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> | |
a86c6181 | 40 | #include <asm/uaccess.h> |
6873fa0d | 41 | #include <linux/fiemap.h> |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
4a092d73 | 43 | #include "ext4_extents.h" |
f19d5870 | 44 | #include "xattr.h" |
a86c6181 | 45 | |
0562e0ba JZ |
46 | #include <trace/events/ext4.h> |
47 | ||
5f95d21f LC |
48 | /* |
49 | * used by extent splitting. | |
50 | */ | |
51 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
52 | due to ENOSPC */ | |
53 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
54 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
55 | ||
dee1f973 DM |
56 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
57 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
58 | ||
7ac5990d DW |
59 | static __le32 ext4_extent_block_csum(struct inode *inode, |
60 | struct ext4_extent_header *eh) | |
61 | { | |
62 | struct ext4_inode_info *ei = EXT4_I(inode); | |
63 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
64 | __u32 csum; | |
65 | ||
66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
67 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
68 | return cpu_to_le32(csum); | |
69 | } | |
70 | ||
71 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
72 | struct ext4_extent_header *eh) | |
73 | { | |
74 | struct ext4_extent_tail *et; | |
75 | ||
76 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
77 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
78 | return 1; | |
79 | ||
80 | et = find_ext4_extent_tail(eh); | |
81 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
82 | return 0; | |
83 | return 1; | |
84 | } | |
85 | ||
86 | static void ext4_extent_block_csum_set(struct inode *inode, | |
87 | struct ext4_extent_header *eh) | |
88 | { | |
89 | struct ext4_extent_tail *et; | |
90 | ||
91 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
92 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
93 | return; | |
94 | ||
95 | et = find_ext4_extent_tail(eh); | |
96 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
97 | } | |
98 | ||
d583fb87 AH |
99 | static int ext4_split_extent(handle_t *handle, |
100 | struct inode *inode, | |
101 | struct ext4_ext_path *path, | |
102 | struct ext4_map_blocks *map, | |
103 | int split_flag, | |
104 | int flags); | |
105 | ||
5f95d21f LC |
106 | static int ext4_split_extent_at(handle_t *handle, |
107 | struct inode *inode, | |
108 | struct ext4_ext_path *path, | |
109 | ext4_lblk_t split, | |
110 | int split_flag, | |
111 | int flags); | |
112 | ||
91dd8c11 | 113 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 114 | struct extent_status *newes); |
91dd8c11 | 115 | |
487caeef JK |
116 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
117 | struct inode *inode, | |
118 | int needed) | |
a86c6181 AT |
119 | { |
120 | int err; | |
121 | ||
0390131b FM |
122 | if (!ext4_handle_valid(handle)) |
123 | return 0; | |
a86c6181 | 124 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
125 | return 0; |
126 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 127 | if (err <= 0) |
9102e4fa | 128 | return err; |
487caeef | 129 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
130 | if (err == 0) |
131 | err = -EAGAIN; | |
487caeef JK |
132 | |
133 | return err; | |
a86c6181 AT |
134 | } |
135 | ||
136 | /* | |
137 | * could return: | |
138 | * - EROFS | |
139 | * - ENOMEM | |
140 | */ | |
141 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
142 | struct ext4_ext_path *path) | |
143 | { | |
144 | if (path->p_bh) { | |
145 | /* path points to block */ | |
146 | return ext4_journal_get_write_access(handle, path->p_bh); | |
147 | } | |
148 | /* path points to leaf/index in inode body */ | |
149 | /* we use in-core data, no need to protect them */ | |
150 | return 0; | |
151 | } | |
152 | ||
153 | /* | |
154 | * could return: | |
155 | * - EROFS | |
156 | * - ENOMEM | |
157 | * - EIO | |
158 | */ | |
2656497b DW |
159 | int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle, |
160 | struct inode *inode, struct ext4_ext_path *path) | |
a86c6181 AT |
161 | { |
162 | int err; | |
163 | if (path->p_bh) { | |
7ac5990d | 164 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 165 | /* path points to block */ |
9ea7a0df TT |
166 | err = __ext4_handle_dirty_metadata(where, line, handle, |
167 | inode, path->p_bh); | |
a86c6181 AT |
168 | } else { |
169 | /* path points to leaf/index in inode body */ | |
170 | err = ext4_mark_inode_dirty(handle, inode); | |
171 | } | |
172 | return err; | |
173 | } | |
174 | ||
f65e6fba | 175 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 176 | struct ext4_ext_path *path, |
725d26d3 | 177 | ext4_lblk_t block) |
a86c6181 | 178 | { |
a86c6181 | 179 | if (path) { |
81fdbb4a | 180 | int depth = path->p_depth; |
a86c6181 | 181 | struct ext4_extent *ex; |
a86c6181 | 182 | |
ad4fb9ca KM |
183 | /* |
184 | * Try to predict block placement assuming that we are | |
185 | * filling in a file which will eventually be | |
186 | * non-sparse --- i.e., in the case of libbfd writing | |
187 | * an ELF object sections out-of-order but in a way | |
188 | * the eventually results in a contiguous object or | |
189 | * executable file, or some database extending a table | |
190 | * space file. However, this is actually somewhat | |
191 | * non-ideal if we are writing a sparse file such as | |
192 | * qemu or KVM writing a raw image file that is going | |
193 | * to stay fairly sparse, since it will end up | |
194 | * fragmenting the file system's free space. Maybe we | |
195 | * should have some hueristics or some way to allow | |
196 | * userspace to pass a hint to file system, | |
b8d6568a | 197 | * especially if the latter case turns out to be |
ad4fb9ca KM |
198 | * common. |
199 | */ | |
7e028976 | 200 | ex = path[depth].p_ext; |
ad4fb9ca KM |
201 | if (ex) { |
202 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
203 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
204 | ||
205 | if (block > ext_block) | |
206 | return ext_pblk + (block - ext_block); | |
207 | else | |
208 | return ext_pblk - (ext_block - block); | |
209 | } | |
a86c6181 | 210 | |
d0d856e8 RD |
211 | /* it looks like index is empty; |
212 | * try to find starting block from index itself */ | |
a86c6181 AT |
213 | if (path[depth].p_bh) |
214 | return path[depth].p_bh->b_blocknr; | |
215 | } | |
216 | ||
217 | /* OK. use inode's group */ | |
f86186b4 | 218 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
219 | } |
220 | ||
654b4908 AK |
221 | /* |
222 | * Allocation for a meta data block | |
223 | */ | |
f65e6fba | 224 | static ext4_fsblk_t |
654b4908 | 225 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 226 | struct ext4_ext_path *path, |
55f020db | 227 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 228 | { |
f65e6fba | 229 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
230 | |
231 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
232 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
233 | NULL, err); | |
a86c6181 AT |
234 | return newblock; |
235 | } | |
236 | ||
55ad63bf | 237 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
238 | { |
239 | int size; | |
240 | ||
241 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
242 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 243 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
244 | if (!check && size > 6) |
245 | size = 6; | |
a86c6181 AT |
246 | #endif |
247 | return size; | |
248 | } | |
249 | ||
55ad63bf | 250 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
251 | { |
252 | int size; | |
253 | ||
254 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
255 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 256 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
257 | if (!check && size > 5) |
258 | size = 5; | |
a86c6181 AT |
259 | #endif |
260 | return size; | |
261 | } | |
262 | ||
55ad63bf | 263 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
264 | { |
265 | int size; | |
266 | ||
267 | size = sizeof(EXT4_I(inode)->i_data); | |
268 | size -= sizeof(struct ext4_extent_header); | |
269 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 270 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
271 | if (!check && size > 3) |
272 | size = 3; | |
a86c6181 AT |
273 | #endif |
274 | return size; | |
275 | } | |
276 | ||
55ad63bf | 277 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
278 | { |
279 | int size; | |
280 | ||
281 | size = sizeof(EXT4_I(inode)->i_data); | |
282 | size -= sizeof(struct ext4_extent_header); | |
283 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 284 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
285 | if (!check && size > 4) |
286 | size = 4; | |
a86c6181 AT |
287 | #endif |
288 | return size; | |
289 | } | |
290 | ||
d2a17637 MC |
291 | /* |
292 | * Calculate the number of metadata blocks needed | |
293 | * to allocate @blocks | |
294 | * Worse case is one block per extent | |
295 | */ | |
01f49d0b | 296 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 297 | { |
9d0be502 | 298 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 299 | int idxs; |
d2a17637 | 300 | |
9d0be502 TT |
301 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
302 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
303 | |
304 | /* | |
9d0be502 TT |
305 | * If the new delayed allocation block is contiguous with the |
306 | * previous da block, it can share index blocks with the | |
307 | * previous block, so we only need to allocate a new index | |
308 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
309 | * an additional index block, and at ldxs**3 blocks, yet | |
310 | * another index blocks. | |
d2a17637 | 311 | */ |
9d0be502 TT |
312 | if (ei->i_da_metadata_calc_len && |
313 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
314 | int num = 0; |
315 | ||
9d0be502 TT |
316 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
317 | num++; | |
318 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
319 | num++; | |
320 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
321 | num++; | |
322 | ei->i_da_metadata_calc_len = 0; | |
323 | } else | |
324 | ei->i_da_metadata_calc_len++; | |
325 | ei->i_da_metadata_calc_last_lblock++; | |
326 | return num; | |
327 | } | |
d2a17637 | 328 | |
9d0be502 TT |
329 | /* |
330 | * In the worst case we need a new set of index blocks at | |
331 | * every level of the inode's extent tree. | |
332 | */ | |
333 | ei->i_da_metadata_calc_len = 1; | |
334 | ei->i_da_metadata_calc_last_lblock = lblock; | |
335 | return ext_depth(inode) + 1; | |
d2a17637 MC |
336 | } |
337 | ||
c29c0ae7 AT |
338 | static int |
339 | ext4_ext_max_entries(struct inode *inode, int depth) | |
340 | { | |
341 | int max; | |
342 | ||
343 | if (depth == ext_depth(inode)) { | |
344 | if (depth == 0) | |
55ad63bf | 345 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 346 | else |
55ad63bf | 347 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
348 | } else { |
349 | if (depth == 0) | |
55ad63bf | 350 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 351 | else |
55ad63bf | 352 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
353 | } |
354 | ||
355 | return max; | |
356 | } | |
357 | ||
56b19868 AK |
358 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
359 | { | |
bf89d16f | 360 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 361 | int len = ext4_ext_get_actual_len(ext); |
5946d089 EG |
362 | ext4_lblk_t lblock = le32_to_cpu(ext->ee_block); |
363 | ext4_lblk_t last = lblock + len - 1; | |
e84a26ce | 364 | |
5946d089 | 365 | if (lblock > last) |
31d4f3a2 | 366 | return 0; |
6fd058f7 | 367 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
368 | } |
369 | ||
370 | static int ext4_valid_extent_idx(struct inode *inode, | |
371 | struct ext4_extent_idx *ext_idx) | |
372 | { | |
bf89d16f | 373 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 374 | |
6fd058f7 | 375 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
376 | } |
377 | ||
378 | static int ext4_valid_extent_entries(struct inode *inode, | |
379 | struct ext4_extent_header *eh, | |
380 | int depth) | |
381 | { | |
56b19868 AK |
382 | unsigned short entries; |
383 | if (eh->eh_entries == 0) | |
384 | return 1; | |
385 | ||
386 | entries = le16_to_cpu(eh->eh_entries); | |
387 | ||
388 | if (depth == 0) { | |
389 | /* leaf entries */ | |
81fdbb4a | 390 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
5946d089 EG |
391 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
392 | ext4_fsblk_t pblock = 0; | |
393 | ext4_lblk_t lblock = 0; | |
394 | ext4_lblk_t prev = 0; | |
395 | int len = 0; | |
56b19868 AK |
396 | while (entries) { |
397 | if (!ext4_valid_extent(inode, ext)) | |
398 | return 0; | |
5946d089 EG |
399 | |
400 | /* Check for overlapping extents */ | |
401 | lblock = le32_to_cpu(ext->ee_block); | |
402 | len = ext4_ext_get_actual_len(ext); | |
403 | if ((lblock <= prev) && prev) { | |
404 | pblock = ext4_ext_pblock(ext); | |
405 | es->s_last_error_block = cpu_to_le64(pblock); | |
406 | return 0; | |
407 | } | |
56b19868 AK |
408 | ext++; |
409 | entries--; | |
5946d089 | 410 | prev = lblock + len - 1; |
56b19868 AK |
411 | } |
412 | } else { | |
81fdbb4a | 413 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
414 | while (entries) { |
415 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
416 | return 0; | |
417 | ext_idx++; | |
418 | entries--; | |
419 | } | |
420 | } | |
421 | return 1; | |
422 | } | |
423 | ||
c398eda0 TT |
424 | static int __ext4_ext_check(const char *function, unsigned int line, |
425 | struct inode *inode, struct ext4_extent_header *eh, | |
c349179b | 426 | int depth, ext4_fsblk_t pblk) |
c29c0ae7 AT |
427 | { |
428 | const char *error_msg; | |
429 | int max = 0; | |
430 | ||
431 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
432 | error_msg = "invalid magic"; | |
433 | goto corrupted; | |
434 | } | |
435 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
436 | error_msg = "unexpected eh_depth"; | |
437 | goto corrupted; | |
438 | } | |
439 | if (unlikely(eh->eh_max == 0)) { | |
440 | error_msg = "invalid eh_max"; | |
441 | goto corrupted; | |
442 | } | |
443 | max = ext4_ext_max_entries(inode, depth); | |
444 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
445 | error_msg = "too large eh_max"; | |
446 | goto corrupted; | |
447 | } | |
448 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
449 | error_msg = "invalid eh_entries"; | |
450 | goto corrupted; | |
451 | } | |
56b19868 AK |
452 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
453 | error_msg = "invalid extent entries"; | |
454 | goto corrupted; | |
455 | } | |
7ac5990d DW |
456 | /* Verify checksum on non-root extent tree nodes */ |
457 | if (ext_depth(inode) != depth && | |
458 | !ext4_extent_block_csum_verify(inode, eh)) { | |
459 | error_msg = "extent tree corrupted"; | |
460 | goto corrupted; | |
461 | } | |
c29c0ae7 AT |
462 | return 0; |
463 | ||
464 | corrupted: | |
c398eda0 | 465 | ext4_error_inode(inode, function, line, 0, |
c349179b TT |
466 | "pblk %llu bad header/extent: %s - magic %x, " |
467 | "entries %u, max %u(%u), depth %u(%u)", | |
468 | (unsigned long long) pblk, error_msg, | |
469 | le16_to_cpu(eh->eh_magic), | |
470 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
471 | max, le16_to_cpu(eh->eh_depth), depth); | |
c29c0ae7 AT |
472 | return -EIO; |
473 | } | |
474 | ||
c349179b TT |
475 | #define ext4_ext_check(inode, eh, depth, pblk) \ |
476 | __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk)) | |
c29c0ae7 | 477 | |
7a262f7c AK |
478 | int ext4_ext_check_inode(struct inode *inode) |
479 | { | |
c349179b | 480 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
7a262f7c AK |
481 | } |
482 | ||
7d7ea89e TT |
483 | static struct buffer_head * |
484 | __read_extent_tree_block(const char *function, unsigned int line, | |
107a7bd3 TT |
485 | struct inode *inode, ext4_fsblk_t pblk, int depth, |
486 | int flags) | |
f8489128 | 487 | { |
7d7ea89e TT |
488 | struct buffer_head *bh; |
489 | int err; | |
490 | ||
491 | bh = sb_getblk(inode->i_sb, pblk); | |
492 | if (unlikely(!bh)) | |
493 | return ERR_PTR(-ENOMEM); | |
f8489128 | 494 | |
7d7ea89e TT |
495 | if (!bh_uptodate_or_lock(bh)) { |
496 | trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); | |
497 | err = bh_submit_read(bh); | |
498 | if (err < 0) | |
499 | goto errout; | |
500 | } | |
7869a4a6 | 501 | if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE)) |
7d7ea89e TT |
502 | return bh; |
503 | err = __ext4_ext_check(function, line, inode, | |
c349179b | 504 | ext_block_hdr(bh), depth, pblk); |
7d7ea89e TT |
505 | if (err) |
506 | goto errout; | |
f8489128 | 507 | set_buffer_verified(bh); |
107a7bd3 TT |
508 | /* |
509 | * If this is a leaf block, cache all of its entries | |
510 | */ | |
511 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { | |
512 | struct ext4_extent_header *eh = ext_block_hdr(bh); | |
513 | struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); | |
514 | ext4_lblk_t prev = 0; | |
515 | int i; | |
516 | ||
517 | for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { | |
518 | unsigned int status = EXTENT_STATUS_WRITTEN; | |
519 | ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); | |
520 | int len = ext4_ext_get_actual_len(ex); | |
521 | ||
522 | if (prev && (prev != lblk)) | |
523 | ext4_es_cache_extent(inode, prev, | |
524 | lblk - prev, ~0, | |
525 | EXTENT_STATUS_HOLE); | |
526 | ||
527 | if (ext4_ext_is_uninitialized(ex)) | |
528 | status = EXTENT_STATUS_UNWRITTEN; | |
529 | ext4_es_cache_extent(inode, lblk, len, | |
530 | ext4_ext_pblock(ex), status); | |
531 | prev = lblk + len; | |
532 | } | |
533 | } | |
7d7ea89e TT |
534 | return bh; |
535 | errout: | |
536 | put_bh(bh); | |
537 | return ERR_PTR(err); | |
538 | ||
f8489128 DW |
539 | } |
540 | ||
107a7bd3 TT |
541 | #define read_extent_tree_block(inode, pblk, depth, flags) \ |
542 | __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \ | |
543 | (depth), (flags)) | |
f8489128 | 544 | |
7869a4a6 TT |
545 | /* |
546 | * This function is called to cache a file's extent information in the | |
547 | * extent status tree | |
548 | */ | |
549 | int ext4_ext_precache(struct inode *inode) | |
550 | { | |
551 | struct ext4_inode_info *ei = EXT4_I(inode); | |
552 | struct ext4_ext_path *path = NULL; | |
553 | struct buffer_head *bh; | |
554 | int i = 0, depth, ret = 0; | |
555 | ||
556 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
