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" |
4a092d73 | 44 | #include "ext4_extents.h" |
f19d5870 | 45 | #include "xattr.h" |
a86c6181 | 46 | |
0562e0ba JZ |
47 | #include <trace/events/ext4.h> |
48 | ||
5f95d21f LC |
49 | /* |
50 | * used by extent splitting. | |
51 | */ | |
52 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
53 | due to ENOSPC */ | |
54 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
55 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
56 | ||
dee1f973 DM |
57 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
58 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
59 | ||
7ac5990d DW |
60 | static __le32 ext4_extent_block_csum(struct inode *inode, |
61 | struct ext4_extent_header *eh) | |
62 | { | |
63 | struct ext4_inode_info *ei = EXT4_I(inode); | |
64 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
65 | __u32 csum; | |
66 | ||
67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
68 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
69 | return cpu_to_le32(csum); | |
70 | } | |
71 | ||
72 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
73 | struct ext4_extent_header *eh) | |
74 | { | |
75 | struct ext4_extent_tail *et; | |
76 | ||
77 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
78 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
79 | return 1; | |
80 | ||
81 | et = find_ext4_extent_tail(eh); | |
82 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
83 | return 0; | |
84 | return 1; | |
85 | } | |
86 | ||
87 | static void ext4_extent_block_csum_set(struct inode *inode, | |
88 | struct ext4_extent_header *eh) | |
89 | { | |
90 | struct ext4_extent_tail *et; | |
91 | ||
92 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
93 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
94 | return; | |
95 | ||
96 | et = find_ext4_extent_tail(eh); | |
97 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
98 | } | |
99 | ||
d583fb87 AH |
100 | static int ext4_split_extent(handle_t *handle, |
101 | struct inode *inode, | |
102 | struct ext4_ext_path *path, | |
103 | struct ext4_map_blocks *map, | |
104 | int split_flag, | |
105 | int flags); | |
106 | ||
5f95d21f LC |
107 | static int ext4_split_extent_at(handle_t *handle, |
108 | struct inode *inode, | |
109 | struct ext4_ext_path *path, | |
110 | ext4_lblk_t split, | |
111 | int split_flag, | |
112 | int flags); | |
113 | ||
91dd8c11 LC |
114 | static int ext4_find_delayed_extent(struct inode *inode, |
115 | struct ext4_ext_cache *newex); | |
116 | ||
487caeef JK |
117 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
118 | struct inode *inode, | |
119 | int needed) | |
a86c6181 AT |
120 | { |
121 | int err; | |
122 | ||
0390131b FM |
123 | if (!ext4_handle_valid(handle)) |
124 | return 0; | |
a86c6181 | 125 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
126 | return 0; |
127 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 128 | if (err <= 0) |
9102e4fa | 129 | return err; |
487caeef | 130 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
131 | if (err == 0) |
132 | err = -EAGAIN; | |
487caeef JK |
133 | |
134 | return err; | |
a86c6181 AT |
135 | } |
136 | ||
137 | /* | |
138 | * could return: | |
139 | * - EROFS | |
140 | * - ENOMEM | |
141 | */ | |
142 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
143 | struct ext4_ext_path *path) | |
144 | { | |
145 | if (path->p_bh) { | |
146 | /* path points to block */ | |
147 | return ext4_journal_get_write_access(handle, path->p_bh); | |
148 | } | |
149 | /* path points to leaf/index in inode body */ | |
150 | /* we use in-core data, no need to protect them */ | |
151 | return 0; | |
152 | } | |
153 | ||
154 | /* | |
155 | * could return: | |
156 | * - EROFS | |
157 | * - ENOMEM | |
158 | * - EIO | |
159 | */ | |
9ea7a0df TT |
160 | #define ext4_ext_dirty(handle, inode, path) \ |
161 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
162 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
163 | handle_t *handle, struct inode *inode, | |
164 | struct ext4_ext_path *path) | |
a86c6181 AT |
165 | { |
166 | int err; | |
167 | if (path->p_bh) { | |
7ac5990d | 168 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 169 | /* path points to block */ |
9ea7a0df TT |
170 | err = __ext4_handle_dirty_metadata(where, line, handle, |
171 | inode, path->p_bh); | |
a86c6181 AT |
172 | } else { |
173 | /* path points to leaf/index in inode body */ | |
174 | err = ext4_mark_inode_dirty(handle, inode); | |
175 | } | |
176 | return err; | |
177 | } | |
178 | ||
f65e6fba | 179 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 180 | struct ext4_ext_path *path, |
725d26d3 | 181 | ext4_lblk_t block) |
a86c6181 | 182 | { |
a86c6181 | 183 | if (path) { |
81fdbb4a | 184 | int depth = path->p_depth; |
a86c6181 | 185 | struct ext4_extent *ex; |
a86c6181 | 186 | |
ad4fb9ca KM |
187 | /* |
188 | * Try to predict block placement assuming that we are | |
189 | * filling in a file which will eventually be | |
190 | * non-sparse --- i.e., in the case of libbfd writing | |
191 | * an ELF object sections out-of-order but in a way | |
192 | * the eventually results in a contiguous object or | |
193 | * executable file, or some database extending a table | |
194 | * space file. However, this is actually somewhat | |
195 | * non-ideal if we are writing a sparse file such as | |
196 | * qemu or KVM writing a raw image file that is going | |
197 | * to stay fairly sparse, since it will end up | |
198 | * fragmenting the file system's free space. Maybe we | |
199 | * should have some hueristics or some way to allow | |
200 | * userspace to pass a hint to file system, | |
b8d6568a | 201 | * especially if the latter case turns out to be |
ad4fb9ca KM |
202 | * common. |
203 | */ | |
7e028976 | 204 | ex = path[depth].p_ext; |
ad4fb9ca KM |
205 | if (ex) { |
206 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
207 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
208 | ||
209 | if (block > ext_block) | |
210 | return ext_pblk + (block - ext_block); | |
211 | else | |
212 | return ext_pblk - (ext_block - block); | |
213 | } | |
a86c6181 | 214 | |
d0d856e8 RD |
215 | /* it looks like index is empty; |
216 | * try to find starting block from index itself */ | |
a86c6181 AT |
217 | if (path[depth].p_bh) |
218 | return path[depth].p_bh->b_blocknr; | |
219 | } | |
220 | ||
221 | /* OK. use inode's group */ | |
f86186b4 | 222 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
223 | } |
224 | ||
654b4908 AK |
225 | /* |
226 | * Allocation for a meta data block | |
227 | */ | |
f65e6fba | 228 | static ext4_fsblk_t |
654b4908 | 229 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 230 | struct ext4_ext_path *path, |
55f020db | 231 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 232 | { |
f65e6fba | 233 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
234 | |
235 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
236 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
237 | NULL, err); | |
a86c6181 AT |
238 | return newblock; |
239 | } | |
240 | ||
55ad63bf | 241 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
242 | { |
243 | int size; | |
244 | ||
245 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
246 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 247 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
248 | if (!check && size > 6) |
249 | size = 6; | |
a86c6181 AT |
250 | #endif |
251 | return size; | |
252 | } | |
253 | ||
55ad63bf | 254 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
255 | { |
256 | int size; | |
257 | ||
258 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
259 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 260 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
261 | if (!check && size > 5) |
262 | size = 5; | |
a86c6181 AT |
263 | #endif |
264 | return size; | |
265 | } | |
266 | ||
55ad63bf | 267 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
268 | { |
269 | int size; | |
270 | ||
271 | size = sizeof(EXT4_I(inode)->i_data); | |
272 | size -= sizeof(struct ext4_extent_header); | |
273 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 274 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
275 | if (!check && size > 3) |
276 | size = 3; | |
a86c6181 AT |
277 | #endif |
278 | return size; | |
279 | } | |
280 | ||
55ad63bf | 281 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
282 | { |
283 | int size; | |
284 | ||
285 | size = sizeof(EXT4_I(inode)->i_data); | |
286 | size -= sizeof(struct ext4_extent_header); | |
287 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 288 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
289 | if (!check && size > 4) |
290 | size = 4; | |
a86c6181 AT |
291 | #endif |
292 | return size; | |
293 | } | |
294 | ||
d2a17637 MC |
295 | /* |
296 | * Calculate the number of metadata blocks needed | |
297 | * to allocate @blocks | |
298 | * Worse case is one block per extent | |
299 | */ | |
01f49d0b | 300 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 301 | { |
9d0be502 | 302 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 303 | int idxs; |
d2a17637 | 304 | |
9d0be502 TT |
305 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
306 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
307 | |
308 | /* | |
9d0be502 TT |
309 | * If the new delayed allocation block is contiguous with the |
310 | * previous da block, it can share index blocks with the | |
311 | * previous block, so we only need to allocate a new index | |
312 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
313 | * an additional index block, and at ldxs**3 blocks, yet | |
314 | * another index blocks. | |
d2a17637 | 315 | */ |
9d0be502 TT |
316 | if (ei->i_da_metadata_calc_len && |
317 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
318 | int num = 0; |
319 | ||
9d0be502 TT |
320 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
321 | num++; | |
322 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
323 | num++; | |
324 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
325 | num++; | |
326 | ei->i_da_metadata_calc_len = 0; | |
327 | } else | |
328 | ei->i_da_metadata_calc_len++; | |
329 | ei->i_da_metadata_calc_last_lblock++; | |
330 | return num; | |
331 | } | |
d2a17637 | 332 | |
9d0be502 TT |
333 | /* |
334 | * In the worst case we need a new set of index blocks at | |
335 | * every level of the inode's extent tree. | |
336 | */ | |
337 | ei->i_da_metadata_calc_len = 1; | |
338 | ei->i_da_metadata_calc_last_lblock = lblock; | |
339 | return ext_depth(inode) + 1; | |
d2a17637 MC |
340 | } |
341 | ||
c29c0ae7 AT |
342 | static int |
343 | ext4_ext_max_entries(struct inode *inode, int depth) | |
344 | { | |
345 | int max; | |
346 | ||
347 | if (depth == ext_depth(inode)) { | |
348 | if (depth == 0) | |
55ad63bf | 349 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 350 | else |
55ad63bf | 351 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
352 | } else { |
353 | if (depth == 0) | |
55ad63bf | 354 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 355 | else |
55ad63bf | 356 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
357 | } |
358 | ||
359 | return max; | |
360 | } | |
361 | ||
56b19868 AK |
362 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
363 | { | |
bf89d16f | 364 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 365 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 366 | |
31d4f3a2 TT |
367 | if (len == 0) |
368 | return 0; | |
6fd058f7 | 369 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
370 | } |
371 | ||
372 | static int ext4_valid_extent_idx(struct inode *inode, | |
373 | struct ext4_extent_idx *ext_idx) | |
374 | { | |
bf89d16f | 375 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 376 | |
6fd058f7 | 377 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
378 | } |
379 | ||
380 | static int ext4_valid_extent_entries(struct inode *inode, | |
381 | struct ext4_extent_header *eh, | |
382 | int depth) | |
383 | { | |
56b19868 AK |
384 | unsigned short entries; |
385 | if (eh->eh_entries == 0) | |
386 | return 1; | |
387 | ||
388 | entries = le16_to_cpu(eh->eh_entries); | |
389 | ||
390 | if (depth == 0) { | |
391 | /* leaf entries */ | |
81fdbb4a | 392 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
56b19868 AK |
393 | while (entries) { |
394 | if (!ext4_valid_extent(inode, ext)) | |
395 | return 0; | |
396 | ext++; | |
397 | entries--; | |
398 | } | |
399 | } else { | |
81fdbb4a | 400 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
401 | while (entries) { |
402 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
403 | return 0; | |
404 | ext_idx++; | |
405 | entries--; | |
406 | } | |
407 | } | |
408 | return 1; | |
409 | } | |
410 | ||
c398eda0 TT |
411 | static int __ext4_ext_check(const char *function, unsigned int line, |
412 | struct inode *inode, struct ext4_extent_header *eh, | |
413 | int depth) | |
c29c0ae7 AT |
414 | { |
415 | const char *error_msg; | |
416 | int max = 0; | |
417 | ||
418 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
419 | error_msg = "invalid magic"; | |
420 | goto corrupted; | |
421 | } | |
422 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
423 | error_msg = "unexpected eh_depth"; | |
424 | goto corrupted; | |
425 | } | |
426 | if (unlikely(eh->eh_max == 0)) { | |
427 | error_msg = "invalid eh_max"; | |
428 | goto corrupted; | |
429 | } | |
430 | max = ext4_ext_max_entries(inode, depth); | |
431 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
432 | error_msg = "too large eh_max"; | |
433 | goto corrupted; | |
434 | } | |
435 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
436 | error_msg = "invalid eh_entries"; | |
437 | goto corrupted; | |
438 | } | |
56b19868 AK |
439 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
440 | error_msg = "invalid extent entries"; | |
441 | goto corrupted; | |
442 | } | |
7ac5990d DW |
443 | /* Verify checksum on non-root extent tree nodes */ |
444 | if (ext_depth(inode) != depth && | |
445 | !ext4_extent_block_csum_verify(inode, eh)) { | |
446 | error_msg = "extent tree corrupted"; | |
447 | goto corrupted; | |
448 | } | |
c29c0ae7 AT |
449 | return 0; |
450 | ||
451 | corrupted: | |
c398eda0 | 452 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 453 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 454 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 455 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
456 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
457 | max, le16_to_cpu(eh->eh_depth), depth); | |
458 | ||
459 | return -EIO; | |
460 | } | |
461 | ||
56b19868 | 462 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 463 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 464 | |
7a262f7c AK |
465 | int ext4_ext_check_inode(struct inode *inode) |
466 | { | |
467 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
468 | } | |
469 | ||
f8489128 DW |
470 | static int __ext4_ext_check_block(const char *function, unsigned int line, |
471 | struct inode *inode, | |
472 | struct ext4_extent_header *eh, | |
473 | int depth, | |
474 | struct buffer_head *bh) | |
475 | { | |
476 | int ret; | |
477 | ||
478 | if (buffer_verified(bh)) | |
479 | return 0; | |
480 | ret = ext4_ext_check(inode, eh, depth); | |
481 | if (ret) | |
482 | return ret; | |
483 | set_buffer_verified(bh); | |
484 | return ret; | |
485 | } | |
486 | ||
487 | #define ext4_ext_check_block(inode, eh, depth, bh) \ | |
488 | __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh) | |
489 | ||
a86c6181 AT |
490 | #ifdef EXT_DEBUG |
491 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
492 | { | |
493 | int k, l = path->p_depth; | |
494 | ||
495 | ext_debug("path:"); | |
496 | for (k = 0; k <= l; k++, path++) { | |
497 | if (path->p_idx) { | |
2ae02107 | 498 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 499 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 500 | } else if (path->p_ext) { |
553f9008 | 501 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 502 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 503 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 504 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 505 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
506 | } else |
507 | ext_debug(" []"); | |
508 | } | |
509 | ext_debug("\n"); | |
510 | } | |
511 | ||
512 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
513 | { | |
514 | int depth = ext_depth(inode); | |
515 | struct ext4_extent_header *eh; | |
516 | struct ext4_extent *ex; | |
517 | int i; | |
518 | ||
519 | if (!path) | |
520 | return; | |
521 | ||
522 | eh = path[depth].p_hdr; | |
523 | ex = EXT_FIRST_EXTENT(eh); | |
524 | ||
553f9008 M |
525 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
526 | ||
a86c6181 | 527 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
528 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
529 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 530 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
531 | } |
532 | ext_debug("\n"); | |
533 | } | |
1b16da77 YY |
534 | |
535 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
536 | ext4_fsblk_t newblock, int level) | |
537 | { | |
538 | int depth = ext_depth(inode); | |
539 | struct ext4_extent *ex; | |
540 | ||
541 | if (depth != level) { | |
542 | struct ext4_extent_idx *idx; | |
543 | idx = path[level].p_idx; | |
544 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
545 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
546 | le32_to_cpu(idx->ei_block), | |
547 | ext4_idx_pblock(idx), | |
548 | newblock); | |
549 | idx++; | |
550 | } | |
551 | ||
552 | return; | |
553 | } | |
554 | ||
555 | ex = path[depth].p_ext; | |
556 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
557 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
558 | le32_to_cpu(ex->ee_block), | |
559 | ext4_ext_pblock(ex), | |
560 | ext4_ext_is_uninitialized(ex), | |
561 | ext4_ext_get_actual_len(ex), | |
562 | newblock); | |
563 | ex++; | |
564 | } | |
565 | } | |
566 | ||
a86c6181 | 567 | #else |
af5bc92d TT |
568 | #define ext4_ext_show_path(inode, path) |
569 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 570 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
571 | #endif |
572 | ||
b35905c1 | 573 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
574 | { |
575 | int depth = path->p_depth; | |
576 | int i; | |
577 | ||
578 | for (i = 0; i <= depth; i++, path++) | |
579 | if (path->p_bh) { | |
580 | brelse(path->p_bh); | |
581 | path->p_bh = NULL; | |
582 | } | |
583 | } | |
584 | ||
585 | /* | |
d0d856e8 RD |
586 | * ext4_ext_binsearch_idx: |
587 | * binary search for the closest index of the given block | |
c29c0ae7 | 588 | * the header must be checked before calling this |
a86c6181 AT |
589 | */ |
590 | static void | |
725d26d3 AK |
591 | ext4_ext_binsearch_idx(struct inode *inode, |
592 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
593 | { |
594 | struct ext4_extent_header *eh = path->p_hdr; | |
595 | struct ext4_extent_idx *r, *l, *m; | |
596 | ||
a86c6181 | 597 | |
bba90743 | 598 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
599 | |
600 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 601 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
602 | while (l <= r) { |
603 | m = l + (r - l) / 2; | |
604 | if (block < le32_to_cpu(m->ei_block)) | |
605 | r = m - 1; | |
606 | else | |
607 | l = m + 1; | |
26d535ed DM |
608 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
609 | m, le32_to_cpu(m->ei_block), | |
610 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
611 | } |
612 | ||
613 | path->p_idx = l - 1; | |
4a3c3a51 | 614 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 615 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
616 | |
617 | #ifdef CHECK_BINSEARCH | |
618 | { | |
619 | struct ext4_extent_idx *chix, *ix; | |
620 | int k; | |
621 | ||
622 | chix = ix = EXT_FIRST_INDEX(eh); | |
623 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
624 | if (k != 0 && | |
625 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
626 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
627 | "first=0x%p\n", k, | |
628 | ix, EXT_FIRST_INDEX(eh)); | |
629 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
630 | le32_to_cpu(ix->ei_block), |
631 | le32_to_cpu(ix[-1].ei_block)); | |
632 | } | |
633 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 634 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
635 | if (block < le32_to_cpu(ix->ei_block)) |
636 | break; | |
637 | chix = ix; | |
638 | } | |
639 | BUG_ON(chix != path->p_idx); | |
640 | } | |
641 | #endif | |
642 | ||
643 | } | |
644 | ||
645 | /* | |
d0d856e8 RD |
646 | * ext4_ext_binsearch: |
647 | * binary search for closest extent of the given block | |
c29c0ae7 | 648 | * the header must be checked before calling this |
a86c6181 AT |
649 | */ |
650 | static void | |
725d26d3 AK |
651 | ext4_ext_binsearch(struct inode *inode, |
652 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
653 | { |
654 | struct ext4_extent_header *eh = path->p_hdr; | |
655 | struct ext4_extent *r, *l, *m; | |
656 | ||
a86c6181 AT |
657 | if (eh->eh_entries == 0) { |
658 | /* | |
d0d856e8 RD |
659 | * this leaf is empty: |
660 | * we get such a leaf in split/add case | |
a86c6181 AT |
661 | */ |
662 | return; | |
663 | } | |
664 | ||
bba90743 | 665 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
666 | |
667 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 668 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
669 | |
670 | while (l <= r) { | |
671 | m = l + (r - l) / 2; | |
672 | if (block < le32_to_cpu(m->ee_block)) | |
673 | r = m - 1; | |
674 | else | |
675 | l = m + 1; | |
26d535ed DM |
676 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
677 | m, le32_to_cpu(m->ee_block), | |
678 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
679 | } |
680 | ||
681 | path->p_ext = l - 1; | |
553f9008 | 682 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 683 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 684 | ext4_ext_pblock(path->p_ext), |
553f9008 | 685 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 686 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
687 | |
688 | #ifdef CHECK_BINSEARCH | |
689 | { | |
690 | struct ext4_extent *chex, *ex; | |
691 | int k; | |
692 | ||
693 | chex = ex = EXT_FIRST_EXTENT(eh); | |
694 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
695 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 696 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
697 | if (block < le32_to_cpu(ex->ee_block)) |
698 | break; | |
699 | chex = ex; | |
700 | } | |
701 | BUG_ON(chex != path->p_ext); | |
702 | } | |
703 | #endif | |
704 | ||
705 | } | |
706 | ||
707 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
708 | { | |
709 | struct ext4_extent_header *eh; | |
710 | ||
711 | eh = ext_inode_hdr(inode); | |
712 | eh->eh_depth = 0; | |
