]>
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 | ||
dee1f973 DM |
55 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
56 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
57 | ||
7ac5990d DW |
58 | static __le32 ext4_extent_block_csum(struct inode *inode, |
59 | struct ext4_extent_header *eh) | |
60 | { | |
61 | struct ext4_inode_info *ei = EXT4_I(inode); | |
62 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
63 | __u32 csum; | |
64 | ||
65 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
66 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
67 | return cpu_to_le32(csum); | |
68 | } | |
69 | ||
70 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
71 | struct ext4_extent_header *eh) | |
72 | { | |
73 | struct ext4_extent_tail *et; | |
74 | ||
75 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
76 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
77 | return 1; | |
78 | ||
79 | et = find_ext4_extent_tail(eh); | |
80 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
81 | return 0; | |
82 | return 1; | |
83 | } | |
84 | ||
85 | static void ext4_extent_block_csum_set(struct inode *inode, | |
86 | struct ext4_extent_header *eh) | |
87 | { | |
88 | struct ext4_extent_tail *et; | |
89 | ||
90 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
91 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
92 | return; | |
93 | ||
94 | et = find_ext4_extent_tail(eh); | |
95 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
96 | } | |
97 | ||
d583fb87 AH |
98 | static int ext4_split_extent(handle_t *handle, |
99 | struct inode *inode, | |
100 | struct ext4_ext_path *path, | |
101 | struct ext4_map_blocks *map, | |
102 | int split_flag, | |
103 | int flags); | |
104 | ||
5f95d21f LC |
105 | static int ext4_split_extent_at(handle_t *handle, |
106 | struct inode *inode, | |
107 | struct ext4_ext_path *path, | |
108 | ext4_lblk_t split, | |
109 | int split_flag, | |
110 | int flags); | |
111 | ||
487caeef JK |
112 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
113 | struct inode *inode, | |
114 | int needed) | |
a86c6181 AT |
115 | { |
116 | int err; | |
117 | ||
0390131b FM |
118 | if (!ext4_handle_valid(handle)) |
119 | return 0; | |
a86c6181 | 120 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
121 | return 0; |
122 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 123 | if (err <= 0) |
9102e4fa | 124 | return err; |
487caeef | 125 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
126 | if (err == 0) |
127 | err = -EAGAIN; | |
487caeef JK |
128 | |
129 | return err; | |
a86c6181 AT |
130 | } |
131 | ||
132 | /* | |
133 | * could return: | |
134 | * - EROFS | |
135 | * - ENOMEM | |
136 | */ | |
137 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
138 | struct ext4_ext_path *path) | |
139 | { | |
140 | if (path->p_bh) { | |
141 | /* path points to block */ | |
142 | return ext4_journal_get_write_access(handle, path->p_bh); | |
143 | } | |
144 | /* path points to leaf/index in inode body */ | |
145 | /* we use in-core data, no need to protect them */ | |
146 | return 0; | |
147 | } | |
148 | ||
149 | /* | |
150 | * could return: | |
151 | * - EROFS | |
152 | * - ENOMEM | |
153 | * - EIO | |
154 | */ | |
9ea7a0df TT |
155 | #define ext4_ext_dirty(handle, inode, path) \ |
156 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
157 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
158 | handle_t *handle, struct inode *inode, | |
159 | struct ext4_ext_path *path) | |
a86c6181 AT |
160 | { |
161 | int err; | |
162 | if (path->p_bh) { | |
7ac5990d | 163 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 164 | /* path points to block */ |
9ea7a0df TT |
165 | err = __ext4_handle_dirty_metadata(where, line, handle, |
166 | inode, path->p_bh); | |
a86c6181 AT |
167 | } else { |
168 | /* path points to leaf/index in inode body */ | |
169 | err = ext4_mark_inode_dirty(handle, inode); | |
170 | } | |
171 | return err; | |
172 | } | |
173 | ||
f65e6fba | 174 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 175 | struct ext4_ext_path *path, |
725d26d3 | 176 | ext4_lblk_t block) |
a86c6181 | 177 | { |
a86c6181 | 178 | if (path) { |
81fdbb4a | 179 | int depth = path->p_depth; |
a86c6181 | 180 | struct ext4_extent *ex; |
a86c6181 | 181 | |
ad4fb9ca KM |
182 | /* |
183 | * Try to predict block placement assuming that we are | |
184 | * filling in a file which will eventually be | |
185 | * non-sparse --- i.e., in the case of libbfd writing | |
186 | * an ELF object sections out-of-order but in a way | |
187 | * the eventually results in a contiguous object or | |
188 | * executable file, or some database extending a table | |
189 | * space file. However, this is actually somewhat | |
190 | * non-ideal if we are writing a sparse file such as | |
191 | * qemu or KVM writing a raw image file that is going | |
192 | * to stay fairly sparse, since it will end up | |
193 | * fragmenting the file system's free space. Maybe we | |
194 | * should have some hueristics or some way to allow | |
195 | * userspace to pass a hint to file system, | |
b8d6568a | 196 | * especially if the latter case turns out to be |
ad4fb9ca KM |
197 | * common. |
198 | */ | |
7e028976 | 199 | ex = path[depth].p_ext; |
ad4fb9ca KM |
200 | if (ex) { |
201 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
202 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
203 | ||
204 | if (block > ext_block) | |
205 | return ext_pblk + (block - ext_block); | |
206 | else | |
207 | return ext_pblk - (ext_block - block); | |
208 | } | |
a86c6181 | 209 | |
d0d856e8 RD |
210 | /* it looks like index is empty; |
211 | * try to find starting block from index itself */ | |
a86c6181 AT |
212 | if (path[depth].p_bh) |
213 | return path[depth].p_bh->b_blocknr; | |
214 | } | |
215 | ||
216 | /* OK. use inode's group */ | |
f86186b4 | 217 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
218 | } |
219 | ||
654b4908 AK |
220 | /* |
221 | * Allocation for a meta data block | |
222 | */ | |
f65e6fba | 223 | static ext4_fsblk_t |
654b4908 | 224 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 225 | struct ext4_ext_path *path, |
55f020db | 226 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 227 | { |
f65e6fba | 228 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
229 | |
230 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
231 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
232 | NULL, err); | |
a86c6181 AT |
233 | return newblock; |
234 | } | |
235 | ||
55ad63bf | 236 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
237 | { |
238 | int size; | |
239 | ||
240 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
241 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 242 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
243 | if (!check && size > 6) |
244 | size = 6; | |
a86c6181 AT |
245 | #endif |
246 | return size; | |
247 | } | |
248 | ||
55ad63bf | 249 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
250 | { |
251 | int size; | |
252 | ||
253 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
254 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 255 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
256 | if (!check && size > 5) |
257 | size = 5; | |
a86c6181 AT |
258 | #endif |
259 | return size; | |
260 | } | |
261 | ||
55ad63bf | 262 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
263 | { |
264 | int size; | |
265 | ||
266 | size = sizeof(EXT4_I(inode)->i_data); | |
267 | size -= sizeof(struct ext4_extent_header); | |
268 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 269 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
270 | if (!check && size > 3) |
271 | size = 3; | |
a86c6181 AT |
272 | #endif |
273 | return size; | |
274 | } | |
275 | ||
55ad63bf | 276 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
277 | { |
278 | int size; | |
279 | ||
280 | size = sizeof(EXT4_I(inode)->i_data); | |
281 | size -= sizeof(struct ext4_extent_header); | |
282 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 283 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
284 | if (!check && size > 4) |
285 | size = 4; | |
a86c6181 AT |
286 | #endif |
287 | return size; | |
288 | } | |
289 | ||
d2a17637 MC |
290 | /* |
291 | * Calculate the number of metadata blocks needed | |
292 | * to allocate @blocks | |
293 | * Worse case is one block per extent | |
294 | */ | |
01f49d0b | 295 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 296 | { |
9d0be502 | 297 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 298 | int idxs; |
d2a17637 | 299 | |
9d0be502 TT |
300 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
301 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
302 | |
303 | /* | |
9d0be502 TT |
304 | * If the new delayed allocation block is contiguous with the |
305 | * previous da block, it can share index blocks with the | |
306 | * previous block, so we only need to allocate a new index | |
307 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
308 | * an additional index block, and at ldxs**3 blocks, yet | |
309 | * another index blocks. | |
d2a17637 | 310 | */ |
9d0be502 TT |
311 | if (ei->i_da_metadata_calc_len && |
312 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
313 | int num = 0; |
314 | ||
9d0be502 TT |
315 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
316 | num++; | |
317 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
318 | num++; | |
319 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
320 | num++; | |
321 | ei->i_da_metadata_calc_len = 0; | |
322 | } else | |
323 | ei->i_da_metadata_calc_len++; | |
324 | ei->i_da_metadata_calc_last_lblock++; | |
325 | return num; | |
326 | } | |
d2a17637 | 327 | |
9d0be502 TT |
328 | /* |
329 | * In the worst case we need a new set of index blocks at | |
330 | * every level of the inode's extent tree. | |
331 | */ | |
332 | ei->i_da_metadata_calc_len = 1; | |
333 | ei->i_da_metadata_calc_last_lblock = lblock; | |
334 | return ext_depth(inode) + 1; | |
d2a17637 MC |
335 | } |
336 | ||
c29c0ae7 AT |
337 | static int |
338 | ext4_ext_max_entries(struct inode *inode, int depth) | |
339 | { | |
340 | int max; | |
341 | ||
342 | if (depth == ext_depth(inode)) { | |
343 | if (depth == 0) | |
55ad63bf | 344 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 345 | else |
55ad63bf | 346 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
347 | } else { |
348 | if (depth == 0) | |
55ad63bf | 349 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 350 | else |
55ad63bf | 351 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
352 | } |
353 | ||
354 | return max; | |
355 | } | |
356 | ||
56b19868 AK |
357 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
358 | { | |
bf89d16f | 359 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 360 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 361 | |
31d4f3a2 TT |
362 | if (len == 0) |
363 | return 0; | |
6fd058f7 | 364 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
365 | } |
366 | ||
367 | static int ext4_valid_extent_idx(struct inode *inode, | |
368 | struct ext4_extent_idx *ext_idx) | |
369 | { | |
bf89d16f | 370 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 371 | |
6fd058f7 | 372 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
373 | } |
374 | ||
375 | static int ext4_valid_extent_entries(struct inode *inode, | |
376 | struct ext4_extent_header *eh, | |
377 | int depth) | |
378 | { | |
56b19868 AK |
379 | unsigned short entries; |
380 | if (eh->eh_entries == 0) | |
381 | return 1; | |
382 | ||
383 | entries = le16_to_cpu(eh->eh_entries); | |
384 | ||
385 | if (depth == 0) { | |
386 | /* leaf entries */ | |
81fdbb4a | 387 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
56b19868 AK |
388 | while (entries) { |
389 | if (!ext4_valid_extent(inode, ext)) | |
390 | return 0; | |
391 | ext++; | |
392 | entries--; | |
393 | } | |
394 | } else { | |
81fdbb4a | 395 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
396 | while (entries) { |
397 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
398 | return 0; | |
399 | ext_idx++; | |
400 | entries--; | |
401 | } | |
402 | } | |
403 | return 1; | |
404 | } | |
405 | ||
c398eda0 TT |
406 | static int __ext4_ext_check(const char *function, unsigned int line, |
407 | struct inode *inode, struct ext4_extent_header *eh, | |
408 | int depth) | |
c29c0ae7 AT |
409 | { |
410 | const char *error_msg; | |
411 | int max = 0; | |
412 | ||
413 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
414 | error_msg = "invalid magic"; | |
415 | goto corrupted; | |
416 | } | |
417 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
418 | error_msg = "unexpected eh_depth"; | |
419 | goto corrupted; | |
420 | } | |
421 | if (unlikely(eh->eh_max == 0)) { | |
422 | error_msg = "invalid eh_max"; | |
423 | goto corrupted; | |
424 | } | |
425 | max = ext4_ext_max_entries(inode, depth); | |
426 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
427 | error_msg = "too large eh_max"; | |
428 | goto corrupted; | |
429 | } | |
430 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
431 | error_msg = "invalid eh_entries"; | |
432 | goto corrupted; | |
433 | } | |
56b19868 AK |
434 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
435 | error_msg = "invalid extent entries"; | |
436 | goto corrupted; | |
437 | } | |
7ac5990d DW |
438 | /* Verify checksum on non-root extent tree nodes */ |
439 | if (ext_depth(inode) != depth && | |
440 | !ext4_extent_block_csum_verify(inode, eh)) { | |
441 | error_msg = "extent tree corrupted"; | |
442 | goto corrupted; | |
443 | } | |
c29c0ae7 AT |
444 | return 0; |
445 | ||
446 | corrupted: | |
c398eda0 | 447 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 448 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 449 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 450 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
451 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
452 | max, le16_to_cpu(eh->eh_depth), depth); | |
453 | ||
454 | return -EIO; | |
455 | } | |
456 | ||
56b19868 | 457 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 458 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 459 | |
7a262f7c AK |
460 | int ext4_ext_check_inode(struct inode *inode) |
461 | { | |
462 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
463 | } | |
464 | ||
f8489128 DW |
465 | static int __ext4_ext_check_block(const char *function, unsigned int line, |
466 | struct inode *inode, | |
467 | struct ext4_extent_header *eh, | |
468 | int depth, | |
469 | struct buffer_head *bh) | |
470 | { | |
471 | int ret; | |
472 | ||
473 | if (buffer_verified(bh)) | |
474 | return 0; | |
475 | ret = ext4_ext_check(inode, eh, depth); | |
476 | if (ret) | |
477 | return ret; | |
478 | set_buffer_verified(bh); | |
479 | return ret; | |
480 | } | |
481 | ||
482 | #define ext4_ext_check_block(inode, eh, depth, bh) \ | |
483 | __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh) | |
484 | ||
a86c6181 AT |
485 | #ifdef EXT_DEBUG |
486 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
487 | { | |
488 | int k, l = path->p_depth; | |
489 | ||
490 | ext_debug("path:"); | |
491 | for (k = 0; k <= l; k++, path++) { | |
492 | if (path->p_idx) { | |
2ae02107 | 493 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 494 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 495 | } else if (path->p_ext) { |
553f9008 | 496 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 497 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 498 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 499 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 500 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
501 | } else |
502 | ext_debug(" []"); | |
503 | } | |
504 | ext_debug("\n"); | |
505 | } | |
506 | ||
507 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
508 | { | |
509 | int depth = ext_depth(inode); | |
510 | struct ext4_extent_header *eh; | |
511 | struct ext4_extent *ex; | |
512 | int i; | |
513 | ||
514 | if (!path) | |
515 | return; | |
516 | ||
517 | eh = path[depth].p_hdr; | |
518 | ex = EXT_FIRST_EXTENT(eh); | |
519 | ||
553f9008 M |
520 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
521 | ||
a86c6181 | 522 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
523 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
524 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 525 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
526 | } |
527 | ext_debug("\n"); | |
528 | } | |
1b16da77 YY |
529 | |
530 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
531 | ext4_fsblk_t newblock, int level) | |
532 | { | |
533 | int depth = ext_depth(inode); | |
534 | struct ext4_extent *ex; | |
535 | ||
536 | if (depth != level) { | |
537 | struct ext4_extent_idx *idx; | |
538 | idx = path[level].p_idx; | |
539 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
540 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
541 | le32_to_cpu(idx->ei_block), | |
542 | ext4_idx_pblock(idx), | |
543 | newblock); | |
544 | idx++; | |
545 | } | |
546 | ||
547 | return; | |
548 | } | |
549 | ||
550 | ex = path[depth].p_ext; | |
551 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
552 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
553 | le32_to_cpu(ex->ee_block), | |
554 | ext4_ext_pblock(ex), | |
555 | ext4_ext_is_uninitialized(ex), | |
556 | ext4_ext_get_actual_len(ex), | |
557 | newblock); | |
558 | ex++; | |
559 | } | |
560 | } | |
561 | ||
a86c6181 | 562 | #else |
af5bc92d TT |
563 | #define ext4_ext_show_path(inode, path) |
564 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 565 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
566 | #endif |
567 | ||
b35905c1 | 568 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
569 | { |
570 | int depth = path->p_depth; | |
571 | int i; | |
572 | ||
573 | for (i = 0; i <= depth; i++, path++) | |
574 | if (path->p_bh) { | |
575 | brelse(path->p_bh); | |
576 | path->p_bh = NULL; | |
577 | } | |
578 | } | |
579 | ||
580 | /* | |
d0d856e8 RD |
581 | * ext4_ext_binsearch_idx: |
582 | * binary search for the closest index of the given block | |
c29c0ae7 | 583 | * the header must be checked before calling this |
a86c6181 AT |
584 | */ |
585 | static void | |
725d26d3 AK |
586 | ext4_ext_binsearch_idx(struct inode *inode, |
587 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
588 | { |
589 | struct ext4_extent_header *eh = path->p_hdr; | |
590 | struct ext4_extent_idx *r, *l, *m; | |
591 | ||
a86c6181 | 592 | |
bba90743 | 593 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
594 | |
595 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 596 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
597 | while (l <= r) { |
598 | m = l + (r - l) / 2; | |
599 | if (block < le32_to_cpu(m->ei_block)) | |
600 | r = m - 1; | |
601 | else | |
602 | l = m + 1; | |
26d535ed DM |
603 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
604 | m, le32_to_cpu(m->ei_block), | |
605 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
606 | } |
607 | ||
608 | path->p_idx = l - 1; | |
4a3c3a51 | 609 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 610 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
611 | |
612 | #ifdef CHECK_BINSEARCH | |
613 | { | |
614 | struct ext4_extent_idx *chix, *ix; | |
615 | int k; | |
616 | ||
617 | chix = ix = EXT_FIRST_INDEX(eh); | |
618 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
619 | if (k != 0 && | |
620 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
621 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
622 | "first=0x%p\n", k, | |
623 | ix, EXT_FIRST_INDEX(eh)); | |
624 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
625 | le32_to_cpu(ix->ei_block), |
626 | le32_to_cpu(ix[-1].ei_block)); | |
627 | } | |
628 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 629 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
630 | if (block < le32_to_cpu(ix->ei_block)) |
631 | break; | |
632 | chix = ix; | |
633 | } | |
634 | BUG_ON(chix != path->p_idx); | |
635 | } | |
636 | #endif | |
637 | ||
638 | } | |
639 | ||
640 | /* | |
d0d856e8 RD |
641 | * ext4_ext_binsearch: |
642 | * binary search for closest extent of the given block | |
c29c0ae7 | 643 | * the header must be checked before calling this |
a86c6181 AT |
644 | */ |
645 | static void | |
725d26d3 AK |
646 | ext4_ext_binsearch(struct inode *inode, |
647 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
648 | { |
649 | struct ext4_extent_header *eh = path->p_hdr; | |
650 | struct ext4_extent *r, *l, *m; | |
651 | ||
a86c6181 AT |
652 | if (eh->eh_entries == 0) { |
653 | /* | |
d0d856e8 RD |
654 | * this leaf is empty: |
655 | * we get such a leaf in split/add case | |
a86c6181 AT |
656 | */ |
657 | return; | |
658 | } | |
659 | ||
bba90743 | 660 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
661 | |
662 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 663 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
664 | |
665 | while (l <= r) { | |
666 | m = l + (r - l) / 2; | |
667 | if (block < le32_to_cpu(m->ee_block)) | |
668 | r = m - 1; | |
669 | else | |
670 | l = m + 1; | |
26d535ed DM |
671 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
672 | m, le32_to_cpu(m->ee_block), | |
673 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
674 | } |
675 | ||
676 | path->p_ext = l - 1; | |
553f9008 | 677 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 678 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 679 | ext4_ext_pblock(path->p_ext), |
553f9008 | 680 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 681 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
682 | |
683 | #ifdef CHECK_BINSEARCH | |
684 | { | |
685 | struct ext4_extent *chex, *ex; | |
686 | int k; | |
687 | ||
688 | chex = ex = EXT_FIRST_EXTENT(eh); | |
689 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
690 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 691 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
692 | if (block < le32_to_cpu(ex->ee_block)) |
693 | break; | |
694 | chex = ex; | |
695 | } | |
696 | BUG_ON(chex != path->p_ext); | |
697 | } | |
698 | #endif | |
699 | ||
700 | } | |
701 | ||
702 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
703 | { | |
704 | struct ext4_extent_header *eh; | |
705 | ||
706 | eh = ext_inode_hdr(inode); | |
707 | eh->eh_depth = 0; | |
708 | eh->eh_entries = 0; | |
709 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 710 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
711 | ext4_mark_inode_dirty(handle, inode); |
712 | ext4_ext_invalidate_cache(inode); | |
713 | return 0; | |
714 | } | |
715 | ||
716 | struct ext4_ext_path * | |
725d26d3 AK |
717 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
718 | struct ext4_ext_path *path) | |
a86c6181 AT |
719 | { |
720 | struct ext4_extent_header *eh; | |
721 | struct buffer_head *bh; | |
722 | short int depth, i, ppos = 0, alloc = 0; | |
723 | ||
724 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 725 | depth = ext_depth(inode); |
a86c6181 AT |
726 | |
727 | /* account possible depth increase */ | |
728 | if (!path) { | |
5d4958f9 | 729 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
730 | GFP_NOFS); |
731 | if (!path) | |
732 | return ERR_PTR(-ENOMEM); | |
733 | alloc = 1; | |
734 | } | |
a86c6181 | 735 | path[0].p_hdr = eh; |
1973adcb | 736 | path[0].p_bh = NULL; |
a86c6181 | 737 | |
c29c0ae7 | 738 | i = depth; |
a86c6181 AT |
739 | /* walk through the tree */ |
740 | while (i) { | |
741 | ext_debug("depth %d: num %d, max %d\n", | |
742 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 743 | |
a86c6181 | 744 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 745 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
