]>
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; | |
860d21e2 | 728 | int ret; |
a86c6181 AT |
729 | |
730 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 731 | depth = ext_depth(inode); |
a86c6181 AT |
732 | |
733 | /* account possible depth increase */ | |
734 | if (!path) { | |
5d4958f9 | 735 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
736 | GFP_NOFS); |
737 | if (!path) | |
738 | return ERR_PTR(-ENOMEM); | |
739 | alloc = 1; | |
740 | } | |
a86c6181 | 741 | path[0].p_hdr = eh; |
1973adcb | 742 | path[0].p_bh = NULL; |
a86c6181 | 743 | |
c29c0ae7 | 744 | i = depth; |
a86c6181 AT |
745 | /* walk through the tree */ |
746 | while (i) { | |
747 | ext_debug("depth %d: num %d, max %d\n", | |
748 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 749 | |
a86c6181 | 750 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 751 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
752 | path[ppos].p_depth = i; |
753 | path[ppos].p_ext = NULL; | |
754 | ||
7a262f7c | 755 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
860d21e2 TT |
756 | if (unlikely(!bh)) { |
757 | ret = -ENOMEM; | |
a86c6181 | 758 | goto err; |
860d21e2 | 759 | } |
7a262f7c | 760 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
761 | trace_ext4_ext_load_extent(inode, block, |
762 | path[ppos].p_block); | |
860d21e2 TT |
763 | ret = bh_submit_read(bh); |
764 | if (ret < 0) { | |
7a262f7c AK |
765 | put_bh(bh); |
766 | goto err; | |
767 | } | |
7a262f7c | 768 | } |
a86c6181 AT |
769 | eh = ext_block_hdr(bh); |
770 | ppos++; | |
273df556 FM |
771 | if (unlikely(ppos > depth)) { |
772 | put_bh(bh); | |
773 | EXT4_ERROR_INODE(inode, | |
774 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 775 | ret = -EIO; |
273df556 FM |
776 | goto err; |
777 | } | |
a86c6181 AT |
778 | path[ppos].p_bh = bh; |
779 | path[ppos].p_hdr = eh; | |
780 | i--; | |
781 | ||
860d21e2 TT |
782 | ret = ext4_ext_check_block(inode, eh, i, bh); |
783 | if (ret < 0) | |
a86c6181 AT |
784 | goto err; |
785 | } | |
786 | ||
787 | path[ppos].p_depth = i; | |
a86c6181 AT |
788 | path[ppos].p_ext = NULL; |
789 | path[ppos].p_idx = NULL; | |
790 | ||
a86c6181 AT |
791 | /* find extent */ |
792 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
793 | /* if not an empty leaf */ |
794 | if (path[ppos].p_ext) | |
bf89d16f | 795 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
796 | |
797 | ext4_ext_show_path(inode, path); | |
798 | ||
799 | return path; | |
800 | ||
801 | err: | |
802 | ext4_ext_drop_refs(path); | |
803 | if (alloc) | |
804 | kfree(path); | |
860d21e2 | 805 | return ERR_PTR(ret); |
a86c6181 AT |
806 | } |
807 | ||
808 | /* | |
d0d856e8 RD |
809 | * ext4_ext_insert_index: |
810 | * insert new index [@logical;@ptr] into the block at @curp; | |
811 | * check where to insert: before @curp or after @curp | |
a86c6181 | 812 | */ |
1f109d5a TT |
813 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
814 | struct ext4_ext_path *curp, | |
815 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
816 | { |
817 | struct ext4_extent_idx *ix; | |
818 | int len, err; | |
819 | ||
7e028976 AM |
820 | err = ext4_ext_get_access(handle, inode, curp); |
821 | if (err) | |
a86c6181 AT |
822 | return err; |
823 | ||
273df556 FM |
824 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
825 | EXT4_ERROR_INODE(inode, | |
826 | "logical %d == ei_block %d!", | |
827 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
828 | return -EIO; | |
829 | } | |
d4620315 RD |
830 | |
831 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
832 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
833 | EXT4_ERROR_INODE(inode, | |
834 | "eh_entries %d >= eh_max %d!", | |
835 | le16_to_cpu(curp->p_hdr->eh_entries), | |
836 | le16_to_cpu(curp->p_hdr->eh_max)); | |
837 | return -EIO; | |
838 | } | |
839 | ||
a86c6181 AT |
840 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
841 | /* insert after */ | |
80e675f9 | 842 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
843 | ix = curp->p_idx + 1; |
844 | } else { | |
845 | /* insert before */ | |
80e675f9 | 846 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
847 | ix = curp->p_idx; |
848 | } | |
849 | ||
80e675f9 EG |
850 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
851 | BUG_ON(len < 0); | |
852 | if (len > 0) { | |
853 | ext_debug("insert new index %d: " | |
854 | "move %d indices from 0x%p to 0x%p\n", | |
855 | logical, len, ix, ix + 1); | |
856 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
857 | } | |
858 | ||
f472e026 TM |
859 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
860 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
861 | return -EIO; | |
862 | } | |
863 | ||
a86c6181 | 864 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 865 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 866 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 867 | |
273df556 FM |
868 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
869 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
870 | return -EIO; | |
871 | } | |
a86c6181 AT |
872 | |
873 | err = ext4_ext_dirty(handle, inode, curp); | |
874 | ext4_std_error(inode->i_sb, err); | |
875 | ||
876 | return err; | |
877 | } | |
878 | ||
879 | /* | |
d0d856e8 RD |
880 | * ext4_ext_split: |
881 | * inserts new subtree into the path, using free index entry | |
882 | * at depth @at: | |
883 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
884 | * - makes decision where to split | |
885 | * - moves remaining extents and index entries (right to the split point) | |
886 | * into the newly allocated blocks | |
887 | * - initializes subtree | |
a86c6181 AT |
888 | */ |
889 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
890 | unsigned int flags, |
891 | struct ext4_ext_path *path, | |
892 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
893 | { |
894 | struct buffer_head *bh = NULL; | |
895 | int depth = ext_depth(inode); | |
896 | struct ext4_extent_header *neh; | |
897 | struct ext4_extent_idx *fidx; | |
a86c6181 | 898 | int i = at, k, m, a; |
f65e6fba | 899 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 900 | __le32 border; |
f65e6fba | 901 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
902 | int err = 0; |
903 | ||
904 | /* make decision: where to split? */ | |
d0d856e8 | 905 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 906 | |
d0d856e8 | 907 | /* if current leaf will be split, then we should use |
a86c6181 | 908 | * border from split point */ |
273df556 FM |
909 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
910 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
911 | return -EIO; | |
912 | } | |
a86c6181 AT |
913 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
914 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 915 | ext_debug("leaf will be split." |
a86c6181 | 916 | " next leaf starts at %d\n", |
8c55e204 | 917 | le32_to_cpu(border)); |
a86c6181 AT |
918 | } else { |
919 | border = newext->ee_block; | |
920 | ext_debug("leaf will be added." | |
921 | " next leaf starts at %d\n", | |
8c55e204 | 922 | le32_to_cpu(border)); |
a86c6181 AT |
923 | } |
924 | ||
925 | /* | |
d0d856e8 RD |
926 | * If error occurs, then we break processing |
927 | * and mark filesystem read-only. index won't | |
a86c6181 | 928 | * be inserted and tree will be in consistent |
d0d856e8 | 929 | * state. Next mount will repair buffers too. |
a86c6181 AT |
930 | */ |
931 | ||
932 | /* | |
d0d856e8 RD |
933 | * Get array to track all allocated blocks. |
934 | * We need this to handle errors and free blocks | |
935 | * upon them. | |
a86c6181 | 936 | */ |
5d4958f9 | 937 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
938 | if (!ablocks) |
939 | return -ENOMEM; | |
a86c6181 AT |
940 | |
941 | /* allocate all needed blocks */ | |
942 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
943 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 944 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 945 | newext, &err, flags); |
a86c6181 AT |
946 | if (newblock == 0) |
947 | goto cleanup; | |
948 | ablocks[a] = newblock; | |
949 | } | |
950 | ||
951 | /* initialize new leaf */ | |
952 | newblock = ablocks[--a]; | |
273df556 FM |
953 | if (unlikely(newblock == 0)) { |
954 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
955 | err = -EIO; | |
956 | goto cleanup; | |
957 | } | |
a86c6181 | 958 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 959 | if (unlikely(!bh)) { |
860d21e2 | 960 | err = -ENOMEM; |
a86c6181 AT |
961 | goto cleanup; |
962 | } | |
963 | lock_buffer(bh); | |
964 | ||
7e028976 AM |
965 | err = ext4_journal_get_create_access(handle, bh); |
966 | if (err) | |
a86c6181 AT |
967 | goto cleanup; |
968 | ||
969 | neh = ext_block_hdr(bh); | |
970 | neh->eh_entries = 0; | |
55ad63bf | 971 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
972 | neh->eh_magic = EXT4_EXT_MAGIC; |
973 | neh->eh_depth = 0; | |
a86c6181 | 974 | |
d0d856e8 | 975 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
976 | if (unlikely(path[depth].p_hdr->eh_entries != |
977 | path[depth].p_hdr->eh_max)) { | |
978 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
979 | path[depth].p_hdr->eh_entries, | |
980 | path[depth].p_hdr->eh_max); | |
981 | err = -EIO; | |
982 | goto cleanup; | |
983 | } | |
a86c6181 | 984 | /* start copy from next extent */ |
1b16da77 YY |
985 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
986 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 987 | if (m) { |
1b16da77 YY |
988 | struct ext4_extent *ex; |
989 | ex = EXT_FIRST_EXTENT(neh); | |
990 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 991 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
992 | } |
993 | ||
7ac5990d | 994 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
995 | set_buffer_uptodate(bh); |
996 | unlock_buffer(bh); | |
997 | ||
0390131b | 998 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 999 | if (err) |
a86c6181 AT |
1000 | goto cleanup; |
1001 | brelse(bh); | |
1002 | bh = NULL; | |
1003 | ||
1004 | /* correct old leaf */ | |
1005 | if (m) { | |
7e028976 AM |
1006 | err = ext4_ext_get_access(handle, inode, path + depth); |
1007 | if (err) | |
a86c6181 | 1008 | goto cleanup; |
e8546d06 | 1009 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1010 | err = ext4_ext_dirty(handle, inode, path + depth); |
1011 | if (err) | |
a86c6181 AT |
1012 | goto cleanup; |
1013 | ||
1014 | } | |
1015 | ||
1016 | /* create intermediate indexes */ | |
1017 | k = depth - at - 1; | |
273df556 FM |
1018 | if (unlikely(k < 0)) { |
1019 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1020 | err = -EIO; | |
1021 | goto cleanup; | |
1022 | } | |
a86c6181 AT |
1023 | if (k) |
1024 | ext_debug("create %d intermediate indices\n", k); | |
1025 | /* insert new index into current index block */ | |
1026 | /* current depth stored in i var */ | |
1027 | i = depth - 1; | |
1028 | while (k--) { | |
1029 | oldblock = newblock; | |
1030 | newblock = ablocks[--a]; | |
bba90743 | 1031 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1032 | if (unlikely(!bh)) { |
860d21e2 | 1033 | err = -ENOMEM; |
a86c6181 AT |
1034 | goto cleanup; |
1035 | } | |
1036 | lock_buffer(bh); | |
1037 | ||
7e028976 AM |
1038 | err = ext4_journal_get_create_access(handle, bh); |
1039 | if (err) | |
a86c6181 AT |
1040 | goto cleanup; |
1041 | ||
1042 | neh = ext_block_hdr(bh); | |
1043 | neh->eh_entries = cpu_to_le16(1); | |
1044 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1045 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1046 | neh->eh_depth = cpu_to_le16(depth - i); |
1047 | fidx = EXT_FIRST_INDEX(neh); | |
1048 | fidx->ei_block = border; | |
f65e6fba | 1049 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1050 | |
bba90743 ES |
1051 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1052 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1053 | |
1b16da77 | 1054 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1055 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1056 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1057 | EXT4_ERROR_INODE(inode, | |
1058 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1059 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1060 | err = -EIO; | |
1061 | goto cleanup; | |
1062 | } | |
1b16da77 YY |
1063 | /* start copy indexes */ |
1064 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1065 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1066 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1067 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1068 | if (m) { |
1b16da77 | 1069 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1070 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1071 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1072 | } |
7ac5990d | 1073 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1074 | set_buffer_uptodate(bh); |
1075 | unlock_buffer(bh); | |
1076 | ||
0390131b | 1077 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1078 | if (err) |
a86c6181 AT |
1079 | goto cleanup; |
1080 | brelse(bh); | |
1081 | bh = NULL; | |
1082 | ||
1083 | /* correct old index */ | |
1084 | if (m) { | |
1085 | err = ext4_ext_get_access(handle, inode, path + i); | |
1086 | if (err) | |
1087 | goto cleanup; | |
e8546d06 | 1088 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1089 | err = ext4_ext_dirty(handle, inode, path + i); |
1090 | if (err) | |
1091 | goto cleanup; | |
1092 | } | |
1093 | ||
1094 | i--; | |
1095 | } | |
1096 | ||
1097 | /* insert new index */ | |
a86c6181 AT |
1098 | err = ext4_ext_insert_index(handle, inode, path + at, |
1099 | le32_to_cpu(border), newblock); | |
1100 | ||
1101 | cleanup: | |
1102 | if (bh) { | |
1103 | if (buffer_locked(bh)) | |
1104 | unlock_buffer(bh); | |
1105 | brelse(bh); | |
1106 | } | |
1107 | ||
1108 | if (err) { | |
1109 | /* free all allocated blocks in error case */ | |
1110 | for (i = 0; i < depth; i++) { | |
1111 | if (!ablocks[i]) | |
1112 | continue; | |
7dc57615 | 1113 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1114 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1115 | } |
1116 | } | |
1117 | kfree(ablocks); | |
1118 | ||
1119 | return err; | |
1120 | } | |
1121 | ||
1122 | /* | |
d0d856e8 RD |
1123 | * ext4_ext_grow_indepth: |
1124 | * implements tree growing procedure: | |
1125 | * - allocates new block | |
1126 | * - moves top-level data (index block or leaf) into the new block | |
1127 | * - initializes new top-level, creating index that points to the | |
1128 | * just created block | |
a86c6181 AT |
1129 | */ |
1130 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1131 | unsigned int flags, |
55f020db | 1132 | struct ext4_extent *newext) |
a86c6181 | 1133 | { |
a86c6181 | 1134 | struct ext4_extent_header *neh; |
a86c6181 | 1135 | struct buffer_head *bh; |
f65e6fba | 1136 | ext4_fsblk_t newblock; |
a86c6181 AT |
1137 | int err = 0; |
1138 | ||
1939dd84 | 1139 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1140 | newext, &err, flags); |
a86c6181 AT |
1141 | if (newblock == 0) |
1142 | return err; | |
1143 | ||
1144 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1145 | if (unlikely(!bh)) |
860d21e2 | 1146 | return -ENOMEM; |
a86c6181 AT |
1147 | lock_buffer(bh); |
1148 | ||
7e028976 AM |
1149 | err = ext4_journal_get_create_access(handle, bh); |
1150 | if (err) { | |
a86c6181 AT |
1151 | unlock_buffer(bh); |
1152 | goto out; | |
1153 | } | |
1154 | ||
1155 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1156 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1157 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1158 | |
1159 | /* set size of new block */ | |
1160 | neh = ext_block_hdr(bh); | |
1161 | /* old root could have indexes or leaves | |
1162 | * so calculate e_max right way */ | |
1163 | if (ext_depth(inode)) | |
55ad63bf | 1164 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1165 | else |
55ad63bf | 1166 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1167 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1168 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1169 | set_buffer_uptodate(bh); |
1170 | unlock_buffer(bh); | |
1171 | ||
0390131b | 1172 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1173 | if (err) |
a86c6181 AT |
1174 | goto out; |
1175 | ||
1939dd84 | 1176 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1177 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1178 | neh->eh_entries = cpu_to_le16(1); |
1179 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1180 | if (neh->eh_depth == 0) { | |
1181 | /* Root extent block becomes index block */ | |
1182 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1183 | EXT_FIRST_INDEX(neh)->ei_block = | |
1184 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1185 | } | |
2ae02107 | 1186 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1187 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1188 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1189 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1190 | |
ba39ebb6 | 1191 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1192 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1193 | out: |
1194 | brelse(bh); | |
1195 | ||
1196 | return err; | |
1197 | } | |
1198 | ||
1199 | /* | |
d0d856e8 RD |
1200 | * ext4_ext_create_new_leaf: |
1201 | * finds empty index and adds new leaf. | |
1202 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1203 | */ |
1204 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1205 | unsigned int flags, |
1206 | struct ext4_ext_path *path, | |
1207 | struct ext4_extent *newext) | |
a86c6181 AT |
1208 | { |
1209 | struct ext4_ext_path *curp; | |
1210 | int depth, i, err = 0; | |
1211 | ||
1212 | repeat: | |
1213 | i = depth = ext_depth(inode); | |
1214 | ||
1215 | /* walk up to the tree and look for free index entry */ | |
1216 | curp = path + depth; | |
1217 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1218 | i--; | |
1219 | curp--; | |
1220 | } | |
1221 | ||
d0d856e8 RD |
1222 | /* we use already allocated block for index block, |
1223 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1224 | if (EXT_HAS_FREE_INDEX(curp)) { |
1225 | /* if we found index with free entry, then use that | |
1226 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1227 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1228 | if (err) |
1229 | goto out; | |
a86c6181 AT |
1230 | |
1231 | /* refill path */ | |
1232 | ext4_ext_drop_refs(path); | |
1233 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1234 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1235 | path); | |
a86c6181 AT |
1236 | if (IS_ERR(path)) |
1237 | err = PTR_ERR(path); | |
1238 | } else { | |
1239 | /* tree is full, time to grow in depth */ | |
1939dd84 | 1240 | err = ext4_ext_grow_indepth(handle, inode, flags, newext); |
a86c6181 AT |
1241 | if (err) |
1242 | goto out; | |
1243 | ||
1244 | /* refill path */ | |
1245 | ext4_ext_drop_refs(path); | |
1246 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1247 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1248 | path); | |
a86c6181 AT |
1249 | if (IS_ERR(path)) { |
1250 | err = PTR_ERR(path); | |
1251 | goto out; | |
1252 | } | |
1253 | ||
1254 | /* | |
d0d856e8 RD |
1255 | * only first (depth 0 -> 1) produces free space; |
1256 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1257 | */ |
1258 | depth = ext_depth(inode); | |
1259 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1260 | /* now we need to split */ |
a86c6181 AT |
1261 | goto repeat; |
1262 | } | |
1263 | } | |
1264 | ||
1265 | out: | |
1266 | return err; | |
1267 | } | |
1268 | ||
1988b51e AT |
1269 | /* |
1270 | * search the closest allocated block to the left for *logical | |
1271 | * and returns it at @logical + it's physical address at @phys | |
1272 | * if *logical is the smallest allocated block, the function | |
1273 | * returns 0 at @phys | |
1274 | * return value contains 0 (success) or error code | |
1275 | */ | |
1f109d5a TT |
1276 | static int ext4_ext_search_left(struct inode *inode, |
1277 | struct ext4_ext_path *path, | |
1278 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1279 | { |
1280 | struct ext4_extent_idx *ix; | |
1281 | struct ext4_extent *ex; | |
b939e376 | 1282 | int depth, ee_len; |
