 | } |
1819 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1820 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1821 | } | |
1822 | ||
1823 | /* | |
d0d856e8 RD |
1824 | * There is no free space in the found leaf. |
1825 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1826 | */ |
55f020db AH |
1827 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1828 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1829 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1830 | if (err) |
1831 | goto cleanup; | |
1832 | depth = ext_depth(inode); | |
1833 | eh = path[depth].p_hdr; | |
1834 | ||
1835 | has_space: | |
1836 | nearex = path[depth].p_ext; | |
1837 | ||
7e028976 AM |
1838 | err = ext4_ext_get_access(handle, inode, path + depth); |
1839 | if (err) | |
a86c6181 AT |
1840 | goto cleanup; |
1841 | ||
1842 | if (!nearex) { | |
1843 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1844 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1845 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1846 | ext4_ext_pblock(newext), |
553f9008 | 1847 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1848 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1849 | nearex = EXT_FIRST_EXTENT(eh); |
1850 | } else { | |
1851 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1852 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1853 | /* Insert after */ |
32de6756 YY |
1854 | ext_debug("insert %u:%llu:[%d]%d before: " |
1855 | "nearest %p\n", | |
80e675f9 EG |
1856 | le32_to_cpu(newext->ee_block), |
1857 | ext4_ext_pblock(newext), | |
1858 | ext4_ext_is_uninitialized(newext), | |
1859 | ext4_ext_get_actual_len(newext), | |
1860 | nearex); | |
1861 | nearex++; | |
1862 | } else { | |
1863 | /* Insert before */ | |
1864 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
1865 | ext_debug("insert %u:%llu:[%d]%d after: " |
1866 | "nearest %p\n", | |
8c55e204 | 1867 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1868 | ext4_ext_pblock(newext), |
553f9008 | 1869 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1870 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
1871 | nearex); |
1872 | } | |
1873 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
1874 | if (len > 0) { | |
32de6756 | 1875 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
1876 | "move %d extents from 0x%p to 0x%p\n", |
1877 | le32_to_cpu(newext->ee_block), | |
1878 | ext4_ext_pblock(newext), | |
1879 | ext4_ext_is_uninitialized(newext), | |
1880 | ext4_ext_get_actual_len(newext), | |
1881 | len, nearex, nearex + 1); | |
1882 | memmove(nearex + 1, nearex, | |
1883 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 1884 | } |
a86c6181 AT |
1885 | } |
1886 | ||
e8546d06 | 1887 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 1888 | path[depth].p_ext = nearex; |
a86c6181 | 1889 | nearex->ee_block = newext->ee_block; |
bf89d16f | 1890 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1891 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1892 | |
1893 | merge: | |
e7bcf823 | 1894 | /* try to merge extents */ |
744692dc | 1895 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
0031462b | 1896 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 | 1897 | |
a86c6181 AT |
1898 | |
1899 | /* time to correct all indexes above */ | |
1900 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1901 | if (err) | |
1902 | goto cleanup; | |
1903 | ||
1904 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1905 | ||
1906 | cleanup: | |
1907 | if (npath) { | |
1908 | ext4_ext_drop_refs(npath); | |
1909 | kfree(npath); | |
1910 | } | |
a86c6181 AT |
1911 | ext4_ext_invalidate_cache(inode); |
1912 | return err; | |
1913 | } | |
1914 | ||
1f109d5a TT |
1915 | static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1916 | ext4_lblk_t num, ext_prepare_callback func, | |
1917 | void *cbdata) | |
6873fa0d ES |
1918 | { |
1919 | struct ext4_ext_path *path = NULL; | |
1920 | struct ext4_ext_cache cbex; | |
1921 | struct ext4_extent *ex; | |
1922 | ext4_lblk_t next, start = 0, end = 0; | |
1923 | ext4_lblk_t last = block + num; | |
1924 | int depth, exists, err = 0; | |
1925 | ||
1926 | BUG_ON(func == NULL); | |
1927 | BUG_ON(inode == NULL); | |
1928 | ||
f17722f9 | 1929 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1930 | num = last - block; |
1931 | /* find extent for this block */ | |
fab3a549 | 1932 | down_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d | 1933 | path = ext4_ext_find_extent(inode, block, path); |
fab3a549 | 1934 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1935 | if (IS_ERR(path)) { |
1936 | err = PTR_ERR(path); | |
1937 | path = NULL; | |
1938 | break; | |
1939 | } | |
1940 | ||
1941 | depth = ext_depth(inode); | |
273df556 FM |
1942 | if (unlikely(path[depth].p_hdr == NULL)) { |
1943 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1944 | err = -EIO; | |
1945 | break; | |
1946 | } | |
6873fa0d ES |
1947 | ex = path[depth].p_ext; |
1948 | next = ext4_ext_next_allocated_block(path); | |
1949 | ||
1950 | exists = 0; | |
1951 | if (!ex) { | |
1952 | /* there is no extent yet, so try to allocate | |
1953 | * all requested space */ | |
1954 | start = block; | |
1955 | end = block + num; | |
1956 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1957 | /* need to allocate space before found extent */ | |
1958 | start = block; | |
1959 | end = le32_to_cpu(ex->ee_block); | |
1960 | if (block + num < end) | |
1961 | end = block + num; | |
1962 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1963 | + ext4_ext_get_actual_len(ex)) { | |
1964 | /* need to allocate space after found extent */ | |
1965 | start = block; | |
1966 | end = block + num; | |
1967 | if (end >= next) | |
1968 | end = next; | |
1969 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1970 | /* | |
1971 | * some part of requested space is covered | |
1972 | * by found extent | |
1973 | */ | |
1974 | start = block; | |
1975 | end = le32_to_cpu(ex->ee_block) | |
1976 | + ext4_ext_get_actual_len(ex); | |
1977 | if (block + num < end) | |
1978 | end = block + num; | |
1979 | exists = 1; | |
1980 | } else { | |
1981 | BUG(); | |
1982 | } | |
1983 | BUG_ON(end <= start); | |
1984 | ||
1985 | if (!exists) { | |
1986 | cbex.ec_block = start; | |
1987 | cbex.ec_len = end - start; | |
1988 | cbex.ec_start = 0; | |
6873fa0d ES |
1989 | } else { |
1990 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1991 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
bf89d16f | 1992 | cbex.ec_start = ext4_ext_pblock(ex); |
6873fa0d ES |
1993 | } |
1994 | ||
273df556 FM |
1995 | if (unlikely(cbex.ec_len == 0)) { |
1996 | EXT4_ERROR_INODE(inode, "cbex.ec_len == 0"); | |
1997 | err = -EIO; | |
1998 | break; | |
1999 | } | |
c03f8aa9 | 2000 | err = func(inode, next, &cbex, ex, cbdata); |
6873fa0d ES |
2001 | ext4_ext_drop_refs(path); |
2002 | ||
2003 | if (err < 0) | |
2004 | break; | |
2005 | ||
2006 | if (err == EXT_REPEAT) | |
2007 | continue; | |
2008 | else if (err == EXT_BREAK) { | |
2009 | err = 0; | |
2010 | break; | |
2011 | } | |
2012 | ||
2013 | if (ext_depth(inode) != depth) { | |
2014 | /* depth was changed. we have to realloc path */ | |
2015 | kfree(path); | |
2016 | path = NULL; | |
2017 | } | |
2018 | ||
2019 | block = cbex.ec_block + cbex.ec_len; | |
2020 | } | |
2021 | ||
2022 | if (path) { | |
2023 | ext4_ext_drop_refs(path); | |
2024 | kfree(path); | |
2025 | } | |
2026 | ||
2027 | return err; | |
2028 | } | |
2029 | ||
09b88252 | 2030 | static void |
725d26d3 | 2031 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 2032 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
2033 | { |
2034 | struct ext4_ext_cache *cex; | |
2035 | BUG_ON(len == 0); | |
2ec0ae3a | 2036 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
d8990240 | 2037 | trace_ext4_ext_put_in_cache(inode, block, len, start); |
a86c6181 | 2038 | cex = &EXT4_I(inode)->i_cached_extent; |
a86c6181 AT |
2039 | cex->ec_block = block; |
2040 | cex->ec_len = len; | |
2041 | cex->ec_start = start; | |
2ec0ae3a | 2042 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
2043 | } |
2044 | ||
2045 | /* | |
d0d856e8 RD |
2046 | * ext4_ext_put_gap_in_cache: |
2047 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2048 | * and cache this gap |
2049 | */ | |
09b88252 | 2050 | static void |
a86c6181 | 2051 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2052 | ext4_lblk_t block) |
a86c6181 AT |
2053 | { |
2054 | int depth = ext_depth(inode); | |
725d26d3 AK |
2055 | unsigned long len; |
2056 | ext4_lblk_t lblock; | |
a86c6181 AT |
2057 | struct ext4_extent *ex; |
2058 | ||
2059 | ex = path[depth].p_ext; | |
2060 | if (ex == NULL) { | |
2061 | /* there is no extent yet, so gap is [0;-] */ | |
2062 | lblock = 0; | |
f17722f9 | 2063 | len = EXT_MAX_BLOCKS; |
a86c6181 AT |
2064 | ext_debug("cache gap(whole file):"); |
2065 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2066 | lblock = block; | |
2067 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2068 | ext_debug("cache gap(before): %u [%u:%u]", |
2069 | block, | |
2070 | le32_to_cpu(ex->ee_block), | |
2071 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 2072 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2073 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2074 | ext4_lblk_t next; |
8c55e204 | 2075 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2076 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2077 | |
2078 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2079 | ext_debug("cache gap(after): [%u:%u] %u", |
2080 | le32_to_cpu(ex->ee_block), | |
2081 | ext4_ext_get_actual_len(ex), | |
2082 | block); | |
725d26d3 AK |
2083 | BUG_ON(next == lblock); |
2084 | len = next - lblock; | |
a86c6181 AT |
2085 | } else { |
2086 | lblock = len = 0; | |
2087 | BUG(); | |
2088 | } | |
2089 | ||
bba90743 | 2090 | ext_debug(" -> %u:%lu\n", lblock, len); |
b05e6ae5 | 2091 | ext4_ext_put_in_cache(inode, lblock, len, 0); |
a86c6181 AT |
2092 | } |
2093 | ||
b05e6ae5 | 2094 | /* |
b7ca1e8e | 2095 | * ext4_ext_check_cache() |
a4bb6b64 AH |
2096 | * Checks to see if the given block is in the cache. |
2097 | * If it is, the cached extent is stored in the given | |
2098 | * cache extent pointer. If the cached extent is a hole, | |
2099 | * this routine should be used instead of | |
2100 | * ext4_ext_in_cache if the calling function needs to | |
2101 | * know the size of the hole. | |
2102 | * | |
2103 | * @inode: The files inode | |
2104 | * @block: The block to look for in the cache | |
2105 | * @ex: Pointer where the cached extent will be stored | |
2106 | * if it contains block | |
2107 | * | |
b05e6ae5 TT |
2108 | * Return 0 if cache is invalid; 1 if the cache is valid |
2109 | */ | |
a4bb6b64 AH |
2110 | static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block, |
2111 | struct ext4_ext_cache *ex){ | |
a86c6181 | 2112 | struct ext4_ext_cache *cex; |
77f4135f | 2113 | struct ext4_sb_info *sbi; |
b05e6ae5 | 2114 | int ret = 0; |
a86c6181 | 2115 | |
60e6679e | 2116 | /* |
2ec0ae3a TT |
2117 | * We borrow i_block_reservation_lock to protect i_cached_extent |
2118 | */ | |
2119 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
a86c6181 | 2120 | cex = &EXT4_I(inode)->i_cached_extent; |
77f4135f | 2121 | sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2122 | |
2123 | /* has cache valid data? */ | |
b05e6ae5 | 2124 | if (cex->ec_len == 0) |
2ec0ae3a | 2125 | goto errout; |
a86c6181 | 2126 | |
731eb1a0 | 2127 | if (in_range(block, cex->ec_block, cex->ec_len)) { |
a4bb6b64 | 2128 | memcpy(ex, cex, sizeof(struct ext4_ext_cache)); |
bba90743 ES |
2129 | ext_debug("%u cached by %u:%u:%llu\n", |
2130 | block, | |
2131 | cex->ec_block, cex->ec_len, cex->ec_start); | |
b05e6ae5 | 2132 | ret = 1; |
a86c6181 | 2133 | } |
2ec0ae3a | 2134 | errout: |
d8990240 | 2135 | trace_ext4_ext_in_cache(inode, block, ret); |
2ec0ae3a TT |
2136 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2137 | return ret; | |
a86c6181 AT |
2138 | } |
2139 | ||
a4bb6b64 AH |
2140 | /* |
2141 | * ext4_ext_in_cache() | |
2142 | * Checks to see if the given block is in the cache. | |
2143 | * If it is, the cached extent is stored in the given | |
2144 | * extent pointer. | |
2145 | * | |
2146 | * @inode: The files inode | |
2147 | * @block: The block to look for in the cache | |
2148 | * @ex: Pointer where the cached extent will be stored | |
2149 | * if it contains block | |
2150 | * | |
2151 | * Return 0 if cache is invalid; 1 if the cache is valid | |
2152 | */ | |
2153 | static int | |
2154 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, | |
2155 | struct ext4_extent *ex) | |
2156 | { | |
2157 | struct ext4_ext_cache cex; | |
2158 | int ret = 0; | |
2159 | ||
2160 | if (ext4_ext_check_cache(inode, block, &cex)) { | |
2161 | ex->ee_block = cpu_to_le32(cex.ec_block); | |
2162 | ext4_ext_store_pblock(ex, cex.ec_start); | |
2163 | ex->ee_len = cpu_to_le16(cex.ec_len); | |
2164 | ret = 1; | |
2165 | } | |
2166 | ||
2167 | return ret; | |
2168 | } | |
2169 | ||
2170 | ||
a86c6181 | 2171 | /* |
d0d856e8 RD |
2172 | * ext4_ext_rm_idx: |
2173 | * removes index from the index block. | |
a86c6181 | 2174 | */ |
1d03ec98 | 2175 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
2176 | struct ext4_ext_path *path) |
2177 | { | |
a86c6181 | 2178 | int err; |
f65e6fba | 2179 | ext4_fsblk_t leaf; |
a86c6181 AT |
2180 | |
2181 | /* free index block */ | |
2182 | path--; | |
bf89d16f | 2183 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2184 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2185 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2186 | return -EIO; | |
2187 | } | |
7e028976 AM |
2188 | err = ext4_ext_get_access(handle, inode, path); |
2189 | if (err) | |
a86c6181 | 2190 | return err; |
0e1147b0 RD |
2191 | |
2192 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2193 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2194 | len *= sizeof(struct ext4_extent_idx); | |
2195 | memmove(path->p_idx, path->p_idx + 1, len); | |
2196 | } | |
2197 | ||
e8546d06 | 2198 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2199 | err = ext4_ext_dirty(handle, inode, path); |
2200 | if (err) | |
a86c6181 | 2201 | return err; |
2ae02107 | 2202 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2203 | trace_ext4_ext_rm_idx(inode, leaf); |
2204 | ||
7dc57615 | 2205 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2206 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
a86c6181 AT |
2207 | return err; |
2208 | } | |
2209 | ||
2210 | /* | |
ee12b630 MC |
2211 | * ext4_ext_calc_credits_for_single_extent: |
2212 | * This routine returns max. credits that needed to insert an extent | |
2213 | * to the extent tree. | |
2214 | * When pass the actual path, the caller should calculate credits | |
2215 | * under i_data_sem. | |
a86c6181 | 2216 | */ |
525f4ed8 | 2217 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2218 | struct ext4_ext_path *path) |
2219 | { | |
a86c6181 | 2220 | if (path) { |
ee12b630 | 2221 | int depth = ext_depth(inode); |
f3bd1f3f | 2222 | int ret = 0; |
ee12b630 | 2223 | |
a86c6181 | 2224 | /* probably there is space in leaf? */ |
a86c6181 | 2225 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2226 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2227 | |
ee12b630 MC |
2228 | /* |
2229 | * There are some space in the leaf tree, no | |