746 | path[ppos].p_depth = i; |
747 | path[ppos].p_ext = NULL; | |
748 | ||
7a262f7c AK |
749 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
750 | if (unlikely(!bh)) | |
a86c6181 | 751 | goto err; |
7a262f7c | 752 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
753 | trace_ext4_ext_load_extent(inode, block, |
754 | path[ppos].p_block); | |
7a262f7c AK |
755 | if (bh_submit_read(bh) < 0) { |
756 | put_bh(bh); | |
757 | goto err; | |
758 | } | |
7a262f7c | 759 | } |
a86c6181 AT |
760 | eh = ext_block_hdr(bh); |
761 | ppos++; | |
273df556 FM |
762 | if (unlikely(ppos > depth)) { |
763 | put_bh(bh); | |
764 | EXT4_ERROR_INODE(inode, | |
765 | "ppos %d > depth %d", ppos, depth); | |
766 | goto err; | |
767 | } | |
a86c6181 AT |
768 | path[ppos].p_bh = bh; |
769 | path[ppos].p_hdr = eh; | |
770 | i--; | |
771 | ||
f8489128 | 772 | if (ext4_ext_check_block(inode, eh, i, bh)) |
a86c6181 AT |
773 | goto err; |
774 | } | |
775 | ||
776 | path[ppos].p_depth = i; | |
a86c6181 AT |
777 | path[ppos].p_ext = NULL; |
778 | path[ppos].p_idx = NULL; | |
779 | ||
a86c6181 AT |
780 | /* find extent */ |
781 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
782 | /* if not an empty leaf */ |
783 | if (path[ppos].p_ext) | |
bf89d16f | 784 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
785 | |
786 | ext4_ext_show_path(inode, path); | |
787 | ||
788 | return path; | |
789 | ||
790 | err: | |
791 | ext4_ext_drop_refs(path); | |
792 | if (alloc) | |
793 | kfree(path); | |
794 | return ERR_PTR(-EIO); | |
795 | } | |
796 | ||
797 | /* | |
d0d856e8 RD |
798 | * ext4_ext_insert_index: |
799 | * insert new index [@logical;@ptr] into the block at @curp; | |
800 | * check where to insert: before @curp or after @curp | |
a86c6181 | 801 | */ |
1f109d5a TT |
802 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
803 | struct ext4_ext_path *curp, | |
804 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
805 | { |
806 | struct ext4_extent_idx *ix; | |
807 | int len, err; | |
808 | ||
7e028976 AM |
809 | err = ext4_ext_get_access(handle, inode, curp); |
810 | if (err) | |
a86c6181 AT |
811 | return err; |
812 | ||
273df556 FM |
813 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
814 | EXT4_ERROR_INODE(inode, | |
815 | "logical %d == ei_block %d!", | |
816 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
817 | return -EIO; | |
818 | } | |
d4620315 RD |
819 | |
820 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
821 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
822 | EXT4_ERROR_INODE(inode, | |
823 | "eh_entries %d >= eh_max %d!", | |
824 | le16_to_cpu(curp->p_hdr->eh_entries), | |
825 | le16_to_cpu(curp->p_hdr->eh_max)); | |
826 | return -EIO; | |
827 | } | |
828 | ||
a86c6181 AT |
829 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
830 | /* insert after */ | |
80e675f9 | 831 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
832 | ix = curp->p_idx + 1; |
833 | } else { | |
834 | /* insert before */ | |
80e675f9 | 835 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
836 | ix = curp->p_idx; |
837 | } | |
838 | ||
80e675f9 EG |
839 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
840 | BUG_ON(len < 0); | |
841 | if (len > 0) { | |
842 | ext_debug("insert new index %d: " | |
843 | "move %d indices from 0x%p to 0x%p\n", | |
844 | logical, len, ix, ix + 1); | |
845 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
846 | } | |
847 | ||
f472e026 TM |
848 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
849 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
850 | return -EIO; | |
851 | } | |
852 | ||
a86c6181 | 853 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 854 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 855 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 856 | |
273df556 FM |
857 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
858 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
859 | return -EIO; | |
860 | } | |
a86c6181 AT |
861 | |
862 | err = ext4_ext_dirty(handle, inode, curp); | |
863 | ext4_std_error(inode->i_sb, err); | |
864 | ||
865 | return err; | |
866 | } | |
867 | ||
868 | /* | |
d0d856e8 RD |
869 | * ext4_ext_split: |
870 | * inserts new subtree into the path, using free index entry | |
871 | * at depth @at: | |
872 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
873 | * - makes decision where to split | |
874 | * - moves remaining extents and index entries (right to the split point) | |
875 | * into the newly allocated blocks | |
876 | * - initializes subtree | |
a86c6181 AT |
877 | */ |
878 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
879 | unsigned int flags, |
880 | struct ext4_ext_path *path, | |
881 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
882 | { |
883 | struct buffer_head *bh = NULL; | |
884 | int depth = ext_depth(inode); | |
885 | struct ext4_extent_header *neh; | |
886 | struct ext4_extent_idx *fidx; | |
a86c6181 | 887 | int i = at, k, m, a; |
f65e6fba | 888 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 889 | __le32 border; |
f65e6fba | 890 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
891 | int err = 0; |
892 | ||
893 | /* make decision: where to split? */ | |
d0d856e8 | 894 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 895 | |
d0d856e8 | 896 | /* if current leaf will be split, then we should use |
a86c6181 | 897 | * border from split point */ |
273df556 FM |
898 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
899 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
900 | return -EIO; | |
901 | } | |
a86c6181 AT |
902 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
903 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 904 | ext_debug("leaf will be split." |
a86c6181 | 905 | " next leaf starts at %d\n", |
8c55e204 | 906 | le32_to_cpu(border)); |
a86c6181 AT |
907 | } else { |
908 | border = newext->ee_block; | |
909 | ext_debug("leaf will be added." | |
910 | " next leaf starts at %d\n", | |
8c55e204 | 911 | le32_to_cpu(border)); |
a86c6181 AT |
912 | } |
913 | ||
914 | /* | |
d0d856e8 RD |
915 | * If error occurs, then we break processing |
916 | * and mark filesystem read-only. index won't | |
a86c6181 | 917 | * be inserted and tree will be in consistent |
d0d856e8 | 918 | * state. Next mount will repair buffers too. |
a86c6181 AT |
919 | */ |
920 | ||
921 | /* | |
d0d856e8 RD |
922 | * Get array to track all allocated blocks. |
923 | * We need this to handle errors and free blocks | |
924 | * upon them. | |
a86c6181 | 925 | */ |
5d4958f9 | 926 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
927 | if (!ablocks) |
928 | return -ENOMEM; | |
a86c6181 AT |
929 | |
930 | /* allocate all needed blocks */ | |
931 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
932 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 933 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 934 | newext, &err, flags); |
a86c6181 AT |
935 | if (newblock == 0) |
936 | goto cleanup; | |
937 | ablocks[a] = newblock; | |
938 | } | |
939 | ||
940 | /* initialize new leaf */ | |
941 | newblock = ablocks[--a]; | |
273df556 FM |
942 | if (unlikely(newblock == 0)) { |
943 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
944 | err = -EIO; | |
945 | goto cleanup; | |
946 | } | |
a86c6181 AT |
947 | bh = sb_getblk(inode->i_sb, newblock); |
948 | if (!bh) { | |
949 | err = -EIO; | |
950 | goto cleanup; | |
951 | } | |
952 | lock_buffer(bh); | |
953 | ||
7e028976 AM |
954 | err = ext4_journal_get_create_access(handle, bh); |
955 | if (err) | |
a86c6181 AT |
956 | goto cleanup; |
957 | ||
958 | neh = ext_block_hdr(bh); | |
959 | neh->eh_entries = 0; | |
55ad63bf | 960 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
961 | neh->eh_magic = EXT4_EXT_MAGIC; |
962 | neh->eh_depth = 0; | |
a86c6181 | 963 | |
d0d856e8 | 964 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
965 | if (unlikely(path[depth].p_hdr->eh_entries != |
966 | path[depth].p_hdr->eh_max)) { | |
967 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
968 | path[depth].p_hdr->eh_entries, | |
969 | path[depth].p_hdr->eh_max); | |
970 | err = -EIO; | |
971 | goto cleanup; | |
972 | } | |
a86c6181 | 973 | /* start copy from next extent */ |
1b16da77 YY |
974 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
975 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 976 | if (m) { |
1b16da77 YY |
977 | struct ext4_extent *ex; |
978 | ex = EXT_FIRST_EXTENT(neh); | |
979 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 980 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
981 | } |
982 | ||
7ac5990d | 983 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
984 | set_buffer_uptodate(bh); |
985 | unlock_buffer(bh); | |
986 | ||
0390131b | 987 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 988 | if (err) |
a86c6181 AT |
989 | goto cleanup; |
990 | brelse(bh); | |
991 | bh = NULL; | |
992 | ||
993 | /* correct old leaf */ | |
994 | if (m) { | |
7e028976 AM |
995 | err = ext4_ext_get_access(handle, inode, path + depth); |
996 | if (err) | |
a86c6181 | 997 | goto cleanup; |
e8546d06 | 998 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
999 | err = ext4_ext_dirty(handle, inode, path + depth); |
1000 | if (err) | |
a86c6181 AT |
1001 | goto cleanup; |
1002 | ||
1003 | } | |
1004 | ||
1005 | /* create intermediate indexes */ | |
1006 | k = depth - at - 1; | |
273df556 FM |
1007 | if (unlikely(k < 0)) { |
1008 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1009 | err = -EIO; | |
1010 | goto cleanup; | |
1011 | } | |
a86c6181 AT |
1012 | if (k) |
1013 | ext_debug("create %d intermediate indices\n", k); | |
1014 | /* insert new index into current index block */ | |
1015 | /* current depth stored in i var */ | |
1016 | i = depth - 1; | |
1017 | while (k--) { | |
1018 | oldblock = newblock; | |
1019 | newblock = ablocks[--a]; | |
bba90743 | 1020 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
1021 | if (!bh) { |
1022 | err = -EIO; | |
1023 | goto cleanup; | |
1024 | } | |
1025 | lock_buffer(bh); | |
1026 | ||
7e028976 AM |
1027 | err = ext4_journal_get_create_access(handle, bh); |
1028 | if (err) | |
a86c6181 AT |
1029 | goto cleanup; |
1030 | ||
1031 | neh = ext_block_hdr(bh); | |
1032 | neh->eh_entries = cpu_to_le16(1); | |
1033 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1034 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1035 | neh->eh_depth = cpu_to_le16(depth - i); |
1036 | fidx = EXT_FIRST_INDEX(neh); | |
1037 | fidx->ei_block = border; | |
f65e6fba | 1038 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1039 | |
bba90743 ES |
1040 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1041 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1042 | |
1b16da77 | 1043 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1044 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1045 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1046 | EXT4_ERROR_INODE(inode, | |
1047 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1048 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1049 | err = -EIO; | |
1050 | goto cleanup; | |
1051 | } | |
1b16da77 YY |
1052 | /* start copy indexes */ |
1053 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1054 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1055 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1056 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1057 | if (m) { |
1b16da77 | 1058 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1059 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1060 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1061 | } |
7ac5990d | 1062 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1063 | set_buffer_uptodate(bh); |
1064 | unlock_buffer(bh); | |
1065 | ||
0390131b | 1066 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1067 | if (err) |
a86c6181 AT |
1068 | goto cleanup; |
1069 | brelse(bh); | |
1070 | bh = NULL; | |
1071 | ||
1072 | /* correct old index */ | |
1073 | if (m) { | |
1074 | err = ext4_ext_get_access(handle, inode, path + i); | |
1075 | if (err) | |
1076 | goto cleanup; | |
e8546d06 | 1077 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1078 | err = ext4_ext_dirty(handle, inode, path + i); |
1079 | if (err) | |
1080 | goto cleanup; | |
1081 | } | |
1082 | ||
1083 | i--; | |
1084 | } | |
1085 | ||
1086 | /* insert new index */ | |
a86c6181 AT |
1087 | err = ext4_ext_insert_index(handle, inode, path + at, |
1088 | le32_to_cpu(border), newblock); | |
1089 | ||
1090 | cleanup: | |
1091 | if (bh) { | |
1092 | if (buffer_locked(bh)) | |
1093 | unlock_buffer(bh); | |
1094 | brelse(bh); | |
1095 | } | |
1096 | ||
1097 | if (err) { | |
1098 | /* free all allocated blocks in error case */ | |
1099 | for (i = 0; i < depth; i++) { | |
1100 | if (!ablocks[i]) | |
1101 | continue; | |
7dc57615 | 1102 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1103 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1104 | } |
1105 | } | |
1106 | kfree(ablocks); | |
1107 | ||
1108 | return err; | |
1109 | } | |
1110 | ||
1111 | /* | |
d0d856e8 RD |
1112 | * ext4_ext_grow_indepth: |
1113 | * implements tree growing procedure: | |
1114 | * - allocates new block | |
1115 | * - moves top-level data (index block or leaf) into the new block | |
1116 | * - initializes new top-level, creating index that points to the | |
1117 | * just created block | |
a86c6181 AT |
1118 | */ |
1119 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1120 | unsigned int flags, |
55f020db | 1121 | struct ext4_extent *newext) |
a86c6181 | 1122 | { |
a86c6181 | 1123 | struct ext4_extent_header *neh; |
a86c6181 | 1124 | struct buffer_head *bh; |
f65e6fba | 1125 | ext4_fsblk_t newblock; |
a86c6181 AT |
1126 | int err = 0; |
1127 | ||
1939dd84 | 1128 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1129 | newext, &err, flags); |
a86c6181 AT |
1130 | if (newblock == 0) |
1131 | return err; | |
1132 | ||
1133 | bh = sb_getblk(inode->i_sb, newblock); | |
1134 | if (!bh) { | |
1135 | err = -EIO; | |
1136 | ext4_std_error(inode->i_sb, err); | |
1137 | return err; | |
1138 | } | |
1139 | lock_buffer(bh); | |
1140 | ||
7e028976 AM |
1141 | err = ext4_journal_get_create_access(handle, bh); |
1142 | if (err) { | |
a86c6181 AT |
1143 | unlock_buffer(bh); |
1144 | goto out; | |
1145 | } | |
1146 | ||
1147 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1148 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1149 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1150 | |
1151 | /* set size of new block */ | |
1152 | neh = ext_block_hdr(bh); | |
1153 | /* old root could have indexes or leaves | |
1154 | * so calculate e_max right way */ | |
1155 | if (ext_depth(inode)) | |
55ad63bf | 1156 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1157 | else |
55ad63bf | 1158 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1159 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1160 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1161 | set_buffer_uptodate(bh); |
1162 | unlock_buffer(bh); | |
1163 | ||
0390131b | 1164 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1165 | if (err) |
a86c6181 AT |
1166 | goto out; |
1167 | ||
1939dd84 | 1168 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1169 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1170 | neh->eh_entries = cpu_to_le16(1); |
1171 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1172 | if (neh->eh_depth == 0) { | |
1173 | /* Root extent block becomes index block */ | |
1174 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1175 | EXT_FIRST_INDEX(neh)->ei_block = | |
1176 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1177 | } | |
2ae02107 | 1178 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1179 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1180 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1181 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1182 | |
ba39ebb6 | 1183 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1184 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1185 | out: |
1186 | brelse(bh); | |
1187 | ||
1188 | return err; | |
1189 | } | |
1190 | ||
1191 | /* | |
d0d856e8 RD |
1192 | * ext4_ext_create_new_leaf: |
1193 | * finds empty index and adds new leaf. | |
1194 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1195 | */ |
1196 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1197 | unsigned int flags, |
1198 | struct ext4_ext_path *path, | |
1199 | struct ext4_extent *newext) | |
a86c6181 AT |
1200 | { |
1201 | struct ext4_ext_path *curp; | |
1202 | int depth, i, err = 0; | |
1203 | ||
1204 | repeat: | |
1205 | i = depth = ext_depth(inode); | |
1206 | ||
1207 | /* walk up to the tree and look for free index entry */ | |
1208 | curp = path + depth; | |
1209 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1210 | i--; | |
1211 | curp--; | |
1212 | } | |
1213 | ||
d0d856e8 RD |
1214 | /* we use already allocated block for index block, |
1215 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1216 | if (EXT_HAS_FREE_INDEX(curp)) { |
1217 | /* if we found index with free entry, then use that | |
1218 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1219 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1220 | if (err) |
1221 | goto out; | |
a86c6181 AT |
1222 | |
1223 | /* refill path */ | |
1224 | ext4_ext_drop_refs(path); | |
1225 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1226 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1227 | path); | |
a86c6181 AT |
1228 | if (IS_ERR(path)) |
1229 | err = PTR_ERR(path); | |
1230 | } else { | |
1231 | /* tree is full, time to grow in depth */ | |
1939dd84 | 1232 | err = ext4_ext_grow_indepth(handle, inode, flags, newext); |
a86c6181 AT |
1233 | if (err) |
1234 | goto out; | |
1235 | ||
1236 | /* refill path */ | |
1237 | ext4_ext_drop_refs(path); | |
1238 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1239 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1240 | path); | |
a86c6181 AT |
1241 | if (IS_ERR(path)) { |
1242 | err = PTR_ERR(path); | |
1243 | goto out; | |
1244 | } | |
1245 | ||
1246 | /* | |
d0d856e8 RD |
1247 | * only first (depth 0 -> 1) produces free space; |
1248 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1249 | */ |
1250 | depth = ext_depth(inode); | |
1251 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1252 | /* now we need to split */ |
a86c6181 AT |
1253 | goto repeat; |
1254 | } | |
1255 | } | |
1256 | ||
1257 | out: | |
1258 | return err; | |
1259 | } | |
1260 | ||
1988b51e AT |
1261 | /* |
1262 | * search the closest allocated block to the left for *logical | |
1263 | * and returns it at @logical + it's physical address at @phys | |
1264 | * if *logical is the smallest allocated block, the function | |
1265 | * returns 0 at @phys | |
1266 | * return value contains 0 (success) or error code | |
1267 | */ | |
1f109d5a TT |
1268 | static int ext4_ext_search_left(struct inode *inode, |
1269 | struct ext4_ext_path *path, | |
1270 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1271 | { |
1272 | struct ext4_extent_idx *ix; | |
1273 | struct ext4_extent *ex; | |
b939e376 | 1274 | int depth, ee_len; |
1988b51e | 1275 | |
273df556 FM |
1276 | if (unlikely(path == NULL)) { |
1277 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1278 | return -EIO; | |
1279 | } | |
1988b51e AT |
1280 | depth = path->p_depth; |
1281 | *phys = 0; | |
1282 | ||
1283 | if (depth == 0 && path->p_ext == NULL) | |
1284 | return 0; | |
1285 | ||
1286 | /* usually extent in the path covers blocks smaller | |
1287 | * then *logical, but it can be that extent is the | |
1288 | * first one in the file */ | |
1289 | ||
1290 | ex = path[depth].p_ext; | |
b939e376 | 1291 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1292 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1293 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1294 | EXT4_ERROR_INODE(inode, | |
1295 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1296 | *logical, le32_to_cpu(ex->ee_block)); | |
1297 | return -EIO; | |
1298 | } | |
1988b51e AT |
1299 | while (--depth >= 0) { |
1300 | ix = path[depth].p_idx; | |
273df556 FM |
1301 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1302 | EXT4_ERROR_INODE(inode, | |
1303 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1304 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1305 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1306 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1307 | depth); |
1308 | return -EIO; | |
1309 | } | |
1988b51e AT |
1310 | } |
1311 | return 0; | |
1312 | } | |
1313 | ||
273df556 FM |
1314 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1315 | EXT4_ERROR_INODE(inode, | |
1316 | "logical %d < ee_block %d + ee_len %d!", | |
1317 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1318 | return -EIO; | |
1319 | } | |
1988b51e | 1320 | |
b939e376 | 1321 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1322 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1323 | return 0; |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * search the closest allocated block to the right for *logical | |
1328 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1329 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1330 | * returns 0 at @phys |
1331 | * return value contains 0 (success) or error code | |
1332 | */ | |
1f109d5a TT |
1333 | static int ext4_ext_search_right(struct inode *inode, |
1334 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1335 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1336 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1337 | { |
1338 | struct buffer_head *bh = NULL; | |
1339 | struct ext4_extent_header *eh; | |
1340 | struct ext4_extent_idx *ix; | |
1341 | struct ext4_extent *ex; | |
1342 | ext4_fsblk_t block; | |
395a87bf ES |
1343 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1344 | int ee_len; | |
1988b51e | 1345 | |
273df556 FM |
1346 | if (unlikely(path == NULL)) { |
1347 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1348 | return -EIO; | |
1349 | } | |
1988b51e AT |
1350 | depth = path->p_depth; |
1351 | *phys = 0; | |
1352 | ||
1353 | if (depth == 0 && path->p_ext == NULL) | |
1354 | return 0; | |
1355 | ||
1356 | /* usually extent in the path covers blocks smaller | |
1357 | * then *logical, but it can be that extent is the | |
1358 | * first one in the file */ | |
1359 | ||
1360 | ex = path[depth].p_ext; | |
b939e376 | 1361 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1362 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1363 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1364 | EXT4_ERROR_INODE(inode, | |
1365 | "first_extent(path[%d].p_hdr) != ex", | |
1366 | depth); | |
1367 | return -EIO; | |
1368 | } | |
1988b51e AT |
1369 | while (--depth >= 0) { |
1370 | ix = path[depth].p_idx; | |
273df556 FM |
1371 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1372 | EXT4_ERROR_INODE(inode, | |
1373 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1374 | *logical); | |
1375 | return -EIO; | |
1376 | } | |
1988b51e | 1377 | } |
4d33b1ef | 1378 | goto found_extent; |
1988b51e AT |
1379 | } |
1380 | ||
273df556 FM |
1381 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1382 | EXT4_ERROR_INODE(inode, | |
1383 | "logical %d < ee_block %d + ee_len %d!", | |
1384 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1385 | return -EIO; | |
1386 | } | |
1988b51e AT |
1387 | |
1388 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1389 | /* next allocated block in this leaf */ | |
1390 | ex++; | |
4d33b1ef | 1391 | goto found_extent; |
1988b51e AT |
1392 | } |
1393 | ||
1394 | /* go up and search for index to the right */ | |