557 | return 0; /* not an extent-mapped inode */ | |
558 | ||
559 | down_read(&ei->i_data_sem); | |
560 | depth = ext_depth(inode); | |
561 | ||
562 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
563 | GFP_NOFS); | |
564 | if (path == NULL) { | |
565 | up_read(&ei->i_data_sem); | |
566 | return -ENOMEM; | |
567 | } | |
568 | ||
569 | /* Don't cache anything if there are no external extent blocks */ | |
570 | if (depth == 0) | |
571 | goto out; | |
572 | path[0].p_hdr = ext_inode_hdr(inode); | |
573 | ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0); | |
574 | if (ret) | |
575 | goto out; | |
576 | path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr); | |
577 | while (i >= 0) { | |
578 | /* | |
579 | * If this is a leaf block or we've reached the end of | |
580 | * the index block, go up | |
581 | */ | |
582 | if ((i == depth) || | |
583 | path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) { | |
584 | brelse(path[i].p_bh); | |
585 | path[i].p_bh = NULL; | |
586 | i--; | |
587 | continue; | |
588 | } | |
589 | bh = read_extent_tree_block(inode, | |
590 | ext4_idx_pblock(path[i].p_idx++), | |
591 | depth - i - 1, | |
592 | EXT4_EX_FORCE_CACHE); | |
593 | if (IS_ERR(bh)) { | |
594 | ret = PTR_ERR(bh); | |
595 | break; | |
596 | } | |
597 | i++; | |
598 | path[i].p_bh = bh; | |
599 | path[i].p_hdr = ext_block_hdr(bh); | |
600 | path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr); | |
601 | } | |
602 | ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED); | |
603 | out: | |
604 | up_read(&ei->i_data_sem); | |
605 | ext4_ext_drop_refs(path); | |
606 | kfree(path); | |
607 | return ret; | |
608 | } | |
609 | ||
a86c6181 AT |
610 | #ifdef EXT_DEBUG |
611 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
612 | { | |
613 | int k, l = path->p_depth; | |
614 | ||
615 | ext_debug("path:"); | |
616 | for (k = 0; k <= l; k++, path++) { | |
617 | if (path->p_idx) { | |
2ae02107 | 618 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 619 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 620 | } else if (path->p_ext) { |
553f9008 | 621 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 622 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 623 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 624 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 625 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
626 | } else |
627 | ext_debug(" []"); | |
628 | } | |
629 | ext_debug("\n"); | |
630 | } | |
631 | ||
632 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
633 | { | |
634 | int depth = ext_depth(inode); | |
635 | struct ext4_extent_header *eh; | |
636 | struct ext4_extent *ex; | |
637 | int i; | |
638 | ||
639 | if (!path) | |
640 | return; | |
641 | ||
642 | eh = path[depth].p_hdr; | |
643 | ex = EXT_FIRST_EXTENT(eh); | |
644 | ||
553f9008 M |
645 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
646 | ||
a86c6181 | 647 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
648 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
649 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 650 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
651 | } |
652 | ext_debug("\n"); | |
653 | } | |
1b16da77 YY |
654 | |
655 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
656 | ext4_fsblk_t newblock, int level) | |
657 | { | |
658 | int depth = ext_depth(inode); | |
659 | struct ext4_extent *ex; | |
660 | ||
661 | if (depth != level) { | |
662 | struct ext4_extent_idx *idx; | |
663 | idx = path[level].p_idx; | |
664 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
665 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
666 | le32_to_cpu(idx->ei_block), | |
667 | ext4_idx_pblock(idx), | |
668 | newblock); | |
669 | idx++; | |
670 | } | |
671 | ||
672 | return; | |
673 | } | |
674 | ||
675 | ex = path[depth].p_ext; | |
676 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
677 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
678 | le32_to_cpu(ex->ee_block), | |
679 | ext4_ext_pblock(ex), | |
680 | ext4_ext_is_uninitialized(ex), | |
681 | ext4_ext_get_actual_len(ex), | |
682 | newblock); | |
683 | ex++; | |
684 | } | |
685 | } | |
686 | ||
a86c6181 | 687 | #else |
af5bc92d TT |
688 | #define ext4_ext_show_path(inode, path) |
689 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 690 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
691 | #endif |
692 | ||
b35905c1 | 693 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
694 | { |
695 | int depth = path->p_depth; | |
696 | int i; | |
697 | ||
698 | for (i = 0; i <= depth; i++, path++) | |
699 | if (path->p_bh) { | |
700 | brelse(path->p_bh); | |
701 | path->p_bh = NULL; | |
702 | } | |
703 | } | |
704 | ||
705 | /* | |
d0d856e8 RD |
706 | * ext4_ext_binsearch_idx: |
707 | * binary search for the closest index of the given block | |
c29c0ae7 | 708 | * the header must be checked before calling this |
a86c6181 AT |
709 | */ |
710 | static void | |
725d26d3 AK |
711 | ext4_ext_binsearch_idx(struct inode *inode, |
712 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
713 | { |
714 | struct ext4_extent_header *eh = path->p_hdr; | |
715 | struct ext4_extent_idx *r, *l, *m; | |
716 | ||
a86c6181 | 717 | |
bba90743 | 718 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
719 | |
720 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 721 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
722 | while (l <= r) { |
723 | m = l + (r - l) / 2; | |
724 | if (block < le32_to_cpu(m->ei_block)) | |
725 | r = m - 1; | |
726 | else | |
727 | l = m + 1; | |
26d535ed DM |
728 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
729 | m, le32_to_cpu(m->ei_block), | |
730 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
731 | } |
732 | ||
733 | path->p_idx = l - 1; | |
4a3c3a51 | 734 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 735 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
736 | |
737 | #ifdef CHECK_BINSEARCH | |
738 | { | |
739 | struct ext4_extent_idx *chix, *ix; | |
740 | int k; | |
741 | ||
742 | chix = ix = EXT_FIRST_INDEX(eh); | |
743 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
744 | if (k != 0 && | |
745 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
746 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
747 | "first=0x%p\n", k, | |
748 | ix, EXT_FIRST_INDEX(eh)); | |
749 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
750 | le32_to_cpu(ix->ei_block), |
751 | le32_to_cpu(ix[-1].ei_block)); | |
752 | } | |
753 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 754 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
755 | if (block < le32_to_cpu(ix->ei_block)) |
756 | break; | |
757 | chix = ix; | |
758 | } | |
759 | BUG_ON(chix != path->p_idx); | |
760 | } | |
761 | #endif | |
762 | ||
763 | } | |
764 | ||
765 | /* | |
d0d856e8 RD |
766 | * ext4_ext_binsearch: |
767 | * binary search for closest extent of the given block | |
c29c0ae7 | 768 | * the header must be checked before calling this |
a86c6181 AT |
769 | */ |
770 | static void | |
725d26d3 AK |
771 | ext4_ext_binsearch(struct inode *inode, |
772 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
773 | { |
774 | struct ext4_extent_header *eh = path->p_hdr; | |
775 | struct ext4_extent *r, *l, *m; | |
776 | ||
a86c6181 AT |
777 | if (eh->eh_entries == 0) { |
778 | /* | |
d0d856e8 RD |
779 | * this leaf is empty: |
780 | * we get such a leaf in split/add case | |
a86c6181 AT |
781 | */ |
782 | return; | |
783 | } | |
784 | ||
bba90743 | 785 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
786 | |
787 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 788 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
789 | |
790 | while (l <= r) { | |
791 | m = l + (r - l) / 2; | |
792 | if (block < le32_to_cpu(m->ee_block)) | |
793 | r = m - 1; | |
794 | else | |
795 | l = m + 1; | |
26d535ed DM |
796 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
797 | m, le32_to_cpu(m->ee_block), | |
798 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
799 | } |
800 | ||
801 | path->p_ext = l - 1; | |
553f9008 | 802 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 803 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 804 | ext4_ext_pblock(path->p_ext), |
553f9008 | 805 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 806 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
807 | |
808 | #ifdef CHECK_BINSEARCH | |
809 | { | |
810 | struct ext4_extent *chex, *ex; | |
811 | int k; | |
812 | ||
813 | chex = ex = EXT_FIRST_EXTENT(eh); | |
814 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
815 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 816 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
817 | if (block < le32_to_cpu(ex->ee_block)) |
818 | break; | |
819 | chex = ex; | |
820 | } | |
821 | BUG_ON(chex != path->p_ext); | |
822 | } | |
823 | #endif | |
824 | ||
825 | } | |
826 | ||
827 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
828 | { | |
829 | struct ext4_extent_header *eh; | |
830 | ||
831 | eh = ext_inode_hdr(inode); | |
832 | eh->eh_depth = 0; | |
833 | eh->eh_entries = 0; | |
834 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 835 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 836 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
837 | return 0; |
838 | } | |
839 | ||
840 | struct ext4_ext_path * | |
725d26d3 | 841 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
107a7bd3 | 842 | struct ext4_ext_path *path, int flags) |
a86c6181 AT |
843 | { |
844 | struct ext4_extent_header *eh; | |
845 | struct buffer_head *bh; | |
846 | short int depth, i, ppos = 0, alloc = 0; | |
860d21e2 | 847 | int ret; |
a86c6181 AT |
848 | |
849 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 850 | depth = ext_depth(inode); |
a86c6181 AT |
851 | |
852 | /* account possible depth increase */ | |
853 | if (!path) { | |
5d4958f9 | 854 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
855 | GFP_NOFS); |
856 | if (!path) | |
857 | return ERR_PTR(-ENOMEM); | |
858 | alloc = 1; | |
859 | } | |
a86c6181 | 860 | path[0].p_hdr = eh; |
1973adcb | 861 | path[0].p_bh = NULL; |
a86c6181 | 862 | |
c29c0ae7 | 863 | i = depth; |
a86c6181 AT |
864 | /* walk through the tree */ |
865 | while (i) { | |
866 | ext_debug("depth %d: num %d, max %d\n", | |
867 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 868 | |
a86c6181 | 869 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 870 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
871 | path[ppos].p_depth = i; |
872 | path[ppos].p_ext = NULL; | |
873 | ||
107a7bd3 TT |
874 | bh = read_extent_tree_block(inode, path[ppos].p_block, --i, |
875 | flags); | |
7d7ea89e TT |
876 | if (IS_ERR(bh)) { |
877 | ret = PTR_ERR(bh); | |
a86c6181 | 878 | goto err; |
860d21e2 | 879 | } |
7d7ea89e | 880 | |
a86c6181 AT |
881 | eh = ext_block_hdr(bh); |
882 | ppos++; | |
273df556 FM |
883 | if (unlikely(ppos > depth)) { |
884 | put_bh(bh); | |
885 | EXT4_ERROR_INODE(inode, | |
886 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 887 | ret = -EIO; |
273df556 FM |
888 | goto err; |
889 | } | |
a86c6181 AT |
890 | path[ppos].p_bh = bh; |
891 | path[ppos].p_hdr = eh; | |
a86c6181 AT |
892 | } |
893 | ||
894 | path[ppos].p_depth = i; | |
a86c6181 AT |
895 | path[ppos].p_ext = NULL; |
896 | path[ppos].p_idx = NULL; | |
897 | ||
a86c6181 AT |
898 | /* find extent */ |
899 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
900 | /* if not an empty leaf */ |
901 | if (path[ppos].p_ext) | |
bf89d16f | 902 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
903 | |
904 | ext4_ext_show_path(inode, path); | |
905 | ||
906 | return path; | |
907 | ||
908 | err: | |
909 | ext4_ext_drop_refs(path); | |
910 | if (alloc) | |
911 | kfree(path); | |
860d21e2 | 912 | return ERR_PTR(ret); |
a86c6181 AT |
913 | } |
914 | ||
915 | /* | |
d0d856e8 RD |
916 | * ext4_ext_insert_index: |
917 | * insert new index [@logical;@ptr] into the block at @curp; | |
918 | * check where to insert: before @curp or after @curp | |
a86c6181 | 919 | */ |
1f109d5a TT |
920 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
921 | struct ext4_ext_path *curp, | |
922 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
923 | { |
924 | struct ext4_extent_idx *ix; | |
925 | int len, err; | |
926 | ||
7e028976 AM |
927 | err = ext4_ext_get_access(handle, inode, curp); |
928 | if (err) | |
a86c6181 AT |
929 | return err; |
930 | ||
273df556 FM |
931 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
932 | EXT4_ERROR_INODE(inode, | |
933 | "logical %d == ei_block %d!", | |
934 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
935 | return -EIO; | |
936 | } | |
d4620315 RD |
937 | |
938 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
939 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
940 | EXT4_ERROR_INODE(inode, | |
941 | "eh_entries %d >= eh_max %d!", | |
942 | le16_to_cpu(curp->p_hdr->eh_entries), | |
943 | le16_to_cpu(curp->p_hdr->eh_max)); | |
944 | return -EIO; | |
945 | } | |
946 | ||
a86c6181 AT |
947 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
948 | /* insert after */ | |
80e675f9 | 949 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
950 | ix = curp->p_idx + 1; |
951 | } else { | |
952 | /* insert before */ | |
80e675f9 | 953 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
954 | ix = curp->p_idx; |
955 | } | |
956 | ||
80e675f9 EG |
957 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
958 | BUG_ON(len < 0); | |
959 | if (len > 0) { | |
960 | ext_debug("insert new index %d: " | |
961 | "move %d indices from 0x%p to 0x%p\n", | |
962 | logical, len, ix, ix + 1); | |
963 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
964 | } | |
965 | ||
f472e026 TM |
966 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
967 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
968 | return -EIO; | |
969 | } | |
970 | ||
a86c6181 | 971 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 972 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 973 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 974 | |
273df556 FM |
975 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
976 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
977 | return -EIO; | |
978 | } | |
a86c6181 AT |
979 | |
980 | err = ext4_ext_dirty(handle, inode, curp); | |
981 | ext4_std_error(inode->i_sb, err); | |
982 | ||
983 | return err; | |
984 | } | |
985 | ||
986 | /* | |
d0d856e8 RD |
987 | * ext4_ext_split: |
988 | * inserts new subtree into the path, using free index entry | |
989 | * at depth @at: | |
990 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
991 | * - makes decision where to split | |
992 | * - moves remaining extents and index entries (right to the split point) | |
993 | * into the newly allocated blocks | |
994 | * - initializes subtree | |
a86c6181 AT |
995 | */ |
996 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
997 | unsigned int flags, |
998 | struct ext4_ext_path *path, | |
999 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
1000 | { |
1001 | struct buffer_head *bh = NULL; | |
1002 | int depth = ext_depth(inode); | |
1003 | struct ext4_extent_header *neh; | |
1004 | struct ext4_extent_idx *fidx; | |
a86c6181 | 1005 | int i = at, k, m, a; |
f65e6fba | 1006 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 1007 | __le32 border; |
f65e6fba | 1008 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
1009 | int err = 0; |
1010 | ||
1011 | /* make decision: where to split? */ | |
d0d856e8 | 1012 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 1013 | |
d0d856e8 | 1014 | /* if current leaf will be split, then we should use |
a86c6181 | 1015 | * border from split point */ |
273df556 FM |
1016 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
1017 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
1018 | return -EIO; | |
1019 | } | |
a86c6181 AT |
1020 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
1021 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 1022 | ext_debug("leaf will be split." |
a86c6181 | 1023 | " next leaf starts at %d\n", |
8c55e204 | 1024 | le32_to_cpu(border)); |
a86c6181 AT |
1025 | } else { |
1026 | border = newext->ee_block; | |
1027 | ext_debug("leaf will be added." | |
1028 | " next leaf starts at %d\n", | |
8c55e204 | 1029 | le32_to_cpu(border)); |
a86c6181 AT |
1030 | } |
1031 | ||
1032 | /* | |
d0d856e8 RD |
1033 | * If error occurs, then we break processing |
1034 | * and mark filesystem read-only. index won't | |
a86c6181 | 1035 | * be inserted and tree will be in consistent |
d0d856e8 | 1036 | * state. Next mount will repair buffers too. |
a86c6181 AT |
1037 | */ |
1038 | ||
1039 | /* | |
d0d856e8 RD |
1040 | * Get array to track all allocated blocks. |
1041 | * We need this to handle errors and free blocks | |
1042 | * upon them. | |
a86c6181 | 1043 | */ |
5d4958f9 | 1044 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
1045 | if (!ablocks) |
1046 | return -ENOMEM; | |
a86c6181 AT |
1047 | |
1048 | /* allocate all needed blocks */ | |
1049 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
1050 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 1051 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 1052 | newext, &err, flags); |
a86c6181 AT |
1053 | if (newblock == 0) |
1054 | goto cleanup; | |
1055 | ablocks[a] = newblock; | |
1056 | } | |
1057 | ||
1058 | /* initialize new leaf */ | |
1059 | newblock = ablocks[--a]; | |
273df556 FM |
1060 | if (unlikely(newblock == 0)) { |
1061 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
1062 | err = -EIO; | |
1063 | goto cleanup; | |
1064 | } | |
a86c6181 | 1065 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1066 | if (unlikely(!bh)) { |
860d21e2 | 1067 | err = -ENOMEM; |
a86c6181 AT |
1068 | goto cleanup; |
1069 | } | |
1070 | lock_buffer(bh); | |
1071 | ||
7e028976 AM |
1072 | err = ext4_journal_get_create_access(handle, bh); |
1073 | if (err) | |
a86c6181 AT |
1074 | goto cleanup; |
1075 | ||
1076 | neh = ext_block_hdr(bh); | |
1077 | neh->eh_entries = 0; | |
55ad63bf | 1078 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1079 | neh->eh_magic = EXT4_EXT_MAGIC; |
1080 | neh->eh_depth = 0; | |
a86c6181 | 1081 | |
d0d856e8 | 1082 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
1083 | if (unlikely(path[depth].p_hdr->eh_entries != |
1084 | path[depth].p_hdr->eh_max)) { | |
1085 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
1086 | path[depth].p_hdr->eh_entries, | |