1988b51e | 1283 | |
273df556 FM |
1284 | if (unlikely(path == NULL)) { |
1285 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1286 | return -EIO; | |
1287 | } | |
1988b51e AT |
1288 | depth = path->p_depth; |
1289 | *phys = 0; | |
1290 | ||
1291 | if (depth == 0 && path->p_ext == NULL) | |
1292 | return 0; | |
1293 | ||
1294 | /* usually extent in the path covers blocks smaller | |
1295 | * then *logical, but it can be that extent is the | |
1296 | * first one in the file */ | |
1297 | ||
1298 | ex = path[depth].p_ext; | |
b939e376 | 1299 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1300 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1301 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1302 | EXT4_ERROR_INODE(inode, | |
1303 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1304 | *logical, le32_to_cpu(ex->ee_block)); | |
1305 | return -EIO; | |
1306 | } | |
1988b51e AT |
1307 | while (--depth >= 0) { |
1308 | ix = path[depth].p_idx; | |
273df556 FM |
1309 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1310 | EXT4_ERROR_INODE(inode, | |
1311 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1312 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1313 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1314 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1315 | depth); |
1316 | return -EIO; | |
1317 | } | |
1988b51e AT |
1318 | } |
1319 | return 0; | |
1320 | } | |
1321 | ||
273df556 FM |
1322 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1323 | EXT4_ERROR_INODE(inode, | |
1324 | "logical %d < ee_block %d + ee_len %d!", | |
1325 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1326 | return -EIO; | |
1327 | } | |
1988b51e | 1328 | |
b939e376 | 1329 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1330 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1331 | return 0; |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * search the closest allocated block to the right for *logical | |
1336 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1337 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1338 | * returns 0 at @phys |
1339 | * return value contains 0 (success) or error code | |
1340 | */ | |
1f109d5a TT |
1341 | static int ext4_ext_search_right(struct inode *inode, |
1342 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1343 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1344 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1345 | { |
1346 | struct buffer_head *bh = NULL; | |
1347 | struct ext4_extent_header *eh; | |
1348 | struct ext4_extent_idx *ix; | |
1349 | struct ext4_extent *ex; | |
1350 | ext4_fsblk_t block; | |
395a87bf ES |
1351 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1352 | int ee_len; | |
1988b51e | 1353 | |
273df556 FM |
1354 | if (unlikely(path == NULL)) { |
1355 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1356 | return -EIO; | |
1357 | } | |
1988b51e AT |
1358 | depth = path->p_depth; |
1359 | *phys = 0; | |
1360 | ||
1361 | if (depth == 0 && path->p_ext == NULL) | |
1362 | return 0; | |
1363 | ||
1364 | /* usually extent in the path covers blocks smaller | |
1365 | * then *logical, but it can be that extent is the | |
1366 | * first one in the file */ | |
1367 | ||
1368 | ex = path[depth].p_ext; | |
b939e376 | 1369 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1370 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1371 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1372 | EXT4_ERROR_INODE(inode, | |
1373 | "first_extent(path[%d].p_hdr) != ex", | |
1374 | depth); | |
1375 | return -EIO; | |
1376 | } | |
1988b51e AT |
1377 | while (--depth >= 0) { |
1378 | ix = path[depth].p_idx; | |
273df556 FM |
1379 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1380 | EXT4_ERROR_INODE(inode, | |
1381 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1382 | *logical); | |
1383 | return -EIO; | |
1384 | } | |
1988b51e | 1385 | } |
4d33b1ef | 1386 | goto found_extent; |
1988b51e AT |
1387 | } |
1388 | ||
273df556 FM |
1389 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1390 | EXT4_ERROR_INODE(inode, | |
1391 | "logical %d < ee_block %d + ee_len %d!", | |
1392 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1393 | return -EIO; | |
1394 | } | |
1988b51e AT |
1395 | |
1396 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1397 | /* next allocated block in this leaf */ | |
1398 | ex++; | |
4d33b1ef | 1399 | goto found_extent; |
1988b51e AT |
1400 | } |
1401 | ||
1402 | /* go up and search for index to the right */ | |
1403 | while (--depth >= 0) { | |
1404 | ix = path[depth].p_idx; | |
1405 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1406 | goto got_index; |
1988b51e AT |
1407 | } |
1408 | ||
25f1ee3a WF |
1409 | /* we've gone up to the root and found no index to the right */ |
1410 | return 0; | |
1988b51e | 1411 | |
25f1ee3a | 1412 | got_index: |
1988b51e AT |
1413 | /* we've found index to the right, let's |
1414 | * follow it and find the closest allocated | |
1415 | * block to the right */ | |
1416 | ix++; | |
bf89d16f | 1417 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1418 | while (++depth < path->p_depth) { |
1419 | bh = sb_bread(inode->i_sb, block); | |
1420 | if (bh == NULL) | |
1421 | return -EIO; | |
1422 | eh = ext_block_hdr(bh); | |
395a87bf | 1423 | /* subtract from p_depth to get proper eh_depth */ |
f8489128 DW |
1424 | if (ext4_ext_check_block(inode, eh, |
1425 | path->p_depth - depth, bh)) { | |
1988b51e AT |
1426 | put_bh(bh); |
1427 | return -EIO; | |
1428 | } | |
1429 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1430 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1431 | put_bh(bh); |
1432 | } | |
1433 | ||
1434 | bh = sb_bread(inode->i_sb, block); | |
1435 | if (bh == NULL) | |
1436 | return -EIO; | |
1437 | eh = ext_block_hdr(bh); | |
f8489128 | 1438 | if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) { |
1988b51e AT |
1439 | put_bh(bh); |
1440 | return -EIO; | |
1441 | } | |
1442 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1443 | found_extent: |
1988b51e | 1444 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1445 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1446 | *ret_ex = ex; |
1447 | if (bh) | |
1448 | put_bh(bh); | |
1988b51e | 1449 | return 0; |
1988b51e AT |
1450 | } |
1451 | ||
a86c6181 | 1452 | /* |
d0d856e8 | 1453 | * ext4_ext_next_allocated_block: |
f17722f9 | 1454 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1455 | * NOTE: it considers block number from index entry as |
1456 | * allocated block. Thus, index entries have to be consistent | |
1457 | * with leaves. | |
a86c6181 | 1458 | */ |
725d26d3 | 1459 | static ext4_lblk_t |
a86c6181 AT |
1460 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1461 | { | |
1462 | int depth; | |
1463 | ||
1464 | BUG_ON(path == NULL); | |
1465 | depth = path->p_depth; | |
1466 | ||
1467 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1468 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1469 | |
1470 | while (depth >= 0) { | |
1471 | if (depth == path->p_depth) { | |
1472 | /* leaf */ | |
6f8ff537 CW |
1473 | if (path[depth].p_ext && |
1474 | path[depth].p_ext != | |
a86c6181 AT |
1475 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1476 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1477 | } else { | |
1478 | /* index */ | |
1479 | if (path[depth].p_idx != | |
1480 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1481 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1482 | } | |
1483 | depth--; | |
1484 | } | |
1485 | ||
f17722f9 | 1486 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1487 | } |
1488 | ||
1489 | /* | |
d0d856e8 | 1490 | * ext4_ext_next_leaf_block: |
f17722f9 | 1491 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1492 | */ |
5718789d | 1493 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1494 | { |
1495 | int depth; | |
1496 | ||
1497 | BUG_ON(path == NULL); | |
1498 | depth = path->p_depth; | |
1499 | ||
1500 | /* zero-tree has no leaf blocks at all */ | |
1501 | if (depth == 0) | |
f17722f9 | 1502 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1503 | |
1504 | /* go to index block */ | |
1505 | depth--; | |
1506 | ||
1507 | while (depth >= 0) { | |
1508 | if (path[depth].p_idx != | |
1509 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1510 | return (ext4_lblk_t) |
1511 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1512 | depth--; |
1513 | } | |
1514 | ||
f17722f9 | 1515 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1516 | } |
1517 | ||
1518 | /* | |
d0d856e8 RD |
1519 | * ext4_ext_correct_indexes: |
1520 | * if leaf gets modified and modified extent is first in the leaf, | |
1521 | * then we have to correct all indexes above. | |
a86c6181 AT |
1522 | * TODO: do we need to correct tree in all cases? |
1523 | */ | |
1d03ec98 | 1524 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1525 | struct ext4_ext_path *path) |
1526 | { | |
1527 | struct ext4_extent_header *eh; | |
1528 | int depth = ext_depth(inode); | |
1529 | struct ext4_extent *ex; | |
1530 | __le32 border; | |
1531 | int k, err = 0; | |
1532 | ||
1533 | eh = path[depth].p_hdr; | |
1534 | ex = path[depth].p_ext; | |
273df556 FM |
1535 | |
1536 | if (unlikely(ex == NULL || eh == NULL)) { | |
1537 | EXT4_ERROR_INODE(inode, | |
1538 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1539 | return -EIO; | |
1540 | } | |
a86c6181 AT |
1541 | |
1542 | if (depth == 0) { | |
1543 | /* there is no tree at all */ | |
1544 | return 0; | |
1545 | } | |
1546 | ||
1547 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1548 | /* we correct tree if first leaf got modified only */ | |
1549 | return 0; | |
1550 | } | |
1551 | ||
1552 | /* | |
d0d856e8 | 1553 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1554 | */ |
1555 | k = depth - 1; | |
1556 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1557 | err = ext4_ext_get_access(handle, inode, path + k); |
1558 | if (err) | |
a86c6181 AT |
1559 | return err; |
1560 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1561 | err = ext4_ext_dirty(handle, inode, path + k); |
1562 | if (err) | |
a86c6181 AT |
1563 | return err; |
1564 | ||
1565 | while (k--) { | |
1566 | /* change all left-side indexes */ | |
1567 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1568 | break; | |
7e028976 AM |
1569 | err = ext4_ext_get_access(handle, inode, path + k); |
1570 | if (err) | |
a86c6181 AT |
1571 | break; |
1572 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1573 | err = ext4_ext_dirty(handle, inode, path + k); |
1574 | if (err) | |
a86c6181 AT |
1575 | break; |
1576 | } | |
1577 | ||
1578 | return err; | |
1579 | } | |
1580 | ||
748de673 | 1581 | int |
a86c6181 AT |
1582 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1583 | struct ext4_extent *ex2) | |
1584 | { | |
749269fa | 1585 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1586 | |
1587 | /* | |
1588 | * Make sure that either both extents are uninitialized, or | |
1589 | * both are _not_. | |
1590 | */ | |
1591 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1592 | return 0; | |
1593 | ||
749269fa AA |
1594 | if (ext4_ext_is_uninitialized(ex1)) |
1595 | max_len = EXT_UNINIT_MAX_LEN; | |
1596 | else | |
1597 | max_len = EXT_INIT_MAX_LEN; | |
1598 | ||
a2df2a63 AA |
1599 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1600 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1601 | ||
1602 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1603 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1604 | return 0; |
1605 | ||
471d4011 SB |
1606 | /* |
1607 | * To allow future support for preallocated extents to be added | |
1608 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1609 | * this can result in the top bit of ee_len being set. |
471d4011 | 1610 | */ |
749269fa | 1611 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1612 | return 0; |
bbf2f9fb | 1613 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1614 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1615 | return 0; |
1616 | #endif | |
1617 | ||
bf89d16f | 1618 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1619 | return 1; |
1620 | return 0; | |
1621 | } | |
1622 | ||
56055d3a AA |
1623 | /* |
1624 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1625 | * It always tries to merge towards right. If you want to merge towards | |
1626 | * left, pass "ex - 1" as argument instead of "ex". | |
1627 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1628 | * 1 if they got merged. | |
1629 | */ | |
197217a5 | 1630 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1631 | struct ext4_ext_path *path, |
1632 | struct ext4_extent *ex) | |
56055d3a AA |
1633 | { |
1634 | struct ext4_extent_header *eh; | |
1635 | unsigned int depth, len; | |
1636 | int merge_done = 0; | |
1637 | int uninitialized = 0; | |
1638 | ||
1639 | depth = ext_depth(inode); | |
1640 | BUG_ON(path[depth].p_hdr == NULL); | |
1641 | eh = path[depth].p_hdr; | |
1642 | ||
1643 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1644 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1645 | break; | |
1646 | /* merge with next extent! */ | |
1647 | if (ext4_ext_is_uninitialized(ex)) | |
1648 | uninitialized = 1; | |
1649 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1650 | + ext4_ext_get_actual_len(ex + 1)); | |
1651 | if (uninitialized) | |
1652 | ext4_ext_mark_uninitialized(ex); | |
1653 | ||
1654 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1655 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1656 | * sizeof(struct ext4_extent); | |
1657 | memmove(ex + 1, ex + 2, len); | |
1658 | } | |
e8546d06 | 1659 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1660 | merge_done = 1; |
1661 | WARN_ON(eh->eh_entries == 0); | |
1662 | if (!eh->eh_entries) | |
24676da4 | 1663 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1664 | } |
1665 | ||
1666 | return merge_done; | |
1667 | } | |
1668 | ||
ecb94f5f TT |
1669 | /* |
1670 | * This function does a very simple check to see if we can collapse | |
1671 | * an extent tree with a single extent tree leaf block into the inode. | |
1672 | */ | |
1673 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1674 | struct inode *inode, | |
1675 | struct ext4_ext_path *path) | |
1676 | { | |
1677 | size_t s; | |
1678 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1679 | ext4_fsblk_t blk; | |
1680 | ||
1681 | if ((path[0].p_depth != 1) || | |
1682 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1683 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1684 | return; | |
1685 | ||
1686 | /* | |
1687 | * We need to modify the block allocation bitmap and the block | |
1688 | * group descriptor to release the extent tree block. If we | |
1689 | * can't get the journal credits, give up. | |
1690 | */ | |
1691 | if (ext4_journal_extend(handle, 2)) | |
1692 | return; | |
1693 | ||
1694 | /* | |
1695 | * Copy the extent data up to the inode | |
1696 | */ | |
1697 | blk = ext4_idx_pblock(path[0].p_idx); | |
1698 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1699 | sizeof(struct ext4_extent_idx); | |
1700 | s += sizeof(struct ext4_extent_header); | |
1701 | ||
1702 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1703 | path[0].p_depth = 0; | |
1704 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1705 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1706 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1707 | ||
1708 | brelse(path[1].p_bh); | |
1709 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
1710 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); | |
1711 | } | |
1712 | ||
197217a5 YY |
1713 | /* |
1714 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1715 | * return 1 if merge left else 0. | |
1716 | */ | |
ecb94f5f TT |
1717 | static void ext4_ext_try_to_merge(handle_t *handle, |
1718 | struct inode *inode, | |
197217a5 YY |
1719 | struct ext4_ext_path *path, |
1720 | struct ext4_extent *ex) { | |
1721 | struct ext4_extent_header *eh; | |
1722 | unsigned int depth; | |
1723 | int merge_done = 0; | |
197217a5 YY |
1724 | |
1725 | depth = ext_depth(inode); | |
1726 | BUG_ON(path[depth].p_hdr == NULL); | |
1727 | eh = path[depth].p_hdr; | |
1728 | ||
1729 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1730 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1731 | ||
1732 | if (!merge_done) | |
ecb94f5f | 1733 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1734 | |
ecb94f5f | 1735 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1736 | } |
1737 | ||
25d14f98 AA |
1738 | /* |
1739 | * check if a portion of the "newext" extent overlaps with an | |
1740 | * existing extent. | |
1741 | * | |
1742 | * If there is an overlap discovered, it updates the length of the newext | |
1743 | * such that there will be no overlap, and then returns 1. | |
1744 | * If there is no overlap found, it returns 0. | |
1745 | */ | |
4d33b1ef TT |
1746 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1747 | struct inode *inode, | |
1f109d5a TT |
1748 | struct ext4_extent *newext, |
1749 | struct ext4_ext_path *path) | |
25d14f98 | 1750 | { |
725d26d3 | 1751 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1752 | unsigned int depth, len1; |
1753 | unsigned int ret = 0; | |
1754 | ||
1755 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1756 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1757 | depth = ext_depth(inode); |
1758 | if (!path[depth].p_ext) | |
1759 | goto out; | |
1760 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1761 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1762 | |
1763 | /* | |
1764 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1765 | * is before the requested block(s) |
25d14f98 AA |
1766 | */ |
1767 | if (b2 < b1) { | |
1768 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1769 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1770 | goto out; |
4d33b1ef | 1771 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1772 | } |
1773 | ||
725d26d3 | 1774 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1775 | if (b1 + len1 < b1) { |
f17722f9 | 1776 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1777 | newext->ee_len = cpu_to_le16(len1); |
1778 | ret = 1; | |
1779 | } | |
1780 | ||
1781 | /* check for overlap */ | |
1782 | if (b1 + len1 > b2) { | |
1783 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1784 | ret = 1; | |
1785 | } | |
1786 | out: | |
1787 | return ret; | |
1788 | } | |
1789 | ||
a86c6181 | 1790 | /* |
d0d856e8 RD |
1791 | * ext4_ext_insert_extent: |
1792 | * tries to merge requsted extent into the existing extent or | |
1793 | * inserts requested extent as new one into the tree, | |
1794 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1795 | */ |
1796 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1797 | struct ext4_ext_path *path, | |
0031462b | 1798 | struct ext4_extent *newext, int flag) |
a86c6181 | 1799 | { |
af5bc92d | 1800 | struct ext4_extent_header *eh; |
a86c6181 AT |
1801 | struct ext4_extent *ex, *fex; |
1802 | struct ext4_extent *nearex; /* nearest extent */ | |
1803 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1804 | int depth, len, err; |
1805 | ext4_lblk_t next; | |
a2df2a63 | 1806 | unsigned uninitialized = 0; |
55f020db | 1807 | int flags = 0; |
a86c6181 | 1808 | |
273df556 FM |
1809 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1810 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1811 | return -EIO; | |
1812 | } | |
a86c6181 AT |
1813 | depth = ext_depth(inode); |
1814 | ex = path[depth].p_ext; | |
273df556 FM |
1815 | if (unlikely(path[depth].p_hdr == NULL)) { |
1816 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1817 | return -EIO; | |
1818 | } | |
a86c6181 AT |
1819 | |
1820 | /* try to insert block into found extent and return */ | |
744692dc | 1821 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1822 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
32de6756 | 1823 | ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n", |
bf89d16f TT |
1824 | ext4_ext_is_uninitialized(newext), |
1825 | ext4_ext_get_actual_len(newext), | |
1826 | le32_to_cpu(ex->ee_block), | |
1827 | ext4_ext_is_uninitialized(ex), | |
1828 | ext4_ext_get_actual_len(ex), | |