2230 | * need to account for leaf block credit | |
2231 | * | |
2232 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2233 | * and other metadata blocks still need to be |
ee12b630 MC |
2234 | * accounted. |
2235 | */ | |
525f4ed8 | 2236 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2237 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2238 | return ret; |
ee12b630 MC |
2239 | } |
2240 | } | |
a86c6181 | 2241 | |
525f4ed8 | 2242 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2243 | } |
a86c6181 | 2244 | |
ee12b630 MC |
2245 | /* |
2246 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2247 | * | |
2248 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2249 | * in the worse case, each tree level index/leaf need to be changed | |
2250 | * if the tree split due to insert a new extent, then the old tree | |
2251 | * index/leaf need to be updated too | |
2252 | * | |
2253 | * If the nrblocks are discontiguous, they could cause | |
2254 | * the whole tree split more than once, but this is really rare. | |
2255 | */ | |
525f4ed8 | 2256 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2257 | { |
2258 | int index; | |
2259 | int depth = ext_depth(inode); | |
a86c6181 | 2260 | |
ee12b630 MC |
2261 | if (chunk) |
2262 | index = depth * 2; | |
2263 | else | |
2264 | index = depth * 3; | |
a86c6181 | 2265 | |
ee12b630 | 2266 | return index; |
a86c6181 AT |
2267 | } |
2268 | ||
2269 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2270 | struct ext4_extent *ex, |
2271 | ext4_fsblk_t *partial_cluster, | |
2272 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2273 | { |
0aa06000 | 2274 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2275 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2276 | ext4_fsblk_t pblk; |
e6362609 | 2277 | int flags = EXT4_FREE_BLOCKS_FORGET; |
a86c6181 | 2278 | |
c9de560d | 2279 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
e6362609 | 2280 | flags |= EXT4_FREE_BLOCKS_METADATA; |
0aa06000 TT |
2281 | /* |
2282 | * For bigalloc file systems, we never free a partial cluster | |
2283 | * at the beginning of the extent. Instead, we make a note | |
2284 | * that we tried freeing the cluster, and check to see if we | |
2285 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2286 | * or at the end of the ext4_truncate() operation. | |
2287 | */ | |
2288 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2289 | ||
d8990240 | 2290 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2291 | /* |
2292 | * If we have a partial cluster, and it's different from the | |
2293 | * cluster of the last block, we need to explicitly free the | |
2294 | * partial cluster here. | |
2295 | */ | |
2296 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2297 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2298 | ext4_free_blocks(handle, inode, NULL, | |
2299 | EXT4_C2B(sbi, *partial_cluster), | |
2300 | sbi->s_cluster_ratio, flags); | |
2301 | *partial_cluster = 0; | |
2302 | } | |
2303 | ||
a86c6181 AT |
2304 | #ifdef EXTENTS_STATS |
2305 | { | |
2306 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2307 | spin_lock(&sbi->s_ext_stats_lock); |
2308 | sbi->s_ext_blocks += ee_len; | |
2309 | sbi->s_ext_extents++; | |
2310 | if (ee_len < sbi->s_ext_min) | |
2311 | sbi->s_ext_min = ee_len; | |
2312 | if (ee_len > sbi->s_ext_max) | |
2313 | sbi->s_ext_max = ee_len; | |
2314 | if (ext_depth(inode) > sbi->s_depth_max) | |
2315 | sbi->s_depth_max = ext_depth(inode); | |
2316 | spin_unlock(&sbi->s_ext_stats_lock); | |
2317 | } | |
2318 | #endif | |
2319 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2320 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2321 | /* tail removal */ |
725d26d3 | 2322 | ext4_lblk_t num; |
725d26d3 | 2323 | |
a2df2a63 | 2324 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2325 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2326 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2327 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2328 | /* | |
2329 | * If the block range to be freed didn't start at the | |
2330 | * beginning of a cluster, and we removed the entire | |
2331 | * extent, save the partial cluster here, since we | |
2332 | * might need to delete if we determine that the | |
2333 | * truncate operation has removed all of the blocks in | |
2334 | * the cluster. | |
2335 | */ | |
2336 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2337 | (ee_len == num)) | |
2338 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2339 | else | |
2340 | *partial_cluster = 0; | |
a86c6181 | 2341 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2342 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2343 | /* head removal */ |
2344 | ext4_lblk_t num; | |
2345 | ext4_fsblk_t start; | |
2346 | ||
2347 | num = to - from; | |
2348 | start = ext4_ext_pblock(ex); | |
2349 | ||
2350 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
ee90d57e | 2351 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
d583fb87 | 2352 | |
a86c6181 | 2353 | } else { |
725d26d3 AK |
2354 | printk(KERN_INFO "strange request: removal(2) " |
2355 | "%u-%u from %u:%u\n", | |
2356 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2357 | } |
2358 | return 0; | |
2359 | } | |
2360 | ||
d583fb87 AH |
2361 | |
2362 | /* | |
2363 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2364 | * blocks appearing between "start" and "end", and splits the extents | |
2365 | * if "start" and "end" appear in the same extent | |
2366 | * | |
2367 | * @handle: The journal handle | |
2368 | * @inode: The files inode | |
2369 | * @path: The path to the leaf | |
2370 | * @start: The first block to remove | |
2371 | * @end: The last block to remove | |
2372 | */ | |
a86c6181 AT |
2373 | static int |
2374 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2375 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2376 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2377 | { |
0aa06000 | 2378 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2379 | int err = 0, correct_index = 0; |
2380 | int depth = ext_depth(inode), credits; | |
2381 | struct ext4_extent_header *eh; | |
750c9c47 | 2382 | ext4_lblk_t a, b; |
725d26d3 AK |
2383 | unsigned num; |
2384 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2385 | unsigned short ex_ee_len; |
a2df2a63 | 2386 | unsigned uninitialized = 0; |
a86c6181 AT |
2387 | struct ext4_extent *ex; |
2388 | ||
c29c0ae7 | 2389 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2390 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2391 | if (!path[depth].p_hdr) |
2392 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2393 | eh = path[depth].p_hdr; | |
273df556 FM |
2394 | if (unlikely(path[depth].p_hdr == NULL)) { |
2395 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2396 | return -EIO; | |
2397 | } | |
a86c6181 AT |
2398 | /* find where to start removing */ |
2399 | ex = EXT_LAST_EXTENT(eh); | |
2400 | ||
2401 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2402 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2403 | |
d8990240 AK |
2404 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2405 | ||
a86c6181 AT |
2406 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2407 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2408 | |
2409 | if (ext4_ext_is_uninitialized(ex)) | |
2410 | uninitialized = 1; | |
2411 | else | |
2412 | uninitialized = 0; | |
2413 | ||
553f9008 M |
2414 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2415 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2416 | path[depth].p_ext = ex; |
2417 | ||
2418 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2419 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2420 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2421 | |
2422 | ext_debug(" border %u:%u\n", a, b); | |
2423 | ||
d583fb87 | 2424 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2425 | if (end < ex_ee_block) { |
d583fb87 AH |
2426 | ex--; |
2427 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2428 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2429 | continue; | |
750c9c47 | 2430 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2431 | EXT4_ERROR_INODE(inode, |
2432 | "can not handle truncate %u:%u " | |
2433 | "on extent %u:%u", | |
2434 | start, end, ex_ee_block, | |
2435 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2436 | err = -EIO; |
2437 | goto out; | |
a86c6181 AT |
2438 | } else if (a != ex_ee_block) { |
2439 | /* remove tail of the extent */ | |
750c9c47 | 2440 | num = a - ex_ee_block; |
a86c6181 AT |
2441 | } else { |
2442 | /* remove whole extent: excellent! */ | |
a86c6181 | 2443 | num = 0; |
a86c6181 | 2444 | } |
34071da7 TT |
2445 | /* |
2446 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2447 | * descriptor) for each block group; assume two block | |
2448 | * groups plus ex_ee_len/blocks_per_block_group for | |
2449 | * the worst case | |
2450 | */ | |
2451 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2452 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2453 | correct_index = 1; | |
2454 | credits += (ext_depth(inode)) + 1; | |
2455 | } | |
5aca07eb | 2456 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2457 | |
487caeef | 2458 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2459 | if (err) |
a86c6181 | 2460 | goto out; |
a86c6181 AT |
2461 | |
2462 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2463 | if (err) | |
2464 | goto out; | |
2465 | ||
0aa06000 TT |
2466 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2467 | a, b); | |
a86c6181 AT |
2468 | if (err) |
2469 | goto out; | |
2470 | ||
750c9c47 | 2471 | if (num == 0) |
d0d856e8 | 2472 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2473 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2474 | |
a86c6181 | 2475 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2476 | /* |
2477 | * Do not mark uninitialized if all the blocks in the | |
2478 | * extent have been removed. | |
2479 | */ | |
2480 | if (uninitialized && num) | |
a2df2a63 | 2481 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2482 | /* |
2483 | * If the extent was completely released, | |
2484 | * we need to remove it from the leaf | |
2485 | */ | |
2486 | if (num == 0) { | |
f17722f9 | 2487 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2488 | /* |
2489 | * For hole punching, we need to scoot all the | |
2490 | * extents up when an extent is removed so that | |
2491 | * we dont have blank extents in the middle | |
2492 | */ | |
2493 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2494 | sizeof(struct ext4_extent)); | |
2495 | ||
2496 | /* Now get rid of the one at the end */ | |
2497 | memset(EXT_LAST_EXTENT(eh), 0, | |
2498 | sizeof(struct ext4_extent)); | |
2499 | } | |
2500 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2501 | } else |
2502 | *partial_cluster = 0; | |
d583fb87 | 2503 | |
750c9c47 DM |
2504 | err = ext4_ext_dirty(handle, inode, path + depth); |
2505 | if (err) | |
2506 | goto out; | |
2507 | ||
bf52c6f7 | 2508 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2509 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2510 | ex--; |
2511 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2512 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2513 | } |
2514 | ||
2515 | if (correct_index && eh->eh_entries) | |
2516 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2517 | ||
0aa06000 TT |
2518 | /* |
2519 | * If there is still a entry in the leaf node, check to see if | |
2520 | * it references the partial cluster. This is the only place | |
2521 | * where it could; if it doesn't, we can free the cluster. | |
2522 | */ | |
2523 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2524 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2525 | *partial_cluster)) { | |
2526 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2527 | ||
2528 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2529 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2530 | ||
2531 | ext4_free_blocks(handle, inode, NULL, | |
2532 | EXT4_C2B(sbi, *partial_cluster), | |
2533 | sbi->s_cluster_ratio, flags); | |
2534 | *partial_cluster = 0; | |
2535 | } | |
2536 | ||
a86c6181 AT |
2537 | /* if this leaf is free, then we should |
2538 | * remove it from index block above */ | |
2539 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2540 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2541 | ||
2542 | out: | |
2543 | return err; | |
2544 | } | |
2545 | ||
2546 | /* | |
d0d856e8 RD |
2547 | * ext4_ext_more_to_rm: |
2548 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2549 | */ |
09b88252 | 2550 | static int |
a86c6181 AT |
2551 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2552 | { | |
2553 | BUG_ON(path->p_idx == NULL); | |
2554 | ||
2555 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2556 | return 0; | |
2557 | ||
2558 | /* | |
d0d856e8 | 2559 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2560 | * so we have to consider current index for truncation |
2561 | */ | |
2562 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2563 | return 0; | |
2564 | return 1; | |
2565 | } | |
2566 | ||
5f95d21f LC |
2567 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2568 | ext4_lblk_t end) | |
a86c6181 AT |
2569 | { |
2570 | struct super_block *sb = inode->i_sb; | |
2571 | int depth = ext_depth(inode); | |
968dee77 | 2572 | struct ext4_ext_path *path = NULL; |
0aa06000 | 2573 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2574 | handle_t *handle; |
968dee77 | 2575 | int i = 0, err; |
a86c6181 | 2576 | |
5f95d21f | 2577 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2578 | |
2579 | /* probably first extent we're gonna free will be last in block */ | |
2580 | handle = ext4_journal_start(inode, depth + 1); | |
2581 | if (IS_ERR(handle)) | |
2582 | return PTR_ERR(handle); | |
2583 | ||
0617b83f | 2584 | again: |
a86c6181 AT |
2585 | ext4_ext_invalidate_cache(inode); |
2586 | ||
d8990240 AK |
2587 | trace_ext4_ext_remove_space(inode, start, depth); |
2588 | ||
5f95d21f LC |
2589 | /* |
2590 | * Check if we are removing extents inside the extent tree. If that | |
2591 | * is the case, we are going to punch a hole inside the extent tree | |
2592 | * so we have to check whether we need to split the extent covering | |
2593 | * the last block to remove so we can easily remove the part of it | |
2594 | * in ext4_ext_rm_leaf(). | |
2595 | */ | |
2596 | if (end < EXT_MAX_BLOCKS - 1) { | |
2597 | struct ext4_extent *ex; | |
2598 | ext4_lblk_t ee_block; | |
2599 | ||
2600 | /* find extent for this block */ | |
2601 | path = ext4_ext_find_extent(inode, end, NULL); | |
2602 | if (IS_ERR(path)) { | |
2603 | ext4_journal_stop(handle); | |
2604 | return PTR_ERR(path); | |
2605 | } | |
2606 | depth = ext_depth(inode); | |
2607 | ex = path[depth].p_ext; | |
968dee77 AS |
2608 | if (!ex) { |
2609 | ext4_ext_drop_refs(path); | |
2610 | kfree(path); | |
2611 | path = NULL; | |
5f95d21f | 2612 | goto cont; |
968dee77 | 2613 | } |
5f95d21f LC |
2614 | |
2615 | ee_block = le32_to_cpu(ex->ee_block); | |
2616 | ||
2617 | /* | |
2618 | * See if the last block is inside the extent, if so split | |
2619 | * the extent at 'end' block so we can easily remove the | |
2620 | * tail of the first part of the split extent in | |
2621 | * ext4_ext_rm_leaf(). | |
2622 | */ | |
2623 | if (end >= ee_block && | |