1395 | while (--depth >= 0) { | |
1396 | ix = path[depth].p_idx; | |
1397 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1398 | goto got_index; |
1988b51e AT |
1399 | } |
1400 | ||
25f1ee3a WF |
1401 | /* we've gone up to the root and found no index to the right */ |
1402 | return 0; | |
1988b51e | 1403 | |
25f1ee3a | 1404 | got_index: |
1988b51e AT |
1405 | /* we've found index to the right, let's |
1406 | * follow it and find the closest allocated | |
1407 | * block to the right */ | |
1408 | ix++; | |
bf89d16f | 1409 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1410 | while (++depth < path->p_depth) { |
1411 | bh = sb_bread(inode->i_sb, block); | |
1412 | if (bh == NULL) | |
1413 | return -EIO; | |
1414 | eh = ext_block_hdr(bh); | |
395a87bf | 1415 | /* subtract from p_depth to get proper eh_depth */ |
f8489128 DW |
1416 | if (ext4_ext_check_block(inode, eh, |
1417 | path->p_depth - depth, bh)) { | |
1988b51e AT |
1418 | put_bh(bh); |
1419 | return -EIO; | |
1420 | } | |
1421 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1422 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1423 | put_bh(bh); |
1424 | } | |
1425 | ||
1426 | bh = sb_bread(inode->i_sb, block); | |
1427 | if (bh == NULL) | |
1428 | return -EIO; | |
1429 | eh = ext_block_hdr(bh); | |
f8489128 | 1430 | if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) { |
1988b51e AT |
1431 | put_bh(bh); |
1432 | return -EIO; | |
1433 | } | |
1434 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1435 | found_extent: |
1988b51e | 1436 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1437 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1438 | *ret_ex = ex; |
1439 | if (bh) | |
1440 | put_bh(bh); | |
1988b51e | 1441 | return 0; |
1988b51e AT |
1442 | } |
1443 | ||
a86c6181 | 1444 | /* |
d0d856e8 | 1445 | * ext4_ext_next_allocated_block: |
f17722f9 | 1446 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1447 | * NOTE: it considers block number from index entry as |
1448 | * allocated block. Thus, index entries have to be consistent | |
1449 | * with leaves. | |
a86c6181 | 1450 | */ |
725d26d3 | 1451 | static ext4_lblk_t |
a86c6181 AT |
1452 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1453 | { | |
1454 | int depth; | |
1455 | ||
1456 | BUG_ON(path == NULL); | |
1457 | depth = path->p_depth; | |
1458 | ||
1459 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1460 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1461 | |
1462 | while (depth >= 0) { | |
1463 | if (depth == path->p_depth) { | |
1464 | /* leaf */ | |
6f8ff537 CW |
1465 | if (path[depth].p_ext && |
1466 | path[depth].p_ext != | |
a86c6181 AT |
1467 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1468 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1469 | } else { | |
1470 | /* index */ | |
1471 | if (path[depth].p_idx != | |
1472 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1473 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1474 | } | |
1475 | depth--; | |
1476 | } | |
1477 | ||
f17722f9 | 1478 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1479 | } |
1480 | ||
1481 | /* | |
d0d856e8 | 1482 | * ext4_ext_next_leaf_block: |
f17722f9 | 1483 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1484 | */ |
5718789d | 1485 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1486 | { |
1487 | int depth; | |
1488 | ||
1489 | BUG_ON(path == NULL); | |
1490 | depth = path->p_depth; | |
1491 | ||
1492 | /* zero-tree has no leaf blocks at all */ | |
1493 | if (depth == 0) | |
f17722f9 | 1494 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1495 | |
1496 | /* go to index block */ | |
1497 | depth--; | |
1498 | ||
1499 | while (depth >= 0) { | |
1500 | if (path[depth].p_idx != | |
1501 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1502 | return (ext4_lblk_t) |
1503 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1504 | depth--; |
1505 | } | |
1506 | ||
f17722f9 | 1507 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1508 | } |
1509 | ||
1510 | /* | |
d0d856e8 RD |
1511 | * ext4_ext_correct_indexes: |
1512 | * if leaf gets modified and modified extent is first in the leaf, | |
1513 | * then we have to correct all indexes above. | |
a86c6181 AT |
1514 | * TODO: do we need to correct tree in all cases? |
1515 | */ | |
1d03ec98 | 1516 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1517 | struct ext4_ext_path *path) |
1518 | { | |
1519 | struct ext4_extent_header *eh; | |
1520 | int depth = ext_depth(inode); | |
1521 | struct ext4_extent *ex; | |
1522 | __le32 border; | |
1523 | int k, err = 0; | |
1524 | ||
1525 | eh = path[depth].p_hdr; | |
1526 | ex = path[depth].p_ext; | |
273df556 FM |
1527 | |
1528 | if (unlikely(ex == NULL || eh == NULL)) { | |
1529 | EXT4_ERROR_INODE(inode, | |
1530 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1531 | return -EIO; | |
1532 | } | |
a86c6181 AT |
1533 | |
1534 | if (depth == 0) { | |
1535 | /* there is no tree at all */ | |
1536 | return 0; | |
1537 | } | |
1538 | ||
1539 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1540 | /* we correct tree if first leaf got modified only */ | |
1541 | return 0; | |
1542 | } | |
1543 | ||
1544 | /* | |
d0d856e8 | 1545 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1546 | */ |
1547 | k = depth - 1; | |
1548 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1549 | err = ext4_ext_get_access(handle, inode, path + k); |
1550 | if (err) | |
a86c6181 AT |
1551 | return err; |
1552 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1553 | err = ext4_ext_dirty(handle, inode, path + k); |
1554 | if (err) | |
a86c6181 AT |
1555 | return err; |
1556 | ||
1557 | while (k--) { | |
1558 | /* change all left-side indexes */ | |
1559 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1560 | break; | |
7e028976 AM |
1561 | err = ext4_ext_get_access(handle, inode, path + k); |
1562 | if (err) | |
a86c6181 AT |
1563 | break; |
1564 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1565 | err = ext4_ext_dirty(handle, inode, path + k); |
1566 | if (err) | |
a86c6181 AT |
1567 | break; |
1568 | } | |
1569 | ||
1570 | return err; | |
1571 | } | |
1572 | ||
748de673 | 1573 | int |
a86c6181 AT |
1574 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1575 | struct ext4_extent *ex2) | |
1576 | { | |
749269fa | 1577 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1578 | |
1579 | /* | |
1580 | * Make sure that either both extents are uninitialized, or | |
1581 | * both are _not_. | |
1582 | */ | |
1583 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1584 | return 0; | |
1585 | ||
749269fa AA |
1586 | if (ext4_ext_is_uninitialized(ex1)) |
1587 | max_len = EXT_UNINIT_MAX_LEN; | |
1588 | else | |
1589 | max_len = EXT_INIT_MAX_LEN; | |
1590 | ||
a2df2a63 AA |
1591 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1592 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1593 | ||
1594 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1595 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1596 | return 0; |
1597 | ||
471d4011 SB |
1598 | /* |
1599 | * To allow future support for preallocated extents to be added | |
1600 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1601 | * this can result in the top bit of ee_len being set. |
471d4011 | 1602 | */ |
749269fa | 1603 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1604 | return 0; |
bbf2f9fb | 1605 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1606 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1607 | return 0; |
1608 | #endif | |
1609 | ||
bf89d16f | 1610 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1611 | return 1; |
1612 | return 0; | |
1613 | } | |
1614 | ||
56055d3a AA |
1615 | /* |
1616 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1617 | * It always tries to merge towards right. If you want to merge towards | |
1618 | * left, pass "ex - 1" as argument instead of "ex". | |
1619 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1620 | * 1 if they got merged. | |
1621 | */ | |
197217a5 | 1622 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1623 | struct ext4_ext_path *path, |
1624 | struct ext4_extent *ex) | |
56055d3a AA |
1625 | { |
1626 | struct ext4_extent_header *eh; | |
1627 | unsigned int depth, len; | |
1628 | int merge_done = 0; | |
1629 | int uninitialized = 0; | |
1630 | ||
1631 | depth = ext_depth(inode); | |
1632 | BUG_ON(path[depth].p_hdr == NULL); | |
1633 | eh = path[depth].p_hdr; | |
1634 | ||
1635 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1636 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1637 | break; | |
1638 | /* merge with next extent! */ | |
1639 | if (ext4_ext_is_uninitialized(ex)) | |
1640 | uninitialized = 1; | |
1641 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1642 | + ext4_ext_get_actual_len(ex + 1)); | |
1643 | if (uninitialized) | |
1644 | ext4_ext_mark_uninitialized(ex); | |
1645 | ||
1646 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1647 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1648 | * sizeof(struct ext4_extent); | |
1649 | memmove(ex + 1, ex + 2, len); | |
1650 | } | |
e8546d06 | 1651 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1652 | merge_done = 1; |
1653 | WARN_ON(eh->eh_entries == 0); | |
1654 | if (!eh->eh_entries) | |
24676da4 | 1655 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1656 | } |
1657 | ||
1658 | return merge_done; | |
1659 | } | |
1660 | ||
ecb94f5f TT |
1661 | /* |
1662 | * This function does a very simple check to see if we can collapse | |
1663 | * an extent tree with a single extent tree leaf block into the inode. | |
1664 | */ | |
1665 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1666 | struct inode *inode, | |
1667 | struct ext4_ext_path *path) | |
1668 | { | |
1669 | size_t s; | |
1670 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1671 | ext4_fsblk_t blk; | |
1672 | ||
1673 | if ((path[0].p_depth != 1) || | |
1674 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1675 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1676 | return; | |
1677 | ||
1678 | /* | |
1679 | * We need to modify the block allocation bitmap and the block | |
1680 | * group descriptor to release the extent tree block. If we | |
1681 | * can't get the journal credits, give up. | |
1682 | */ | |
1683 | if (ext4_journal_extend(handle, 2)) | |
1684 | return; | |
1685 | ||
1686 | /* | |
1687 | * Copy the extent data up to the inode | |
1688 | */ | |
1689 | blk = ext4_idx_pblock(path[0].p_idx); | |
1690 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1691 | sizeof(struct ext4_extent_idx); | |
1692 | s += sizeof(struct ext4_extent_header); | |
1693 | ||
1694 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1695 | path[0].p_depth = 0; | |
1696 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1697 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1698 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1699 | ||
1700 | brelse(path[1].p_bh); | |
1701 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
1702 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); | |
1703 | } | |
1704 | ||
197217a5 YY |
1705 | /* |
1706 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1707 | * return 1 if merge left else 0. | |
1708 | */ | |
ecb94f5f TT |
1709 | static void ext4_ext_try_to_merge(handle_t *handle, |
1710 | struct inode *inode, | |
197217a5 YY |
1711 | struct ext4_ext_path *path, |
1712 | struct ext4_extent *ex) { | |
1713 | struct ext4_extent_header *eh; | |
1714 | unsigned int depth; | |
1715 | int merge_done = 0; | |
197217a5 YY |
1716 | |
1717 | depth = ext_depth(inode); | |
1718 | BUG_ON(path[depth].p_hdr == NULL); | |
1719 | eh = path[depth].p_hdr; | |
1720 | ||
1721 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1722 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1723 | ||
1724 | if (!merge_done) | |
ecb94f5f | 1725 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1726 | |
ecb94f5f | 1727 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1728 | } |
1729 | ||
25d14f98 AA |
1730 | /* |
1731 | * check if a portion of the "newext" extent overlaps with an | |
1732 | * existing extent. | |
1733 | * | |
1734 | * If there is an overlap discovered, it updates the length of the newext | |
1735 | * such that there will be no overlap, and then returns 1. | |
1736 | * If there is no overlap found, it returns 0. | |
1737 | */ | |
4d33b1ef TT |
1738 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1739 | struct inode *inode, | |
1f109d5a TT |
1740 | struct ext4_extent *newext, |
1741 | struct ext4_ext_path *path) | |
25d14f98 | 1742 | { |
725d26d3 | 1743 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1744 | unsigned int depth, len1; |
1745 | unsigned int ret = 0; | |
1746 | ||
1747 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1748 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1749 | depth = ext_depth(inode); |
1750 | if (!path[depth].p_ext) | |
1751 | goto out; | |
1752 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1753 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1754 | |
1755 | /* | |
1756 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1757 | * is before the requested block(s) |
25d14f98 AA |
1758 | */ |
1759 | if (b2 < b1) { | |
1760 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1761 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1762 | goto out; |
4d33b1ef | 1763 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1764 | } |
1765 | ||
725d26d3 | 1766 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1767 | if (b1 + len1 < b1) { |
f17722f9 | 1768 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1769 | newext->ee_len = cpu_to_le16(len1); |
1770 | ret = 1; | |
1771 | } | |
1772 | ||
1773 | /* check for overlap */ | |
1774 | if (b1 + len1 > b2) { | |
1775 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1776 | ret = 1; | |
1777 | } | |
1778 | out: | |
1779 | return ret; | |
1780 | } | |
1781 | ||
a86c6181 | 1782 | /* |
d0d856e8 RD |
1783 | * ext4_ext_insert_extent: |
1784 | * tries to merge requsted extent into the existing extent or | |
1785 | * inserts requested extent as new one into the tree, | |
1786 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1787 | */ |
1788 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1789 | struct ext4_ext_path *path, | |
0031462b | 1790 | struct ext4_extent *newext, int flag) |
a86c6181 | 1791 | { |
af5bc92d | 1792 | struct ext4_extent_header *eh; |
a86c6181 AT |
1793 | struct ext4_extent *ex, *fex; |
1794 | struct ext4_extent *nearex; /* nearest extent */ | |
1795 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1796 | int depth, len, err; |
1797 | ext4_lblk_t next; | |
a2df2a63 | 1798 | unsigned uninitialized = 0; |
55f020db | 1799 | int flags = 0; |
a86c6181 | 1800 | |
273df556 FM |
1801 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1802 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1803 | return -EIO; | |
1804 | } | |
a86c6181 AT |
1805 | depth = ext_depth(inode); |
1806 | ex = path[depth].p_ext; | |
273df556 FM |
1807 | if (unlikely(path[depth].p_hdr == NULL)) { |
1808 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1809 | return -EIO; | |
1810 | } | |
a86c6181 AT |
1811 | |
1812 | /* try to insert block into found extent and return */ | |
744692dc | 1813 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1814 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
32de6756 | 1815 | ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n", |
bf89d16f TT |
1816 | ext4_ext_is_uninitialized(newext), |
1817 | ext4_ext_get_actual_len(newext), | |
1818 | le32_to_cpu(ex->ee_block), | |
1819 | ext4_ext_is_uninitialized(ex), | |
1820 | ext4_ext_get_actual_len(ex), | |
1821 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1822 | err = ext4_ext_get_access(handle, inode, path + depth); |
1823 | if (err) | |
a86c6181 | 1824 | return err; |
a2df2a63 AA |
1825 | |
1826 | /* | |
1827 | * ext4_can_extents_be_merged should have checked that either | |
1828 | * both extents are uninitialized, or both aren't. Thus we | |
1829 | * need to check only one of them here. | |
1830 | */ | |
1831 | if (ext4_ext_is_uninitialized(ex)) | |
1832 | uninitialized = 1; | |
1833 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1834 | + ext4_ext_get_actual_len(newext)); | |
1835 | if (uninitialized) | |
1836 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1837 | eh = path[depth].p_hdr; |
1838 | nearex = ex; | |
1839 | goto merge; | |
1840 | } | |
1841 | ||
a86c6181 AT |
1842 | depth = ext_depth(inode); |
1843 | eh = path[depth].p_hdr; | |
1844 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1845 | goto has_space; | |
1846 | ||
1847 | /* probably next leaf has space for us? */ | |
1848 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1849 | next = EXT_MAX_BLOCKS; |
1850 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1851 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1852 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1853 | ext_debug("next leaf block - %u\n", next); |
a86c6181 AT |
1854 | BUG_ON(npath != NULL); |
1855 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1856 | if (IS_ERR(npath)) | |
1857 | return PTR_ERR(npath); | |
1858 | BUG_ON(npath->p_depth != path->p_depth); | |
1859 | eh = npath[depth].p_hdr; | |
1860 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1861 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1862 | le16_to_cpu(eh->eh_entries)); |
1863 | path = npath; | |
ffb505ff | 1864 | goto has_space; |
a86c6181 AT |
1865 | } |
1866 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1867 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1868 | } | |
1869 | ||
1870 | /* | |
d0d856e8 RD |
1871 | * There is no free space in the found leaf. |
1872 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1873 | */ |
55f020db AH |
1874 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1875 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1876 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1877 | if (err) |
1878 | goto cleanup; | |
1879 | depth = ext_depth(inode); | |
1880 | eh = path[depth].p_hdr; | |
1881 | ||
1882 | has_space: | |
1883 | nearex = path[depth].p_ext; | |
1884 | ||
7e028976 AM |
1885 | err = ext4_ext_get_access(handle, inode, path + depth); |
1886 | if (err) | |
a86c6181 AT |
1887 | goto cleanup; |
1888 | ||
1889 | if (!nearex) { | |
1890 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1891 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1892 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1893 | ext4_ext_pblock(newext), |
553f9008 | 1894 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1895 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1896 | nearex = EXT_FIRST_EXTENT(eh); |
1897 | } else { | |
1898 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1899 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1900 | /* Insert after */ |
32de6756 YY |
1901 | ext_debug("insert %u:%llu:[%d]%d before: " |
1902 | "nearest %p\n", | |
80e675f9 EG |
1903 | le32_to_cpu(newext->ee_block), |
1904 | ext4_ext_pblock(newext), | |
1905 | ext4_ext_is_uninitialized(newext), | |
1906 | ext4_ext_get_actual_len(newext), | |
1907 | nearex); | |
1908 | nearex++; | |
1909 | } else { | |
1910 | /* Insert before */ | |
1911 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
1912 | ext_debug("insert %u:%llu:[%d]%d after: " |
1913 | "nearest %p\n", | |
8c55e204 | 1914 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1915 | ext4_ext_pblock(newext), |
553f9008 | 1916 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1917 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
1918 | nearex); |
1919 | } | |
1920 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
1921 | if (len > 0) { | |
32de6756 | 1922 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
1923 | "move %d extents from 0x%p to 0x%p\n", |
1924 | le32_to_cpu(newext->ee_block), | |
1925 | ext4_ext_pblock(newext), | |
1926 | ext4_ext_is_uninitialized(newext), | |
1927 | ext4_ext_get_actual_len(newext), | |
1928 | len, nearex, nearex + 1); | |
1929 | memmove(nearex + 1, nearex, | |
1930 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 1931 | } |
a86c6181 AT |
1932 | } |
1933 | ||
e8546d06 | 1934 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 1935 | path[depth].p_ext = nearex; |
a86c6181 | 1936 | nearex->ee_block = newext->ee_block; |
bf89d16f | 1937 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1938 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1939 | |
1940 | merge: | |
e7bcf823 | 1941 | /* try to merge extents */ |
744692dc | 1942 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 1943 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 1944 | |
a86c6181 AT |
1945 | |
1946 | /* time to correct all indexes above */ | |
1947 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1948 | if (err) | |
1949 | goto cleanup; | |
1950 | ||
ecb94f5f | 1951 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
1952 | |
1953 | cleanup: | |
1954 | if (npath) { | |
1955 | ext4_ext_drop_refs(npath); | |
1956 | kfree(npath); | |
1957 | } | |
a86c6181 AT |
1958 | ext4_ext_invalidate_cache(inode); |
1959 | return err; | |
1960 | } | |
1961 | ||
1f109d5a TT |
1962 | static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1963 | ext4_lblk_t num, ext_prepare_callback func, | |
1964 | void *cbdata) | |
6873fa0d ES |
1965 | { |
1966 | struct ext4_ext_path *path = NULL; | |
1967 | struct ext4_ext_cache cbex; | |
1968 | struct ext4_extent *ex; | |
1969 | ext4_lblk_t next, start = 0, end = 0; | |
1970 | ext4_lblk_t last = block + num; | |
1971 | int depth, exists, err = 0; | |
1972 | ||
1973 | BUG_ON(func == NULL); | |
1974 | BUG_ON(inode == NULL); | |
1975 | ||
f17722f9 | 1976 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1977 | num = last - block; |
1978 | /* find extent for this block */ | |
fab3a549 | 1979 | down_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d | 1980 | path = ext4_ext_find_extent(inode, block, path); |
fab3a549 | 1981 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1982 | if (IS_ERR(path)) { |
1983 | err = PTR_ERR(path); | |
1984 | path = NULL; | |
1985 | break; | |
1986 | } | |
1987 | ||
1988 | depth = ext_depth(inode); | |
273df556 FM |
1989 | if (unlikely(path[depth].p_hdr == NULL)) { |
1990 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1991 | err = -EIO; | |
1992 | break; | |
1993 | } | |
6873fa0d ES |
1994 | ex = path[depth].p_ext; |
1995 | next = ext4_ext_next_allocated_block(path); | |
1996 | ||
1997 | exists = 0; | |
1998 | if (!ex) { | |
1999 | /* there is no extent yet, so try to allocate | |
2000 | * all requested space */ | |
2001 | start = block; | |
2002 | end = block + num; | |
2003 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2004 | /* need to allocate space before found extent */ | |
2005 | start = block; | |
2006 | end = le32_to_cpu(ex->ee_block); | |
2007 | if (block + num < end) | |
2008 | end = block + num; | |
2009 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2010 | + ext4_ext_get_actual_len(ex)) { | |
2011 | /* need to allocate space after found extent */ | |
2012 | start = block; | |
2013 | end = block + num; | |
2014 | if (end >= next) | |
2015 | end = next; | |
2016 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2017 | /* | |
2018 | * some part of requested space is covered | |
2019 | * by found extent | |
2020 | */ | |
2021 | start = block; | |
2022 | end = le32_to_cpu(ex->ee_block) | |
2023 | + ext4_ext_get_actual_len(ex); | |
2024 | if (block + num < end) | |
2025 | end = block + num; | |
2026 | exists = 1; | |
2027 | } else { | |
2028 | BUG(); | |
2029 | } | |
2030 | BUG_ON(end <= start); | |