1829 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1830 | err = ext4_ext_get_access(handle, inode, path + depth); |
1831 | if (err) | |
a86c6181 | 1832 | return err; |
a2df2a63 AA |
1833 | |
1834 | /* | |
1835 | * ext4_can_extents_be_merged should have checked that either | |
1836 | * both extents are uninitialized, or both aren't. Thus we | |
1837 | * need to check only one of them here. | |
1838 | */ | |
1839 | if (ext4_ext_is_uninitialized(ex)) | |
1840 | uninitialized = 1; | |
1841 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1842 | + ext4_ext_get_actual_len(newext)); | |
1843 | if (uninitialized) | |
1844 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1845 | eh = path[depth].p_hdr; |
1846 | nearex = ex; | |
1847 | goto merge; | |
1848 | } | |
1849 | ||
a86c6181 AT |
1850 | depth = ext_depth(inode); |
1851 | eh = path[depth].p_hdr; | |
1852 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1853 | goto has_space; | |
1854 | ||
1855 | /* probably next leaf has space for us? */ | |
1856 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1857 | next = EXT_MAX_BLOCKS; |
1858 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1859 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1860 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1861 | ext_debug("next leaf block - %u\n", next); |
a86c6181 AT |
1862 | BUG_ON(npath != NULL); |
1863 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1864 | if (IS_ERR(npath)) | |
1865 | return PTR_ERR(npath); | |
1866 | BUG_ON(npath->p_depth != path->p_depth); | |
1867 | eh = npath[depth].p_hdr; | |
1868 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1869 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1870 | le16_to_cpu(eh->eh_entries)); |
1871 | path = npath; | |
ffb505ff | 1872 | goto has_space; |
a86c6181 AT |
1873 | } |
1874 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1875 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1876 | } | |
1877 | ||
1878 | /* | |
d0d856e8 RD |
1879 | * There is no free space in the found leaf. |
1880 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1881 | */ |
55f020db AH |
1882 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1883 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1884 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1885 | if (err) |
1886 | goto cleanup; | |
1887 | depth = ext_depth(inode); | |
1888 | eh = path[depth].p_hdr; | |
1889 | ||
1890 | has_space: | |
1891 | nearex = path[depth].p_ext; | |
1892 | ||
7e028976 AM |
1893 | err = ext4_ext_get_access(handle, inode, path + depth); |
1894 | if (err) | |
a86c6181 AT |
1895 | goto cleanup; |
1896 | ||
1897 | if (!nearex) { | |
1898 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1899 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1900 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1901 | ext4_ext_pblock(newext), |
553f9008 | 1902 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1903 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1904 | nearex = EXT_FIRST_EXTENT(eh); |
1905 | } else { | |
1906 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1907 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1908 | /* Insert after */ |
32de6756 YY |
1909 | ext_debug("insert %u:%llu:[%d]%d before: " |
1910 | "nearest %p\n", | |
80e675f9 EG |
1911 | le32_to_cpu(newext->ee_block), |
1912 | ext4_ext_pblock(newext), | |
1913 | ext4_ext_is_uninitialized(newext), | |
1914 | ext4_ext_get_actual_len(newext), | |
1915 | nearex); | |
1916 | nearex++; | |
1917 | } else { | |
1918 | /* Insert before */ | |
1919 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
1920 | ext_debug("insert %u:%llu:[%d]%d after: " |
1921 | "nearest %p\n", | |
8c55e204 | 1922 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1923 | ext4_ext_pblock(newext), |
553f9008 | 1924 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1925 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
1926 | nearex); |
1927 | } | |
1928 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
1929 | if (len > 0) { | |
32de6756 | 1930 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
1931 | "move %d extents from 0x%p to 0x%p\n", |
1932 | le32_to_cpu(newext->ee_block), | |
1933 | ext4_ext_pblock(newext), | |
1934 | ext4_ext_is_uninitialized(newext), | |
1935 | ext4_ext_get_actual_len(newext), | |
1936 | len, nearex, nearex + 1); | |
1937 | memmove(nearex + 1, nearex, | |
1938 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 1939 | } |
a86c6181 AT |
1940 | } |
1941 | ||
e8546d06 | 1942 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 1943 | path[depth].p_ext = nearex; |
a86c6181 | 1944 | nearex->ee_block = newext->ee_block; |
bf89d16f | 1945 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1946 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1947 | |
1948 | merge: | |
e7bcf823 | 1949 | /* try to merge extents */ |
744692dc | 1950 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 1951 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 1952 | |
a86c6181 AT |
1953 | |
1954 | /* time to correct all indexes above */ | |
1955 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1956 | if (err) | |
1957 | goto cleanup; | |
1958 | ||
ecb94f5f | 1959 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
1960 | |
1961 | cleanup: | |
1962 | if (npath) { | |
1963 | ext4_ext_drop_refs(npath); | |
1964 | kfree(npath); | |
1965 | } | |
a86c6181 AT |
1966 | ext4_ext_invalidate_cache(inode); |
1967 | return err; | |
1968 | } | |
1969 | ||
91dd8c11 LC |
1970 | static int ext4_fill_fiemap_extents(struct inode *inode, |
1971 | ext4_lblk_t block, ext4_lblk_t num, | |
1972 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
1973 | { |
1974 | struct ext4_ext_path *path = NULL; | |
06348679 | 1975 | struct ext4_ext_cache newex; |
6873fa0d | 1976 | struct ext4_extent *ex; |
91dd8c11 | 1977 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 1978 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
1979 | int exists, depth = 0, err = 0; |
1980 | unsigned int flags = 0; | |
1981 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 1982 | |
f17722f9 | 1983 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1984 | num = last - block; |
1985 | /* find extent for this block */ | |
fab3a549 | 1986 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
1987 | |
1988 | if (path && ext_depth(inode) != depth) { | |
1989 | /* depth was changed. we have to realloc path */ | |
1990 | kfree(path); | |
1991 | path = NULL; | |
1992 | } | |
1993 | ||
6873fa0d ES |
1994 | path = ext4_ext_find_extent(inode, block, path); |
1995 | if (IS_ERR(path)) { | |
91dd8c11 | 1996 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1997 | err = PTR_ERR(path); |
1998 | path = NULL; | |
1999 | break; | |
2000 | } | |
2001 | ||
2002 | depth = ext_depth(inode); | |
273df556 | 2003 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2004 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2005 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2006 | err = -EIO; | |
2007 | break; | |
2008 | } | |
6873fa0d ES |
2009 | ex = path[depth].p_ext; |
2010 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2011 | ext4_ext_drop_refs(path); |
6873fa0d | 2012 | |
91dd8c11 | 2013 | flags = 0; |
6873fa0d ES |
2014 | exists = 0; |
2015 | if (!ex) { | |
2016 | /* there is no extent yet, so try to allocate | |
2017 | * all requested space */ | |
2018 | start = block; | |
2019 | end = block + num; | |
2020 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2021 | /* need to allocate space before found extent */ | |
2022 | start = block; | |
2023 | end = le32_to_cpu(ex->ee_block); | |
2024 | if (block + num < end) | |
2025 | end = block + num; | |
2026 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2027 | + ext4_ext_get_actual_len(ex)) { | |
2028 | /* need to allocate space after found extent */ | |
2029 | start = block; | |
2030 | end = block + num; | |
2031 | if (end >= next) | |
2032 | end = next; | |
2033 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2034 | /* | |
2035 | * some part of requested space is covered | |
2036 | * by found extent | |
2037 | */ | |
2038 | start = block; | |
2039 | end = le32_to_cpu(ex->ee_block) | |
2040 | + ext4_ext_get_actual_len(ex); | |
2041 | if (block + num < end) | |
2042 | end = block + num; | |
2043 | exists = 1; | |
2044 | } else { | |
2045 | BUG(); | |
2046 | } | |
2047 | BUG_ON(end <= start); | |
2048 | ||
2049 | if (!exists) { | |
06348679 LC |
2050 | newex.ec_block = start; |
2051 | newex.ec_len = end - start; | |
2052 | newex.ec_start = 0; | |
6873fa0d | 2053 | } else { |
06348679 LC |
2054 | newex.ec_block = le32_to_cpu(ex->ee_block); |
2055 | newex.ec_len = ext4_ext_get_actual_len(ex); | |
2056 | newex.ec_start = ext4_ext_pblock(ex); | |
91dd8c11 LC |
2057 | if (ext4_ext_is_uninitialized(ex)) |
2058 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
6873fa0d ES |
2059 | } |
2060 | ||
91dd8c11 | 2061 | /* |
06348679 LC |
2062 | * Find delayed extent and update newex accordingly. We call |
2063 | * it even in !exists case to find out whether newex is the | |
91dd8c11 LC |
2064 | * last existing extent or not. |
2065 | */ | |
06348679 | 2066 | next_del = ext4_find_delayed_extent(inode, &newex); |
91dd8c11 LC |
2067 | if (!exists && next_del) { |
2068 | exists = 1; | |
2069 | flags |= FIEMAP_EXTENT_DELALLOC; | |
2070 | } | |
2071 | up_read(&EXT4_I(inode)->i_data_sem); | |
2072 | ||
06348679 LC |
2073 | if (unlikely(newex.ec_len == 0)) { |
2074 | EXT4_ERROR_INODE(inode, "newex.ec_len == 0"); | |
273df556 FM |
2075 | err = -EIO; |
2076 | break; | |
2077 | } | |
6873fa0d | 2078 | |
f7fec032 ZL |
2079 | /* |
2080 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2081 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2082 | * possible that an extent is with unwritten and delayed | |
2083 | * status due to when an extent is delayed allocated and | |
2084 | * is allocated by fallocate status tree will track both of | |
2085 | * them in a extent. | |
2086 | * | |
2087 | * So we could return a unwritten and delayed extent, and | |
2088 | * its block is equal to 'next'. | |
2089 | */ | |
2090 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2091 | flags |= FIEMAP_EXTENT_LAST; |
2092 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2093 | next != EXT_MAX_BLOCKS)) { | |
2094 | EXT4_ERROR_INODE(inode, | |
2095 | "next extent == %u, next " | |
2096 | "delalloc extent = %u", | |
2097 | next, next_del); | |
2098 | err = -EIO; | |
2099 | break; | |
2100 | } | |
6873fa0d ES |
2101 | } |
2102 | ||
91dd8c11 LC |
2103 | if (exists) { |
2104 | err = fiemap_fill_next_extent(fieinfo, | |
06348679 LC |
2105 | (__u64)newex.ec_block << blksize_bits, |
2106 | (__u64)newex.ec_start << blksize_bits, | |
2107 | (__u64)newex.ec_len << blksize_bits, | |
91dd8c11 LC |
2108 | flags); |
2109 | if (err < 0) | |
2110 | break; | |
2111 | if (err == 1) { | |
2112 | err = 0; | |
2113 | break; | |
2114 | } | |
6873fa0d ES |
2115 | } |
2116 | ||
06348679 | 2117 | block = newex.ec_block + newex.ec_len; |
6873fa0d ES |
2118 | } |
2119 | ||
2120 | if (path) { | |
2121 | ext4_ext_drop_refs(path); | |
2122 | kfree(path); | |
2123 | } | |
2124 | ||
2125 | return err; | |
2126 | } | |
2127 | ||
09b88252 | 2128 | static void |
725d26d3 | 2129 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
b05e6ae5 | 2130 | __u32 len, ext4_fsblk_t start) |
a86c6181 AT |
2131 | { |
2132 | struct ext4_ext_cache *cex; | |
2133 | BUG_ON(len == 0); | |
2ec0ae3a | 2134 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
d8990240 | 2135 | trace_ext4_ext_put_in_cache(inode, block, len, start); |
a86c6181 | 2136 | cex = &EXT4_I(inode)->i_cached_extent; |
a86c6181 AT |
2137 | cex->ec_block = block; |
2138 | cex->ec_len = len; | |
2139 | cex->ec_start = start; | |
2ec0ae3a | 2140 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
a86c6181 AT |
2141 | } |
2142 | ||
2143 | /* | |
d0d856e8 RD |
2144 | * ext4_ext_put_gap_in_cache: |
2145 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2146 | * and cache this gap |
2147 | */ | |
09b88252 | 2148 | static void |
a86c6181 | 2149 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2150 | ext4_lblk_t block) |
a86c6181 AT |
2151 | { |
2152 | int depth = ext_depth(inode); | |
725d26d3 AK |
2153 | unsigned long len; |
2154 | ext4_lblk_t lblock; | |
a86c6181 AT |
2155 | struct ext4_extent *ex; |
2156 | ||
2157 | ex = path[depth].p_ext; | |
2158 | if (ex == NULL) { | |
2159 | /* there is no extent yet, so gap is [0;-] */ | |
2160 | lblock = 0; | |
f17722f9 | 2161 | len = EXT_MAX_BLOCKS; |
a86c6181 AT |
2162 | ext_debug("cache gap(whole file):"); |
2163 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2164 | lblock = block; | |
2165 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2166 | ext_debug("cache gap(before): %u [%u:%u]", |
2167 | block, | |
2168 | le32_to_cpu(ex->ee_block), | |
2169 | ext4_ext_get_actual_len(ex)); | |
d100eef2 ZL |
2170 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2171 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2172 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2173 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2174 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2175 | ext4_lblk_t next; |
8c55e204 | 2176 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2177 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2178 | |
2179 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2180 | ext_debug("cache gap(after): [%u:%u] %u", |
2181 | le32_to_cpu(ex->ee_block), | |
2182 | ext4_ext_get_actual_len(ex), | |
2183 | block); | |
725d26d3 AK |
2184 | BUG_ON(next == lblock); |
2185 | len = next - lblock; | |
d100eef2 ZL |
2186 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2187 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2188 | EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2189 | } else { |
2190 | lblock = len = 0; | |
2191 | BUG(); | |
2192 | } | |
2193 | ||
bba90743 | 2194 | ext_debug(" -> %u:%lu\n", lblock, len); |
b05e6ae5 | 2195 | ext4_ext_put_in_cache(inode, lblock, len, 0); |
a86c6181 AT |
2196 | } |
2197 | ||
b05e6ae5 | 2198 | /* |
63fedaf1 | 2199 | * ext4_ext_in_cache() |
a4bb6b64 AH |
2200 | * Checks to see if the given block is in the cache. |
2201 | * If it is, the cached extent is stored in the given | |
63fedaf1 | 2202 | * cache extent pointer. |
a4bb6b64 AH |
2203 | * |
2204 | * @inode: The files inode | |
2205 | * @block: The block to look for in the cache | |
2206 | * @ex: Pointer where the cached extent will be stored | |
2207 | * if it contains block | |
2208 | * | |
b05e6ae5 TT |
2209 | * Return 0 if cache is invalid; 1 if the cache is valid |
2210 | */ | |
63fedaf1 LC |
2211 | static int |
2212 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, | |
2213 | struct ext4_extent *ex) | |
2214 | { | |
a86c6181 | 2215 | struct ext4_ext_cache *cex; |
b05e6ae5 | 2216 | int ret = 0; |
a86c6181 | 2217 | |
60e6679e | 2218 | /* |
2ec0ae3a TT |
2219 | * We borrow i_block_reservation_lock to protect i_cached_extent |
2220 | */ | |
2221 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
a86c6181 AT |
2222 | cex = &EXT4_I(inode)->i_cached_extent; |
2223 | ||
2224 | /* has cache valid data? */ | |
b05e6ae5 | 2225 | if (cex->ec_len == 0) |
2ec0ae3a | 2226 | goto errout; |
a86c6181 | 2227 | |
731eb1a0 | 2228 | if (in_range(block, cex->ec_block, cex->ec_len)) { |
63fedaf1 LC |
2229 | ex->ee_block = cpu_to_le32(cex->ec_block); |
2230 | ext4_ext_store_pblock(ex, cex->ec_start); | |
2231 | ex->ee_len = cpu_to_le16(cex->ec_len); | |
bba90743 ES |
2232 | ext_debug("%u cached by %u:%u:%llu\n", |
2233 | block, | |
2234 | cex->ec_block, cex->ec_len, cex->ec_start); | |
b05e6ae5 | 2235 | ret = 1; |
a86c6181 | 2236 | } |
2ec0ae3a | 2237 | errout: |
d8990240 | 2238 | trace_ext4_ext_in_cache(inode, block, ret); |
2ec0ae3a TT |
2239 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
2240 | return ret; | |
a86c6181 AT |
2241 | } |
2242 | ||
2243 | /* | |
d0d856e8 RD |
2244 | * ext4_ext_rm_idx: |
2245 | * removes index from the index block. | |
a86c6181 | 2246 | */ |
1d03ec98 | 2247 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2248 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2249 | { |
a86c6181 | 2250 | int err; |
f65e6fba | 2251 | ext4_fsblk_t leaf; |
a86c6181 AT |
2252 | |
2253 | /* free index block */ | |
c36575e6 FL |
2254 | depth--; |
2255 | path = path + depth; | |
bf89d16f | 2256 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2257 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2258 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2259 | return -EIO; | |
2260 | } | |
7e028976 AM |
2261 | err = ext4_ext_get_access(handle, inode, path); |
2262 | if (err) | |
a86c6181 | 2263 | return err; |
0e1147b0 RD |
2264 | |
2265 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2266 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2267 | len *= sizeof(struct ext4_extent_idx); | |
2268 | memmove(path->p_idx, path->p_idx + 1, len); | |
2269 | } | |
2270 | ||
e8546d06 | 2271 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2272 | err = ext4_ext_dirty(handle, inode, path); |
2273 | if (err) | |
a86c6181 | 2274 | return err; |
2ae02107 | 2275 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2276 | trace_ext4_ext_rm_idx(inode, leaf); |
2277 | ||
7dc57615 | 2278 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2279 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2280 | |
2281 | while (--depth >= 0) { | |
2282 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2283 | break; | |
2284 | path--; | |
2285 | err = ext4_ext_get_access(handle, inode, path); | |
2286 | if (err) | |
2287 | break; | |
2288 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2289 | err = ext4_ext_dirty(handle, inode, path); | |
2290 | if (err) | |
2291 | break; | |
2292 | } | |
a86c6181 AT |
2293 | return err; |
2294 | } | |
2295 | ||
2296 | /* | |
ee12b630 MC |
2297 | * ext4_ext_calc_credits_for_single_extent: |
2298 | * This routine returns max. credits that needed to insert an extent | |
2299 | * to the extent tree. | |
2300 | * When pass the actual path, the caller should calculate credits | |
2301 | * under i_data_sem. | |
a86c6181 | 2302 | */ |
525f4ed8 | 2303 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2304 | struct ext4_ext_path *path) |
2305 | { | |
a86c6181 | 2306 | if (path) { |
ee12b630 | 2307 | int depth = ext_depth(inode); |
f3bd1f3f | 2308 | int ret = 0; |
ee12b630 | 2309 | |
a86c6181 | 2310 | /* probably there is space in leaf? */ |
a86c6181 | 2311 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2312 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2313 | |
ee12b630 MC |
2314 | /* |
2315 | * There are some space in the leaf tree, no | |
2316 | * need to account for leaf block credit | |
2317 | * | |
2318 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2319 | * and other metadata blocks still need to be |
ee12b630 MC |
2320 | * accounted. |
2321 | */ | |
525f4ed8 | 2322 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2323 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2324 | return ret; |
ee12b630 MC |
2325 | } |
2326 | } | |
a86c6181 | 2327 | |
525f4ed8 | 2328 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2329 | } |
a86c6181 | 2330 | |
ee12b630 MC |
2331 | /* |
2332 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2333 | * | |
2334 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2335 | * in the worse case, each tree level index/leaf need to be changed | |
2336 | * if the tree split due to insert a new extent, then the old tree | |
2337 | * index/leaf need to be updated too | |
2338 | * | |
2339 | * If the nrblocks are discontiguous, they could cause | |
2340 | * the whole tree split more than once, but this is really rare. | |
2341 | */ | |
525f4ed8 | 2342 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2343 | { |
2344 | int index; | |
f19d5870 TM |
2345 | int depth; |
2346 | ||
2347 | /* If we are converting the inline data, only one is needed here. */ | |
2348 | if (ext4_has_inline_data(inode)) | |
2349 | return 1; | |
2350 | ||
2351 | depth = ext_depth(inode); | |
a86c6181 | 2352 | |
ee12b630 MC |
2353 | if (chunk) |
2354 | index = depth * 2; | |
2355 | else | |
2356 | index = depth * 3; | |
a86c6181 | 2357 | |