713 | eh->eh_entries = 0; | |
714 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 715 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
716 | ext4_mark_inode_dirty(handle, inode); |
717 | ext4_ext_invalidate_cache(inode); | |
718 | return 0; | |
719 | } | |
720 | ||
721 | struct ext4_ext_path * | |
725d26d3 AK |
722 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
723 | struct ext4_ext_path *path) | |
a86c6181 AT |
724 | { |
725 | struct ext4_extent_header *eh; | |
726 | struct buffer_head *bh; | |
727 | short int depth, i, ppos = 0, alloc = 0; | |
728 | ||
729 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 730 | depth = ext_depth(inode); |
a86c6181 AT |
731 | |
732 | /* account possible depth increase */ | |
733 | if (!path) { | |
5d4958f9 | 734 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
735 | GFP_NOFS); |
736 | if (!path) | |
737 | return ERR_PTR(-ENOMEM); | |
738 | alloc = 1; | |
739 | } | |
a86c6181 | 740 | path[0].p_hdr = eh; |
1973adcb | 741 | path[0].p_bh = NULL; |
a86c6181 | 742 | |
c29c0ae7 | 743 | i = depth; |
a86c6181 AT |
744 | /* walk through the tree */ |
745 | while (i) { | |
746 | ext_debug("depth %d: num %d, max %d\n", | |
747 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 748 | |
a86c6181 | 749 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 750 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
751 | path[ppos].p_depth = i; |
752 | path[ppos].p_ext = NULL; | |
753 | ||
7a262f7c AK |
754 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
755 | if (unlikely(!bh)) | |
a86c6181 | 756 | goto err; |
7a262f7c | 757 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
758 | trace_ext4_ext_load_extent(inode, block, |
759 | path[ppos].p_block); | |
7a262f7c AK |
760 | if (bh_submit_read(bh) < 0) { |
761 | put_bh(bh); | |
762 | goto err; | |
763 | } | |
7a262f7c | 764 | } |
a86c6181 AT |
765 | eh = ext_block_hdr(bh); |
766 | ppos++; | |
273df556 FM |
767 | if (unlikely(ppos > depth)) { |
768 | put_bh(bh); | |
769 | EXT4_ERROR_INODE(inode, | |
770 | "ppos %d > depth %d", ppos, depth); | |
771 | goto err; | |
772 | } | |
a86c6181 AT |
773 | path[ppos].p_bh = bh; |
774 | path[ppos].p_hdr = eh; | |
775 | i--; | |
776 | ||
f8489128 | 777 | if (ext4_ext_check_block(inode, eh, i, bh)) |
a86c6181 AT |
778 | goto err; |
779 | } | |
780 | ||
781 | path[ppos].p_depth = i; | |
a86c6181 AT |
782 | path[ppos].p_ext = NULL; |
783 | path[ppos].p_idx = NULL; | |
784 | ||
a86c6181 AT |
785 | /* find extent */ |
786 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
787 | /* if not an empty leaf */ |
788 | if (path[ppos].p_ext) | |
bf89d16f | 789 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
790 | |
791 | ext4_ext_show_path(inode, path); | |
792 | ||
793 | return path; | |
794 | ||
795 | err: | |
796 | ext4_ext_drop_refs(path); | |
797 | if (alloc) | |
798 | kfree(path); | |
799 | return ERR_PTR(-EIO); | |
800 | } | |
801 | ||
802 | /* | |
d0d856e8 RD |
803 | * ext4_ext_insert_index: |
804 | * insert new index [@logical;@ptr] into the block at @curp; | |
805 | * check where to insert: before @curp or after @curp | |
a86c6181 | 806 | */ |
1f109d5a TT |
807 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
808 | struct ext4_ext_path *curp, | |
809 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
810 | { |
811 | struct ext4_extent_idx *ix; | |
812 | int len, err; | |
813 | ||
7e028976 AM |
814 | err = ext4_ext_get_access(handle, inode, curp); |
815 | if (err) | |
a86c6181 AT |
816 | return err; |
817 | ||
273df556 FM |
818 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
819 | EXT4_ERROR_INODE(inode, | |
820 | "logical %d == ei_block %d!", | |
821 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
822 | return -EIO; | |
823 | } | |
d4620315 RD |
824 | |
825 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
826 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
827 | EXT4_ERROR_INODE(inode, | |
828 | "eh_entries %d >= eh_max %d!", | |
829 | le16_to_cpu(curp->p_hdr->eh_entries), | |
830 | le16_to_cpu(curp->p_hdr->eh_max)); | |
831 | return -EIO; | |
832 | } | |
833 | ||
a86c6181 AT |
834 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
835 | /* insert after */ | |
80e675f9 | 836 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
837 | ix = curp->p_idx + 1; |
838 | } else { | |
839 | /* insert before */ | |
80e675f9 | 840 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
841 | ix = curp->p_idx; |
842 | } | |
843 | ||
80e675f9 EG |
844 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
845 | BUG_ON(len < 0); | |
846 | if (len > 0) { | |
847 | ext_debug("insert new index %d: " | |
848 | "move %d indices from 0x%p to 0x%p\n", | |
849 | logical, len, ix, ix + 1); | |
850 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
851 | } | |
852 | ||
f472e026 TM |
853 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
854 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
855 | return -EIO; | |
856 | } | |
857 | ||
a86c6181 | 858 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 859 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 860 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 861 | |
273df556 FM |
862 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
863 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
864 | return -EIO; | |
865 | } | |
a86c6181 AT |
866 | |
867 | err = ext4_ext_dirty(handle, inode, curp); | |
868 | ext4_std_error(inode->i_sb, err); | |
869 | ||
870 | return err; | |
871 | } | |
872 | ||
873 | /* | |
d0d856e8 RD |
874 | * ext4_ext_split: |
875 | * inserts new subtree into the path, using free index entry | |
876 | * at depth @at: | |
877 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
878 | * - makes decision where to split | |
879 | * - moves remaining extents and index entries (right to the split point) | |
880 | * into the newly allocated blocks | |
881 | * - initializes subtree | |
a86c6181 AT |
882 | */ |
883 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
884 | unsigned int flags, |
885 | struct ext4_ext_path *path, | |
886 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
887 | { |
888 | struct buffer_head *bh = NULL; | |
889 | int depth = ext_depth(inode); | |
890 | struct ext4_extent_header *neh; | |
891 | struct ext4_extent_idx *fidx; | |
a86c6181 | 892 | int i = at, k, m, a; |
f65e6fba | 893 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 894 | __le32 border; |
f65e6fba | 895 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
896 | int err = 0; |
897 | ||
898 | /* make decision: where to split? */ | |
d0d856e8 | 899 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 900 | |
d0d856e8 | 901 | /* if current leaf will be split, then we should use |
a86c6181 | 902 | * border from split point */ |
273df556 FM |
903 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
904 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
905 | return -EIO; | |
906 | } | |
a86c6181 AT |
907 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
908 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 909 | ext_debug("leaf will be split." |
a86c6181 | 910 | " next leaf starts at %d\n", |
8c55e204 | 911 | le32_to_cpu(border)); |
a86c6181 AT |
912 | } else { |
913 | border = newext->ee_block; | |
914 | ext_debug("leaf will be added." | |
915 | " next leaf starts at %d\n", | |
8c55e204 | 916 | le32_to_cpu(border)); |
a86c6181 AT |
917 | } |
918 | ||
919 | /* | |
d0d856e8 RD |
920 | * If error occurs, then we break processing |
921 | * and mark filesystem read-only. index won't | |
a86c6181 | 922 | * be inserted and tree will be in consistent |
d0d856e8 | 923 | * state. Next mount will repair buffers too. |
a86c6181 AT |
924 | */ |
925 | ||
926 | /* | |
d0d856e8 RD |
927 | * Get array to track all allocated blocks. |
928 | * We need this to handle errors and free blocks | |
929 | * upon them. | |
a86c6181 | 930 | */ |
5d4958f9 | 931 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
932 | if (!ablocks) |
933 | return -ENOMEM; | |
a86c6181 AT |
934 | |
935 | /* allocate all needed blocks */ | |
936 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
937 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 938 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 939 | newext, &err, flags); |
a86c6181 AT |
940 | if (newblock == 0) |
941 | goto cleanup; | |
942 | ablocks[a] = newblock; | |
943 | } | |
944 | ||
945 | /* initialize new leaf */ | |
946 | newblock = ablocks[--a]; | |
273df556 FM |
947 | if (unlikely(newblock == 0)) { |
948 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
949 | err = -EIO; | |
950 | goto cleanup; | |
951 | } | |
a86c6181 AT |
952 | bh = sb_getblk(inode->i_sb, newblock); |
953 | if (!bh) { | |
954 | err = -EIO; | |
955 | goto cleanup; | |
956 | } | |
957 | lock_buffer(bh); | |
958 | ||
7e028976 AM |
959 | err = ext4_journal_get_create_access(handle, bh); |
960 | if (err) | |
a86c6181 AT |
961 | goto cleanup; |
962 | ||
963 | neh = ext_block_hdr(bh); | |
964 | neh->eh_entries = 0; | |
55ad63bf | 965 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
966 | neh->eh_magic = EXT4_EXT_MAGIC; |
967 | neh->eh_depth = 0; | |
a86c6181 | 968 | |
d0d856e8 | 969 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
970 | if (unlikely(path[depth].p_hdr->eh_entries != |
971 | path[depth].p_hdr->eh_max)) { | |
972 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
973 | path[depth].p_hdr->eh_entries, | |
974 | path[depth].p_hdr->eh_max); | |
975 | err = -EIO; | |
976 | goto cleanup; | |
977 | } | |
a86c6181 | 978 | /* start copy from next extent */ |
1b16da77 YY |
979 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
980 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 981 | if (m) { |
1b16da77 YY |
982 | struct ext4_extent *ex; |
983 | ex = EXT_FIRST_EXTENT(neh); | |
984 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 985 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
986 | } |
987 | ||
7ac5990d | 988 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
989 | set_buffer_uptodate(bh); |
990 | unlock_buffer(bh); | |
991 | ||
0390131b | 992 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 993 | if (err) |
a86c6181 AT |
994 | goto cleanup; |
995 | brelse(bh); | |
996 | bh = NULL; | |
997 | ||
998 | /* correct old leaf */ | |
999 | if (m) { | |
7e028976 AM |
1000 | err = ext4_ext_get_access(handle, inode, path + depth); |
1001 | if (err) | |
a86c6181 | 1002 | goto cleanup; |
e8546d06 | 1003 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1004 | err = ext4_ext_dirty(handle, inode, path + depth); |
1005 | if (err) | |
a86c6181 AT |
1006 | goto cleanup; |
1007 | ||
1008 | } | |
1009 | ||
1010 | /* create intermediate indexes */ | |
1011 | k = depth - at - 1; | |
273df556 FM |
1012 | if (unlikely(k < 0)) { |
1013 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1014 | err = -EIO; | |
1015 | goto cleanup; | |
1016 | } | |
a86c6181 AT |
1017 | if (k) |
1018 | ext_debug("create %d intermediate indices\n", k); | |
1019 | /* insert new index into current index block */ | |
1020 | /* current depth stored in i var */ | |
1021 | i = depth - 1; | |
1022 | while (k--) { | |
1023 | oldblock = newblock; | |
1024 | newblock = ablocks[--a]; | |
bba90743 | 1025 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
1026 | if (!bh) { |
1027 | err = -EIO; | |
1028 | goto cleanup; | |
1029 | } | |
1030 | lock_buffer(bh); | |
1031 | ||
7e028976 AM |
1032 | err = ext4_journal_get_create_access(handle, bh); |
1033 | if (err) | |
a86c6181 AT |
1034 | goto cleanup; |
1035 | ||
1036 | neh = ext_block_hdr(bh); | |
1037 | neh->eh_entries = cpu_to_le16(1); | |
1038 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1039 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1040 | neh->eh_depth = cpu_to_le16(depth - i); |
1041 | fidx = EXT_FIRST_INDEX(neh); | |
1042 | fidx->ei_block = border; | |
f65e6fba | 1043 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1044 | |
bba90743 ES |
1045 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1046 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1047 | |
1b16da77 | 1048 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1049 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1050 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1051 | EXT4_ERROR_INODE(inode, | |
1052 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1053 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1054 | err = -EIO; | |
1055 | goto cleanup; | |
1056 | } | |
1b16da77 YY |
1057 | /* start copy indexes */ |
1058 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1059 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1060 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1061 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1062 | if (m) { |
1b16da77 | 1063 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1064 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1065 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1066 | } |
7ac5990d | 1067 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1068 | set_buffer_uptodate(bh); |
1069 | unlock_buffer(bh); | |
1070 | ||
0390131b | 1071 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1072 | if (err) |
a86c6181 AT |
1073 | goto cleanup; |
1074 | brelse(bh); | |
1075 | bh = NULL; | |
1076 | ||
1077 | /* correct old index */ | |
1078 | if (m) { | |
1079 | err = ext4_ext_get_access(handle, inode, path + i); | |
1080 | if (err) | |
1081 | goto cleanup; | |
e8546d06 | 1082 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1083 | err = ext4_ext_dirty(handle, inode, path + i); |
1084 | if (err) | |
1085 | goto cleanup; | |
1086 | } | |
1087 | ||
1088 | i--; | |
1089 | } | |
1090 | ||
1091 | /* insert new index */ | |
a86c6181 AT |
1092 | err = ext4_ext_insert_index(handle, inode, path + at, |
1093 | le32_to_cpu(border), newblock); | |
1094 | ||
1095 | cleanup: | |
1096 | if (bh) { | |
1097 | if (buffer_locked(bh)) | |
1098 | unlock_buffer(bh); | |
1099 | brelse(bh); | |
1100 | } | |
1101 | ||
1102 | if (err) { | |
1103 | /* free all allocated blocks in error case */ | |
1104 | for (i = 0; i < depth; i++) { | |
1105 | if (!ablocks[i]) | |
1106 | continue; | |
7dc57615 | 1107 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1108 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1109 | } |
1110 | } | |
1111 | kfree(ablocks); | |
1112 | ||
1113 | return err; | |
1114 | } | |
1115 | ||
1116 | /* | |
d0d856e8 RD |
1117 | * ext4_ext_grow_indepth: |
1118 | * implements tree growing procedure: | |
1119 | * - allocates new block | |
1120 | * - moves top-level data (index block or leaf) into the new block | |
1121 | * - initializes new top-level, creating index that points to the | |
1122 | * just created block | |
a86c6181 AT |
1123 | */ |
1124 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1125 | unsigned int flags, |
55f020db | 1126 | struct ext4_extent *newext) |
a86c6181 | 1127 | { |
a86c6181 | 1128 | struct ext4_extent_header *neh; |
a86c6181 | 1129 | struct buffer_head *bh; |
f65e6fba | 1130 | ext4_fsblk_t newblock; |
a86c6181 AT |
1131 | int err = 0; |
1132 | ||
1939dd84 | 1133 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1134 | newext, &err, flags); |
a86c6181 AT |
1135 | if (newblock == 0) |
1136 | return err; | |
1137 | ||
1138 | bh = sb_getblk(inode->i_sb, newblock); | |
1139 | if (!bh) { | |
1140 | err = -EIO; | |
1141 | ext4_std_error(inode->i_sb, err); | |
1142 | return err; | |
1143 | } | |
1144 | lock_buffer(bh); | |
1145 | ||
7e028976 AM |
1146 | err = ext4_journal_get_create_access(handle, bh); |
1147 | if (err) { | |
a86c6181 AT |
1148 | unlock_buffer(bh); |
1149 | goto out; | |
1150 | } | |
1151 | ||
1152 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1153 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1154 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1155 | |
1156 | /* set size of new block */ | |
1157 | neh = ext_block_hdr(bh); | |
1158 | /* old root could have indexes or leaves | |
1159 | * so calculate e_max right way */ | |
1160 | if (ext_depth(inode)) | |
55ad63bf | 1161 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1162 | else |
55ad63bf | 1163 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1164 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1165 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1166 | set_buffer_uptodate(bh); |
1167 | unlock_buffer(bh); | |
1168 | ||
0390131b | 1169 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1170 | if (err) |
a86c6181 AT |
1171 | goto out; |
1172 | ||
1939dd84 | 1173 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1174 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1175 | neh->eh_entries = cpu_to_le16(1); |
1176 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1177 | if (neh->eh_depth == 0) { | |
1178 | /* Root extent block becomes index block */ | |
1179 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1180 | EXT_FIRST_INDEX(neh)->ei_block = | |
1181 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1182 | } | |
2ae02107 | 1183 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1184 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1185 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1186 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1187 | |
ba39ebb6 | 1188 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1189 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1190 | out: |
1191 | brelse(bh); | |
1192 | ||
1193 | return err; | |
1194 | } | |
1195 | ||
1196 | /* | |
d0d856e8 RD |
1197 | * ext4_ext_create_new_leaf: |
1198 | * finds empty index and adds new leaf. | |
1199 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1200 | */ |
1201 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1202 | unsigned int flags, |
1203 | struct ext4_ext_path *path, | |
1204 | struct ext4_extent *newext) | |
a86c6181 AT |
1205 | { |
1206 | struct ext4_ext_path *curp; | |
1207 | int depth, i, err = 0; | |
1208 | ||
1209 | repeat: | |
1210 | i = depth = ext_depth(inode); | |
1211 | ||
1212 | /* walk up to the tree and look for free index entry */ | |
1213 | curp = path + depth; | |
1214 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1215 | i--; | |
1216 | curp--; | |
1217 | } | |
1218 | ||
d0d856e8 RD |
1219 | /* we use already allocated block for index block, |
1220 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1221 | if (EXT_HAS_FREE_INDEX(curp)) { |
1222 | /* if we found index with free entry, then use that | |
1223 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1224 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1225 | if (err) |
1226 | goto out; | |
a86c6181 AT |
1227 | |
1228 | /* refill path */ | |
1229 | ext4_ext_drop_refs(path); | |
1230 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1231 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1232 | path); | |
a86c6181 AT |
1233 | if (IS_ERR(path)) |
1234 | err = PTR_ERR(path); | |
1235 | } else { | |
1236 | /* tree is full, time to grow in depth */ | |
1939dd84 | 1237 | err = ext4_ext_grow_indepth(handle, inode, flags, newext); |
a86c6181 AT |
1238 | if (err) |
1239 | goto out; | |
1240 | ||
1241 | /* refill path */ | |
1242 | ext4_ext_drop_refs(path); | |
1243 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1244 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1245 | path); | |
a86c6181 AT |
1246 | if (IS_ERR(path)) { |
1247 | err = PTR_ERR(path); | |
1248 | goto out; | |
1249 | } | |
1250 | ||
1251 | /* | |
d0d856e8 RD |
1252 | * only first (depth 0 -> 1) produces free space; |
1253 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1254 | */ |
1255 | depth = ext_depth(inode); | |
1256 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1257 | /* now we need to split */ |
a86c6181 AT |
1258 | goto repeat; |
1259 | } | |
1260 | } | |
1261 | ||
1262 | out: | |
1263 | return err; | |
1264 | } | |
1265 | ||
1988b51e AT |
1266 | /* |
1267 | * search the closest allocated block to the left for *logical | |
1268 | * and returns it at @logical + it's physical address at @phys | |
1269 | * if *logical is the smallest allocated block, the function | |
1270 | * returns 0 at @phys | |
1271 | * return value contains 0 (success) or error code | |
1272 | */ | |
1f109d5a TT |
1273 | static int ext4_ext_search_left(struct inode *inode, |
1274 | struct ext4_ext_path *path, | |
1275 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1276 | { |
1277 | struct ext4_extent_idx *ix; | |
1278 | struct ext4_extent *ex; | |
b939e376 | 1279 | int depth, ee_len; |
1988b51e | 1280 | |
273df556 FM |
1281 | if (unlikely(path == NULL)) { |
1282 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1283 | return -EIO; | |
1284 | } | |
1988b51e AT |
1285 | depth = path->p_depth; |
1286 | *phys = 0; | |
1287 | ||
1288 | if (depth == 0 && path->p_ext == NULL) | |
1289 | return 0; | |
1290 | ||
1291 | /* usually extent in the path covers blocks smaller | |
1292 | * then *logical, but it can be that extent is the | |
1293 | * first one in the file */ | |
1294 | ||
1295 | ex = path[depth].p_ext; | |
b939e376 | 1296 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1297 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1298 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1299 | EXT4_ERROR_INODE(inode, | |
1300 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1301 | *logical, le32_to_cpu(ex->ee_block)); | |
1302 | return -EIO; | |
1303 | } | |
1988b51e AT |
1304 | while (--depth >= 0) { |
1305 | ix = path[depth].p_idx; | |
273df556 FM |
1306 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1307 | EXT4_ERROR_INODE(inode, | |
1308 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1309 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1310 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1311 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1312 | depth); |
1313 | return -EIO; | |
1314 | } | |
1988b51e AT |
1315 | } |
1316 | return 0; | |
1317 | } | |
1318 | ||
273df556 FM |
1319 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1320 | EXT4_ERROR_INODE(inode, | |
1321 | "logical %d < ee_block %d + ee_len %d!", | |
1322 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1323 | return -EIO; | |
1324 | } | |
1988b51e | 1325 | |
b939e376 | 1326 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1327 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1328 | return 0; |
1329 | } | |
1330 | ||
1331 | /* | |
1332 | * search the closest allocated block to the right for *logical | |
1333 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1334 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1335 | * returns 0 at @phys |
1336 | * return value contains 0 (success) or error code | |
1337 | */ | |
1f109d5a TT |
1338 | static int ext4_ext_search_right(struct inode *inode, |
1339 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1340 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1341 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1342 | { |