2031 | ||
2032 | if (!exists) { | |
2033 | cbex.ec_block = start; | |
2034 | cbex.ec_len = end - start; | |
2035 | cbex.ec_start = 0; | |
6873fa0d ES |
2036 | } else { |
2037 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
2038 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
bf89d16f | 2039 | cbex.ec_start = ext4_ext_pblock(ex); |
6873fa0d ES |
2040 | } |
2041 | ||
273df556 FM |
2042 | if (unlikely(cbex.ec_len == 0)) { |
2043 | EXT4_ERROR_INODE(inode, "cbex.ec_len == 0"); | |
2044 | err = -EIO; | |
2045 | break; | |
2046 | } | |
c03f8aa9 | 2047 | err = func(inode, next, &cbex, ex, cbdata); |
6873fa0d ES |
2048 | ext4_ext_drop_refs(path); |
2049 | ||
2050 | if (err < 0) | |
2051 | break; | |
2052 | ||
2053 | if (err == EXT_REPEAT) | |
2054 | continue; | |
2055 | else if (err == EXT_BREAK) { | |
2056 | err = 0; | |
2057 | break; | |
2058 | } | |
2059 | ||
2060 | if (ext_depth(inode) != depth) { | |
2061 | /* depth was changed. we have to realloc path */ | |
2062 | kfree(path); | |
2063 | path = NULL; | |
2064 | } | |
2065 | ||
2066 | block = cbex.ec_block + cbex.ec_len; | |
2067 | } | |
2068 | ||
2069 | if (path) { | |
2070 | ext4_ext_drop_refs(path); | |
2071 | kfree(path); | |
2072 | } | |
2073 | ||
2074 | return err; | |
2075 | } | |
2076 | ||
09b88252 | 2077 | static void |
725d26d3 | 2078 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 2079 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
2080 | { |
2081 | struct ext4_ext_cache *cex; | |
2082 | BUG_ON(len == 0); | |
2ec0ae3a | 2083 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
d8990240 | 2084 | trace_ext4_ext_put_in_cache(inode, block, len, start); |
a86c6181 | 2085 | cex = &EXT4_I(inode)->i_cached_extent; |
a86c6181 AT |
2086 | cex->ec_block = block; |
2087 | cex->ec_len = len; | |
2088 | cex->ec_start = start; | |
2ec0ae3a | 2089 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
2090 | } |
2091 | ||
2092 | /* | |
d0d856e8 RD |
2093 | * ext4_ext_put_gap_in_cache: |
2094 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2095 | * and cache this gap |
2096 | */ | |
09b88252 | 2097 | static void |
a86c6181 | 2098 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2099 | ext4_lblk_t block) |
a86c6181 AT |
2100 | { |
2101 | int depth = ext_depth(inode); | |
725d26d3 AK |
2102 | unsigned long len; |
2103 | ext4_lblk_t lblock; | |
a86c6181 AT |
2104 | struct ext4_extent *ex; |
2105 | ||
2106 | ex = path[depth].p_ext; | |
2107 | if (ex == NULL) { | |
2108 | /* there is no extent yet, so gap is [0;-] */ | |
2109 | lblock = 0; | |
f17722f9 | 2110 | len = EXT_MAX_BLOCKS; |
a86c6181 AT |
2111 | ext_debug("cache gap(whole file):"); |
2112 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2113 | lblock = block; | |
2114 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2115 | ext_debug("cache gap(before): %u [%u:%u]", |
2116 | block, | |
2117 | le32_to_cpu(ex->ee_block), | |
2118 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 2119 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2120 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2121 | ext4_lblk_t next; |
8c55e204 | 2122 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2123 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2124 | |
2125 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2126 | ext_debug("cache gap(after): [%u:%u] %u", |
2127 | le32_to_cpu(ex->ee_block), | |
2128 | ext4_ext_get_actual_len(ex), | |
2129 | block); | |
725d26d3 AK |
2130 | BUG_ON(next == lblock); |
2131 | len = next - lblock; | |
a86c6181 AT |
2132 | } else { |
2133 | lblock = len = 0; | |
2134 | BUG(); | |
2135 | } | |
2136 | ||
bba90743 | 2137 | ext_debug(" -> %u:%lu\n", lblock, len); |
b05e6ae5 | 2138 | ext4_ext_put_in_cache(inode, lblock, len, 0); |
a86c6181 AT |
2139 | } |
2140 | ||
b05e6ae5 | 2141 | /* |
63fedaf1 | 2142 | * ext4_ext_in_cache() |
a4bb6b64 AH |
2143 | * Checks to see if the given block is in the cache. |
2144 | * If it is, the cached extent is stored in the given | |
63fedaf1 | 2145 | * cache extent pointer. |
a4bb6b64 AH |
2146 | * |
2147 | * @inode: The files inode | |
2148 | * @block: The block to look for in the cache | |
2149 | * @ex: Pointer where the cached extent will be stored | |
2150 | * if it contains block | |
2151 | * | |
b05e6ae5 TT |
2152 | * Return 0 if cache is invalid; 1 if the cache is valid |
2153 | */ | |
63fedaf1 LC |
2154 | static int |
2155 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, | |
2156 | struct ext4_extent *ex) | |
2157 | { | |
a86c6181 | 2158 | struct ext4_ext_cache *cex; |
77f4135f | 2159 | struct ext4_sb_info *sbi; |
b05e6ae5 | 2160 | int ret = 0; |
a86c6181 | 2161 | |
60e6679e | 2162 | /* |
2ec0ae3a TT |
2163 | * We borrow i_block_reservation_lock to protect i_cached_extent |
2164 | */ | |
2165 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
a86c6181 | 2166 | cex = &EXT4_I(inode)->i_cached_extent; |
77f4135f | 2167 | sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2168 | |
2169 | /* has cache valid data? */ | |
b05e6ae5 | 2170 | if (cex->ec_len == 0) |
2ec0ae3a | 2171 | goto errout; |
a86c6181 | 2172 | |
731eb1a0 | 2173 | if (in_range(block, cex->ec_block, cex->ec_len)) { |
63fedaf1 LC |
2174 | ex->ee_block = cpu_to_le32(cex->ec_block); |
2175 | ext4_ext_store_pblock(ex, cex->ec_start); | |
2176 | ex->ee_len = cpu_to_le16(cex->ec_len); | |
bba90743 ES |
2177 | ext_debug("%u cached by %u:%u:%llu\n", |
2178 | block, | |
2179 | cex->ec_block, cex->ec_len, cex->ec_start); | |
b05e6ae5 | 2180 | ret = 1; |
a86c6181 | 2181 | } |
2ec0ae3a | 2182 | errout: |
d8990240 | 2183 | trace_ext4_ext_in_cache(inode, block, ret); |
2ec0ae3a TT |
2184 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2185 | return ret; | |
a86c6181 AT |
2186 | } |
2187 | ||
2188 | /* | |
d0d856e8 RD |
2189 | * ext4_ext_rm_idx: |
2190 | * removes index from the index block. | |
a86c6181 | 2191 | */ |
1d03ec98 | 2192 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
2193 | struct ext4_ext_path *path) |
2194 | { | |
a86c6181 | 2195 | int err; |
f65e6fba | 2196 | ext4_fsblk_t leaf; |
a86c6181 AT |
2197 | |
2198 | /* free index block */ | |
2199 | path--; | |
bf89d16f | 2200 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2201 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2202 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2203 | return -EIO; | |
2204 | } | |
7e028976 AM |
2205 | err = ext4_ext_get_access(handle, inode, path); |
2206 | if (err) | |
a86c6181 | 2207 | return err; |
0e1147b0 RD |
2208 | |
2209 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2210 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2211 | len *= sizeof(struct ext4_extent_idx); | |
2212 | memmove(path->p_idx, path->p_idx + 1, len); | |
2213 | } | |
2214 | ||
e8546d06 | 2215 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2216 | err = ext4_ext_dirty(handle, inode, path); |
2217 | if (err) | |
a86c6181 | 2218 | return err; |
2ae02107 | 2219 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2220 | trace_ext4_ext_rm_idx(inode, leaf); |
2221 | ||
7dc57615 | 2222 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2223 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
a86c6181 AT |
2224 | return err; |
2225 | } | |
2226 | ||
2227 | /* | |
ee12b630 MC |
2228 | * ext4_ext_calc_credits_for_single_extent: |
2229 | * This routine returns max. credits that needed to insert an extent | |
2230 | * to the extent tree. | |
2231 | * When pass the actual path, the caller should calculate credits | |
2232 | * under i_data_sem. | |
a86c6181 | 2233 | */ |
525f4ed8 | 2234 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2235 | struct ext4_ext_path *path) |
2236 | { | |
a86c6181 | 2237 | if (path) { |
ee12b630 | 2238 | int depth = ext_depth(inode); |
f3bd1f3f | 2239 | int ret = 0; |
ee12b630 | 2240 | |
a86c6181 | 2241 | /* probably there is space in leaf? */ |
a86c6181 | 2242 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2243 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2244 | |
ee12b630 MC |
2245 | /* |
2246 | * There are some space in the leaf tree, no | |
2247 | * need to account for leaf block credit | |
2248 | * | |
2249 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2250 | * and other metadata blocks still need to be |
ee12b630 MC |
2251 | * accounted. |
2252 | */ | |
525f4ed8 | 2253 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2254 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2255 | return ret; |
ee12b630 MC |
2256 | } |
2257 | } | |
a86c6181 | 2258 | |
525f4ed8 | 2259 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2260 | } |
a86c6181 | 2261 | |
ee12b630 MC |
2262 | /* |
2263 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2264 | * | |
2265 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2266 | * in the worse case, each tree level index/leaf need to be changed | |
2267 | * if the tree split due to insert a new extent, then the old tree | |
2268 | * index/leaf need to be updated too | |
2269 | * | |
2270 | * If the nrblocks are discontiguous, they could cause | |
2271 | * the whole tree split more than once, but this is really rare. | |
2272 | */ | |
525f4ed8 | 2273 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2274 | { |
2275 | int index; | |
2276 | int depth = ext_depth(inode); | |
a86c6181 | 2277 | |
ee12b630 MC |
2278 | if (chunk) |
2279 | index = depth * 2; | |
2280 | else | |
2281 | index = depth * 3; | |
a86c6181 | 2282 | |
ee12b630 | 2283 | return index; |
a86c6181 AT |
2284 | } |
2285 | ||
2286 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2287 | struct ext4_extent *ex, |
2288 | ext4_fsblk_t *partial_cluster, | |
2289 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2290 | { |
0aa06000 | 2291 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2292 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2293 | ext4_fsblk_t pblk; |
18888cf0 | 2294 | int flags = 0; |
a86c6181 | 2295 | |
c9de560d | 2296 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
18888cf0 AS |
2297 | flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2298 | else if (ext4_should_journal_data(inode)) | |
2299 | flags |= EXT4_FREE_BLOCKS_FORGET; | |
2300 | ||
0aa06000 TT |
2301 | /* |
2302 | * For bigalloc file systems, we never free a partial cluster | |
2303 | * at the beginning of the extent. Instead, we make a note | |
2304 | * that we tried freeing the cluster, and check to see if we | |
2305 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2306 | * or at the end of the ext4_truncate() operation. | |
2307 | */ | |
2308 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2309 | ||
d8990240 | 2310 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2311 | /* |
2312 | * If we have a partial cluster, and it's different from the | |
2313 | * cluster of the last block, we need to explicitly free the | |
2314 | * partial cluster here. | |
2315 | */ | |
2316 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2317 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2318 | ext4_free_blocks(handle, inode, NULL, | |
2319 | EXT4_C2B(sbi, *partial_cluster), | |
2320 | sbi->s_cluster_ratio, flags); | |
2321 | *partial_cluster = 0; | |
2322 | } | |
2323 | ||
a86c6181 AT |
2324 | #ifdef EXTENTS_STATS |
2325 | { | |
2326 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2327 | spin_lock(&sbi->s_ext_stats_lock); |
2328 | sbi->s_ext_blocks += ee_len; | |
2329 | sbi->s_ext_extents++; | |
2330 | if (ee_len < sbi->s_ext_min) | |
2331 | sbi->s_ext_min = ee_len; | |
2332 | if (ee_len > sbi->s_ext_max) | |
2333 | sbi->s_ext_max = ee_len; | |
2334 | if (ext_depth(inode) > sbi->s_depth_max) | |
2335 | sbi->s_depth_max = ext_depth(inode); | |
2336 | spin_unlock(&sbi->s_ext_stats_lock); | |
2337 | } | |
2338 | #endif | |
2339 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2340 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2341 | /* tail removal */ |
725d26d3 | 2342 | ext4_lblk_t num; |
725d26d3 | 2343 | |
a2df2a63 | 2344 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2345 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2346 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2347 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2348 | /* | |
2349 | * If the block range to be freed didn't start at the | |
2350 | * beginning of a cluster, and we removed the entire | |
2351 | * extent, save the partial cluster here, since we | |
2352 | * might need to delete if we determine that the | |
2353 | * truncate operation has removed all of the blocks in | |
2354 | * the cluster. | |
2355 | */ | |
2356 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2357 | (ee_len == num)) | |
2358 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2359 | else | |
2360 | *partial_cluster = 0; | |
a86c6181 | 2361 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2362 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2363 | /* head removal */ |
2364 | ext4_lblk_t num; | |
2365 | ext4_fsblk_t start; | |
2366 | ||
2367 | num = to - from; | |
2368 | start = ext4_ext_pblock(ex); | |
2369 | ||
2370 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
ee90d57e | 2371 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
d583fb87 | 2372 | |
a86c6181 | 2373 | } else { |
725d26d3 AK |
2374 | printk(KERN_INFO "strange request: removal(2) " |
2375 | "%u-%u from %u:%u\n", | |
2376 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2377 | } |
2378 | return 0; | |
2379 | } | |
2380 | ||
d583fb87 AH |
2381 | |
2382 | /* | |
2383 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2384 | * blocks appearing between "start" and "end", and splits the extents | |
2385 | * if "start" and "end" appear in the same extent | |
2386 | * | |
2387 | * @handle: The journal handle | |
2388 | * @inode: The files inode | |
2389 | * @path: The path to the leaf | |
2390 | * @start: The first block to remove | |
2391 | * @end: The last block to remove | |
2392 | */ | |
a86c6181 AT |
2393 | static int |
2394 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2395 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2396 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2397 | { |
0aa06000 | 2398 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2399 | int err = 0, correct_index = 0; |
2400 | int depth = ext_depth(inode), credits; | |
2401 | struct ext4_extent_header *eh; | |
750c9c47 | 2402 | ext4_lblk_t a, b; |
725d26d3 AK |
2403 | unsigned num; |
2404 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2405 | unsigned short ex_ee_len; |
a2df2a63 | 2406 | unsigned uninitialized = 0; |
a86c6181 AT |
2407 | struct ext4_extent *ex; |
2408 | ||
c29c0ae7 | 2409 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2410 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2411 | if (!path[depth].p_hdr) |
2412 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2413 | eh = path[depth].p_hdr; | |
273df556 FM |
2414 | if (unlikely(path[depth].p_hdr == NULL)) { |
2415 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2416 | return -EIO; | |
2417 | } | |
a86c6181 AT |
2418 | /* find where to start removing */ |
2419 | ex = EXT_LAST_EXTENT(eh); | |
2420 | ||
2421 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2422 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2423 | |
d8990240 AK |
2424 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2425 | ||
a86c6181 AT |
2426 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2427 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2428 | |
2429 | if (ext4_ext_is_uninitialized(ex)) | |
2430 | uninitialized = 1; | |
2431 | else | |
2432 | uninitialized = 0; | |
2433 | ||
553f9008 M |
2434 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2435 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2436 | path[depth].p_ext = ex; |
2437 | ||
2438 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2439 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2440 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2441 | |
2442 | ext_debug(" border %u:%u\n", a, b); | |
2443 | ||
d583fb87 | 2444 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2445 | if (end < ex_ee_block) { |
d583fb87 AH |
2446 | ex--; |
2447 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2448 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2449 | continue; | |
750c9c47 | 2450 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2451 | EXT4_ERROR_INODE(inode, |
2452 | "can not handle truncate %u:%u " | |
2453 | "on extent %u:%u", | |
2454 | start, end, ex_ee_block, | |
2455 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2456 | err = -EIO; |
2457 | goto out; | |
a86c6181 AT |
2458 | } else if (a != ex_ee_block) { |
2459 | /* remove tail of the extent */ | |
750c9c47 | 2460 | num = a - ex_ee_block; |
a86c6181 AT |
2461 | } else { |
2462 | /* remove whole extent: excellent! */ | |
a86c6181 | 2463 | num = 0; |
a86c6181 | 2464 | } |
34071da7 TT |
2465 | /* |
2466 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2467 | * descriptor) for each block group; assume two block | |
2468 | * groups plus ex_ee_len/blocks_per_block_group for | |
2469 | * the worst case | |
2470 | */ | |
2471 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2472 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2473 | correct_index = 1; | |
2474 | credits += (ext_depth(inode)) + 1; | |
2475 | } | |
5aca07eb | 2476 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2477 | |
487caeef | 2478 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2479 | if (err) |
a86c6181 | 2480 | goto out; |
a86c6181 AT |
2481 | |
2482 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2483 | if (err) | |
2484 | goto out; | |
2485 | ||
0aa06000 TT |
2486 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2487 | a, b); | |
a86c6181 AT |
2488 | if (err) |
2489 | goto out; | |
2490 | ||
750c9c47 | 2491 | if (num == 0) |
d0d856e8 | 2492 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2493 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2494 | |
a86c6181 | 2495 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2496 | /* |
2497 | * Do not mark uninitialized if all the blocks in the | |
2498 | * extent have been removed. | |
2499 | */ | |
2500 | if (uninitialized && num) | |
a2df2a63 | 2501 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2502 | /* |
2503 | * If the extent was completely released, | |
2504 | * we need to remove it from the leaf | |
2505 | */ | |
2506 | if (num == 0) { | |
f17722f9 | 2507 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2508 | /* |
2509 | * For hole punching, we need to scoot all the | |
2510 | * extents up when an extent is removed so that | |
2511 | * we dont have blank extents in the middle | |
2512 | */ | |
2513 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2514 | sizeof(struct ext4_extent)); | |
2515 | ||
2516 | /* Now get rid of the one at the end */ | |
2517 | memset(EXT_LAST_EXTENT(eh), 0, | |
2518 | sizeof(struct ext4_extent)); | |
2519 | } | |
2520 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2521 | } else |
2522 | *partial_cluster = 0; | |
d583fb87 | 2523 | |
750c9c47 DM |
2524 | err = ext4_ext_dirty(handle, inode, path + depth); |
2525 | if (err) | |
2526 | goto out; | |
2527 | ||
bf52c6f7 | 2528 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2529 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2530 | ex--; |
2531 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2532 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2533 | } |
2534 | ||
2535 | if (correct_index && eh->eh_entries) | |
2536 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2537 | ||
0aa06000 TT |
2538 | /* |
2539 | * If there is still a entry in the leaf node, check to see if | |
2540 | * it references the partial cluster. This is the only place | |
2541 | * where it could; if it doesn't, we can free the cluster. | |
2542 | */ | |
2543 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2544 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2545 | *partial_cluster)) { | |
2546 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2547 | ||
2548 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2549 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2550 | ||
2551 | ext4_free_blocks(handle, inode, NULL, | |
2552 | EXT4_C2B(sbi, *partial_cluster), | |
2553 | sbi->s_cluster_ratio, flags); | |
2554 | *partial_cluster = 0; | |
2555 | } | |
2556 | ||
a86c6181 AT |
2557 | /* if this leaf is free, then we should |
2558 | * remove it from index block above */ | |
2559 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2560 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2561 | ||
2562 | out: | |
2563 | return err; | |
2564 | } | |
2565 | ||
2566 | /* | |
d0d856e8 RD |
2567 | * ext4_ext_more_to_rm: |
2568 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2569 | */ |
09b88252 | 2570 | static int |
a86c6181 AT |
2571 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2572 | { | |
2573 | BUG_ON(path->p_idx == NULL); | |
2574 | ||
2575 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2576 | return 0; | |
2577 | ||
2578 | /* | |
d0d856e8 | 2579 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2580 | * so we have to consider current index for truncation |
2581 | */ | |
2582 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2583 | return 0; | |
2584 | return 1; | |
2585 | } | |
2586 | ||
5f95d21f LC |
2587 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2588 | ext4_lblk_t end) | |
a86c6181 AT |
2589 | { |
2590 | struct super_block *sb = inode->i_sb; | |
2591 | int depth = ext_depth(inode); | |
968dee77 | 2592 | struct ext4_ext_path *path = NULL; |
0aa06000 | 2593 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2594 | handle_t *handle; |
6f2080e6 | 2595 | int i = 0, err = 0; |
a86c6181 | 2596 | |
5f95d21f | 2597 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2598 | |
2599 | /* probably first extent we're gonna free will be last in block */ | |
2600 | handle = ext4_journal_start(inode, depth + 1); | |
2601 | if (IS_ERR(handle)) | |
2602 | return PTR_ERR(handle); | |
2603 | ||
0617b83f | 2604 | again: |
a86c6181 AT |
2605 | ext4_ext_invalidate_cache(inode); |
2606 | ||
d8990240 AK |
2607 | trace_ext4_ext_remove_space(inode, start, depth); |
2608 | ||
5f95d21f LC |
2609 | /* |
2610 | * Check if we are removing extents inside the extent tree. If that | |
2611 | * is the case, we are going to punch a hole inside the extent tree | |
2612 | * so we have to check whether we need to split the extent covering | |
2613 | * the last block to remove so we can easily remove the part of it | |
2614 | * in ext4_ext_rm_leaf(). | |
2615 | */ | |
2616 | if (end < EXT_MAX_BLOCKS - 1) { | |
2617 | struct ext4_extent *ex; | |
2618 | ext4_lblk_t ee_block; | |
2619 | ||
2620 | /* find extent for this block */ | |
2621 | path = ext4_ext_find_extent(inode, end, NULL); | |
2622 | if (IS_ERR(path)) { | |
2623 | ext4_journal_stop(handle); | |
2624 | return PTR_ERR(path); | |
2625 | } | |
2626 | depth = ext_depth(inode); | |
6f2080e6 | 2627 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2628 | ex = path[depth].p_ext; |
968dee77 | 2629 | if (!ex) { |
6f2080e6 DM |
2630 | if (depth) { |
2631 | EXT4_ERROR_INODE(inode, | |
2632 | "path[%d].p_hdr == NULL", | |
2633 | depth); | |
2634 | err = -EIO; | |
2635 | } | |
2636 | goto out; | |
968dee77 | 2637 | } |
5f95d21f LC |
2638 | |
2639 | ee_block = le32_to_cpu(ex->ee_block); | |
2640 | ||
2641 | /* | |
2642 | * See if the last block is inside the extent, if so split | |
2643 | * the extent at 'end' block so we can easily remove the | |
2644 | * tail of the first part of the split extent in | |
2645 | * ext4_ext_rm_leaf(). | |
2646 | */ | |
2647 | if (end >= ee_block && | |