ee12b630 | 2358 | return index; |
a86c6181 AT |
2359 | } |
2360 | ||
2361 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2362 | struct ext4_extent *ex, |
2363 | ext4_fsblk_t *partial_cluster, | |
2364 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2365 | { |
0aa06000 | 2366 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2367 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2368 | ext4_fsblk_t pblk; |
18888cf0 | 2369 | int flags = 0; |
a86c6181 | 2370 | |
c9de560d | 2371 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
18888cf0 AS |
2372 | flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2373 | else if (ext4_should_journal_data(inode)) | |
2374 | flags |= EXT4_FREE_BLOCKS_FORGET; | |
2375 | ||
0aa06000 TT |
2376 | /* |
2377 | * For bigalloc file systems, we never free a partial cluster | |
2378 | * at the beginning of the extent. Instead, we make a note | |
2379 | * that we tried freeing the cluster, and check to see if we | |
2380 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2381 | * or at the end of the ext4_truncate() operation. | |
2382 | */ | |
2383 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2384 | ||
d8990240 | 2385 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2386 | /* |
2387 | * If we have a partial cluster, and it's different from the | |
2388 | * cluster of the last block, we need to explicitly free the | |
2389 | * partial cluster here. | |
2390 | */ | |
2391 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2392 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2393 | ext4_free_blocks(handle, inode, NULL, | |
2394 | EXT4_C2B(sbi, *partial_cluster), | |
2395 | sbi->s_cluster_ratio, flags); | |
2396 | *partial_cluster = 0; | |
2397 | } | |
2398 | ||
a86c6181 AT |
2399 | #ifdef EXTENTS_STATS |
2400 | { | |
2401 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2402 | spin_lock(&sbi->s_ext_stats_lock); |
2403 | sbi->s_ext_blocks += ee_len; | |
2404 | sbi->s_ext_extents++; | |
2405 | if (ee_len < sbi->s_ext_min) | |
2406 | sbi->s_ext_min = ee_len; | |
2407 | if (ee_len > sbi->s_ext_max) | |
2408 | sbi->s_ext_max = ee_len; | |
2409 | if (ext_depth(inode) > sbi->s_depth_max) | |
2410 | sbi->s_depth_max = ext_depth(inode); | |
2411 | spin_unlock(&sbi->s_ext_stats_lock); | |
2412 | } | |
2413 | #endif | |
2414 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2415 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2416 | /* tail removal */ |
725d26d3 | 2417 | ext4_lblk_t num; |
725d26d3 | 2418 | |
a2df2a63 | 2419 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2420 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2421 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2422 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2423 | /* | |
2424 | * If the block range to be freed didn't start at the | |
2425 | * beginning of a cluster, and we removed the entire | |
2426 | * extent, save the partial cluster here, since we | |
2427 | * might need to delete if we determine that the | |
2428 | * truncate operation has removed all of the blocks in | |
2429 | * the cluster. | |
2430 | */ | |
2431 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2432 | (ee_len == num)) | |
2433 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2434 | else | |
2435 | *partial_cluster = 0; | |
a86c6181 | 2436 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2437 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2438 | /* head removal */ |
2439 | ext4_lblk_t num; | |
2440 | ext4_fsblk_t start; | |
2441 | ||
2442 | num = to - from; | |
2443 | start = ext4_ext_pblock(ex); | |
2444 | ||
2445 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
ee90d57e | 2446 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
d583fb87 | 2447 | |
a86c6181 | 2448 | } else { |
725d26d3 AK |
2449 | printk(KERN_INFO "strange request: removal(2) " |
2450 | "%u-%u from %u:%u\n", | |
2451 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2452 | } |
2453 | return 0; | |
2454 | } | |
2455 | ||
d583fb87 AH |
2456 | |
2457 | /* | |
2458 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2459 | * blocks appearing between "start" and "end", and splits the extents | |
2460 | * if "start" and "end" appear in the same extent | |
2461 | * | |
2462 | * @handle: The journal handle | |
2463 | * @inode: The files inode | |
2464 | * @path: The path to the leaf | |
2465 | * @start: The first block to remove | |
2466 | * @end: The last block to remove | |
2467 | */ | |
a86c6181 AT |
2468 | static int |
2469 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2470 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2471 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2472 | { |
0aa06000 | 2473 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2474 | int err = 0, correct_index = 0; |
2475 | int depth = ext_depth(inode), credits; | |
2476 | struct ext4_extent_header *eh; | |
750c9c47 | 2477 | ext4_lblk_t a, b; |
725d26d3 AK |
2478 | unsigned num; |
2479 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2480 | unsigned short ex_ee_len; |
a2df2a63 | 2481 | unsigned uninitialized = 0; |
a86c6181 AT |
2482 | struct ext4_extent *ex; |
2483 | ||
c29c0ae7 | 2484 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2485 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2486 | if (!path[depth].p_hdr) |
2487 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2488 | eh = path[depth].p_hdr; | |
273df556 FM |
2489 | if (unlikely(path[depth].p_hdr == NULL)) { |
2490 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2491 | return -EIO; | |
2492 | } | |
a86c6181 AT |
2493 | /* find where to start removing */ |
2494 | ex = EXT_LAST_EXTENT(eh); | |
2495 | ||
2496 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2497 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2498 | |
d8990240 AK |
2499 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2500 | ||
a86c6181 AT |
2501 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2502 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2503 | |
2504 | if (ext4_ext_is_uninitialized(ex)) | |
2505 | uninitialized = 1; | |
2506 | else | |
2507 | uninitialized = 0; | |
2508 | ||
553f9008 M |
2509 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2510 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2511 | path[depth].p_ext = ex; |
2512 | ||
2513 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2514 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2515 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2516 | |
2517 | ext_debug(" border %u:%u\n", a, b); | |
2518 | ||
d583fb87 | 2519 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2520 | if (end < ex_ee_block) { |
d583fb87 AH |
2521 | ex--; |
2522 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2523 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2524 | continue; | |
750c9c47 | 2525 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2526 | EXT4_ERROR_INODE(inode, |
2527 | "can not handle truncate %u:%u " | |
2528 | "on extent %u:%u", | |
2529 | start, end, ex_ee_block, | |
2530 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2531 | err = -EIO; |
2532 | goto out; | |
a86c6181 AT |
2533 | } else if (a != ex_ee_block) { |
2534 | /* remove tail of the extent */ | |
750c9c47 | 2535 | num = a - ex_ee_block; |
a86c6181 AT |
2536 | } else { |
2537 | /* remove whole extent: excellent! */ | |
a86c6181 | 2538 | num = 0; |
a86c6181 | 2539 | } |
34071da7 TT |
2540 | /* |
2541 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2542 | * descriptor) for each block group; assume two block | |
2543 | * groups plus ex_ee_len/blocks_per_block_group for | |
2544 | * the worst case | |
2545 | */ | |
2546 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2547 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2548 | correct_index = 1; | |
2549 | credits += (ext_depth(inode)) + 1; | |
2550 | } | |
5aca07eb | 2551 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2552 | |
487caeef | 2553 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2554 | if (err) |
a86c6181 | 2555 | goto out; |
a86c6181 AT |
2556 | |
2557 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2558 | if (err) | |
2559 | goto out; | |
2560 | ||
0aa06000 TT |
2561 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2562 | a, b); | |
a86c6181 AT |
2563 | if (err) |
2564 | goto out; | |
2565 | ||
750c9c47 | 2566 | if (num == 0) |
d0d856e8 | 2567 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2568 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2569 | |
a86c6181 | 2570 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2571 | /* |
2572 | * Do not mark uninitialized if all the blocks in the | |
2573 | * extent have been removed. | |
2574 | */ | |
2575 | if (uninitialized && num) | |
a2df2a63 | 2576 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2577 | /* |
2578 | * If the extent was completely released, | |
2579 | * we need to remove it from the leaf | |
2580 | */ | |
2581 | if (num == 0) { | |
f17722f9 | 2582 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2583 | /* |
2584 | * For hole punching, we need to scoot all the | |
2585 | * extents up when an extent is removed so that | |
2586 | * we dont have blank extents in the middle | |
2587 | */ | |
2588 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2589 | sizeof(struct ext4_extent)); | |
2590 | ||
2591 | /* Now get rid of the one at the end */ | |
2592 | memset(EXT_LAST_EXTENT(eh), 0, | |
2593 | sizeof(struct ext4_extent)); | |
2594 | } | |
2595 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2596 | } else |
2597 | *partial_cluster = 0; | |
d583fb87 | 2598 | |
750c9c47 DM |
2599 | err = ext4_ext_dirty(handle, inode, path + depth); |
2600 | if (err) | |
2601 | goto out; | |
2602 | ||
bf52c6f7 | 2603 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2604 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2605 | ex--; |
2606 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2607 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2608 | } |
2609 | ||
2610 | if (correct_index && eh->eh_entries) | |
2611 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2612 | ||
0aa06000 TT |
2613 | /* |
2614 | * If there is still a entry in the leaf node, check to see if | |
2615 | * it references the partial cluster. This is the only place | |
2616 | * where it could; if it doesn't, we can free the cluster. | |
2617 | */ | |
2618 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2619 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2620 | *partial_cluster)) { | |
2621 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2622 | ||
2623 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2624 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2625 | ||
2626 | ext4_free_blocks(handle, inode, NULL, | |
2627 | EXT4_C2B(sbi, *partial_cluster), | |
2628 | sbi->s_cluster_ratio, flags); | |
2629 | *partial_cluster = 0; | |
2630 | } | |
2631 | ||
a86c6181 AT |
2632 | /* if this leaf is free, then we should |
2633 | * remove it from index block above */ | |
2634 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2635 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2636 | |
2637 | out: | |
2638 | return err; | |
2639 | } | |
2640 | ||
2641 | /* | |
d0d856e8 RD |
2642 | * ext4_ext_more_to_rm: |
2643 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2644 | */ |
09b88252 | 2645 | static int |
a86c6181 AT |
2646 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2647 | { | |
2648 | BUG_ON(path->p_idx == NULL); | |
2649 | ||
2650 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2651 | return 0; | |
2652 | ||
2653 | /* | |
d0d856e8 | 2654 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2655 | * so we have to consider current index for truncation |
2656 | */ | |
2657 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2658 | return 0; | |
2659 | return 1; | |
2660 | } | |
2661 | ||
5f95d21f LC |
2662 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2663 | ext4_lblk_t end) | |
a86c6181 AT |
2664 | { |
2665 | struct super_block *sb = inode->i_sb; | |
2666 | int depth = ext_depth(inode); | |
968dee77 | 2667 | struct ext4_ext_path *path = NULL; |
0aa06000 | 2668 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2669 | handle_t *handle; |
6f2080e6 | 2670 | int i = 0, err = 0; |
a86c6181 | 2671 | |
5f95d21f | 2672 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2673 | |
2674 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2675 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2676 | if (IS_ERR(handle)) |
2677 | return PTR_ERR(handle); | |
2678 | ||
0617b83f | 2679 | again: |
a86c6181 AT |
2680 | ext4_ext_invalidate_cache(inode); |
2681 | ||
d8990240 AK |
2682 | trace_ext4_ext_remove_space(inode, start, depth); |
2683 | ||
5f95d21f LC |
2684 | /* |
2685 | * Check if we are removing extents inside the extent tree. If that | |
2686 | * is the case, we are going to punch a hole inside the extent tree | |
2687 | * so we have to check whether we need to split the extent covering | |
2688 | * the last block to remove so we can easily remove the part of it | |
2689 | * in ext4_ext_rm_leaf(). | |
2690 | */ | |
2691 | if (end < EXT_MAX_BLOCKS - 1) { | |
2692 | struct ext4_extent *ex; | |
2693 | ext4_lblk_t ee_block; | |
2694 | ||
2695 | /* find extent for this block */ | |
2696 | path = ext4_ext_find_extent(inode, end, NULL); | |
2697 | if (IS_ERR(path)) { | |
2698 | ext4_journal_stop(handle); | |
2699 | return PTR_ERR(path); | |
2700 | } | |
2701 | depth = ext_depth(inode); | |
6f2080e6 | 2702 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2703 | ex = path[depth].p_ext; |
968dee77 | 2704 | if (!ex) { |
6f2080e6 DM |
2705 | if (depth) { |
2706 | EXT4_ERROR_INODE(inode, | |
2707 | "path[%d].p_hdr == NULL", | |
2708 | depth); | |
2709 | err = -EIO; | |
2710 | } | |
2711 | goto out; | |
968dee77 | 2712 | } |
5f95d21f LC |
2713 | |
2714 | ee_block = le32_to_cpu(ex->ee_block); | |
2715 | ||
2716 | /* | |
2717 | * See if the last block is inside the extent, if so split | |
2718 | * the extent at 'end' block so we can easily remove the | |
2719 | * tail of the first part of the split extent in | |
2720 | * ext4_ext_rm_leaf(). | |
2721 | */ | |
2722 | if (end >= ee_block && | |
2723 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2724 | int split_flag = 0; | |
2725 | ||
2726 | if (ext4_ext_is_uninitialized(ex)) | |
2727 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2728 | EXT4_EXT_MARK_UNINIT2; | |
2729 | ||
2730 | /* | |
2731 | * Split the extent in two so that 'end' is the last | |
2732 | * block in the first new extent | |
2733 | */ | |
2734 | err = ext4_split_extent_at(handle, inode, path, | |
2735 | end + 1, split_flag, | |
2736 | EXT4_GET_BLOCKS_PRE_IO | | |
2737 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
2738 | ||
2739 | if (err < 0) | |
2740 | goto out; | |
2741 | } | |
5f95d21f | 2742 | } |
a86c6181 | 2743 | /* |
d0d856e8 RD |
2744 | * We start scanning from right side, freeing all the blocks |
2745 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2746 | */ |
0617b83f | 2747 | depth = ext_depth(inode); |
968dee77 AS |
2748 | if (path) { |
2749 | int k = i = depth; | |
2750 | while (--k > 0) | |
2751 | path[k].p_block = | |
2752 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2753 | } else { | |
2754 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2755 | GFP_NOFS); | |
2756 | if (path == NULL) { | |
2757 | ext4_journal_stop(handle); | |
2758 | return -ENOMEM; | |
2759 | } | |
2760 | path[0].p_depth = depth; | |
2761 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2762 | i = 0; |
5f95d21f | 2763 | |
968dee77 AS |
2764 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
2765 | err = -EIO; | |
2766 | goto out; | |
2767 | } | |
a86c6181 | 2768 | } |
968dee77 | 2769 | err = 0; |
a86c6181 AT |
2770 | |
2771 | while (i >= 0 && err == 0) { | |
2772 | if (i == depth) { | |
2773 | /* this is leaf block */ | |
d583fb87 | 2774 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2775 | &partial_cluster, start, |
5f95d21f | 2776 | end); |
d0d856e8 | 2777 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2778 | brelse(path[i].p_bh); |
2779 | path[i].p_bh = NULL; | |
2780 | i--; | |
2781 | continue; | |
2782 | } | |
2783 | ||
2784 | /* this is index block */ | |
2785 | if (!path[i].p_hdr) { | |
2786 | ext_debug("initialize header\n"); | |
2787 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2788 | } |
2789 | ||
a86c6181 | 2790 | if (!path[i].p_idx) { |
d0d856e8 | 2791 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2792 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2793 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2794 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2795 | path[i].p_hdr, | |
2796 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2797 | } else { | |
d0d856e8 | 2798 | /* we were already here, see at next index */ |
a86c6181 AT |
2799 | path[i].p_idx--; |
2800 | } | |
2801 | ||
2802 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2803 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2804 | path[i].p_idx); | |
2805 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2806 | struct buffer_head *bh; |
a86c6181 | 2807 | /* go to the next level */ |
2ae02107 | 2808 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2809 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2810 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2811 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2812 | if (!bh) { |
a86c6181 AT |
2813 | /* should we reset i_size? */ |
2814 | err = -EIO; | |
2815 | break; | |
2816 | } | |
c29c0ae7 AT |
2817 | if (WARN_ON(i + 1 > depth)) { |
2818 | err = -EIO; | |
2819 | break; | |
2820 | } | |
f8489128 DW |
2821 | if (ext4_ext_check_block(inode, ext_block_hdr(bh), |
2822 | depth - i - 1, bh)) { | |
c29c0ae7 AT |
2823 | err = -EIO; |
2824 | break; | |
2825 | } | |
2826 | path[i + 1].p_bh = bh; | |
a86c6181 | 2827 | |
d0d856e8 RD |
2828 | /* save actual number of indexes since this |
2829 | * number is changed at the next iteration */ | |
a86c6181 AT |
2830 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2831 | i++; | |
2832 | } else { | |
d0d856e8 | 2833 | /* we finished processing this index, go up */ |
a86c6181 | 2834 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2835 | /* index is empty, remove it; |
a86c6181 AT |
2836 | * handle must be already prepared by the |
2837 | * truncatei_leaf() */ | |
c36575e6 | 2838 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2839 | } |
d0d856e8 | 2840 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2841 | brelse(path[i].p_bh); |
2842 | path[i].p_bh = NULL; | |
2843 | i--; | |
2844 | ext_debug("return to level %d\n", i); | |
2845 | } | |
2846 | } | |
2847 | ||
d8990240 AK |
2848 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2849 | path->p_hdr->eh_entries); | |
2850 | ||
7b415bf6 AK |
2851 | /* If we still have something in the partial cluster and we have removed |
2852 | * even the first extent, then we should free the blocks in the partial | |
2853 | * cluster as well. */ | |
2854 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2855 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2856 | ||
2857 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2858 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2859 | ||
2860 | ext4_free_blocks(handle, inode, NULL, | |
2861 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2862 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2863 | partial_cluster = 0; | |
2864 | } | |
2865 | ||
a86c6181 AT |
2866 | /* TODO: flexible tree reduction should be here */ |
2867 | if (path->p_hdr->eh_entries == 0) { | |
2868 | /* | |
d0d856e8 RD |
2869 | * truncate to zero freed all the tree, |
2870 | * so we need to correct eh_depth | |
a86c6181 AT |
2871 | */ |
2872 | err = ext4_ext_get_access(handle, inode, path); | |
2873 | if (err == 0) { | |
2874 | ext_inode_hdr(inode)->eh_depth = 0; | |
2875 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2876 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2877 | err = ext4_ext_dirty(handle, inode, path); |
2878 | } | |
2879 | } | |
2880 | out: | |
a86c6181 AT |
2881 | ext4_ext_drop_refs(path); |
2882 | kfree(path); | |
968dee77 AS |