1087 | path[depth].p_hdr->eh_max); | |
1088 | err = -EIO; | |
1089 | goto cleanup; | |
1090 | } | |
a86c6181 | 1091 | /* start copy from next extent */ |
1b16da77 YY |
1092 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
1093 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 1094 | if (m) { |
1b16da77 YY |
1095 | struct ext4_extent *ex; |
1096 | ex = EXT_FIRST_EXTENT(neh); | |
1097 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 1098 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1099 | } |
1100 | ||
7ac5990d | 1101 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1102 | set_buffer_uptodate(bh); |
1103 | unlock_buffer(bh); | |
1104 | ||
0390131b | 1105 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1106 | if (err) |
a86c6181 AT |
1107 | goto cleanup; |
1108 | brelse(bh); | |
1109 | bh = NULL; | |
1110 | ||
1111 | /* correct old leaf */ | |
1112 | if (m) { | |
7e028976 AM |
1113 | err = ext4_ext_get_access(handle, inode, path + depth); |
1114 | if (err) | |
a86c6181 | 1115 | goto cleanup; |
e8546d06 | 1116 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1117 | err = ext4_ext_dirty(handle, inode, path + depth); |
1118 | if (err) | |
a86c6181 AT |
1119 | goto cleanup; |
1120 | ||
1121 | } | |
1122 | ||
1123 | /* create intermediate indexes */ | |
1124 | k = depth - at - 1; | |
273df556 FM |
1125 | if (unlikely(k < 0)) { |
1126 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1127 | err = -EIO; | |
1128 | goto cleanup; | |
1129 | } | |
a86c6181 AT |
1130 | if (k) |
1131 | ext_debug("create %d intermediate indices\n", k); | |
1132 | /* insert new index into current index block */ | |
1133 | /* current depth stored in i var */ | |
1134 | i = depth - 1; | |
1135 | while (k--) { | |
1136 | oldblock = newblock; | |
1137 | newblock = ablocks[--a]; | |
bba90743 | 1138 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1139 | if (unlikely(!bh)) { |
860d21e2 | 1140 | err = -ENOMEM; |
a86c6181 AT |
1141 | goto cleanup; |
1142 | } | |
1143 | lock_buffer(bh); | |
1144 | ||
7e028976 AM |
1145 | err = ext4_journal_get_create_access(handle, bh); |
1146 | if (err) | |
a86c6181 AT |
1147 | goto cleanup; |
1148 | ||
1149 | neh = ext_block_hdr(bh); | |
1150 | neh->eh_entries = cpu_to_le16(1); | |
1151 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1152 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1153 | neh->eh_depth = cpu_to_le16(depth - i); |
1154 | fidx = EXT_FIRST_INDEX(neh); | |
1155 | fidx->ei_block = border; | |
f65e6fba | 1156 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1157 | |
bba90743 ES |
1158 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1159 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1160 | |
1b16da77 | 1161 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1162 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1163 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1164 | EXT4_ERROR_INODE(inode, | |
1165 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1166 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1167 | err = -EIO; | |
1168 | goto cleanup; | |
1169 | } | |
1b16da77 YY |
1170 | /* start copy indexes */ |
1171 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1172 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1173 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1174 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1175 | if (m) { |
1b16da77 | 1176 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1177 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1178 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1179 | } |
7ac5990d | 1180 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1181 | set_buffer_uptodate(bh); |
1182 | unlock_buffer(bh); | |
1183 | ||
0390131b | 1184 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1185 | if (err) |
a86c6181 AT |
1186 | goto cleanup; |
1187 | brelse(bh); | |
1188 | bh = NULL; | |
1189 | ||
1190 | /* correct old index */ | |
1191 | if (m) { | |
1192 | err = ext4_ext_get_access(handle, inode, path + i); | |
1193 | if (err) | |
1194 | goto cleanup; | |
e8546d06 | 1195 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1196 | err = ext4_ext_dirty(handle, inode, path + i); |
1197 | if (err) | |
1198 | goto cleanup; | |
1199 | } | |
1200 | ||
1201 | i--; | |
1202 | } | |
1203 | ||
1204 | /* insert new index */ | |
a86c6181 AT |
1205 | err = ext4_ext_insert_index(handle, inode, path + at, |
1206 | le32_to_cpu(border), newblock); | |
1207 | ||
1208 | cleanup: | |
1209 | if (bh) { | |
1210 | if (buffer_locked(bh)) | |
1211 | unlock_buffer(bh); | |
1212 | brelse(bh); | |
1213 | } | |
1214 | ||
1215 | if (err) { | |
1216 | /* free all allocated blocks in error case */ | |
1217 | for (i = 0; i < depth; i++) { | |
1218 | if (!ablocks[i]) | |
1219 | continue; | |
7dc57615 | 1220 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1221 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1222 | } |
1223 | } | |
1224 | kfree(ablocks); | |
1225 | ||
1226 | return err; | |
1227 | } | |
1228 | ||
1229 | /* | |
d0d856e8 RD |
1230 | * ext4_ext_grow_indepth: |
1231 | * implements tree growing procedure: | |
1232 | * - allocates new block | |
1233 | * - moves top-level data (index block or leaf) into the new block | |
1234 | * - initializes new top-level, creating index that points to the | |
1235 | * just created block | |
a86c6181 AT |
1236 | */ |
1237 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1238 | unsigned int flags, |
55f020db | 1239 | struct ext4_extent *newext) |
a86c6181 | 1240 | { |
a86c6181 | 1241 | struct ext4_extent_header *neh; |
a86c6181 | 1242 | struct buffer_head *bh; |
f65e6fba | 1243 | ext4_fsblk_t newblock; |
a86c6181 AT |
1244 | int err = 0; |
1245 | ||
1939dd84 | 1246 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1247 | newext, &err, flags); |
a86c6181 AT |
1248 | if (newblock == 0) |
1249 | return err; | |
1250 | ||
1251 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1252 | if (unlikely(!bh)) |
860d21e2 | 1253 | return -ENOMEM; |
a86c6181 AT |
1254 | lock_buffer(bh); |
1255 | ||
7e028976 AM |
1256 | err = ext4_journal_get_create_access(handle, bh); |
1257 | if (err) { | |
a86c6181 AT |
1258 | unlock_buffer(bh); |
1259 | goto out; | |
1260 | } | |
1261 | ||
1262 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1263 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1264 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1265 | |
1266 | /* set size of new block */ | |
1267 | neh = ext_block_hdr(bh); | |
1268 | /* old root could have indexes or leaves | |
1269 | * so calculate e_max right way */ | |
1270 | if (ext_depth(inode)) | |
55ad63bf | 1271 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1272 | else |
55ad63bf | 1273 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1274 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1275 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1276 | set_buffer_uptodate(bh); |
1277 | unlock_buffer(bh); | |
1278 | ||
0390131b | 1279 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1280 | if (err) |
a86c6181 AT |
1281 | goto out; |
1282 | ||
1939dd84 | 1283 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1284 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1285 | neh->eh_entries = cpu_to_le16(1); |
1286 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1287 | if (neh->eh_depth == 0) { | |
1288 | /* Root extent block becomes index block */ | |
1289 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1290 | EXT_FIRST_INDEX(neh)->ei_block = | |
1291 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1292 | } | |
2ae02107 | 1293 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1294 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1295 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1296 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1297 | |
ba39ebb6 | 1298 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1299 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1300 | out: |
1301 | brelse(bh); | |
1302 | ||
1303 | return err; | |
1304 | } | |
1305 | ||
1306 | /* | |
d0d856e8 RD |
1307 | * ext4_ext_create_new_leaf: |
1308 | * finds empty index and adds new leaf. | |
1309 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1310 | */ |
1311 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
107a7bd3 TT |
1312 | unsigned int mb_flags, |
1313 | unsigned int gb_flags, | |
55f020db AH |
1314 | struct ext4_ext_path *path, |
1315 | struct ext4_extent *newext) | |
a86c6181 AT |
1316 | { |
1317 | struct ext4_ext_path *curp; | |
1318 | int depth, i, err = 0; | |
1319 | ||
1320 | repeat: | |
1321 | i = depth = ext_depth(inode); | |
1322 | ||
1323 | /* walk up to the tree and look for free index entry */ | |
1324 | curp = path + depth; | |
1325 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1326 | i--; | |
1327 | curp--; | |
1328 | } | |
1329 | ||
d0d856e8 RD |
1330 | /* we use already allocated block for index block, |
1331 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1332 | if (EXT_HAS_FREE_INDEX(curp)) { |
1333 | /* if we found index with free entry, then use that | |
1334 | * entry: create all needed subtree and add new leaf */ | |
107a7bd3 | 1335 | err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
787e0981 SF |
1336 | if (err) |
1337 | goto out; | |
a86c6181 AT |
1338 | |
1339 | /* refill path */ | |
1340 | ext4_ext_drop_refs(path); | |
1341 | path = ext4_ext_find_extent(inode, | |
725d26d3 | 1342 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
107a7bd3 | 1343 | path, gb_flags); |
a86c6181 AT |
1344 | if (IS_ERR(path)) |
1345 | err = PTR_ERR(path); | |
1346 | } else { | |
1347 | /* tree is full, time to grow in depth */ | |
107a7bd3 | 1348 | err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext); |
a86c6181 AT |
1349 | if (err) |
1350 | goto out; | |
1351 | ||
1352 | /* refill path */ | |
1353 | ext4_ext_drop_refs(path); | |
1354 | path = ext4_ext_find_extent(inode, | |
725d26d3 | 1355 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
107a7bd3 | 1356 | path, gb_flags); |
a86c6181 AT |
1357 | if (IS_ERR(path)) { |
1358 | err = PTR_ERR(path); | |
1359 | goto out; | |
1360 | } | |
1361 | ||
1362 | /* | |
d0d856e8 RD |
1363 | * only first (depth 0 -> 1) produces free space; |
1364 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1365 | */ |
1366 | depth = ext_depth(inode); | |
1367 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1368 | /* now we need to split */ |
a86c6181 AT |
1369 | goto repeat; |
1370 | } | |
1371 | } | |
1372 | ||
1373 | out: | |
1374 | return err; | |
1375 | } | |
1376 | ||
1988b51e AT |
1377 | /* |
1378 | * search the closest allocated block to the left for *logical | |
1379 | * and returns it at @logical + it's physical address at @phys | |
1380 | * if *logical is the smallest allocated block, the function | |
1381 | * returns 0 at @phys | |
1382 | * return value contains 0 (success) or error code | |
1383 | */ | |
1f109d5a TT |
1384 | static int ext4_ext_search_left(struct inode *inode, |
1385 | struct ext4_ext_path *path, | |
1386 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1387 | { |
1388 | struct ext4_extent_idx *ix; | |
1389 | struct ext4_extent *ex; | |
b939e376 | 1390 | int depth, ee_len; |
1988b51e | 1391 | |
273df556 FM |
1392 | if (unlikely(path == NULL)) { |
1393 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1394 | return -EIO; | |
1395 | } | |
1988b51e AT |
1396 | depth = path->p_depth; |
1397 | *phys = 0; | |
1398 | ||
1399 | if (depth == 0 && path->p_ext == NULL) | |
1400 | return 0; | |
1401 | ||
1402 | /* usually extent in the path covers blocks smaller | |
1403 | * then *logical, but it can be that extent is the | |
1404 | * first one in the file */ | |
1405 | ||
1406 | ex = path[depth].p_ext; | |
b939e376 | 1407 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1408 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1409 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1410 | EXT4_ERROR_INODE(inode, | |
1411 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1412 | *logical, le32_to_cpu(ex->ee_block)); | |
1413 | return -EIO; | |
1414 | } | |
1988b51e AT |
1415 | while (--depth >= 0) { |
1416 | ix = path[depth].p_idx; | |
273df556 FM |
1417 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1418 | EXT4_ERROR_INODE(inode, | |
1419 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1420 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1421 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1422 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1423 | depth); |
1424 | return -EIO; | |
1425 | } | |
1988b51e AT |
1426 | } |
1427 | return 0; | |
1428 | } | |
1429 | ||
273df556 FM |
1430 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1431 | EXT4_ERROR_INODE(inode, | |
1432 | "logical %d < ee_block %d + ee_len %d!", | |
1433 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1434 | return -EIO; | |
1435 | } | |
1988b51e | 1436 | |
b939e376 | 1437 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1438 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1439 | return 0; |
1440 | } | |
1441 | ||
1442 | /* | |
1443 | * search the closest allocated block to the right for *logical | |
1444 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1445 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1446 | * returns 0 at @phys |
1447 | * return value contains 0 (success) or error code | |
1448 | */ | |
1f109d5a TT |
1449 | static int ext4_ext_search_right(struct inode *inode, |
1450 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1451 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1452 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1453 | { |
1454 | struct buffer_head *bh = NULL; | |
1455 | struct ext4_extent_header *eh; | |
1456 | struct ext4_extent_idx *ix; | |
1457 | struct ext4_extent *ex; | |
1458 | ext4_fsblk_t block; | |
395a87bf ES |
1459 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1460 | int ee_len; | |
1988b51e | 1461 | |
273df556 FM |
1462 | if (unlikely(path == NULL)) { |
1463 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1464 | return -EIO; | |
1465 | } | |
1988b51e AT |
1466 | depth = path->p_depth; |
1467 | *phys = 0; | |
1468 | ||
1469 | if (depth == 0 && path->p_ext == NULL) | |
1470 | return 0; | |
1471 | ||
1472 | /* usually extent in the path covers blocks smaller | |
1473 | * then *logical, but it can be that extent is the | |
1474 | * first one in the file */ | |
1475 | ||
1476 | ex = path[depth].p_ext; | |
b939e376 | 1477 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1478 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1479 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1480 | EXT4_ERROR_INODE(inode, | |
1481 | "first_extent(path[%d].p_hdr) != ex", | |
1482 | depth); | |
1483 | return -EIO; | |
1484 | } | |
1988b51e AT |
1485 | while (--depth >= 0) { |
1486 | ix = path[depth].p_idx; | |
273df556 FM |
1487 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1488 | EXT4_ERROR_INODE(inode, | |
1489 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1490 | *logical); | |
1491 | return -EIO; | |
1492 | } | |
1988b51e | 1493 | } |
4d33b1ef | 1494 | goto found_extent; |
1988b51e AT |
1495 | } |
1496 | ||
273df556 FM |
1497 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1498 | EXT4_ERROR_INODE(inode, | |
1499 | "logical %d < ee_block %d + ee_len %d!", | |
1500 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1501 | return -EIO; | |
1502 | } | |
1988b51e AT |
1503 | |
1504 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1505 | /* next allocated block in this leaf */ | |
1506 | ex++; | |
4d33b1ef | 1507 | goto found_extent; |
1988b51e AT |
1508 | } |
1509 | ||
1510 | /* go up and search for index to the right */ | |
1511 | while (--depth >= 0) { | |
1512 | ix = path[depth].p_idx; | |
1513 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1514 | goto got_index; |
1988b51e AT |
1515 | } |
1516 | ||
25f1ee3a WF |
1517 | /* we've gone up to the root and found no index to the right */ |
1518 | return 0; | |
1988b51e | 1519 | |
25f1ee3a | 1520 | got_index: |
1988b51e AT |
1521 | /* we've found index to the right, let's |
1522 | * follow it and find the closest allocated | |
1523 | * block to the right */ | |
1524 | ix++; | |
bf89d16f | 1525 | block = ext4_idx_pblock(ix); |
1988b51e | 1526 | while (++depth < path->p_depth) { |
395a87bf | 1527 | /* subtract from p_depth to get proper eh_depth */ |
7d7ea89e | 1528 | bh = read_extent_tree_block(inode, block, |
107a7bd3 | 1529 | path->p_depth - depth, 0); |
7d7ea89e TT |
1530 | if (IS_ERR(bh)) |
1531 | return PTR_ERR(bh); | |
1532 | eh = ext_block_hdr(bh); | |
1988b51e | 1533 | ix = EXT_FIRST_INDEX(eh); |
bf89d16f | 1534 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1535 | put_bh(bh); |
1536 | } | |
1537 | ||
107a7bd3 | 1538 | bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0); |
7d7ea89e TT |
1539 | if (IS_ERR(bh)) |
1540 | return PTR_ERR(bh); | |
1988b51e | 1541 | eh = ext_block_hdr(bh); |
1988b51e | 1542 | ex = EXT_FIRST_EXTENT(eh); |
4d33b1ef | 1543 | found_extent: |
1988b51e | 1544 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1545 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1546 | *ret_ex = ex; |
1547 | if (bh) | |
1548 | put_bh(bh); | |
1988b51e | 1549 | return 0; |
1988b51e AT |
1550 | } |
1551 | ||
a86c6181 | 1552 | /* |
d0d856e8 | 1553 | * ext4_ext_next_allocated_block: |
f17722f9 | 1554 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1555 | * NOTE: it considers block number from index entry as |
1556 | * allocated block. Thus, index entries have to be consistent | |
1557 | * with leaves. | |
a86c6181 | 1558 | */ |
725d26d3 | 1559 | static ext4_lblk_t |
a86c6181 AT |
1560 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1561 | { | |
1562 | int depth; | |
1563 | ||
1564 | BUG_ON(path == NULL); | |
1565 | depth = path->p_depth; | |
1566 | ||
1567 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1568 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1569 | |
1570 | while (depth >= 0) { | |
1571 | if (depth == path->p_depth) { | |
1572 | /* leaf */ | |
6f8ff537 CW |
1573 | if (path[depth].p_ext && |
1574 | path[depth].p_ext != | |
a86c6181 AT |
1575 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1576 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1577 | } else { | |
1578 | /* index */ | |
1579 | if (path[depth].p_idx != | |