2648 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2649 | int split_flag = 0; | |
2650 | ||
2651 | if (ext4_ext_is_uninitialized(ex)) | |
2652 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2653 | EXT4_EXT_MARK_UNINIT2; | |
2654 | ||
2655 | /* | |
2656 | * Split the extent in two so that 'end' is the last | |
2657 | * block in the first new extent | |
2658 | */ | |
2659 | err = ext4_split_extent_at(handle, inode, path, | |
2660 | end + 1, split_flag, | |
2661 | EXT4_GET_BLOCKS_PRE_IO | | |
2662 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
2663 | ||
2664 | if (err < 0) | |
2665 | goto out; | |
2666 | } | |
5f95d21f | 2667 | } |
a86c6181 | 2668 | /* |
d0d856e8 RD |
2669 | * We start scanning from right side, freeing all the blocks |
2670 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2671 | */ |
0617b83f | 2672 | depth = ext_depth(inode); |
968dee77 AS |
2673 | if (path) { |
2674 | int k = i = depth; | |
2675 | while (--k > 0) | |
2676 | path[k].p_block = | |
2677 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2678 | } else { | |
2679 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2680 | GFP_NOFS); | |
2681 | if (path == NULL) { | |
2682 | ext4_journal_stop(handle); | |
2683 | return -ENOMEM; | |
2684 | } | |
2685 | path[0].p_depth = depth; | |
2686 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2687 | i = 0; |
5f95d21f | 2688 | |
968dee77 AS |
2689 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
2690 | err = -EIO; | |
2691 | goto out; | |
2692 | } | |
a86c6181 | 2693 | } |
968dee77 | 2694 | err = 0; |
a86c6181 AT |
2695 | |
2696 | while (i >= 0 && err == 0) { | |
2697 | if (i == depth) { | |
2698 | /* this is leaf block */ | |
d583fb87 | 2699 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2700 | &partial_cluster, start, |
5f95d21f | 2701 | end); |
d0d856e8 | 2702 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2703 | brelse(path[i].p_bh); |
2704 | path[i].p_bh = NULL; | |
2705 | i--; | |
2706 | continue; | |
2707 | } | |
2708 | ||
2709 | /* this is index block */ | |
2710 | if (!path[i].p_hdr) { | |
2711 | ext_debug("initialize header\n"); | |
2712 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2713 | } |
2714 | ||
a86c6181 | 2715 | if (!path[i].p_idx) { |
d0d856e8 | 2716 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2717 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2718 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2719 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2720 | path[i].p_hdr, | |
2721 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2722 | } else { | |
d0d856e8 | 2723 | /* we were already here, see at next index */ |
a86c6181 AT |
2724 | path[i].p_idx--; |
2725 | } | |
2726 | ||
2727 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2728 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2729 | path[i].p_idx); | |
2730 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2731 | struct buffer_head *bh; |
a86c6181 | 2732 | /* go to the next level */ |
2ae02107 | 2733 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2734 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2735 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2736 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2737 | if (!bh) { |
a86c6181 AT |
2738 | /* should we reset i_size? */ |
2739 | err = -EIO; | |
2740 | break; | |
2741 | } | |
c29c0ae7 AT |
2742 | if (WARN_ON(i + 1 > depth)) { |
2743 | err = -EIO; | |
2744 | break; | |
2745 | } | |
f8489128 DW |
2746 | if (ext4_ext_check_block(inode, ext_block_hdr(bh), |
2747 | depth - i - 1, bh)) { | |
c29c0ae7 AT |
2748 | err = -EIO; |
2749 | break; | |
2750 | } | |
2751 | path[i + 1].p_bh = bh; | |
a86c6181 | 2752 | |
d0d856e8 RD |
2753 | /* save actual number of indexes since this |
2754 | * number is changed at the next iteration */ | |
a86c6181 AT |
2755 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2756 | i++; | |
2757 | } else { | |
d0d856e8 | 2758 | /* we finished processing this index, go up */ |
a86c6181 | 2759 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2760 | /* index is empty, remove it; |
a86c6181 AT |
2761 | * handle must be already prepared by the |
2762 | * truncatei_leaf() */ | |
2763 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2764 | } | |
d0d856e8 | 2765 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2766 | brelse(path[i].p_bh); |
2767 | path[i].p_bh = NULL; | |
2768 | i--; | |
2769 | ext_debug("return to level %d\n", i); | |
2770 | } | |
2771 | } | |
2772 | ||
d8990240 AK |
2773 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2774 | path->p_hdr->eh_entries); | |
2775 | ||
7b415bf6 AK |
2776 | /* If we still have something in the partial cluster and we have removed |
2777 | * even the first extent, then we should free the blocks in the partial | |
2778 | * cluster as well. */ | |
2779 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2780 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2781 | ||
2782 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2783 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2784 | ||
2785 | ext4_free_blocks(handle, inode, NULL, | |
2786 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2787 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2788 | partial_cluster = 0; | |
2789 | } | |
2790 | ||
a86c6181 AT |
2791 | /* TODO: flexible tree reduction should be here */ |
2792 | if (path->p_hdr->eh_entries == 0) { | |
2793 | /* | |
d0d856e8 RD |
2794 | * truncate to zero freed all the tree, |
2795 | * so we need to correct eh_depth | |
a86c6181 AT |
2796 | */ |
2797 | err = ext4_ext_get_access(handle, inode, path); | |
2798 | if (err == 0) { | |
2799 | ext_inode_hdr(inode)->eh_depth = 0; | |
2800 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2801 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2802 | err = ext4_ext_dirty(handle, inode, path); |
2803 | } | |
2804 | } | |
2805 | out: | |
a86c6181 AT |
2806 | ext4_ext_drop_refs(path); |
2807 | kfree(path); | |
968dee77 AS |
2808 | if (err == -EAGAIN) { |
2809 | path = NULL; | |
0617b83f | 2810 | goto again; |
968dee77 | 2811 | } |
a86c6181 AT |
2812 | ext4_journal_stop(handle); |
2813 | ||
2814 | return err; | |
2815 | } | |
2816 | ||
2817 | /* | |
2818 | * called at mount time | |
2819 | */ | |
2820 | void ext4_ext_init(struct super_block *sb) | |
2821 | { | |
2822 | /* | |
2823 | * possible initialization would be here | |
2824 | */ | |
2825 | ||
83982b6f | 2826 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2827 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2828 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2829 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2830 | ", aggressive tests" |
a86c6181 AT |
2831 | #endif |
2832 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2833 | ", check binsearch" |
a86c6181 AT |
2834 | #endif |
2835 | #ifdef EXTENTS_STATS | |
92b97816 | 2836 | ", stats" |
a86c6181 | 2837 | #endif |
92b97816 | 2838 | "\n"); |
90576c0b | 2839 | #endif |
a86c6181 AT |
2840 | #ifdef EXTENTS_STATS |
2841 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2842 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2843 | EXT4_SB(sb)->s_ext_max = 0; | |
2844 | #endif | |
2845 | } | |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * called at umount time | |
2850 | */ | |
2851 | void ext4_ext_release(struct super_block *sb) | |
2852 | { | |
83982b6f | 2853 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2854 | return; |
2855 | ||
2856 | #ifdef EXTENTS_STATS | |
2857 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2858 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2859 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2860 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2861 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2862 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2863 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2864 | } | |
2865 | #endif | |
2866 | } | |
2867 | ||
093a088b AK |
2868 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2869 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2870 | { | |
2407518d LC |
2871 | ext4_fsblk_t ee_pblock; |
2872 | unsigned int ee_len; | |
b720303d | 2873 | int ret; |
093a088b | 2874 | |
093a088b | 2875 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2876 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2877 | |
a107e5a3 | 2878 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2879 | if (ret > 0) |
2880 | ret = 0; | |
093a088b | 2881 | |
2407518d | 2882 | return ret; |
093a088b AK |
2883 | } |
2884 | ||
47ea3bb5 YY |
2885 | /* |
2886 | * ext4_split_extent_at() splits an extent at given block. | |
2887 | * | |
2888 | * @handle: the journal handle | |
2889 | * @inode: the file inode | |
2890 | * @path: the path to the extent | |
2891 | * @split: the logical block where the extent is splitted. | |
2892 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2893 | * the states(init or uninit) of new extents. | |
2894 | * @flags: flags used to insert new extent to extent tree. | |
2895 | * | |
2896 | * | |
2897 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2898 | * of which are deterimined by split_flag. | |
2899 | * | |
2900 | * There are two cases: | |
2901 | * a> the extent are splitted into two extent. | |
2902 | * b> split is not needed, and just mark the extent. | |
2903 | * | |
2904 | * return 0 on success. | |
2905 | */ | |
2906 | static int ext4_split_extent_at(handle_t *handle, | |
2907 | struct inode *inode, | |
2908 | struct ext4_ext_path *path, | |
2909 | ext4_lblk_t split, | |
2910 | int split_flag, | |
2911 | int flags) | |
2912 | { | |
2913 | ext4_fsblk_t newblock; | |
2914 | ext4_lblk_t ee_block; | |
2915 | struct ext4_extent *ex, newex, orig_ex; | |
2916 | struct ext4_extent *ex2 = NULL; | |
2917 | unsigned int ee_len, depth; | |
2918 | int err = 0; | |
2919 | ||
dee1f973 DM |
2920 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
2921 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
2922 | ||
47ea3bb5 YY |
2923 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
2924 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2925 | ||
2926 | ext4_ext_show_leaf(inode, path); | |
2927 | ||
2928 | depth = ext_depth(inode); | |
2929 | ex = path[depth].p_ext; | |
2930 | ee_block = le32_to_cpu(ex->ee_block); | |
2931 | ee_len = ext4_ext_get_actual_len(ex); | |
2932 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2933 | ||
2934 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
2935 | ||
2936 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2937 | if (err) | |
2938 | goto out; | |
2939 | ||
2940 | if (split == ee_block) { | |
2941 | /* | |
2942 | * case b: block @split is the block that the extent begins with | |
2943 | * then we just change the state of the extent, and splitting | |
2944 | * is not needed. | |
2945 | */ | |
2946 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2947 | ext4_ext_mark_uninitialized(ex); | |
2948 | else | |
2949 | ext4_ext_mark_initialized(ex); | |
2950 | ||
2951 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 2952 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 2953 | |
ecb94f5f | 2954 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
2955 | goto out; |
2956 | } | |
2957 | ||
2958 | /* case a */ | |
2959 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
2960 | ex->ee_len = cpu_to_le16(split - ee_block); | |
2961 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
2962 | ext4_ext_mark_uninitialized(ex); | |
2963 | ||
2964 | /* | |
2965 | * path may lead to new leaf, not to original leaf any more | |
2966 | * after ext4_ext_insert_extent() returns, | |
2967 | */ | |
2968 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2969 | if (err) | |
2970 | goto fix_extent_len; | |
2971 | ||
2972 | ex2 = &newex; | |
2973 | ex2->ee_block = cpu_to_le32(split); | |
2974 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
2975 | ext4_ext_store_pblock(ex2, newblock); | |
2976 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2977 | ext4_ext_mark_uninitialized(ex2); | |
2978 | ||
2979 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
2980 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 DM |
2981 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
2982 | if (split_flag & EXT4_EXT_DATA_VALID1) | |
2983 | err = ext4_ext_zeroout(inode, ex2); | |
2984 | else | |
2985 | err = ext4_ext_zeroout(inode, ex); | |
2986 | } else | |
2987 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2988 | ||
47ea3bb5 YY |
2989 | if (err) |
2990 | goto fix_extent_len; | |
2991 | /* update the extent length and mark as initialized */ | |
af1584f5 | 2992 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
2993 | ext4_ext_try_to_merge(handle, inode, path, ex); |
2994 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
47ea3bb5 YY |
2995 | goto out; |
2996 | } else if (err) | |
2997 | goto fix_extent_len; | |
2998 | ||
2999 | out: | |
3000 | ext4_ext_show_leaf(inode, path); | |
3001 | return err; | |
3002 | ||
3003 | fix_extent_len: | |
3004 | ex->ee_len = orig_ex.ee_len; | |
3005 | ext4_ext_dirty(handle, inode, path + depth); | |
3006 | return err; | |
3007 | } | |
3008 | ||
3009 | /* | |
3010 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3011 | * by @map as split_flags indicates | |
3012 | * | |
3013 | * It may result in splitting the extent into multiple extents (upto three) | |
3014 | * There are three possibilities: | |
3015 | * a> There is no split required | |
3016 | * b> Splits in two extents: Split is happening at either end of the extent | |
3017 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3018 | * | |
3019 | */ | |
3020 | static int ext4_split_extent(handle_t *handle, | |
3021 | struct inode *inode, | |
3022 | struct ext4_ext_path *path, | |
3023 | struct ext4_map_blocks *map, | |
3024 | int split_flag, | |
3025 | int flags) | |
3026 | { | |
3027 | ext4_lblk_t ee_block; | |
3028 | struct ext4_extent *ex; | |
3029 | unsigned int ee_len, depth; | |
3030 | int err = 0; | |
3031 | int uninitialized; | |
3032 | int split_flag1, flags1; | |
3033 | ||
3034 | depth = ext_depth(inode); | |
3035 | ex = path[depth].p_ext; | |
3036 | ee_block = le32_to_cpu(ex->ee_block); | |
3037 | ee_len = ext4_ext_get_actual_len(ex); | |
3038 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3039 | ||
3040 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3041 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3042 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3043 | if (uninitialized) | |
3044 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3045 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3046 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3047 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3048 | err = ext4_split_extent_at(handle, inode, path, |
3049 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3050 | if (err) |
3051 | goto out; | |
47ea3bb5 YY |
3052 | } |
3053 | ||
3054 | ext4_ext_drop_refs(path); | |
3055 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3056 | if (IS_ERR(path)) | |
3057 | return PTR_ERR(path); | |
3058 | ||
3059 | if (map->m_lblk >= ee_block) { | |
dee1f973 DM |
3060 | split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3061 | EXT4_EXT_DATA_VALID2); | |
47ea3bb5 YY |
3062 | if (uninitialized) |
3063 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
3064 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3065 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
3066 | err = ext4_split_extent_at(handle, inode, path, | |
3067 | map->m_lblk, split_flag1, flags); | |
3068 | if (err) | |
3069 | goto out; | |
3070 | } | |
3071 | ||
3072 | ext4_ext_show_leaf(inode, path); | |
3073 | out: | |
3074 | return err ? err : map->m_len; | |
3075 | } | |
3076 | ||
56055d3a | 3077 | /* |
e35fd660 | 3078 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3079 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3080 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3081 | * uninitialized). |
3082 | * There are three possibilities: | |
3083 | * a> There is no split required: Entire extent should be initialized | |
3084 | * b> Splits in two extents: Write is happening at either end of the extent | |
3085 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3086 | * |
3087 | * Pre-conditions: | |
3088 | * - The extent pointed to by 'path' is uninitialized. | |
3089 | * - The extent pointed to by 'path' contains a superset | |
3090 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3091 | * | |
3092 | * Post-conditions on success: | |
3093 | * - the returned value is the number of blocks beyond map->l_lblk | |
3094 | * that are allocated and initialized. | |
3095 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3096 | */ |
725d26d3 | 3097 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3098 | struct inode *inode, |
3099 | struct ext4_map_blocks *map, | |
3100 | struct ext4_ext_path *path) | |
56055d3a | 3101 | { |
67a5da56 | 3102 | struct ext4_sb_info *sbi; |
6f91bc5f | 3103 | struct ext4_extent_header *eh; |
667eff35 YY |
3104 | struct ext4_map_blocks split_map; |
3105 | struct ext4_extent zero_ex; | |
3106 | struct ext4_extent *ex; | |
21ca087a | 3107 | ext4_lblk_t ee_block, eof_block; |
f85b287a | 3108 | unsigned int ee_len, depth; |
67a5da56 | 3109 | int allocated, max_zeroout = 0; |
56055d3a | 3110 | int err = 0; |
667eff35 | 3111 | int split_flag = 0; |
21ca087a DM |
3112 | |
3113 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3114 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3115 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a | 3116 | |
67a5da56 | 3117 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3118 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3119 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3120 | if (eof_block < map->m_lblk + map->m_len) |
3121 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3122 | |
3123 | depth = ext_depth(inode); | |
6f91bc5f | 3124 | eh = path[depth].p_hdr; |
56055d3a AA |
3125 | ex = path[depth].p_ext; |
3126 | ee_block = le32_to_cpu(ex->ee_block); | |
3127 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3128 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 3129 | |
6f91bc5f EG |
3130 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3131 | ||
3132 | /* Pre-conditions */ | |
3133 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3134 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3135 | |
3136 | /* | |
3137 | * Attempt to transfer newly initialized blocks from the currently | |
3138 | * uninitialized extent to its left neighbor. This is much cheaper | |
3139 | * than an insertion followed by a merge as those involve costly | |
3140 | * memmove() calls. This is the common case in steady state for | |
3141 | * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append | |
3142 | * writes. | |
3143 | * | |
3144 | * Limitations of the current logic: | |
3145 | * - L1: we only deal with writes at the start of the extent. | |
3146 | * The approach could be extended to writes at the end | |
3147 | * of the extent but this scenario was deemed less common. | |
3148 | * - L2: we do not deal with writes covering the whole extent. | |
3149 | * This would require removing the extent if the transfer | |
3150 | * is possible. | |
3151 | * - L3: we only attempt to merge with an extent stored in the | |
3152 | * same extent tree node. | |
3153 | */ | |
3154 | if ((map->m_lblk == ee_block) && /*L1*/ | |
3155 | (map->m_len < ee_len) && /*L2*/ | |
3156 | (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/ | |
3157 | struct ext4_extent *prev_ex; | |
3158 | ext4_lblk_t prev_lblk; | |
3159 | ext4_fsblk_t prev_pblk, ee_pblk; | |
3160 | unsigned int prev_len, write_len; | |
3161 | ||
3162 | prev_ex = ex - 1; | |
3163 | prev_lblk = le32_to_cpu(prev_ex->ee_block); | |
3164 | prev_len = ext4_ext_get_actual_len(prev_ex); | |
3165 | prev_pblk = ext4_ext_pblock(prev_ex); | |
3166 | ee_pblk = ext4_ext_pblock(ex); | |
3167 | write_len = map->m_len; | |
3168 | ||
3169 | /* | |
3170 | * A transfer of blocks from 'ex' to 'prev_ex' is allowed | |
3171 | * upon those conditions: | |
3172 | * - C1: prev_ex is initialized, | |
3173 | * - C2: prev_ex is logically abutting ex, | |
3174 | * - C3: prev_ex is physically abutting ex, | |
3175 | * - C4: prev_ex can receive the additional blocks without | |
3176 | * overflowing the (initialized) length limit. | |
3177 | */ | |
3178 | if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/ | |
3179 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ | |
3180 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
3181 | (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/ | |
3182 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3183 | if (err) | |
3184 | goto out; | |
3185 | ||
3186 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3187 | map, ex, prev_ex); | |
3188 | ||
3189 | /* Shift the start of ex by 'write_len' blocks */ | |
3190 | ex->ee_block = cpu_to_le32(ee_block + write_len); | |
3191 | ext4_ext_store_pblock(ex, ee_pblk + write_len); | |
3192 | ex->ee_len = cpu_to_le16(ee_len - write_len); | |
3193 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3194 | ||
3195 | /* Extend prev_ex by 'write_len' blocks */ | |
3196 | prev_ex->ee_len = cpu_to_le16(prev_len + write_len); | |
3197 | ||
3198 | /* Mark the block containing both extents as dirty */ | |
3199 | ext4_ext_dirty(handle, inode, path + depth); | |
3200 | ||
3201 | /* Update path to point to the right extent */ | |
3202 | path[depth].p_ext = prev_ex; | |
3203 | ||
3204 | /* Result: number of initialized blocks past m_lblk */ | |
3205 | allocated = write_len; | |
3206 | goto out; | |
3207 | } | |
3208 | } | |
3209 | ||
667eff35 | 3210 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3211 | /* |
3212 | * It is safe to convert extent to initialized via explicit | |
3213 | * zeroout only if extent is fully insde i_size or new_size. | |
3214 | */ | |
667eff35 | 3215 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3216 | |
67a5da56 ZL |
3217 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3218 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
3219 | inode->i_sb->s_blocksize_bits; | |
3220 | ||
3221 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3222 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3223 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3224 | if (err) |
d03856bd | 3225 | goto out; |
d03856bd AK |
3226 | |
3227 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3228 | if (err) | |
3229 | goto out; | |
667eff35 | 3230 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3231 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3232 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3233 | goto out; |
56055d3a | 3234 | } |
667eff35 | 3235 | |
56055d3a | 3236 | /* |
667eff35 YY |
3237 | * four cases: |
3238 | * 1. split the extent into three extents. | |
3239 | * 2. split the extent into two extents, zeroout the first half. | |
3240 | * 3. split the extent into two extents, zeroout the second half. | |
3241 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3242 | */ |
667eff35 YY |
3243 | split_map.m_lblk = map->m_lblk; |
3244 | split_map.m_len = map->m_len; | |
3245 | ||
67a5da56 ZL |
3246 | if (max_zeroout && (allocated > map->m_len)) { |
3247 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3248 | /* case 3 */ |
3249 | zero_ex.ee_block = | |
9b940f8e AH |
3250 | cpu_to_le32(map->m_lblk); |
3251 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3252 | ext4_ext_store_pblock(&zero_ex, |
3253 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3254 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3255 | if (err) |