1343 | struct buffer_head *bh = NULL; | |
1344 | struct ext4_extent_header *eh; | |
1345 | struct ext4_extent_idx *ix; | |
1346 | struct ext4_extent *ex; | |
1347 | ext4_fsblk_t block; | |
395a87bf ES |
1348 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1349 | int ee_len; | |
1988b51e | 1350 | |
273df556 FM |
1351 | if (unlikely(path == NULL)) { |
1352 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1353 | return -EIO; | |
1354 | } | |
1988b51e AT |
1355 | depth = path->p_depth; |
1356 | *phys = 0; | |
1357 | ||
1358 | if (depth == 0 && path->p_ext == NULL) | |
1359 | return 0; | |
1360 | ||
1361 | /* usually extent in the path covers blocks smaller | |
1362 | * then *logical, but it can be that extent is the | |
1363 | * first one in the file */ | |
1364 | ||
1365 | ex = path[depth].p_ext; | |
b939e376 | 1366 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1367 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1368 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1369 | EXT4_ERROR_INODE(inode, | |
1370 | "first_extent(path[%d].p_hdr) != ex", | |
1371 | depth); | |
1372 | return -EIO; | |
1373 | } | |
1988b51e AT |
1374 | while (--depth >= 0) { |
1375 | ix = path[depth].p_idx; | |
273df556 FM |
1376 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1377 | EXT4_ERROR_INODE(inode, | |
1378 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1379 | *logical); | |
1380 | return -EIO; | |
1381 | } | |
1988b51e | 1382 | } |
4d33b1ef | 1383 | goto found_extent; |
1988b51e AT |
1384 | } |
1385 | ||
273df556 FM |
1386 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1387 | EXT4_ERROR_INODE(inode, | |
1388 | "logical %d < ee_block %d + ee_len %d!", | |
1389 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1390 | return -EIO; | |
1391 | } | |
1988b51e AT |
1392 | |
1393 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1394 | /* next allocated block in this leaf */ | |
1395 | ex++; | |
4d33b1ef | 1396 | goto found_extent; |
1988b51e AT |
1397 | } |
1398 | ||
1399 | /* go up and search for index to the right */ | |
1400 | while (--depth >= 0) { | |
1401 | ix = path[depth].p_idx; | |
1402 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1403 | goto got_index; |
1988b51e AT |
1404 | } |
1405 | ||
25f1ee3a WF |
1406 | /* we've gone up to the root and found no index to the right */ |
1407 | return 0; | |
1988b51e | 1408 | |
25f1ee3a | 1409 | got_index: |
1988b51e AT |
1410 | /* we've found index to the right, let's |
1411 | * follow it and find the closest allocated | |
1412 | * block to the right */ | |
1413 | ix++; | |
bf89d16f | 1414 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1415 | while (++depth < path->p_depth) { |
1416 | bh = sb_bread(inode->i_sb, block); | |
1417 | if (bh == NULL) | |
1418 | return -EIO; | |
1419 | eh = ext_block_hdr(bh); | |
395a87bf | 1420 | /* subtract from p_depth to get proper eh_depth */ |
f8489128 DW |
1421 | if (ext4_ext_check_block(inode, eh, |
1422 | path->p_depth - depth, bh)) { | |
1988b51e AT |
1423 | put_bh(bh); |
1424 | return -EIO; | |
1425 | } | |
1426 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1427 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1428 | put_bh(bh); |
1429 | } | |
1430 | ||
1431 | bh = sb_bread(inode->i_sb, block); | |
1432 | if (bh == NULL) | |
1433 | return -EIO; | |
1434 | eh = ext_block_hdr(bh); | |
f8489128 | 1435 | if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) { |
1988b51e AT |
1436 | put_bh(bh); |
1437 | return -EIO; | |
1438 | } | |
1439 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1440 | found_extent: |
1988b51e | 1441 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1442 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1443 | *ret_ex = ex; |
1444 | if (bh) | |
1445 | put_bh(bh); | |
1988b51e | 1446 | return 0; |
1988b51e AT |
1447 | } |
1448 | ||
a86c6181 | 1449 | /* |
d0d856e8 | 1450 | * ext4_ext_next_allocated_block: |
f17722f9 | 1451 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1452 | * NOTE: it considers block number from index entry as |
1453 | * allocated block. Thus, index entries have to be consistent | |
1454 | * with leaves. | |
a86c6181 | 1455 | */ |
725d26d3 | 1456 | static ext4_lblk_t |
a86c6181 AT |
1457 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1458 | { | |
1459 | int depth; | |
1460 | ||
1461 | BUG_ON(path == NULL); | |
1462 | depth = path->p_depth; | |
1463 | ||
1464 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1465 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1466 | |
1467 | while (depth >= 0) { | |
1468 | if (depth == path->p_depth) { | |
1469 | /* leaf */ | |
6f8ff537 CW |
1470 | if (path[depth].p_ext && |
1471 | path[depth].p_ext != | |
a86c6181 AT |
1472 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1473 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1474 | } else { | |
1475 | /* index */ | |
1476 | if (path[depth].p_idx != | |
1477 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1478 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1479 | } | |
1480 | depth--; | |
1481 | } | |
1482 | ||
f17722f9 | 1483 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1484 | } |
1485 | ||
1486 | /* | |
d0d856e8 | 1487 | * ext4_ext_next_leaf_block: |
f17722f9 | 1488 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1489 | */ |
5718789d | 1490 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1491 | { |
1492 | int depth; | |
1493 | ||
1494 | BUG_ON(path == NULL); | |
1495 | depth = path->p_depth; | |
1496 | ||
1497 | /* zero-tree has no leaf blocks at all */ | |
1498 | if (depth == 0) | |
f17722f9 | 1499 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1500 | |
1501 | /* go to index block */ | |
1502 | depth--; | |
1503 | ||
1504 | while (depth >= 0) { | |
1505 | if (path[depth].p_idx != | |
1506 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1507 | return (ext4_lblk_t) |
1508 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1509 | depth--; |
1510 | } | |
1511 | ||
f17722f9 | 1512 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1513 | } |
1514 | ||
1515 | /* | |
d0d856e8 RD |
1516 | * ext4_ext_correct_indexes: |
1517 | * if leaf gets modified and modified extent is first in the leaf, | |
1518 | * then we have to correct all indexes above. | |
a86c6181 AT |
1519 | * TODO: do we need to correct tree in all cases? |
1520 | */ | |
1d03ec98 | 1521 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1522 | struct ext4_ext_path *path) |
1523 | { | |
1524 | struct ext4_extent_header *eh; | |
1525 | int depth = ext_depth(inode); | |
1526 | struct ext4_extent *ex; | |
1527 | __le32 border; | |
1528 | int k, err = 0; | |
1529 | ||
1530 | eh = path[depth].p_hdr; | |
1531 | ex = path[depth].p_ext; | |
273df556 FM |
1532 | |
1533 | if (unlikely(ex == NULL || eh == NULL)) { | |
1534 | EXT4_ERROR_INODE(inode, | |
1535 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1536 | return -EIO; | |
1537 | } | |
a86c6181 AT |
1538 | |
1539 | if (depth == 0) { | |
1540 | /* there is no tree at all */ | |
1541 | return 0; | |
1542 | } | |
1543 | ||
1544 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1545 | /* we correct tree if first leaf got modified only */ | |
1546 | return 0; | |
1547 | } | |
1548 | ||
1549 | /* | |
d0d856e8 | 1550 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1551 | */ |
1552 | k = depth - 1; | |
1553 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1554 | err = ext4_ext_get_access(handle, inode, path + k); |
1555 | if (err) | |
a86c6181 AT |
1556 | return err; |
1557 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1558 | err = ext4_ext_dirty(handle, inode, path + k); |
1559 | if (err) | |
a86c6181 AT |
1560 | return err; |
1561 | ||
1562 | while (k--) { | |
1563 | /* change all left-side indexes */ | |
1564 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1565 | break; | |
7e028976 AM |
1566 | err = ext4_ext_get_access(handle, inode, path + k); |
1567 | if (err) | |
a86c6181 AT |
1568 | break; |
1569 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1570 | err = ext4_ext_dirty(handle, inode, path + k); |
1571 | if (err) | |
a86c6181 AT |
1572 | break; |
1573 | } | |
1574 | ||
1575 | return err; | |
1576 | } | |
1577 | ||
748de673 | 1578 | int |
a86c6181 AT |
1579 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1580 | struct ext4_extent *ex2) | |
1581 | { | |
749269fa | 1582 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1583 | |
1584 | /* | |
1585 | * Make sure that either both extents are uninitialized, or | |
1586 | * both are _not_. | |
1587 | */ | |
1588 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1589 | return 0; | |
1590 | ||
749269fa AA |
1591 | if (ext4_ext_is_uninitialized(ex1)) |
1592 | max_len = EXT_UNINIT_MAX_LEN; | |
1593 | else | |
1594 | max_len = EXT_INIT_MAX_LEN; | |
1595 | ||
a2df2a63 AA |
1596 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1597 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1598 | ||
1599 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1600 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1601 | return 0; |
1602 | ||
471d4011 SB |
1603 | /* |
1604 | * To allow future support for preallocated extents to be added | |
1605 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1606 | * this can result in the top bit of ee_len being set. |
471d4011 | 1607 | */ |
749269fa | 1608 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1609 | return 0; |
bbf2f9fb | 1610 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1611 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1612 | return 0; |
1613 | #endif | |
1614 | ||
bf89d16f | 1615 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1616 | return 1; |
1617 | return 0; | |
1618 | } | |
1619 | ||
56055d3a AA |
1620 | /* |
1621 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1622 | * It always tries to merge towards right. If you want to merge towards | |
1623 | * left, pass "ex - 1" as argument instead of "ex". | |
1624 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1625 | * 1 if they got merged. | |
1626 | */ | |
197217a5 | 1627 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1628 | struct ext4_ext_path *path, |
1629 | struct ext4_extent *ex) | |
56055d3a AA |
1630 | { |
1631 | struct ext4_extent_header *eh; | |
1632 | unsigned int depth, len; | |
1633 | int merge_done = 0; | |
1634 | int uninitialized = 0; | |
1635 | ||
1636 | depth = ext_depth(inode); | |
1637 | BUG_ON(path[depth].p_hdr == NULL); | |
1638 | eh = path[depth].p_hdr; | |
1639 | ||
1640 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1641 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1642 | break; | |
1643 | /* merge with next extent! */ | |
1644 | if (ext4_ext_is_uninitialized(ex)) | |
1645 | uninitialized = 1; | |
1646 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1647 | + ext4_ext_get_actual_len(ex + 1)); | |
1648 | if (uninitialized) | |
1649 | ext4_ext_mark_uninitialized(ex); | |
1650 | ||
1651 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1652 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1653 | * sizeof(struct ext4_extent); | |
1654 | memmove(ex + 1, ex + 2, len); | |
1655 | } | |
e8546d06 | 1656 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1657 | merge_done = 1; |
1658 | WARN_ON(eh->eh_entries == 0); | |
1659 | if (!eh->eh_entries) | |
24676da4 | 1660 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1661 | } |
1662 | ||
1663 | return merge_done; | |
1664 | } | |
1665 | ||
ecb94f5f TT |
1666 | /* |
1667 | * This function does a very simple check to see if we can collapse | |
1668 | * an extent tree with a single extent tree leaf block into the inode. | |
1669 | */ | |
1670 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1671 | struct inode *inode, | |
1672 | struct ext4_ext_path *path) | |
1673 | { | |
1674 | size_t s; | |
1675 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1676 | ext4_fsblk_t blk; | |
1677 | ||
1678 | if ((path[0].p_depth != 1) || | |
1679 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1680 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1681 | return; | |
1682 | ||
1683 | /* | |
1684 | * We need to modify the block allocation bitmap and the block | |
1685 | * group descriptor to release the extent tree block. If we | |
1686 | * can't get the journal credits, give up. | |
1687 | */ | |
1688 | if (ext4_journal_extend(handle, 2)) | |
1689 | return; | |
1690 | ||
1691 | /* | |
1692 | * Copy the extent data up to the inode | |
1693 | */ | |
1694 | blk = ext4_idx_pblock(path[0].p_idx); | |
1695 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1696 | sizeof(struct ext4_extent_idx); | |
1697 | s += sizeof(struct ext4_extent_header); | |
1698 | ||
1699 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1700 | path[0].p_depth = 0; | |
1701 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1702 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1703 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1704 | ||
1705 | brelse(path[1].p_bh); | |
1706 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
1707 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); | |
1708 | } | |
1709 | ||
197217a5 YY |
1710 | /* |
1711 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1712 | * return 1 if merge left else 0. | |
1713 | */ | |
ecb94f5f TT |
1714 | static void ext4_ext_try_to_merge(handle_t *handle, |
1715 | struct inode *inode, | |
197217a5 YY |
1716 | struct ext4_ext_path *path, |
1717 | struct ext4_extent *ex) { | |
1718 | struct ext4_extent_header *eh; | |
1719 | unsigned int depth; | |
1720 | int merge_done = 0; | |
197217a5 YY |
1721 | |
1722 | depth = ext_depth(inode); | |
1723 | BUG_ON(path[depth].p_hdr == NULL); | |
1724 | eh = path[depth].p_hdr; | |
1725 | ||
1726 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1727 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1728 | ||
1729 | if (!merge_done) | |
ecb94f5f | 1730 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1731 | |
ecb94f5f | 1732 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1733 | } |
1734 | ||
25d14f98 AA |
1735 | /* |
1736 | * check if a portion of the "newext" extent overlaps with an | |
1737 | * existing extent. | |
1738 | * | |
1739 | * If there is an overlap discovered, it updates the length of the newext | |
1740 | * such that there will be no overlap, and then returns 1. | |
1741 | * If there is no overlap found, it returns 0. | |
1742 | */ | |
4d33b1ef TT |
1743 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1744 | struct inode *inode, | |
1f109d5a TT |
1745 | struct ext4_extent *newext, |
1746 | struct ext4_ext_path *path) | |
25d14f98 | 1747 | { |
725d26d3 | 1748 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1749 | unsigned int depth, len1; |
1750 | unsigned int ret = 0; | |
1751 | ||
1752 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1753 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1754 | depth = ext_depth(inode); |
1755 | if (!path[depth].p_ext) | |
1756 | goto out; | |
1757 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1758 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1759 | |
1760 | /* | |
1761 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1762 | * is before the requested block(s) |
25d14f98 AA |
1763 | */ |
1764 | if (b2 < b1) { | |
1765 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1766 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1767 | goto out; |
4d33b1ef | 1768 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1769 | } |
1770 | ||
725d26d3 | 1771 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1772 | if (b1 + len1 < b1) { |
f17722f9 | 1773 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1774 | newext->ee_len = cpu_to_le16(len1); |
1775 | ret = 1; | |
1776 | } | |
1777 | ||
1778 | /* check for overlap */ | |
1779 | if (b1 + len1 > b2) { | |
1780 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1781 | ret = 1; | |
1782 | } | |
1783 | out: | |
1784 | return ret; | |
1785 | } | |
1786 | ||
a86c6181 | 1787 | /* |
d0d856e8 RD |
1788 | * ext4_ext_insert_extent: |
1789 | * tries to merge requsted extent into the existing extent or | |
1790 | * inserts requested extent as new one into the tree, | |
1791 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1792 | */ |
1793 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1794 | struct ext4_ext_path *path, | |
0031462b | 1795 | struct ext4_extent *newext, int flag) |
a86c6181 | 1796 | { |
af5bc92d | 1797 | struct ext4_extent_header *eh; |
a86c6181 AT |
1798 | struct ext4_extent *ex, *fex; |
1799 | struct ext4_extent *nearex; /* nearest extent */ | |
1800 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1801 | int depth, len, err; |
1802 | ext4_lblk_t next; | |
a2df2a63 | 1803 | unsigned uninitialized = 0; |
55f020db | 1804 | int flags = 0; |
a86c6181 | 1805 | |
273df556 FM |
1806 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1807 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1808 | return -EIO; | |
1809 | } | |
a86c6181 AT |
1810 | depth = ext_depth(inode); |
1811 | ex = path[depth].p_ext; | |
273df556 FM |
1812 | if (unlikely(path[depth].p_hdr == NULL)) { |
1813 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1814 | return -EIO; | |
1815 | } | |
a86c6181 AT |
1816 | |
1817 | /* try to insert block into found extent and return */ | |
744692dc | 1818 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1819 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
32de6756 | 1820 | ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n", |
bf89d16f TT |
1821 | ext4_ext_is_uninitialized(newext), |
1822 | ext4_ext_get_actual_len(newext), | |
1823 | le32_to_cpu(ex->ee_block), | |
1824 | ext4_ext_is_uninitialized(ex), | |
1825 | ext4_ext_get_actual_len(ex), | |
1826 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1827 | err = ext4_ext_get_access(handle, inode, path + depth); |
1828 | if (err) | |
a86c6181 | 1829 | return err; |
a2df2a63 AA |
1830 | |
1831 | /* | |
1832 | * ext4_can_extents_be_merged should have checked that either | |
1833 | * both extents are uninitialized, or both aren't. Thus we | |
1834 | * need to check only one of them here. | |
1835 | */ | |
1836 | if (ext4_ext_is_uninitialized(ex)) | |
1837 | uninitialized = 1; | |
1838 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1839 | + ext4_ext_get_actual_len(newext)); | |
1840 | if (uninitialized) | |
1841 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1842 | eh = path[depth].p_hdr; |
1843 | nearex = ex; | |
1844 | goto merge; | |
1845 | } | |
1846 | ||
a86c6181 AT |
1847 | depth = ext_depth(inode); |
1848 | eh = path[depth].p_hdr; | |
1849 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1850 | goto has_space; | |
1851 | ||
1852 | /* probably next leaf has space for us? */ | |
1853 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1854 | next = EXT_MAX_BLOCKS; |
1855 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1856 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1857 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1858 | ext_debug("next leaf block - %u\n", next); |
a86c6181 AT |
1859 | BUG_ON(npath != NULL); |
1860 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1861 | if (IS_ERR(npath)) | |
1862 | return PTR_ERR(npath); | |
1863 | BUG_ON(npath->p_depth != path->p_depth); | |
1864 | eh = npath[depth].p_hdr; | |
1865 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1866 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1867 | le16_to_cpu(eh->eh_entries)); |
1868 | path = npath; | |
ffb505ff | 1869 | goto has_space; |
a86c6181 AT |
1870 | } |
1871 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1872 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1873 | } | |
1874 | ||
1875 | /* | |
d0d856e8 RD |
1876 | * There is no free space in the found leaf. |
1877 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1878 | */ |
55f020db AH |
1879 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1880 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1881 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1882 | if (err) |
1883 | goto cleanup; | |
1884 | depth = ext_depth(inode); | |
1885 | eh = path[depth].p_hdr; | |
1886 | ||
1887 | has_space: | |
1888 | nearex = path[depth].p_ext; | |
1889 | ||
7e028976 AM |
1890 | err = ext4_ext_get_access(handle, inode, path + depth); |
1891 | if (err) | |
a86c6181 AT |
1892 | goto cleanup; |
1893 | ||
1894 | if (!nearex) { | |
1895 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1896 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1897 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1898 | ext4_ext_pblock(newext), |
553f9008 | 1899 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1900 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1901 | nearex = EXT_FIRST_EXTENT(eh); |
1902 | } else { | |
1903 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1904 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1905 | /* Insert after */ |
32de6756 YY |
1906 | ext_debug("insert %u:%llu:[%d]%d before: " |
1907 | "nearest %p\n", | |
80e675f9 EG |
1908 | le32_to_cpu(newext->ee_block), |
1909 | ext4_ext_pblock(newext), | |
1910 | ext4_ext_is_uninitialized(newext), | |
1911 | ext4_ext_get_actual_len(newext), | |
1912 | nearex); | |
1913 | nearex++; | |
1914 | } else { | |
1915 | /* Insert before */ | |
1916 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
1917 | ext_debug("insert %u:%llu:[%d]%d after: " |
1918 | "nearest %p\n", | |
8c55e204 | 1919 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1920 | ext4_ext_pblock(newext), |
553f9008 | 1921 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1922 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
1923 | nearex); |
1924 | } | |
1925 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
1926 | if (len > 0) { | |
32de6756 | 1927 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
1928 | "move %d extents from 0x%p to 0x%p\n", |
1929 | le32_to_cpu(newext->ee_block), | |
1930 | ext4_ext_pblock(newext), | |
1931 | ext4_ext_is_uninitialized(newext), | |
1932 | ext4_ext_get_actual_len(newext), | |
1933 | len, nearex, nearex + 1); | |
1934 | memmove(nearex + 1, nearex, | |
1935 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 1936 | } |
a86c6181 AT |
1937 | } |
1938 | ||
e8546d06 | 1939 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 1940 | path[depth].p_ext = nearex; |
a86c6181 | 1941 | nearex->ee_block = newext->ee_block; |
bf89d16f | 1942 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1943 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1944 | |
1945 | merge: | |
e7bcf823 | 1946 | /* try to merge extents */ |
744692dc | 1947 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 1948 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 1949 | |
a86c6181 AT |
1950 | |