2624 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2625 | int split_flag = 0; | |
2626 | ||
2627 | if (ext4_ext_is_uninitialized(ex)) | |
2628 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2629 | EXT4_EXT_MARK_UNINIT2; | |
2630 | ||
2631 | /* | |
2632 | * Split the extent in two so that 'end' is the last | |
2633 | * block in the first new extent | |
2634 | */ | |
2635 | err = ext4_split_extent_at(handle, inode, path, | |
2636 | end + 1, split_flag, | |
2637 | EXT4_GET_BLOCKS_PRE_IO | | |
2638 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
2639 | ||
2640 | if (err < 0) | |
2641 | goto out; | |
2642 | } | |
5f95d21f LC |
2643 | } |
2644 | cont: | |
2645 | ||
a86c6181 | 2646 | /* |
d0d856e8 RD |
2647 | * We start scanning from right side, freeing all the blocks |
2648 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2649 | */ |
0617b83f | 2650 | depth = ext_depth(inode); |
968dee77 AS |
2651 | if (path) { |
2652 | int k = i = depth; | |
2653 | while (--k > 0) | |
2654 | path[k].p_block = | |
2655 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2656 | } else { | |
2657 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2658 | GFP_NOFS); | |
2659 | if (path == NULL) { | |
2660 | ext4_journal_stop(handle); | |
2661 | return -ENOMEM; | |
2662 | } | |
2663 | path[0].p_depth = depth; | |
2664 | path[0].p_hdr = ext_inode_hdr(inode); | |
5f95d21f | 2665 | |
968dee77 AS |
2666 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
2667 | err = -EIO; | |
2668 | goto out; | |
2669 | } | |
a86c6181 | 2670 | } |
968dee77 | 2671 | err = 0; |
a86c6181 AT |
2672 | |
2673 | while (i >= 0 && err == 0) { | |
2674 | if (i == depth) { | |
2675 | /* this is leaf block */ | |
d583fb87 | 2676 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2677 | &partial_cluster, start, |
5f95d21f | 2678 | end); |
d0d856e8 | 2679 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2680 | brelse(path[i].p_bh); |
2681 | path[i].p_bh = NULL; | |
2682 | i--; | |
2683 | continue; | |
2684 | } | |
2685 | ||
2686 | /* this is index block */ | |
2687 | if (!path[i].p_hdr) { | |
2688 | ext_debug("initialize header\n"); | |
2689 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2690 | } |
2691 | ||
a86c6181 | 2692 | if (!path[i].p_idx) { |
d0d856e8 | 2693 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2694 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2695 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2696 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2697 | path[i].p_hdr, | |
2698 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2699 | } else { | |
d0d856e8 | 2700 | /* we were already here, see at next index */ |
a86c6181 AT |
2701 | path[i].p_idx--; |
2702 | } | |
2703 | ||
2704 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2705 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2706 | path[i].p_idx); | |
2707 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2708 | struct buffer_head *bh; |
a86c6181 | 2709 | /* go to the next level */ |
2ae02107 | 2710 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2711 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2712 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2713 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2714 | if (!bh) { |
a86c6181 AT |
2715 | /* should we reset i_size? */ |
2716 | err = -EIO; | |
2717 | break; | |
2718 | } | |
c29c0ae7 AT |
2719 | if (WARN_ON(i + 1 > depth)) { |
2720 | err = -EIO; | |
2721 | break; | |
2722 | } | |
f8489128 DW |
2723 | if (ext4_ext_check_block(inode, ext_block_hdr(bh), |
2724 | depth - i - 1, bh)) { | |
c29c0ae7 AT |
2725 | err = -EIO; |
2726 | break; | |
2727 | } | |
2728 | path[i + 1].p_bh = bh; | |
a86c6181 | 2729 | |
d0d856e8 RD |
2730 | /* save actual number of indexes since this |
2731 | * number is changed at the next iteration */ | |
a86c6181 AT |
2732 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2733 | i++; | |
2734 | } else { | |
d0d856e8 | 2735 | /* we finished processing this index, go up */ |
a86c6181 | 2736 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2737 | /* index is empty, remove it; |
a86c6181 AT |
2738 | * handle must be already prepared by the |
2739 | * truncatei_leaf() */ | |
2740 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2741 | } | |
d0d856e8 | 2742 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2743 | brelse(path[i].p_bh); |
2744 | path[i].p_bh = NULL; | |
2745 | i--; | |
2746 | ext_debug("return to level %d\n", i); | |
2747 | } | |
2748 | } | |
2749 | ||
d8990240 AK |
2750 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2751 | path->p_hdr->eh_entries); | |
2752 | ||
7b415bf6 AK |
2753 | /* If we still have something in the partial cluster and we have removed |
2754 | * even the first extent, then we should free the blocks in the partial | |
2755 | * cluster as well. */ | |
2756 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2757 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2758 | ||
2759 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2760 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2761 | ||
2762 | ext4_free_blocks(handle, inode, NULL, | |
2763 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2764 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2765 | partial_cluster = 0; | |
2766 | } | |
2767 | ||
a86c6181 AT |
2768 | /* TODO: flexible tree reduction should be here */ |
2769 | if (path->p_hdr->eh_entries == 0) { | |
2770 | /* | |
d0d856e8 RD |
2771 | * truncate to zero freed all the tree, |
2772 | * so we need to correct eh_depth | |
a86c6181 AT |
2773 | */ |
2774 | err = ext4_ext_get_access(handle, inode, path); | |
2775 | if (err == 0) { | |
2776 | ext_inode_hdr(inode)->eh_depth = 0; | |
2777 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2778 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2779 | err = ext4_ext_dirty(handle, inode, path); |
2780 | } | |
2781 | } | |
2782 | out: | |
a86c6181 AT |
2783 | ext4_ext_drop_refs(path); |
2784 | kfree(path); | |
968dee77 AS |
2785 | if (err == -EAGAIN) { |
2786 | path = NULL; | |
0617b83f | 2787 | goto again; |
968dee77 | 2788 | } |
a86c6181 AT |
2789 | ext4_journal_stop(handle); |
2790 | ||
2791 | return err; | |
2792 | } | |
2793 | ||
2794 | /* | |
2795 | * called at mount time | |
2796 | */ | |
2797 | void ext4_ext_init(struct super_block *sb) | |
2798 | { | |
2799 | /* | |
2800 | * possible initialization would be here | |
2801 | */ | |
2802 | ||
83982b6f | 2803 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2804 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2805 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2806 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2807 | ", aggressive tests" |
a86c6181 AT |
2808 | #endif |
2809 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2810 | ", check binsearch" |
a86c6181 AT |
2811 | #endif |
2812 | #ifdef EXTENTS_STATS | |
92b97816 | 2813 | ", stats" |
a86c6181 | 2814 | #endif |
92b97816 | 2815 | "\n"); |
90576c0b | 2816 | #endif |
a86c6181 AT |
2817 | #ifdef EXTENTS_STATS |
2818 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2819 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2820 | EXT4_SB(sb)->s_ext_max = 0; | |
2821 | #endif | |
2822 | } | |
2823 | } | |
2824 | ||
2825 | /* | |
2826 | * called at umount time | |
2827 | */ | |
2828 | void ext4_ext_release(struct super_block *sb) | |
2829 | { | |
83982b6f | 2830 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2831 | return; |
2832 | ||
2833 | #ifdef EXTENTS_STATS | |
2834 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2835 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2836 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2837 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2838 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2839 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2840 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2841 | } | |
2842 | #endif | |
2843 | } | |
2844 | ||
093a088b AK |
2845 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2846 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2847 | { | |
2407518d LC |
2848 | ext4_fsblk_t ee_pblock; |
2849 | unsigned int ee_len; | |
b720303d | 2850 | int ret; |
093a088b | 2851 | |
093a088b | 2852 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2853 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2854 | |
a107e5a3 | 2855 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2856 | if (ret > 0) |
2857 | ret = 0; | |
093a088b | 2858 | |
2407518d | 2859 | return ret; |
093a088b AK |
2860 | } |
2861 | ||
47ea3bb5 YY |
2862 | /* |
2863 | * ext4_split_extent_at() splits an extent at given block. | |
2864 | * | |
2865 | * @handle: the journal handle | |
2866 | * @inode: the file inode | |
2867 | * @path: the path to the extent | |
2868 | * @split: the logical block where the extent is splitted. | |
2869 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2870 | * the states(init or uninit) of new extents. | |
2871 | * @flags: flags used to insert new extent to extent tree. | |
2872 | * | |
2873 | * | |
2874 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2875 | * of which are deterimined by split_flag. | |
2876 | * | |
2877 | * There are two cases: | |
2878 | * a> the extent are splitted into two extent. | |
2879 | * b> split is not needed, and just mark the extent. | |
2880 | * | |
2881 | * return 0 on success. | |
2882 | */ | |
2883 | static int ext4_split_extent_at(handle_t *handle, | |
2884 | struct inode *inode, | |
2885 | struct ext4_ext_path *path, | |
2886 | ext4_lblk_t split, | |
2887 | int split_flag, | |
2888 | int flags) | |
2889 | { | |
2890 | ext4_fsblk_t newblock; | |
2891 | ext4_lblk_t ee_block; | |
2892 | struct ext4_extent *ex, newex, orig_ex; | |
2893 | struct ext4_extent *ex2 = NULL; | |
2894 | unsigned int ee_len, depth; | |
2895 | int err = 0; | |
2896 | ||
2897 | ext_debug("ext4_split_extents_at: inode %lu, logical" | |
2898 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2899 | ||
2900 | ext4_ext_show_leaf(inode, path); | |
2901 | ||
2902 | depth = ext_depth(inode); | |
2903 | ex = path[depth].p_ext; | |
2904 | ee_block = le32_to_cpu(ex->ee_block); | |
2905 | ee_len = ext4_ext_get_actual_len(ex); | |
2906 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2907 | ||
2908 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
2909 | ||
2910 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2911 | if (err) | |
2912 | goto out; | |
2913 | ||
2914 | if (split == ee_block) { | |
2915 | /* | |
2916 | * case b: block @split is the block that the extent begins with | |
2917 | * then we just change the state of the extent, and splitting | |
2918 | * is not needed. | |
2919 | */ | |
2920 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2921 | ext4_ext_mark_uninitialized(ex); | |
2922 | else | |
2923 | ext4_ext_mark_initialized(ex); | |
2924 | ||
2925 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
2926 | ext4_ext_try_to_merge(inode, path, ex); | |
2927 | ||
2928 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2929 | goto out; | |
2930 | } | |
2931 | ||
2932 | /* case a */ | |
2933 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
2934 | ex->ee_len = cpu_to_le16(split - ee_block); | |
2935 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
2936 | ext4_ext_mark_uninitialized(ex); | |
2937 | ||
2938 | /* | |
2939 | * path may lead to new leaf, not to original leaf any more | |
2940 | * after ext4_ext_insert_extent() returns, | |
2941 | */ | |
2942 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2943 | if (err) | |
2944 | goto fix_extent_len; | |
2945 | ||
2946 | ex2 = &newex; | |
2947 | ex2->ee_block = cpu_to_le32(split); | |
2948 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
2949 | ext4_ext_store_pblock(ex2, newblock); | |
2950 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2951 | ext4_ext_mark_uninitialized(ex2); | |
2952 | ||
2953 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
2954 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
2955 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2956 | if (err) | |
2957 | goto fix_extent_len; | |
2958 | /* update the extent length and mark as initialized */ | |
af1584f5 | 2959 | ex->ee_len = cpu_to_le16(ee_len); |
47ea3bb5 YY |
2960 | ext4_ext_try_to_merge(inode, path, ex); |
2961 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2962 | goto out; | |
2963 | } else if (err) | |
2964 | goto fix_extent_len; | |
2965 | ||
2966 | out: | |
2967 | ext4_ext_show_leaf(inode, path); | |
2968 | return err; | |
2969 | ||
2970 | fix_extent_len: | |
2971 | ex->ee_len = orig_ex.ee_len; | |
2972 | ext4_ext_dirty(handle, inode, path + depth); | |
2973 | return err; | |
2974 | } | |
2975 | ||
2976 | /* | |
2977 | * ext4_split_extents() splits an extent and mark extent which is covered | |
2978 | * by @map as split_flags indicates | |
2979 | * | |
2980 | * It may result in splitting the extent into multiple extents (upto three) | |
2981 | * There are three possibilities: | |
2982 | * a> There is no split required | |
2983 | * b> Splits in two extents: Split is happening at either end of the extent | |
2984 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
2985 | * | |
2986 | */ | |
2987 | static int ext4_split_extent(handle_t *handle, | |
2988 | struct inode *inode, | |
2989 | struct ext4_ext_path *path, | |
2990 | struct ext4_map_blocks *map, | |
2991 | int split_flag, | |
2992 | int flags) | |
2993 | { | |
2994 | ext4_lblk_t ee_block; | |
2995 | struct ext4_extent *ex; | |
2996 | unsigned int ee_len, depth; | |
2997 | int err = 0; | |
2998 | int uninitialized; | |
2999 | int split_flag1, flags1; | |
3000 | ||
3001 | depth = ext_depth(inode); | |
3002 | ex = path[depth].p_ext; | |
3003 | ee_block = le32_to_cpu(ex->ee_block); | |
3004 | ee_len = ext4_ext_get_actual_len(ex); | |
3005 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3006 | ||
3007 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
3008 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
3009 | EXT4_EXT_MAY_ZEROOUT : 0; | |
3010 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; | |
3011 | if (uninitialized) | |
3012 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3013 | EXT4_EXT_MARK_UNINIT2; | |
3014 | err = ext4_split_extent_at(handle, inode, path, | |
3015 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3016 | if (err) |
3017 | goto out; | |
47ea3bb5 YY |
3018 | } |
3019 | ||
3020 | ext4_ext_drop_refs(path); | |
3021 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3022 | if (IS_ERR(path)) | |
3023 | return PTR_ERR(path); | |
3024 | ||
3025 | if (map->m_lblk >= ee_block) { | |
3026 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ? | |
3027 | EXT4_EXT_MAY_ZEROOUT : 0; | |
3028 | if (uninitialized) | |
3029 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
3030 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3031 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
3032 | err = ext4_split_extent_at(handle, inode, path, | |
3033 | map->m_lblk, split_flag1, flags); | |
3034 | if (err) | |
3035 | goto out; | |
3036 | } | |
3037 | ||
3038 | ext4_ext_show_leaf(inode, path); | |
3039 | out: | |
3040 | return err ? err : map->m_len; | |
3041 | } | |
3042 | ||
3977c965 | 3043 | #define EXT4_EXT_ZERO_LEN 7 |
56055d3a | 3044 | /* |