3256 | goto out; | |
667eff35 YY |
3257 | split_map.m_lblk = map->m_lblk; |
3258 | split_map.m_len = allocated; | |
67a5da56 | 3259 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3260 | /* case 2 */ |
3261 | if (map->m_lblk != ee_block) { | |
3262 | zero_ex.ee_block = ex->ee_block; | |
3263 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3264 | ee_block); | |
3265 | ext4_ext_store_pblock(&zero_ex, | |
3266 | ext4_ext_pblock(ex)); | |
3267 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3268 | if (err) | |
3269 | goto out; | |
3270 | } | |
3271 | ||
667eff35 | 3272 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3273 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3274 | allocated = map->m_len; | |
56055d3a AA |
3275 | } |
3276 | } | |
667eff35 YY |
3277 | |
3278 | allocated = ext4_split_extent(handle, inode, path, | |
67a5da56 | 3279 | &split_map, split_flag, 0); |
667eff35 YY |
3280 | if (allocated < 0) |
3281 | err = allocated; | |
3282 | ||
56055d3a AA |
3283 | out: |
3284 | return err ? err : allocated; | |
3285 | } | |
3286 | ||
0031462b | 3287 | /* |
e35fd660 | 3288 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3289 | * ext4_get_blocks_dio_write() when DIO to write |
3290 | * to an uninitialized extent. | |
3291 | * | |
fd018fe8 | 3292 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3293 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3294 | * There are three possibilities: |
3295 | * a> There is no split required: Entire extent should be uninitialized | |
3296 | * b> Splits in two extents: Write is happening at either end of the extent | |
3297 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3298 | * | |
3299 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3300 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3301 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3302 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3303 | * into three uninitialized extent(at most). After IO complete, the part |
3304 | * being filled will be convert to initialized by the end_io callback function | |
3305 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3306 | * |
3307 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3308 | */ |
3309 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3310 | struct inode *inode, | |
e35fd660 | 3311 | struct ext4_map_blocks *map, |
0031462b | 3312 | struct ext4_ext_path *path, |
0031462b MC |
3313 | int flags) |
3314 | { | |
667eff35 YY |
3315 | ext4_lblk_t eof_block; |
3316 | ext4_lblk_t ee_block; | |
3317 | struct ext4_extent *ex; | |
3318 | unsigned int ee_len; | |
3319 | int split_flag = 0, depth; | |
21ca087a DM |
3320 | |
3321 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3322 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3323 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3324 | |
3325 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3326 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3327 | if (eof_block < map->m_lblk + map->m_len) |
3328 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3329 | /* |
3330 | * It is safe to convert extent to initialized via explicit | |
3331 | * zeroout only if extent is fully insde i_size or new_size. | |
3332 | */ | |
667eff35 YY |
3333 | depth = ext_depth(inode); |
3334 | ex = path[depth].p_ext; | |
3335 | ee_block = le32_to_cpu(ex->ee_block); | |
3336 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3337 | |
667eff35 YY |
3338 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3339 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3340 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3341 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3342 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3343 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3344 | } |
197217a5 | 3345 | |
c7064ef1 | 3346 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3347 | struct inode *inode, |
3348 | struct ext4_map_blocks *map, | |
3349 | struct ext4_ext_path *path) | |
0031462b MC |
3350 | { |
3351 | struct ext4_extent *ex; | |
dee1f973 DM |
3352 | ext4_lblk_t ee_block; |
3353 | unsigned int ee_len; | |
0031462b MC |
3354 | int depth; |
3355 | int err = 0; | |
0031462b MC |
3356 | |
3357 | depth = ext_depth(inode); | |
0031462b | 3358 | ex = path[depth].p_ext; |
dee1f973 DM |
3359 | ee_block = le32_to_cpu(ex->ee_block); |
3360 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3361 | |
197217a5 YY |
3362 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3363 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3364 | (unsigned long long)ee_block, ee_len); |
3365 | ||
3366 | /* If extent is larger than requested then split is required */ | |
3367 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
3368 | err = ext4_split_unwritten_extents(handle, inode, map, path, | |
3369 | EXT4_GET_BLOCKS_CONVERT); | |
3370 | if (err < 0) | |
3371 | goto out; | |
3372 | ext4_ext_drop_refs(path); | |
3373 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3374 | if (IS_ERR(path)) { | |
3375 | err = PTR_ERR(path); | |
3376 | goto out; | |
3377 | } | |
3378 | depth = ext_depth(inode); | |
3379 | ex = path[depth].p_ext; | |
3380 | } | |
197217a5 | 3381 | |
0031462b MC |
3382 | err = ext4_ext_get_access(handle, inode, path + depth); |
3383 | if (err) | |
3384 | goto out; | |
3385 | /* first mark the extent as initialized */ | |
3386 | ext4_ext_mark_initialized(ex); | |
3387 | ||
197217a5 YY |
3388 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3389 | * borders are not changed | |
0031462b | 3390 | */ |
ecb94f5f | 3391 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3392 | |
0031462b | 3393 | /* Mark modified extent as dirty */ |
ecb94f5f | 3394 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3395 | out: |
3396 | ext4_ext_show_leaf(inode, path); | |
3397 | return err; | |
3398 | } | |
3399 | ||
515f41c3 AK |
3400 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3401 | sector_t block, int count) | |
3402 | { | |
3403 | int i; | |
3404 | for (i = 0; i < count; i++) | |
3405 | unmap_underlying_metadata(bdev, block + i); | |
3406 | } | |
3407 | ||
58590b06 TT |
3408 | /* |
3409 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3410 | */ | |
3411 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3412 | ext4_lblk_t lblk, |
58590b06 TT |
3413 | struct ext4_ext_path *path, |
3414 | unsigned int len) | |
3415 | { | |
3416 | int i, depth; | |
3417 | struct ext4_extent_header *eh; | |
65922cb5 | 3418 | struct ext4_extent *last_ex; |
58590b06 TT |
3419 | |
3420 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3421 | return 0; | |
3422 | ||
3423 | depth = ext_depth(inode); | |
3424 | eh = path[depth].p_hdr; | |
58590b06 | 3425 | |
afcff5d8 LC |
3426 | /* |
3427 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3428 | * do not care for this case anymore. Simply remove the flag | |
3429 | * if there are no extents. | |
3430 | */ | |
3431 | if (unlikely(!eh->eh_entries)) | |
3432 | goto out; | |
58590b06 TT |
3433 | last_ex = EXT_LAST_EXTENT(eh); |
3434 | /* | |
3435 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3436 | * last block in the last extent in the file. We test this by | |
3437 | * first checking to see if the caller to | |
3438 | * ext4_ext_get_blocks() was interested in the last block (or | |
3439 | * a block beyond the last block) in the current extent. If | |
3440 | * this turns out to be false, we can bail out from this | |
3441 | * function immediately. | |
3442 | */ | |
d002ebf1 | 3443 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3444 | ext4_ext_get_actual_len(last_ex)) |
3445 | return 0; | |
3446 | /* | |
3447 | * If the caller does appear to be planning to write at or | |
3448 | * beyond the end of the current extent, we then test to see | |
3449 | * if the current extent is the last extent in the file, by | |
3450 | * checking to make sure it was reached via the rightmost node | |
3451 | * at each level of the tree. | |
3452 | */ | |
3453 | for (i = depth-1; i >= 0; i--) | |
3454 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3455 | return 0; | |
afcff5d8 | 3456 | out: |
58590b06 TT |
3457 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3458 | return ext4_mark_inode_dirty(handle, inode); | |
3459 | } | |
3460 | ||
7b415bf6 AK |
3461 | /** |
3462 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3463 | * | |
3464 | * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns | |
3465 | * whether there are any buffers marked for delayed allocation. It returns '1' | |
3466 | * on the first delalloc'ed buffer head found. If no buffer head in the given | |
3467 | * range is marked for delalloc, it returns 0. | |
3468 | * lblk_start should always be <= lblk_end. | |
3469 | * search_hint_reverse is to indicate that searching in reverse from lblk_end to | |
3470 | * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed | |
3471 | * block sooner). This is useful when blocks are truncated sequentially from | |
3472 | * lblk_start towards lblk_end. | |
3473 | */ | |
3474 | static int ext4_find_delalloc_range(struct inode *inode, | |
3475 | ext4_lblk_t lblk_start, | |
3476 | ext4_lblk_t lblk_end, | |
3477 | int search_hint_reverse) | |
3478 | { | |
3479 | struct address_space *mapping = inode->i_mapping; | |
3480 | struct buffer_head *head, *bh = NULL; | |
3481 | struct page *page; | |
3482 | ext4_lblk_t i, pg_lblk; | |
3483 | pgoff_t index; | |
3484 | ||
8c48f7e8 RD |
3485 | if (!test_opt(inode->i_sb, DELALLOC)) |
3486 | return 0; | |
3487 | ||
7b415bf6 AK |
3488 | /* reverse search wont work if fs block size is less than page size */ |
3489 | if (inode->i_blkbits < PAGE_CACHE_SHIFT) | |
3490 | search_hint_reverse = 0; | |
3491 | ||
3492 | if (search_hint_reverse) | |
3493 | i = lblk_end; | |
3494 | else | |
3495 | i = lblk_start; | |
3496 | ||
3497 | index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
3498 | ||
3499 | while ((i >= lblk_start) && (i <= lblk_end)) { | |
3500 | page = find_get_page(mapping, index); | |
5356f261 | 3501 | if (!page) |
7b415bf6 AK |
3502 | goto nextpage; |
3503 | ||
7b415bf6 AK |
3504 | if (!page_has_buffers(page)) |
3505 | goto nextpage; | |
3506 | ||
3507 | head = page_buffers(page); | |
3508 | if (!head) | |
3509 | goto nextpage; | |
3510 | ||
3511 | bh = head; | |
3512 | pg_lblk = index << (PAGE_CACHE_SHIFT - | |
3513 | inode->i_blkbits); | |
3514 | do { | |
3515 | if (unlikely(pg_lblk < lblk_start)) { | |
3516 | /* | |
3517 | * This is possible when fs block size is less | |
3518 | * than page size and our cluster starts/ends in | |
3519 | * middle of the page. So we need to skip the | |
3520 | * initial few blocks till we reach the 'lblk' | |
3521 | */ | |
3522 | pg_lblk++; | |
3523 | continue; | |
3524 | } | |
3525 | ||
5356f261 AK |
3526 | /* Check if the buffer is delayed allocated and that it |
3527 | * is not yet mapped. (when da-buffers are mapped during | |
3528 | * their writeout, their da_mapped bit is set.) | |
3529 | */ | |
3530 | if (buffer_delay(bh) && !buffer_da_mapped(bh)) { | |
7b415bf6 | 3531 | page_cache_release(page); |
d8990240 AK |
3532 | trace_ext4_find_delalloc_range(inode, |
3533 | lblk_start, lblk_end, | |
3534 | search_hint_reverse, | |
3535 | 1, i); | |
7b415bf6 AK |
3536 | return 1; |
3537 | } | |
3538 | if (search_hint_reverse) | |
3539 | i--; | |
3540 | else | |
3541 | i++; | |
3542 | } while ((i >= lblk_start) && (i <= lblk_end) && | |
3543 | ((bh = bh->b_this_page) != head)); | |
3544 | nextpage: | |
3545 | if (page) | |
3546 | page_cache_release(page); | |
3547 | /* | |
3548 | * Move to next page. 'i' will be the first lblk in the next | |
3549 | * page. | |
3550 | */ | |
3551 | if (search_hint_reverse) | |
3552 | index--; | |
3553 | else | |
3554 | index++; | |
3555 | i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
3556 | } | |
3557 | ||
d8990240 AK |
3558 | trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end, |
3559 | search_hint_reverse, 0, 0); | |
7b415bf6 AK |
3560 | return 0; |
3561 | } | |
3562 | ||
3563 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk, | |
3564 | int search_hint_reverse) | |
3565 | { | |
3566 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3567 | ext4_lblk_t lblk_start, lblk_end; | |
3568 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3569 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3570 | ||
3571 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end, | |
3572 | search_hint_reverse); | |
3573 | } | |
3574 | ||
3575 | /** | |
3576 | * Determines how many complete clusters (out of those specified by the 'map') | |
3577 | * are under delalloc and were reserved quota for. | |
3578 | * This function is called when we are writing out the blocks that were | |
3579 | * originally written with their allocation delayed, but then the space was | |
3580 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3581 | * The cases to look for are: | |
3582 | * ('=' indicated delayed allocated blocks | |
3583 | * '-' indicates non-delayed allocated blocks) | |
3584 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3585 | * are not delalloc'ed. | |
3586 | * Ex: | |
3587 | * |----c---=|====c====|====c====|===-c----| | |
3588 | * |++++++ allocated ++++++| | |
3589 | * ==> 4 complete clusters in above example | |
3590 | * | |
3591 | * (b) partial cluster (outside of allocated range) towards either end is | |
3592 | * marked for delayed allocation. In this case, we will exclude that | |
3593 | * cluster. | |
3594 | * Ex: | |
3595 | * |----====c========|========c========| | |
3596 | * |++++++ allocated ++++++| | |
3597 | * ==> 1 complete clusters in above example | |
3598 | * | |
3599 | * Ex: | |
3600 | * |================c================| | |
3601 | * |++++++ allocated ++++++| | |
3602 | * ==> 0 complete clusters in above example | |
3603 | * | |
3604 | * The ext4_da_update_reserve_space will be called only if we | |
3605 | * determine here that there were some "entire" clusters that span | |
3606 | * this 'allocated' range. | |
3607 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3608 | * without ever calling ext4_find_delalloc_range. | |
3609 | */ | |
3610 | static unsigned int | |
3611 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3612 | unsigned int num_blks) | |
3613 | { | |
3614 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3615 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3616 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3617 | unsigned int allocated_clusters = 0; | |
3618 | ||
3619 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3620 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3621 | ||
3622 | /* max possible clusters for this allocation */ | |
3623 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3624 | ||
d8990240 AK |
3625 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3626 | ||
7b415bf6 AK |
3627 | /* Check towards left side */ |
3628 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3629 | if (c_offset) { | |
3630 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3631 | lblk_to = lblk_from + c_offset - 1; | |
3632 | ||
3633 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0)) | |
3634 | allocated_clusters--; | |
3635 | } | |
3636 | ||
3637 | /* Now check towards right. */ | |
3638 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3639 | if (allocated_clusters && c_offset) { | |
3640 | lblk_from = lblk_start + num_blks; | |
3641 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3642 | ||
3643 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0)) | |
3644 | allocated_clusters--; | |
3645 | } | |
3646 | ||
3647 | return allocated_clusters; | |
3648 | } | |
3649 | ||
0031462b MC |
3650 | static int |
3651 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3652 | struct ext4_map_blocks *map, |
0031462b | 3653 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3654 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3655 | { |
3656 | int ret = 0; | |
3657 | int err = 0; | |
f45ee3a1 | 3658 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3659 | |
88635ca2 ZL |
3660 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3661 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3662 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3663 | flags, allocated); |
3664 | ext4_ext_show_leaf(inode, path); | |
3665 | ||
b5645534 ZL |
3666 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3667 | allocated, newblock); | |
d8990240 | 3668 | |
c7064ef1 | 3669 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3670 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3671 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3672 | path, flags); | |
82e54229 DM |
3673 | if (ret <= 0) |
3674 | goto out; | |
5f524950 M |
3675 | /* |
3676 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3677 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3678 | * completed |
3679 | */ | |
0edeb71d TM |
3680 | if (io) |
3681 | ext4_set_io_unwritten_flag(inode, io); | |
3682 | else | |
19f5fb7a | 3683 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
744692dc | 3684 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3685 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3686 | goto out; |
3687 | } | |
c7064ef1 | 3688 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3689 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3690 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3691 | path); |
58590b06 | 3692 | if (ret >= 0) { |
b436b9be | 3693 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3694 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3695 | path, map->m_len); | |
58590b06 TT |
3696 | } else |
3697 | err = ret; | |
0031462b MC |
3698 | goto out2; |
3699 | } | |
3700 | /* buffered IO case */ | |
3701 | /* | |
3702 | * repeat fallocate creation request | |
3703 | * we already have an unwritten extent | |
3704 | */ | |
3705 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | |
3706 | goto map_out; | |
3707 | ||
3708 | /* buffered READ or buffered write_begin() lookup */ | |
3709 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3710 | /* | |
3711 | * We have blocks reserved already. We | |
3712 | * return allocated blocks so that delalloc | |
3713 | * won't do block reservation for us. But | |
3714 | * the buffer head will be unmapped so that | |
3715 | * a read from the block returns 0s. | |
3716 | */ | |
e35fd660 | 3717 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3718 | goto out1; |
3719 | } | |
3720 | ||
3721 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3722 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
a4e5d88b | 3723 | if (ret >= 0) |
b436b9be | 3724 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3725 | out: |
3726 | if (ret <= 0) { | |
3727 | err = ret; | |
3728 | goto out2; | |
3729 | } else | |
3730 | allocated = ret; | |
e35fd660 | 3731 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3732 | /* |
3733 | * if we allocated more blocks than requested | |
3734 | * we need to make sure we unmap the extra block | |
3735 | * allocated. The actual needed block will get | |
3736 | * unmapped later when we find the buffer_head marked | |
3737 | * new. | |
3738 | */ | |
e35fd660 | 3739 | if (allocated > map->m_len) { |
515f41c3 | 3740 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3741 | newblock + map->m_len, |
3742 | allocated - map->m_len); | |
3743 | allocated = map->m_len; | |
515f41c3 | 3744 | } |
5f634d06 AK |
3745 | |
3746 | /* | |
3747 | * If we have done fallocate with the offset that is already | |
3748 | * delayed allocated, we would have block reservation | |
3749 | * and quota reservation done in the delayed write path. | |
3750 | * But fallocate would have already updated quota and block | |
3751 | * count for this offset. So cancel these reservation | |
3752 | */ | |
7b415bf6 AK |
3753 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3754 | unsigned int reserved_clusters; | |
3755 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3756 | map->m_lblk, map->m_len); | |
3757 | if (reserved_clusters) | |
3758 | ext4_da_update_reserve_space(inode, | |
3759 | reserved_clusters, | |
3760 | 0); | |
3761 | } | |
5f634d06 | 3762 | |
0031462b | 3763 | map_out: |
e35fd660 | 3764 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3765 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3766 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3767 | map->m_len); | |
3768 | if (err < 0) | |
3769 | goto out2; | |
3770 | } | |
0031462b | 3771 | out1: |
e35fd660 TT |
3772 | if (allocated > map->m_len) |
3773 | allocated = map->m_len; | |
0031462b | 3774 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3775 | map->m_pblk = newblock; |
3776 | map->m_len = allocated; | |
0031462b MC |
3777 | out2: |
3778 | if (path) { | |
3779 | ext4_ext_drop_refs(path); | |
3780 | kfree(path); | |
3781 | } | |
3782 | return err ? err : allocated; | |
3783 | } | |
58590b06 | 3784 | |
4d33b1ef TT |
3785 | /* |
3786 | * get_implied_cluster_alloc - check to see if the requested | |
3787 | * allocation (in the map structure) overlaps with a cluster already | |
3788 | * allocated in an extent. | |
d8990240 | 3789 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3790 | * @map The requested lblk->pblk mapping |
3791 | * @ex The extent structure which might contain an implied | |
3792 | * cluster allocation | |
3793 | * | |
3794 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3795 | * find blocks that were already in the inode's extent tree. Hence, | |
3796 | * we know that the beginning of the requested region cannot overlap | |
3797 | * the extent from the inode's extent tree. There are three cases we | |
3798 | * want to catch. The first is this case: | |
3799 | * | |
3800 | * |--- cluster # N--| | |
3801 | * |--- extent ---| |---- requested region ---| | |
3802 | * |==========| | |
3803 | * | |
3804 | * The second case that we need to test for is this one: | |
3805 | * | |
3806 | * |--------- cluster # N ----------------| | |
3807 | * |--- requested region --| |------- extent ----| | |
3808 | * |=======================| | |
3809 | * | |
3810 | * The third case is when the requested region lies between two extents | |
3811 | * within the same cluster: | |
3812 | * |------------- cluster # N-------------| | |
3813 | * |----- ex -----| |---- ex_right ----| | |
3814 | * |------ requested region ------| | |
3815 | * |================| | |
3816 | * | |
3817 | * In each of the above cases, we need to set the map->m_pblk and | |
3818 | * map->m_len so it corresponds to the return the extent labelled as | |
3819 | * "|====|" from cluster #N, since it is already in use for data in | |
3820 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3821 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3822 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3823 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3824 | * by calling ext4_mb_new_blocks(). | |
3825 | */ | |
d8990240 | 3826 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3827 | struct ext4_map_blocks *map, |
3828 | struct ext4_extent *ex, | |
3829 | struct ext4_ext_path *path) | |
3830 | { | |
d8990240 | 3831 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3832 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3833 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3834 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3835 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3836 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3837 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3838 | ||
3839 | /* The extent passed in that we are trying to match */ | |