1951 | /* time to correct all indexes above */ | |
1952 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1953 | if (err) | |
1954 | goto cleanup; | |
1955 | ||
ecb94f5f | 1956 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
1957 | |
1958 | cleanup: | |
1959 | if (npath) { | |
1960 | ext4_ext_drop_refs(npath); | |
1961 | kfree(npath); | |
1962 | } | |
a86c6181 AT |
1963 | ext4_ext_invalidate_cache(inode); |
1964 | return err; | |
1965 | } | |
1966 | ||
91dd8c11 LC |
1967 | static int ext4_fill_fiemap_extents(struct inode *inode, |
1968 | ext4_lblk_t block, ext4_lblk_t num, | |
1969 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
1970 | { |
1971 | struct ext4_ext_path *path = NULL; | |
06348679 | 1972 | struct ext4_ext_cache newex; |
6873fa0d | 1973 | struct ext4_extent *ex; |
91dd8c11 | 1974 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 1975 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
1976 | int exists, depth = 0, err = 0; |
1977 | unsigned int flags = 0; | |
1978 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 1979 | |
f17722f9 | 1980 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1981 | num = last - block; |
1982 | /* find extent for this block */ | |
fab3a549 | 1983 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
1984 | |
1985 | if (path && ext_depth(inode) != depth) { | |
1986 | /* depth was changed. we have to realloc path */ | |
1987 | kfree(path); | |
1988 | path = NULL; | |
1989 | } | |
1990 | ||
6873fa0d ES |
1991 | path = ext4_ext_find_extent(inode, block, path); |
1992 | if (IS_ERR(path)) { | |
91dd8c11 | 1993 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1994 | err = PTR_ERR(path); |
1995 | path = NULL; | |
1996 | break; | |
1997 | } | |
1998 | ||
1999 | depth = ext_depth(inode); | |
273df556 | 2000 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2001 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2002 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2003 | err = -EIO; | |
2004 | break; | |
2005 | } | |
6873fa0d ES |
2006 | ex = path[depth].p_ext; |
2007 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2008 | ext4_ext_drop_refs(path); |
6873fa0d | 2009 | |
91dd8c11 | 2010 | flags = 0; |
6873fa0d ES |
2011 | exists = 0; |
2012 | if (!ex) { | |
2013 | /* there is no extent yet, so try to allocate | |
2014 | * all requested space */ | |
2015 | start = block; | |
2016 | end = block + num; | |
2017 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2018 | /* need to allocate space before found extent */ | |
2019 | start = block; | |
2020 | end = le32_to_cpu(ex->ee_block); | |
2021 | if (block + num < end) | |
2022 | end = block + num; | |
2023 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2024 | + ext4_ext_get_actual_len(ex)) { | |
2025 | /* need to allocate space after found extent */ | |
2026 | start = block; | |
2027 | end = block + num; | |
2028 | if (end >= next) | |
2029 | end = next; | |
2030 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2031 | /* | |
2032 | * some part of requested space is covered | |
2033 | * by found extent | |
2034 | */ | |
2035 | start = block; | |
2036 | end = le32_to_cpu(ex->ee_block) | |
2037 | + ext4_ext_get_actual_len(ex); | |
2038 | if (block + num < end) | |
2039 | end = block + num; | |
2040 | exists = 1; | |
2041 | } else { | |
2042 | BUG(); | |
2043 | } | |
2044 | BUG_ON(end <= start); | |
2045 | ||
2046 | if (!exists) { | |
06348679 LC |
2047 | newex.ec_block = start; |
2048 | newex.ec_len = end - start; | |
2049 | newex.ec_start = 0; | |
6873fa0d | 2050 | } else { |
06348679 LC |
2051 | newex.ec_block = le32_to_cpu(ex->ee_block); |
2052 | newex.ec_len = ext4_ext_get_actual_len(ex); | |
2053 | newex.ec_start = ext4_ext_pblock(ex); | |
91dd8c11 LC |
2054 | if (ext4_ext_is_uninitialized(ex)) |
2055 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
6873fa0d ES |
2056 | } |
2057 | ||
91dd8c11 | 2058 | /* |
06348679 LC |
2059 | * Find delayed extent and update newex accordingly. We call |
2060 | * it even in !exists case to find out whether newex is the | |
91dd8c11 LC |
2061 | * last existing extent or not. |
2062 | */ | |
06348679 | 2063 | next_del = ext4_find_delayed_extent(inode, &newex); |
91dd8c11 LC |
2064 | if (!exists && next_del) { |
2065 | exists = 1; | |
2066 | flags |= FIEMAP_EXTENT_DELALLOC; | |
2067 | } | |
2068 | up_read(&EXT4_I(inode)->i_data_sem); | |
2069 | ||
06348679 LC |
2070 | if (unlikely(newex.ec_len == 0)) { |
2071 | EXT4_ERROR_INODE(inode, "newex.ec_len == 0"); | |
273df556 FM |
2072 | err = -EIO; |
2073 | break; | |
2074 | } | |
6873fa0d | 2075 | |
91dd8c11 LC |
2076 | /* This is possible iff next == next_del == EXT_MAX_BLOCKS */ |
2077 | if (next == next_del) { | |
2078 | flags |= FIEMAP_EXTENT_LAST; | |
2079 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2080 | next != EXT_MAX_BLOCKS)) { | |
2081 | EXT4_ERROR_INODE(inode, | |
2082 | "next extent == %u, next " | |
2083 | "delalloc extent = %u", | |
2084 | next, next_del); | |
2085 | err = -EIO; | |
2086 | break; | |
2087 | } | |
6873fa0d ES |
2088 | } |
2089 | ||
91dd8c11 LC |
2090 | if (exists) { |
2091 | err = fiemap_fill_next_extent(fieinfo, | |
06348679 LC |
2092 | (__u64)newex.ec_block << blksize_bits, |
2093 | (__u64)newex.ec_start << blksize_bits, | |
2094 | (__u64)newex.ec_len << blksize_bits, | |
91dd8c11 LC |
2095 | flags); |
2096 | if (err < 0) | |
2097 | break; | |
2098 | if (err == 1) { | |
2099 | err = 0; | |
2100 | break; | |
2101 | } | |
6873fa0d ES |
2102 | } |
2103 | ||
06348679 | 2104 | block = newex.ec_block + newex.ec_len; |
6873fa0d ES |
2105 | } |
2106 | ||
2107 | if (path) { | |
2108 | ext4_ext_drop_refs(path); | |
2109 | kfree(path); | |
2110 | } | |
2111 | ||
2112 | return err; | |
2113 | } | |
2114 | ||
09b88252 | 2115 | static void |
725d26d3 | 2116 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 2117 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
2118 | { |
2119 | struct ext4_ext_cache *cex; | |
2120 | BUG_ON(len == 0); | |
2ec0ae3a | 2121 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
d8990240 | 2122 | trace_ext4_ext_put_in_cache(inode, block, len, start); |
a86c6181 | 2123 | cex = &EXT4_I(inode)->i_cached_extent; |
a86c6181 AT |
2124 | cex->ec_block = block; |
2125 | cex->ec_len = len; | |
2126 | cex->ec_start = start; | |
2ec0ae3a | 2127 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
2128 | } |
2129 | ||
2130 | /* | |
d0d856e8 RD |
2131 | * ext4_ext_put_gap_in_cache: |
2132 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2133 | * and cache this gap |
2134 | */ | |
09b88252 | 2135 | static void |
a86c6181 | 2136 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2137 | ext4_lblk_t block) |
a86c6181 AT |
2138 | { |
2139 | int depth = ext_depth(inode); | |
725d26d3 AK |
2140 | unsigned long len; |
2141 | ext4_lblk_t lblock; | |
a86c6181 AT |
2142 | struct ext4_extent *ex; |
2143 | ||
2144 | ex = path[depth].p_ext; | |
2145 | if (ex == NULL) { | |
2146 | /* there is no extent yet, so gap is [0;-] */ | |
2147 | lblock = 0; | |
f17722f9 | 2148 | len = EXT_MAX_BLOCKS; |
a86c6181 AT |
2149 | ext_debug("cache gap(whole file):"); |
2150 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2151 | lblock = block; | |
2152 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2153 | ext_debug("cache gap(before): %u [%u:%u]", |
2154 | block, | |
2155 | le32_to_cpu(ex->ee_block), | |
2156 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 2157 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2158 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2159 | ext4_lblk_t next; |
8c55e204 | 2160 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2161 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2162 | |
2163 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2164 | ext_debug("cache gap(after): [%u:%u] %u", |
2165 | le32_to_cpu(ex->ee_block), | |
2166 | ext4_ext_get_actual_len(ex), | |
2167 | block); | |
725d26d3 AK |
2168 | BUG_ON(next == lblock); |
2169 | len = next - lblock; | |
a86c6181 AT |
2170 | } else { |
2171 | lblock = len = 0; | |
2172 | BUG(); | |
2173 | } | |
2174 | ||
bba90743 | 2175 | ext_debug(" -> %u:%lu\n", lblock, len); |
b05e6ae5 | 2176 | ext4_ext_put_in_cache(inode, lblock, len, 0); |
a86c6181 AT |
2177 | } |
2178 | ||
b05e6ae5 | 2179 | /* |
63fedaf1 | 2180 | * ext4_ext_in_cache() |
a4bb6b64 AH |
2181 | * Checks to see if the given block is in the cache. |
2182 | * If it is, the cached extent is stored in the given | |
63fedaf1 | 2183 | * cache extent pointer. |
a4bb6b64 AH |
2184 | * |
2185 | * @inode: The files inode | |
2186 | * @block: The block to look for in the cache | |
2187 | * @ex: Pointer where the cached extent will be stored | |
2188 | * if it contains block | |
2189 | * | |
b05e6ae5 TT |
2190 | * Return 0 if cache is invalid; 1 if the cache is valid |
2191 | */ | |
63fedaf1 LC |
2192 | static int |
2193 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, | |
2194 | struct ext4_extent *ex) | |
2195 | { | |
a86c6181 | 2196 | struct ext4_ext_cache *cex; |
77f4135f | 2197 | struct ext4_sb_info *sbi; |
b05e6ae5 | 2198 | int ret = 0; |
a86c6181 | 2199 | |
60e6679e | 2200 | /* |
2ec0ae3a TT |
2201 | * We borrow i_block_reservation_lock to protect i_cached_extent |
2202 | */ | |
2203 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
a86c6181 | 2204 | cex = &EXT4_I(inode)->i_cached_extent; |
77f4135f | 2205 | sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2206 | |
2207 | /* has cache valid data? */ | |
b05e6ae5 | 2208 | if (cex->ec_len == 0) |
2ec0ae3a | 2209 | goto errout; |
a86c6181 | 2210 | |
731eb1a0 | 2211 | if (in_range(block, cex->ec_block, cex->ec_len)) { |
63fedaf1 LC |
2212 | ex->ee_block = cpu_to_le32(cex->ec_block); |
2213 | ext4_ext_store_pblock(ex, cex->ec_start); | |
2214 | ex->ee_len = cpu_to_le16(cex->ec_len); | |
bba90743 ES |
2215 | ext_debug("%u cached by %u:%u:%llu\n", |
2216 | block, | |
2217 | cex->ec_block, cex->ec_len, cex->ec_start); | |
b05e6ae5 | 2218 | ret = 1; |
a86c6181 | 2219 | } |
2ec0ae3a | 2220 | errout: |
d8990240 | 2221 | trace_ext4_ext_in_cache(inode, block, ret); |
2ec0ae3a TT |
2222 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2223 | return ret; | |
a86c6181 AT |
2224 | } |
2225 | ||
2226 | /* | |
d0d856e8 RD |
2227 | * ext4_ext_rm_idx: |
2228 | * removes index from the index block. | |
a86c6181 | 2229 | */ |
1d03ec98 | 2230 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
2231 | struct ext4_ext_path *path) |
2232 | { | |
a86c6181 | 2233 | int err; |
f65e6fba | 2234 | ext4_fsblk_t leaf; |
a86c6181 AT |
2235 | |
2236 | /* free index block */ | |
2237 | path--; | |
bf89d16f | 2238 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2239 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2240 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2241 | return -EIO; | |
2242 | } | |
7e028976 AM |
2243 | err = ext4_ext_get_access(handle, inode, path); |
2244 | if (err) | |
a86c6181 | 2245 | return err; |
0e1147b0 RD |
2246 | |
2247 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2248 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2249 | len *= sizeof(struct ext4_extent_idx); | |
2250 | memmove(path->p_idx, path->p_idx + 1, len); | |
2251 | } | |
2252 | ||
e8546d06 | 2253 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2254 | err = ext4_ext_dirty(handle, inode, path); |
2255 | if (err) | |
a86c6181 | 2256 | return err; |
2ae02107 | 2257 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2258 | trace_ext4_ext_rm_idx(inode, leaf); |
2259 | ||
7dc57615 | 2260 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2261 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
a86c6181 AT |
2262 | return err; |
2263 | } | |
2264 | ||
2265 | /* | |
ee12b630 MC |
2266 | * ext4_ext_calc_credits_for_single_extent: |
2267 | * This routine returns max. credits that needed to insert an extent | |
2268 | * to the extent tree. | |
2269 | * When pass the actual path, the caller should calculate credits | |
2270 | * under i_data_sem. | |
a86c6181 | 2271 | */ |
525f4ed8 | 2272 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2273 | struct ext4_ext_path *path) |
2274 | { | |
a86c6181 | 2275 | if (path) { |
ee12b630 | 2276 | int depth = ext_depth(inode); |
f3bd1f3f | 2277 | int ret = 0; |
ee12b630 | 2278 | |
a86c6181 | 2279 | /* probably there is space in leaf? */ |
a86c6181 | 2280 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2281 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2282 | |
ee12b630 MC |
2283 | /* |
2284 | * There are some space in the leaf tree, no | |
2285 | * need to account for leaf block credit | |
2286 | * | |
2287 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2288 | * and other metadata blocks still need to be |
ee12b630 MC |
2289 | * accounted. |
2290 | */ | |
525f4ed8 | 2291 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2292 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2293 | return ret; |
ee12b630 MC |
2294 | } |
2295 | } | |
a86c6181 | 2296 | |
525f4ed8 | 2297 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2298 | } |
a86c6181 | 2299 | |
ee12b630 MC |
2300 | /* |
2301 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2302 | * | |
2303 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2304 | * in the worse case, each tree level index/leaf need to be changed | |
2305 | * if the tree split due to insert a new extent, then the old tree | |
2306 | * index/leaf need to be updated too | |
2307 | * | |
2308 | * If the nrblocks are discontiguous, they could cause | |
2309 | * the whole tree split more than once, but this is really rare. | |
2310 | */ | |
525f4ed8 | 2311 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2312 | { |
2313 | int index; | |
f19d5870 TM |
2314 | int depth; |
2315 | ||
2316 | /* If we are converting the inline data, only one is needed here. */ | |
2317 | if (ext4_has_inline_data(inode)) | |
2318 | return 1; | |
2319 | ||
2320 | depth = ext_depth(inode); | |
a86c6181 | 2321 | |
ee12b630 MC |
2322 | if (chunk) |
2323 | index = depth * 2; | |
2324 | else | |
2325 | index = depth * 3; | |
a86c6181 | 2326 | |
ee12b630 | 2327 | return index; |
a86c6181 AT |
2328 | } |
2329 | ||
2330 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2331 | struct ext4_extent *ex, |
2332 | ext4_fsblk_t *partial_cluster, | |
2333 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2334 | { |
0aa06000 | 2335 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2336 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2337 | ext4_fsblk_t pblk; |
18888cf0 | 2338 | int flags = 0; |
a86c6181 | 2339 | |
c9de560d | 2340 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
18888cf0 AS |
2341 | flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2342 | else if (ext4_should_journal_data(inode)) | |
2343 | flags |= EXT4_FREE_BLOCKS_FORGET; | |
2344 | ||
0aa06000 TT |
2345 | /* |
2346 | * For bigalloc file systems, we never free a partial cluster | |
2347 | * at the beginning of the extent. Instead, we make a note | |
2348 | * that we tried freeing the cluster, and check to see if we | |
2349 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2350 | * or at the end of the ext4_truncate() operation. | |
2351 | */ | |
2352 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2353 | ||
d8990240 | 2354 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2355 | /* |
2356 | * If we have a partial cluster, and it's different from the | |
2357 | * cluster of the last block, we need to explicitly free the | |
2358 | * partial cluster here. | |
2359 | */ | |
2360 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2361 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2362 | ext4_free_blocks(handle, inode, NULL, | |
2363 | EXT4_C2B(sbi, *partial_cluster), | |
2364 | sbi->s_cluster_ratio, flags); | |
2365 | *partial_cluster = 0; | |
2366 | } | |
2367 | ||
a86c6181 AT |
2368 | #ifdef EXTENTS_STATS |
2369 | { | |
2370 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2371 | spin_lock(&sbi->s_ext_stats_lock); |
2372 | sbi->s_ext_blocks += ee_len; | |
2373 | sbi->s_ext_extents++; | |
2374 | if (ee_len < sbi->s_ext_min) | |
2375 | sbi->s_ext_min = ee_len; | |
2376 | if (ee_len > sbi->s_ext_max) | |
2377 | sbi->s_ext_max = ee_len; | |
2378 | if (ext_depth(inode) > sbi->s_depth_max) | |
2379 | sbi->s_depth_max = ext_depth(inode); | |
2380 | spin_unlock(&sbi->s_ext_stats_lock); | |
2381 | } | |
2382 | #endif | |
2383 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2384 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2385 | /* tail removal */ |
725d26d3 | 2386 | ext4_lblk_t num; |
725d26d3 | 2387 | |
a2df2a63 | 2388 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2389 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2390 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2391 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2392 | /* | |
2393 | * If the block range to be freed didn't start at the | |
2394 | * beginning of a cluster, and we removed the entire | |
2395 | * extent, save the partial cluster here, since we | |
2396 | * might need to delete if we determine that the | |
2397 | * truncate operation has removed all of the blocks in | |
2398 | * the cluster. | |
2399 | */ | |
2400 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2401 | (ee_len == num)) | |
2402 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2403 | else | |
2404 | *partial_cluster = 0; | |
a86c6181 | 2405 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2406 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2407 | /* head removal */ |
2408 | ext4_lblk_t num; | |
2409 | ext4_fsblk_t start; | |
2410 | ||
2411 | num = to - from; | |
2412 | start = ext4_ext_pblock(ex); | |
2413 | ||
2414 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
ee90d57e | 2415 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
d583fb87 | 2416 | |
a86c6181 | 2417 | } else { |
725d26d3 AK |
2418 | printk(KERN_INFO "strange request: removal(2) " |
2419 | "%u-%u from %u:%u\n", | |
2420 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2421 | } |
2422 | return 0; | |
2423 | } | |
2424 | ||
d583fb87 AH |
2425 | |
2426 | /* | |
2427 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2428 | * blocks appearing between "start" and "end", and splits the extents | |
2429 | * if "start" and "end" appear in the same extent | |
2430 | * | |
2431 | * @handle: The journal handle | |
2432 | * @inode: The files inode | |
2433 | * @path: The path to the leaf | |
2434 | * @start: The first block to remove | |
2435 | * @end: The last block to remove | |
2436 | */ | |
a86c6181 AT |
2437 | static int |
2438 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2439 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2440 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2441 | { |
0aa06000 | 2442 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2443 | int err = 0, correct_index = 0; |
2444 | int depth = ext_depth(inode), credits; | |
2445 | struct ext4_extent_header *eh; | |
750c9c47 | 2446 | ext4_lblk_t a, b; |
725d26d3 AK |
2447 | unsigned num; |
2448 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2449 | unsigned short ex_ee_len; |
a2df2a63 | 2450 | unsigned uninitialized = 0; |
a86c6181 AT |
2451 | struct ext4_extent *ex; |
2452 | ||
c29c0ae7 | 2453 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2454 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2455 | if (!path[depth].p_hdr) |
2456 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2457 | eh = path[depth].p_hdr; | |
273df556 FM |
2458 | if (unlikely(path[depth].p_hdr == NULL)) { |
2459 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2460 | return -EIO; | |
2461 | } | |
a86c6181 AT |
2462 | /* find where to start removing */ |
2463 | ex = EXT_LAST_EXTENT(eh); | |
2464 | ||
2465 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2466 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2467 | |
d8990240 AK |
2468 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2469 | ||
a86c6181 AT |
2470 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2471 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2472 | |
2473 | if (ext4_ext_is_uninitialized(ex)) | |
2474 | uninitialized = 1; | |
2475 | else | |
2476 | uninitialized = 0; | |
2477 | ||
553f9008 M |
2478 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2479 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2480 | path[depth].p_ext = ex; |
2481 | ||
2482 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2483 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2484 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2485 | |
2486 | ext_debug(" border %u:%u\n", a, b); | |
2487 | ||
d583fb87 | 2488 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2489 | if (end < ex_ee_block) { |
d583fb87 AH |
2490 | ex--; |
2491 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2492 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2493 | continue; | |
750c9c47 | 2494 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2495 | EXT4_ERROR_INODE(inode, |
2496 | "can not handle truncate %u:%u " | |
2497 | "on extent %u:%u", | |
2498 | start, end, ex_ee_block, | |
2499 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2500 | err = -EIO; |
2501 | goto out; | |
a86c6181 AT |
2502 | } else if (a != ex_ee_block) { |
2503 | /* remove tail of the extent */ | |
750c9c47 | 2504 | num = a - ex_ee_block; |
a86c6181 AT |
2505 | } else { |
2506 | /* remove whole extent: excellent! */ | |
a86c6181 | 2507 | num = 0; |
a86c6181 | 2508 | } |
34071da7 TT |
2509 | /* |
2510 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2511 | * descriptor) for each block group; assume two block | |
2512 | * groups plus ex_ee_len/blocks_per_block_group for | |
2513 | * the worst case | |
2514 | */ | |
2515 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2516 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2517 | correct_index = 1; | |
2518 | credits += (ext_depth(inode)) + 1; | |
2519 | } | |
5aca07eb | 2520 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2521 | |
487caeef | 2522 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2523 | if (err) |
a86c6181 | 2524 | goto out; |
a86c6181 AT |
2525 | |
2526 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2527 | if (err) | |
2528 | goto out; | |
2529 | ||
0aa06000 TT |
2530 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2531 | a, b); | |
a86c6181 AT |
2532 | if (err) |
2533 | goto out; | |
2534 | ||
750c9c47 | 2535 | if (num == 0) |
d0d856e8 | 2536 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2537 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2538 | |
a86c6181 | 2539 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2540 | /* |
2541 | * Do not mark uninitialized if all the blocks in the | |
2542 | * extent have been removed. | |
2543 | */ | |
2544 | if (uninitialized && num) | |
a2df2a63 | 2545 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2546 | /* |