2883 | if (err == -EAGAIN) { |
2884 | path = NULL; | |
0617b83f | 2885 | goto again; |
968dee77 | 2886 | } |
a86c6181 AT |
2887 | ext4_journal_stop(handle); |
2888 | ||
2889 | return err; | |
2890 | } | |
2891 | ||
2892 | /* | |
2893 | * called at mount time | |
2894 | */ | |
2895 | void ext4_ext_init(struct super_block *sb) | |
2896 | { | |
2897 | /* | |
2898 | * possible initialization would be here | |
2899 | */ | |
2900 | ||
83982b6f | 2901 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2902 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2903 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2904 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2905 | ", aggressive tests" |
a86c6181 AT |
2906 | #endif |
2907 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2908 | ", check binsearch" |
a86c6181 AT |
2909 | #endif |
2910 | #ifdef EXTENTS_STATS | |
92b97816 | 2911 | ", stats" |
a86c6181 | 2912 | #endif |
92b97816 | 2913 | "\n"); |
90576c0b | 2914 | #endif |
a86c6181 AT |
2915 | #ifdef EXTENTS_STATS |
2916 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2917 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2918 | EXT4_SB(sb)->s_ext_max = 0; | |
2919 | #endif | |
2920 | } | |
2921 | } | |
2922 | ||
2923 | /* | |
2924 | * called at umount time | |
2925 | */ | |
2926 | void ext4_ext_release(struct super_block *sb) | |
2927 | { | |
83982b6f | 2928 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2929 | return; |
2930 | ||
2931 | #ifdef EXTENTS_STATS | |
2932 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2933 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2934 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2935 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2936 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2937 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2938 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2939 | } | |
2940 | #endif | |
2941 | } | |
2942 | ||
093a088b AK |
2943 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2944 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2945 | { | |
2407518d LC |
2946 | ext4_fsblk_t ee_pblock; |
2947 | unsigned int ee_len; | |
b720303d | 2948 | int ret; |
093a088b | 2949 | |
093a088b | 2950 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2951 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2952 | |
a107e5a3 | 2953 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2954 | if (ret > 0) |
2955 | ret = 0; | |
093a088b | 2956 | |
2407518d | 2957 | return ret; |
093a088b AK |
2958 | } |
2959 | ||
47ea3bb5 YY |
2960 | /* |
2961 | * ext4_split_extent_at() splits an extent at given block. | |
2962 | * | |
2963 | * @handle: the journal handle | |
2964 | * @inode: the file inode | |
2965 | * @path: the path to the extent | |
2966 | * @split: the logical block where the extent is splitted. | |
2967 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2968 | * the states(init or uninit) of new extents. | |
2969 | * @flags: flags used to insert new extent to extent tree. | |
2970 | * | |
2971 | * | |
2972 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2973 | * of which are deterimined by split_flag. | |
2974 | * | |
2975 | * There are two cases: | |
2976 | * a> the extent are splitted into two extent. | |
2977 | * b> split is not needed, and just mark the extent. | |
2978 | * | |
2979 | * return 0 on success. | |
2980 | */ | |
2981 | static int ext4_split_extent_at(handle_t *handle, | |
2982 | struct inode *inode, | |
2983 | struct ext4_ext_path *path, | |
2984 | ext4_lblk_t split, | |
2985 | int split_flag, | |
2986 | int flags) | |
2987 | { | |
2988 | ext4_fsblk_t newblock; | |
2989 | ext4_lblk_t ee_block; | |
2990 | struct ext4_extent *ex, newex, orig_ex; | |
2991 | struct ext4_extent *ex2 = NULL; | |
2992 | unsigned int ee_len, depth; | |
2993 | int err = 0; | |
2994 | ||
dee1f973 DM |
2995 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
2996 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
2997 | ||
47ea3bb5 YY |
2998 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
2999 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3000 | ||
3001 | ext4_ext_show_leaf(inode, path); | |
3002 | ||
3003 | depth = ext_depth(inode); | |
3004 | ex = path[depth].p_ext; | |
3005 | ee_block = le32_to_cpu(ex->ee_block); | |
3006 | ee_len = ext4_ext_get_actual_len(ex); | |
3007 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3008 | ||
3009 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
3010 | ||
3011 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3012 | if (err) | |
3013 | goto out; | |
3014 | ||
3015 | if (split == ee_block) { | |
3016 | /* | |
3017 | * case b: block @split is the block that the extent begins with | |
3018 | * then we just change the state of the extent, and splitting | |
3019 | * is not needed. | |
3020 | */ | |
3021 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3022 | ext4_ext_mark_uninitialized(ex); | |
3023 | else | |
3024 | ext4_ext_mark_initialized(ex); | |
3025 | ||
3026 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3027 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3028 | |
ecb94f5f | 3029 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3030 | goto out; |
3031 | } | |
3032 | ||
3033 | /* case a */ | |
3034 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3035 | ex->ee_len = cpu_to_le16(split - ee_block); | |
3036 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
3037 | ext4_ext_mark_uninitialized(ex); | |
3038 | ||
3039 | /* | |
3040 | * path may lead to new leaf, not to original leaf any more | |
3041 | * after ext4_ext_insert_extent() returns, | |
3042 | */ | |
3043 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3044 | if (err) | |
3045 | goto fix_extent_len; | |
3046 | ||
3047 | ex2 = &newex; | |
3048 | ex2->ee_block = cpu_to_le32(split); | |
3049 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3050 | ext4_ext_store_pblock(ex2, newblock); | |
3051 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3052 | ext4_ext_mark_uninitialized(ex2); | |
3053 | ||
3054 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
3055 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 DM |
3056 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
3057 | if (split_flag & EXT4_EXT_DATA_VALID1) | |
3058 | err = ext4_ext_zeroout(inode, ex2); | |
3059 | else | |
3060 | err = ext4_ext_zeroout(inode, ex); | |
3061 | } else | |
3062 | err = ext4_ext_zeroout(inode, &orig_ex); | |
3063 | ||
47ea3bb5 YY |
3064 | if (err) |
3065 | goto fix_extent_len; | |
3066 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3067 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3068 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3069 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
47ea3bb5 YY |
3070 | goto out; |
3071 | } else if (err) | |
3072 | goto fix_extent_len; | |
3073 | ||
3074 | out: | |
3075 | ext4_ext_show_leaf(inode, path); | |
3076 | return err; | |
3077 | ||
3078 | fix_extent_len: | |
3079 | ex->ee_len = orig_ex.ee_len; | |
3080 | ext4_ext_dirty(handle, inode, path + depth); | |
3081 | return err; | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3086 | * by @map as split_flags indicates | |
3087 | * | |
3088 | * It may result in splitting the extent into multiple extents (upto three) | |
3089 | * There are three possibilities: | |
3090 | * a> There is no split required | |
3091 | * b> Splits in two extents: Split is happening at either end of the extent | |
3092 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3093 | * | |
3094 | */ | |
3095 | static int ext4_split_extent(handle_t *handle, | |
3096 | struct inode *inode, | |
3097 | struct ext4_ext_path *path, | |
3098 | struct ext4_map_blocks *map, | |
3099 | int split_flag, | |
3100 | int flags) | |
3101 | { | |
3102 | ext4_lblk_t ee_block; | |
3103 | struct ext4_extent *ex; | |
3104 | unsigned int ee_len, depth; | |
3105 | int err = 0; | |
3106 | int uninitialized; | |
3107 | int split_flag1, flags1; | |
3108 | ||
3109 | depth = ext_depth(inode); | |
3110 | ex = path[depth].p_ext; | |
3111 | ee_block = le32_to_cpu(ex->ee_block); | |
3112 | ee_len = ext4_ext_get_actual_len(ex); | |
3113 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3114 | ||
3115 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3116 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3117 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3118 | if (uninitialized) | |
3119 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3120 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3121 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3122 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3123 | err = ext4_split_extent_at(handle, inode, path, |
3124 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3125 | if (err) |
3126 | goto out; | |
47ea3bb5 YY |
3127 | } |
3128 | ||
3129 | ext4_ext_drop_refs(path); | |
3130 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3131 | if (IS_ERR(path)) | |
3132 | return PTR_ERR(path); | |
3133 | ||
3134 | if (map->m_lblk >= ee_block) { | |
dee1f973 DM |
3135 | split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3136 | EXT4_EXT_DATA_VALID2); | |
47ea3bb5 YY |
3137 | if (uninitialized) |
3138 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; | |
3139 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3140 | split_flag1 |= EXT4_EXT_MARK_UNINIT2; | |
3141 | err = ext4_split_extent_at(handle, inode, path, | |
3142 | map->m_lblk, split_flag1, flags); | |
3143 | if (err) | |
3144 | goto out; | |
3145 | } | |
3146 | ||
3147 | ext4_ext_show_leaf(inode, path); | |
3148 | out: | |
3149 | return err ? err : map->m_len; | |
3150 | } | |
3151 | ||
56055d3a | 3152 | /* |
e35fd660 | 3153 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3154 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3155 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3156 | * uninitialized). |
3157 | * There are three possibilities: | |
3158 | * a> There is no split required: Entire extent should be initialized | |
3159 | * b> Splits in two extents: Write is happening at either end of the extent | |
3160 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3161 | * |
3162 | * Pre-conditions: | |
3163 | * - The extent pointed to by 'path' is uninitialized. | |
3164 | * - The extent pointed to by 'path' contains a superset | |
3165 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3166 | * | |
3167 | * Post-conditions on success: | |
3168 | * - the returned value is the number of blocks beyond map->l_lblk | |
3169 | * that are allocated and initialized. | |
3170 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3171 | */ |
725d26d3 | 3172 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3173 | struct inode *inode, |
3174 | struct ext4_map_blocks *map, | |
3175 | struct ext4_ext_path *path) | |
56055d3a | 3176 | { |
67a5da56 | 3177 | struct ext4_sb_info *sbi; |
6f91bc5f | 3178 | struct ext4_extent_header *eh; |
667eff35 YY |
3179 | struct ext4_map_blocks split_map; |
3180 | struct ext4_extent zero_ex; | |
3181 | struct ext4_extent *ex; | |
21ca087a | 3182 | ext4_lblk_t ee_block, eof_block; |
f85b287a | 3183 | unsigned int ee_len, depth; |
67a5da56 | 3184 | int allocated, max_zeroout = 0; |
56055d3a | 3185 | int err = 0; |
667eff35 | 3186 | int split_flag = 0; |
21ca087a DM |
3187 | |
3188 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3189 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3190 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a | 3191 | |
67a5da56 | 3192 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3193 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3194 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3195 | if (eof_block < map->m_lblk + map->m_len) |
3196 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3197 | |
3198 | depth = ext_depth(inode); | |
6f91bc5f | 3199 | eh = path[depth].p_hdr; |
56055d3a AA |
3200 | ex = path[depth].p_ext; |
3201 | ee_block = le32_to_cpu(ex->ee_block); | |
3202 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3203 | allocated = ee_len - (map->m_lblk - ee_block); |
56055d3a | 3204 | |
6f91bc5f EG |
3205 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3206 | ||
3207 | /* Pre-conditions */ | |
3208 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3209 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3210 | |
3211 | /* | |
3212 | * Attempt to transfer newly initialized blocks from the currently | |
3213 | * uninitialized extent to its left neighbor. This is much cheaper | |
3214 | * than an insertion followed by a merge as those involve costly | |
3215 | * memmove() calls. This is the common case in steady state for | |
3216 | * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append | |
3217 | * writes. | |
3218 | * | |
3219 | * Limitations of the current logic: | |
3220 | * - L1: we only deal with writes at the start of the extent. | |
3221 | * The approach could be extended to writes at the end | |
3222 | * of the extent but this scenario was deemed less common. | |
3223 | * - L2: we do not deal with writes covering the whole extent. | |
3224 | * This would require removing the extent if the transfer | |
3225 | * is possible. | |
3226 | * - L3: we only attempt to merge with an extent stored in the | |
3227 | * same extent tree node. | |
3228 | */ | |
3229 | if ((map->m_lblk == ee_block) && /*L1*/ | |
3230 | (map->m_len < ee_len) && /*L2*/ | |
3231 | (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/ | |
3232 | struct ext4_extent *prev_ex; | |
3233 | ext4_lblk_t prev_lblk; | |
3234 | ext4_fsblk_t prev_pblk, ee_pblk; | |
3235 | unsigned int prev_len, write_len; | |
3236 | ||
3237 | prev_ex = ex - 1; | |
3238 | prev_lblk = le32_to_cpu(prev_ex->ee_block); | |
3239 | prev_len = ext4_ext_get_actual_len(prev_ex); | |
3240 | prev_pblk = ext4_ext_pblock(prev_ex); | |
3241 | ee_pblk = ext4_ext_pblock(ex); | |
3242 | write_len = map->m_len; | |
3243 | ||
3244 | /* | |
3245 | * A transfer of blocks from 'ex' to 'prev_ex' is allowed | |
3246 | * upon those conditions: | |
3247 | * - C1: prev_ex is initialized, | |
3248 | * - C2: prev_ex is logically abutting ex, | |
3249 | * - C3: prev_ex is physically abutting ex, | |
3250 | * - C4: prev_ex can receive the additional blocks without | |
3251 | * overflowing the (initialized) length limit. | |
3252 | */ | |
3253 | if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/ | |
3254 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ | |
3255 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
3256 | (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/ | |
3257 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3258 | if (err) | |
3259 | goto out; | |
3260 | ||
3261 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3262 | map, ex, prev_ex); | |
3263 | ||
3264 | /* Shift the start of ex by 'write_len' blocks */ | |
3265 | ex->ee_block = cpu_to_le32(ee_block + write_len); | |
3266 | ext4_ext_store_pblock(ex, ee_pblk + write_len); | |
3267 | ex->ee_len = cpu_to_le16(ee_len - write_len); | |
3268 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3269 | ||
3270 | /* Extend prev_ex by 'write_len' blocks */ | |
3271 | prev_ex->ee_len = cpu_to_le16(prev_len + write_len); | |
3272 | ||
3273 | /* Mark the block containing both extents as dirty */ | |
3274 | ext4_ext_dirty(handle, inode, path + depth); | |
3275 | ||
3276 | /* Update path to point to the right extent */ | |
3277 | path[depth].p_ext = prev_ex; | |
3278 | ||
3279 | /* Result: number of initialized blocks past m_lblk */ | |
3280 | allocated = write_len; | |
3281 | goto out; | |
3282 | } | |
3283 | } | |
3284 | ||
667eff35 | 3285 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3286 | /* |
3287 | * It is safe to convert extent to initialized via explicit | |
3288 | * zeroout only if extent is fully insde i_size or new_size. | |
3289 | */ | |
667eff35 | 3290 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3291 | |
67a5da56 ZL |
3292 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3293 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
3294 | inode->i_sb->s_blocksize_bits; | |
3295 | ||
3296 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3297 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3298 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3299 | if (err) |
d03856bd | 3300 | goto out; |
d03856bd AK |
3301 | |
3302 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3303 | if (err) | |
3304 | goto out; | |
667eff35 | 3305 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3306 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3307 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3308 | goto out; |
56055d3a | 3309 | } |
667eff35 | 3310 | |
56055d3a | 3311 | /* |
667eff35 YY |
3312 | * four cases: |
3313 | * 1. split the extent into three extents. | |
3314 | * 2. split the extent into two extents, zeroout the first half. | |
3315 | * 3. split the extent into two extents, zeroout the second half. | |
3316 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3317 | */ |
667eff35 YY |
3318 | split_map.m_lblk = map->m_lblk; |
3319 | split_map.m_len = map->m_len; | |
3320 | ||
67a5da56 ZL |
3321 | if (max_zeroout && (allocated > map->m_len)) { |
3322 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3323 | /* case 3 */ |
3324 | zero_ex.ee_block = | |
9b940f8e AH |
3325 | cpu_to_le32(map->m_lblk); |
3326 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3327 | ext4_ext_store_pblock(&zero_ex, |
3328 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3329 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3330 | if (err) |
3331 | goto out; | |
667eff35 YY |
3332 | split_map.m_lblk = map->m_lblk; |
3333 | split_map.m_len = allocated; | |
67a5da56 | 3334 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3335 | /* case 2 */ |
3336 | if (map->m_lblk != ee_block) { | |
3337 | zero_ex.ee_block = ex->ee_block; | |
3338 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3339 | ee_block); | |
3340 | ext4_ext_store_pblock(&zero_ex, | |
3341 | ext4_ext_pblock(ex)); | |
3342 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3343 | if (err) | |
3344 | goto out; | |
3345 | } | |
3346 | ||
667eff35 | 3347 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3348 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3349 | allocated = map->m_len; | |
56055d3a AA |
3350 | } |
3351 | } | |
667eff35 YY |
3352 | |
3353 | allocated = ext4_split_extent(handle, inode, path, | |
67a5da56 | 3354 | &split_map, split_flag, 0); |
667eff35 YY |
3355 | if (allocated < 0) |
3356 | err = allocated; | |
3357 | ||
56055d3a AA |
3358 | out: |
3359 | return err ? err : allocated; | |
3360 | } | |
3361 | ||
0031462b | 3362 | /* |
e35fd660 | 3363 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3364 | * ext4_get_blocks_dio_write() when DIO to write |
3365 | * to an uninitialized extent. | |
3366 | * | |
fd018fe8 | 3367 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3368 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3369 | * There are three possibilities: |
3370 | * a> There is no split required: Entire extent should be uninitialized | |
3371 | * b> Splits in two extents: Write is happening at either end of the extent | |
3372 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3373 | * | |
3374 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3375 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3376 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3377 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3378 | * into three uninitialized extent(at most). After IO complete, the part |
3379 | * being filled will be convert to initialized by the end_io callback function | |
3380 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3381 | * |
3382 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3383 | */ |
3384 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3385 | struct inode *inode, | |
e35fd660 | 3386 | struct ext4_map_blocks *map, |
0031462b | 3387 | struct ext4_ext_path *path, |
0031462b MC |
3388 | int flags) |
3389 | { | |
667eff35 YY |
3390 | ext4_lblk_t eof_block; |
3391 | ext4_lblk_t ee_block; | |
3392 | struct ext4_extent *ex; | |
3393 | unsigned int ee_len; | |
3394 | int split_flag = 0, depth; | |