3840 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3841 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3842 | ||
3843 | /* The requested region passed into ext4_map_blocks() */ | |
3844 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3845 | |
3846 | if ((rr_cluster_start == ex_cluster_end) || | |
3847 | (rr_cluster_start == ex_cluster_start)) { | |
3848 | if (rr_cluster_start == ex_cluster_end) | |
3849 | ee_start += ee_len - 1; | |
3850 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3851 | c_offset; | |
3852 | map->m_len = min(map->m_len, | |
3853 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3854 | /* | |
3855 | * Check for and handle this case: | |
3856 | * | |
3857 | * |--------- cluster # N-------------| | |
3858 | * |------- extent ----| | |
3859 | * |--- requested region ---| | |
3860 | * |===========| | |
3861 | */ | |
3862 | ||
3863 | if (map->m_lblk < ee_block) | |
3864 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3865 | ||
3866 | /* | |
3867 | * Check for the case where there is already another allocated | |
3868 | * block to the right of 'ex' but before the end of the cluster. | |
3869 | * | |
3870 | * |------------- cluster # N-------------| | |
3871 | * |----- ex -----| |---- ex_right ----| | |
3872 | * |------ requested region ------| | |
3873 | * |================| | |
3874 | */ | |
3875 | if (map->m_lblk > ee_block) { | |
3876 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3877 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3878 | } | |
d8990240 AK |
3879 | |
3880 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3881 | return 1; |
3882 | } | |
d8990240 AK |
3883 | |
3884 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3885 | return 0; |
3886 | } | |
3887 | ||
3888 | ||
c278bfec | 3889 | /* |
f5ab0d1f MC |
3890 | * Block allocation/map/preallocation routine for extents based files |
3891 | * | |
3892 | * | |
c278bfec | 3893 | * Need to be called with |
0e855ac8 AK |
3894 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3895 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3896 | * |
3897 | * return > 0, number of of blocks already mapped/allocated | |
3898 | * if create == 0 and these are pre-allocated blocks | |
3899 | * buffer head is unmapped | |
3900 | * otherwise blocks are mapped | |
3901 | * | |
3902 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3903 | * buffer head is unmapped | |
3904 | * | |
3905 | * return < 0, error case. | |
c278bfec | 3906 | */ |
e35fd660 TT |
3907 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3908 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3909 | { |
3910 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3911 | struct ext4_extent newex, *ex, *ex2; |
3912 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3913 | ext4_fsblk_t newblock = 0; |
37794732 | 3914 | int free_on_err = 0, err = 0, depth; |
4d33b1ef | 3915 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 3916 | unsigned int allocated_clusters = 0; |
c9de560d | 3917 | struct ext4_allocation_request ar; |
f45ee3a1 | 3918 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 3919 | ext4_lblk_t cluster_offset; |
82e54229 | 3920 | int set_unwritten = 0; |
a86c6181 | 3921 | |
84fe3bef | 3922 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3923 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3924 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3925 | |
3926 | /* check in cache */ | |
7877191c | 3927 | if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) { |
b05e6ae5 | 3928 | if (!newex.ee_start_lo && !newex.ee_start_hi) { |
7b415bf6 AK |
3929 | if ((sbi->s_cluster_ratio > 1) && |
3930 | ext4_find_delalloc_cluster(inode, map->m_lblk, 0)) | |
3931 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
3932 | ||
c2177057 | 3933 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3934 | /* |
3935 | * block isn't allocated yet and | |
3936 | * user doesn't want to allocate it | |
3937 | */ | |
a86c6181 AT |
3938 | goto out2; |
3939 | } | |
3940 | /* we should allocate requested block */ | |
b05e6ae5 | 3941 | } else { |
a86c6181 | 3942 | /* block is already allocated */ |
7b415bf6 AK |
3943 | if (sbi->s_cluster_ratio > 1) |
3944 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
e35fd660 | 3945 | newblock = map->m_lblk |
8c55e204 | 3946 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3947 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3948 | /* number of remaining blocks in the extent */ |
b939e376 | 3949 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3950 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3951 | goto out; |
a86c6181 AT |
3952 | } |
3953 | } | |
3954 | ||
3955 | /* find extent for this block */ | |
e35fd660 | 3956 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3957 | if (IS_ERR(path)) { |
3958 | err = PTR_ERR(path); | |
3959 | path = NULL; | |
3960 | goto out2; | |
3961 | } | |
3962 | ||
3963 | depth = ext_depth(inode); | |
3964 | ||
3965 | /* | |
d0d856e8 RD |
3966 | * consistent leaf must not be empty; |
3967 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3968 | * this is why assert can't be put in ext4_ext_find_extent() |
3969 | */ | |
273df556 FM |
3970 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3971 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3972 | "lblock: %lu, depth: %d pblock %lld", |
3973 | (unsigned long) map->m_lblk, depth, | |
3974 | path[depth].p_block); | |
034fb4c9 SP |
3975 | err = -EIO; |
3976 | goto out2; | |
3977 | } | |
a86c6181 | 3978 | |
7e028976 AM |
3979 | ex = path[depth].p_ext; |
3980 | if (ex) { | |
725d26d3 | 3981 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3982 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3983 | unsigned short ee_len; |
471d4011 SB |
3984 | |
3985 | /* | |
471d4011 | 3986 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3987 | * we split out initialized portions during a write. |
471d4011 | 3988 | */ |
a2df2a63 | 3989 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
3990 | |
3991 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
3992 | ||
d0d856e8 | 3993 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
3994 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
3995 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 3996 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3997 | allocated = ee_len - (map->m_lblk - ee_block); |
3998 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3999 | ee_block, ee_len, newblock); | |
56055d3a | 4000 | |
e861304b | 4001 | /* |
7877191c LC |
4002 | * Do not put uninitialized extent |
4003 | * in the cache | |
e861304b | 4004 | */ |
7877191c LC |
4005 | if (!ext4_ext_is_uninitialized(ex)) { |
4006 | ext4_ext_put_in_cache(inode, ee_block, | |
4007 | ee_len, ee_start); | |
4008 | goto out; | |
f7d0d379 | 4009 | } |
37794732 | 4010 | allocated = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
4011 | handle, inode, map, path, flags, |
4012 | allocated, newblock); | |
37794732 | 4013 | goto out3; |
a86c6181 AT |
4014 | } |
4015 | } | |
4016 | ||
7b415bf6 AK |
4017 | if ((sbi->s_cluster_ratio > 1) && |
4018 | ext4_find_delalloc_cluster(inode, map->m_lblk, 0)) | |
4019 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
4020 | ||
a86c6181 | 4021 | /* |
d0d856e8 | 4022 | * requested block isn't allocated yet; |
a86c6181 AT |
4023 | * we couldn't try to create block if create flag is zero |
4024 | */ | |
c2177057 | 4025 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4026 | /* |
4027 | * put just found gap into cache to speed up | |
4028 | * subsequent requests | |
4029 | */ | |
e35fd660 | 4030 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
4031 | goto out2; |
4032 | } | |
4d33b1ef | 4033 | |
a86c6181 | 4034 | /* |
c2ea3fde | 4035 | * Okay, we need to do block allocation. |
63f57933 | 4036 | */ |
7b415bf6 | 4037 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4038 | newex.ee_block = cpu_to_le32(map->m_lblk); |
4039 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
4040 | ||
4041 | /* | |
4042 | * If we are doing bigalloc, check to see if the extent returned | |
4043 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4044 | */ | |
4045 | if (cluster_offset && ex && | |
d8990240 | 4046 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4047 | ar.len = allocated = map->m_len; |
4048 | newblock = map->m_pblk; | |
7b415bf6 | 4049 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4050 | goto got_allocated_blocks; |
4051 | } | |
a86c6181 | 4052 | |
c9de560d | 4053 | /* find neighbour allocated blocks */ |
e35fd660 | 4054 | ar.lleft = map->m_lblk; |
c9de560d AT |
4055 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4056 | if (err) | |
4057 | goto out2; | |
e35fd660 | 4058 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4059 | ex2 = NULL; |
4060 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4061 | if (err) |
4062 | goto out2; | |
25d14f98 | 4063 | |
4d33b1ef TT |
4064 | /* Check if the extent after searching to the right implies a |
4065 | * cluster we can use. */ | |
4066 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4067 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4068 | ar.len = allocated = map->m_len; |
4069 | newblock = map->m_pblk; | |
7b415bf6 | 4070 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4071 | goto got_allocated_blocks; |
4072 | } | |
4073 | ||
749269fa AA |
4074 | /* |
4075 | * See if request is beyond maximum number of blocks we can have in | |
4076 | * a single extent. For an initialized extent this limit is | |
4077 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4078 | * EXT_UNINIT_MAX_LEN. | |
4079 | */ | |
e35fd660 | 4080 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4081 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4082 | map->m_len = EXT_INIT_MAX_LEN; |
4083 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4084 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4085 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4086 | |
e35fd660 | 4087 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4088 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4089 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4090 | if (err) |
b939e376 | 4091 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4092 | else |
e35fd660 | 4093 | allocated = map->m_len; |
c9de560d AT |
4094 | |
4095 | /* allocate new block */ | |
4096 | ar.inode = inode; | |
e35fd660 TT |
4097 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4098 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4099 | /* |
4100 | * We calculate the offset from the beginning of the cluster | |
4101 | * for the logical block number, since when we allocate a | |
4102 | * physical cluster, the physical block should start at the | |
4103 | * same offset from the beginning of the cluster. This is | |
4104 | * needed so that future calls to get_implied_cluster_alloc() | |
4105 | * work correctly. | |
4106 | */ | |
4107 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4108 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4109 | ar.goal -= offset; | |
4110 | ar.logical -= offset; | |
c9de560d AT |
4111 | if (S_ISREG(inode->i_mode)) |
4112 | ar.flags = EXT4_MB_HINT_DATA; | |
4113 | else | |
4114 | /* disable in-core preallocation for non-regular files */ | |
4115 | ar.flags = 0; | |
556b27ab VH |
4116 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4117 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4118 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4119 | if (!newblock) |
4120 | goto out2; | |
84fe3bef | 4121 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4122 | ar.goal, newblock, allocated); |
4d33b1ef | 4123 | free_on_err = 1; |
7b415bf6 | 4124 | allocated_clusters = ar.len; |
4d33b1ef TT |
4125 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4126 | if (ar.len > allocated) | |
4127 | ar.len = allocated; | |
a86c6181 | 4128 | |
4d33b1ef | 4129 | got_allocated_blocks: |
a86c6181 | 4130 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4131 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4132 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4133 | /* Mark uninitialized */ |
4134 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4135 | ext4_ext_mark_uninitialized(&newex); |
8d5d02e6 | 4136 | /* |
744692dc | 4137 | * io_end structure was created for every IO write to an |
25985edc | 4138 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4139 | * here we flag the IO that really needs the conversion. |
5f524950 | 4140 | * For non asycn direct IO case, flag the inode state |
25985edc | 4141 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4142 | */ |
82e54229 DM |
4143 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4144 | set_unwritten = 1; | |
744692dc | 4145 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4146 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4147 | } |
c8d46e41 | 4148 | |
a4e5d88b DM |
4149 | err = 0; |
4150 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4151 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4152 | path, ar.len); | |
575a1d4b JZ |
4153 | if (!err) |
4154 | err = ext4_ext_insert_extent(handle, inode, path, | |
4155 | &newex, flags); | |
82e54229 DM |
4156 | |
4157 | if (!err && set_unwritten) { | |
4158 | if (io) | |
4159 | ext4_set_io_unwritten_flag(inode, io); | |
4160 | else | |
4161 | ext4_set_inode_state(inode, | |
4162 | EXT4_STATE_DIO_UNWRITTEN); | |
4163 | } | |
4164 | ||
4d33b1ef | 4165 | if (err && free_on_err) { |
7132de74 MP |
4166 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4167 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4168 | /* free data blocks we just allocated */ |
c9de560d AT |
4169 | /* not a good idea to call discard here directly, |
4170 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4171 | ext4_discard_preallocations(inode); |
7dc57615 | 4172 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4173 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4174 | goto out2; |
315054f0 | 4175 | } |
a86c6181 | 4176 | |
a86c6181 | 4177 | /* previous routine could use block we allocated */ |
bf89d16f | 4178 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4179 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4180 | if (allocated > map->m_len) |
4181 | allocated = map->m_len; | |
4182 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4183 | |
5f634d06 AK |
4184 | /* |
4185 | * Update reserved blocks/metadata blocks after successful | |
4186 | * block allocation which had been deferred till now. | |
4187 | */ | |
7b415bf6 | 4188 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4189 | unsigned int reserved_clusters; |
7b415bf6 | 4190 | /* |
81fdbb4a | 4191 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4192 | */ |
4193 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4194 | map->m_lblk, allocated); | |
4195 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4196 | if (reserved_clusters) { | |
4197 | /* | |
4198 | * We have clusters reserved for this range. | |
4199 | * But since we are not doing actual allocation | |
4200 | * and are simply using blocks from previously | |
4201 | * allocated cluster, we should release the | |
4202 | * reservation and not claim quota. | |
4203 | */ | |
4204 | ext4_da_update_reserve_space(inode, | |
4205 | reserved_clusters, 0); | |
4206 | } | |
4207 | } else { | |
4208 | BUG_ON(allocated_clusters < reserved_clusters); | |
4209 | /* We will claim quota for all newly allocated blocks.*/ | |
4210 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4211 | 1); | |
4212 | if (reserved_clusters < allocated_clusters) { | |
5356f261 | 4213 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4214 | int reservation = allocated_clusters - |
4215 | reserved_clusters; | |
4216 | /* | |
4217 | * It seems we claimed few clusters outside of | |
4218 | * the range of this allocation. We should give | |
4219 | * it back to the reservation pool. This can | |
4220 | * happen in the following case: | |
4221 | * | |
4222 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4223 | * cluster has 4 blocks. Thus, the clusters | |
4224 | * are [0-3],[4-7],[8-11]... | |
4225 | * * First comes delayed allocation write for | |
4226 | * logical blocks 10 & 11. Since there were no | |
4227 | * previous delayed allocated blocks in the | |
4228 | * range [8-11], we would reserve 1 cluster | |
4229 | * for this write. | |
4230 | * * Next comes write for logical blocks 3 to 8. | |
4231 | * In this case, we will reserve 2 clusters | |
4232 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4233 | * that range has a delayed allocated blocks. | |
4234 | * Thus total reserved clusters now becomes 3. | |
4235 | * * Now, during the delayed allocation writeout | |
4236 | * time, we will first write blocks [3-8] and | |
4237 | * allocate 3 clusters for writing these | |
4238 | * blocks. Also, we would claim all these | |
4239 | * three clusters above. | |
4240 | * * Now when we come here to writeout the | |
4241 | * blocks [10-11], we would expect to claim | |
4242 | * the reservation of 1 cluster we had made | |
4243 | * (and we would claim it since there are no | |
4244 | * more delayed allocated blocks in the range | |
4245 | * [8-11]. But our reserved cluster count had | |
4246 | * already gone to 0. | |
4247 | * | |
4248 | * Thus, at the step 4 above when we determine | |
4249 | * that there are still some unwritten delayed | |
4250 | * allocated blocks outside of our current | |
4251 | * block range, we should increment the | |
4252 | * reserved clusters count so that when the | |
4253 | * remaining blocks finally gets written, we | |
4254 | * could claim them. | |
4255 | */ | |
5356f261 AK |
4256 | dquot_reserve_block(inode, |
4257 | EXT4_C2B(sbi, reservation)); | |
4258 | spin_lock(&ei->i_block_reservation_lock); | |
4259 | ei->i_reserved_data_blocks += reservation; | |
4260 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 AK |
4261 | } |
4262 | } | |
4263 | } | |
5f634d06 | 4264 | |
b436b9be JK |
4265 | /* |
4266 | * Cache the extent and update transaction to commit on fdatasync only | |
4267 | * when it is _not_ an uninitialized extent. | |
4268 | */ | |
4269 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 4270 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
4271 | ext4_update_inode_fsync_trans(handle, inode, 1); |
4272 | } else | |
4273 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 4274 | out: |
e35fd660 TT |
4275 | if (allocated > map->m_len) |
4276 | allocated = map->m_len; | |
a86c6181 | 4277 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4278 | map->m_flags |= EXT4_MAP_MAPPED; |
4279 | map->m_pblk = newblock; | |
4280 | map->m_len = allocated; | |
a86c6181 AT |
4281 | out2: |
4282 | if (path) { | |
4283 | ext4_ext_drop_refs(path); | |
4284 | kfree(path); | |
4285 | } | |
e861304b | 4286 | |
37794732 | 4287 | out3: |
19b303d8 | 4288 | trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated); |
e7b319e3 | 4289 | |
7877191c | 4290 | return err ? err : allocated; |
a86c6181 AT |
4291 | } |
4292 | ||
cf108bca | 4293 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
4294 | { |
4295 | struct address_space *mapping = inode->i_mapping; | |
4296 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 4297 | ext4_lblk_t last_block; |
a86c6181 | 4298 | handle_t *handle; |
189e868f | 4299 | loff_t page_len; |
a86c6181 AT |
4300 | int err = 0; |
4301 | ||
3889fd57 JZ |
4302 | /* |
4303 | * finish any pending end_io work so we won't run the risk of | |
4304 | * converting any truncated blocks to initialized later | |
4305 | */ | |
c278531d | 4306 | ext4_flush_unwritten_io(inode); |
3889fd57 | 4307 | |
a86c6181 AT |
4308 | /* |
4309 | * probably first extent we're gonna free will be last in block | |
4310 | */ | |
f3bd1f3f | 4311 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 4312 | handle = ext4_journal_start(inode, err); |
cf108bca | 4313 | if (IS_ERR(handle)) |
a86c6181 | 4314 | return; |
a86c6181 | 4315 | |
189e868f AH |
4316 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
4317 | page_len = PAGE_CACHE_SIZE - | |
4318 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4319 | ||
4320 | err = ext4_discard_partial_page_buffers(handle, | |
4321 | mapping, inode->i_size, page_len, 0); | |
4322 | ||
4323 | if (err) | |
4324 | goto out_stop; | |
4325 | } | |
a86c6181 | 4326 | |
9ddfc3dc JK |
4327 | if (ext4_orphan_add(handle, inode)) |
4328 | goto out_stop; | |
4329 | ||
0e855ac8 | 4330 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
4331 | ext4_ext_invalidate_cache(inode); |
4332 | ||
c2ea3fde | 4333 | ext4_discard_preallocations(inode); |
c9de560d | 4334 | |
a86c6181 | 4335 | /* |
d0d856e8 RD |
4336 | * TODO: optimization is possible here. |
4337 | * Probably we need not scan at all, | |
4338 | * because page truncation is enough. | |
a86c6181 | 4339 | */ |
a86c6181 AT |
4340 | |
4341 | /* we have to know where to truncate from in crash case */ | |
4342 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4343 | ext4_mark_inode_dirty(handle, inode); | |
4344 | ||
4345 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4346 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
51865fda ZL |
4347 | err = ext4_es_remove_extent(inode, last_block, |
4348 | EXT_MAX_BLOCKS - last_block); | |
5f95d21f | 4349 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
a86c6181 AT |
4350 | |
4351 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
4352 | * transaction synchronous. |
4353 | */ | |
a86c6181 | 4354 | if (IS_SYNC(inode)) |
0390131b | 4355 | ext4_handle_sync(handle); |
a86c6181 | 4356 | |
9ddfc3dc | 4357 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
4358 | |
4359 | out_stop: | |
a86c6181 | 4360 | /* |
d0d856e8 | 4361 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
4362 | * then we need to clear up the orphan record which we created above. |
4363 | * However, if this was a real unlink then we were called by | |
4364 | * ext4_delete_inode(), and we allow that function to clean up the | |
4365 | * orphan info for us. | |
4366 | */ | |
4367 | if (inode->i_nlink) | |
4368 | ext4_orphan_del(handle, inode); | |
4369 | ||
ef737728 SR |
4370 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4371 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
4372 | ext4_journal_stop(handle); |
4373 | } | |
4374 | ||
fd28784a AK |
4375 | static void ext4_falloc_update_inode(struct inode *inode, |
4376 | int mode, loff_t new_size, int update_ctime) | |
4377 | { | |
4378 | struct timespec now; | |
4379 | ||
4380 | if (update_ctime) { | |
4381 | now = current_fs_time(inode->i_sb); | |
4382 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4383 | inode->i_ctime = now; | |
4384 | } | |
4385 | /* | |
4386 | * Update only when preallocation was requested beyond | |
4387 | * the file size. | |
4388 | */ | |
cf17fea6 AK |
4389 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4390 | if (new_size > i_size_read(inode)) | |
4391 | i_size_write(inode, new_size); | |
4392 | if (new_size > EXT4_I(inode)->i_disksize) | |
4393 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4394 | } else { |
4395 | /* | |
4396 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4397 | * can proceed even if the new size is the same as i_size. | |
4398 | */ | |
4399 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4400 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4401 | } |
4402 | ||
4403 | } | |
4404 | ||
a2df2a63 | 4405 | /* |
2fe17c10 | 4406 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4407 | * operation, which gets called from sys_fallocate system call. |
4408 | * For block-mapped files, posix_fallocate should fall back to the method | |