2547 | * If the extent was completely released, | |
2548 | * we need to remove it from the leaf | |
2549 | */ | |
2550 | if (num == 0) { | |
f17722f9 | 2551 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2552 | /* |
2553 | * For hole punching, we need to scoot all the | |
2554 | * extents up when an extent is removed so that | |
2555 | * we dont have blank extents in the middle | |
2556 | */ | |
2557 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2558 | sizeof(struct ext4_extent)); | |
2559 | ||
2560 | /* Now get rid of the one at the end */ | |
2561 | memset(EXT_LAST_EXTENT(eh), 0, | |
2562 | sizeof(struct ext4_extent)); | |
2563 | } | |
2564 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2565 | } else |
2566 | *partial_cluster = 0; | |
d583fb87 | 2567 | |
750c9c47 DM |
2568 | err = ext4_ext_dirty(handle, inode, path + depth); |
2569 | if (err) | |
2570 | goto out; | |
2571 | ||
bf52c6f7 | 2572 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2573 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2574 | ex--; |
2575 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2576 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2577 | } |
2578 | ||
2579 | if (correct_index && eh->eh_entries) | |
2580 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2581 | ||
0aa06000 TT |
2582 | /* |
2583 | * If there is still a entry in the leaf node, check to see if | |
2584 | * it references the partial cluster. This is the only place | |
2585 | * where it could; if it doesn't, we can free the cluster. | |
2586 | */ | |
2587 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2588 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2589 | *partial_cluster)) { | |
2590 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2591 | ||
2592 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2593 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2594 | ||
2595 | ext4_free_blocks(handle, inode, NULL, | |
2596 | EXT4_C2B(sbi, *partial_cluster), | |
2597 | sbi->s_cluster_ratio, flags); | |
2598 | *partial_cluster = 0; | |
2599 | } | |
2600 | ||
a86c6181 AT |
2601 | /* if this leaf is free, then we should |
2602 | * remove it from index block above */ | |
2603 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2604 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2605 | ||
2606 | out: | |
2607 | return err; | |
2608 | } | |
2609 | ||
2610 | /* | |
d0d856e8 RD |
2611 | * ext4_ext_more_to_rm: |
2612 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2613 | */ |
09b88252 | 2614 | static int |
a86c6181 AT |
2615 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2616 | { | |
2617 | BUG_ON(path->p_idx == NULL); | |
2618 | ||
2619 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2620 | return 0; | |
2621 | ||
2622 | /* | |
d0d856e8 | 2623 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2624 | * so we have to consider current index for truncation |
2625 | */ | |
2626 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2627 | return 0; | |
2628 | return 1; | |
2629 | } | |
2630 | ||
5f95d21f LC |
2631 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2632 | ext4_lblk_t end) | |
a86c6181 AT |
2633 | { |
2634 | struct super_block *sb = inode->i_sb; | |
2635 | int depth = ext_depth(inode); | |
968dee77 | 2636 | struct ext4_ext_path *path = NULL; |
0aa06000 | 2637 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2638 | handle_t *handle; |
6f2080e6 | 2639 | int i = 0, err = 0; |
a86c6181 | 2640 | |
5f95d21f | 2641 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2642 | |
2643 | /* probably first extent we're gonna free will be last in block */ | |
2644 | handle = ext4_journal_start(inode, depth + 1); | |
2645 | if (IS_ERR(handle)) | |
2646 | return PTR_ERR(handle); | |
2647 | ||
0617b83f | 2648 | again: |
a86c6181 AT |
2649 | ext4_ext_invalidate_cache(inode); |
2650 | ||
d8990240 AK |
2651 | trace_ext4_ext_remove_space(inode, start, depth); |
2652 | ||
5f95d21f LC |
2653 | /* |
2654 | * Check if we are removing extents inside the extent tree. If that | |
2655 | * is the case, we are going to punch a hole inside the extent tree | |
2656 | * so we have to check whether we need to split the extent covering | |
2657 | * the last block to remove so we can easily remove the part of it | |
2658 | * in ext4_ext_rm_leaf(). | |
2659 | */ | |
2660 | if (end < EXT_MAX_BLOCKS - 1) { | |
2661 | struct ext4_extent *ex; | |
2662 | ext4_lblk_t ee_block; | |
2663 | ||
2664 | /* find extent for this block */ | |
2665 | path = ext4_ext_find_extent(inode, end, NULL); | |
2666 | if (IS_ERR(path)) { | |
2667 | ext4_journal_stop(handle); | |
2668 | return PTR_ERR(path); | |
2669 | } | |
2670 | depth = ext_depth(inode); | |
6f2080e6 | 2671 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2672 | ex = path[depth].p_ext; |
968dee77 | 2673 | if (!ex) { |
6f2080e6 DM |
2674 | if (depth) { |
2675 | EXT4_ERROR_INODE(inode, | |
2676 | "path[%d].p_hdr == NULL", | |
2677 | depth); | |
2678 | err = -EIO; | |
2679 | } | |
2680 | goto out; | |
968dee77 | 2681 | } |
5f95d21f LC |
2682 | |
2683 | ee_block = le32_to_cpu(ex->ee_block); | |
2684 | ||
2685 | /* | |
2686 | * See if the last block is inside the extent, if so split | |
2687 | * the extent at 'end' block so we can easily remove the | |
2688 | * tail of the first part of the split extent in | |
2689 | * ext4_ext_rm_leaf(). | |
2690 | */ | |
2691 | if (end >= ee_block && | |
2692 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2693 | int split_flag = 0; | |
2694 | ||
2695 | if (ext4_ext_is_uninitialized(ex)) | |
2696 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2697 | EXT4_EXT_MARK_UNINIT2; | |
2698 | ||
2699 | /* | |
2700 | * Split the extent in two so that 'end' is the last | |
2701 | * block in the first new extent | |
2702 | */ | |
2703 | err = ext4_split_extent_at(handle, inode, path, | |
2704 | end + 1, split_flag, | |
2705 | EXT4_GET_BLOCKS_PRE_IO | | |
2706 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
2707 | ||
2708 | if (err < 0) | |
2709 | goto out; | |
2710 | } | |
5f95d21f | 2711 | } |
a86c6181 | 2712 | /* |
d0d856e8 RD |
2713 | * We start scanning from right side, freeing all the blocks |
2714 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2715 | */ |
0617b83f | 2716 | depth = ext_depth(inode); |
968dee77 AS |
2717 | if (path) { |
2718 | int k = i = depth; | |
2719 | while (--k > 0) | |
2720 | path[k].p_block = | |
2721 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2722 | } else { | |
2723 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2724 | GFP_NOFS); | |
2725 | if (path == NULL) { | |
2726 | ext4_journal_stop(handle); | |
2727 | return -ENOMEM; | |
2728 | } | |
2729 | path[0].p_depth = depth; | |
2730 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2731 | i = 0; |
5f95d21f | 2732 | |
968dee77 AS |
2733 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
2734 | err = -EIO; | |
2735 | goto out; | |
2736 | } | |
a86c6181 | 2737 | } |
968dee77 | 2738 | err = 0; |
a86c6181 AT |
2739 | |
2740 | while (i >= 0 && err == 0) { | |
2741 | if (i == depth) { | |
2742 | /* this is leaf block */ | |
d583fb87 | 2743 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2744 | &partial_cluster, start, |
5f95d21f | 2745 | end); |
d0d856e8 | 2746 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2747 | brelse(path[i].p_bh); |
2748 | path[i].p_bh = NULL; | |
2749 | i--; | |
2750 | continue; | |
2751 | } | |
2752 | ||
2753 | /* this is index block */ | |
2754 | if (!path[i].p_hdr) { | |
2755 | ext_debug("initialize header\n"); | |
2756 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2757 | } |
2758 | ||
a86c6181 | 2759 | if (!path[i].p_idx) { |
d0d856e8 | 2760 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2761 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2762 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2763 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2764 | path[i].p_hdr, | |
2765 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2766 | } else { | |
d0d856e8 | 2767 | /* we were already here, see at next index */ |
a86c6181 AT |
2768 | path[i].p_idx--; |
2769 | } | |
2770 | ||
2771 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2772 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2773 | path[i].p_idx); | |
2774 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2775 | struct buffer_head *bh; |
a86c6181 | 2776 | /* go to the next level */ |
2ae02107 | 2777 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2778 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2779 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2780 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2781 | if (!bh) { |
a86c6181 AT |
2782 | /* should we reset i_size? */ |
2783 | err = -EIO; | |
2784 | break; | |
2785 | } | |
c29c0ae7 AT |
2786 | if (WARN_ON(i + 1 > depth)) { |
2787 | err = -EIO; | |
2788 | break; | |
2789 | } | |
f8489128 DW |
2790 | if (ext4_ext_check_block(inode, ext_block_hdr(bh), |
2791 | depth - i - 1, bh)) { | |
c29c0ae7 AT |
2792 | err = -EIO; |
2793 | break; | |
2794 | } | |
2795 | path[i + 1].p_bh = bh; | |
a86c6181 | 2796 | |
d0d856e8 RD |
2797 | /* save actual number of indexes since this |
2798 | * number is changed at the next iteration */ | |
a86c6181 AT |
2799 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2800 | i++; | |
2801 | } else { | |
d0d856e8 | 2802 | /* we finished processing this index, go up */ |
a86c6181 | 2803 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2804 | /* index is empty, remove it; |
a86c6181 AT |
2805 | * handle must be already prepared by the |
2806 | * truncatei_leaf() */ | |
2807 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2808 | } | |
d0d856e8 | 2809 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2810 | brelse(path[i].p_bh); |
2811 | path[i].p_bh = NULL; | |
2812 | i--; | |
2813 | ext_debug("return to level %d\n", i); | |
2814 | } | |
2815 | } | |
2816 | ||
d8990240 AK |
2817 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2818 | path->p_hdr->eh_entries); | |
2819 | ||
7b415bf6 AK |
2820 | /* If we still have something in the partial cluster and we have removed |
2821 | * even the first extent, then we should free the blocks in the partial | |
2822 | * cluster as well. */ | |
2823 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2824 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2825 | ||
2826 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2827 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2828 | ||
2829 | ext4_free_blocks(handle, inode, NULL, | |
2830 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2831 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2832 | partial_cluster = 0; | |
2833 | } | |
2834 | ||
a86c6181 AT |
2835 | /* TODO: flexible tree reduction should be here */ |
2836 | if (path->p_hdr->eh_entries == 0) { | |
2837 | /* | |
d0d856e8 RD |
2838 | * truncate to zero freed all the tree, |
2839 | * so we need to correct eh_depth | |
a86c6181 AT |
2840 | */ |
2841 | err = ext4_ext_get_access(handle, inode, path); | |
2842 | if (err == 0) { | |
2843 | ext_inode_hdr(inode)->eh_depth = 0; | |
2844 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2845 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2846 | err = ext4_ext_dirty(handle, inode, path); |
2847 | } | |
2848 | } | |
2849 | out: | |
a86c6181 AT |
2850 | ext4_ext_drop_refs(path); |
2851 | kfree(path); | |
968dee77 AS |
2852 | if (err == -EAGAIN) { |
2853 | path = NULL; | |
0617b83f | 2854 | goto again; |
968dee77 | 2855 | } |
a86c6181 AT |
2856 | ext4_journal_stop(handle); |
2857 | ||
2858 | return err; | |
2859 | } | |
2860 | ||
2861 | /* | |
2862 | * called at mount time | |
2863 | */ | |
2864 | void ext4_ext_init(struct super_block *sb) | |
2865 | { | |
2866 | /* | |
2867 | * possible initialization would be here | |
2868 | */ | |
2869 | ||
83982b6f | 2870 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2871 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2872 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2873 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2874 | ", aggressive tests" |
a86c6181 AT |
2875 | #endif |
2876 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2877 | ", check binsearch" |
a86c6181 AT |
2878 | #endif |
2879 | #ifdef EXTENTS_STATS | |
92b97816 | 2880 | ", stats" |
a86c6181 | 2881 | #endif |
92b97816 | 2882 | "\n"); |
90576c0b | 2883 | #endif |
a86c6181 AT |
2884 | #ifdef EXTENTS_STATS |
2885 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2886 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2887 | EXT4_SB(sb)->s_ext_max = 0; | |
2888 | #endif | |
2889 | } | |
2890 | } | |
2891 | ||
2892 | /* | |
2893 | * called at umount time | |
2894 | */ | |
2895 | void ext4_ext_release(struct super_block *sb) | |
2896 | { | |
83982b6f | 2897 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2898 | return; |
2899 | ||
2900 | #ifdef EXTENTS_STATS | |
2901 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2902 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2903 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2904 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2905 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2906 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2907 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2908 | } | |
2909 | #endif | |
2910 | } | |
2911 | ||
093a088b AK |
2912 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2913 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2914 | { | |
2407518d LC |
2915 | ext4_fsblk_t ee_pblock; |
2916 | unsigned int ee_len; | |
b720303d | 2917 | int ret; |
093a088b | 2918 | |
093a088b | 2919 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2920 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2921 | |
a107e5a3 | 2922 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2923 | if (ret > 0) |
2924 | ret = 0; | |
093a088b | 2925 | |
2407518d | 2926 | return ret; |
093a088b AK |
2927 | } |
2928 | ||
47ea3bb5 YY |
2929 | /* |
2930 | * ext4_split_extent_at() splits an extent at given block. | |
2931 | * | |
2932 | * @handle: the journal handle | |
2933 | * @inode: the file inode | |
2934 | * @path: the path to the extent | |
2935 | * @split: the logical block where the extent is splitted. | |
2936 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2937 | * the states(init or uninit) of new extents. | |
2938 | * @flags: flags used to insert new extent to extent tree. | |
2939 | * | |
2940 | * | |
2941 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2942 | * of which are deterimined by split_flag. | |
2943 | * | |
2944 | * There are two cases: | |
2945 | * a> the extent are splitted into two extent. | |
2946 | * b> split is not needed, and just mark the extent. | |
2947 | * | |
2948 | * return 0 on success. | |
2949 | */ | |
2950 | static int ext4_split_extent_at(handle_t *handle, | |
2951 | struct inode *inode, | |
2952 | struct ext4_ext_path *path, | |
2953 | ext4_lblk_t split, | |
2954 | int split_flag, | |
2955 | int flags) | |
2956 | { | |
2957 | ext4_fsblk_t newblock; | |
2958 | ext4_lblk_t ee_block; | |
2959 | struct ext4_extent *ex, newex, orig_ex; | |
2960 | struct ext4_extent *ex2 = NULL; | |
2961 | unsigned int ee_len, depth; | |
2962 | int err = 0; | |
2963 | ||
dee1f973 DM |
2964 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
2965 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
2966 | ||
47ea3bb5 YY |
2967 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
2968 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2969 | ||
2970 | ext4_ext_show_leaf(inode, path); | |
2971 | ||
2972 | depth = ext_depth(inode); | |
2973 | ex = path[depth].p_ext; | |
2974 | ee_block = le32_to_cpu(ex->ee_block); | |
2975 | ee_len = ext4_ext_get_actual_len(ex); | |
2976 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2977 | ||
2978 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
2979 | ||
2980 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2981 | if (err) | |
2982 | goto out; | |
2983 | ||
2984 | if (split == ee_block) { | |
2985 | /* | |
2986 | * case b: block @split is the block that the extent begins with | |
2987 | * then we just change the state of the extent, and splitting | |
2988 | * is not needed. | |
2989 | */ | |
2990 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2991 | ext4_ext_mark_uninitialized(ex); | |
2992 | else | |
2993 | ext4_ext_mark_initialized(ex); | |
2994 | ||
2995 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 2996 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 2997 | |
ecb94f5f | 2998 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
2999 | goto out; |
3000 | } | |
3001 | ||
3002 | /* case a */ | |
3003 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3004 | ex->ee_len = cpu_to_le16(split - ee_block); | |
3005 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
3006 | ext4_ext_mark_uninitialized(ex); | |
3007 | ||
3008 | /* | |
3009 | * path may lead to new leaf, not to original leaf any more | |
3010 | * after ext4_ext_insert_extent() returns, | |
3011 | */ | |
3012 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3013 | if (err) | |
3014 | goto fix_extent_len; | |
3015 | ||
3016 | ex2 = &newex; | |
3017 | ex2->ee_block = cpu_to_le32(split); | |
3018 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3019 | ext4_ext_store_pblock(ex2, newblock); | |
3020 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3021 | ext4_ext_mark_uninitialized(ex2); | |
3022 | ||
3023 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
3024 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 DM |
3025 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
3026 | if (split_flag & EXT4_EXT_DATA_VALID1) | |
3027 | err = ext4_ext_zeroout(inode, ex2); | |
3028 | else | |
3029 | err = ext4_ext_zeroout(inode, ex); | |
3030 | } else | |
3031 | err = ext4_ext_zeroout(inode, &orig_ex); | |
3032 | ||
47ea3bb5 YY |
3033 | if (err) |
3034 | goto fix_extent_len; | |
3035 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3036 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3037 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3038 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
47ea3bb5 YY |
3039 | goto out; |
3040 | } else if (err) | |
3041 | goto fix_extent_len; | |
3042 | ||
3043 | out: | |
3044 | ext4_ext_show_leaf(inode, path); | |
3045 | return err; | |
3046 | ||
3047 | fix_extent_len: | |
3048 | ex->ee_len = orig_ex.ee_len; | |
3049 | ext4_ext_dirty(handle, inode, path + depth); | |
3050 | return err; | |
3051 | } | |
3052 | ||
3053 | /* | |
3054 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3055 | * by @map as split_flags indicates | |
3056 | * | |
3057 | * It may result in splitting the extent into multiple extents (upto three) | |
3058 | * There are three possibilities: | |
3059 | * a> There is no split required | |
3060 | * b> Splits in two extents: Split is happening at either end of the extent | |
3061 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3062 | * | |
3063 | */ | |
3064 | static int ext4_split_extent(handle_t *handle, | |
3065 | struct inode *inode, | |
3066 | struct ext4_ext_path *path, | |
3067 | struct ext4_map_blocks *map, | |
3068 | int split_flag, | |
3069 | int flags) | |
3070 | { | |
3071 | ext4_lblk_t ee_block; | |
3072 | struct ext4_extent *ex; | |
3073 | unsigned int ee_len, depth; | |
3074 | int err = 0; | |
3075 | int uninitialized; | |
3076 | int split_flag1, flags1; | |
3077 | ||
3078 | depth = ext_depth(inode); | |
3079 | ex = path[depth].p_ext; | |
3080 | ee_block = le32_to_cpu(ex->ee_block); | |
3081 | ee_len = ext4_ext_get_actual_len(ex); | |
3082 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3083 | ||
3084 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3085 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3086 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3087 | if (uninitialized) | |
3088 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3089 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3090 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3091 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3092 | err = ext4_split_extent_at(handle, inode, path, |
3093 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3094 | if (err) |
3095 | goto out; | |
47ea3bb5 YY |
3096 | } |
3097 | ||
3098 | ext4_ext_drop_refs(path); | |
3099 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3100 | if (IS_ERR(path)) | |
3101 | return PTR_ERR(path); | |
3102 | ||
3103 | if (map->m_lblk >= ee_block) { | |
dee1f973 DM |
3104 | split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3105 | EXT4_EXT_DATA_VALID2); | |
47ea3bb5 YY |
3106 | if (uninitialized) |
3107 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
3108 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3109 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
3110 | err = ext4_split_extent_at(handle, inode, path, | |
3111 | map->m_lblk, split_flag1, flags); | |
3112 | if (err) | |
3113 | goto out; | |
3114 | } | |
3115 | ||
3116 | ext4_ext_show_leaf(inode, path); | |
3117 | out: | |
3118 | return err ? err : map->m_len; | |
3119 | } | |
3120 | ||
56055d3a | 3121 | /* |
e35fd660 | 3122 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3123 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3124 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3125 | * uninitialized). |
3126 | * There are three possibilities: | |
3127 | * a> There is no split required: Entire extent should be initialized | |
3128 | * b> Splits in two extents: Write is happening at either end of the extent | |
3129 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3130 | * |
3131 | * Pre-conditions: | |
3132 | * - The extent pointed to by 'path' is uninitialized. | |
3133 | * - The extent pointed to by 'path' contains a superset | |
3134 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3135 | * | |
3136 | * Post-conditions on success: | |
3137 | * - the returned value is the number of blocks beyond map->l_lblk | |
3138 | * that are allocated and initialized. | |
3139 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3140 | */ |
725d26d3 | 3141 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3142 | struct inode *inode, |
3143 | struct ext4_map_blocks *map, | |
3144 | struct ext4_ext_path *path) | |
56055d3a | 3145 | { |
67a5da56 | 3146 | struct ext4_sb_info *sbi; |
6f91bc5f | 3147 | struct ext4_extent_header *eh; |
667eff35 YY |
3148 | struct ext4_map_blocks split_map; |
3149 | struct ext4_extent zero_ex; | |
3150 | struct ext4_extent *ex; | |
21ca087a | 3151 | ext4_lblk_t ee_block, eof_block; |
f85b287a | 3152 | unsigned int ee_len, depth; |
67a5da56 | 3153 | int allocated, max_zeroout = 0; |
56055d3a | 3154 | int err = 0; |
667eff35 | 3155 | int split_flag = 0; |
21ca087a DM |
3156 | |