1580 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1581 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1582 | } | |
1583 | depth--; | |
1584 | } | |
1585 | ||
f17722f9 | 1586 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1587 | } |
1588 | ||
1589 | /* | |
d0d856e8 | 1590 | * ext4_ext_next_leaf_block: |
f17722f9 | 1591 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1592 | */ |
5718789d | 1593 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1594 | { |
1595 | int depth; | |
1596 | ||
1597 | BUG_ON(path == NULL); | |
1598 | depth = path->p_depth; | |
1599 | ||
1600 | /* zero-tree has no leaf blocks at all */ | |
1601 | if (depth == 0) | |
f17722f9 | 1602 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1603 | |
1604 | /* go to index block */ | |
1605 | depth--; | |
1606 | ||
1607 | while (depth >= 0) { | |
1608 | if (path[depth].p_idx != | |
1609 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1610 | return (ext4_lblk_t) |
1611 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1612 | depth--; |
1613 | } | |
1614 | ||
f17722f9 | 1615 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1616 | } |
1617 | ||
1618 | /* | |
d0d856e8 RD |
1619 | * ext4_ext_correct_indexes: |
1620 | * if leaf gets modified and modified extent is first in the leaf, | |
1621 | * then we have to correct all indexes above. | |
a86c6181 AT |
1622 | * TODO: do we need to correct tree in all cases? |
1623 | */ | |
1d03ec98 | 1624 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1625 | struct ext4_ext_path *path) |
1626 | { | |
1627 | struct ext4_extent_header *eh; | |
1628 | int depth = ext_depth(inode); | |
1629 | struct ext4_extent *ex; | |
1630 | __le32 border; | |
1631 | int k, err = 0; | |
1632 | ||
1633 | eh = path[depth].p_hdr; | |
1634 | ex = path[depth].p_ext; | |
273df556 FM |
1635 | |
1636 | if (unlikely(ex == NULL || eh == NULL)) { | |
1637 | EXT4_ERROR_INODE(inode, | |
1638 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1639 | return -EIO; | |
1640 | } | |
a86c6181 AT |
1641 | |
1642 | if (depth == 0) { | |
1643 | /* there is no tree at all */ | |
1644 | return 0; | |
1645 | } | |
1646 | ||
1647 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1648 | /* we correct tree if first leaf got modified only */ | |
1649 | return 0; | |
1650 | } | |
1651 | ||
1652 | /* | |
d0d856e8 | 1653 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1654 | */ |
1655 | k = depth - 1; | |
1656 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1657 | err = ext4_ext_get_access(handle, inode, path + k); |
1658 | if (err) | |
a86c6181 AT |
1659 | return err; |
1660 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1661 | err = ext4_ext_dirty(handle, inode, path + k); |
1662 | if (err) | |
a86c6181 AT |
1663 | return err; |
1664 | ||
1665 | while (k--) { | |
1666 | /* change all left-side indexes */ | |
1667 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1668 | break; | |
7e028976 AM |
1669 | err = ext4_ext_get_access(handle, inode, path + k); |
1670 | if (err) | |
a86c6181 AT |
1671 | break; |
1672 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1673 | err = ext4_ext_dirty(handle, inode, path + k); |
1674 | if (err) | |
a86c6181 AT |
1675 | break; |
1676 | } | |
1677 | ||
1678 | return err; | |
1679 | } | |
1680 | ||
748de673 | 1681 | int |
a86c6181 AT |
1682 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1683 | struct ext4_extent *ex2) | |
1684 | { | |
da0169b3 | 1685 | unsigned short ext1_ee_len, ext2_ee_len; |
a2df2a63 AA |
1686 | |
1687 | /* | |
ec22ba8e DM |
1688 | * Make sure that both extents are initialized. We don't merge |
1689 | * uninitialized extents so that we can be sure that end_io code has | |
1690 | * the extent that was written properly split out and conversion to | |
1691 | * initialized is trivial. | |
a2df2a63 | 1692 | */ |
a9b82415 | 1693 | if (ext4_ext_is_uninitialized(ex1) != ext4_ext_is_uninitialized(ex2)) |
a2df2a63 AA |
1694 | return 0; |
1695 | ||
1696 | ext1_ee_len = ext4_ext_get_actual_len(ex1); | |
1697 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1698 | ||
1699 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1700 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1701 | return 0; |
1702 | ||
471d4011 SB |
1703 | /* |
1704 | * To allow future support for preallocated extents to be added | |
1705 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1706 | * this can result in the top bit of ee_len being set. |
471d4011 | 1707 | */ |
da0169b3 | 1708 | if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN) |
471d4011 | 1709 | return 0; |
a9b82415 DW |
1710 | if (ext4_ext_is_uninitialized(ex1) && |
1711 | (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) || | |
1712 | atomic_read(&EXT4_I(inode)->i_unwritten) || | |
1713 | (ext1_ee_len + ext2_ee_len > EXT_UNINIT_MAX_LEN))) | |
1714 | return 0; | |
bbf2f9fb | 1715 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1716 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1717 | return 0; |
1718 | #endif | |
1719 | ||
bf89d16f | 1720 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1721 | return 1; |
1722 | return 0; | |
1723 | } | |
1724 | ||
56055d3a AA |
1725 | /* |
1726 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1727 | * It always tries to merge towards right. If you want to merge towards | |
1728 | * left, pass "ex - 1" as argument instead of "ex". | |
1729 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1730 | * 1 if they got merged. | |
1731 | */ | |
197217a5 | 1732 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1733 | struct ext4_ext_path *path, |
1734 | struct ext4_extent *ex) | |
56055d3a AA |
1735 | { |
1736 | struct ext4_extent_header *eh; | |
1737 | unsigned int depth, len; | |
a9b82415 | 1738 | int merge_done = 0, uninit; |
56055d3a AA |
1739 | |
1740 | depth = ext_depth(inode); | |
1741 | BUG_ON(path[depth].p_hdr == NULL); | |
1742 | eh = path[depth].p_hdr; | |
1743 | ||
1744 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1745 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1746 | break; | |
1747 | /* merge with next extent! */ | |
a9b82415 | 1748 | uninit = ext4_ext_is_uninitialized(ex); |
56055d3a AA |
1749 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
1750 | + ext4_ext_get_actual_len(ex + 1)); | |
a9b82415 DW |
1751 | if (uninit) |
1752 | ext4_ext_mark_uninitialized(ex); | |
56055d3a AA |
1753 | |
1754 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1755 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1756 | * sizeof(struct ext4_extent); | |
1757 | memmove(ex + 1, ex + 2, len); | |
1758 | } | |
e8546d06 | 1759 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1760 | merge_done = 1; |
1761 | WARN_ON(eh->eh_entries == 0); | |
1762 | if (!eh->eh_entries) | |
24676da4 | 1763 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1764 | } |
1765 | ||
1766 | return merge_done; | |
1767 | } | |
1768 | ||
ecb94f5f TT |
1769 | /* |
1770 | * This function does a very simple check to see if we can collapse | |
1771 | * an extent tree with a single extent tree leaf block into the inode. | |
1772 | */ | |
1773 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1774 | struct inode *inode, | |
1775 | struct ext4_ext_path *path) | |
1776 | { | |
1777 | size_t s; | |
1778 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1779 | ext4_fsblk_t blk; | |
1780 | ||
1781 | if ((path[0].p_depth != 1) || | |
1782 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1783 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1784 | return; | |
1785 | ||
1786 | /* | |
1787 | * We need to modify the block allocation bitmap and the block | |
1788 | * group descriptor to release the extent tree block. If we | |
1789 | * can't get the journal credits, give up. | |
1790 | */ | |
1791 | if (ext4_journal_extend(handle, 2)) | |
1792 | return; | |
1793 | ||
1794 | /* | |
1795 | * Copy the extent data up to the inode | |
1796 | */ | |
1797 | blk = ext4_idx_pblock(path[0].p_idx); | |
1798 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1799 | sizeof(struct ext4_extent_idx); | |
1800 | s += sizeof(struct ext4_extent_header); | |
1801 | ||
1802 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1803 | path[0].p_depth = 0; | |
1804 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1805 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1806 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1807 | ||
1808 | brelse(path[1].p_bh); | |
1809 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
7d734532 JK |
1810 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET | |
1811 | EXT4_FREE_BLOCKS_RESERVE); | |
ecb94f5f TT |
1812 | } |
1813 | ||
197217a5 YY |
1814 | /* |
1815 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1816 | * return 1 if merge left else 0. | |
1817 | */ | |
ecb94f5f TT |
1818 | static void ext4_ext_try_to_merge(handle_t *handle, |
1819 | struct inode *inode, | |
197217a5 YY |
1820 | struct ext4_ext_path *path, |
1821 | struct ext4_extent *ex) { | |
1822 | struct ext4_extent_header *eh; | |
1823 | unsigned int depth; | |
1824 | int merge_done = 0; | |
197217a5 YY |
1825 | |
1826 | depth = ext_depth(inode); | |
1827 | BUG_ON(path[depth].p_hdr == NULL); | |
1828 | eh = path[depth].p_hdr; | |
1829 | ||
1830 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1831 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1832 | ||
1833 | if (!merge_done) | |
ecb94f5f | 1834 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1835 | |
ecb94f5f | 1836 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1837 | } |
1838 | ||
25d14f98 AA |
1839 | /* |
1840 | * check if a portion of the "newext" extent overlaps with an | |
1841 | * existing extent. | |
1842 | * | |
1843 | * If there is an overlap discovered, it updates the length of the newext | |
1844 | * such that there will be no overlap, and then returns 1. | |
1845 | * If there is no overlap found, it returns 0. | |
1846 | */ | |
4d33b1ef TT |
1847 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1848 | struct inode *inode, | |
1f109d5a TT |
1849 | struct ext4_extent *newext, |
1850 | struct ext4_ext_path *path) | |
25d14f98 | 1851 | { |
725d26d3 | 1852 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1853 | unsigned int depth, len1; |
1854 | unsigned int ret = 0; | |
1855 | ||
1856 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1857 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1858 | depth = ext_depth(inode); |
1859 | if (!path[depth].p_ext) | |
1860 | goto out; | |
f5a44db5 | 1861 | b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block)); |
25d14f98 AA |
1862 | |
1863 | /* | |
1864 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1865 | * is before the requested block(s) |
25d14f98 AA |
1866 | */ |
1867 | if (b2 < b1) { | |
1868 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1869 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1870 | goto out; |
f5a44db5 | 1871 | b2 = EXT4_LBLK_CMASK(sbi, b2); |
25d14f98 AA |
1872 | } |
1873 | ||
725d26d3 | 1874 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1875 | if (b1 + len1 < b1) { |
f17722f9 | 1876 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1877 | newext->ee_len = cpu_to_le16(len1); |
1878 | ret = 1; | |
1879 | } | |
1880 | ||
1881 | /* check for overlap */ | |
1882 | if (b1 + len1 > b2) { | |
1883 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1884 | ret = 1; | |
1885 | } | |
1886 | out: | |
1887 | return ret; | |
1888 | } | |
1889 | ||
a86c6181 | 1890 | /* |
d0d856e8 RD |
1891 | * ext4_ext_insert_extent: |
1892 | * tries to merge requsted extent into the existing extent or | |
1893 | * inserts requested extent as new one into the tree, | |
1894 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1895 | */ |
1896 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1897 | struct ext4_ext_path *path, | |
107a7bd3 | 1898 | struct ext4_extent *newext, int gb_flags) |
a86c6181 | 1899 | { |
af5bc92d | 1900 | struct ext4_extent_header *eh; |
a86c6181 AT |
1901 | struct ext4_extent *ex, *fex; |
1902 | struct ext4_extent *nearex; /* nearest extent */ | |
1903 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1904 | int depth, len, err; |
1905 | ext4_lblk_t next; | |
a9b82415 | 1906 | int mb_flags = 0, uninit; |
a86c6181 | 1907 | |
273df556 FM |
1908 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1909 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1910 | return -EIO; | |
1911 | } | |
a86c6181 AT |
1912 | depth = ext_depth(inode); |
1913 | ex = path[depth].p_ext; | |
be8981be | 1914 | eh = path[depth].p_hdr; |
273df556 FM |
1915 | if (unlikely(path[depth].p_hdr == NULL)) { |
1916 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1917 | return -EIO; | |
1918 | } | |
a86c6181 AT |
1919 | |
1920 | /* try to insert block into found extent and return */ | |
107a7bd3 | 1921 | if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1922 | |
1923 | /* | |
be8981be LC |
1924 | * Try to see whether we should rather test the extent on |
1925 | * right from ex, or from the left of ex. This is because | |
1926 | * ext4_ext_find_extent() can return either extent on the | |
1927 | * left, or on the right from the searched position. This | |
1928 | * will make merging more effective. | |
a2df2a63 | 1929 | */ |
be8981be LC |
1930 | if (ex < EXT_LAST_EXTENT(eh) && |
1931 | (le32_to_cpu(ex->ee_block) + | |
1932 | ext4_ext_get_actual_len(ex) < | |
1933 | le32_to_cpu(newext->ee_block))) { | |
1934 | ex += 1; | |
1935 | goto prepend; | |
1936 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1937 | (le32_to_cpu(newext->ee_block) + | |
1938 | ext4_ext_get_actual_len(newext) < | |
1939 | le32_to_cpu(ex->ee_block))) | |
1940 | ex -= 1; | |
1941 | ||
1942 | /* Try to append newex to the ex */ | |
1943 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1944 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
1945 | "(from %llu)\n", | |
1946 | ext4_ext_is_uninitialized(newext), | |
1947 | ext4_ext_get_actual_len(newext), | |
1948 | le32_to_cpu(ex->ee_block), | |
1949 | ext4_ext_is_uninitialized(ex), | |
1950 | ext4_ext_get_actual_len(ex), | |
1951 | ext4_ext_pblock(ex)); | |
1952 | err = ext4_ext_get_access(handle, inode, | |
1953 | path + depth); | |
1954 | if (err) | |
1955 | return err; | |
a9b82415 | 1956 | uninit = ext4_ext_is_uninitialized(ex); |
be8981be | 1957 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
a2df2a63 | 1958 | + ext4_ext_get_actual_len(newext)); |
a9b82415 DW |
1959 | if (uninit) |
1960 | ext4_ext_mark_uninitialized(ex); | |
be8981be LC |
1961 | eh = path[depth].p_hdr; |
1962 | nearex = ex; | |
1963 | goto merge; | |
1964 | } | |
1965 | ||
1966 | prepend: | |
1967 | /* Try to prepend newex to the ex */ | |
1968 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
1969 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
1970 | "(from %llu)\n", | |
1971 | le32_to_cpu(newext->ee_block), | |
1972 | ext4_ext_is_uninitialized(newext), | |
1973 | ext4_ext_get_actual_len(newext), | |
1974 | le32_to_cpu(ex->ee_block), | |
1975 | ext4_ext_is_uninitialized(ex), | |
1976 | ext4_ext_get_actual_len(ex), | |
1977 | ext4_ext_pblock(ex)); | |
1978 | err = ext4_ext_get_access(handle, inode, | |
1979 | path + depth); | |
1980 | if (err) | |
1981 | return err; | |
1982 | ||
a9b82415 | 1983 | uninit = ext4_ext_is_uninitialized(ex); |
be8981be LC |
1984 | ex->ee_block = newext->ee_block; |
1985 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
1986 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1987 | + ext4_ext_get_actual_len(newext)); | |
a9b82415 DW |
1988 | if (uninit) |
1989 | ext4_ext_mark_uninitialized(ex); | |
be8981be LC |
1990 | eh = path[depth].p_hdr; |
1991 | nearex = ex; | |
1992 | goto merge; | |
1993 | } | |
a86c6181 AT |
1994 | } |
1995 | ||
a86c6181 AT |
1996 | depth = ext_depth(inode); |
1997 | eh = path[depth].p_hdr; | |
1998 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1999 | goto has_space; | |
2000 | ||
2001 | /* probably next leaf has space for us? */ | |
2002 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
2003 | next = EXT_MAX_BLOCKS; |
2004 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 2005 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 2006 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 2007 | ext_debug("next leaf block - %u\n", next); |
a86c6181 | 2008 | BUG_ON(npath != NULL); |
107a7bd3 | 2009 | npath = ext4_ext_find_extent(inode, next, NULL, 0); |
a86c6181 AT |
2010 | if (IS_ERR(npath)) |
2011 | return PTR_ERR(npath); | |
2012 | BUG_ON(npath->p_depth != path->p_depth); | |
2013 | eh = npath[depth].p_hdr; | |
2014 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 2015 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
2016 | le16_to_cpu(eh->eh_entries)); |
2017 | path = npath; | |
ffb505ff | 2018 | goto has_space; |
a86c6181 AT |
2019 | } |
2020 | ext_debug("next leaf has no free space(%d,%d)\n", | |
2021 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
2022 | } | |
2023 | ||
2024 | /* | |
d0d856e8 RD |
2025 | * There is no free space in the found leaf. |
2026 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 2027 | */ |
107a7bd3 TT |
2028 | if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
2029 | mb_flags = EXT4_MB_USE_RESERVED; | |
2030 | err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, | |
2031 | path, newext); | |
a86c6181 AT |
2032 | if (err) |
2033 | goto cleanup; | |
2034 | depth = ext_depth(inode); | |
2035 | eh = path[depth].p_hdr; | |
2036 | ||
2037 | has_space: | |
2038 | nearex = path[depth].p_ext; | |
2039 | ||
7e028976 AM |
2040 | err = ext4_ext_get_access(handle, inode, path + depth); |
2041 | if (err) | |
a86c6181 AT |
2042 | goto cleanup; |
2043 | ||
2044 | if (!nearex) { | |
2045 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 2046 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 2047 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2048 | ext4_ext_pblock(newext), |
553f9008 | 2049 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 2050 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
2051 | nearex = EXT_FIRST_EXTENT(eh); |
2052 | } else { | |
2053 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 2054 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 2055 | /* Insert after */ |
32de6756 YY |
2056 | ext_debug("insert %u:%llu:[%d]%d before: " |
2057 | "nearest %p\n", | |
80e675f9 EG |
2058 | le32_to_cpu(newext->ee_block), |