4409 | * of writing zeroes to the required new blocks (the same behavior which is | |
4410 | * expected for file systems which do not support fallocate() system call). | |
4411 | */ | |
2fe17c10 | 4412 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4413 | { |
2fe17c10 | 4414 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 4415 | handle_t *handle; |
fd28784a | 4416 | loff_t new_size; |
498e5f24 | 4417 | unsigned int max_blocks; |
a2df2a63 AA |
4418 | int ret = 0; |
4419 | int ret2 = 0; | |
4420 | int retries = 0; | |
a4e5d88b | 4421 | int flags; |
2ed88685 | 4422 | struct ext4_map_blocks map; |
a2df2a63 AA |
4423 | unsigned int credits, blkbits = inode->i_blkbits; |
4424 | ||
4425 | /* | |
4426 | * currently supporting (pre)allocate mode for extent-based | |
4427 | * files _only_ | |
4428 | */ | |
12e9b892 | 4429 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
a2df2a63 AA |
4430 | return -EOPNOTSUPP; |
4431 | ||
a4bb6b64 AH |
4432 | /* Return error if mode is not supported */ |
4433 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4434 | return -EOPNOTSUPP; | |
4435 | ||
4436 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4437 | return ext4_punch_hole(file, offset, len); | |
4438 | ||
0562e0ba | 4439 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4440 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4441 | /* |
4442 | * We can't just convert len to max_blocks because | |
4443 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4444 | */ | |
a2df2a63 | 4445 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4446 | - map.m_lblk; |
a2df2a63 | 4447 | /* |
f3bd1f3f | 4448 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4449 | */ |
f3bd1f3f | 4450 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4451 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4452 | ret = inode_newsize_ok(inode, (len + offset)); |
4453 | if (ret) { | |
4454 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4455 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4456 | return ret; |
4457 | } | |
3c6fe770 | 4458 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4459 | if (mode & FALLOC_FL_KEEP_SIZE) |
4460 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4461 | /* |
4462 | * Don't normalize the request if it can fit in one extent so | |
4463 | * that it doesn't get unnecessarily split into multiple | |
4464 | * extents. | |
4465 | */ | |
4466 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4467 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f DM |
4468 | |
4469 | /* Prevent race condition between unwritten */ | |
4470 | ext4_flush_unwritten_io(inode); | |
a2df2a63 AA |
4471 | retry: |
4472 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4473 | map.m_lblk = map.m_lblk + ret; |
4474 | map.m_len = max_blocks = max_blocks - ret; | |
a2df2a63 AA |
4475 | handle = ext4_journal_start(inode, credits); |
4476 | if (IS_ERR(handle)) { | |
4477 | ret = PTR_ERR(handle); | |
4478 | break; | |
4479 | } | |
a4e5d88b | 4480 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4481 | if (ret <= 0) { |
2c98615d AK |
4482 | #ifdef EXT4FS_DEBUG |
4483 | WARN_ON(ret <= 0); | |
e35fd660 | 4484 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
2c98615d | 4485 | "returned error inode#%lu, block=%u, " |
9fd9784c | 4486 | "max_blocks=%u", __func__, |
a6371b63 | 4487 | inode->i_ino, map.m_lblk, max_blocks); |
2c98615d | 4488 | #endif |
a2df2a63 AA |
4489 | ext4_mark_inode_dirty(handle, inode); |
4490 | ret2 = ext4_journal_stop(handle); | |
4491 | break; | |
4492 | } | |
2ed88685 | 4493 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4494 | blkbits) >> blkbits)) |
4495 | new_size = offset + len; | |
4496 | else | |
29ae07b7 | 4497 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4498 | |
fd28784a | 4499 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4500 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4501 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4502 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4503 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4504 | ret2 = ext4_journal_stop(handle); |
4505 | if (ret2) | |
4506 | break; | |
4507 | } | |
fd28784a AK |
4508 | if (ret == -ENOSPC && |
4509 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4510 | ret = 0; | |
a2df2a63 | 4511 | goto retry; |
a2df2a63 | 4512 | } |
55bd725a | 4513 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4514 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4515 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4516 | return ret > 0 ? ret2 : ret; |
4517 | } | |
6873fa0d | 4518 | |
0031462b MC |
4519 | /* |
4520 | * This function convert a range of blocks to written extents | |
4521 | * The caller of this function will pass the start offset and the size. | |
4522 | * all unwritten extents within this range will be converted to | |
4523 | * written extents. | |
4524 | * | |
4525 | * This function is called from the direct IO end io call back | |
4526 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4527 | * Returns 0 on success. |
0031462b MC |
4528 | */ |
4529 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4530 | ssize_t len) |
0031462b MC |
4531 | { |
4532 | handle_t *handle; | |
0031462b MC |
4533 | unsigned int max_blocks; |
4534 | int ret = 0; | |
4535 | int ret2 = 0; | |
2ed88685 | 4536 | struct ext4_map_blocks map; |
0031462b MC |
4537 | unsigned int credits, blkbits = inode->i_blkbits; |
4538 | ||
2ed88685 | 4539 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4540 | /* |
4541 | * We can't just convert len to max_blocks because | |
4542 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4543 | */ | |
2ed88685 TT |
4544 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4545 | map.m_lblk); | |
0031462b MC |
4546 | /* |
4547 | * credits to insert 1 extent into extent tree | |
4548 | */ | |
4549 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4550 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4551 | map.m_lblk += ret; |
4552 | map.m_len = (max_blocks -= ret); | |
0031462b MC |
4553 | handle = ext4_journal_start(inode, credits); |
4554 | if (IS_ERR(handle)) { | |
4555 | ret = PTR_ERR(handle); | |
4556 | break; | |
4557 | } | |
2ed88685 | 4558 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4559 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
0031462b MC |
4560 | if (ret <= 0) { |
4561 | WARN_ON(ret <= 0); | |
92b97816 TT |
4562 | ext4_msg(inode->i_sb, KERN_ERR, |
4563 | "%s:%d: inode #%lu: block %u: len %u: " | |
4564 | "ext4_ext_map_blocks returned %d", | |
4565 | __func__, __LINE__, inode->i_ino, map.m_lblk, | |
4566 | map.m_len, ret); | |
0031462b MC |
4567 | } |
4568 | ext4_mark_inode_dirty(handle, inode); | |
4569 | ret2 = ext4_journal_stop(handle); | |
4570 | if (ret <= 0 || ret2 ) | |
4571 | break; | |
4572 | } | |
4573 | return ret > 0 ? ret2 : ret; | |
4574 | } | |
6d9c85eb | 4575 | |
6873fa0d ES |
4576 | /* |
4577 | * Callback function called for each extent to gather FIEMAP information. | |
4578 | */ | |
c03f8aa9 | 4579 | static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next, |
6873fa0d ES |
4580 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
4581 | void *data) | |
4582 | { | |
6873fa0d ES |
4583 | __u64 logical; |
4584 | __u64 physical; | |
4585 | __u64 length; | |
4586 | __u32 flags = 0; | |
6d9c85eb YY |
4587 | int ret = 0; |
4588 | struct fiemap_extent_info *fieinfo = data; | |
4589 | unsigned char blksize_bits; | |
6873fa0d | 4590 | |
6d9c85eb YY |
4591 | blksize_bits = inode->i_sb->s_blocksize_bits; |
4592 | logical = (__u64)newex->ec_block << blksize_bits; | |
6873fa0d | 4593 | |
b05e6ae5 | 4594 | if (newex->ec_start == 0) { |
6d9c85eb YY |
4595 | /* |
4596 | * No extent in extent-tree contains block @newex->ec_start, | |
4597 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
4598 | * | |
4599 | * Holes or delayed-extents are processed as follows. | |
4600 | * 1. lookup dirty pages with specified range in pagecache. | |
4601 | * If no page is got, then there is no delayed-extent and | |
4602 | * return with EXT_CONTINUE. | |
4603 | * 2. find the 1st mapped buffer, | |
4604 | * 3. check if the mapped buffer is both in the request range | |
4605 | * and a delayed buffer. If not, there is no delayed-extent, | |
4606 | * then return. | |
4607 | * 4. a delayed-extent is found, the extent will be collected. | |
4608 | */ | |
4609 | ext4_lblk_t end = 0; | |
4610 | pgoff_t last_offset; | |
4611 | pgoff_t offset; | |
4612 | pgoff_t index; | |
b221349f | 4613 | pgoff_t start_index = 0; |
6d9c85eb | 4614 | struct page **pages = NULL; |
6873fa0d | 4615 | struct buffer_head *bh = NULL; |
6d9c85eb YY |
4616 | struct buffer_head *head = NULL; |
4617 | unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *); | |
4618 | ||
4619 | pages = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4620 | if (pages == NULL) | |
4621 | return -ENOMEM; | |
6873fa0d ES |
4622 | |
4623 | offset = logical >> PAGE_SHIFT; | |
6d9c85eb YY |
4624 | repeat: |
4625 | last_offset = offset; | |
4626 | head = NULL; | |
4627 | ret = find_get_pages_tag(inode->i_mapping, &offset, | |
4628 | PAGECACHE_TAG_DIRTY, nr_pages, pages); | |
4629 | ||
4630 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4631 | /* First time, try to find a mapped buffer. */ | |
4632 | if (ret == 0) { | |
4633 | out: | |
4634 | for (index = 0; index < ret; index++) | |
4635 | page_cache_release(pages[index]); | |
4636 | /* just a hole. */ | |
4637 | kfree(pages); | |
4638 | return EXT_CONTINUE; | |
4639 | } | |
b221349f | 4640 | index = 0; |
6873fa0d | 4641 | |
b221349f | 4642 | next_page: |
6d9c85eb | 4643 | /* Try to find the 1st mapped buffer. */ |
b221349f | 4644 | end = ((__u64)pages[index]->index << PAGE_SHIFT) >> |
6d9c85eb | 4645 | blksize_bits; |
b221349f | 4646 | if (!page_has_buffers(pages[index])) |
6d9c85eb | 4647 | goto out; |
b221349f | 4648 | head = page_buffers(pages[index]); |
6d9c85eb YY |
4649 | if (!head) |
4650 | goto out; | |
6873fa0d | 4651 | |
b221349f | 4652 | index++; |
6d9c85eb YY |
4653 | bh = head; |
4654 | do { | |
b221349f YY |
4655 | if (end >= newex->ec_block + |
4656 | newex->ec_len) | |
4657 | /* The buffer is out of | |
4658 | * the request range. | |
4659 | */ | |
4660 | goto out; | |
4661 | ||
4662 | if (buffer_mapped(bh) && | |
4663 | end >= newex->ec_block) { | |
4664 | start_index = index - 1; | |
6d9c85eb | 4665 | /* get the 1st mapped buffer. */ |
6d9c85eb YY |
4666 | goto found_mapped_buffer; |
4667 | } | |
b221349f | 4668 | |
6d9c85eb YY |
4669 | bh = bh->b_this_page; |
4670 | end++; | |
4671 | } while (bh != head); | |
6873fa0d | 4672 | |
b221349f YY |
4673 | /* No mapped buffer in the range found in this page, |
4674 | * We need to look up next page. | |
4675 | */ | |
4676 | if (index >= ret) { | |
4677 | /* There is no page left, but we need to limit | |
4678 | * newex->ec_len. | |
4679 | */ | |
4680 | newex->ec_len = end - newex->ec_block; | |
4681 | goto out; | |
4682 | } | |
4683 | goto next_page; | |
6873fa0d | 4684 | } else { |
6d9c85eb YY |
4685 | /*Find contiguous delayed buffers. */ |
4686 | if (ret > 0 && pages[0]->index == last_offset) | |
4687 | head = page_buffers(pages[0]); | |
4688 | bh = head; | |
b221349f YY |
4689 | index = 1; |
4690 | start_index = 0; | |
6873fa0d | 4691 | } |
6d9c85eb YY |
4692 | |
4693 | found_mapped_buffer: | |
4694 | if (bh != NULL && buffer_delay(bh)) { | |
4695 | /* 1st or contiguous delayed buffer found. */ | |
4696 | if (!(flags & FIEMAP_EXTENT_DELALLOC)) { | |
4697 | /* | |
4698 | * 1st delayed buffer found, record | |
4699 | * the start of extent. | |
4700 | */ | |
4701 | flags |= FIEMAP_EXTENT_DELALLOC; | |
4702 | newex->ec_block = end; | |
4703 | logical = (__u64)end << blksize_bits; | |
4704 | } | |
4705 | /* Find contiguous delayed buffers. */ | |
4706 | do { | |
4707 | if (!buffer_delay(bh)) | |
4708 | goto found_delayed_extent; | |
4709 | bh = bh->b_this_page; | |
4710 | end++; | |
4711 | } while (bh != head); | |
4712 | ||
b221349f | 4713 | for (; index < ret; index++) { |
6d9c85eb YY |
4714 | if (!page_has_buffers(pages[index])) { |
4715 | bh = NULL; | |
4716 | break; | |
4717 | } | |
4718 | head = page_buffers(pages[index]); | |
4719 | if (!head) { | |
4720 | bh = NULL; | |
4721 | break; | |
4722 | } | |
b221349f | 4723 | |
6d9c85eb | 4724 | if (pages[index]->index != |
b221349f YY |
4725 | pages[start_index]->index + index |
4726 | - start_index) { | |
6d9c85eb YY |
4727 | /* Blocks are not contiguous. */ |
4728 | bh = NULL; | |
4729 | break; | |
4730 | } | |
4731 | bh = head; | |
4732 | do { | |
4733 | if (!buffer_delay(bh)) | |
4734 | /* Delayed-extent ends. */ | |
4735 | goto found_delayed_extent; | |
4736 | bh = bh->b_this_page; | |
4737 | end++; | |
4738 | } while (bh != head); | |
4739 | } | |
4740 | } else if (!(flags & FIEMAP_EXTENT_DELALLOC)) | |
4741 | /* a hole found. */ | |
4742 | goto out; | |
4743 | ||
4744 | found_delayed_extent: | |
4745 | newex->ec_len = min(end - newex->ec_block, | |
4746 | (ext4_lblk_t)EXT_INIT_MAX_LEN); | |
4747 | if (ret == nr_pages && bh != NULL && | |
4748 | newex->ec_len < EXT_INIT_MAX_LEN && | |
4749 | buffer_delay(bh)) { | |
4750 | /* Have not collected an extent and continue. */ | |
4751 | for (index = 0; index < ret; index++) | |
4752 | page_cache_release(pages[index]); | |
4753 | goto repeat; | |
6873fa0d | 4754 | } |
6d9c85eb YY |
4755 | |
4756 | for (index = 0; index < ret; index++) | |
4757 | page_cache_release(pages[index]); | |
4758 | kfree(pages); | |
6873fa0d ES |
4759 | } |
4760 | ||
4761 | physical = (__u64)newex->ec_start << blksize_bits; | |
4762 | length = (__u64)newex->ec_len << blksize_bits; | |
4763 | ||
4764 | if (ex && ext4_ext_is_uninitialized(ex)) | |
4765 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
4766 | ||
c03f8aa9 | 4767 | if (next == EXT_MAX_BLOCKS) |
6873fa0d ES |
4768 | flags |= FIEMAP_EXTENT_LAST; |
4769 | ||
6d9c85eb | 4770 | ret = fiemap_fill_next_extent(fieinfo, logical, physical, |
6873fa0d | 4771 | length, flags); |
6d9c85eb YY |
4772 | if (ret < 0) |
4773 | return ret; | |
4774 | if (ret == 1) | |
6873fa0d | 4775 | return EXT_BREAK; |
6873fa0d ES |
4776 | return EXT_CONTINUE; |
4777 | } | |
6873fa0d ES |
4778 | /* fiemap flags we can handle specified here */ |
4779 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4780 | ||
3a06d778 AK |
4781 | static int ext4_xattr_fiemap(struct inode *inode, |
4782 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4783 | { |
4784 | __u64 physical = 0; | |
4785 | __u64 length; | |
4786 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4787 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4788 | int error = 0; | |
4789 | ||
4790 | /* in-inode? */ | |
19f5fb7a | 4791 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4792 | struct ext4_iloc iloc; |
4793 | int offset; /* offset of xattr in inode */ | |
4794 | ||
4795 | error = ext4_get_inode_loc(inode, &iloc); | |
4796 | if (error) | |
4797 | return error; | |
4798 | physical = iloc.bh->b_blocknr << blockbits; | |
4799 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4800 | EXT4_I(inode)->i_extra_isize; | |
4801 | physical += offset; | |
4802 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4803 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4804 | brelse(iloc.bh); |
6873fa0d ES |
4805 | } else { /* external block */ |
4806 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4807 | length = inode->i_sb->s_blocksize; | |
4808 | } | |
4809 | ||
4810 | if (physical) | |
4811 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4812 | length, flags); | |
4813 | return (error < 0 ? error : 0); | |
4814 | } | |
4815 | ||
a4bb6b64 AH |
4816 | /* |
4817 | * ext4_ext_punch_hole | |
4818 | * | |
4819 | * Punches a hole of "length" bytes in a file starting | |
4820 | * at byte "offset" | |
4821 | * | |
4822 | * @inode: The inode of the file to punch a hole in | |
4823 | * @offset: The starting byte offset of the hole | |
4824 | * @length: The length of the hole | |
4825 | * | |
4826 | * Returns the number of blocks removed or negative on err | |
4827 | */ | |
4828 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4829 | { | |
4830 | struct inode *inode = file->f_path.dentry->d_inode; | |
4831 | struct super_block *sb = inode->i_sb; | |
5f95d21f | 4832 | ext4_lblk_t first_block, stop_block; |
a4bb6b64 | 4833 | struct address_space *mapping = inode->i_mapping; |
a4bb6b64 | 4834 | handle_t *handle; |
ba06208a AH |
4835 | loff_t first_page, last_page, page_len; |
4836 | loff_t first_page_offset, last_page_offset; | |
5f95d21f | 4837 | int credits, err = 0; |
a4bb6b64 | 4838 | |
02d262df DM |
4839 | /* |
4840 | * Write out all dirty pages to avoid race conditions | |
4841 | * Then release them. | |
4842 | */ | |
4843 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4844 | err = filemap_write_and_wait_range(mapping, | |
4845 | offset, offset + length - 1); | |
4846 | ||
4847 | if (err) | |
4848 | return err; | |
4849 | } | |
4850 | ||
4851 | mutex_lock(&inode->i_mutex); | |
4852 | /* It's not possible punch hole on append only file */ | |
4853 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
4854 | err = -EPERM; | |
4855 | goto out_mutex; | |
4856 | } | |
4857 | if (IS_SWAPFILE(inode)) { | |
4858 | err = -ETXTBSY; | |
4859 | goto out_mutex; | |
4860 | } | |
4861 | ||
2be4751b AH |
4862 | /* No need to punch hole beyond i_size */ |
4863 | if (offset >= inode->i_size) | |
02d262df | 4864 | goto out_mutex; |
2be4751b AH |
4865 | |
4866 | /* | |
4867 | * If the hole extends beyond i_size, set the hole | |
4868 | * to end after the page that contains i_size | |
4869 | */ | |
4870 | if (offset + length > inode->i_size) { | |
4871 | length = inode->i_size + | |
4872 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4873 | offset; | |
4874 | } | |
4875 | ||
a4bb6b64 AH |
4876 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4877 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4878 | ||
4879 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4880 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4881 | ||
a4bb6b64 AH |
4882 | /* Now release the pages */ |
4883 | if (last_page_offset > first_page_offset) { | |
5e44f8c3 HD |
4884 | truncate_pagecache_range(inode, first_page_offset, |
4885 | last_page_offset - 1); | |
a4bb6b64 AH |
4886 | } |
4887 | ||
02d262df DM |
4888 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
4889 | ext4_inode_block_unlocked_dio(inode); | |
c278531d | 4890 | err = ext4_flush_unwritten_io(inode); |
28a535f9 | 4891 | if (err) |
02d262df | 4892 | goto out_dio; |
c278531d | 4893 | inode_dio_wait(inode); |
a4bb6b64 AH |
4894 | |
4895 | credits = ext4_writepage_trans_blocks(inode); | |
4896 | handle = ext4_journal_start(inode, credits); | |
02d262df DM |
4897 | if (IS_ERR(handle)) { |
4898 | err = PTR_ERR(handle); | |
4899 | goto out_dio; | |
4900 | } | |
a4bb6b64 | 4901 | |
a4bb6b64 AH |
4902 | |
4903 | /* | |
ba06208a AH |
4904 | * Now we need to zero out the non-page-aligned data in the |
4905 | * pages at the start and tail of the hole, and unmap the buffer | |
4906 | * heads for the block aligned regions of the page that were | |
4907 | * completely zeroed. | |
a4bb6b64 | 4908 | */ |
ba06208a AH |
4909 | if (first_page > last_page) { |
4910 | /* | |
4911 | * If the file space being truncated is contained within a page | |
4912 | * just zero out and unmap the middle of that page | |
4913 | */ | |
4914 | err = ext4_discard_partial_page_buffers(handle, | |
4915 | mapping, offset, length, 0); | |
4916 | ||
4917 | if (err) | |
4918 | goto out; | |
4919 | } else { | |
4920 | /* | |
4921 | * zero out and unmap the partial page that contains | |
4922 | * the start of the hole | |
4923 | */ | |
4924 | page_len = first_page_offset - offset; | |
4925 | if (page_len > 0) { | |
4926 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4927 | offset, page_len, 0); | |
4928 | if (err) | |
4929 | goto out; | |
4930 | } | |
4931 | ||
4932 | /* | |
4933 | * zero out and unmap the partial page that contains | |
4934 | * the end of the hole | |
4935 | */ | |
4936 | page_len = offset + length - last_page_offset; | |
4937 | if (page_len > 0) { | |
4938 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4939 | last_page_offset, page_len, 0); | |
4940 | if (err) | |
4941 | goto out; | |
a4bb6b64 AH |
4942 | } |
4943 | } | |
4944 | ||
2be4751b AH |
4945 | /* |
4946 | * If i_size is contained in the last page, we need to | |
4947 | * unmap and zero the partial page after i_size | |
4948 | */ | |
4949 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4950 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4951 | ||
4952 | page_len = PAGE_CACHE_SIZE - | |
4953 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4954 | ||
4955 | if (page_len > 0) { | |
4956 | err = ext4_discard_partial_page_buffers(handle, | |
4957 | mapping, inode->i_size, page_len, 0); | |
4958 | ||
4959 | if (err) | |
4960 | goto out; | |
4961 | } | |
4962 | } | |
4963 | ||
5f95d21f LC |
4964 | first_block = (offset + sb->s_blocksize - 1) >> |
4965 | EXT4_BLOCK_SIZE_BITS(sb); | |
4966 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4967 | ||
a4bb6b64 | 4968 | /* If there are no blocks to remove, return now */ |
5f95d21f | 4969 | if (first_block >= stop_block) |
a4bb6b64 AH |
4970 | goto out; |
4971 | ||
4972 | down_write(&EXT4_I(inode)->i_data_sem); | |
4973 | ext4_ext_invalidate_cache(inode); | |
4974 | ext4_discard_preallocations(inode); | |
4975 | ||
51865fda ZL |
4976 | err = ext4_es_remove_extent(inode, first_block, |
4977 | stop_block - first_block); | |
5f95d21f | 4978 | err = ext4_ext_remove_space(inode, first_block, stop_block - 1); |
a4bb6b64 | 4979 | |
5f95d21f LC |
4980 | ext4_ext_invalidate_cache(inode); |
4981 | ext4_discard_preallocations(inode); | |
a4bb6b64 AH |
4982 | |
4983 | if (IS_SYNC(inode)) | |
4984 | ext4_handle_sync(handle); | |
4985 | ||
4986 | up_write(&EXT4_I(inode)->i_data_sem); | |
4987 | ||
4988 | out: | |
a4bb6b64 AH |
4989 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4990 | ext4_mark_inode_dirty(handle, inode); | |
4991 | ext4_journal_stop(handle); | |
02d262df DM |
4992 | out_dio: |
4993 | ext4_inode_resume_unlocked_dio(inode); | |
4994 | out_mutex: | |
4995 | mutex_unlock(&inode->i_mutex); | |
a4bb6b64 AH |
4996 | return err; |
4997 | } | |
6873fa0d ES |
4998 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4999 | __u64 start, __u64 len) | |
5000 | { | |
5001 | ext4_lblk_t start_blk; | |
6873fa0d ES |
5002 | int error = 0; |
5003 | ||
5004 | /* fallback to generic here if not in extents fmt */ | |
12e9b892 | 5005 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
5006 | return generic_block_fiemap(inode, fieinfo, start, len, |
5007 | ext4_get_block); | |
5008 | ||
5009 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
5010 | return -EBADR; | |
5011 | ||
5012 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
5013 | error = ext4_xattr_fiemap(inode, fieinfo); | |
5014 | } else { | |
aca92ff6 LM |
5015 | ext4_lblk_t len_blks; |
5016 | __u64 last_blk; | |
5017 | ||
6873fa0d | 5018 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 5019 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
5020 | if (last_blk >= EXT_MAX_BLOCKS) |
5021 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 5022 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
5023 | |
5024 | /* | |
5025 | * Walk the extent tree gathering extent information. | |
5026 | * ext4_ext_fiemap_cb will push extents back to user. | |
5027 | */ | |
6873fa0d ES |
5028 | error = ext4_ext_walk_space(inode, start_blk, len_blks, |
5029 | ext4_ext_fiemap_cb, fieinfo); | |
6873fa0d ES |
5030 | } |
5031 | ||
5032 | return error; | |
5033 | } |