3157 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3158 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3159 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a | 3160 | |
67a5da56 | 3161 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3162 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3163 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3164 | if (eof_block < map->m_lblk + map->m_len) |
3165 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3166 | |
3167 | depth = ext_depth(inode); | |
6f91bc5f | 3168 | eh = path[depth].p_hdr; |
56055d3a AA |
3169 | ex = path[depth].p_ext; |
3170 | ee_block = le32_to_cpu(ex->ee_block); | |
3171 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3172 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 3173 | |
6f91bc5f EG |
3174 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3175 | ||
3176 | /* Pre-conditions */ | |
3177 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3178 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3179 | |
3180 | /* | |
3181 | * Attempt to transfer newly initialized blocks from the currently | |
3182 | * uninitialized extent to its left neighbor. This is much cheaper | |
3183 | * than an insertion followed by a merge as those involve costly | |
3184 | * memmove() calls. This is the common case in steady state for | |
3185 | * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append | |
3186 | * writes. | |
3187 | * | |
3188 | * Limitations of the current logic: | |
3189 | * - L1: we only deal with writes at the start of the extent. | |
3190 | * The approach could be extended to writes at the end | |
3191 | * of the extent but this scenario was deemed less common. | |
3192 | * - L2: we do not deal with writes covering the whole extent. | |
3193 | * This would require removing the extent if the transfer | |
3194 | * is possible. | |
3195 | * - L3: we only attempt to merge with an extent stored in the | |
3196 | * same extent tree node. | |
3197 | */ | |
3198 | if ((map->m_lblk == ee_block) && /*L1*/ | |
3199 | (map->m_len < ee_len) && /*L2*/ | |
3200 | (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/ | |
3201 | struct ext4_extent *prev_ex; | |
3202 | ext4_lblk_t prev_lblk; | |
3203 | ext4_fsblk_t prev_pblk, ee_pblk; | |
3204 | unsigned int prev_len, write_len; | |
3205 | ||
3206 | prev_ex = ex - 1; | |
3207 | prev_lblk = le32_to_cpu(prev_ex->ee_block); | |
3208 | prev_len = ext4_ext_get_actual_len(prev_ex); | |
3209 | prev_pblk = ext4_ext_pblock(prev_ex); | |
3210 | ee_pblk = ext4_ext_pblock(ex); | |
3211 | write_len = map->m_len; | |
3212 | ||
3213 | /* | |
3214 | * A transfer of blocks from 'ex' to 'prev_ex' is allowed | |
3215 | * upon those conditions: | |
3216 | * - C1: prev_ex is initialized, | |
3217 | * - C2: prev_ex is logically abutting ex, | |
3218 | * - C3: prev_ex is physically abutting ex, | |
3219 | * - C4: prev_ex can receive the additional blocks without | |
3220 | * overflowing the (initialized) length limit. | |
3221 | */ | |
3222 | if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/ | |
3223 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ | |
3224 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
3225 | (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/ | |
3226 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3227 | if (err) | |
3228 | goto out; | |
3229 | ||
3230 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3231 | map, ex, prev_ex); | |
3232 | ||
3233 | /* Shift the start of ex by 'write_len' blocks */ | |
3234 | ex->ee_block = cpu_to_le32(ee_block + write_len); | |
3235 | ext4_ext_store_pblock(ex, ee_pblk + write_len); | |
3236 | ex->ee_len = cpu_to_le16(ee_len - write_len); | |
3237 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3238 | ||
3239 | /* Extend prev_ex by 'write_len' blocks */ | |
3240 | prev_ex->ee_len = cpu_to_le16(prev_len + write_len); | |
3241 | ||
3242 | /* Mark the block containing both extents as dirty */ | |
3243 | ext4_ext_dirty(handle, inode, path + depth); | |
3244 | ||
3245 | /* Update path to point to the right extent */ | |
3246 | path[depth].p_ext = prev_ex; | |
3247 | ||
3248 | /* Result: number of initialized blocks past m_lblk */ | |
3249 | allocated = write_len; | |
3250 | goto out; | |
3251 | } | |
3252 | } | |
3253 | ||
667eff35 | 3254 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3255 | /* |
3256 | * It is safe to convert extent to initialized via explicit | |
3257 | * zeroout only if extent is fully insde i_size or new_size. | |
3258 | */ | |
667eff35 | 3259 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3260 | |
67a5da56 ZL |
3261 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3262 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
3263 | inode->i_sb->s_blocksize_bits; | |
3264 | ||
3265 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3266 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3267 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3268 | if (err) |
d03856bd | 3269 | goto out; |
d03856bd AK |
3270 | |
3271 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3272 | if (err) | |
3273 | goto out; | |
667eff35 | 3274 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3275 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3276 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3277 | goto out; |
56055d3a | 3278 | } |
667eff35 | 3279 | |
56055d3a | 3280 | /* |
667eff35 YY |
3281 | * four cases: |
3282 | * 1. split the extent into three extents. | |
3283 | * 2. split the extent into two extents, zeroout the first half. | |
3284 | * 3. split the extent into two extents, zeroout the second half. | |
3285 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3286 | */ |
667eff35 YY |
3287 | split_map.m_lblk = map->m_lblk; |
3288 | split_map.m_len = map->m_len; | |
3289 | ||
67a5da56 ZL |
3290 | if (max_zeroout && (allocated > map->m_len)) { |
3291 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3292 | /* case 3 */ |
3293 | zero_ex.ee_block = | |
9b940f8e AH |
3294 | cpu_to_le32(map->m_lblk); |
3295 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3296 | ext4_ext_store_pblock(&zero_ex, |
3297 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3298 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3299 | if (err) |
3300 | goto out; | |
667eff35 YY |
3301 | split_map.m_lblk = map->m_lblk; |
3302 | split_map.m_len = allocated; | |
67a5da56 | 3303 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3304 | /* case 2 */ |
3305 | if (map->m_lblk != ee_block) { | |
3306 | zero_ex.ee_block = ex->ee_block; | |
3307 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3308 | ee_block); | |
3309 | ext4_ext_store_pblock(&zero_ex, | |
3310 | ext4_ext_pblock(ex)); | |
3311 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3312 | if (err) | |
3313 | goto out; | |
3314 | } | |
3315 | ||
667eff35 | 3316 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3317 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3318 | allocated = map->m_len; | |
56055d3a AA |
3319 | } |
3320 | } | |
667eff35 YY |
3321 | |
3322 | allocated = ext4_split_extent(handle, inode, path, | |
67a5da56 | 3323 | &split_map, split_flag, 0); |
667eff35 YY |
3324 | if (allocated < 0) |
3325 | err = allocated; | |
3326 | ||
56055d3a AA |
3327 | out: |
3328 | return err ? err : allocated; | |
3329 | } | |
3330 | ||
0031462b | 3331 | /* |
e35fd660 | 3332 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3333 | * ext4_get_blocks_dio_write() when DIO to write |
3334 | * to an uninitialized extent. | |
3335 | * | |
fd018fe8 | 3336 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3337 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3338 | * There are three possibilities: |
3339 | * a> There is no split required: Entire extent should be uninitialized | |
3340 | * b> Splits in two extents: Write is happening at either end of the extent | |
3341 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3342 | * | |
3343 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3344 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3345 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3346 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3347 | * into three uninitialized extent(at most). After IO complete, the part |
3348 | * being filled will be convert to initialized by the end_io callback function | |
3349 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3350 | * |
3351 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3352 | */ |
3353 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3354 | struct inode *inode, | |
e35fd660 | 3355 | struct ext4_map_blocks *map, |
0031462b | 3356 | struct ext4_ext_path *path, |
0031462b MC |
3357 | int flags) |
3358 | { | |
667eff35 YY |
3359 | ext4_lblk_t eof_block; |
3360 | ext4_lblk_t ee_block; | |
3361 | struct ext4_extent *ex; | |
3362 | unsigned int ee_len; | |
3363 | int split_flag = 0, depth; | |
21ca087a DM |
3364 | |
3365 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3366 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3367 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3368 | |
3369 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3370 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3371 | if (eof_block < map->m_lblk + map->m_len) |
3372 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3373 | /* |
3374 | * It is safe to convert extent to initialized via explicit | |
3375 | * zeroout only if extent is fully insde i_size or new_size. | |
3376 | */ | |
667eff35 YY |
3377 | depth = ext_depth(inode); |
3378 | ex = path[depth].p_ext; | |
3379 | ee_block = le32_to_cpu(ex->ee_block); | |
3380 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3381 | |
667eff35 YY |
3382 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3383 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3384 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3385 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3386 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3387 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3388 | } |
197217a5 | 3389 | |
c7064ef1 | 3390 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3391 | struct inode *inode, |
3392 | struct ext4_map_blocks *map, | |
3393 | struct ext4_ext_path *path) | |
0031462b MC |
3394 | { |
3395 | struct ext4_extent *ex; | |
dee1f973 DM |
3396 | ext4_lblk_t ee_block; |
3397 | unsigned int ee_len; | |
0031462b MC |
3398 | int depth; |
3399 | int err = 0; | |
0031462b MC |
3400 | |
3401 | depth = ext_depth(inode); | |
0031462b | 3402 | ex = path[depth].p_ext; |
dee1f973 DM |
3403 | ee_block = le32_to_cpu(ex->ee_block); |
3404 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3405 | |
197217a5 YY |
3406 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3407 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3408 | (unsigned long long)ee_block, ee_len); |
3409 | ||
3410 | /* If extent is larger than requested then split is required */ | |
3411 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
3412 | err = ext4_split_unwritten_extents(handle, inode, map, path, | |
3413 | EXT4_GET_BLOCKS_CONVERT); | |
3414 | if (err < 0) | |
3415 | goto out; | |
3416 | ext4_ext_drop_refs(path); | |
3417 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3418 | if (IS_ERR(path)) { | |
3419 | err = PTR_ERR(path); | |
3420 | goto out; | |
3421 | } | |
3422 | depth = ext_depth(inode); | |
3423 | ex = path[depth].p_ext; | |
3424 | } | |
197217a5 | 3425 | |
0031462b MC |
3426 | err = ext4_ext_get_access(handle, inode, path + depth); |
3427 | if (err) | |
3428 | goto out; | |
3429 | /* first mark the extent as initialized */ | |
3430 | ext4_ext_mark_initialized(ex); | |
3431 | ||
197217a5 YY |
3432 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3433 | * borders are not changed | |
0031462b | 3434 | */ |
ecb94f5f | 3435 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3436 | |
0031462b | 3437 | /* Mark modified extent as dirty */ |
ecb94f5f | 3438 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3439 | out: |
3440 | ext4_ext_show_leaf(inode, path); | |
3441 | return err; | |
3442 | } | |
3443 | ||
515f41c3 AK |
3444 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3445 | sector_t block, int count) | |
3446 | { | |
3447 | int i; | |
3448 | for (i = 0; i < count; i++) | |
3449 | unmap_underlying_metadata(bdev, block + i); | |
3450 | } | |
3451 | ||
58590b06 TT |
3452 | /* |
3453 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3454 | */ | |
3455 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3456 | ext4_lblk_t lblk, |
58590b06 TT |
3457 | struct ext4_ext_path *path, |
3458 | unsigned int len) | |
3459 | { | |
3460 | int i, depth; | |
3461 | struct ext4_extent_header *eh; | |
65922cb5 | 3462 | struct ext4_extent *last_ex; |
58590b06 TT |
3463 | |
3464 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3465 | return 0; | |
3466 | ||
3467 | depth = ext_depth(inode); | |
3468 | eh = path[depth].p_hdr; | |
58590b06 | 3469 | |
afcff5d8 LC |
3470 | /* |
3471 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3472 | * do not care for this case anymore. Simply remove the flag | |
3473 | * if there are no extents. | |
3474 | */ | |
3475 | if (unlikely(!eh->eh_entries)) | |
3476 | goto out; | |
58590b06 TT |
3477 | last_ex = EXT_LAST_EXTENT(eh); |
3478 | /* | |
3479 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3480 | * last block in the last extent in the file. We test this by | |
3481 | * first checking to see if the caller to | |
3482 | * ext4_ext_get_blocks() was interested in the last block (or | |
3483 | * a block beyond the last block) in the current extent. If | |
3484 | * this turns out to be false, we can bail out from this | |
3485 | * function immediately. | |
3486 | */ | |
d002ebf1 | 3487 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3488 | ext4_ext_get_actual_len(last_ex)) |
3489 | return 0; | |
3490 | /* | |
3491 | * If the caller does appear to be planning to write at or | |
3492 | * beyond the end of the current extent, we then test to see | |
3493 | * if the current extent is the last extent in the file, by | |
3494 | * checking to make sure it was reached via the rightmost node | |
3495 | * at each level of the tree. | |
3496 | */ | |
3497 | for (i = depth-1; i >= 0; i--) | |
3498 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3499 | return 0; | |
afcff5d8 | 3500 | out: |
58590b06 TT |
3501 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3502 | return ext4_mark_inode_dirty(handle, inode); | |
3503 | } | |
3504 | ||
7b415bf6 AK |
3505 | /** |
3506 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3507 | * | |
7d1b1fbc | 3508 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 AK |
3509 | */ |
3510 | static int ext4_find_delalloc_range(struct inode *inode, | |
3511 | ext4_lblk_t lblk_start, | |
7d1b1fbc | 3512 | ext4_lblk_t lblk_end) |
7b415bf6 | 3513 | { |
7d1b1fbc | 3514 | struct extent_status es; |
7b415bf6 | 3515 | |
7d1b1fbc ZL |
3516 | es.start = lblk_start; |
3517 | ext4_es_find_extent(inode, &es); | |
3518 | if (es.len == 0) | |
3519 | return 0; /* there is no delay extent in this tree */ | |
3520 | else if (es.start <= lblk_start && lblk_start < es.start + es.len) | |
3521 | return 1; | |
3522 | else if (lblk_start <= es.start && es.start <= lblk_end) | |
3523 | return 1; | |
7b415bf6 | 3524 | else |
7d1b1fbc | 3525 | return 0; |
7b415bf6 AK |
3526 | } |
3527 | ||
7d1b1fbc | 3528 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3529 | { |
3530 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3531 | ext4_lblk_t lblk_start, lblk_end; | |
3532 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3533 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3534 | ||
7d1b1fbc | 3535 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3536 | } |
3537 | ||
3538 | /** | |
3539 | * Determines how many complete clusters (out of those specified by the 'map') | |
3540 | * are under delalloc and were reserved quota for. | |
3541 | * This function is called when we are writing out the blocks that were | |
3542 | * originally written with their allocation delayed, but then the space was | |
3543 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3544 | * The cases to look for are: | |
3545 | * ('=' indicated delayed allocated blocks | |
3546 | * '-' indicates non-delayed allocated blocks) | |
3547 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3548 | * are not delalloc'ed. | |
3549 | * Ex: | |
3550 | * |----c---=|====c====|====c====|===-c----| | |
3551 | * |++++++ allocated ++++++| | |
3552 | * ==> 4 complete clusters in above example | |
3553 | * | |
3554 | * (b) partial cluster (outside of allocated range) towards either end is | |
3555 | * marked for delayed allocation. In this case, we will exclude that | |
3556 | * cluster. | |
3557 | * Ex: | |
3558 | * |----====c========|========c========| | |
3559 | * |++++++ allocated ++++++| | |
3560 | * ==> 1 complete clusters in above example | |
3561 | * | |
3562 | * Ex: | |
3563 | * |================c================| | |
3564 | * |++++++ allocated ++++++| | |
3565 | * ==> 0 complete clusters in above example | |
3566 | * | |
3567 | * The ext4_da_update_reserve_space will be called only if we | |
3568 | * determine here that there were some "entire" clusters that span | |
3569 | * this 'allocated' range. | |
3570 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3571 | * without ever calling ext4_find_delalloc_range. | |
3572 | */ | |
3573 | static unsigned int | |
3574 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3575 | unsigned int num_blks) | |
3576 | { | |
3577 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3578 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3579 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3580 | unsigned int allocated_clusters = 0; | |
3581 | ||
3582 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3583 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3584 | ||
3585 | /* max possible clusters for this allocation */ | |
3586 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3587 | ||
d8990240 AK |
3588 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3589 | ||
7b415bf6 AK |
3590 | /* Check towards left side */ |
3591 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3592 | if (c_offset) { | |
3593 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3594 | lblk_to = lblk_from + c_offset - 1; | |
3595 | ||
7d1b1fbc | 3596 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3597 | allocated_clusters--; |
3598 | } | |
3599 | ||
3600 | /* Now check towards right. */ | |
3601 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3602 | if (allocated_clusters && c_offset) { | |
3603 | lblk_from = lblk_start + num_blks; | |
3604 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3605 | ||
7d1b1fbc | 3606 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3607 | allocated_clusters--; |
3608 | } | |
3609 | ||
3610 | return allocated_clusters; | |
3611 | } | |
3612 | ||
0031462b MC |
3613 | static int |
3614 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3615 | struct ext4_map_blocks *map, |
0031462b | 3616 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3617 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3618 | { |
3619 | int ret = 0; | |
3620 | int err = 0; | |
f45ee3a1 | 3621 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3622 | |
88635ca2 ZL |
3623 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3624 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3625 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3626 | flags, allocated); |
3627 | ext4_ext_show_leaf(inode, path); | |
3628 | ||
b5645534 ZL |
3629 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3630 | allocated, newblock); | |
d8990240 | 3631 | |
c7064ef1 | 3632 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3633 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3634 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3635 | path, flags); | |
82e54229 DM |
3636 | if (ret <= 0) |
3637 | goto out; | |
5f524950 M |
3638 | /* |
3639 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3640 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3641 | * completed |
3642 | */ | |
0edeb71d TM |
3643 | if (io) |
3644 | ext4_set_io_unwritten_flag(inode, io); | |
3645 | else | |
19f5fb7a | 3646 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
744692dc | 3647 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3648 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3649 | goto out; |
3650 | } | |
c7064ef1 | 3651 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3652 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3653 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3654 | path); |
58590b06 | 3655 | if (ret >= 0) { |
b436b9be | 3656 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3657 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3658 | path, map->m_len); | |
58590b06 TT |
3659 | } else |
3660 | err = ret; | |
0031462b MC |
3661 | goto out2; |
3662 | } | |
3663 | /* buffered IO case */ | |
3664 | /* | |
3665 | * repeat fallocate creation request | |
3666 | * we already have an unwritten extent | |
3667 | */ | |
3668 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | |
3669 | goto map_out; | |
3670 | ||
3671 | /* buffered READ or buffered write_begin() lookup */ | |
3672 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3673 | /* | |
3674 | * We have blocks reserved already. We | |
3675 | * return allocated blocks so that delalloc | |
3676 | * won't do block reservation for us. But | |
3677 | * the buffer head will be unmapped so that | |
3678 | * a read from the block returns 0s. | |
3679 | */ | |
e35fd660 | 3680 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3681 | goto out1; |
3682 | } | |
3683 | ||
3684 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3685 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
a4e5d88b | 3686 | if (ret >= 0) |
b436b9be | 3687 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3688 | out: |
3689 | if (ret <= 0) { | |
3690 | err = ret; | |
3691 | goto out2; | |
3692 | } else | |
3693 | allocated = ret; | |
e35fd660 | 3694 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3695 | /* |
3696 | * if we allocated more blocks than requested | |
3697 | * we need to make sure we unmap the extra block | |
3698 | * allocated. The actual needed block will get | |
3699 | * unmapped later when we find the buffer_head marked | |
3700 | * new. | |
3701 | */ | |
e35fd660 | 3702 | if (allocated > map->m_len) { |
515f41c3 | 3703 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3704 | newblock + map->m_len, |
3705 | allocated - map->m_len); | |
3706 | allocated = map->m_len; | |
515f41c3 | 3707 | } |
5f634d06 AK |
3708 | |
3709 | /* | |
3710 | * If we have done fallocate with the offset that is already | |
3711 | * delayed allocated, we would have block reservation | |