2059 | ext4_ext_pblock(newext), | |
2060 | ext4_ext_is_uninitialized(newext), | |
2061 | ext4_ext_get_actual_len(newext), | |
2062 | nearex); | |
2063 | nearex++; | |
2064 | } else { | |
2065 | /* Insert before */ | |
2066 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
2067 | ext_debug("insert %u:%llu:[%d]%d after: " |
2068 | "nearest %p\n", | |
8c55e204 | 2069 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2070 | ext4_ext_pblock(newext), |
553f9008 | 2071 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 2072 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
2073 | nearex); |
2074 | } | |
2075 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
2076 | if (len > 0) { | |
32de6756 | 2077 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
2078 | "move %d extents from 0x%p to 0x%p\n", |
2079 | le32_to_cpu(newext->ee_block), | |
2080 | ext4_ext_pblock(newext), | |
2081 | ext4_ext_is_uninitialized(newext), | |
2082 | ext4_ext_get_actual_len(newext), | |
2083 | len, nearex, nearex + 1); | |
2084 | memmove(nearex + 1, nearex, | |
2085 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2086 | } |
a86c6181 AT |
2087 | } |
2088 | ||
e8546d06 | 2089 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2090 | path[depth].p_ext = nearex; |
a86c6181 | 2091 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2092 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2093 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2094 | |
2095 | merge: | |
e7bcf823 | 2096 | /* try to merge extents */ |
107a7bd3 | 2097 | if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2098 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2099 | |
a86c6181 AT |
2100 | |
2101 | /* time to correct all indexes above */ | |
2102 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2103 | if (err) | |
2104 | goto cleanup; | |
2105 | ||
ecb94f5f | 2106 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2107 | |
2108 | cleanup: | |
2109 | if (npath) { | |
2110 | ext4_ext_drop_refs(npath); | |
2111 | kfree(npath); | |
2112 | } | |
a86c6181 AT |
2113 | return err; |
2114 | } | |
2115 | ||
91dd8c11 LC |
2116 | static int ext4_fill_fiemap_extents(struct inode *inode, |
2117 | ext4_lblk_t block, ext4_lblk_t num, | |
2118 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
2119 | { |
2120 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 2121 | struct ext4_extent *ex; |
69eb33dc | 2122 | struct extent_status es; |
91dd8c11 | 2123 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 2124 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
2125 | int exists, depth = 0, err = 0; |
2126 | unsigned int flags = 0; | |
2127 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 2128 | |
f17722f9 | 2129 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
2130 | num = last - block; |
2131 | /* find extent for this block */ | |
fab3a549 | 2132 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
2133 | |
2134 | if (path && ext_depth(inode) != depth) { | |
2135 | /* depth was changed. we have to realloc path */ | |
2136 | kfree(path); | |
2137 | path = NULL; | |
2138 | } | |
2139 | ||
107a7bd3 | 2140 | path = ext4_ext_find_extent(inode, block, path, 0); |
6873fa0d | 2141 | if (IS_ERR(path)) { |
91dd8c11 | 2142 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
2143 | err = PTR_ERR(path); |
2144 | path = NULL; | |
2145 | break; | |
2146 | } | |
2147 | ||
2148 | depth = ext_depth(inode); | |
273df556 | 2149 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2150 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2151 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2152 | err = -EIO; | |
2153 | break; | |
2154 | } | |
6873fa0d ES |
2155 | ex = path[depth].p_ext; |
2156 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2157 | ext4_ext_drop_refs(path); |
6873fa0d | 2158 | |
91dd8c11 | 2159 | flags = 0; |
6873fa0d ES |
2160 | exists = 0; |
2161 | if (!ex) { | |
2162 | /* there is no extent yet, so try to allocate | |
2163 | * all requested space */ | |
2164 | start = block; | |
2165 | end = block + num; | |
2166 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2167 | /* need to allocate space before found extent */ | |
2168 | start = block; | |
2169 | end = le32_to_cpu(ex->ee_block); | |
2170 | if (block + num < end) | |
2171 | end = block + num; | |
2172 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2173 | + ext4_ext_get_actual_len(ex)) { | |
2174 | /* need to allocate space after found extent */ | |
2175 | start = block; | |
2176 | end = block + num; | |
2177 | if (end >= next) | |
2178 | end = next; | |
2179 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2180 | /* | |
2181 | * some part of requested space is covered | |
2182 | * by found extent | |
2183 | */ | |
2184 | start = block; | |
2185 | end = le32_to_cpu(ex->ee_block) | |
2186 | + ext4_ext_get_actual_len(ex); | |
2187 | if (block + num < end) | |
2188 | end = block + num; | |
2189 | exists = 1; | |
2190 | } else { | |
2191 | BUG(); | |
2192 | } | |
2193 | BUG_ON(end <= start); | |
2194 | ||
2195 | if (!exists) { | |
69eb33dc ZL |
2196 | es.es_lblk = start; |
2197 | es.es_len = end - start; | |
2198 | es.es_pblk = 0; | |
6873fa0d | 2199 | } else { |
69eb33dc ZL |
2200 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2201 | es.es_len = ext4_ext_get_actual_len(ex); | |
2202 | es.es_pblk = ext4_ext_pblock(ex); | |
91dd8c11 LC |
2203 | if (ext4_ext_is_uninitialized(ex)) |
2204 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
6873fa0d ES |
2205 | } |
2206 | ||
91dd8c11 | 2207 | /* |
69eb33dc ZL |
2208 | * Find delayed extent and update es accordingly. We call |
2209 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2210 | * last existing extent or not. |
2211 | */ | |
69eb33dc | 2212 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2213 | if (!exists && next_del) { |
2214 | exists = 1; | |
72dac95d JL |
2215 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2216 | FIEMAP_EXTENT_UNKNOWN); | |
91dd8c11 LC |
2217 | } |
2218 | up_read(&EXT4_I(inode)->i_data_sem); | |
2219 | ||
69eb33dc ZL |
2220 | if (unlikely(es.es_len == 0)) { |
2221 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
273df556 FM |
2222 | err = -EIO; |
2223 | break; | |
2224 | } | |
6873fa0d | 2225 | |
f7fec032 ZL |
2226 | /* |
2227 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2228 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2229 | * possible that an extent is with unwritten and delayed | |
2230 | * status due to when an extent is delayed allocated and | |
2231 | * is allocated by fallocate status tree will track both of | |
2232 | * them in a extent. | |
2233 | * | |
2234 | * So we could return a unwritten and delayed extent, and | |
2235 | * its block is equal to 'next'. | |
2236 | */ | |
2237 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2238 | flags |= FIEMAP_EXTENT_LAST; |
2239 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2240 | next != EXT_MAX_BLOCKS)) { | |
2241 | EXT4_ERROR_INODE(inode, | |
2242 | "next extent == %u, next " | |
2243 | "delalloc extent = %u", | |
2244 | next, next_del); | |
2245 | err = -EIO; | |
2246 | break; | |
2247 | } | |
6873fa0d ES |
2248 | } |
2249 | ||
91dd8c11 LC |
2250 | if (exists) { |
2251 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2252 | (__u64)es.es_lblk << blksize_bits, |
2253 | (__u64)es.es_pblk << blksize_bits, | |
2254 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2255 | flags); |
2256 | if (err < 0) | |
2257 | break; | |
2258 | if (err == 1) { | |
2259 | err = 0; | |
2260 | break; | |
2261 | } | |
6873fa0d ES |
2262 | } |
2263 | ||
69eb33dc | 2264 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2265 | } |
2266 | ||
2267 | if (path) { | |
2268 | ext4_ext_drop_refs(path); | |
2269 | kfree(path); | |
2270 | } | |
2271 | ||
2272 | return err; | |
2273 | } | |
2274 | ||
a86c6181 | 2275 | /* |
d0d856e8 RD |
2276 | * ext4_ext_put_gap_in_cache: |
2277 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2278 | * and cache this gap |
2279 | */ | |
09b88252 | 2280 | static void |
a86c6181 | 2281 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2282 | ext4_lblk_t block) |
a86c6181 AT |
2283 | { |
2284 | int depth = ext_depth(inode); | |
27b1b228 AS |
2285 | unsigned long len = 0; |
2286 | ext4_lblk_t lblock = 0; | |
a86c6181 AT |
2287 | struct ext4_extent *ex; |
2288 | ||
2289 | ex = path[depth].p_ext; | |
2290 | if (ex == NULL) { | |
69eb33dc ZL |
2291 | /* |
2292 | * there is no extent yet, so gap is [0;-] and we | |
2293 | * don't cache it | |
2294 | */ | |
a86c6181 AT |
2295 | ext_debug("cache gap(whole file):"); |
2296 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2297 | lblock = block; | |
2298 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2299 | ext_debug("cache gap(before): %u [%u:%u]", |
2300 | block, | |
2301 | le32_to_cpu(ex->ee_block), | |
2302 | ext4_ext_get_actual_len(ex)); | |
d100eef2 ZL |
2303 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2304 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2305 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2306 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2307 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2308 | ext4_lblk_t next; |
8c55e204 | 2309 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2310 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2311 | |
2312 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2313 | ext_debug("cache gap(after): [%u:%u] %u", |
2314 | le32_to_cpu(ex->ee_block), | |
2315 | ext4_ext_get_actual_len(ex), | |
2316 | block); | |
725d26d3 AK |
2317 | BUG_ON(next == lblock); |
2318 | len = next - lblock; | |
d100eef2 ZL |
2319 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2320 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2321 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2322 | } else { |
a86c6181 AT |
2323 | BUG(); |
2324 | } | |
2325 | ||
bba90743 | 2326 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
2327 | } |
2328 | ||
2329 | /* | |
d0d856e8 RD |
2330 | * ext4_ext_rm_idx: |
2331 | * removes index from the index block. | |
a86c6181 | 2332 | */ |
1d03ec98 | 2333 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2334 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2335 | { |
a86c6181 | 2336 | int err; |
f65e6fba | 2337 | ext4_fsblk_t leaf; |
a86c6181 AT |
2338 | |
2339 | /* free index block */ | |
c36575e6 FL |
2340 | depth--; |
2341 | path = path + depth; | |
bf89d16f | 2342 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2343 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2344 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2345 | return -EIO; | |
2346 | } | |
7e028976 AM |
2347 | err = ext4_ext_get_access(handle, inode, path); |
2348 | if (err) | |
a86c6181 | 2349 | return err; |
0e1147b0 RD |
2350 | |
2351 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2352 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2353 | len *= sizeof(struct ext4_extent_idx); | |
2354 | memmove(path->p_idx, path->p_idx + 1, len); | |
2355 | } | |
2356 | ||
e8546d06 | 2357 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2358 | err = ext4_ext_dirty(handle, inode, path); |
2359 | if (err) | |
a86c6181 | 2360 | return err; |
2ae02107 | 2361 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2362 | trace_ext4_ext_rm_idx(inode, leaf); |
2363 | ||
7dc57615 | 2364 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2365 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2366 | |
2367 | while (--depth >= 0) { | |
2368 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2369 | break; | |
2370 | path--; | |
2371 | err = ext4_ext_get_access(handle, inode, path); | |
2372 | if (err) | |
2373 | break; | |
2374 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2375 | err = ext4_ext_dirty(handle, inode, path); | |
2376 | if (err) | |
2377 | break; | |
2378 | } | |
a86c6181 AT |
2379 | return err; |
2380 | } | |
2381 | ||
2382 | /* | |
ee12b630 MC |
2383 | * ext4_ext_calc_credits_for_single_extent: |
2384 | * This routine returns max. credits that needed to insert an extent | |
2385 | * to the extent tree. | |
2386 | * When pass the actual path, the caller should calculate credits | |
2387 | * under i_data_sem. | |
a86c6181 | 2388 | */ |
525f4ed8 | 2389 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2390 | struct ext4_ext_path *path) |
2391 | { | |
a86c6181 | 2392 | if (path) { |
ee12b630 | 2393 | int depth = ext_depth(inode); |
f3bd1f3f | 2394 | int ret = 0; |
ee12b630 | 2395 | |
a86c6181 | 2396 | /* probably there is space in leaf? */ |
a86c6181 | 2397 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2398 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2399 | |
ee12b630 MC |
2400 | /* |
2401 | * There are some space in the leaf tree, no | |
2402 | * need to account for leaf block credit | |
2403 | * | |
2404 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2405 | * and other metadata blocks still need to be |
ee12b630 MC |
2406 | * accounted. |
2407 | */ | |
525f4ed8 | 2408 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2409 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2410 | return ret; |
ee12b630 MC |
2411 | } |
2412 | } | |
a86c6181 | 2413 | |
525f4ed8 | 2414 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2415 | } |
a86c6181 | 2416 | |
ee12b630 | 2417 | /* |
fffb2739 | 2418 | * How many index/leaf blocks need to change/allocate to add @extents extents? |
ee12b630 | 2419 | * |
fffb2739 JK |
2420 | * If we add a single extent, then in the worse case, each tree level |
2421 | * index/leaf need to be changed in case of the tree split. | |
ee12b630 | 2422 | * |
fffb2739 JK |
2423 | * If more extents are inserted, they could cause the whole tree split more |
2424 | * than once, but this is really rare. | |
ee12b630 | 2425 | */ |
fffb2739 | 2426 | int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
ee12b630 MC |
2427 | { |
2428 | int index; | |
f19d5870 TM |
2429 | int depth; |
2430 | ||
2431 | /* If we are converting the inline data, only one is needed here. */ | |
2432 | if (ext4_has_inline_data(inode)) | |
2433 | return 1; | |
2434 | ||
2435 | depth = ext_depth(inode); | |
a86c6181 | 2436 | |
fffb2739 | 2437 | if (extents <= 1) |
ee12b630 MC |
2438 | index = depth * 2; |
2439 | else | |
2440 | index = depth * 3; | |
a86c6181 | 2441 | |
ee12b630 | 2442 | return index; |
a86c6181 AT |
2443 | } |
2444 | ||
981250ca TT |
2445 | static inline int get_default_free_blocks_flags(struct inode *inode) |
2446 | { | |
2447 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2448 | return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; | |
2449 | else if (ext4_should_journal_data(inode)) | |
2450 | return EXT4_FREE_BLOCKS_FORGET; | |
2451 | return 0; | |
2452 | } | |
2453 | ||
a86c6181 | 2454 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
0aa06000 | 2455 | struct ext4_extent *ex, |
d23142c6 | 2456 | long long *partial_cluster, |
0aa06000 | 2457 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2458 | { |
0aa06000 | 2459 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2460 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2461 | ext4_fsblk_t pblk; |
981250ca | 2462 | int flags = get_default_free_blocks_flags(inode); |
18888cf0 | 2463 | |
0aa06000 TT |
2464 | /* |
2465 | * For bigalloc file systems, we never free a partial cluster | |
2466 | * at the beginning of the extent. Instead, we make a note | |
2467 | * that we tried freeing the cluster, and check to see if we | |
2468 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2469 | * or at the end of the ext4_truncate() operation. | |
2470 | */ | |
2471 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2472 | ||
d8990240 | 2473 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2474 | /* |
2475 | * If we have a partial cluster, and it's different from the | |
2476 | * cluster of the last block, we need to explicitly free the | |
2477 | * partial cluster here. | |
2478 | */ | |
2479 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
d23142c6 LC |
2480 | if ((*partial_cluster > 0) && |
2481 | (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
0aa06000 TT |
2482 | ext4_free_blocks(handle, inode, NULL, |
2483 | EXT4_C2B(sbi, *partial_cluster), | |
2484 | sbi->s_cluster_ratio, flags); | |
2485 | *partial_cluster = 0; | |
2486 | } | |
2487 | ||
a86c6181 AT |
2488 | #ifdef EXTENTS_STATS |
2489 | { | |
2490 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2491 | spin_lock(&sbi->s_ext_stats_lock); |
2492 | sbi->s_ext_blocks += ee_len; | |
2493 | sbi->s_ext_extents++; | |
2494 | if (ee_len < sbi->s_ext_min) | |
2495 | sbi->s_ext_min = ee_len; | |
2496 | if (ee_len > sbi->s_ext_max) | |
2497 | sbi->s_ext_max = ee_len; | |
2498 | if (ext_depth(inode) > sbi->s_depth_max) | |
2499 | sbi->s_depth_max = ext_depth(inode); | |
2500 | spin_unlock(&sbi->s_ext_stats_lock); | |
2501 | } | |
2502 | #endif | |
2503 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2504 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2505 | /* tail removal */ |
725d26d3 | 2506 | ext4_lblk_t num; |
d23142c6 | 2507 | unsigned int unaligned; |
725d26d3 | 2508 | |
a2df2a63 | 2509 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 | 2510 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
d23142c6 LC |
2511 | /* |
2512 | * Usually we want to free partial cluster at the end of the | |
2513 | * extent, except for the situation when the cluster is still | |
2514 | * used by any other extent (partial_cluster is negative). | |
2515 | */ | |
2516 | if (*partial_cluster < 0 && | |
2517 | -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1)) | |
2518 | flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; | |
2519 | ||
2520 | ext_debug("free last %u blocks starting %llu partial %lld\n", | |
2521 | num, pblk, *partial_cluster); | |
0aa06000 TT |
2522 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); |
2523 | /* | |
2524 | * If the block range to be freed didn't start at the | |
2525 | * beginning of a cluster, and we removed the entire | |
d23142c6 LC |
2526 | * extent and the cluster is not used by any other extent, |
2527 | * save the partial cluster here, since we might need to | |
2528 | * delete if we determine that the truncate operation has | |
2529 | * removed all of the blocks in the cluster. | |
2530 | * | |
2531 | * On the other hand, if we did not manage to free the whole | |
2532 | * extent, we have to mark the cluster as used (store negative | |