e35fd660 | 3045 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3046 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3047 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3048 | * uninitialized). |
3049 | * There are three possibilities: | |
3050 | * a> There is no split required: Entire extent should be initialized | |
3051 | * b> Splits in two extents: Write is happening at either end of the extent | |
3052 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3053 | * |
3054 | * Pre-conditions: | |
3055 | * - The extent pointed to by 'path' is uninitialized. | |
3056 | * - The extent pointed to by 'path' contains a superset | |
3057 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3058 | * | |
3059 | * Post-conditions on success: | |
3060 | * - the returned value is the number of blocks beyond map->l_lblk | |
3061 | * that are allocated and initialized. | |
3062 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3063 | */ |
725d26d3 | 3064 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3065 | struct inode *inode, |
3066 | struct ext4_map_blocks *map, | |
3067 | struct ext4_ext_path *path) | |
56055d3a | 3068 | { |
6f91bc5f | 3069 | struct ext4_extent_header *eh; |
667eff35 YY |
3070 | struct ext4_map_blocks split_map; |
3071 | struct ext4_extent zero_ex; | |
3072 | struct ext4_extent *ex; | |
21ca087a | 3073 | ext4_lblk_t ee_block, eof_block; |
f85b287a DC |
3074 | unsigned int ee_len, depth; |
3075 | int allocated; | |
56055d3a | 3076 | int err = 0; |
667eff35 | 3077 | int split_flag = 0; |
21ca087a DM |
3078 | |
3079 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3080 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3081 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3082 | |
3083 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3084 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3085 | if (eof_block < map->m_lblk + map->m_len) |
3086 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3087 | |
3088 | depth = ext_depth(inode); | |
6f91bc5f | 3089 | eh = path[depth].p_hdr; |
56055d3a AA |
3090 | ex = path[depth].p_ext; |
3091 | ee_block = le32_to_cpu(ex->ee_block); | |
3092 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3093 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 3094 | |
6f91bc5f EG |
3095 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3096 | ||
3097 | /* Pre-conditions */ | |
3098 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3099 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3100 | |
3101 | /* | |
3102 | * Attempt to transfer newly initialized blocks from the currently | |
3103 | * uninitialized extent to its left neighbor. This is much cheaper | |
3104 | * than an insertion followed by a merge as those involve costly | |
3105 | * memmove() calls. This is the common case in steady state for | |
3106 | * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append | |
3107 | * writes. | |
3108 | * | |
3109 | * Limitations of the current logic: | |
3110 | * - L1: we only deal with writes at the start of the extent. | |
3111 | * The approach could be extended to writes at the end | |
3112 | * of the extent but this scenario was deemed less common. | |
3113 | * - L2: we do not deal with writes covering the whole extent. | |
3114 | * This would require removing the extent if the transfer | |
3115 | * is possible. | |
3116 | * - L3: we only attempt to merge with an extent stored in the | |
3117 | * same extent tree node. | |
3118 | */ | |
3119 | if ((map->m_lblk == ee_block) && /*L1*/ | |
3120 | (map->m_len < ee_len) && /*L2*/ | |
3121 | (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/ | |
3122 | struct ext4_extent *prev_ex; | |
3123 | ext4_lblk_t prev_lblk; | |
3124 | ext4_fsblk_t prev_pblk, ee_pblk; | |
3125 | unsigned int prev_len, write_len; | |
3126 | ||
3127 | prev_ex = ex - 1; | |
3128 | prev_lblk = le32_to_cpu(prev_ex->ee_block); | |
3129 | prev_len = ext4_ext_get_actual_len(prev_ex); | |
3130 | prev_pblk = ext4_ext_pblock(prev_ex); | |
3131 | ee_pblk = ext4_ext_pblock(ex); | |
3132 | write_len = map->m_len; | |
3133 | ||
3134 | /* | |
3135 | * A transfer of blocks from 'ex' to 'prev_ex' is allowed | |
3136 | * upon those conditions: | |
3137 | * - C1: prev_ex is initialized, | |
3138 | * - C2: prev_ex is logically abutting ex, | |
3139 | * - C3: prev_ex is physically abutting ex, | |
3140 | * - C4: prev_ex can receive the additional blocks without | |
3141 | * overflowing the (initialized) length limit. | |
3142 | */ | |
3143 | if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/ | |
3144 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ | |
3145 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
3146 | (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/ | |
3147 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3148 | if (err) | |
3149 | goto out; | |
3150 | ||
3151 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3152 | map, ex, prev_ex); | |
3153 | ||
3154 | /* Shift the start of ex by 'write_len' blocks */ | |
3155 | ex->ee_block = cpu_to_le32(ee_block + write_len); | |
3156 | ext4_ext_store_pblock(ex, ee_pblk + write_len); | |
3157 | ex->ee_len = cpu_to_le16(ee_len - write_len); | |
3158 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3159 | ||
3160 | /* Extend prev_ex by 'write_len' blocks */ | |
3161 | prev_ex->ee_len = cpu_to_le16(prev_len + write_len); | |
3162 | ||
3163 | /* Mark the block containing both extents as dirty */ | |
3164 | ext4_ext_dirty(handle, inode, path + depth); | |
3165 | ||
3166 | /* Update path to point to the right extent */ | |
3167 | path[depth].p_ext = prev_ex; | |
3168 | ||
3169 | /* Result: number of initialized blocks past m_lblk */ | |
3170 | allocated = write_len; | |
3171 | goto out; | |
3172 | } | |
3173 | } | |
3174 | ||
667eff35 | 3175 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3176 | /* |
3177 | * It is safe to convert extent to initialized via explicit | |
3178 | * zeroout only if extent is fully insde i_size or new_size. | |
3179 | */ | |
667eff35 | 3180 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3181 | |
3977c965 | 3182 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
667eff35 YY |
3183 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN && |
3184 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3185 | err = ext4_ext_zeroout(inode, ex); | |
3977c965 | 3186 | if (err) |
d03856bd | 3187 | goto out; |
d03856bd AK |
3188 | |
3189 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3190 | if (err) | |
3191 | goto out; | |
667eff35 YY |
3192 | ext4_ext_mark_initialized(ex); |
3193 | ext4_ext_try_to_merge(inode, path, ex); | |
3194 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3195 | goto out; | |
56055d3a | 3196 | } |
667eff35 | 3197 | |
56055d3a | 3198 | /* |
667eff35 YY |
3199 | * four cases: |
3200 | * 1. split the extent into three extents. | |
3201 | * 2. split the extent into two extents, zeroout the first half. | |
3202 | * 3. split the extent into two extents, zeroout the second half. | |
3203 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3204 | */ |
667eff35 YY |
3205 | split_map.m_lblk = map->m_lblk; |
3206 | split_map.m_len = map->m_len; | |
3207 | ||
3208 | if (allocated > map->m_len) { | |
3209 | if (allocated <= EXT4_EXT_ZERO_LEN && | |
3210 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3211 | /* case 3 */ | |
3212 | zero_ex.ee_block = | |
9b940f8e AH |
3213 | cpu_to_le32(map->m_lblk); |
3214 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3215 | ext4_ext_store_pblock(&zero_ex, |
3216 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3217 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3218 | if (err) |
3219 | goto out; | |
667eff35 YY |
3220 | split_map.m_lblk = map->m_lblk; |
3221 | split_map.m_len = allocated; | |
3222 | } else if ((map->m_lblk - ee_block + map->m_len < | |
3223 | EXT4_EXT_ZERO_LEN) && | |
3224 | (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
3225 | /* case 2 */ | |
3226 | if (map->m_lblk != ee_block) { | |
3227 | zero_ex.ee_block = ex->ee_block; | |
3228 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3229 | ee_block); | |
3230 | ext4_ext_store_pblock(&zero_ex, | |
3231 | ext4_ext_pblock(ex)); | |
3232 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3233 | if (err) | |
3234 | goto out; | |
3235 | } | |
3236 | ||
667eff35 | 3237 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3238 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3239 | allocated = map->m_len; | |
56055d3a AA |
3240 | } |
3241 | } | |
667eff35 YY |
3242 | |
3243 | allocated = ext4_split_extent(handle, inode, path, | |
3244 | &split_map, split_flag, 0); | |
3245 | if (allocated < 0) | |
3246 | err = allocated; | |
3247 | ||
56055d3a AA |
3248 | out: |
3249 | return err ? err : allocated; | |
3250 | } | |
3251 | ||
0031462b | 3252 | /* |
e35fd660 | 3253 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3254 | * ext4_get_blocks_dio_write() when DIO to write |
3255 | * to an uninitialized extent. | |
3256 | * | |
fd018fe8 | 3257 | * Writing to an uninitialized extent may result in splitting the uninitialized |
b595076a | 3258 | * extent into multiple /initialized uninitialized extents (up to three) |
0031462b MC |
3259 | * There are three possibilities: |
3260 | * a> There is no split required: Entire extent should be uninitialized | |
3261 | * b> Splits in two extents: Write is happening at either end of the extent | |
3262 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3263 | * | |
3264 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3265 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3266 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3267 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3268 | * into three uninitialized extent(at most). After IO complete, the part |
3269 | * being filled will be convert to initialized by the end_io callback function | |
3270 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3271 | * |
3272 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3273 | */ |
3274 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3275 | struct inode *inode, | |
e35fd660 | 3276 | struct ext4_map_blocks *map, |
0031462b | 3277 | struct ext4_ext_path *path, |
0031462b MC |
3278 | int flags) |
3279 | { | |
667eff35 YY |
3280 | ext4_lblk_t eof_block; |
3281 | ext4_lblk_t ee_block; | |
3282 | struct ext4_extent *ex; | |
3283 | unsigned int ee_len; | |
3284 | int split_flag = 0, depth; | |
21ca087a DM |
3285 | |
3286 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3287 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3288 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3289 | |
3290 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3291 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3292 | if (eof_block < map->m_lblk + map->m_len) |
3293 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3294 | /* |
3295 | * It is safe to convert extent to initialized via explicit | |
3296 | * zeroout only if extent is fully insde i_size or new_size. | |
3297 | */ | |
667eff35 YY |
3298 | depth = ext_depth(inode); |
3299 | ex = path[depth].p_ext; | |
3300 | ee_block = le32_to_cpu(ex->ee_block); | |
3301 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3302 | |
667eff35 YY |
3303 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3304 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
0031462b | 3305 | |
667eff35 YY |
3306 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3307 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3308 | } |
197217a5 | 3309 | |
c7064ef1 | 3310 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
0031462b MC |
3311 | struct inode *inode, |
3312 | struct ext4_ext_path *path) | |
3313 | { | |
3314 | struct ext4_extent *ex; | |
0031462b MC |
3315 | int depth; |
3316 | int err = 0; | |
0031462b MC |
3317 | |
3318 | depth = ext_depth(inode); | |
0031462b MC |
3319 | ex = path[depth].p_ext; |
3320 | ||
197217a5 YY |
3321 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3322 | "block %llu, max_blocks %u\n", inode->i_ino, | |
3323 | (unsigned long long)le32_to_cpu(ex->ee_block), | |
3324 | ext4_ext_get_actual_len(ex)); | |
3325 | ||
0031462b MC |
3326 | err = ext4_ext_get_access(handle, inode, path + depth); |
3327 | if (err) | |
3328 | goto out; | |
3329 | /* first mark the extent as initialized */ | |
3330 | ext4_ext_mark_initialized(ex); | |
3331 | ||
197217a5 YY |
3332 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3333 | * borders are not changed | |
0031462b | 3334 | */ |
197217a5 YY |
3335 | ext4_ext_try_to_merge(inode, path, ex); |
3336 | ||
0031462b MC |
3337 | /* Mark modified extent as dirty */ |
3338 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3339 | out: | |
3340 | ext4_ext_show_leaf(inode, path); | |
3341 | return err; | |
3342 | } | |
3343 | ||
515f41c3 AK |
3344 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3345 | sector_t block, int count) | |
3346 | { | |
3347 | int i; | |
3348 | for (i = 0; i < count; i++) | |
3349 | unmap_underlying_metadata(bdev, block + i); | |
3350 | } | |
3351 | ||
58590b06 TT |
3352 | /* |
3353 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3354 | */ | |
3355 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3356 | ext4_lblk_t lblk, |
58590b06 TT |
3357 | struct ext4_ext_path *path, |
3358 | unsigned int len) | |
3359 | { | |
3360 | int i, depth; | |
3361 | struct ext4_extent_header *eh; | |
65922cb5 | 3362 | struct ext4_extent *last_ex; |
58590b06 TT |
3363 | |
3364 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3365 | return 0; | |
3366 | ||
3367 | depth = ext_depth(inode); | |
3368 | eh = path[depth].p_hdr; | |
58590b06 | 3369 | |
afcff5d8 LC |
3370 | /* |
3371 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3372 | * do not care for this case anymore. Simply remove the flag | |
3373 | * if there are no extents. | |
3374 | */ | |
3375 | if (unlikely(!eh->eh_entries)) | |
3376 | goto out; | |
58590b06 TT |
3377 | last_ex = EXT_LAST_EXTENT(eh); |
3378 | /* | |
3379 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3380 | * last block in the last extent in the file. We test this by | |
3381 | * first checking to see if the caller to | |
3382 | * ext4_ext_get_blocks() was interested in the last block (or | |
3383 | * a block beyond the last block) in the current extent. If | |
3384 | * this turns out to be false, we can bail out from this | |
3385 | * function immediately. | |
3386 | */ | |
d002ebf1 | 3387 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3388 | ext4_ext_get_actual_len(last_ex)) |
3389 | return 0; | |
3390 | /* | |
3391 | * If the caller does appear to be planning to write at or | |
3392 | * beyond the end of the current extent, we then test to see | |
3393 | * if the current extent is the last extent in the file, by | |
3394 | * checking to make sure it was reached via the rightmost node | |
3395 | * at each level of the tree. | |
3396 | */ | |
3397 | for (i = depth-1; i >= 0; i--) | |
3398 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3399 | return 0; | |
afcff5d8 | 3400 | out: |
58590b06 TT |
3401 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3402 | return ext4_mark_inode_dirty(handle, inode); | |
3403 | } | |
3404 | ||
7b415bf6 AK |
3405 | /** |
3406 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3407 | * | |
3408 | * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns | |
3409 | * whether there are any buffers marked for delayed allocation. It returns '1' | |
3410 | * on the first delalloc'ed buffer head found. If no buffer head in the given | |
3411 | * range is marked for delalloc, it returns 0. | |
3412 | * lblk_start should always be <= lblk_end. | |
3413 | * search_hint_reverse is to indicate that searching in reverse from lblk_end to | |
3414 | * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed | |
3415 | * block sooner). This is useful when blocks are truncated sequentially from | |
3416 | * lblk_start towards lblk_end. | |
3417 | */ | |
3418 | static int ext4_find_delalloc_range(struct inode *inode, | |
3419 | ext4_lblk_t lblk_start, | |
3420 | ext4_lblk_t lblk_end, | |
3421 | int search_hint_reverse) | |
3422 | { | |
3423 | struct address_space *mapping = inode->i_mapping; | |
3424 | struct buffer_head *head, *bh = NULL; | |
3425 | struct page *page; | |
3426 | ext4_lblk_t i, pg_lblk; | |
3427 | pgoff_t index; | |
3428 | ||
8c48f7e8 RD |
3429 | if (!test_opt(inode->i_sb, DELALLOC)) |
3430 | return 0; | |
3431 | ||
7b415bf6 AK |
3432 | /* reverse search wont work if fs block size is less than page size */ |
3433 | if (inode->i_blkbits < PAGE_CACHE_SHIFT) | |
3434 | search_hint_reverse = 0; | |
3435 | ||
3436 | if (search_hint_reverse) | |
3437 | i = lblk_end; | |
3438 | else | |
3439 | i = lblk_start; | |
3440 | ||
3441 | index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
3442 | ||
3443 | while ((i >= lblk_start) && (i <= lblk_end)) { | |
3444 | page = find_get_page(mapping, index); | |
5356f261 | 3445 | if (!page) |
7b415bf6 AK |
3446 | goto nextpage; |
3447 | ||
7b415bf6 AK |
3448 | if (!page_has_buffers(page)) |
3449 | goto nextpage; | |
3450 | ||
3451 | head = page_buffers(page); | |
3452 | if (!head) | |
3453 | goto nextpage; | |
3454 | ||
3455 | bh = head; | |
3456 | pg_lblk = index << (PAGE_CACHE_SHIFT - | |
3457 | inode->i_blkbits); | |
3458 | do { | |
3459 | if (unlikely(pg_lblk < lblk_start)) { | |
3460 | /* | |
3461 | * This is possible when fs block size is less | |
3462 | * than page size and our cluster starts/ends in | |
3463 | * middle of the page. So we need to skip the | |
3464 | * initial few blocks till we reach the 'lblk' | |
3465 | */ | |
3466 | pg_lblk++; | |
3467 | continue; | |
3468 | } | |
3469 | ||
5356f261 AK |
3470 | /* Check if the buffer is delayed allocated and that it |
3471 | * is not yet mapped. (when da-buffers are mapped during | |
3472 | * their writeout, their da_mapped bit is set.) | |
3473 | */ | |
3474 | if (buffer_delay(bh) && !buffer_da_mapped(bh)) { | |
7b415bf6 | 3475 | page_cache_release(page); |
d8990240 AK |
3476 | trace_ext4_find_delalloc_range(inode, |
3477 | lblk_start, lblk_end, | |
3478 | search_hint_reverse, | |
3479 | 1, i); | |
7b415bf6 AK |
3480 | return 1; |
3481 | } | |
3482 | if (search_hint_reverse) | |
3483 | i--; | |
3484 | else | |
3485 | i++; | |
3486 | } while ((i >= lblk_start) && (i <= lblk_end) && | |
3487 | ((bh = bh->b_this_page) != head)); | |
3488 | nextpage: | |
3489 | if (page) | |
3490 | page_cache_release(page); | |
3491 | /* | |
3492 | * Move to next page. 'i' will be the first lblk in the next | |
3493 | * page. | |
3494 | */ | |
3495 | if (search_hint_reverse) | |
3496 | index--; | |
3497 | else | |
3498 | index++; | |
3499 | i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
3500 | } | |
3501 | ||
d8990240 AK |
3502 | trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end, |
3503 | search_hint_reverse, 0, 0); | |
7b415bf6 AK |
3504 | return 0; |
3505 | } | |
3506 | ||
3507 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk, | |
3508 | int search_hint_reverse) | |
3509 | { | |
3510 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3511 | ext4_lblk_t lblk_start, lblk_end; | |
3512 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3513 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3514 | ||
3515 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end, | |
3516 | search_hint_reverse); | |
3517 | } | |
3518 | ||
3519 | /** | |
3520 | * Determines how many complete clusters (out of those specified by the 'map') | |
3521 | * are under delalloc and were reserved quota for. | |
3522 | * This function is called when we are writing out the blocks that were | |
3523 | * originally written with their allocation delayed, but then the space was | |
3524 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3525 | * The cases to look for are: | |
3526 | * ('=' indicated delayed allocated blocks | |
3527 | * '-' indicates non-delayed allocated blocks) | |
3528 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3529 | * are not delalloc'ed. | |
3530 | * Ex: | |
3531 | * |----c---=|====c====|====c====|===-c----| | |
3532 | * |++++++ allocated ++++++| | |
3533 | * ==> 4 complete clusters in above example | |
3534 | * | |
3535 | * (b) partial cluster (outside of allocated range) towards either end is | |
3536 | * marked for delayed allocation. In this case, we will exclude that | |
3537 | * cluster. | |
3538 | * Ex: | |
3539 | * |----====c========|========c========| | |
3540 | * |++++++ allocated ++++++| | |
3541 | * ==> 1 complete clusters in above example | |
3542 | * | |
3543 | * Ex: | |
3544 | * |================c================| | |
3545 | * |++++++ allocated ++++++| | |
3546 | * ==> 0 complete clusters in above example | |
3547 | * | |
3548 | * The ext4_da_update_reserve_space will be called only if we | |
3549 | * determine here that there were some "entire" clusters that span | |
3550 | * this 'allocated' range. | |
3551 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3552 | * without ever calling ext4_find_delalloc_range. | |
3553 | */ | |
3554 | static unsigned int | |
3555 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3556 | unsigned int num_blks) | |
3557 | { | |
3558 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3559 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3560 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3561 | unsigned int allocated_clusters = 0; | |
3562 | ||
3563 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3564 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3565 | ||
3566 | /* max possible clusters for this allocation */ | |
3567 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3568 | ||
d8990240 AK |
3569 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3570 | ||
7b415bf6 AK |
3571 | /* Check towards left side */ |
3572 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3573 | if (c_offset) { | |
3574 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3575 | lblk_to = lblk_from + c_offset - 1; | |
3576 | ||
3577 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0)) | |
3578 | allocated_clusters--; | |
3579 | } | |
3580 | ||
3581 | /* Now check towards right. */ | |
3582 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3583 | if (allocated_clusters && c_offset) { | |
3584 | lblk_from = lblk_start + num_blks; | |
3585 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3586 | ||
3587 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0)) | |
3588 | allocated_clusters--; | |
3589 | } | |
3590 | ||
3591 | return allocated_clusters; | |
3592 | } | |
3593 | ||
0031462b MC |
3594 | static int |
3595 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3596 | struct ext4_map_blocks *map, |
0031462b | 3597 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3598 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3599 | { |
3600 | int ret = 0; | |
3601 | int err = 0; | |
8d5d02e6 | 3602 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
0031462b | 3603 | |
88635ca2 ZL |
3604 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3605 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3606 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3607 | flags, allocated); |
3608 | ext4_ext_show_leaf(inode, path); | |
3609 | ||
d8990240 AK |
3610 | trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated, |
3611 | newblock); | |
3612 | ||
c7064ef1 | 3613 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3614 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3615 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3616 | path, flags); | |
5f524950 M |
3617 | /* |
3618 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3619 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3620 | * completed |
3621 | */ | |
0edeb71d TM |
3622 | if (io) |
3623 | ext4_set_io_unwritten_flag(inode, io); | |
3624 | else | |
19f5fb7a | 3625 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
744692dc | 3626 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3627 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3628 | goto out; |
3629 | } | |
c7064ef1 | 3630 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3631 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
c7064ef1 | 3632 | ret = ext4_convert_unwritten_extents_endio(handle, inode, |
0031462b | 3633 | path); |
58590b06 | 3634 | if (ret >= 0) { |
b436b9be | 3635 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3636 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3637 | path, map->m_len); | |
58590b06 TT |
3638 | } else |
3639 | err = ret; | |
0031462b MC |
3640 | goto out2; |
3641 | } | |
3642 | /* buffered IO case */ | |
3643 | /* | |
3644 | * repeat fallocate creation request | |
3645 | * we already have an unwritten extent | |
3646 | */ | |
3647 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | |
3648 | goto map_out; | |
3649 | ||
3650 | /* buffered READ or buffered write_begin() lookup */ | |
3651 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3652 | /* | |
3653 | * We have blocks reserved already. We | |
3654 | * return allocated blocks so that delalloc | |
3655 | * won't do block reservation for us. But | |
3656 | * the buffer head will be unmapped so that | |
3657 | * a read from the block returns 0s. | |
3658 | */ | |
e35fd660 | 3659 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3660 | goto out1; |
3661 | } | |
3662 | ||
3663 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3664 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
a4e5d88b | 3665 | if (ret >= 0) |
b436b9be | 3666 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3667 | out: |
3668 | if (ret <= 0) { | |
3669 | err = ret; | |
3670 | goto out2; | |
3671 | } else | |
3672 | allocated = ret; | |
e35fd660 | 3673 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3674 | /* |
3675 | * if we allocated more blocks than requested | |
3676 | * we need to make sure we unmap the extra block | |
3677 | * allocated. The actual needed block will get | |
3678 | * unmapped later when we find the buffer_head marked | |
3679 | * new. | |
3680 | */ | |
e35fd660 | 3681 | if (allocated > map->m_len) { |
515f41c3 | 3682 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3683 | newblock + map->m_len, |
3684 | allocated - map->m_len); | |
3685 | allocated = map->m_len; | |
515f41c3 | 3686 | } |
5f634d06 AK |
3687 | |
3688 | /* | |
3689 | * If we have done fallocate with the offset that is already | |
3690 | * delayed allocated, we would have block reservation | |
3691 | * and quota reservation done in the delayed write path. | |
3692 | * But fallocate would have already updated quota and block | |
3693 | * count for this offset. So cancel these reservation | |
3694 | */ | |
7b415bf6 AK |
3695 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3696 | unsigned int reserved_clusters; | |
3697 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3698 | map->m_lblk, map->m_len); | |
3699 | if (reserved_clusters) | |
3700 | ext4_da_update_reserve_space(inode, | |
3701 | reserved_clusters, | |
3702 | 0); | |
3703 | } | |
5f634d06 | 3704 | |
0031462b | 3705 | map_out: |
e35fd660 | 3706 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3707 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3708 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3709 | map->m_len); | |
3710 | if (err < 0) | |
3711 | goto out2; | |
3712 | } | |
0031462b | 3713 | out1: |
e35fd660 TT |
3714 | if (allocated > map->m_len) |
3715 | allocated = map->m_len; | |
0031462b | 3716 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3717 | map->m_pblk = newblock; |
3718 | map->m_len = allocated; | |
0031462b MC |
3719 | out2: |
3720 | if (path) { | |
3721 | ext4_ext_drop_refs(path); | |
3722 | kfree(path); | |
3723 | } | |
3724 | return err ? err : allocated; | |
3725 | } | |
58590b06 | 3726 | |
4d33b1ef TT |
3727 | /* |
3728 | * get_implied_cluster_alloc - check to see if the requested | |
3729 | * allocation (in the map structure) overlaps with a cluster already | |
3730 | * allocated in an extent. | |
d8990240 | 3731 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3732 | * @map The requested lblk->pblk mapping |
3733 | * @ex The extent structure which might contain an implied | |
3734 | * cluster allocation | |
3735 | * | |
3736 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3737 | * find blocks that were already in the inode's extent tree. Hence, | |
3738 | * we know that the beginning of the requested region cannot overlap | |
3739 | * the extent from the inode's extent tree. There are three cases we | |
3740 | * want to catch. The first is this case: | |
3741 | * | |
3742 | * |--- cluster # N--| | |
3743 | * |--- extent ---| |---- requested region ---| | |
3744 | * |==========| | |
3745 | * | |
3746 | * The second case that we need to test for is this one: | |
3747 | * | |
3748 | * |--------- cluster # N ----------------| | |
3749 | * |--- requested region --| |------- extent ----| | |
3750 | * |=======================| | |
3751 | * | |
3752 | * The third case is when the requested region lies between two extents | |
3753 | * within the same cluster: | |
3754 | * |------------- cluster # N-------------| | |
3755 | * |----- ex -----| |---- ex_right ----| | |
3756 | * |------ requested region ------| | |
3757 | * |================| | |
3758 | * | |
3759 | * In each of the above cases, we need to set the map->m_pblk and | |
3760 | * map->m_len so it corresponds to the return the extent labelled as | |
3761 | * "|====|" from cluster #N, since it is already in use for data in | |
3762 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3763 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3764 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3765 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3766 | * by calling ext4_mb_new_blocks(). | |
3767 | */ | |
d8990240 | 3768 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3769 | struct ext4_map_blocks *map, |
3770 | struct ext4_extent *ex, | |
3771 | struct ext4_ext_path *path) | |
3772 | { | |
d8990240 | 3773 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3774 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3775 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3776 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3777 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3778 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3779 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3780 | ||
3781 | /* The extent passed in that we are trying to match */ | |
3782 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3783 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3784 | ||
3785 | /* The requested region passed into ext4_map_blocks() */ | |