3712 | * and quota reservation done in the delayed write path. | |
3713 | * But fallocate would have already updated quota and block | |
3714 | * count for this offset. So cancel these reservation | |
3715 | */ | |
7b415bf6 AK |
3716 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3717 | unsigned int reserved_clusters; | |
3718 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3719 | map->m_lblk, map->m_len); | |
3720 | if (reserved_clusters) | |
3721 | ext4_da_update_reserve_space(inode, | |
3722 | reserved_clusters, | |
3723 | 0); | |
3724 | } | |
5f634d06 | 3725 | |
0031462b | 3726 | map_out: |
e35fd660 | 3727 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3728 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3729 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3730 | map->m_len); | |
3731 | if (err < 0) | |
3732 | goto out2; | |
3733 | } | |
0031462b | 3734 | out1: |
e35fd660 TT |
3735 | if (allocated > map->m_len) |
3736 | allocated = map->m_len; | |
0031462b | 3737 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3738 | map->m_pblk = newblock; |
3739 | map->m_len = allocated; | |
0031462b MC |
3740 | out2: |
3741 | if (path) { | |
3742 | ext4_ext_drop_refs(path); | |
3743 | kfree(path); | |
3744 | } | |
3745 | return err ? err : allocated; | |
3746 | } | |
58590b06 | 3747 | |
4d33b1ef TT |
3748 | /* |
3749 | * get_implied_cluster_alloc - check to see if the requested | |
3750 | * allocation (in the map structure) overlaps with a cluster already | |
3751 | * allocated in an extent. | |
d8990240 | 3752 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3753 | * @map The requested lblk->pblk mapping |
3754 | * @ex The extent structure which might contain an implied | |
3755 | * cluster allocation | |
3756 | * | |
3757 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3758 | * find blocks that were already in the inode's extent tree. Hence, | |
3759 | * we know that the beginning of the requested region cannot overlap | |
3760 | * the extent from the inode's extent tree. There are three cases we | |
3761 | * want to catch. The first is this case: | |
3762 | * | |
3763 | * |--- cluster # N--| | |
3764 | * |--- extent ---| |---- requested region ---| | |
3765 | * |==========| | |
3766 | * | |
3767 | * The second case that we need to test for is this one: | |
3768 | * | |
3769 | * |--------- cluster # N ----------------| | |
3770 | * |--- requested region --| |------- extent ----| | |
3771 | * |=======================| | |
3772 | * | |
3773 | * The third case is when the requested region lies between two extents | |
3774 | * within the same cluster: | |
3775 | * |------------- cluster # N-------------| | |
3776 | * |----- ex -----| |---- ex_right ----| | |
3777 | * |------ requested region ------| | |
3778 | * |================| | |
3779 | * | |
3780 | * In each of the above cases, we need to set the map->m_pblk and | |
3781 | * map->m_len so it corresponds to the return the extent labelled as | |
3782 | * "|====|" from cluster #N, since it is already in use for data in | |
3783 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3784 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3785 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3786 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3787 | * by calling ext4_mb_new_blocks(). | |
3788 | */ | |
d8990240 | 3789 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3790 | struct ext4_map_blocks *map, |
3791 | struct ext4_extent *ex, | |
3792 | struct ext4_ext_path *path) | |
3793 | { | |
d8990240 | 3794 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3795 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3796 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3797 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3798 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3799 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3800 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3801 | ||
3802 | /* The extent passed in that we are trying to match */ | |
3803 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3804 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3805 | ||
3806 | /* The requested region passed into ext4_map_blocks() */ | |
3807 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3808 | |
3809 | if ((rr_cluster_start == ex_cluster_end) || | |
3810 | (rr_cluster_start == ex_cluster_start)) { | |
3811 | if (rr_cluster_start == ex_cluster_end) | |
3812 | ee_start += ee_len - 1; | |
3813 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3814 | c_offset; | |
3815 | map->m_len = min(map->m_len, | |
3816 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3817 | /* | |
3818 | * Check for and handle this case: | |
3819 | * | |
3820 | * |--------- cluster # N-------------| | |
3821 | * |------- extent ----| | |
3822 | * |--- requested region ---| | |
3823 | * |===========| | |
3824 | */ | |
3825 | ||
3826 | if (map->m_lblk < ee_block) | |
3827 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3828 | ||
3829 | /* | |
3830 | * Check for the case where there is already another allocated | |
3831 | * block to the right of 'ex' but before the end of the cluster. | |
3832 | * | |
3833 | * |------------- cluster # N-------------| | |
3834 | * |----- ex -----| |---- ex_right ----| | |
3835 | * |------ requested region ------| | |
3836 | * |================| | |
3837 | */ | |
3838 | if (map->m_lblk > ee_block) { | |
3839 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3840 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3841 | } | |
d8990240 AK |
3842 | |
3843 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3844 | return 1; |
3845 | } | |
d8990240 AK |
3846 | |
3847 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3848 | return 0; |
3849 | } | |
3850 | ||
3851 | ||
c278bfec | 3852 | /* |
f5ab0d1f MC |
3853 | * Block allocation/map/preallocation routine for extents based files |
3854 | * | |
3855 | * | |
c278bfec | 3856 | * Need to be called with |
0e855ac8 AK |
3857 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3858 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3859 | * |
3860 | * return > 0, number of of blocks already mapped/allocated | |
3861 | * if create == 0 and these are pre-allocated blocks | |
3862 | * buffer head is unmapped | |
3863 | * otherwise blocks are mapped | |
3864 | * | |
3865 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3866 | * buffer head is unmapped | |
3867 | * | |
3868 | * return < 0, error case. | |
c278bfec | 3869 | */ |
e35fd660 TT |
3870 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3871 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3872 | { |
3873 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3874 | struct ext4_extent newex, *ex, *ex2; |
3875 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3876 | ext4_fsblk_t newblock = 0; |
37794732 | 3877 | int free_on_err = 0, err = 0, depth; |
4d33b1ef | 3878 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 3879 | unsigned int allocated_clusters = 0; |
c9de560d | 3880 | struct ext4_allocation_request ar; |
f45ee3a1 | 3881 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 3882 | ext4_lblk_t cluster_offset; |
82e54229 | 3883 | int set_unwritten = 0; |
a86c6181 | 3884 | |
84fe3bef | 3885 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3886 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3887 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3888 | |
3889 | /* check in cache */ | |
7877191c | 3890 | if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) { |
b05e6ae5 | 3891 | if (!newex.ee_start_lo && !newex.ee_start_hi) { |
7b415bf6 | 3892 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 3893 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
3894 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
3895 | ||
c2177057 | 3896 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3897 | /* |
3898 | * block isn't allocated yet and | |
3899 | * user doesn't want to allocate it | |
3900 | */ | |
a86c6181 AT |
3901 | goto out2; |
3902 | } | |
3903 | /* we should allocate requested block */ | |
b05e6ae5 | 3904 | } else { |
a86c6181 | 3905 | /* block is already allocated */ |
7b415bf6 AK |
3906 | if (sbi->s_cluster_ratio > 1) |
3907 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
e35fd660 | 3908 | newblock = map->m_lblk |
8c55e204 | 3909 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3910 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3911 | /* number of remaining blocks in the extent */ |
b939e376 | 3912 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3913 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3914 | goto out; |
a86c6181 AT |
3915 | } |
3916 | } | |
3917 | ||
3918 | /* find extent for this block */ | |
e35fd660 | 3919 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3920 | if (IS_ERR(path)) { |
3921 | err = PTR_ERR(path); | |
3922 | path = NULL; | |
3923 | goto out2; | |
3924 | } | |
3925 | ||
3926 | depth = ext_depth(inode); | |
3927 | ||
3928 | /* | |
d0d856e8 RD |
3929 | * consistent leaf must not be empty; |
3930 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3931 | * this is why assert can't be put in ext4_ext_find_extent() |
3932 | */ | |
273df556 FM |
3933 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3934 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3935 | "lblock: %lu, depth: %d pblock %lld", |
3936 | (unsigned long) map->m_lblk, depth, | |
3937 | path[depth].p_block); | |
034fb4c9 SP |
3938 | err = -EIO; |
3939 | goto out2; | |
3940 | } | |
a86c6181 | 3941 | |
7e028976 AM |
3942 | ex = path[depth].p_ext; |
3943 | if (ex) { | |
725d26d3 | 3944 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3945 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3946 | unsigned short ee_len; |
471d4011 SB |
3947 | |
3948 | /* | |
471d4011 | 3949 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3950 | * we split out initialized portions during a write. |
471d4011 | 3951 | */ |
a2df2a63 | 3952 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
3953 | |
3954 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
3955 | ||
d0d856e8 | 3956 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
3957 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
3958 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 3959 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3960 | allocated = ee_len - (map->m_lblk - ee_block); |
3961 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3962 | ee_block, ee_len, newblock); | |
56055d3a | 3963 | |
e861304b | 3964 | /* |
7877191c LC |
3965 | * Do not put uninitialized extent |
3966 | * in the cache | |
e861304b | 3967 | */ |
7877191c LC |
3968 | if (!ext4_ext_is_uninitialized(ex)) { |
3969 | ext4_ext_put_in_cache(inode, ee_block, | |
3970 | ee_len, ee_start); | |
3971 | goto out; | |
f7d0d379 | 3972 | } |
37794732 | 3973 | allocated = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
3974 | handle, inode, map, path, flags, |
3975 | allocated, newblock); | |
37794732 | 3976 | goto out3; |
a86c6181 AT |
3977 | } |
3978 | } | |
3979 | ||
7b415bf6 | 3980 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 3981 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
3982 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
3983 | ||
a86c6181 | 3984 | /* |
d0d856e8 | 3985 | * requested block isn't allocated yet; |
a86c6181 AT |
3986 | * we couldn't try to create block if create flag is zero |
3987 | */ | |
c2177057 | 3988 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3989 | /* |
3990 | * put just found gap into cache to speed up | |
3991 | * subsequent requests | |
3992 | */ | |
e35fd660 | 3993 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
3994 | goto out2; |
3995 | } | |
4d33b1ef | 3996 | |
a86c6181 | 3997 | /* |
c2ea3fde | 3998 | * Okay, we need to do block allocation. |
63f57933 | 3999 | */ |
7b415bf6 | 4000 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4001 | newex.ee_block = cpu_to_le32(map->m_lblk); |
4002 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
4003 | ||
4004 | /* | |
4005 | * If we are doing bigalloc, check to see if the extent returned | |
4006 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4007 | */ | |
4008 | if (cluster_offset && ex && | |
d8990240 | 4009 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4010 | ar.len = allocated = map->m_len; |
4011 | newblock = map->m_pblk; | |
7b415bf6 | 4012 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4013 | goto got_allocated_blocks; |
4014 | } | |
a86c6181 | 4015 | |
c9de560d | 4016 | /* find neighbour allocated blocks */ |
e35fd660 | 4017 | ar.lleft = map->m_lblk; |
c9de560d AT |
4018 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4019 | if (err) | |
4020 | goto out2; | |
e35fd660 | 4021 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4022 | ex2 = NULL; |
4023 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4024 | if (err) |
4025 | goto out2; | |
25d14f98 | 4026 | |
4d33b1ef TT |
4027 | /* Check if the extent after searching to the right implies a |
4028 | * cluster we can use. */ | |
4029 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4030 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4031 | ar.len = allocated = map->m_len; |
4032 | newblock = map->m_pblk; | |
7b415bf6 | 4033 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4034 | goto got_allocated_blocks; |
4035 | } | |
4036 | ||
749269fa AA |
4037 | /* |
4038 | * See if request is beyond maximum number of blocks we can have in | |
4039 | * a single extent. For an initialized extent this limit is | |
4040 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4041 | * EXT_UNINIT_MAX_LEN. | |
4042 | */ | |
e35fd660 | 4043 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4044 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4045 | map->m_len = EXT_INIT_MAX_LEN; |
4046 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4047 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4048 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4049 | |
e35fd660 | 4050 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4051 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4052 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4053 | if (err) |
b939e376 | 4054 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4055 | else |
e35fd660 | 4056 | allocated = map->m_len; |
c9de560d AT |
4057 | |
4058 | /* allocate new block */ | |
4059 | ar.inode = inode; | |
e35fd660 TT |
4060 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4061 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4062 | /* |
4063 | * We calculate the offset from the beginning of the cluster | |
4064 | * for the logical block number, since when we allocate a | |
4065 | * physical cluster, the physical block should start at the | |
4066 | * same offset from the beginning of the cluster. This is | |
4067 | * needed so that future calls to get_implied_cluster_alloc() | |
4068 | * work correctly. | |
4069 | */ | |
4070 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4071 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4072 | ar.goal -= offset; | |
4073 | ar.logical -= offset; | |
c9de560d AT |
4074 | if (S_ISREG(inode->i_mode)) |
4075 | ar.flags = EXT4_MB_HINT_DATA; | |
4076 | else | |
4077 | /* disable in-core preallocation for non-regular files */ | |
4078 | ar.flags = 0; | |
556b27ab VH |
4079 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4080 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4081 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4082 | if (!newblock) |
4083 | goto out2; | |
84fe3bef | 4084 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4085 | ar.goal, newblock, allocated); |
4d33b1ef | 4086 | free_on_err = 1; |
7b415bf6 | 4087 | allocated_clusters = ar.len; |
4d33b1ef TT |
4088 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4089 | if (ar.len > allocated) | |
4090 | ar.len = allocated; | |
a86c6181 | 4091 | |
4d33b1ef | 4092 | got_allocated_blocks: |
a86c6181 | 4093 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4094 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4095 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4096 | /* Mark uninitialized */ |
4097 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4098 | ext4_ext_mark_uninitialized(&newex); |
8d5d02e6 | 4099 | /* |
744692dc | 4100 | * io_end structure was created for every IO write to an |
25985edc | 4101 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4102 | * here we flag the IO that really needs the conversion. |
5f524950 | 4103 | * For non asycn direct IO case, flag the inode state |
25985edc | 4104 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4105 | */ |
82e54229 DM |
4106 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4107 | set_unwritten = 1; | |
744692dc | 4108 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4109 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4110 | } |
c8d46e41 | 4111 | |
a4e5d88b DM |
4112 | err = 0; |
4113 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4114 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4115 | path, ar.len); | |
575a1d4b JZ |
4116 | if (!err) |
4117 | err = ext4_ext_insert_extent(handle, inode, path, | |
4118 | &newex, flags); | |
82e54229 DM |
4119 | |
4120 | if (!err && set_unwritten) { | |
4121 | if (io) | |
4122 | ext4_set_io_unwritten_flag(inode, io); | |
4123 | else | |
4124 | ext4_set_inode_state(inode, | |
4125 | EXT4_STATE_DIO_UNWRITTEN); | |
4126 | } | |
4127 | ||
4d33b1ef | 4128 | if (err && free_on_err) { |
7132de74 MP |
4129 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4130 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4131 | /* free data blocks we just allocated */ |
c9de560d AT |
4132 | /* not a good idea to call discard here directly, |
4133 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4134 | ext4_discard_preallocations(inode); |
7dc57615 | 4135 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4136 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4137 | goto out2; |
315054f0 | 4138 | } |
a86c6181 | 4139 | |
a86c6181 | 4140 | /* previous routine could use block we allocated */ |
bf89d16f | 4141 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4142 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4143 | if (allocated > map->m_len) |
4144 | allocated = map->m_len; | |
4145 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4146 | |
5f634d06 AK |
4147 | /* |
4148 | * Update reserved blocks/metadata blocks after successful | |
4149 | * block allocation which had been deferred till now. | |
4150 | */ | |
7b415bf6 | 4151 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4152 | unsigned int reserved_clusters; |
7b415bf6 | 4153 | /* |
81fdbb4a | 4154 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4155 | */ |
4156 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4157 | map->m_lblk, allocated); | |
4158 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4159 | if (reserved_clusters) { | |
4160 | /* | |
4161 | * We have clusters reserved for this range. | |
4162 | * But since we are not doing actual allocation | |
4163 | * and are simply using blocks from previously | |
4164 | * allocated cluster, we should release the | |
4165 | * reservation and not claim quota. | |
4166 | */ | |
4167 | ext4_da_update_reserve_space(inode, | |
4168 | reserved_clusters, 0); | |
4169 | } | |
4170 | } else { | |
4171 | BUG_ON(allocated_clusters < reserved_clusters); | |
4172 | /* We will claim quota for all newly allocated blocks.*/ | |
4173 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4174 | 1); | |
4175 | if (reserved_clusters < allocated_clusters) { | |
5356f261 | 4176 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4177 | int reservation = allocated_clusters - |
4178 | reserved_clusters; | |
4179 | /* | |
4180 | * It seems we claimed few clusters outside of | |
4181 | * the range of this allocation. We should give | |
4182 | * it back to the reservation pool. This can | |
4183 | * happen in the following case: | |
4184 | * | |
4185 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4186 | * cluster has 4 blocks. Thus, the clusters | |
4187 | * are [0-3],[4-7],[8-11]... | |
4188 | * * First comes delayed allocation write for | |
4189 | * logical blocks 10 & 11. Since there were no | |
4190 | * previous delayed allocated blocks in the | |
4191 | * range [8-11], we would reserve 1 cluster | |
4192 | * for this write. | |
4193 | * * Next comes write for logical blocks 3 to 8. | |
4194 | * In this case, we will reserve 2 clusters | |
4195 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4196 | * that range has a delayed allocated blocks. | |
4197 | * Thus total reserved clusters now becomes 3. | |
4198 | * * Now, during the delayed allocation writeout | |
4199 | * time, we will first write blocks [3-8] and | |
4200 | * allocate 3 clusters for writing these | |
4201 | * blocks. Also, we would claim all these | |
4202 | * three clusters above. | |
4203 | * * Now when we come here to writeout the | |
4204 | * blocks [10-11], we would expect to claim | |
4205 | * the reservation of 1 cluster we had made | |
4206 | * (and we would claim it since there are no | |
4207 | * more delayed allocated blocks in the range | |
4208 | * [8-11]. But our reserved cluster count had | |
4209 | * already gone to 0. | |
4210 | * | |
4211 | * Thus, at the step 4 above when we determine | |
4212 | * that there are still some unwritten delayed | |
4213 | * allocated blocks outside of our current | |
4214 | * block range, we should increment the | |
4215 | * reserved clusters count so that when the | |
4216 | * remaining blocks finally gets written, we | |
4217 | * could claim them. | |
4218 | */ | |
5356f261 AK |
4219 | dquot_reserve_block(inode, |
4220 | EXT4_C2B(sbi, reservation)); | |
4221 | spin_lock(&ei->i_block_reservation_lock); | |
4222 | ei->i_reserved_data_blocks += reservation; | |
4223 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 AK |
4224 | } |
4225 | } | |
4226 | } | |
5f634d06 | 4227 | |
b436b9be JK |
4228 | /* |
4229 | * Cache the extent and update transaction to commit on fdatasync only | |
4230 | * when it is _not_ an uninitialized extent. | |
4231 | */ | |
4232 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 4233 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
4234 | ext4_update_inode_fsync_trans(handle, inode, 1); |
4235 | } else | |
4236 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 4237 | out: |
e35fd660 TT |
4238 | if (allocated > map->m_len) |
4239 | allocated = map->m_len; | |
a86c6181 | 4240 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4241 | map->m_flags |= EXT4_MAP_MAPPED; |
4242 | map->m_pblk = newblock; | |
4243 | map->m_len = allocated; | |
a86c6181 AT |
4244 | out2: |
4245 | if (path) { | |
4246 | ext4_ext_drop_refs(path); | |
4247 | kfree(path); | |