21ca087a DM |
3395 | |
3396 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3397 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3398 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3399 | |
3400 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3401 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3402 | if (eof_block < map->m_lblk + map->m_len) |
3403 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3404 | /* |
3405 | * It is safe to convert extent to initialized via explicit | |
3406 | * zeroout only if extent is fully insde i_size or new_size. | |
3407 | */ | |
667eff35 YY |
3408 | depth = ext_depth(inode); |
3409 | ex = path[depth].p_ext; | |
3410 | ee_block = le32_to_cpu(ex->ee_block); | |
3411 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3412 | |
667eff35 YY |
3413 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3414 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3415 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3416 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3417 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3418 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3419 | } |
197217a5 | 3420 | |
c7064ef1 | 3421 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3422 | struct inode *inode, |
3423 | struct ext4_map_blocks *map, | |
3424 | struct ext4_ext_path *path) | |
0031462b MC |
3425 | { |
3426 | struct ext4_extent *ex; | |
dee1f973 DM |
3427 | ext4_lblk_t ee_block; |
3428 | unsigned int ee_len; | |
0031462b MC |
3429 | int depth; |
3430 | int err = 0; | |
0031462b MC |
3431 | |
3432 | depth = ext_depth(inode); | |
0031462b | 3433 | ex = path[depth].p_ext; |
dee1f973 DM |
3434 | ee_block = le32_to_cpu(ex->ee_block); |
3435 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3436 | |
197217a5 YY |
3437 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3438 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3439 | (unsigned long long)ee_block, ee_len); |
3440 | ||
3441 | /* If extent is larger than requested then split is required */ | |
3442 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
3443 | err = ext4_split_unwritten_extents(handle, inode, map, path, | |
3444 | EXT4_GET_BLOCKS_CONVERT); | |
3445 | if (err < 0) | |
3446 | goto out; | |
3447 | ext4_ext_drop_refs(path); | |
3448 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3449 | if (IS_ERR(path)) { | |
3450 | err = PTR_ERR(path); | |
3451 | goto out; | |
3452 | } | |
3453 | depth = ext_depth(inode); | |
3454 | ex = path[depth].p_ext; | |
3455 | } | |
197217a5 | 3456 | |
0031462b MC |
3457 | err = ext4_ext_get_access(handle, inode, path + depth); |
3458 | if (err) | |
3459 | goto out; | |
3460 | /* first mark the extent as initialized */ | |
3461 | ext4_ext_mark_initialized(ex); | |
3462 | ||
197217a5 YY |
3463 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3464 | * borders are not changed | |
0031462b | 3465 | */ |
ecb94f5f | 3466 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3467 | |
0031462b | 3468 | /* Mark modified extent as dirty */ |
ecb94f5f | 3469 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3470 | out: |
3471 | ext4_ext_show_leaf(inode, path); | |
3472 | return err; | |
3473 | } | |
3474 | ||
515f41c3 AK |
3475 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3476 | sector_t block, int count) | |
3477 | { | |
3478 | int i; | |
3479 | for (i = 0; i < count; i++) | |
3480 | unmap_underlying_metadata(bdev, block + i); | |
3481 | } | |
3482 | ||
58590b06 TT |
3483 | /* |
3484 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3485 | */ | |
3486 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3487 | ext4_lblk_t lblk, |
58590b06 TT |
3488 | struct ext4_ext_path *path, |
3489 | unsigned int len) | |
3490 | { | |
3491 | int i, depth; | |
3492 | struct ext4_extent_header *eh; | |
65922cb5 | 3493 | struct ext4_extent *last_ex; |
58590b06 TT |
3494 | |
3495 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3496 | return 0; | |
3497 | ||
3498 | depth = ext_depth(inode); | |
3499 | eh = path[depth].p_hdr; | |
58590b06 | 3500 | |
afcff5d8 LC |
3501 | /* |
3502 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3503 | * do not care for this case anymore. Simply remove the flag | |
3504 | * if there are no extents. | |
3505 | */ | |
3506 | if (unlikely(!eh->eh_entries)) | |
3507 | goto out; | |
58590b06 TT |
3508 | last_ex = EXT_LAST_EXTENT(eh); |
3509 | /* | |
3510 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3511 | * last block in the last extent in the file. We test this by | |
3512 | * first checking to see if the caller to | |
3513 | * ext4_ext_get_blocks() was interested in the last block (or | |
3514 | * a block beyond the last block) in the current extent. If | |
3515 | * this turns out to be false, we can bail out from this | |
3516 | * function immediately. | |
3517 | */ | |
d002ebf1 | 3518 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3519 | ext4_ext_get_actual_len(last_ex)) |
3520 | return 0; | |
3521 | /* | |
3522 | * If the caller does appear to be planning to write at or | |
3523 | * beyond the end of the current extent, we then test to see | |
3524 | * if the current extent is the last extent in the file, by | |
3525 | * checking to make sure it was reached via the rightmost node | |
3526 | * at each level of the tree. | |
3527 | */ | |
3528 | for (i = depth-1; i >= 0; i--) | |
3529 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3530 | return 0; | |
afcff5d8 | 3531 | out: |
58590b06 TT |
3532 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3533 | return ext4_mark_inode_dirty(handle, inode); | |
3534 | } | |
3535 | ||
7b415bf6 AK |
3536 | /** |
3537 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3538 | * | |
7d1b1fbc | 3539 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3540 | */ |
f7fec032 ZL |
3541 | int ext4_find_delalloc_range(struct inode *inode, |
3542 | ext4_lblk_t lblk_start, | |
3543 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3544 | { |
7d1b1fbc | 3545 | struct extent_status es; |
7b415bf6 | 3546 | |
be401363 | 3547 | ext4_es_find_delayed_extent(inode, lblk_start, &es); |
06b0c886 | 3548 | if (es.es_len == 0) |
7d1b1fbc | 3549 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3550 | else if (es.es_lblk <= lblk_start && |
3551 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3552 | return 1; |
06b0c886 | 3553 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3554 | return 1; |
7b415bf6 | 3555 | else |
7d1b1fbc | 3556 | return 0; |
7b415bf6 AK |
3557 | } |
3558 | ||
7d1b1fbc | 3559 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3560 | { |
3561 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3562 | ext4_lblk_t lblk_start, lblk_end; | |
3563 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3564 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3565 | ||
7d1b1fbc | 3566 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3567 | } |
3568 | ||
3569 | /** | |
3570 | * Determines how many complete clusters (out of those specified by the 'map') | |
3571 | * are under delalloc and were reserved quota for. | |
3572 | * This function is called when we are writing out the blocks that were | |
3573 | * originally written with their allocation delayed, but then the space was | |
3574 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3575 | * The cases to look for are: | |
3576 | * ('=' indicated delayed allocated blocks | |
3577 | * '-' indicates non-delayed allocated blocks) | |
3578 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3579 | * are not delalloc'ed. | |
3580 | * Ex: | |
3581 | * |----c---=|====c====|====c====|===-c----| | |
3582 | * |++++++ allocated ++++++| | |
3583 | * ==> 4 complete clusters in above example | |
3584 | * | |
3585 | * (b) partial cluster (outside of allocated range) towards either end is | |
3586 | * marked for delayed allocation. In this case, we will exclude that | |
3587 | * cluster. | |
3588 | * Ex: | |
3589 | * |----====c========|========c========| | |
3590 | * |++++++ allocated ++++++| | |
3591 | * ==> 1 complete clusters in above example | |
3592 | * | |
3593 | * Ex: | |
3594 | * |================c================| | |
3595 | * |++++++ allocated ++++++| | |
3596 | * ==> 0 complete clusters in above example | |
3597 | * | |
3598 | * The ext4_da_update_reserve_space will be called only if we | |
3599 | * determine here that there were some "entire" clusters that span | |
3600 | * this 'allocated' range. | |
3601 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3602 | * without ever calling ext4_find_delalloc_range. | |
3603 | */ | |
3604 | static unsigned int | |
3605 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3606 | unsigned int num_blks) | |
3607 | { | |
3608 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3609 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3610 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3611 | unsigned int allocated_clusters = 0; | |
3612 | ||
3613 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3614 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3615 | ||
3616 | /* max possible clusters for this allocation */ | |
3617 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3618 | ||
d8990240 AK |
3619 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3620 | ||
7b415bf6 AK |
3621 | /* Check towards left side */ |
3622 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3623 | if (c_offset) { | |
3624 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3625 | lblk_to = lblk_from + c_offset - 1; | |
3626 | ||
7d1b1fbc | 3627 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3628 | allocated_clusters--; |
3629 | } | |
3630 | ||
3631 | /* Now check towards right. */ | |
3632 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3633 | if (allocated_clusters && c_offset) { | |
3634 | lblk_from = lblk_start + num_blks; | |
3635 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3636 | ||
7d1b1fbc | 3637 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3638 | allocated_clusters--; |
3639 | } | |
3640 | ||
3641 | return allocated_clusters; | |
3642 | } | |
3643 | ||
0031462b MC |
3644 | static int |
3645 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3646 | struct ext4_map_blocks *map, |
0031462b | 3647 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3648 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3649 | { |
3650 | int ret = 0; | |
3651 | int err = 0; | |
f45ee3a1 | 3652 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3653 | |
88635ca2 ZL |
3654 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3655 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3656 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3657 | flags, allocated); |
3658 | ext4_ext_show_leaf(inode, path); | |
3659 | ||
b5645534 ZL |
3660 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3661 | allocated, newblock); | |
d8990240 | 3662 | |
c7064ef1 | 3663 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3664 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3665 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3666 | path, flags); | |
82e54229 DM |
3667 | if (ret <= 0) |
3668 | goto out; | |
5f524950 M |
3669 | /* |
3670 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3671 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3672 | * completed |
3673 | */ | |
0edeb71d TM |
3674 | if (io) |
3675 | ext4_set_io_unwritten_flag(inode, io); | |
3676 | else | |
19f5fb7a | 3677 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
a25a4e1a | 3678 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
744692dc | 3679 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3680 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3681 | goto out; |
3682 | } | |
c7064ef1 | 3683 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3684 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3685 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3686 | path); |
58590b06 | 3687 | if (ret >= 0) { |
b436b9be | 3688 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3689 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3690 | path, map->m_len); | |
58590b06 TT |
3691 | } else |
3692 | err = ret; | |
0031462b MC |
3693 | goto out2; |
3694 | } | |
3695 | /* buffered IO case */ | |
3696 | /* | |
3697 | * repeat fallocate creation request | |
3698 | * we already have an unwritten extent | |
3699 | */ | |
a25a4e1a ZL |
3700 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) { |
3701 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
0031462b | 3702 | goto map_out; |
a25a4e1a | 3703 | } |
0031462b MC |
3704 | |
3705 | /* buffered READ or buffered write_begin() lookup */ | |
3706 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3707 | /* | |
3708 | * We have blocks reserved already. We | |
3709 | * return allocated blocks so that delalloc | |
3710 | * won't do block reservation for us. But | |
3711 | * the buffer head will be unmapped so that | |
3712 | * a read from the block returns 0s. | |
3713 | */ | |
e35fd660 | 3714 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3715 | goto out1; |
3716 | } | |
3717 | ||
3718 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3719 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
a4e5d88b | 3720 | if (ret >= 0) |
b436b9be | 3721 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3722 | out: |
3723 | if (ret <= 0) { | |
3724 | err = ret; | |
3725 | goto out2; | |
3726 | } else | |
3727 | allocated = ret; | |
e35fd660 | 3728 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3729 | /* |
3730 | * if we allocated more blocks than requested | |
3731 | * we need to make sure we unmap the extra block | |
3732 | * allocated. The actual needed block will get | |
3733 | * unmapped later when we find the buffer_head marked | |
3734 | * new. | |
3735 | */ | |
e35fd660 | 3736 | if (allocated > map->m_len) { |
515f41c3 | 3737 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3738 | newblock + map->m_len, |
3739 | allocated - map->m_len); | |
3740 | allocated = map->m_len; | |
515f41c3 | 3741 | } |
5f634d06 AK |
3742 | |
3743 | /* | |
3744 | * If we have done fallocate with the offset that is already | |
3745 | * delayed allocated, we would have block reservation | |
3746 | * and quota reservation done in the delayed write path. | |
3747 | * But fallocate would have already updated quota and block | |
3748 | * count for this offset. So cancel these reservation | |
3749 | */ | |
7b415bf6 AK |
3750 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3751 | unsigned int reserved_clusters; | |
3752 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3753 | map->m_lblk, map->m_len); | |
3754 | if (reserved_clusters) | |
3755 | ext4_da_update_reserve_space(inode, | |
3756 | reserved_clusters, | |
3757 | 0); | |
3758 | } | |
5f634d06 | 3759 | |
0031462b | 3760 | map_out: |
e35fd660 | 3761 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3762 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3763 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3764 | map->m_len); | |
3765 | if (err < 0) | |
3766 | goto out2; | |
3767 | } | |
0031462b | 3768 | out1: |
e35fd660 TT |
3769 | if (allocated > map->m_len) |
3770 | allocated = map->m_len; | |
0031462b | 3771 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3772 | map->m_pblk = newblock; |
3773 | map->m_len = allocated; | |
0031462b MC |
3774 | out2: |
3775 | if (path) { | |
3776 | ext4_ext_drop_refs(path); | |
3777 | kfree(path); | |
3778 | } | |
3779 | return err ? err : allocated; | |
3780 | } | |
58590b06 | 3781 | |
4d33b1ef TT |
3782 | /* |
3783 | * get_implied_cluster_alloc - check to see if the requested | |
3784 | * allocation (in the map structure) overlaps with a cluster already | |
3785 | * allocated in an extent. | |
d8990240 | 3786 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3787 | * @map The requested lblk->pblk mapping |
3788 | * @ex The extent structure which might contain an implied | |
3789 | * cluster allocation | |
3790 | * | |
3791 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3792 | * find blocks that were already in the inode's extent tree. Hence, | |
3793 | * we know that the beginning of the requested region cannot overlap | |
3794 | * the extent from the inode's extent tree. There are three cases we | |
3795 | * want to catch. The first is this case: | |
3796 | * | |
3797 | * |--- cluster # N--| | |
3798 | * |--- extent ---| |---- requested region ---| | |
3799 | * |==========| | |
3800 | * | |
3801 | * The second case that we need to test for is this one: | |
3802 | * | |
3803 | * |--------- cluster # N ----------------| | |
3804 | * |--- requested region --| |------- extent ----| | |
3805 | * |=======================| | |
3806 | * | |
3807 | * The third case is when the requested region lies between two extents | |
3808 | * within the same cluster: | |
3809 | * |------------- cluster # N-------------| | |
3810 | * |----- ex -----| |---- ex_right ----| | |
3811 | * |------ requested region ------| | |
3812 | * |================| | |
3813 | * | |
3814 | * In each of the above cases, we need to set the map->m_pblk and | |
3815 | * map->m_len so it corresponds to the return the extent labelled as | |
3816 | * "|====|" from cluster #N, since it is already in use for data in | |
3817 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3818 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3819 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3820 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3821 | * by calling ext4_mb_new_blocks(). | |
3822 | */ | |
d8990240 | 3823 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3824 | struct ext4_map_blocks *map, |
3825 | struct ext4_extent *ex, | |
3826 | struct ext4_ext_path *path) | |
3827 | { | |
d8990240 | 3828 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3829 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3830 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3831 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3832 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3833 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3834 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3835 | ||
3836 | /* The extent passed in that we are trying to match */ | |
3837 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3838 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3839 | ||
3840 | /* The requested region passed into ext4_map_blocks() */ | |
3841 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3842 | |
3843 | if ((rr_cluster_start == ex_cluster_end) || | |
3844 | (rr_cluster_start == ex_cluster_start)) { | |
3845 | if (rr_cluster_start == ex_cluster_end) | |
3846 | ee_start += ee_len - 1; | |
3847 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3848 | c_offset; | |
3849 | map->m_len = min(map->m_len, | |
3850 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3851 | /* | |
3852 | * Check for and handle this case: | |
3853 | * | |
3854 | * |--------- cluster # N-------------| | |
3855 | * |------- extent ----| | |
3856 | * |--- requested region ---| | |
3857 | * |===========| | |
3858 | */ | |
3859 | ||
3860 | if (map->m_lblk < ee_block) | |
3861 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3862 | ||
3863 | /* | |
3864 | * Check for the case where there is already another allocated | |
3865 | * block to the right of 'ex' but before the end of the cluster. | |
3866 | * | |
3867 | * |------------- cluster # N-------------| | |
3868 | * |----- ex -----| |---- ex_right ----| | |
3869 | * |------ requested region ------| | |
3870 | * |================| | |
3871 | */ | |
3872 | if (map->m_lblk > ee_block) { | |
3873 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3874 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3875 | } | |
d8990240 AK |
3876 | |
3877 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3878 | return 1; |
3879 | } | |
d8990240 AK |
3880 | |
3881 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3882 | return 0; |
3883 | } | |
3884 | ||
3885 | ||
c278bfec | 3886 | /* |
f5ab0d1f MC |
3887 | * Block allocation/map/preallocation routine for extents based files |
3888 | * | |
3889 | * | |
c278bfec | 3890 | * Need to be called with |
0e855ac8 AK |