3786 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3787 | |
3788 | if ((rr_cluster_start == ex_cluster_end) || | |
3789 | (rr_cluster_start == ex_cluster_start)) { | |
3790 | if (rr_cluster_start == ex_cluster_end) | |
3791 | ee_start += ee_len - 1; | |
3792 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3793 | c_offset; | |
3794 | map->m_len = min(map->m_len, | |
3795 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3796 | /* | |
3797 | * Check for and handle this case: | |
3798 | * | |
3799 | * |--------- cluster # N-------------| | |
3800 | * |------- extent ----| | |
3801 | * |--- requested region ---| | |
3802 | * |===========| | |
3803 | */ | |
3804 | ||
3805 | if (map->m_lblk < ee_block) | |
3806 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3807 | ||
3808 | /* | |
3809 | * Check for the case where there is already another allocated | |
3810 | * block to the right of 'ex' but before the end of the cluster. | |
3811 | * | |
3812 | * |------------- cluster # N-------------| | |
3813 | * |----- ex -----| |---- ex_right ----| | |
3814 | * |------ requested region ------| | |
3815 | * |================| | |
3816 | */ | |
3817 | if (map->m_lblk > ee_block) { | |
3818 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3819 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3820 | } | |
d8990240 AK |
3821 | |
3822 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3823 | return 1; |
3824 | } | |
d8990240 AK |
3825 | |
3826 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3827 | return 0; |
3828 | } | |
3829 | ||
3830 | ||
c278bfec | 3831 | /* |
f5ab0d1f MC |
3832 | * Block allocation/map/preallocation routine for extents based files |
3833 | * | |
3834 | * | |
c278bfec | 3835 | * Need to be called with |
0e855ac8 AK |
3836 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3837 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3838 | * |
3839 | * return > 0, number of of blocks already mapped/allocated | |
3840 | * if create == 0 and these are pre-allocated blocks | |
3841 | * buffer head is unmapped | |
3842 | * otherwise blocks are mapped | |
3843 | * | |
3844 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3845 | * buffer head is unmapped | |
3846 | * | |
3847 | * return < 0, error case. | |
c278bfec | 3848 | */ |
e35fd660 TT |
3849 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3850 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3851 | { |
3852 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3853 | struct ext4_extent newex, *ex, *ex2; |
3854 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3855 | ext4_fsblk_t newblock = 0; |
4d33b1ef TT |
3856 | int free_on_err = 0, err = 0, depth, ret; |
3857 | unsigned int allocated = 0, offset = 0; | |
81fdbb4a | 3858 | unsigned int allocated_clusters = 0; |
c9de560d | 3859 | struct ext4_allocation_request ar; |
8d5d02e6 | 3860 | ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio; |
4d33b1ef | 3861 | ext4_lblk_t cluster_offset; |
a86c6181 | 3862 | |
84fe3bef | 3863 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3864 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3865 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3866 | |
3867 | /* check in cache */ | |
7877191c | 3868 | if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) { |
b05e6ae5 | 3869 | if (!newex.ee_start_lo && !newex.ee_start_hi) { |
7b415bf6 AK |
3870 | if ((sbi->s_cluster_ratio > 1) && |
3871 | ext4_find_delalloc_cluster(inode, map->m_lblk, 0)) | |
3872 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
3873 | ||
c2177057 | 3874 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3875 | /* |
3876 | * block isn't allocated yet and | |
3877 | * user doesn't want to allocate it | |
3878 | */ | |
a86c6181 AT |
3879 | goto out2; |
3880 | } | |
3881 | /* we should allocate requested block */ | |
b05e6ae5 | 3882 | } else { |
a86c6181 | 3883 | /* block is already allocated */ |
7b415bf6 AK |
3884 | if (sbi->s_cluster_ratio > 1) |
3885 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
e35fd660 | 3886 | newblock = map->m_lblk |
8c55e204 | 3887 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3888 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3889 | /* number of remaining blocks in the extent */ |
b939e376 | 3890 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3891 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3892 | goto out; |
a86c6181 AT |
3893 | } |
3894 | } | |
3895 | ||
3896 | /* find extent for this block */ | |
e35fd660 | 3897 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3898 | if (IS_ERR(path)) { |
3899 | err = PTR_ERR(path); | |
3900 | path = NULL; | |
3901 | goto out2; | |
3902 | } | |
3903 | ||
3904 | depth = ext_depth(inode); | |
3905 | ||
3906 | /* | |
d0d856e8 RD |
3907 | * consistent leaf must not be empty; |
3908 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3909 | * this is why assert can't be put in ext4_ext_find_extent() |
3910 | */ | |
273df556 FM |
3911 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3912 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3913 | "lblock: %lu, depth: %d pblock %lld", |
3914 | (unsigned long) map->m_lblk, depth, | |
3915 | path[depth].p_block); | |
034fb4c9 SP |
3916 | err = -EIO; |
3917 | goto out2; | |
3918 | } | |
a86c6181 | 3919 | |
7e028976 AM |
3920 | ex = path[depth].p_ext; |
3921 | if (ex) { | |
725d26d3 | 3922 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3923 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3924 | unsigned short ee_len; |
471d4011 SB |
3925 | |
3926 | /* | |
471d4011 | 3927 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3928 | * we split out initialized portions during a write. |
471d4011 | 3929 | */ |
a2df2a63 | 3930 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
3931 | |
3932 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
3933 | ||
d0d856e8 | 3934 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
3935 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
3936 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 3937 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3938 | allocated = ee_len - (map->m_lblk - ee_block); |
3939 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3940 | ee_block, ee_len, newblock); | |
56055d3a | 3941 | |
e861304b | 3942 | /* |
7877191c LC |
3943 | * Do not put uninitialized extent |
3944 | * in the cache | |
e861304b | 3945 | */ |
7877191c LC |
3946 | if (!ext4_ext_is_uninitialized(ex)) { |
3947 | ext4_ext_put_in_cache(inode, ee_block, | |
3948 | ee_len, ee_start); | |
3949 | goto out; | |
f7d0d379 | 3950 | } |
7877191c LC |
3951 | ret = ext4_ext_handle_uninitialized_extents( |
3952 | handle, inode, map, path, flags, | |
3953 | allocated, newblock); | |
3954 | return ret; | |
a86c6181 AT |
3955 | } |
3956 | } | |
3957 | ||
7b415bf6 AK |
3958 | if ((sbi->s_cluster_ratio > 1) && |
3959 | ext4_find_delalloc_cluster(inode, map->m_lblk, 0)) | |
3960 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
3961 | ||
a86c6181 | 3962 | /* |
d0d856e8 | 3963 | * requested block isn't allocated yet; |
a86c6181 AT |
3964 | * we couldn't try to create block if create flag is zero |
3965 | */ | |
c2177057 | 3966 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3967 | /* |
3968 | * put just found gap into cache to speed up | |
3969 | * subsequent requests | |
3970 | */ | |
e35fd660 | 3971 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
3972 | goto out2; |
3973 | } | |
4d33b1ef | 3974 | |
a86c6181 | 3975 | /* |
c2ea3fde | 3976 | * Okay, we need to do block allocation. |
63f57933 | 3977 | */ |
7b415bf6 | 3978 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
3979 | newex.ee_block = cpu_to_le32(map->m_lblk); |
3980 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
3981 | ||
3982 | /* | |
3983 | * If we are doing bigalloc, check to see if the extent returned | |
3984 | * by ext4_ext_find_extent() implies a cluster we can use. | |
3985 | */ | |
3986 | if (cluster_offset && ex && | |
d8990240 | 3987 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
3988 | ar.len = allocated = map->m_len; |
3989 | newblock = map->m_pblk; | |
7b415bf6 | 3990 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
3991 | goto got_allocated_blocks; |
3992 | } | |
a86c6181 | 3993 | |
c9de560d | 3994 | /* find neighbour allocated blocks */ |
e35fd660 | 3995 | ar.lleft = map->m_lblk; |
c9de560d AT |
3996 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
3997 | if (err) | |
3998 | goto out2; | |
e35fd660 | 3999 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4000 | ex2 = NULL; |
4001 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4002 | if (err) |
4003 | goto out2; | |
25d14f98 | 4004 | |
4d33b1ef TT |
4005 | /* Check if the extent after searching to the right implies a |
4006 | * cluster we can use. */ | |
4007 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4008 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4009 | ar.len = allocated = map->m_len; |
4010 | newblock = map->m_pblk; | |
7b415bf6 | 4011 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4012 | goto got_allocated_blocks; |
4013 | } | |
4014 | ||
749269fa AA |
4015 | /* |
4016 | * See if request is beyond maximum number of blocks we can have in | |
4017 | * a single extent. For an initialized extent this limit is | |
4018 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4019 | * EXT_UNINIT_MAX_LEN. | |
4020 | */ | |
e35fd660 | 4021 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4022 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4023 | map->m_len = EXT_INIT_MAX_LEN; |
4024 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4025 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4026 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4027 | |
e35fd660 | 4028 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4029 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4030 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4031 | if (err) |
b939e376 | 4032 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4033 | else |
e35fd660 | 4034 | allocated = map->m_len; |
c9de560d AT |
4035 | |
4036 | /* allocate new block */ | |
4037 | ar.inode = inode; | |
e35fd660 TT |
4038 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4039 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4040 | /* |
4041 | * We calculate the offset from the beginning of the cluster | |
4042 | * for the logical block number, since when we allocate a | |
4043 | * physical cluster, the physical block should start at the | |
4044 | * same offset from the beginning of the cluster. This is | |
4045 | * needed so that future calls to get_implied_cluster_alloc() | |
4046 | * work correctly. | |
4047 | */ | |
4048 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4049 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4050 | ar.goal -= offset; | |
4051 | ar.logical -= offset; | |
c9de560d AT |
4052 | if (S_ISREG(inode->i_mode)) |
4053 | ar.flags = EXT4_MB_HINT_DATA; | |
4054 | else | |
4055 | /* disable in-core preallocation for non-regular files */ | |
4056 | ar.flags = 0; | |
556b27ab VH |
4057 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4058 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4059 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4060 | if (!newblock) |
4061 | goto out2; | |
84fe3bef | 4062 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4063 | ar.goal, newblock, allocated); |
4d33b1ef | 4064 | free_on_err = 1; |
7b415bf6 | 4065 | allocated_clusters = ar.len; |
4d33b1ef TT |
4066 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4067 | if (ar.len > allocated) | |
4068 | ar.len = allocated; | |
a86c6181 | 4069 | |
4d33b1ef | 4070 | got_allocated_blocks: |
a86c6181 | 4071 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4072 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4073 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4074 | /* Mark uninitialized */ |
4075 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4076 | ext4_ext_mark_uninitialized(&newex); |
8d5d02e6 | 4077 | /* |
744692dc | 4078 | * io_end structure was created for every IO write to an |
25985edc | 4079 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4080 | * here we flag the IO that really needs the conversion. |
5f524950 | 4081 | * For non asycn direct IO case, flag the inode state |
25985edc | 4082 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4083 | */ |
744692dc | 4084 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
0edeb71d TM |
4085 | if (io) |
4086 | ext4_set_io_unwritten_flag(inode, io); | |
4087 | else | |
19f5fb7a TT |
4088 | ext4_set_inode_state(inode, |
4089 | EXT4_STATE_DIO_UNWRITTEN); | |
5f524950 | 4090 | } |
744692dc | 4091 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4092 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4093 | } |
c8d46e41 | 4094 | |
a4e5d88b DM |
4095 | err = 0; |
4096 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4097 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4098 | path, ar.len); | |
575a1d4b JZ |
4099 | if (!err) |
4100 | err = ext4_ext_insert_extent(handle, inode, path, | |
4101 | &newex, flags); | |
4d33b1ef | 4102 | if (err && free_on_err) { |
7132de74 MP |
4103 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4104 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4105 | /* free data blocks we just allocated */ |
c9de560d AT |
4106 | /* not a good idea to call discard here directly, |
4107 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4108 | ext4_discard_preallocations(inode); |
7dc57615 | 4109 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4110 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4111 | goto out2; |
315054f0 | 4112 | } |
a86c6181 | 4113 | |
a86c6181 | 4114 | /* previous routine could use block we allocated */ |
bf89d16f | 4115 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4116 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4117 | if (allocated > map->m_len) |
4118 | allocated = map->m_len; | |
4119 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4120 | |
5f634d06 AK |
4121 | /* |
4122 | * Update reserved blocks/metadata blocks after successful | |
4123 | * block allocation which had been deferred till now. | |
4124 | */ | |
7b415bf6 | 4125 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4126 | unsigned int reserved_clusters; |
7b415bf6 | 4127 | /* |
81fdbb4a | 4128 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4129 | */ |
4130 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4131 | map->m_lblk, allocated); | |
4132 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4133 | if (reserved_clusters) { | |
4134 | /* | |
4135 | * We have clusters reserved for this range. | |
4136 | * But since we are not doing actual allocation | |
4137 | * and are simply using blocks from previously | |
4138 | * allocated cluster, we should release the | |
4139 | * reservation and not claim quota. | |
4140 | */ | |
4141 | ext4_da_update_reserve_space(inode, | |
4142 | reserved_clusters, 0); | |
4143 | } | |
4144 | } else { | |
4145 | BUG_ON(allocated_clusters < reserved_clusters); | |
4146 | /* We will claim quota for all newly allocated blocks.*/ | |
4147 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4148 | 1); | |
4149 | if (reserved_clusters < allocated_clusters) { | |
5356f261 | 4150 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4151 | int reservation = allocated_clusters - |