4248 | } | |
e861304b | 4249 | |
37794732 | 4250 | out3: |
19b303d8 | 4251 | trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated); |
e7b319e3 | 4252 | |
7877191c | 4253 | return err ? err : allocated; |
a86c6181 AT |
4254 | } |
4255 | ||
cf108bca | 4256 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
4257 | { |
4258 | struct address_space *mapping = inode->i_mapping; | |
4259 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 4260 | ext4_lblk_t last_block; |
a86c6181 | 4261 | handle_t *handle; |
189e868f | 4262 | loff_t page_len; |
a86c6181 AT |
4263 | int err = 0; |
4264 | ||
3889fd57 JZ |
4265 | /* |
4266 | * finish any pending end_io work so we won't run the risk of | |
4267 | * converting any truncated blocks to initialized later | |
4268 | */ | |
c278531d | 4269 | ext4_flush_unwritten_io(inode); |
3889fd57 | 4270 | |
a86c6181 AT |
4271 | /* |
4272 | * probably first extent we're gonna free will be last in block | |
4273 | */ | |
f3bd1f3f | 4274 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 4275 | handle = ext4_journal_start(inode, err); |
cf108bca | 4276 | if (IS_ERR(handle)) |
a86c6181 | 4277 | return; |
a86c6181 | 4278 | |
189e868f AH |
4279 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
4280 | page_len = PAGE_CACHE_SIZE - | |
4281 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4282 | ||
4283 | err = ext4_discard_partial_page_buffers(handle, | |
4284 | mapping, inode->i_size, page_len, 0); | |
4285 | ||
4286 | if (err) | |
4287 | goto out_stop; | |
4288 | } | |
a86c6181 | 4289 | |
9ddfc3dc JK |
4290 | if (ext4_orphan_add(handle, inode)) |
4291 | goto out_stop; | |
4292 | ||
0e855ac8 | 4293 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
4294 | ext4_ext_invalidate_cache(inode); |
4295 | ||
c2ea3fde | 4296 | ext4_discard_preallocations(inode); |
c9de560d | 4297 | |
a86c6181 | 4298 | /* |
d0d856e8 RD |
4299 | * TODO: optimization is possible here. |
4300 | * Probably we need not scan at all, | |
4301 | * because page truncation is enough. | |
a86c6181 | 4302 | */ |
a86c6181 AT |
4303 | |
4304 | /* we have to know where to truncate from in crash case */ | |
4305 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4306 | ext4_mark_inode_dirty(handle, inode); | |
4307 | ||
4308 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4309 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
51865fda ZL |
4310 | err = ext4_es_remove_extent(inode, last_block, |
4311 | EXT_MAX_BLOCKS - last_block); | |
5f95d21f | 4312 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
a86c6181 AT |
4313 | |
4314 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
4315 | * transaction synchronous. |
4316 | */ | |
a86c6181 | 4317 | if (IS_SYNC(inode)) |
0390131b | 4318 | ext4_handle_sync(handle); |
a86c6181 | 4319 | |
9ddfc3dc | 4320 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
4321 | |
4322 | out_stop: | |
a86c6181 | 4323 | /* |
d0d856e8 | 4324 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
4325 | * then we need to clear up the orphan record which we created above. |
4326 | * However, if this was a real unlink then we were called by | |
4327 | * ext4_delete_inode(), and we allow that function to clean up the | |
4328 | * orphan info for us. | |
4329 | */ | |
4330 | if (inode->i_nlink) | |
4331 | ext4_orphan_del(handle, inode); | |
4332 | ||
ef737728 SR |
4333 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4334 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
4335 | ext4_journal_stop(handle); |
4336 | } | |
4337 | ||
fd28784a AK |
4338 | static void ext4_falloc_update_inode(struct inode *inode, |
4339 | int mode, loff_t new_size, int update_ctime) | |
4340 | { | |
4341 | struct timespec now; | |
4342 | ||
4343 | if (update_ctime) { | |
4344 | now = current_fs_time(inode->i_sb); | |
4345 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4346 | inode->i_ctime = now; | |
4347 | } | |
4348 | /* | |
4349 | * Update only when preallocation was requested beyond | |
4350 | * the file size. | |
4351 | */ | |
cf17fea6 AK |
4352 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4353 | if (new_size > i_size_read(inode)) | |
4354 | i_size_write(inode, new_size); | |
4355 | if (new_size > EXT4_I(inode)->i_disksize) | |
4356 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4357 | } else { |
4358 | /* | |
4359 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4360 | * can proceed even if the new size is the same as i_size. | |
4361 | */ | |
4362 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4363 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4364 | } |
4365 | ||
4366 | } | |
4367 | ||
a2df2a63 | 4368 | /* |
2fe17c10 | 4369 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4370 | * operation, which gets called from sys_fallocate system call. |
4371 | * For block-mapped files, posix_fallocate should fall back to the method | |
4372 | * of writing zeroes to the required new blocks (the same behavior which is | |
4373 | * expected for file systems which do not support fallocate() system call). | |
4374 | */ | |
2fe17c10 | 4375 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4376 | { |
2fe17c10 | 4377 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 4378 | handle_t *handle; |
fd28784a | 4379 | loff_t new_size; |
498e5f24 | 4380 | unsigned int max_blocks; |
a2df2a63 AA |
4381 | int ret = 0; |
4382 | int ret2 = 0; | |
4383 | int retries = 0; | |
a4e5d88b | 4384 | int flags; |
2ed88685 | 4385 | struct ext4_map_blocks map; |
a2df2a63 AA |
4386 | unsigned int credits, blkbits = inode->i_blkbits; |
4387 | ||
4388 | /* | |
4389 | * currently supporting (pre)allocate mode for extent-based | |
4390 | * files _only_ | |
4391 | */ | |
12e9b892 | 4392 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
a2df2a63 AA |
4393 | return -EOPNOTSUPP; |
4394 | ||
a4bb6b64 AH |
4395 | /* Return error if mode is not supported */ |
4396 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4397 | return -EOPNOTSUPP; | |
4398 | ||
4399 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4400 | return ext4_punch_hole(file, offset, len); | |
4401 | ||
0562e0ba | 4402 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4403 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4404 | /* |
4405 | * We can't just convert len to max_blocks because | |
4406 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4407 | */ | |
a2df2a63 | 4408 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4409 | - map.m_lblk; |
a2df2a63 | 4410 | /* |
f3bd1f3f | 4411 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4412 | */ |
f3bd1f3f | 4413 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4414 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4415 | ret = inode_newsize_ok(inode, (len + offset)); |
4416 | if (ret) { | |
4417 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4418 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4419 | return ret; |
4420 | } | |
3c6fe770 | 4421 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4422 | if (mode & FALLOC_FL_KEEP_SIZE) |
4423 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4424 | /* |
4425 | * Don't normalize the request if it can fit in one extent so | |
4426 | * that it doesn't get unnecessarily split into multiple | |
4427 | * extents. | |
4428 | */ | |
4429 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4430 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f DM |
4431 | |
4432 | /* Prevent race condition between unwritten */ | |
4433 | ext4_flush_unwritten_io(inode); | |
a2df2a63 AA |
4434 | retry: |
4435 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4436 | map.m_lblk = map.m_lblk + ret; |
4437 | map.m_len = max_blocks = max_blocks - ret; | |
a2df2a63 AA |
4438 | handle = ext4_journal_start(inode, credits); |
4439 | if (IS_ERR(handle)) { | |
4440 | ret = PTR_ERR(handle); | |
4441 | break; | |
4442 | } | |
a4e5d88b | 4443 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4444 | if (ret <= 0) { |
2c98615d AK |
4445 | #ifdef EXT4FS_DEBUG |
4446 | WARN_ON(ret <= 0); | |
e35fd660 | 4447 | printk(KERN_ERR "%s: ext4_ext_map_blocks " |
2c98615d | 4448 | "returned error inode#%lu, block=%u, " |
9fd9784c | 4449 | "max_blocks=%u", __func__, |
a6371b63 | 4450 | inode->i_ino, map.m_lblk, max_blocks); |
2c98615d | 4451 | #endif |
a2df2a63 AA |
4452 | ext4_mark_inode_dirty(handle, inode); |
4453 | ret2 = ext4_journal_stop(handle); | |
4454 | break; | |
4455 | } | |
2ed88685 | 4456 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4457 | blkbits) >> blkbits)) |
4458 | new_size = offset + len; | |
4459 | else | |
29ae07b7 | 4460 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4461 | |
fd28784a | 4462 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4463 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4464 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4465 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4466 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4467 | ret2 = ext4_journal_stop(handle); |
4468 | if (ret2) | |
4469 | break; | |
4470 | } | |
fd28784a AK |
4471 | if (ret == -ENOSPC && |
4472 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4473 | ret = 0; | |
a2df2a63 | 4474 | goto retry; |
a2df2a63 | 4475 | } |
55bd725a | 4476 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4477 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4478 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4479 | return ret > 0 ? ret2 : ret; |
4480 | } | |
6873fa0d | 4481 | |
0031462b MC |
4482 | /* |
4483 | * This function convert a range of blocks to written extents | |
4484 | * The caller of this function will pass the start offset and the size. | |
4485 | * all unwritten extents within this range will be converted to | |
4486 | * written extents. | |
4487 | * | |
4488 | * This function is called from the direct IO end io call back | |
4489 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4490 | * Returns 0 on success. |
0031462b MC |
4491 | */ |
4492 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4493 | ssize_t len) |
0031462b MC |
4494 | { |
4495 | handle_t *handle; | |
0031462b MC |
4496 | unsigned int max_blocks; |
4497 | int ret = 0; | |
4498 | int ret2 = 0; | |
2ed88685 | 4499 | struct ext4_map_blocks map; |
0031462b MC |
4500 | unsigned int credits, blkbits = inode->i_blkbits; |
4501 | ||
2ed88685 | 4502 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4503 | /* |
4504 | * We can't just convert len to max_blocks because | |
4505 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4506 | */ | |
2ed88685 TT |
4507 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4508 | map.m_lblk); | |
0031462b MC |
4509 | /* |
4510 | * credits to insert 1 extent into extent tree | |
4511 | */ | |
4512 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4513 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4514 | map.m_lblk += ret; |
4515 | map.m_len = (max_blocks -= ret); | |
0031462b MC |
4516 | handle = ext4_journal_start(inode, credits); |
4517 | if (IS_ERR(handle)) { | |
4518 | ret = PTR_ERR(handle); | |
4519 | break; | |
4520 | } | |
2ed88685 | 4521 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4522 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
0031462b MC |
4523 | if (ret <= 0) { |
4524 | WARN_ON(ret <= 0); | |
92b97816 TT |
4525 | ext4_msg(inode->i_sb, KERN_ERR, |
4526 | "%s:%d: inode #%lu: block %u: len %u: " | |
4527 | "ext4_ext_map_blocks returned %d", | |
4528 | __func__, __LINE__, inode->i_ino, map.m_lblk, | |
4529 | map.m_len, ret); | |
0031462b MC |
4530 | } |
4531 | ext4_mark_inode_dirty(handle, inode); | |
4532 | ret2 = ext4_journal_stop(handle); | |
4533 | if (ret <= 0 || ret2 ) | |
4534 | break; | |
4535 | } | |
4536 | return ret > 0 ? ret2 : ret; | |
4537 | } | |
6d9c85eb | 4538 | |
6873fa0d | 4539 | /* |
91dd8c11 LC |
4540 | * If newex is not existing extent (newex->ec_start equals zero) find |
4541 | * delayed extent at start of newex and update newex accordingly and | |
4542 | * return start of the next delayed extent. | |
4543 | * | |
4544 | * If newex is existing extent (newex->ec_start is not equal zero) | |
4545 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed | |
4546 | * extent found. Leave newex unmodified. | |
6873fa0d | 4547 | */ |
91dd8c11 LC |
4548 | static int ext4_find_delayed_extent(struct inode *inode, |
4549 | struct ext4_ext_cache *newex) | |
6873fa0d | 4550 | { |
b3aff3e3 | 4551 | struct extent_status es; |
b3aff3e3 | 4552 | ext4_lblk_t next_del; |
6873fa0d | 4553 | |
b3aff3e3 ZL |
4554 | es.start = newex->ec_block; |
4555 | next_del = ext4_es_find_extent(inode, &es); | |
6873fa0d | 4556 | |
b05e6ae5 | 4557 | if (newex->ec_start == 0) { |
6d9c85eb YY |
4558 | /* |
4559 | * No extent in extent-tree contains block @newex->ec_start, | |
4560 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
6d9c85eb | 4561 | */ |
b3aff3e3 ZL |
4562 | if (es.len == 0) |
4563 | /* A hole found. */ | |
91dd8c11 | 4564 | return 0; |
b3aff3e3 ZL |
4565 | |
4566 | if (es.start > newex->ec_block) { | |
4567 | /* A hole found. */ | |
4568 | newex->ec_len = min(es.start - newex->ec_block, | |
4569 | newex->ec_len); | |
91dd8c11 | 4570 | return 0; |
6873fa0d | 4571 | } |
6d9c85eb | 4572 | |
b3aff3e3 | 4573 | newex->ec_len = es.start + es.len - newex->ec_block; |
6873fa0d ES |
4574 | } |
4575 | ||
91dd8c11 | 4576 | return next_del; |
6873fa0d | 4577 | } |
6873fa0d ES |
4578 | /* fiemap flags we can handle specified here */ |
4579 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4580 | ||
3a06d778 AK |
4581 | static int ext4_xattr_fiemap(struct inode *inode, |
4582 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4583 | { |
4584 | __u64 physical = 0; | |
4585 | __u64 length; | |
4586 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4587 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4588 | int error = 0; | |
4589 | ||
4590 | /* in-inode? */ | |
19f5fb7a | 4591 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4592 | struct ext4_iloc iloc; |
4593 | int offset; /* offset of xattr in inode */ | |
4594 | ||
4595 | error = ext4_get_inode_loc(inode, &iloc); | |
4596 | if (error) | |
4597 | return error; | |
4598 | physical = iloc.bh->b_blocknr << blockbits; | |
4599 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4600 | EXT4_I(inode)->i_extra_isize; | |
4601 | physical += offset; | |
4602 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4603 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4604 | brelse(iloc.bh); |
6873fa0d ES |
4605 | } else { /* external block */ |
4606 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4607 | length = inode->i_sb->s_blocksize; | |
4608 | } | |
4609 | ||
4610 | if (physical) | |
4611 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4612 | length, flags); | |
4613 | return (error < 0 ? error : 0); | |
4614 | } | |
4615 | ||
a4bb6b64 AH |
4616 | /* |
4617 | * ext4_ext_punch_hole | |
4618 | * | |
4619 | * Punches a hole of "length" bytes in a file starting | |
4620 | * at byte "offset" | |
4621 | * | |
4622 | * @inode: The inode of the file to punch a hole in | |
4623 | * @offset: The starting byte offset of the hole | |
4624 | * @length: The length of the hole | |
4625 | * | |
4626 | * Returns the number of blocks removed or negative on err | |
4627 | */ | |
4628 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4629 | { | |
4630 | struct inode *inode = file->f_path.dentry->d_inode; | |
4631 | struct super_block *sb = inode->i_sb; | |
5f95d21f | 4632 | ext4_lblk_t first_block, stop_block; |
a4bb6b64 | 4633 | struct address_space *mapping = inode->i_mapping; |
a4bb6b64 | 4634 | handle_t *handle; |
ba06208a AH |
4635 | loff_t first_page, last_page, page_len; |
4636 | loff_t first_page_offset, last_page_offset; | |
5f95d21f | 4637 | int credits, err = 0; |
a4bb6b64 | 4638 | |
02d262df DM |
4639 | /* |
4640 | * Write out all dirty pages to avoid race conditions | |
4641 | * Then release them. | |
4642 | */ | |
4643 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4644 | err = filemap_write_and_wait_range(mapping, | |
4645 | offset, offset + length - 1); | |
4646 | ||
4647 | if (err) | |
4648 | return err; | |
4649 | } | |
4650 | ||
4651 | mutex_lock(&inode->i_mutex); | |
4652 | /* It's not possible punch hole on append only file */ | |
4653 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
4654 | err = -EPERM; | |
4655 | goto out_mutex; | |
4656 | } | |
4657 | if (IS_SWAPFILE(inode)) { | |
4658 | err = -ETXTBSY; | |
4659 | goto out_mutex; | |
4660 | } | |
4661 | ||
2be4751b AH |
4662 | /* No need to punch hole beyond i_size */ |
4663 | if (offset >= inode->i_size) | |
02d262df | 4664 | goto out_mutex; |
2be4751b AH |
4665 | |
4666 | /* | |
4667 | * If the hole extends beyond i_size, set the hole | |
4668 | * to end after the page that contains i_size | |
4669 | */ | |
4670 | if (offset + length > inode->i_size) { | |
4671 | length = inode->i_size + | |
4672 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4673 | offset; | |
4674 | } | |
4675 | ||
a4bb6b64 AH |
4676 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4677 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4678 | ||
4679 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4680 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4681 | ||
a4bb6b64 AH |
4682 | /* Now release the pages */ |
4683 | if (last_page_offset > first_page_offset) { | |
5e44f8c3 HD |
4684 | truncate_pagecache_range(inode, first_page_offset, |
4685 | last_page_offset - 1); | |
a4bb6b64 AH |
4686 | } |
4687 | ||
02d262df DM |
4688 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
4689 | ext4_inode_block_unlocked_dio(inode); | |
c278531d | 4690 | err = ext4_flush_unwritten_io(inode); |
28a535f9 | 4691 | if (err) |
02d262df | 4692 | goto out_dio; |
c278531d | 4693 | inode_dio_wait(inode); |
a4bb6b64 AH |
4694 | |
4695 | credits = ext4_writepage_trans_blocks(inode); | |
4696 | handle = ext4_journal_start(inode, credits); | |
02d262df DM |
4697 | if (IS_ERR(handle)) { |
4698 | err = PTR_ERR(handle); | |
4699 | goto out_dio; | |
4700 | } | |
a4bb6b64 | 4701 | |
a4bb6b64 AH |
4702 | |
4703 | /* | |
ba06208a AH |
4704 | * Now we need to zero out the non-page-aligned data in the |
4705 | * pages at the start and tail of the hole, and unmap the buffer | |
4706 | * heads for the block aligned regions of the page that were | |
4707 | * completely zeroed. | |
a4bb6b64 | 4708 | */ |
ba06208a AH |
4709 | if (first_page > last_page) { |
4710 | /* | |
4711 | * If the file space being truncated is contained within a page | |
4712 | * just zero out and unmap the middle of that page | |
4713 | */ | |
4714 | err = ext4_discard_partial_page_buffers(handle, | |
4715 | mapping, offset, length, 0); | |
4716 | ||
4717 | if (err) | |
4718 | goto out; | |
4719 | } else { | |
4720 | /* | |
4721 | * zero out and unmap the partial page that contains | |
4722 | * the start of the hole | |
4723 | */ | |
4724 | page_len = first_page_offset - offset; | |
4725 | if (page_len > 0) { | |
4726 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4727 | offset, page_len, 0); | |
4728 | if (err) | |
4729 | goto out; | |
4730 | } | |
4731 | ||
4732 | /* | |
4733 | * zero out and unmap the partial page that contains | |
4734 | * the end of the hole | |
4735 | */ | |
4736 | page_len = offset + length - last_page_offset; | |
4737 | if (page_len > 0) { | |
4738 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4739 | last_page_offset, page_len, 0); | |
4740 | if (err) | |
4741 | goto out; | |
a4bb6b64 AH |
4742 | } |
4743 | } | |
4744 | ||
2be4751b AH |
4745 | /* |
4746 | * If i_size is contained in the last page, we need to | |
4747 | * unmap and zero the partial page after i_size | |
4748 | */ | |
4749 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4750 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4751 | ||
4752 | page_len = PAGE_CACHE_SIZE - | |
4753 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4754 | ||
4755 | if (page_len > 0) { | |
4756 | err = ext4_discard_partial_page_buffers(handle, | |
4757 | mapping, inode->i_size, page_len, 0); | |
4758 | ||
4759 | if (err) | |
4760 | goto out; | |
4761 | } | |
4762 | } | |
4763 | ||
5f95d21f LC |
4764 | first_block = (offset + sb->s_blocksize - 1) >> |
4765 | EXT4_BLOCK_SIZE_BITS(sb); | |
4766 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4767 | ||
a4bb6b64 | 4768 | /* If there are no blocks to remove, return now */ |
5f95d21f | 4769 | if (first_block >= stop_block) |
a4bb6b64 AH |
4770 | goto out; |
4771 | ||
4772 | down_write(&EXT4_I(inode)->i_data_sem); | |
4773 | ext4_ext_invalidate_cache(inode); | |
4774 | ext4_discard_preallocations(inode); | |
4775 | ||
51865fda ZL |
4776 | err = ext4_es_remove_extent(inode, first_block, |
4777 | stop_block - first_block); | |
5f95d21f | 4778 | err = ext4_ext_remove_space(inode, first_block, stop_block - 1); |
a4bb6b64 | 4779 | |
5f95d21f LC |
4780 | ext4_ext_invalidate_cache(inode); |
4781 | ext4_discard_preallocations(inode); | |
a4bb6b64 AH |
4782 | |
4783 | if (IS_SYNC(inode)) | |
4784 | ext4_handle_sync(handle); | |
4785 | ||
4786 | up_write(&EXT4_I(inode)->i_data_sem); | |
4787 | ||
4788 | out: | |
a4bb6b64 AH |
4789 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4790 | ext4_mark_inode_dirty(handle, inode); | |
4791 | ext4_journal_stop(handle); | |
02d262df DM |
4792 | out_dio: |
4793 | ext4_inode_resume_unlocked_dio(inode); | |
4794 | out_mutex: | |
4795 | mutex_unlock(&inode->i_mutex); | |
a4bb6b64 AH |
4796 | return err; |
4797 | } | |
91dd8c11 | 4798 | |
6873fa0d ES |
4799 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4800 | __u64 start, __u64 len) | |
4801 | { | |
4802 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4803 | int error = 0; |
4804 | ||
4805 | /* fallback to generic here if not in extents fmt */ | |
12e9b892 | 4806 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4807 | return generic_block_fiemap(inode, fieinfo, start, len, |
4808 | ext4_get_block); | |
4809 | ||
4810 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4811 | return -EBADR; | |
4812 | ||
4813 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4814 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4815 | } else { | |
aca92ff6 LM |
4816 | ext4_lblk_t len_blks; |
4817 | __u64 last_blk; | |
4818 | ||
6873fa0d | 4819 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4820 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4821 | if (last_blk >= EXT_MAX_BLOCKS) |
4822 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4823 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4824 | |
4825 | /* | |
91dd8c11 LC |
4826 | * Walk the extent tree gathering extent information |
4827 | * and pushing extents back to the user. | |
6873fa0d | 4828 | */ |
91dd8c11 LC |
4829 | error = ext4_fill_fiemap_extents(inode, start_blk, |
4830 | len_blks, fieinfo); | |
6873fa0d ES |
4831 | } |
4832 | ||
4833 | return error; | |
4834 | } |