3891 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3892 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3893 | * |
3894 | * return > 0, number of of blocks already mapped/allocated | |
3895 | * if create == 0 and these are pre-allocated blocks | |
3896 | * buffer head is unmapped | |
3897 | * otherwise blocks are mapped | |
3898 | * | |
3899 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3900 | * buffer head is unmapped | |
3901 | * | |
3902 | * return < 0, error case. | |
c278bfec | 3903 | */ |
e35fd660 TT |
3904 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3905 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3906 | { |
3907 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3908 | struct ext4_extent newex, *ex, *ex2; |
3909 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3910 | ext4_fsblk_t newblock = 0; |
37794732 | 3911 | int free_on_err = 0, err = 0, depth; |
4d33b1ef | 3912 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 3913 | unsigned int allocated_clusters = 0; |
c9de560d | 3914 | struct ext4_allocation_request ar; |
f45ee3a1 | 3915 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 3916 | ext4_lblk_t cluster_offset; |
82e54229 | 3917 | int set_unwritten = 0; |
a86c6181 | 3918 | |
84fe3bef | 3919 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3920 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3921 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 AT |
3922 | |
3923 | /* check in cache */ | |
7877191c | 3924 | if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) { |
b05e6ae5 | 3925 | if (!newex.ee_start_lo && !newex.ee_start_hi) { |
7b415bf6 | 3926 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 3927 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
3928 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
3929 | ||
c2177057 | 3930 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3931 | /* |
3932 | * block isn't allocated yet and | |
3933 | * user doesn't want to allocate it | |
3934 | */ | |
a86c6181 AT |
3935 | goto out2; |
3936 | } | |
3937 | /* we should allocate requested block */ | |
b05e6ae5 | 3938 | } else { |
a86c6181 | 3939 | /* block is already allocated */ |
7b415bf6 AK |
3940 | if (sbi->s_cluster_ratio > 1) |
3941 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
e35fd660 | 3942 | newblock = map->m_lblk |
8c55e204 | 3943 | - le32_to_cpu(newex.ee_block) |
bf89d16f | 3944 | + ext4_ext_pblock(&newex); |
d0d856e8 | 3945 | /* number of remaining blocks in the extent */ |
b939e376 | 3946 | allocated = ext4_ext_get_actual_len(&newex) - |
e35fd660 | 3947 | (map->m_lblk - le32_to_cpu(newex.ee_block)); |
a86c6181 | 3948 | goto out; |
a86c6181 AT |
3949 | } |
3950 | } | |
3951 | ||
3952 | /* find extent for this block */ | |
e35fd660 | 3953 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3954 | if (IS_ERR(path)) { |
3955 | err = PTR_ERR(path); | |
3956 | path = NULL; | |
3957 | goto out2; | |
3958 | } | |
3959 | ||
3960 | depth = ext_depth(inode); | |
3961 | ||
3962 | /* | |
d0d856e8 RD |
3963 | * consistent leaf must not be empty; |
3964 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3965 | * this is why assert can't be put in ext4_ext_find_extent() |
3966 | */ | |
273df556 FM |
3967 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3968 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3969 | "lblock: %lu, depth: %d pblock %lld", |
3970 | (unsigned long) map->m_lblk, depth, | |
3971 | path[depth].p_block); | |
034fb4c9 SP |
3972 | err = -EIO; |
3973 | goto out2; | |
3974 | } | |
a86c6181 | 3975 | |
7e028976 AM |
3976 | ex = path[depth].p_ext; |
3977 | if (ex) { | |
725d26d3 | 3978 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3979 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3980 | unsigned short ee_len; |
471d4011 SB |
3981 | |
3982 | /* | |
471d4011 | 3983 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3984 | * we split out initialized portions during a write. |
471d4011 | 3985 | */ |
a2df2a63 | 3986 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
3987 | |
3988 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
3989 | ||
d0d856e8 | 3990 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
3991 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
3992 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 3993 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3994 | allocated = ee_len - (map->m_lblk - ee_block); |
3995 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3996 | ee_block, ee_len, newblock); | |
56055d3a | 3997 | |
e861304b | 3998 | /* |
7877191c LC |
3999 | * Do not put uninitialized extent |
4000 | * in the cache | |
e861304b | 4001 | */ |
7877191c LC |
4002 | if (!ext4_ext_is_uninitialized(ex)) { |
4003 | ext4_ext_put_in_cache(inode, ee_block, | |
4004 | ee_len, ee_start); | |
4005 | goto out; | |
f7d0d379 | 4006 | } |
37794732 | 4007 | allocated = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
4008 | handle, inode, map, path, flags, |
4009 | allocated, newblock); | |
37794732 | 4010 | goto out3; |
a86c6181 AT |
4011 | } |
4012 | } | |
4013 | ||
7b415bf6 | 4014 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 4015 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
4016 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4017 | ||
a86c6181 | 4018 | /* |
d0d856e8 | 4019 | * requested block isn't allocated yet; |
a86c6181 AT |
4020 | * we couldn't try to create block if create flag is zero |
4021 | */ | |
c2177057 | 4022 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4023 | /* |
4024 | * put just found gap into cache to speed up | |
4025 | * subsequent requests | |
4026 | */ | |
d100eef2 ZL |
4027 | if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0) |
4028 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); | |
a86c6181 AT |
4029 | goto out2; |
4030 | } | |
4d33b1ef | 4031 | |
a86c6181 | 4032 | /* |
c2ea3fde | 4033 | * Okay, we need to do block allocation. |
63f57933 | 4034 | */ |
7b415bf6 | 4035 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4036 | newex.ee_block = cpu_to_le32(map->m_lblk); |
4037 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
4038 | ||
4039 | /* | |
4040 | * If we are doing bigalloc, check to see if the extent returned | |
4041 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4042 | */ | |
4043 | if (cluster_offset && ex && | |
d8990240 | 4044 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4045 | ar.len = allocated = map->m_len; |
4046 | newblock = map->m_pblk; | |
7b415bf6 | 4047 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4048 | goto got_allocated_blocks; |
4049 | } | |
a86c6181 | 4050 | |
c9de560d | 4051 | /* find neighbour allocated blocks */ |
e35fd660 | 4052 | ar.lleft = map->m_lblk; |
c9de560d AT |
4053 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4054 | if (err) | |
4055 | goto out2; | |
e35fd660 | 4056 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4057 | ex2 = NULL; |
4058 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4059 | if (err) |
4060 | goto out2; | |
25d14f98 | 4061 | |
4d33b1ef TT |
4062 | /* Check if the extent after searching to the right implies a |
4063 | * cluster we can use. */ | |
4064 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4065 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4066 | ar.len = allocated = map->m_len; |
4067 | newblock = map->m_pblk; | |
7b415bf6 | 4068 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4069 | goto got_allocated_blocks; |
4070 | } | |
4071 | ||
749269fa AA |
4072 | /* |
4073 | * See if request is beyond maximum number of blocks we can have in | |
4074 | * a single extent. For an initialized extent this limit is | |
4075 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4076 | * EXT_UNINIT_MAX_LEN. | |
4077 | */ | |
e35fd660 | 4078 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4079 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4080 | map->m_len = EXT_INIT_MAX_LEN; |
4081 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4082 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4083 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4084 | |
e35fd660 | 4085 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4086 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4087 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4088 | if (err) |
b939e376 | 4089 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4090 | else |
e35fd660 | 4091 | allocated = map->m_len; |
c9de560d AT |
4092 | |
4093 | /* allocate new block */ | |
4094 | ar.inode = inode; | |
e35fd660 TT |
4095 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4096 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4097 | /* |
4098 | * We calculate the offset from the beginning of the cluster | |
4099 | * for the logical block number, since when we allocate a | |
4100 | * physical cluster, the physical block should start at the | |
4101 | * same offset from the beginning of the cluster. This is | |
4102 | * needed so that future calls to get_implied_cluster_alloc() | |
4103 | * work correctly. | |
4104 | */ | |
4105 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4106 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4107 | ar.goal -= offset; | |
4108 | ar.logical -= offset; | |
c9de560d AT |
4109 | if (S_ISREG(inode->i_mode)) |
4110 | ar.flags = EXT4_MB_HINT_DATA; | |
4111 | else | |
4112 | /* disable in-core preallocation for non-regular files */ | |
4113 | ar.flags = 0; | |
556b27ab VH |
4114 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4115 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4116 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4117 | if (!newblock) |
4118 | goto out2; | |
84fe3bef | 4119 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4120 | ar.goal, newblock, allocated); |
4d33b1ef | 4121 | free_on_err = 1; |
7b415bf6 | 4122 | allocated_clusters = ar.len; |
4d33b1ef TT |
4123 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4124 | if (ar.len > allocated) | |
4125 | ar.len = allocated; | |
a86c6181 | 4126 | |
4d33b1ef | 4127 | got_allocated_blocks: |
a86c6181 | 4128 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4129 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4130 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4131 | /* Mark uninitialized */ |
4132 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4133 | ext4_ext_mark_uninitialized(&newex); |
a25a4e1a | 4134 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4135 | /* |
744692dc | 4136 | * io_end structure was created for every IO write to an |
25985edc | 4137 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4138 | * here we flag the IO that really needs the conversion. |
5f524950 | 4139 | * For non asycn direct IO case, flag the inode state |
25985edc | 4140 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4141 | */ |
82e54229 DM |
4142 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4143 | set_unwritten = 1; | |
744692dc | 4144 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4145 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4146 | } |
c8d46e41 | 4147 | |
a4e5d88b DM |
4148 | err = 0; |
4149 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4150 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4151 | path, ar.len); | |
575a1d4b JZ |
4152 | if (!err) |
4153 | err = ext4_ext_insert_extent(handle, inode, path, | |
4154 | &newex, flags); | |
82e54229 DM |
4155 | |
4156 | if (!err && set_unwritten) { | |
4157 | if (io) | |
4158 | ext4_set_io_unwritten_flag(inode, io); | |
4159 | else | |
4160 | ext4_set_inode_state(inode, | |
4161 | EXT4_STATE_DIO_UNWRITTEN); | |
4162 | } | |
4163 | ||
4d33b1ef | 4164 | if (err && free_on_err) { |
7132de74 MP |
4165 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4166 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4167 | /* free data blocks we just allocated */ |
c9de560d AT |
4168 | /* not a good idea to call discard here directly, |
4169 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4170 | ext4_discard_preallocations(inode); |
7dc57615 | 4171 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4172 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4173 | goto out2; |
315054f0 | 4174 | } |
a86c6181 | 4175 | |
a86c6181 | 4176 | /* previous routine could use block we allocated */ |
bf89d16f | 4177 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4178 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4179 | if (allocated > map->m_len) |
4180 | allocated = map->m_len; | |
4181 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4182 | |
5f634d06 AK |
4183 | /* |
4184 | * Update reserved blocks/metadata blocks after successful | |
4185 | * block allocation which had been deferred till now. | |
4186 | */ | |
7b415bf6 | 4187 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4188 | unsigned int reserved_clusters; |
7b415bf6 | 4189 | /* |
81fdbb4a | 4190 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4191 | */ |
4192 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4193 | map->m_lblk, allocated); | |
4194 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4195 | if (reserved_clusters) { | |
4196 | /* | |
4197 | * We have clusters reserved for this range. | |
4198 | * But since we are not doing actual allocation | |
4199 | * and are simply using blocks from previously | |
4200 | * allocated cluster, we should release the | |
4201 | * reservation and not claim quota. | |
4202 | */ | |
4203 | ext4_da_update_reserve_space(inode, | |
4204 | reserved_clusters, 0); | |
4205 | } | |
4206 | } else { | |
4207 | BUG_ON(allocated_clusters < reserved_clusters); | |
4208 | /* We will claim quota for all newly allocated blocks.*/ | |
4209 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4210 | 1); | |
4211 | if (reserved_clusters < allocated_clusters) { | |
5356f261 | 4212 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4213 | int reservation = allocated_clusters - |
4214 | reserved_clusters; | |
4215 | /* | |
4216 | * It seems we claimed few clusters outside of | |
4217 | * the range of this allocation. We should give | |
4218 | * it back to the reservation pool. This can | |
4219 | * happen in the following case: | |
4220 | * | |
4221 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4222 | * cluster has 4 blocks. Thus, the clusters | |
4223 | * are [0-3],[4-7],[8-11]... | |
4224 | * * First comes delayed allocation write for | |
4225 | * logical blocks 10 & 11. Since there were no | |
4226 | * previous delayed allocated blocks in the | |
4227 | * range [8-11], we would reserve 1 cluster | |
4228 | * for this write. | |
4229 | * * Next comes write for logical blocks 3 to 8. | |
4230 | * In this case, we will reserve 2 clusters | |
4231 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4232 | * that range has a delayed allocated blocks. | |
4233 | * Thus total reserved clusters now becomes 3. | |
4234 | * * Now, during the delayed allocation writeout | |
4235 | * time, we will first write blocks [3-8] and | |
4236 | * allocate 3 clusters for writing these | |
4237 | * blocks. Also, we would claim all these | |
4238 | * three clusters above. | |
4239 | * * Now when we come here to writeout the | |
4240 | * blocks [10-11], we would expect to claim | |
4241 | * the reservation of 1 cluster we had made | |
4242 | * (and we would claim it since there are no | |
4243 | * more delayed allocated blocks in the range | |
4244 | * [8-11]. But our reserved cluster count had | |
4245 | * already gone to 0. | |
4246 | * | |
4247 | * Thus, at the step 4 above when we determine | |
4248 | * that there are still some unwritten delayed | |
4249 | * allocated blocks outside of our current | |
4250 | * block range, we should increment the | |
4251 | * reserved clusters count so that when the | |
4252 | * remaining blocks finally gets written, we | |
4253 | * could claim them. | |
4254 | */ | |
5356f261 AK |
4255 | dquot_reserve_block(inode, |
4256 | EXT4_C2B(sbi, reservation)); | |
4257 | spin_lock(&ei->i_block_reservation_lock); | |
4258 | ei->i_reserved_data_blocks += reservation; | |
4259 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 AK |
4260 | } |
4261 | } | |
4262 | } | |
5f634d06 | 4263 | |
b436b9be JK |
4264 | /* |
4265 | * Cache the extent and update transaction to commit on fdatasync only | |
4266 | * when it is _not_ an uninitialized extent. | |
4267 | */ | |
4268 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) { | |
b05e6ae5 | 4269 | ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock); |
b436b9be JK |
4270 | ext4_update_inode_fsync_trans(handle, inode, 1); |
4271 | } else | |
4272 | ext4_update_inode_fsync_trans(handle, inode, 0); | |
a86c6181 | 4273 | out: |
e35fd660 TT |
4274 | if (allocated > map->m_len) |
4275 | allocated = map->m_len; | |
a86c6181 | 4276 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4277 | map->m_flags |= EXT4_MAP_MAPPED; |
4278 | map->m_pblk = newblock; | |
4279 | map->m_len = allocated; | |
a86c6181 AT |
4280 | out2: |
4281 | if (path) { | |
4282 | ext4_ext_drop_refs(path); | |
4283 | kfree(path); | |
4284 | } | |
e861304b | 4285 | |
37794732 | 4286 | out3: |
19b303d8 | 4287 | trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated); |
e7b319e3 | 4288 | |
7877191c | 4289 | return err ? err : allocated; |
a86c6181 AT |
4290 | } |
4291 | ||
cf108bca | 4292 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
4293 | { |
4294 | struct address_space *mapping = inode->i_mapping; | |
4295 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 4296 | ext4_lblk_t last_block; |
a86c6181 | 4297 | handle_t *handle; |
189e868f | 4298 | loff_t page_len; |
a86c6181 AT |
4299 | int err = 0; |
4300 | ||
3889fd57 JZ |
4301 | /* |
4302 | * finish any pending end_io work so we won't run the risk of | |
4303 | * converting any truncated blocks to initialized later | |
4304 | */ | |
c278531d | 4305 | ext4_flush_unwritten_io(inode); |
3889fd57 | 4306 | |
a86c6181 AT |
4307 | /* |
4308 | * probably first extent we're gonna free will be last in block | |
4309 | */ | |
f3bd1f3f | 4310 | err = ext4_writepage_trans_blocks(inode); |
9924a92a | 4311 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, err); |
cf108bca | 4312 | if (IS_ERR(handle)) |
a86c6181 | 4313 | return; |
a86c6181 | 4314 | |
189e868f AH |
4315 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
4316 | page_len = PAGE_CACHE_SIZE - | |
4317 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4318 | ||
4319 | err = ext4_discard_partial_page_buffers(handle, | |
4320 | mapping, inode->i_size, page_len, 0); | |
4321 | ||
4322 | if (err) | |
4323 | goto out_stop; | |
4324 | } | |
a86c6181 | 4325 | |
9ddfc3dc JK |
4326 | if (ext4_orphan_add(handle, inode)) |
4327 | goto out_stop; | |
4328 | ||
0e855ac8 | 4329 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
4330 | ext4_ext_invalidate_cache(inode); |
4331 | ||
c2ea3fde | 4332 | ext4_discard_preallocations(inode); |
c9de560d | 4333 | |
a86c6181 | 4334 | /* |
d0d856e8 RD |
4335 | * TODO: optimization is possible here. |
4336 | * Probably we need not scan at all, | |
4337 | * because page truncation is enough. | |
a86c6181 | 4338 | */ |
a86c6181 AT |
4339 | |
4340 | /* we have to know where to truncate from in crash case */ | |
4341 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4342 | ext4_mark_inode_dirty(handle, inode); | |
4343 | ||
4344 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4345 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
51865fda ZL |
4346 | err = ext4_es_remove_extent(inode, last_block, |
4347 | EXT_MAX_BLOCKS - last_block); | |
5f95d21f | 4348 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
a86c6181 AT |
4349 | |
4350 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
4351 | * transaction synchronous. |
4352 | */ | |
a86c6181 | 4353 | if (IS_SYNC(inode)) |
0390131b | 4354 | ext4_handle_sync(handle); |
a86c6181 | 4355 | |
9ddfc3dc | 4356 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
4357 | |
4358 | out_stop: | |
a86c6181 | 4359 | /* |
d0d856e8 | 4360 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
4361 | * then we need to clear up the orphan record which we created above. |
4362 | * However, if this was a real unlink then we were called by | |
4363 | * ext4_delete_inode(), and we allow that function to clean up the | |