2533 | * cluster number in partial_cluster). | |
0aa06000 | 2534 | */ |
f5a44db5 | 2535 | unaligned = EXT4_PBLK_COFF(sbi, pblk); |
d23142c6 LC |
2536 | if (unaligned && (ee_len == num) && |
2537 | (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk)))) | |
0aa06000 | 2538 | *partial_cluster = EXT4_B2C(sbi, pblk); |
d23142c6 LC |
2539 | else if (unaligned) |
2540 | *partial_cluster = -((long long)EXT4_B2C(sbi, pblk)); | |
2541 | else if (*partial_cluster > 0) | |
0aa06000 | 2542 | *partial_cluster = 0; |
78fb9cdf LC |
2543 | } else |
2544 | ext4_error(sbi->s_sb, "strange request: removal(2) " | |
2545 | "%u-%u from %u:%u\n", | |
2546 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2547 | return 0; |
2548 | } | |
2549 | ||
d583fb87 AH |
2550 | |
2551 | /* | |
2552 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2553 | * blocks appearing between "start" and "end", and splits the extents | |
2554 | * if "start" and "end" appear in the same extent | |
2555 | * | |
2556 | * @handle: The journal handle | |
2557 | * @inode: The files inode | |
2558 | * @path: The path to the leaf | |
d23142c6 LC |
2559 | * @partial_cluster: The cluster which we'll have to free if all extents |
2560 | * has been released from it. It gets negative in case | |
2561 | * that the cluster is still used. | |
d583fb87 AH |
2562 | * @start: The first block to remove |
2563 | * @end: The last block to remove | |
2564 | */ | |
a86c6181 AT |
2565 | static int |
2566 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
d23142c6 LC |
2567 | struct ext4_ext_path *path, |
2568 | long long *partial_cluster, | |
0aa06000 | 2569 | ext4_lblk_t start, ext4_lblk_t end) |
a86c6181 | 2570 | { |
0aa06000 | 2571 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2572 | int err = 0, correct_index = 0; |
2573 | int depth = ext_depth(inode), credits; | |
2574 | struct ext4_extent_header *eh; | |
750c9c47 | 2575 | ext4_lblk_t a, b; |
725d26d3 AK |
2576 | unsigned num; |
2577 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2578 | unsigned short ex_ee_len; |
a2df2a63 | 2579 | unsigned uninitialized = 0; |
a86c6181 | 2580 | struct ext4_extent *ex; |
d23142c6 | 2581 | ext4_fsblk_t pblk; |
a86c6181 | 2582 | |
c29c0ae7 | 2583 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2584 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2585 | if (!path[depth].p_hdr) |
2586 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2587 | eh = path[depth].p_hdr; | |
273df556 FM |
2588 | if (unlikely(path[depth].p_hdr == NULL)) { |
2589 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2590 | return -EIO; | |
2591 | } | |
a86c6181 | 2592 | /* find where to start removing */ |
6ae06ff5 AS |
2593 | ex = path[depth].p_ext; |
2594 | if (!ex) | |
2595 | ex = EXT_LAST_EXTENT(eh); | |
a86c6181 AT |
2596 | |
2597 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2598 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2599 | |
d8990240 AK |
2600 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2601 | ||
a86c6181 AT |
2602 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2603 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2604 | |
2605 | if (ext4_ext_is_uninitialized(ex)) | |
2606 | uninitialized = 1; | |
2607 | else | |
2608 | uninitialized = 0; | |
2609 | ||
553f9008 M |
2610 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2611 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2612 | path[depth].p_ext = ex; |
2613 | ||
2614 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2615 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2616 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2617 | |
2618 | ext_debug(" border %u:%u\n", a, b); | |
2619 | ||
d583fb87 | 2620 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2621 | if (end < ex_ee_block) { |
d23142c6 LC |
2622 | /* |
2623 | * We're going to skip this extent and move to another, | |
2624 | * so if this extent is not cluster aligned we have | |
2625 | * to mark the current cluster as used to avoid | |
2626 | * accidentally freeing it later on | |
2627 | */ | |
2628 | pblk = ext4_ext_pblock(ex); | |
f5a44db5 | 2629 | if (EXT4_PBLK_COFF(sbi, pblk)) |
d23142c6 LC |
2630 | *partial_cluster = |
2631 | -((long long)EXT4_B2C(sbi, pblk)); | |
d583fb87 AH |
2632 | ex--; |
2633 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2634 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2635 | continue; | |
750c9c47 | 2636 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2637 | EXT4_ERROR_INODE(inode, |
2638 | "can not handle truncate %u:%u " | |
2639 | "on extent %u:%u", | |
2640 | start, end, ex_ee_block, | |
2641 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2642 | err = -EIO; |
2643 | goto out; | |
a86c6181 AT |
2644 | } else if (a != ex_ee_block) { |
2645 | /* remove tail of the extent */ | |
750c9c47 | 2646 | num = a - ex_ee_block; |
a86c6181 AT |
2647 | } else { |
2648 | /* remove whole extent: excellent! */ | |
a86c6181 | 2649 | num = 0; |
a86c6181 | 2650 | } |
34071da7 TT |
2651 | /* |
2652 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2653 | * descriptor) for each block group; assume two block | |
2654 | * groups plus ex_ee_len/blocks_per_block_group for | |
2655 | * the worst case | |
2656 | */ | |
2657 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2658 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2659 | correct_index = 1; | |
2660 | credits += (ext_depth(inode)) + 1; | |
2661 | } | |
5aca07eb | 2662 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2663 | |
487caeef | 2664 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2665 | if (err) |
a86c6181 | 2666 | goto out; |
a86c6181 AT |
2667 | |
2668 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2669 | if (err) | |
2670 | goto out; | |
2671 | ||
0aa06000 TT |
2672 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2673 | a, b); | |
a86c6181 AT |
2674 | if (err) |
2675 | goto out; | |
2676 | ||
750c9c47 | 2677 | if (num == 0) |
d0d856e8 | 2678 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2679 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2680 | |
a86c6181 | 2681 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2682 | /* |
2683 | * Do not mark uninitialized if all the blocks in the | |
2684 | * extent have been removed. | |
2685 | */ | |
2686 | if (uninitialized && num) | |
a2df2a63 | 2687 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2688 | /* |
2689 | * If the extent was completely released, | |
2690 | * we need to remove it from the leaf | |
2691 | */ | |
2692 | if (num == 0) { | |
f17722f9 | 2693 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2694 | /* |
2695 | * For hole punching, we need to scoot all the | |
2696 | * extents up when an extent is removed so that | |
2697 | * we dont have blank extents in the middle | |
2698 | */ | |
2699 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2700 | sizeof(struct ext4_extent)); | |
2701 | ||
2702 | /* Now get rid of the one at the end */ | |
2703 | memset(EXT_LAST_EXTENT(eh), 0, | |
2704 | sizeof(struct ext4_extent)); | |
2705 | } | |
2706 | le16_add_cpu(&eh->eh_entries, -1); | |
d23142c6 | 2707 | } else if (*partial_cluster > 0) |
0aa06000 | 2708 | *partial_cluster = 0; |
d583fb87 | 2709 | |
750c9c47 DM |
2710 | err = ext4_ext_dirty(handle, inode, path + depth); |
2711 | if (err) | |
2712 | goto out; | |
2713 | ||
bf52c6f7 | 2714 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2715 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2716 | ex--; |
2717 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2718 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2719 | } |
2720 | ||
2721 | if (correct_index && eh->eh_entries) | |
2722 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2723 | ||
0aa06000 | 2724 | /* |
d23142c6 LC |
2725 | * Free the partial cluster only if the current extent does not |
2726 | * reference it. Otherwise we might free used cluster. | |
0aa06000 | 2727 | */ |
d23142c6 | 2728 | if (*partial_cluster > 0 && |
0aa06000 TT |
2729 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != |
2730 | *partial_cluster)) { | |
981250ca | 2731 | int flags = get_default_free_blocks_flags(inode); |
0aa06000 TT |
2732 | |
2733 | ext4_free_blocks(handle, inode, NULL, | |
2734 | EXT4_C2B(sbi, *partial_cluster), | |
2735 | sbi->s_cluster_ratio, flags); | |
2736 | *partial_cluster = 0; | |
2737 | } | |
2738 | ||
a86c6181 AT |
2739 | /* if this leaf is free, then we should |
2740 | * remove it from index block above */ | |
2741 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2742 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2743 | |
2744 | out: | |
2745 | return err; | |
2746 | } | |
2747 | ||
2748 | /* | |
d0d856e8 RD |
2749 | * ext4_ext_more_to_rm: |
2750 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2751 | */ |
09b88252 | 2752 | static int |
a86c6181 AT |
2753 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2754 | { | |
2755 | BUG_ON(path->p_idx == NULL); | |
2756 | ||
2757 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2758 | return 0; | |
2759 | ||
2760 | /* | |
d0d856e8 | 2761 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2762 | * so we have to consider current index for truncation |
2763 | */ | |
2764 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2765 | return 0; | |
2766 | return 1; | |
2767 | } | |
2768 | ||
26a4c0c6 TT |
2769 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2770 | ext4_lblk_t end) | |
a86c6181 AT |
2771 | { |
2772 | struct super_block *sb = inode->i_sb; | |
2773 | int depth = ext_depth(inode); | |
968dee77 | 2774 | struct ext4_ext_path *path = NULL; |
d23142c6 | 2775 | long long partial_cluster = 0; |
a86c6181 | 2776 | handle_t *handle; |
6f2080e6 | 2777 | int i = 0, err = 0; |
a86c6181 | 2778 | |
5f95d21f | 2779 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2780 | |
2781 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2782 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2783 | if (IS_ERR(handle)) |
2784 | return PTR_ERR(handle); | |
2785 | ||
0617b83f | 2786 | again: |
61801325 | 2787 | trace_ext4_ext_remove_space(inode, start, end, depth); |
d8990240 | 2788 | |
5f95d21f LC |
2789 | /* |
2790 | * Check if we are removing extents inside the extent tree. If that | |
2791 | * is the case, we are going to punch a hole inside the extent tree | |
2792 | * so we have to check whether we need to split the extent covering | |
2793 | * the last block to remove so we can easily remove the part of it | |
2794 | * in ext4_ext_rm_leaf(). | |
2795 | */ | |
2796 | if (end < EXT_MAX_BLOCKS - 1) { | |
2797 | struct ext4_extent *ex; | |
2798 | ext4_lblk_t ee_block; | |
2799 | ||
2800 | /* find extent for this block */ | |
107a7bd3 | 2801 | path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE); |
5f95d21f LC |
2802 | if (IS_ERR(path)) { |
2803 | ext4_journal_stop(handle); | |
2804 | return PTR_ERR(path); | |
2805 | } | |
2806 | depth = ext_depth(inode); | |
6f2080e6 | 2807 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2808 | ex = path[depth].p_ext; |
968dee77 | 2809 | if (!ex) { |
6f2080e6 DM |
2810 | if (depth) { |
2811 | EXT4_ERROR_INODE(inode, | |
2812 | "path[%d].p_hdr == NULL", | |
2813 | depth); | |
2814 | err = -EIO; | |
2815 | } | |
2816 | goto out; | |
968dee77 | 2817 | } |
5f95d21f LC |
2818 | |
2819 | ee_block = le32_to_cpu(ex->ee_block); | |
2820 | ||
2821 | /* | |
2822 | * See if the last block is inside the extent, if so split | |
2823 | * the extent at 'end' block so we can easily remove the | |
2824 | * tail of the first part of the split extent in | |
2825 | * ext4_ext_rm_leaf(). | |
2826 | */ | |
2827 | if (end >= ee_block && | |
2828 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2829 | int split_flag = 0; | |
2830 | ||
2831 | if (ext4_ext_is_uninitialized(ex)) | |
2832 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2833 | EXT4_EXT_MARK_UNINIT2; | |
2834 | ||
2835 | /* | |
2836 | * Split the extent in two so that 'end' is the last | |
27dd4385 LC |
2837 | * block in the first new extent. Also we should not |
2838 | * fail removing space due to ENOSPC so try to use | |
2839 | * reserved block if that happens. | |
5f95d21f LC |
2840 | */ |
2841 | err = ext4_split_extent_at(handle, inode, path, | |
27dd4385 | 2842 | end + 1, split_flag, |
107a7bd3 | 2843 | EXT4_EX_NOCACHE | |
27dd4385 LC |
2844 | EXT4_GET_BLOCKS_PRE_IO | |
2845 | EXT4_GET_BLOCKS_METADATA_NOFAIL); | |
5f95d21f LC |
2846 | |
2847 | if (err < 0) | |
2848 | goto out; | |
2849 | } | |
5f95d21f | 2850 | } |
a86c6181 | 2851 | /* |
d0d856e8 RD |
2852 | * We start scanning from right side, freeing all the blocks |
2853 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2854 | */ |
0617b83f | 2855 | depth = ext_depth(inode); |
968dee77 AS |
2856 | if (path) { |
2857 | int k = i = depth; | |
2858 | while (--k > 0) | |
2859 | path[k].p_block = | |
2860 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2861 | } else { | |
2862 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2863 | GFP_NOFS); | |
2864 | if (path == NULL) { | |
2865 | ext4_journal_stop(handle); | |
2866 | return -ENOMEM; | |
2867 | } | |
2868 | path[0].p_depth = depth; | |
2869 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2870 | i = 0; |
5f95d21f | 2871 | |
c349179b | 2872 | if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
968dee77 AS |
2873 | err = -EIO; |
2874 | goto out; | |
2875 | } | |
a86c6181 | 2876 | } |
968dee77 | 2877 | err = 0; |
a86c6181 AT |
2878 | |
2879 | while (i >= 0 && err == 0) { | |
2880 | if (i == depth) { | |
2881 | /* this is leaf block */ | |
d583fb87 | 2882 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2883 | &partial_cluster, start, |
5f95d21f | 2884 | end); |
d0d856e8 | 2885 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2886 | brelse(path[i].p_bh); |
2887 | path[i].p_bh = NULL; | |
2888 | i--; | |
2889 | continue; | |
2890 | } | |
2891 | ||
2892 | /* this is index block */ | |
2893 | if (!path[i].p_hdr) { | |
2894 | ext_debug("initialize header\n"); | |
2895 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2896 | } |
2897 | ||
a86c6181 | 2898 | if (!path[i].p_idx) { |
d0d856e8 | 2899 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2900 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2901 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2902 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2903 | path[i].p_hdr, | |
2904 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2905 | } else { | |
d0d856e8 | 2906 | /* we were already here, see at next index */ |
a86c6181 AT |
2907 | path[i].p_idx--; |
2908 | } | |
2909 | ||
2910 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2911 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2912 | path[i].p_idx); | |
2913 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2914 | struct buffer_head *bh; |
a86c6181 | 2915 | /* go to the next level */ |
2ae02107 | 2916 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2917 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2918 | memset(path + i + 1, 0, sizeof(*path)); |
7d7ea89e | 2919 | bh = read_extent_tree_block(inode, |
107a7bd3 TT |
2920 | ext4_idx_pblock(path[i].p_idx), depth - i - 1, |
2921 | EXT4_EX_NOCACHE); | |
7d7ea89e | 2922 | if (IS_ERR(bh)) { |
a86c6181 | 2923 | /* should we reset i_size? */ |
7d7ea89e | 2924 | err = PTR_ERR(bh); |
a86c6181 AT |
2925 | break; |
2926 | } | |
76828c88 TT |
2927 | /* Yield here to deal with large extent trees. |
2928 | * Should be a no-op if we did IO above. */ | |
2929 | cond_resched(); | |
c29c0ae7 AT |
2930 | if (WARN_ON(i + 1 > depth)) { |
2931 | err = -EIO; | |
c29c0ae7 AT |
2932 | break; |
2933 | } | |
2934 | path[i + 1].p_bh = bh; | |
a86c6181 | 2935 | |
d0d856e8 RD |
2936 | /* save actual number of indexes since this |
2937 | * number is changed at the next iteration */ | |
a86c6181 AT |
2938 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2939 | i++; | |
2940 | } else { | |
d0d856e8 | 2941 | /* we finished processing this index, go up */ |
a86c6181 | 2942 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2943 | /* index is empty, remove it; |
a86c6181 AT |
2944 | * handle must be already prepared by the |
2945 | * truncatei_leaf() */ | |
c36575e6 | 2946 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2947 | } |
d0d856e8 | 2948 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2949 | brelse(path[i].p_bh); |
2950 | path[i].p_bh = NULL; | |
2951 | i--; | |
2952 | ext_debug("return to level %d\n", i); | |
2953 | } | |
2954 | } | |
2955 | ||
61801325 LC |
2956 | trace_ext4_ext_remove_space_done(inode, start, end, depth, |
2957 | partial_cluster, path->p_hdr->eh_entries); | |
d8990240 | 2958 | |
7b415bf6 AK |
2959 | /* If we still have something in the partial cluster and we have removed |
2960 | * even the first extent, then we should free the blocks in the partial | |
2961 | * cluster as well. */ | |
d23142c6 | 2962 | if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) { |
981250ca | 2963 | int flags = get_default_free_blocks_flags(inode); |
7b415bf6 AK |
2964 | |
2965 | ext4_free_blocks(handle, inode, NULL, | |
2966 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2967 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2968 | partial_cluster = 0; | |
2969 | } | |
2970 | ||
a86c6181 AT |
2971 | /* TODO: flexible tree reduction should be here */ |
2972 | if (path->p_hdr->eh_entries == 0) { | |
2973 | /* | |
d0d856e8 RD |
2974 | * truncate to zero freed all the tree, |
2975 | * so we need to correct eh_depth | |
a86c6181 AT |
2976 | */ |
2977 | err = ext4_ext_get_access(handle, inode, path); | |
2978 | if (err == 0) { | |
2979 | ext_inode_hdr(inode)->eh_depth = 0; | |
2980 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2981 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2982 | err = ext4_ext_dirty(handle, inode, path); |
2983 | } | |
2984 | } | |
2985 | out: | |
a86c6181 AT |
2986 | ext4_ext_drop_refs(path); |
2987 | kfree(path); | |
968dee77 AS |
2988 | if (err == -EAGAIN) { |
2989 | path = NULL; | |
0617b83f | 2990 | goto again; |
968dee77 | 2991 | } |
a86c6181 AT |
2992 | ext4_journal_stop(handle); |
2993 | ||
2994 | return err; | |
2995 | } | |
2996 | ||
2997 | /* | |
2998 | * called at mount time | |
2999 | */ | |
3000 | void ext4_ext_init(struct super_block *sb) | |
3001 | { | |
3002 | /* | |
3003 | * possible initialization would be here | |
3004 | */ | |
3005 | ||