4152 | reserved_clusters; | |
4153 | /* | |
4154 | * It seems we claimed few clusters outside of | |
4155 | * the range of this allocation. We should give | |
4156 | * it back to the reservation pool. This can | |
4157 | * happen in the following case: | |
4158 | * | |
4159 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4160 | * cluster has 4 blocks. Thus, the clusters | |
4161 | * are [0-3],[4-7],[8-11]... | |
4162 | * * First comes delayed allocation write for | |
4163 | * logical blocks 10 & 11. Since there were no | |
4164 | * previous delayed allocated blocks in the | |
4165 | * range [8-11], we would reserve 1 cluster | |
4166 | * for this write. | |
4167 | * * Next comes write for logical blocks 3 to 8. | |
4168 | * In this case, we will reserve 2 clusters | |
4169 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4170 | * that range has a delayed allocated blocks. | |
4171 | * Thus total reserved clusters now becomes 3. | |
4172 | * * Now, during the delayed allocation writeout | |
4173 | * time, we will first write blocks [3-8] and | |
4174 | * allocate 3 clusters for writing these | |
4175 | * blocks. Also, we would claim all these | |
4176 | * three clusters above. | |
4177 | * * Now when we come here to writeout the | |
4178 | * blocks [10-11], we would expect to claim | |
4179 | * the reservation of 1 cluster we had made | |
4180 | * (and we would claim it since there are no | |
4181 | * more delayed allocated blocks in the range | |
4182 | * [8-11]. But our reserved cluster count had | |
4183 | * already gone to 0. | |
4184 | * | |
4185 | * Thus, at the step 4 above when we determine | |
4186 | * that there are still some unwritten delayed | |
4187 | * allocated blocks outside of our current | |
4188 | * block range, we should increment the | |
4189 | * reserved clusters count so that when the | |
4190 | * remaining blocks finally gets written, we | |
4191 | * could claim them. | |
4192 | */ | |
5356f261 AK |
4193 | dquot_reserve_block(inode, |
4194 | EXT4_C2B(sbi, reservation)); | |
4195 | spin_lock(&ei->i_block_reservation_lock); | |
4196 | ei->i_reserved_data_blocks += reservation; | |
4197 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 AK |
4198 | } |
4199 | } | |
4200 | } | |
5f634d06 | 4201 | |
b436b9be JK |
4202 | /* |
4203 | * Cache the extent and update transaction to commit on fdatasync only | |
4204 | * when it is _not_ an uninitialized extent. | |
4205 | */ | |
4206 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 4207 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
4208 | ext4_update_inode_fsync_trans(handle, inode, 1); |
4209 | } else | |
4210 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 4211 | out: |
e35fd660 TT |
4212 | if (allocated > map->m_len) |
4213 | allocated = map->m_len; | |
a86c6181 | 4214 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4215 | map->m_flags |= EXT4_MAP_MAPPED; |
4216 | map->m_pblk = newblock; | |
4217 | map->m_len = allocated; | |
a86c6181 AT |
4218 | out2: |
4219 | if (path) { | |
4220 | ext4_ext_drop_refs(path); | |
4221 | kfree(path); | |
4222 | } | |
e861304b | 4223 | |
e7b319e3 | 4224 | trace_ext4_ext_map_blocks_exit(inode, map->m_lblk, |
7877191c | 4225 | newblock, map->m_len, err ? err : allocated); |
e7b319e3 | 4226 | |
7877191c | 4227 | return err ? err : allocated; |
a86c6181 AT |
4228 | } |
4229 | ||
cf108bca | 4230 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
4231 | { |
4232 | struct address_space *mapping = inode->i_mapping; | |
4233 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 4234 | ext4_lblk_t last_block; |
a86c6181 | 4235 | handle_t *handle; |
189e868f | 4236 | loff_t page_len; |
a86c6181 AT |
4237 | int err = 0; |
4238 | ||
3889fd57 JZ |
4239 | /* |
4240 | * finish any pending end_io work so we won't run the risk of | |
4241 | * converting any truncated blocks to initialized later | |
4242 | */ | |
4243 | ext4_flush_completed_IO(inode); | |
4244 | ||
a86c6181 AT |
4245 | /* |
4246 | * probably first extent we're gonna free will be last in block | |
4247 | */ | |
f3bd1f3f | 4248 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 4249 | handle = ext4_journal_start(inode, err); |
cf108bca | 4250 | if (IS_ERR(handle)) |
a86c6181 | 4251 | return; |
a86c6181 | 4252 | |
189e868f AH |
4253 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
4254 | page_len = PAGE_CACHE_SIZE - | |
4255 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4256 | ||
4257 | err = ext4_discard_partial_page_buffers(handle, | |
4258 | mapping, inode->i_size, page_len, 0); | |
4259 | ||
4260 | if (err) | |
4261 | goto out_stop; | |
4262 | } | |
a86c6181 | 4263 | |
9ddfc3dc JK |
4264 | if (ext4_orphan_add(handle, inode)) |
4265 | goto out_stop; | |
4266 | ||
0e855ac8 | 4267 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
4268 | ext4_ext_invalidate_cache(inode); |
4269 | ||
c2ea3fde | 4270 | ext4_discard_preallocations(inode); |
c9de560d | 4271 | |
a86c6181 | 4272 | /* |
d0d856e8 RD |
4273 | * TODO: optimization is possible here. |
4274 | * Probably we need not scan at all, | |
4275 | * because page truncation is enough. | |
a86c6181 | 4276 | */ |
a86c6181 AT |
4277 | |
4278 | /* we have to know where to truncate from in crash case */ | |
4279 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4280 | ext4_mark_inode_dirty(handle, inode); | |
4281 | ||
4282 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4283 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
5f95d21f | 4284 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
a86c6181 AT |
4285 | |
4286 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
4287 | * transaction synchronous. |
4288 | */ | |
a86c6181 | 4289 | if (IS_SYNC(inode)) |
0390131b | 4290 | ext4_handle_sync(handle); |
a86c6181 | 4291 | |
9ddfc3dc | 4292 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
4293 | |
4294 | out_stop: | |
a86c6181 | 4295 | /* |
d0d856e8 | 4296 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
4297 | * then we need to clear up the orphan record which we created above. |
4298 | * However, if this was a real unlink then we were called by | |
4299 | * ext4_delete_inode(), and we allow that function to clean up the | |
4300 | * orphan info for us. | |
4301 | */ | |
4302 | if (inode->i_nlink) | |
4303 | ext4_orphan_del(handle, inode); | |
4304 | ||
ef737728 SR |
4305 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4306 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
4307 | ext4_journal_stop(handle); |
4308 | } | |
4309 | ||
fd28784a AK |
4310 | static void ext4_falloc_update_inode(struct inode *inode, |
4311 | int mode, loff_t new_size, int update_ctime) | |
4312 | { | |
4313 | struct timespec now; | |
4314 | ||
4315 | if (update_ctime) { | |
4316 | now = current_fs_time(inode->i_sb); | |
4317 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4318 | inode->i_ctime = now; | |
4319 | } | |
4320 | /* | |
4321 | * Update only when preallocation was requested beyond | |
4322 | * the file size. | |
4323 | */ | |
cf17fea6 AK |
4324 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4325 | if (new_size > i_size_read(inode)) | |
4326 | i_size_write(inode, new_size); | |
4327 | if (new_size > EXT4_I(inode)->i_disksize) | |
4328 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4329 | } else { |
4330 | /* | |
4331 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4332 | * can proceed even if the new size is the same as i_size. | |
4333 | */ | |
4334 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4335 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4336 | } |
4337 | ||
4338 | } | |
4339 | ||
a2df2a63 | 4340 | /* |
2fe17c10 | 4341 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4342 | * operation, which gets called from sys_fallocate system call. |
4343 | * For block-mapped files, posix_fallocate should fall back to the method | |
4344 | * of writing zeroes to the required new blocks (the same behavior which is | |
4345 | * expected for file systems which do not support fallocate() system call). | |
4346 | */ | |
2fe17c10 | 4347 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4348 | { |
2fe17c10 | 4349 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 4350 | handle_t *handle; |
fd28784a | 4351 | loff_t new_size; |
498e5f24 | 4352 | unsigned int max_blocks; |
a2df2a63 AA |
4353 | int ret = 0; |
4354 | int ret2 = 0; | |
4355 | int retries = 0; | |
a4e5d88b | 4356 | int flags; |
2ed88685 | 4357 | struct ext4_map_blocks map; |
a2df2a63 AA |
4358 | unsigned int credits, blkbits = inode->i_blkbits; |
4359 | ||
4360 | /* | |
4361 | * currently supporting (pre)allocate mode for extent-based | |
4362 | * files _only_ | |
4363 | */ | |
12e9b892 | 4364 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
a2df2a63 AA |
4365 | return -EOPNOTSUPP; |
4366 | ||
a4bb6b64 AH |
4367 | /* Return error if mode is not supported */ |
4368 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4369 | return -EOPNOTSUPP; | |
4370 | ||
4371 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4372 | return ext4_punch_hole(file, offset, len); | |
4373 | ||
0562e0ba | 4374 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4375 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4376 | /* |
4377 | * We can't just convert len to max_blocks because | |
4378 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4379 | */ | |
a2df2a63 | 4380 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4381 | - map.m_lblk; |
a2df2a63 | 4382 | /* |
f3bd1f3f | 4383 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4384 | */ |
f3bd1f3f | 4385 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4386 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4387 | ret = inode_newsize_ok(inode, (len + offset)); |
4388 | if (ret) { | |
4389 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4390 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4391 | return ret; |
4392 | } | |
3c6fe770 | 4393 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4394 | if (mode & FALLOC_FL_KEEP_SIZE) |
4395 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4396 | /* |
4397 | * Don't normalize the request if it can fit in one extent so | |
4398 | * that it doesn't get unnecessarily split into multiple | |
4399 | * extents. | |
4400 | */ | |
4401 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4402 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
a2df2a63 AA |
4403 | retry: |
4404 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4405 | map.m_lblk = map.m_lblk + ret; |
4406 | map.m_len = max_blocks = max_blocks - ret; | |
a2df2a63 AA |
4407 | handle = ext4_journal_start(inode, credits); |
4408 | if (IS_ERR(handle)) { | |
4409 | ret = PTR_ERR(handle); | |
4410 | break; | |
4411 | } | |
a4e5d88b | 4412 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4413 | if (ret <= 0) { |
2c98615d AK |
4414 | #ifdef EXT4FS_DEBUG |
4415 | WARN_ON(ret <= 0); | |
e35fd660 | 4416 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
2c98615d | 4417 | "returned error inode#%lu, block=%u, " |
9fd9784c | 4418 | "max_blocks=%u", __func__, |
a6371b63 | 4419 | inode->i_ino, map.m_lblk, max_blocks); |
2c98615d | 4420 | #endif |
a2df2a63 AA |
4421 | ext4_mark_inode_dirty(handle, inode); |
4422 | ret2 = ext4_journal_stop(handle); | |
4423 | break; | |
4424 | } | |
2ed88685 | 4425 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4426 | blkbits) >> blkbits)) |
4427 | new_size = offset + len; | |
4428 | else | |
29ae07b7 | 4429 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4430 | |
fd28784a | 4431 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4432 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4433 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4434 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4435 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4436 | ret2 = ext4_journal_stop(handle); |
4437 | if (ret2) | |
4438 | break; | |
4439 | } | |
fd28784a AK |
4440 | if (ret == -ENOSPC && |
4441 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4442 | ret = 0; | |
a2df2a63 | 4443 | goto retry; |
a2df2a63 | 4444 | } |
55bd725a | 4445 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4446 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4447 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4448 | return ret > 0 ? ret2 : ret; |
4449 | } | |
6873fa0d | 4450 | |
0031462b MC |
4451 | /* |
4452 | * This function convert a range of blocks to written extents | |
4453 | * The caller of this function will pass the start offset and the size. | |
4454 | * all unwritten extents within this range will be converted to | |
4455 | * written extents. | |
4456 | * | |
4457 | * This function is called from the direct IO end io call back | |
4458 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4459 | * Returns 0 on success. |
0031462b MC |
4460 | */ |
4461 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4462 | ssize_t len) |
0031462b MC |
4463 | { |
4464 | handle_t *handle; | |
0031462b MC |
4465 | unsigned int max_blocks; |
4466 | int ret = 0; | |
4467 | int ret2 = 0; | |
2ed88685 | 4468 | struct ext4_map_blocks map; |
0031462b MC |
4469 | unsigned int credits, blkbits = inode->i_blkbits; |
4470 | ||
2ed88685 | 4471 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4472 | /* |
4473 | * We can't just convert len to max_blocks because | |
4474 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4475 | */ | |
2ed88685 TT |
4476 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4477 | map.m_lblk); | |
0031462b MC |
4478 | /* |
4479 | * credits to insert 1 extent into extent tree | |
4480 | */ | |
4481 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4482 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4483 | map.m_lblk += ret; |
4484 | map.m_len = (max_blocks -= ret); | |
0031462b MC |
4485 | handle = ext4_journal_start(inode, credits); |
4486 | if (IS_ERR(handle)) { | |
4487 | ret = PTR_ERR(handle); | |
4488 | break; | |
4489 | } | |
2ed88685 | 4490 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4491 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
0031462b MC |
4492 | if (ret <= 0) { |
4493 | WARN_ON(ret <= 0); | |
92b97816 TT |
4494 | ext4_msg(inode->i_sb, KERN_ERR, |
4495 | "%s:%d: inode #%lu: block %u: len %u: " | |
4496 | "ext4_ext_map_blocks returned %d", | |
4497 | __func__, __LINE__, inode->i_ino, map.m_lblk, | |
4498 | map.m_len, ret); | |
0031462b MC |
4499 | } |
4500 | ext4_mark_inode_dirty(handle, inode); | |
4501 | ret2 = ext4_journal_stop(handle); | |
4502 | if (ret <= 0 || ret2 ) | |
4503 | break; | |
4504 | } | |
4505 | return ret > 0 ? ret2 : ret; | |
4506 | } | |
6d9c85eb | 4507 | |
6873fa0d ES |
4508 | /* |
4509 | * Callback function called for each extent to gather FIEMAP information. | |
4510 | */ | |
c03f8aa9 | 4511 | static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next, |
6873fa0d ES |
4512 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
4513 | void *data) | |
4514 | { | |
6873fa0d ES |
4515 | __u64 logical; |
4516 | __u64 physical; | |
4517 | __u64 length; | |
4518 | __u32 flags = 0; | |
6d9c85eb YY |
4519 | int ret = 0; |
4520 | struct fiemap_extent_info *fieinfo = data; | |
4521 | unsigned char blksize_bits; | |
6873fa0d | 4522 | |
6d9c85eb YY |
4523 | blksize_bits = inode->i_sb->s_blocksize_bits; |