4364 | * orphan info for us. | |
4365 | */ | |
4366 | if (inode->i_nlink) | |
4367 | ext4_orphan_del(handle, inode); | |
4368 | ||
ef737728 SR |
4369 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4370 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
4371 | ext4_journal_stop(handle); |
4372 | } | |
4373 | ||
fd28784a AK |
4374 | static void ext4_falloc_update_inode(struct inode *inode, |
4375 | int mode, loff_t new_size, int update_ctime) | |
4376 | { | |
4377 | struct timespec now; | |
4378 | ||
4379 | if (update_ctime) { | |
4380 | now = current_fs_time(inode->i_sb); | |
4381 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4382 | inode->i_ctime = now; | |
4383 | } | |
4384 | /* | |
4385 | * Update only when preallocation was requested beyond | |
4386 | * the file size. | |
4387 | */ | |
cf17fea6 AK |
4388 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4389 | if (new_size > i_size_read(inode)) | |
4390 | i_size_write(inode, new_size); | |
4391 | if (new_size > EXT4_I(inode)->i_disksize) | |
4392 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4393 | } else { |
4394 | /* | |
4395 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4396 | * can proceed even if the new size is the same as i_size. | |
4397 | */ | |
4398 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4399 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4400 | } |
4401 | ||
4402 | } | |
4403 | ||
a2df2a63 | 4404 | /* |
2fe17c10 | 4405 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4406 | * operation, which gets called from sys_fallocate system call. |
4407 | * For block-mapped files, posix_fallocate should fall back to the method | |
4408 | * of writing zeroes to the required new blocks (the same behavior which is | |
4409 | * expected for file systems which do not support fallocate() system call). | |
4410 | */ | |
2fe17c10 | 4411 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4412 | { |
2fe17c10 | 4413 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 4414 | handle_t *handle; |
fd28784a | 4415 | loff_t new_size; |
498e5f24 | 4416 | unsigned int max_blocks; |
a2df2a63 AA |
4417 | int ret = 0; |
4418 | int ret2 = 0; | |
4419 | int retries = 0; | |
a4e5d88b | 4420 | int flags; |
2ed88685 | 4421 | struct ext4_map_blocks map; |
a2df2a63 AA |
4422 | unsigned int credits, blkbits = inode->i_blkbits; |
4423 | ||
a4bb6b64 AH |
4424 | /* Return error if mode is not supported */ |
4425 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4426 | return -EOPNOTSUPP; | |
4427 | ||
4428 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4429 | return ext4_punch_hole(file, offset, len); | |
4430 | ||
0c8d414f TM |
4431 | ret = ext4_convert_inline_data(inode); |
4432 | if (ret) | |
4433 | return ret; | |
4434 | ||
8bad6fc8 ZL |
4435 | /* |
4436 | * currently supporting (pre)allocate mode for extent-based | |
4437 | * files _only_ | |
4438 | */ | |
4439 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4440 | return -EOPNOTSUPP; | |
4441 | ||
0562e0ba | 4442 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4443 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4444 | /* |
4445 | * We can't just convert len to max_blocks because | |
4446 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4447 | */ | |
a2df2a63 | 4448 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4449 | - map.m_lblk; |
a2df2a63 | 4450 | /* |
f3bd1f3f | 4451 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4452 | */ |
f3bd1f3f | 4453 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4454 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4455 | ret = inode_newsize_ok(inode, (len + offset)); |
4456 | if (ret) { | |
4457 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4458 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4459 | return ret; |
4460 | } | |
3c6fe770 | 4461 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4462 | if (mode & FALLOC_FL_KEEP_SIZE) |
4463 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4464 | /* |
4465 | * Don't normalize the request if it can fit in one extent so | |
4466 | * that it doesn't get unnecessarily split into multiple | |
4467 | * extents. | |
4468 | */ | |
4469 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4470 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f DM |
4471 | |
4472 | /* Prevent race condition between unwritten */ | |
4473 | ext4_flush_unwritten_io(inode); | |
a2df2a63 AA |
4474 | retry: |
4475 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4476 | map.m_lblk = map.m_lblk + ret; |
4477 | map.m_len = max_blocks = max_blocks - ret; | |
9924a92a TT |
4478 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4479 | credits); | |
a2df2a63 AA |
4480 | if (IS_ERR(handle)) { |
4481 | ret = PTR_ERR(handle); | |
4482 | break; | |
4483 | } | |
a4e5d88b | 4484 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4485 | if (ret <= 0) { |
2c98615d | 4486 | #ifdef EXT4FS_DEBUG |
b06acd38 LC |
4487 | ext4_warning(inode->i_sb, |
4488 | "inode #%lu: block %u: len %u: " | |
4489 | "ext4_ext_map_blocks returned %d", | |
4490 | inode->i_ino, map.m_lblk, | |
4491 | map.m_len, ret); | |
2c98615d | 4492 | #endif |
a2df2a63 AA |
4493 | ext4_mark_inode_dirty(handle, inode); |
4494 | ret2 = ext4_journal_stop(handle); | |
4495 | break; | |
4496 | } | |
2ed88685 | 4497 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4498 | blkbits) >> blkbits)) |
4499 | new_size = offset + len; | |
4500 | else | |
29ae07b7 | 4501 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4502 | |
fd28784a | 4503 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4504 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4505 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4506 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4507 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4508 | ret2 = ext4_journal_stop(handle); |
4509 | if (ret2) | |
4510 | break; | |
4511 | } | |
fd28784a AK |
4512 | if (ret == -ENOSPC && |
4513 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4514 | ret = 0; | |
a2df2a63 | 4515 | goto retry; |
a2df2a63 | 4516 | } |
55bd725a | 4517 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4518 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4519 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4520 | return ret > 0 ? ret2 : ret; |
4521 | } | |
6873fa0d | 4522 | |
0031462b MC |
4523 | /* |
4524 | * This function convert a range of blocks to written extents | |
4525 | * The caller of this function will pass the start offset and the size. | |
4526 | * all unwritten extents within this range will be converted to | |
4527 | * written extents. | |
4528 | * | |
4529 | * This function is called from the direct IO end io call back | |
4530 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4531 | * Returns 0 on success. |
0031462b MC |
4532 | */ |
4533 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4534 | ssize_t len) |
0031462b MC |
4535 | { |
4536 | handle_t *handle; | |
0031462b MC |
4537 | unsigned int max_blocks; |
4538 | int ret = 0; | |
4539 | int ret2 = 0; | |
2ed88685 | 4540 | struct ext4_map_blocks map; |
0031462b MC |
4541 | unsigned int credits, blkbits = inode->i_blkbits; |
4542 | ||
2ed88685 | 4543 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4544 | /* |
4545 | * We can't just convert len to max_blocks because | |
4546 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4547 | */ | |
2ed88685 TT |
4548 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4549 | map.m_lblk); | |
0031462b MC |
4550 | /* |
4551 | * credits to insert 1 extent into extent tree | |
4552 | */ | |
4553 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4554 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4555 | map.m_lblk += ret; |
4556 | map.m_len = (max_blocks -= ret); | |
9924a92a | 4557 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits); |
0031462b MC |
4558 | if (IS_ERR(handle)) { |
4559 | ret = PTR_ERR(handle); | |
4560 | break; | |
4561 | } | |
2ed88685 | 4562 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4563 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4564 | if (ret <= 0) |
4565 | ext4_warning(inode->i_sb, | |
4566 | "inode #%lu: block %u: len %u: " | |
4567 | "ext4_ext_map_blocks returned %d", | |
4568 | inode->i_ino, map.m_lblk, | |
4569 | map.m_len, ret); | |
0031462b MC |
4570 | ext4_mark_inode_dirty(handle, inode); |
4571 | ret2 = ext4_journal_stop(handle); | |
4572 | if (ret <= 0 || ret2 ) | |
4573 | break; | |
4574 | } | |
4575 | return ret > 0 ? ret2 : ret; | |
4576 | } | |
6d9c85eb | 4577 | |
6873fa0d | 4578 | /* |
91dd8c11 LC |
4579 | * If newex is not existing extent (newex->ec_start equals zero) find |
4580 | * delayed extent at start of newex and update newex accordingly and | |
4581 | * return start of the next delayed extent. | |
4582 | * | |
4583 | * If newex is existing extent (newex->ec_start is not equal zero) | |
4584 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed | |
4585 | * extent found. Leave newex unmodified. | |
6873fa0d | 4586 | */ |
91dd8c11 LC |
4587 | static int ext4_find_delayed_extent(struct inode *inode, |
4588 | struct ext4_ext_cache *newex) | |
6873fa0d | 4589 | { |
b3aff3e3 | 4590 | struct extent_status es; |
be401363 | 4591 | ext4_lblk_t block, next_del; |
6873fa0d | 4592 | |
be401363 | 4593 | ext4_es_find_delayed_extent(inode, newex->ec_block, &es); |
6873fa0d | 4594 | |
b05e6ae5 | 4595 | if (newex->ec_start == 0) { |
6d9c85eb YY |
4596 | /* |
4597 | * No extent in extent-tree contains block @newex->ec_start, | |
4598 | * then the block may stay in 1)a hole or 2)delayed-extent. | |
6d9c85eb | 4599 | */ |
06b0c886 | 4600 | if (es.es_len == 0) |
b3aff3e3 | 4601 | /* A hole found. */ |
91dd8c11 | 4602 | return 0; |
b3aff3e3 | 4603 | |
06b0c886 | 4604 | if (es.es_lblk > newex->ec_block) { |
b3aff3e3 | 4605 | /* A hole found. */ |
06b0c886 | 4606 | newex->ec_len = min(es.es_lblk - newex->ec_block, |
b3aff3e3 | 4607 | newex->ec_len); |
91dd8c11 | 4608 | return 0; |
6873fa0d | 4609 | } |
6d9c85eb | 4610 | |
06b0c886 | 4611 | newex->ec_len = es.es_lblk + es.es_len - newex->ec_block; |
6873fa0d ES |
4612 | } |
4613 | ||
be401363 ZL |
4614 | block = newex->ec_block + newex->ec_len; |
4615 | ext4_es_find_delayed_extent(inode, block, &es); | |
4616 | if (es.es_len == 0) | |
4617 | next_del = EXT_MAX_BLOCKS; | |
4618 | else | |
4619 | next_del = es.es_lblk; | |
4620 | ||
91dd8c11 | 4621 | return next_del; |
6873fa0d | 4622 | } |
6873fa0d ES |
4623 | /* fiemap flags we can handle specified here */ |
4624 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4625 | ||
3a06d778 AK |
4626 | static int ext4_xattr_fiemap(struct inode *inode, |
4627 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4628 | { |
4629 | __u64 physical = 0; | |
4630 | __u64 length; | |
4631 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4632 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4633 | int error = 0; | |
4634 | ||
4635 | /* in-inode? */ | |
19f5fb7a | 4636 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4637 | struct ext4_iloc iloc; |
4638 | int offset; /* offset of xattr in inode */ | |
4639 | ||
4640 | error = ext4_get_inode_loc(inode, &iloc); | |
4641 | if (error) | |
4642 | return error; | |
4643 | physical = iloc.bh->b_blocknr << blockbits; | |
4644 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4645 | EXT4_I(inode)->i_extra_isize; | |
4646 | physical += offset; | |
4647 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4648 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4649 | brelse(iloc.bh); |
6873fa0d ES |
4650 | } else { /* external block */ |
4651 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4652 | length = inode->i_sb->s_blocksize; | |
4653 | } | |
4654 | ||
4655 | if (physical) | |
4656 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4657 | length, flags); | |
4658 | return (error < 0 ? error : 0); | |
4659 | } | |
4660 | ||
a4bb6b64 AH |
4661 | /* |
4662 | * ext4_ext_punch_hole | |
4663 | * | |
4664 | * Punches a hole of "length" bytes in a file starting | |
4665 | * at byte "offset" | |
4666 | * | |
4667 | * @inode: The inode of the file to punch a hole in | |
4668 | * @offset: The starting byte offset of the hole | |
4669 | * @length: The length of the hole | |
4670 | * | |
4671 | * Returns the number of blocks removed or negative on err | |
4672 | */ | |
4673 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4674 | { | |
4675 | struct inode *inode = file->f_path.dentry->d_inode; | |
4676 | struct super_block *sb = inode->i_sb; | |
5f95d21f | 4677 | ext4_lblk_t first_block, stop_block; |
a4bb6b64 | 4678 | struct address_space *mapping = inode->i_mapping; |
a4bb6b64 | 4679 | handle_t *handle; |
ba06208a AH |
4680 | loff_t first_page, last_page, page_len; |
4681 | loff_t first_page_offset, last_page_offset; | |
5f95d21f | 4682 | int credits, err = 0; |
a4bb6b64 | 4683 | |
02d262df DM |
4684 | /* |
4685 | * Write out all dirty pages to avoid race conditions | |
4686 | * Then release them. | |
4687 | */ | |
4688 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4689 | err = filemap_write_and_wait_range(mapping, | |
4690 | offset, offset + length - 1); | |
4691 | ||
4692 | if (err) | |
4693 | return err; | |
4694 | } | |
4695 | ||
4696 | mutex_lock(&inode->i_mutex); | |
4697 | /* It's not possible punch hole on append only file */ | |
4698 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
4699 | err = -EPERM; | |
4700 | goto out_mutex; | |
4701 | } | |
4702 | if (IS_SWAPFILE(inode)) { | |
4703 | err = -ETXTBSY; | |
4704 | goto out_mutex; | |
4705 | } | |
4706 | ||
2be4751b AH |
4707 | /* No need to punch hole beyond i_size */ |
4708 | if (offset >= inode->i_size) | |
02d262df | 4709 | goto out_mutex; |
2be4751b AH |
4710 | |
4711 | /* | |
4712 | * If the hole extends beyond i_size, set the hole | |
4713 | * to end after the page that contains i_size | |
4714 | */ | |
4715 | if (offset + length > inode->i_size) { | |
4716 | length = inode->i_size + | |
4717 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4718 | offset; | |
4719 | } | |
4720 | ||
a4bb6b64 AH |
4721 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4722 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4723 | ||
4724 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4725 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4726 | ||
a4bb6b64 AH |
4727 | /* Now release the pages */ |
4728 | if (last_page_offset > first_page_offset) { | |
5e44f8c3 HD |
4729 | truncate_pagecache_range(inode, first_page_offset, |
4730 | last_page_offset - 1); | |
a4bb6b64 AH |
4731 | } |
4732 | ||
02d262df DM |
4733 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
4734 | ext4_inode_block_unlocked_dio(inode); | |
c278531d | 4735 | err = ext4_flush_unwritten_io(inode); |
28a535f9 | 4736 | if (err) |
02d262df | 4737 | goto out_dio; |
c278531d | 4738 | inode_dio_wait(inode); |
a4bb6b64 AH |
4739 | |
4740 | credits = ext4_writepage_trans_blocks(inode); | |
9924a92a | 4741 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
02d262df DM |
4742 | if (IS_ERR(handle)) { |
4743 | err = PTR_ERR(handle); | |
4744 | goto out_dio; | |
4745 | } | |
a4bb6b64 | 4746 | |
a4bb6b64 AH |
4747 | |
4748 | /* | |
ba06208a AH |
4749 | * Now we need to zero out the non-page-aligned data in the |
4750 | * pages at the start and tail of the hole, and unmap the buffer | |
4751 | * heads for the block aligned regions of the page that were | |
4752 | * completely zeroed. | |
a4bb6b64 | 4753 | */ |
ba06208a AH |
4754 | if (first_page > last_page) { |
4755 | /* | |
4756 | * If the file space being truncated is contained within a page | |
4757 | * just zero out and unmap the middle of that page | |
4758 | */ | |
4759 | err = ext4_discard_partial_page_buffers(handle, | |
4760 | mapping, offset, length, 0); | |
4761 | ||
4762 | if (err) | |
4763 | goto out; | |
4764 | } else { | |
4765 | /* | |
4766 | * zero out and unmap the partial page that contains | |
4767 | * the start of the hole | |
4768 | */ | |
4769 | page_len = first_page_offset - offset; | |
4770 | if (page_len > 0) { | |
4771 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4772 | offset, page_len, 0); | |
4773 | if (err) | |
4774 | goto out; | |
4775 | } | |
4776 | ||
4777 | /* | |
4778 | * zero out and unmap the partial page that contains | |
4779 | * the end of the hole | |
4780 | */ | |
4781 | page_len = offset + length - last_page_offset; | |
4782 | if (page_len > 0) { | |
4783 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4784 | last_page_offset, page_len, 0); | |
4785 | if (err) | |
4786 | goto out; | |
a4bb6b64 AH |
4787 | } |
4788 | } | |
4789 | ||
2be4751b AH |
4790 | /* |
4791 | * If i_size is contained in the last page, we need to | |
4792 | * unmap and zero the partial page after i_size | |
4793 | */ | |
4794 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4795 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4796 | ||
4797 | page_len = PAGE_CACHE_SIZE - | |
4798 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4799 | ||
4800 | if (page_len > 0) { | |
4801 | err = ext4_discard_partial_page_buffers(handle, | |
4802 | mapping, inode->i_size, page_len, 0); | |
4803 | ||
4804 | if (err) | |
4805 | goto out; | |
4806 | } | |
4807 | } | |
4808 | ||
5f95d21f LC |
4809 | first_block = (offset + sb->s_blocksize - 1) >> |
4810 | EXT4_BLOCK_SIZE_BITS(sb); | |
4811 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4812 | ||
a4bb6b64 | 4813 | /* If there are no blocks to remove, return now */ |
5f95d21f | 4814 | if (first_block >= stop_block) |
a4bb6b64 AH |
4815 | goto out; |
4816 | ||
4817 | down_write(&EXT4_I(inode)->i_data_sem); | |
4818 | ext4_ext_invalidate_cache(inode); | |
4819 | ext4_discard_preallocations(inode); | |
4820 | ||
51865fda ZL |
4821 | err = ext4_es_remove_extent(inode, first_block, |
4822 | stop_block - first_block); | |
5f95d21f | 4823 | err = ext4_ext_remove_space(inode, first_block, stop_block - 1); |
a4bb6b64 | 4824 | |
5f95d21f LC |
4825 | ext4_ext_invalidate_cache(inode); |
4826 | ext4_discard_preallocations(inode); | |
a4bb6b64 AH |
4827 | |
4828 | if (IS_SYNC(inode)) | |
4829 | ext4_handle_sync(handle); | |
4830 | ||
4831 | up_write(&EXT4_I(inode)->i_data_sem); | |
4832 | ||
4833 | out: | |
a4bb6b64 AH |
4834 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4835 | ext4_mark_inode_dirty(handle, inode); | |
4836 | ext4_journal_stop(handle); | |
02d262df DM |
4837 | out_dio: |
4838 | ext4_inode_resume_unlocked_dio(inode); | |
4839 | out_mutex: | |
4840 | mutex_unlock(&inode->i_mutex); | |
a4bb6b64 AH |
4841 | return err; |
4842 | } | |
91dd8c11 | 4843 | |
6873fa0d ES |
4844 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4845 | __u64 start, __u64 len) | |
4846 | { | |
4847 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4848 | int error = 0; |
4849 | ||
94191985 TM |
4850 | if (ext4_has_inline_data(inode)) { |
4851 | int has_inline = 1; | |
4852 | ||
4853 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline); | |
4854 | ||
4855 | if (has_inline) | |
4856 | return error; | |
4857 | } | |
4858 | ||
6873fa0d | 4859 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 4860 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4861 | return generic_block_fiemap(inode, fieinfo, start, len, |
4862 | ext4_get_block); | |
4863 | ||
4864 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4865 | return -EBADR; | |
4866 | ||
4867 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4868 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4869 | } else { | |
aca92ff6 LM |
4870 | ext4_lblk_t len_blks; |
4871 | __u64 last_blk; | |
4872 | ||
6873fa0d | 4873 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4874 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4875 | if (last_blk >= EXT_MAX_BLOCKS) |
4876 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4877 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4878 | |
4879 | /* | |
91dd8c11 LC |
4880 | * Walk the extent tree gathering extent information |
4881 | * and pushing extents back to the user. | |
6873fa0d | 4882 | */ |
91dd8c11 LC |
4883 | error = ext4_fill_fiemap_extents(inode, start_blk, |
4884 | len_blks, fieinfo); | |
6873fa0d ES |
4885 | } |
4886 | ||
4887 | return error; | |
4888 | } |