83982b6f | 3006 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 3007 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 3008 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 3009 | #ifdef AGGRESSIVE_TEST |
92b97816 | 3010 | ", aggressive tests" |
a86c6181 AT |
3011 | #endif |
3012 | #ifdef CHECK_BINSEARCH | |
92b97816 | 3013 | ", check binsearch" |
a86c6181 AT |
3014 | #endif |
3015 | #ifdef EXTENTS_STATS | |
92b97816 | 3016 | ", stats" |
a86c6181 | 3017 | #endif |
92b97816 | 3018 | "\n"); |
90576c0b | 3019 | #endif |
a86c6181 AT |
3020 | #ifdef EXTENTS_STATS |
3021 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
3022 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
3023 | EXT4_SB(sb)->s_ext_max = 0; | |
3024 | #endif | |
3025 | } | |
3026 | } | |
3027 | ||
3028 | /* | |
3029 | * called at umount time | |
3030 | */ | |
3031 | void ext4_ext_release(struct super_block *sb) | |
3032 | { | |
83982b6f | 3033 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
3034 | return; |
3035 | ||
3036 | #ifdef EXTENTS_STATS | |
3037 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
3038 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3039 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
3040 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
3041 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
3042 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
3043 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
3044 | } | |
3045 | #endif | |
3046 | } | |
3047 | ||
d7b2a00c ZL |
3048 | static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex) |
3049 | { | |
3050 | ext4_lblk_t ee_block; | |
3051 | ext4_fsblk_t ee_pblock; | |
3052 | unsigned int ee_len; | |
3053 | ||
3054 | ee_block = le32_to_cpu(ex->ee_block); | |
3055 | ee_len = ext4_ext_get_actual_len(ex); | |
3056 | ee_pblock = ext4_ext_pblock(ex); | |
3057 | ||
3058 | if (ee_len == 0) | |
3059 | return 0; | |
3060 | ||
3061 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
3062 | EXTENT_STATUS_WRITTEN); | |
3063 | } | |
3064 | ||
093a088b AK |
3065 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
3066 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
3067 | { | |
2407518d LC |
3068 | ext4_fsblk_t ee_pblock; |
3069 | unsigned int ee_len; | |
b720303d | 3070 | int ret; |
093a088b | 3071 | |
093a088b | 3072 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 3073 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 3074 | |
a107e5a3 | 3075 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
3076 | if (ret > 0) |
3077 | ret = 0; | |
093a088b | 3078 | |
2407518d | 3079 | return ret; |
093a088b AK |
3080 | } |
3081 | ||
47ea3bb5 YY |
3082 | /* |
3083 | * ext4_split_extent_at() splits an extent at given block. | |
3084 | * | |
3085 | * @handle: the journal handle | |
3086 | * @inode: the file inode | |
3087 | * @path: the path to the extent | |
3088 | * @split: the logical block where the extent is splitted. | |
3089 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
3090 | * the states(init or uninit) of new extents. | |
3091 | * @flags: flags used to insert new extent to extent tree. | |
3092 | * | |
3093 | * | |
3094 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
3095 | * of which are deterimined by split_flag. | |
3096 | * | |
3097 | * There are two cases: | |
3098 | * a> the extent are splitted into two extent. | |
3099 | * b> split is not needed, and just mark the extent. | |
3100 | * | |
3101 | * return 0 on success. | |
3102 | */ | |
3103 | static int ext4_split_extent_at(handle_t *handle, | |
3104 | struct inode *inode, | |
3105 | struct ext4_ext_path *path, | |
3106 | ext4_lblk_t split, | |
3107 | int split_flag, | |
3108 | int flags) | |
3109 | { | |
3110 | ext4_fsblk_t newblock; | |
3111 | ext4_lblk_t ee_block; | |
adb23551 | 3112 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
3113 | struct ext4_extent *ex2 = NULL; |
3114 | unsigned int ee_len, depth; | |
3115 | int err = 0; | |
3116 | ||
dee1f973 DM |
3117 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
3118 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
3119 | ||
47ea3bb5 YY |
3120 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
3121 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3122 | ||
3123 | ext4_ext_show_leaf(inode, path); | |
3124 | ||
3125 | depth = ext_depth(inode); | |
3126 | ex = path[depth].p_ext; | |
3127 | ee_block = le32_to_cpu(ex->ee_block); | |
3128 | ee_len = ext4_ext_get_actual_len(ex); | |
3129 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3130 | ||
3131 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
357b66fd DM |
3132 | BUG_ON(!ext4_ext_is_uninitialized(ex) && |
3133 | split_flag & (EXT4_EXT_MAY_ZEROOUT | | |
3134 | EXT4_EXT_MARK_UNINIT1 | | |
3135 | EXT4_EXT_MARK_UNINIT2)); | |
47ea3bb5 YY |
3136 | |
3137 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3138 | if (err) | |
3139 | goto out; | |
3140 | ||
3141 | if (split == ee_block) { | |
3142 | /* | |
3143 | * case b: block @split is the block that the extent begins with | |
3144 | * then we just change the state of the extent, and splitting | |
3145 | * is not needed. | |
3146 | */ | |
3147 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3148 | ext4_ext_mark_uninitialized(ex); | |
3149 | else | |
3150 | ext4_ext_mark_initialized(ex); | |
3151 | ||
3152 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3153 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3154 | |
ecb94f5f | 3155 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3156 | goto out; |
3157 | } | |
3158 | ||
3159 | /* case a */ | |
3160 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3161 | ex->ee_len = cpu_to_le16(split - ee_block); | |
3162 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
3163 | ext4_ext_mark_uninitialized(ex); | |
3164 | ||
3165 | /* | |
3166 | * path may lead to new leaf, not to original leaf any more | |
3167 | * after ext4_ext_insert_extent() returns, | |
3168 | */ | |
3169 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3170 | if (err) | |
3171 | goto fix_extent_len; | |
3172 | ||
3173 | ex2 = &newex; | |
3174 | ex2->ee_block = cpu_to_le32(split); | |
3175 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3176 | ext4_ext_store_pblock(ex2, newblock); | |
3177 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3178 | ext4_ext_mark_uninitialized(ex2); | |
3179 | ||
3180 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
3181 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 | 3182 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3183 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3184 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3185 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3186 | zero_ex.ee_len = cpu_to_le16( |
3187 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3188 | ext4_ext_store_pblock(&zero_ex, |
3189 | ext4_ext_pblock(ex2)); | |
3190 | } else { | |
dee1f973 | 3191 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3192 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3193 | zero_ex.ee_len = cpu_to_le16( |
3194 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3195 | ext4_ext_store_pblock(&zero_ex, |
3196 | ext4_ext_pblock(ex)); | |
3197 | } | |
3198 | } else { | |
dee1f973 | 3199 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3200 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3201 | zero_ex.ee_len = cpu_to_le16( |
3202 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3203 | ext4_ext_store_pblock(&zero_ex, |
3204 | ext4_ext_pblock(&orig_ex)); | |
3205 | } | |
dee1f973 | 3206 | |
47ea3bb5 YY |
3207 | if (err) |
3208 | goto fix_extent_len; | |
3209 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3210 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3211 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3212 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3213 | if (err) |
3214 | goto fix_extent_len; | |
3215 | ||
3216 | /* update extent status tree */ | |
d7b2a00c | 3217 | err = ext4_zeroout_es(inode, &zero_ex); |
adb23551 | 3218 | |
47ea3bb5 YY |
3219 | goto out; |
3220 | } else if (err) | |
3221 | goto fix_extent_len; | |
3222 | ||
3223 | out: | |
3224 | ext4_ext_show_leaf(inode, path); | |
3225 | return err; | |
3226 | ||
3227 | fix_extent_len: | |
3228 | ex->ee_len = orig_ex.ee_len; | |
3229 | ext4_ext_dirty(handle, inode, path + depth); | |
3230 | return err; | |
3231 | } | |
3232 | ||
3233 | /* | |
3234 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3235 | * by @map as split_flags indicates | |
3236 | * | |
70261f56 | 3237 | * It may result in splitting the extent into multiple extents (up to three) |
47ea3bb5 YY |
3238 | * There are three possibilities: |
3239 | * a> There is no split required | |
3240 | * b> Splits in two extents: Split is happening at either end of the extent | |
3241 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3242 | * | |
3243 | */ | |
3244 | static int ext4_split_extent(handle_t *handle, | |
3245 | struct inode *inode, | |
3246 | struct ext4_ext_path *path, | |
3247 | struct ext4_map_blocks *map, | |
3248 | int split_flag, | |
3249 | int flags) | |
3250 | { | |
3251 | ext4_lblk_t ee_block; | |
3252 | struct ext4_extent *ex; | |
3253 | unsigned int ee_len, depth; | |
3254 | int err = 0; | |
3255 | int uninitialized; | |
3256 | int split_flag1, flags1; | |
3a225670 | 3257 | int allocated = map->m_len; |
47ea3bb5 YY |
3258 | |
3259 | depth = ext_depth(inode); | |
3260 | ex = path[depth].p_ext; | |
3261 | ee_block = le32_to_cpu(ex->ee_block); | |
3262 | ee_len = ext4_ext_get_actual_len(ex); | |
3263 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3264 | ||
3265 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3266 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3267 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3268 | if (uninitialized) | |
3269 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3270 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3271 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3272 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3273 | err = ext4_split_extent_at(handle, inode, path, |
3274 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3275 | if (err) |
3276 | goto out; | |
3a225670 ZL |
3277 | } else { |
3278 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3279 | } |
357b66fd DM |
3280 | /* |
3281 | * Update path is required because previous ext4_split_extent_at() may | |
3282 | * result in split of original leaf or extent zeroout. | |
3283 | */ | |
47ea3bb5 | 3284 | ext4_ext_drop_refs(path); |
107a7bd3 | 3285 | path = ext4_ext_find_extent(inode, map->m_lblk, path, 0); |
47ea3bb5 YY |
3286 | if (IS_ERR(path)) |
3287 | return PTR_ERR(path); | |
357b66fd DM |
3288 | depth = ext_depth(inode); |
3289 | ex = path[depth].p_ext; | |
3290 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3291 | split_flag1 = 0; | |
47ea3bb5 YY |
3292 | |
3293 | if (map->m_lblk >= ee_block) { | |
357b66fd DM |
3294 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
3295 | if (uninitialized) { | |
47ea3bb5 | 3296 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; |
357b66fd DM |
3297 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3298 | EXT4_EXT_MARK_UNINIT2); | |
3299 | } | |
47ea3bb5 YY |
3300 | err = ext4_split_extent_at(handle, inode, path, |
3301 | map->m_lblk, split_flag1, flags); | |
3302 | if (err) | |
3303 | goto out; | |
3304 | } | |
3305 | ||
3306 | ext4_ext_show_leaf(inode, path); | |
3307 | out: | |
3a225670 | 3308 | return err ? err : allocated; |
47ea3bb5 YY |
3309 | } |
3310 | ||
56055d3a | 3311 | /* |
e35fd660 | 3312 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3313 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3314 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3315 | * uninitialized). |
3316 | * There are three possibilities: | |
3317 | * a> There is no split required: Entire extent should be initialized | |
3318 | * b> Splits in two extents: Write is happening at either end of the extent | |
3319 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3320 | * |
3321 | * Pre-conditions: | |
3322 | * - The extent pointed to by 'path' is uninitialized. | |
3323 | * - The extent pointed to by 'path' contains a superset | |
3324 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3325 | * | |
3326 | * Post-conditions on success: | |
3327 | * - the returned value is the number of blocks beyond map->l_lblk | |
3328 | * that are allocated and initialized. | |
3329 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3330 | */ |
725d26d3 | 3331 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3332 | struct inode *inode, |
3333 | struct ext4_map_blocks *map, | |
27dd4385 LC |
3334 | struct ext4_ext_path *path, |
3335 | int flags) | |
56055d3a | 3336 | { |
67a5da56 | 3337 | struct ext4_sb_info *sbi; |
6f91bc5f | 3338 | struct ext4_extent_header *eh; |
667eff35 YY |
3339 | struct ext4_map_blocks split_map; |
3340 | struct ext4_extent zero_ex; | |
bc2d9db4 | 3341 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3342 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3343 | unsigned int ee_len, depth, map_len = map->m_len; |
3344 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3345 | int err = 0; |
667eff35 | 3346 | int split_flag = 0; |
21ca087a DM |
3347 | |
3348 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3349 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3350 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3351 | |
67a5da56 | 3352 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3353 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3354 | inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3355 | if (eof_block < map->m_lblk + map_len) |
3356 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3357 | |
3358 | depth = ext_depth(inode); | |
6f91bc5f | 3359 | eh = path[depth].p_hdr; |
56055d3a AA |
3360 | ex = path[depth].p_ext; |
3361 | ee_block = le32_to_cpu(ex->ee_block); | |
3362 | ee_len = ext4_ext_get_actual_len(ex); | |
adb23551 | 3363 | zero_ex.ee_len = 0; |
56055d3a | 3364 | |
6f91bc5f EG |
3365 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3366 | ||
3367 | /* Pre-conditions */ | |
3368 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3369 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3370 | |
3371 | /* | |
3372 | * Attempt to transfer newly initialized blocks from the currently | |
bc2d9db4 | 3373 | * uninitialized extent to its neighbor. This is much cheaper |
6f91bc5f | 3374 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3375 | * memmove() calls. Transferring to the left is the common case in |
3376 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3377 | * followed by append writes. | |
6f91bc5f EG |
3378 | * |
3379 | * Limitations of the current logic: | |
bc2d9db4 | 3380 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3381 | * This would require removing the extent if the transfer |
3382 | * is possible. | |
bc2d9db4 | 3383 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3384 | * same extent tree node. |
3385 | */ | |
bc2d9db4 LC |
3386 | if ((map->m_lblk == ee_block) && |
3387 | /* See if we can merge left */ | |
3388 | (map_len < ee_len) && /*L1*/ | |
3389 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3390 | ext4_lblk_t prev_lblk; |
3391 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3392 | unsigned int prev_len; |
6f91bc5f | 3393 | |
bc2d9db4 LC |
3394 | abut_ex = ex - 1; |
3395 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3396 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3397 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3398 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3399 | |
3400 | /* | |
bc2d9db4 | 3401 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3402 | * upon those conditions: |
bc2d9db4 LC |
3403 | * - C1: abut_ex is initialized, |
3404 | * - C2: abut_ex is logically abutting ex, | |
3405 | * - C3: abut_ex is physically abutting ex, | |
3406 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3407 | * overflowing the (initialized) length limit. |
3408 | */ | |
bc2d9db4 | 3409 | if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3410 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3411 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3412 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3413 | err = ext4_ext_get_access(handle, inode, path + depth); |
3414 | if (err) | |
3415 | goto out; | |
3416 | ||
3417 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3418 | map, ex, abut_ex); |
6f91bc5f | 3419 | |
bc2d9db4 LC |
3420 | /* Shift the start of ex by 'map_len' blocks */ |
3421 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3422 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3423 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
6f91bc5f EG |
3424 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ |
3425 | ||
bc2d9db4 LC |
3426 | /* Extend abut_ex by 'map_len' blocks */ |
3427 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3428 | |
bc2d9db4 LC |
3429 | /* Result: number of initialized blocks past m_lblk */ |
3430 | allocated = map_len; | |
3431 | } | |
3432 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3433 | (map_len < ee_len) && /*L1*/ | |
3434 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3435 | /* See if we can merge right */ | |
3436 | ext4_lblk_t next_lblk; | |
3437 | ext4_fsblk_t next_pblk, ee_pblk; | |
3438 | unsigned int next_len; | |
3439 | ||
3440 | abut_ex = ex + 1; | |
3441 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3442 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3443 | next_pblk = ext4_ext_pblock(abut_ex); | |
3444 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3445 | |
bc2d9db4 LC |
3446 | /* |
3447 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3448 | * upon those conditions: | |
3449 | * - C1: abut_ex is initialized, | |
3450 | * - C2: abut_ex is logically abutting ex, | |
3451 | * - C3: abut_ex is physically abutting ex, | |
3452 | * - C4: abut_ex can receive the additional blocks without | |
3453 | * overflowing the (initialized) length limit. | |
3454 | */ | |
3455 | if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/ | |
3456 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ | |
3457 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3458 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3459 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3460 | if (err) | |