4524 | logical = (__u64)newex->ec_block << blksize_bits; | |
6873fa0d | 4525 | |
b05e6ae5 | 4526 | if (newex->ec_start == 0) { |
6d9c85eb YY |
4527 | /* |
4528 | * No extent in extent-tree contains block @newex->ec_start, | |
4529 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
4530 | * | |
4531 | * Holes or delayed-extents are processed as follows. | |
4532 | * 1. lookup dirty pages with specified range in pagecache. | |
4533 | * If no page is got, then there is no delayed-extent and | |
4534 | * return with EXT_CONTINUE. | |
4535 | * 2. find the 1st mapped buffer, | |
4536 | * 3. check if the mapped buffer is both in the request range | |
4537 | * and a delayed buffer. If not, there is no delayed-extent, | |
4538 | * then return. | |
4539 | * 4. a delayed-extent is found, the extent will be collected. | |
4540 | */ | |
4541 | ext4_lblk_t end = 0; | |
4542 | pgoff_t last_offset; | |
4543 | pgoff_t offset; | |
4544 | pgoff_t index; | |
b221349f | 4545 | pgoff_t start_index = 0; |
6d9c85eb | 4546 | struct page **pages = NULL; |
6873fa0d | 4547 | struct buffer_head *bh = NULL; |
6d9c85eb YY |
4548 | struct buffer_head *head = NULL; |
4549 | unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *); | |
4550 | ||
4551 | pages = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4552 | if (pages == NULL) | |
4553 | return -ENOMEM; | |
6873fa0d ES |
4554 | |
4555 | offset = logical >> PAGE_SHIFT; | |
6d9c85eb YY |
4556 | repeat: |
4557 | last_offset = offset; | |
4558 | head = NULL; | |
4559 | ret = find_get_pages_tag(inode->i_mapping, &offset, | |
4560 | PAGECACHE_TAG_DIRTY, nr_pages, pages); | |
4561 | ||
4562 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4563 | /* First time, try to find a mapped buffer. */ | |
4564 | if (ret == 0) { | |
4565 | out: | |
4566 | for (index = 0; index < ret; index++) | |
4567 | page_cache_release(pages[index]); | |
4568 | /* just a hole. */ | |
4569 | kfree(pages); | |
4570 | return EXT_CONTINUE; | |
4571 | } | |
b221349f | 4572 | index = 0; |
6873fa0d | 4573 | |
b221349f | 4574 | next_page: |
6d9c85eb | 4575 | /* Try to find the 1st mapped buffer. */ |
b221349f | 4576 | end = ((__u64)pages[index]->index << PAGE_SHIFT) >> |
6d9c85eb | 4577 | blksize_bits; |
b221349f | 4578 | if (!page_has_buffers(pages[index])) |
6d9c85eb | 4579 | goto out; |
b221349f | 4580 | head = page_buffers(pages[index]); |
6d9c85eb YY |
4581 | if (!head) |
4582 | goto out; | |
6873fa0d | 4583 | |
b221349f | 4584 | index++; |
6d9c85eb YY |
4585 | bh = head; |
4586 | do { | |
b221349f YY |
4587 | if (end >= newex->ec_block + |
4588 | newex->ec_len) | |
4589 | /* The buffer is out of | |
4590 | * the request range. | |
4591 | */ | |
4592 | goto out; | |
4593 | ||
4594 | if (buffer_mapped(bh) && | |
4595 | end >= newex->ec_block) { | |
4596 | start_index = index - 1; | |
6d9c85eb | 4597 | /* get the 1st mapped buffer. */ |
6d9c85eb YY |
4598 | goto found_mapped_buffer; |
4599 | } | |
b221349f | 4600 | |
6d9c85eb YY |
4601 | bh = bh->b_this_page; |
4602 | end++; | |
4603 | } while (bh != head); | |
6873fa0d | 4604 | |
b221349f YY |
4605 | /* No mapped buffer in the range found in this page, |
4606 | * We need to look up next page. | |
4607 | */ | |
4608 | if (index >= ret) { | |
4609 | /* There is no page left, but we need to limit | |
4610 | * newex->ec_len. | |
4611 | */ | |
4612 | newex->ec_len = end - newex->ec_block; | |
4613 | goto out; | |
4614 | } | |
4615 | goto next_page; | |
6873fa0d | 4616 | } else { |
6d9c85eb YY |
4617 | /*Find contiguous delayed buffers. */ |
4618 | if (ret > 0 && pages[0]->index == last_offset) | |
4619 | head = page_buffers(pages[0]); | |
4620 | bh = head; | |
b221349f YY |
4621 | index = 1; |
4622 | start_index = 0; | |
6873fa0d | 4623 | } |
6d9c85eb YY |
4624 | |
4625 | found_mapped_buffer: | |
4626 | if (bh != NULL && buffer_delay(bh)) { | |
4627 | /* 1st or contiguous delayed buffer found. */ | |
4628 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4629 | /* | |
4630 | * 1st delayed buffer found, record | |
4631 | * the start of extent. | |
4632 | */ | |
4633 | flags |= FIEMAP_EXTENT_DELALLOC; | |
4634 | newex->ec_block = end; | |
4635 | logical = (__u64)end << blksize_bits; | |
4636 | } | |
4637 | /* Find contiguous delayed buffers. */ | |
4638 | do { | |
4639 | if (!buffer_delay(bh)) | |
4640 | goto found_delayed_extent; | |
4641 | bh = bh->b_this_page; | |
4642 | end++; | |
4643 | } while (bh != head); | |
4644 | ||
b221349f | 4645 | for (; index < ret; index++) { |
6d9c85eb YY |
4646 | if (!page_has_buffers(pages[index])) { |
4647 | bh = NULL; | |
4648 | break; | |
4649 | } | |
4650 | head = page_buffers(pages[index]); | |
4651 | if (!head) { | |
4652 | bh = NULL; | |
4653 | break; | |
4654 | } | |
b221349f | 4655 | |
6d9c85eb | 4656 | if (pages[index]->index != |
b221349f YY |
4657 | pages[start_index]->index + index |
4658 | - start_index) { | |
6d9c85eb YY |
4659 | /* Blocks are not contiguous. */ |
4660 | bh = NULL; | |
4661 | break; | |
4662 | } | |
4663 | bh = head; | |
4664 | do { | |
4665 | if (!buffer_delay(bh)) | |
4666 | /* Delayed-extent ends. */ | |
4667 | goto found_delayed_extent; | |
4668 | bh = bh->b_this_page; | |
4669 | end++; | |
4670 | } while (bh != head); | |
4671 | } | |
4672 | } else if (!(flags & FIEMAP_EXTENT_DELALLOC)) | |
4673 | /* a hole found. */ | |
4674 | goto out; | |
4675 | ||
4676 | found_delayed_extent: | |
4677 | newex->ec_len = min(end - newex->ec_block, | |
4678 | (ext4_lblk_t)EXT_INIT_MAX_LEN); | |
4679 | if (ret == nr_pages && bh != NULL && | |
4680 | newex->ec_len < EXT_INIT_MAX_LEN && | |
4681 | buffer_delay(bh)) { | |
4682 | /* Have not collected an extent and continue. */ | |
4683 | for (index = 0; index < ret; index++) | |
4684 | page_cache_release(pages[index]); | |
4685 | goto repeat; | |
6873fa0d | 4686 | } |
6d9c85eb YY |
4687 | |
4688 | for (index = 0; index < ret; index++) | |
4689 | page_cache_release(pages[index]); | |
4690 | kfree(pages); | |
6873fa0d ES |
4691 | } |
4692 | ||
4693 | physical = (__u64)newex->ec_start << blksize_bits; | |
4694 | length = (__u64)newex->ec_len << blksize_bits; | |
4695 | ||
4696 | if (ex && ext4_ext_is_uninitialized(ex)) | |
4697 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
4698 | ||
c03f8aa9 | 4699 | if (next == EXT_MAX_BLOCKS) |
6873fa0d ES |
4700 | flags |= FIEMAP_EXTENT_LAST; |
4701 | ||
6d9c85eb | 4702 | ret = fiemap_fill_next_extent(fieinfo, logical, physical, |
6873fa0d | 4703 | length, flags); |
6d9c85eb YY |
4704 | if (ret < 0) |
4705 | return ret; | |
4706 | if (ret == 1) | |
6873fa0d | 4707 | return EXT_BREAK; |
6873fa0d ES |
4708 | return EXT_CONTINUE; |
4709 | } | |
6873fa0d ES |
4710 | /* fiemap flags we can handle specified here */ |
4711 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4712 | ||
3a06d778 AK |
4713 | static int ext4_xattr_fiemap(struct inode *inode, |
4714 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4715 | { |
4716 | __u64 physical = 0; | |
4717 | __u64 length; | |
4718 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4719 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4720 | int error = 0; | |
4721 | ||
4722 | /* in-inode? */ | |
19f5fb7a | 4723 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4724 | struct ext4_iloc iloc; |
4725 | int offset; /* offset of xattr in inode */ | |
4726 | ||
4727 | error = ext4_get_inode_loc(inode, &iloc); | |
4728 | if (error) | |
4729 | return error; | |
4730 | physical = iloc.bh->b_blocknr << blockbits; | |
4731 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4732 | EXT4_I(inode)->i_extra_isize; | |
4733 | physical += offset; | |
4734 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4735 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4736 | brelse(iloc.bh); |
6873fa0d ES |
4737 | } else { /* external block */ |
4738 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4739 | length = inode->i_sb->s_blocksize; | |
4740 | } | |
4741 | ||
4742 | if (physical) | |
4743 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4744 | length, flags); | |
4745 | return (error < 0 ? error : 0); | |
4746 | } | |
4747 | ||
a4bb6b64 AH |
4748 | /* |
4749 | * ext4_ext_punch_hole | |
4750 | * | |
4751 | * Punches a hole of "length" bytes in a file starting | |
4752 | * at byte "offset" | |
4753 | * | |
4754 | * @inode: The inode of the file to punch a hole in | |
4755 | * @offset: The starting byte offset of the hole | |
4756 | * @length: The length of the hole | |
4757 | * | |
4758 | * Returns the number of blocks removed or negative on err | |
4759 | */ | |
4760 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4761 | { | |
4762 | struct inode *inode = file->f_path.dentry->d_inode; | |
4763 | struct super_block *sb = inode->i_sb; | |
5f95d21f | 4764 | ext4_lblk_t first_block, stop_block; |
a4bb6b64 | 4765 | struct address_space *mapping = inode->i_mapping; |
a4bb6b64 | 4766 | handle_t *handle; |
ba06208a AH |
4767 | loff_t first_page, last_page, page_len; |
4768 | loff_t first_page_offset, last_page_offset; | |
5f95d21f | 4769 | int credits, err = 0; |
a4bb6b64 | 4770 | |
2be4751b AH |
4771 | /* No need to punch hole beyond i_size */ |
4772 | if (offset >= inode->i_size) | |
4773 | return 0; | |
4774 | ||
4775 | /* | |
4776 | * If the hole extends beyond i_size, set the hole | |
4777 | * to end after the page that contains i_size | |
4778 | */ | |
4779 | if (offset + length > inode->i_size) { | |
4780 | length = inode->i_size + | |
4781 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4782 | offset; | |
4783 | } | |
4784 | ||
a4bb6b64 AH |
4785 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4786 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4787 | ||
4788 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4789 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4790 | ||
4791 | /* | |
4792 | * Write out all dirty pages to avoid race conditions | |
4793 | * Then release them. | |
4794 | */ | |
4795 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4796 | err = filemap_write_and_wait_range(mapping, | |
2be4751b | 4797 | offset, offset + length - 1); |
a4bb6b64 | 4798 | |
2be4751b AH |
4799 | if (err) |
4800 | return err; | |
a4bb6b64 AH |
4801 | } |
4802 | ||
4803 | /* Now release the pages */ | |
4804 | if (last_page_offset > first_page_offset) { | |
5e44f8c3 HD |
4805 | truncate_pagecache_range(inode, first_page_offset, |
4806 | last_page_offset - 1); | |
a4bb6b64 AH |
4807 | } |
4808 | ||
4809 | /* finish any pending end_io work */ | |
4810 | ext4_flush_completed_IO(inode); | |
4811 | ||
4812 | credits = ext4_writepage_trans_blocks(inode); | |
4813 | handle = ext4_journal_start(inode, credits); | |
4814 | if (IS_ERR(handle)) | |
4815 | return PTR_ERR(handle); | |
4816 | ||
4817 | err = ext4_orphan_add(handle, inode); | |
4818 | if (err) | |
4819 | goto out; | |
4820 | ||
4821 | /* | |
ba06208a AH |
4822 | * Now we need to zero out the non-page-aligned data in the |
4823 | * pages at the start and tail of the hole, and unmap the buffer | |
4824 | * heads for the block aligned regions of the page that were | |
4825 | * completely zeroed. | |
a4bb6b64 | 4826 | */ |
ba06208a AH |
4827 | if (first_page > last_page) { |
4828 | /* | |
4829 | * If the file space being truncated is contained within a page | |
4830 | * just zero out and unmap the middle of that page | |
4831 | */ | |
4832 | err = ext4_discard_partial_page_buffers(handle, | |
4833 | mapping, offset, length, 0); | |
4834 | ||
4835 | if (err) | |
4836 | goto out; | |
4837 | } else { | |
4838 | /* | |
4839 | * zero out and unmap the partial page that contains | |
4840 | * the start of the hole | |
4841 | */ | |
4842 | page_len = first_page_offset - offset; | |
4843 | if (page_len > 0) { | |
4844 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4845 | offset, page_len, 0); | |
4846 | if (err) | |
4847 | goto out; | |
4848 | } | |
4849 | ||
4850 | /* | |
4851 | * zero out and unmap the partial page that contains | |
4852 | * the end of the hole | |
4853 | */ | |
4854 | page_len = offset + length - last_page_offset; | |
4855 | if (page_len > 0) { | |
4856 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4857 | last_page_offset, page_len, 0); | |
4858 | if (err) | |
4859 | goto out; | |
a4bb6b64 AH |
4860 | } |
4861 | } | |
4862 | ||
2be4751b AH |
4863 | /* |
4864 | * If i_size is contained in the last page, we need to | |
4865 | * unmap and zero the partial page after i_size | |
4866 | */ | |
4867 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4868 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4869 | ||
4870 | page_len = PAGE_CACHE_SIZE - | |
4871 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4872 | ||
4873 | if (page_len > 0) { | |
4874 | err = ext4_discard_partial_page_buffers(handle, | |
4875 | mapping, inode->i_size, page_len, 0); | |
4876 | ||
4877 | if (err) | |
4878 | goto out; | |
4879 | } | |
4880 | } | |
4881 | ||
5f95d21f LC |
4882 | first_block = (offset + sb->s_blocksize - 1) >> |
4883 | EXT4_BLOCK_SIZE_BITS(sb); | |
4884 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4885 | ||
a4bb6b64 | 4886 | /* If there are no blocks to remove, return now */ |
5f95d21f | 4887 | if (first_block >= stop_block) |
a4bb6b64 AH |
4888 | goto out; |
4889 | ||
4890 | down_write(&EXT4_I(inode)->i_data_sem); | |
4891 | ext4_ext_invalidate_cache(inode); | |
4892 | ext4_discard_preallocations(inode); | |
4893 | ||
5f95d21f | 4894 | err = ext4_ext_remove_space(inode, first_block, stop_block - 1); |
a4bb6b64 | 4895 | |
5f95d21f LC |
4896 | ext4_ext_invalidate_cache(inode); |
4897 | ext4_discard_preallocations(inode); | |
a4bb6b64 AH |
4898 | |
4899 | if (IS_SYNC(inode)) | |
4900 | ext4_handle_sync(handle); | |
4901 | ||
4902 | up_write(&EXT4_I(inode)->i_data_sem); | |
4903 | ||
4904 | out: | |
4905 | ext4_orphan_del(handle, inode); | |
4906 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
4907 | ext4_mark_inode_dirty(handle, inode); | |
4908 | ext4_journal_stop(handle); | |
4909 | return err; | |
4910 | } | |
6873fa0d ES |
4911 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4912 | __u64 start, __u64 len) | |
4913 | { | |
4914 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4915 | int error = 0; |
4916 | ||
4917 | /* fallback to generic here if not in extents fmt */ | |
12e9b892 | 4918 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4919 | return generic_block_fiemap(inode, fieinfo, start, len, |
4920 | ext4_get_block); | |
4921 | ||
4922 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4923 | return -EBADR; | |
4924 | ||
4925 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4926 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4927 | } else { | |
aca92ff6 LM |
4928 | ext4_lblk_t len_blks; |
4929 | __u64 last_blk; | |
4930 | ||
6873fa0d | 4931 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4932 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4933 | if (last_blk >= EXT_MAX_BLOCKS) |
4934 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4935 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4936 | |
4937 | /* | |
4938 | * Walk the extent tree gathering extent information. | |
4939 | * ext4_ext_fiemap_cb will push extents back to user. | |
4940 | */ | |
6873fa0d ES |
4941 | error = ext4_ext_walk_space(inode, start_blk, len_blks, |
4942 | ext4_ext_fiemap_cb, fieinfo); | |
6873fa0d ES |
4943 | } |
4944 | ||
4945 | return error; | |
4946 | } |