3461 | goto out; | |
3462 | ||
3463 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3464 | map, ex, abut_ex); | |
3465 | ||
3466 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3467 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3468 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3469 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
3470 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3471 | ||
3472 | /* Extend abut_ex by 'map_len' blocks */ | |
3473 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3474 | |
3475 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3476 | allocated = map_len; |
6f91bc5f EG |
3477 | } |
3478 | } | |
bc2d9db4 LC |
3479 | if (allocated) { |
3480 | /* Mark the block containing both extents as dirty */ | |
3481 | ext4_ext_dirty(handle, inode, path + depth); | |
3482 | ||
3483 | /* Update path to point to the right extent */ | |
3484 | path[depth].p_ext = abut_ex; | |
3485 | goto out; | |
3486 | } else | |
3487 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3488 | |
667eff35 | 3489 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3490 | /* |
3491 | * It is safe to convert extent to initialized via explicit | |
9e740568 | 3492 | * zeroout only if extent is fully inside i_size or new_size. |
21ca087a | 3493 | */ |
667eff35 | 3494 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3495 | |
67a5da56 ZL |
3496 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3497 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3498 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 ZL |
3499 | |
3500 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3501 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3502 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3503 | if (err) |
d03856bd | 3504 | goto out; |
adb23551 | 3505 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 | 3506 | zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)); |
adb23551 | 3507 | ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex)); |
d03856bd AK |
3508 | |
3509 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3510 | if (err) | |
3511 | goto out; | |
667eff35 | 3512 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3513 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3514 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3515 | goto out; |
56055d3a | 3516 | } |
667eff35 | 3517 | |
56055d3a | 3518 | /* |
667eff35 YY |
3519 | * four cases: |
3520 | * 1. split the extent into three extents. | |
3521 | * 2. split the extent into two extents, zeroout the first half. | |
3522 | * 3. split the extent into two extents, zeroout the second half. | |
3523 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3524 | */ |
667eff35 YY |
3525 | split_map.m_lblk = map->m_lblk; |
3526 | split_map.m_len = map->m_len; | |
3527 | ||
67a5da56 ZL |
3528 | if (max_zeroout && (allocated > map->m_len)) { |
3529 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3530 | /* case 3 */ |
3531 | zero_ex.ee_block = | |
9b940f8e AH |
3532 | cpu_to_le32(map->m_lblk); |
3533 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3534 | ext4_ext_store_pblock(&zero_ex, |
3535 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3536 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3537 | if (err) |
3538 | goto out; | |
667eff35 YY |
3539 | split_map.m_lblk = map->m_lblk; |
3540 | split_map.m_len = allocated; | |
67a5da56 | 3541 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3542 | /* case 2 */ |
3543 | if (map->m_lblk != ee_block) { | |
3544 | zero_ex.ee_block = ex->ee_block; | |
3545 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3546 | ee_block); | |
3547 | ext4_ext_store_pblock(&zero_ex, | |
3548 | ext4_ext_pblock(ex)); | |
3549 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3550 | if (err) | |
3551 | goto out; | |
3552 | } | |
3553 | ||
667eff35 | 3554 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3555 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3556 | allocated = map->m_len; | |
56055d3a AA |
3557 | } |
3558 | } | |
667eff35 YY |
3559 | |
3560 | allocated = ext4_split_extent(handle, inode, path, | |
27dd4385 | 3561 | &split_map, split_flag, flags); |
667eff35 YY |
3562 | if (allocated < 0) |
3563 | err = allocated; | |
3564 | ||
56055d3a | 3565 | out: |
adb23551 ZL |
3566 | /* If we have gotten a failure, don't zero out status tree */ |
3567 | if (!err) | |
d7b2a00c | 3568 | err = ext4_zeroout_es(inode, &zero_ex); |
56055d3a AA |
3569 | return err ? err : allocated; |
3570 | } | |
3571 | ||
0031462b | 3572 | /* |
e35fd660 | 3573 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3574 | * ext4_get_blocks_dio_write() when DIO to write |
3575 | * to an uninitialized extent. | |
3576 | * | |
fd018fe8 | 3577 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3578 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3579 | * There are three possibilities: |
3580 | * a> There is no split required: Entire extent should be uninitialized | |
3581 | * b> Splits in two extents: Write is happening at either end of the extent | |
3582 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3583 | * | |
3584 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3585 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3586 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3587 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3588 | * into three uninitialized extent(at most). After IO complete, the part |
3589 | * being filled will be convert to initialized by the end_io callback function | |
3590 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3591 | * |
3592 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3593 | */ |
3594 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3595 | struct inode *inode, | |
e35fd660 | 3596 | struct ext4_map_blocks *map, |
0031462b | 3597 | struct ext4_ext_path *path, |
0031462b MC |
3598 | int flags) |
3599 | { | |
667eff35 YY |
3600 | ext4_lblk_t eof_block; |
3601 | ext4_lblk_t ee_block; | |
3602 | struct ext4_extent *ex; | |
3603 | unsigned int ee_len; | |
3604 | int split_flag = 0, depth; | |
21ca087a DM |
3605 | |
3606 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3607 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3608 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3609 | |
3610 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3611 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3612 | if (eof_block < map->m_lblk + map->m_len) |
3613 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3614 | /* |
3615 | * It is safe to convert extent to initialized via explicit | |
3616 | * zeroout only if extent is fully insde i_size or new_size. | |
3617 | */ | |
667eff35 YY |
3618 | depth = ext_depth(inode); |
3619 | ex = path[depth].p_ext; | |
3620 | ee_block = le32_to_cpu(ex->ee_block); | |
3621 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3622 | |
667eff35 YY |
3623 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3624 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3625 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3626 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3627 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3628 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3629 | } |
197217a5 | 3630 | |
c7064ef1 | 3631 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3632 | struct inode *inode, |
3633 | struct ext4_map_blocks *map, | |
3634 | struct ext4_ext_path *path) | |
0031462b MC |
3635 | { |
3636 | struct ext4_extent *ex; | |
dee1f973 DM |
3637 | ext4_lblk_t ee_block; |
3638 | unsigned int ee_len; | |
0031462b MC |
3639 | int depth; |
3640 | int err = 0; | |
0031462b MC |
3641 | |
3642 | depth = ext_depth(inode); | |
0031462b | 3643 | ex = path[depth].p_ext; |
dee1f973 DM |
3644 | ee_block = le32_to_cpu(ex->ee_block); |
3645 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3646 | |
197217a5 YY |
3647 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3648 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3649 | (unsigned long long)ee_block, ee_len); |
3650 | ||
ff95ec22 DM |
3651 | /* If extent is larger than requested it is a clear sign that we still |
3652 | * have some extent state machine issues left. So extent_split is still | |
3653 | * required. | |
3654 | * TODO: Once all related issues will be fixed this situation should be | |
3655 | * illegal. | |
3656 | */ | |
dee1f973 | 3657 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ff95ec22 DM |
3658 | #ifdef EXT4_DEBUG |
3659 | ext4_warning("Inode (%ld) finished: extent logical block %llu," | |
3660 | " len %u; IO logical block %llu, len %u\n", | |
3661 | inode->i_ino, (unsigned long long)ee_block, ee_len, | |
3662 | (unsigned long long)map->m_lblk, map->m_len); | |
3663 | #endif | |
dee1f973 DM |
3664 | err = ext4_split_unwritten_extents(handle, inode, map, path, |
3665 | EXT4_GET_BLOCKS_CONVERT); | |
3666 | if (err < 0) | |
3667 | goto out; | |
3668 | ext4_ext_drop_refs(path); | |
107a7bd3 | 3669 | path = ext4_ext_find_extent(inode, map->m_lblk, path, 0); |
dee1f973 DM |
3670 | if (IS_ERR(path)) { |
3671 | err = PTR_ERR(path); | |
3672 | goto out; | |
3673 | } | |
3674 | depth = ext_depth(inode); | |
3675 | ex = path[depth].p_ext; | |
3676 | } | |
197217a5 | 3677 | |
0031462b MC |
3678 | err = ext4_ext_get_access(handle, inode, path + depth); |
3679 | if (err) | |
3680 | goto out; | |
3681 | /* first mark the extent as initialized */ | |
3682 | ext4_ext_mark_initialized(ex); | |
3683 | ||
197217a5 YY |
3684 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3685 | * borders are not changed | |
0031462b | 3686 | */ |
ecb94f5f | 3687 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3688 | |
0031462b | 3689 | /* Mark modified extent as dirty */ |
ecb94f5f | 3690 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3691 | out: |
3692 | ext4_ext_show_leaf(inode, path); | |
3693 | return err; | |
3694 | } | |
3695 | ||
515f41c3 AK |
3696 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3697 | sector_t block, int count) | |
3698 | { | |
3699 | int i; | |
3700 | for (i = 0; i < count; i++) | |
3701 | unmap_underlying_metadata(bdev, block + i); | |
3702 | } | |
3703 | ||
58590b06 TT |
3704 | /* |
3705 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3706 | */ | |
3707 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3708 | ext4_lblk_t lblk, |
58590b06 TT |
3709 | struct ext4_ext_path *path, |
3710 | unsigned int len) | |
3711 | { | |
3712 | int i, depth; | |
3713 | struct ext4_extent_header *eh; | |
65922cb5 | 3714 | struct ext4_extent *last_ex; |
58590b06 TT |
3715 | |
3716 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3717 | return 0; | |
3718 | ||
3719 | depth = ext_depth(inode); | |
3720 | eh = path[depth].p_hdr; | |
58590b06 | 3721 | |
afcff5d8 LC |
3722 | /* |
3723 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3724 | * do not care for this case anymore. Simply remove the flag | |
3725 | * if there are no extents. | |
3726 | */ | |
3727 | if (unlikely(!eh->eh_entries)) | |
3728 | goto out; | |
58590b06 TT |
3729 | last_ex = EXT_LAST_EXTENT(eh); |
3730 | /* | |
3731 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3732 | * last block in the last extent in the file. We test this by | |
3733 | * first checking to see if the caller to | |
3734 | * ext4_ext_get_blocks() was interested in the last block (or | |
3735 | * a block beyond the last block) in the current extent. If | |
3736 | * this turns out to be false, we can bail out from this | |
3737 | * function immediately. | |
3738 | */ | |
d002ebf1 | 3739 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3740 | ext4_ext_get_actual_len(last_ex)) |
3741 | return 0; | |
3742 | /* | |
3743 | * If the caller does appear to be planning to write at or | |
3744 | * beyond the end of the current extent, we then test to see | |
3745 | * if the current extent is the last extent in the file, by | |
3746 | * checking to make sure it was reached via the rightmost node | |
3747 | * at each level of the tree. | |
3748 | */ | |
3749 | for (i = depth-1; i >= 0; i--) | |
3750 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3751 | return 0; | |
afcff5d8 | 3752 | out: |
58590b06 TT |
3753 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3754 | return ext4_mark_inode_dirty(handle, inode); | |
3755 | } | |
3756 | ||
7b415bf6 AK |
3757 | /** |
3758 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3759 | * | |
7d1b1fbc | 3760 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3761 | */ |
f7fec032 ZL |
3762 | int ext4_find_delalloc_range(struct inode *inode, |
3763 | ext4_lblk_t lblk_start, | |
3764 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3765 | { |
7d1b1fbc | 3766 | struct extent_status es; |
7b415bf6 | 3767 | |
e30b5dca | 3768 | ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es); |
06b0c886 | 3769 | if (es.es_len == 0) |
7d1b1fbc | 3770 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3771 | else if (es.es_lblk <= lblk_start && |
3772 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3773 | return 1; |
06b0c886 | 3774 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3775 | return 1; |
7b415bf6 | 3776 | else |
7d1b1fbc | 3777 | return 0; |
7b415bf6 AK |
3778 | } |
3779 | ||
7d1b1fbc | 3780 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3781 | { |
3782 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3783 | ext4_lblk_t lblk_start, lblk_end; | |
f5a44db5 | 3784 | lblk_start = EXT4_LBLK_CMASK(sbi, lblk); |
7b415bf6 AK |
3785 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; |
3786 | ||
7d1b1fbc | 3787 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3788 | } |
3789 | ||
3790 | /** | |
3791 | * Determines how many complete clusters (out of those specified by the 'map') | |
3792 | * are under delalloc and were reserved quota for. | |
3793 | * This function is called when we are writing out the blocks that were | |
3794 | * originally written with their allocation delayed, but then the space was | |
3795 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3796 | * The cases to look for are: | |
3797 | * ('=' indicated delayed allocated blocks | |
3798 | * '-' indicates non-delayed allocated blocks) | |
3799 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3800 | * are not delalloc'ed. | |
3801 | * Ex: | |
3802 | * |----c---=|====c====|====c====|===-c----| | |
3803 | * |++++++ allocated ++++++| | |
3804 | * ==> 4 complete clusters in above example | |
3805 | * | |
3806 | * (b) partial cluster (outside of allocated range) towards either end is | |
3807 | * marked for delayed allocation. In this case, we will exclude that | |
3808 | * cluster. | |
3809 | * Ex: | |
3810 | * |----====c========|========c========| | |
3811 | * |++++++ allocated ++++++| | |
3812 | * ==> 1 complete clusters in above example | |
3813 | * | |
3814 | * Ex: | |
3815 | * |================c================| | |
3816 | * |++++++ allocated ++++++| | |
3817 | * ==> 0 complete clusters in above example | |
3818 | * | |
3819 | * The ext4_da_update_reserve_space will be called only if we | |
3820 | * determine here that there were some "entire" clusters that span | |
3821 | * this 'allocated' range. | |
3822 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3823 | * without ever calling ext4_find_delalloc_range. | |
3824 | */ | |
3825 | static unsigned int | |
3826 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3827 | unsigned int num_blks) | |
3828 | { | |
3829 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3830 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3831 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3832 | unsigned int allocated_clusters = 0; | |
3833 | ||
3834 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3835 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3836 | ||
3837 | /* max possible clusters for this allocation */ | |
3838 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3839 | ||
d8990240 AK |
3840 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3841 | ||
7b415bf6 | 3842 | /* Check towards left side */ |
f5a44db5 | 3843 | c_offset = EXT4_LBLK_COFF(sbi, lblk_start); |
7b415bf6 | 3844 | if (c_offset) { |
f5a44db5 | 3845 | lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start); |
7b415bf6 AK |
3846 | lblk_to = lblk_from + c_offset - 1; |
3847 | ||
7d1b1fbc | 3848 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3849 | allocated_clusters--; |
3850 | } | |
3851 | ||
3852 | /* Now check towards right. */ | |
f5a44db5 | 3853 | c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks); |
7b415bf6 AK |
3854 | if (allocated_clusters && c_offset) { |
3855 | lblk_from = lblk_start + num_blks; | |
3856 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3857 | ||
7d1b1fbc | 3858 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3859 | allocated_clusters--; |
3860 | } | |
3861 | ||
3862 | return allocated_clusters; | |
3863 | } | |
3864 | ||
0031462b MC |
3865 | static int |
3866 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3867 | struct ext4_map_blocks *map, |
0031462b | 3868 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3869 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3870 | { |
3871 | int ret = 0; | |
3872 | int err = 0; | |
f45ee3a1 | 3873 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3874 | |
88635ca2 ZL |
3875 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3876 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3877 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3878 | flags, allocated); |
3879 | ext4_ext_show_leaf(inode, path); | |
3880 | ||
27dd4385 LC |
3881 | /* |
3882 | * When writing into uninitialized space, we should not fail to | |
3883 | * allocate metadata blocks for the new extent block if needed. | |
3884 | */ | |
3885 | flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
3886 | ||
b5645534 ZL |
3887 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3888 | allocated, newblock); | |
d8990240 | 3889 | |
c7064ef1 | 3890 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3891 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3892 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3893 | path, flags); | |
82e54229 DM |
3894 | if (ret <= 0) |
3895 | goto out; | |
5f524950 M |
3896 | /* |
3897 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3898 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3899 | * completed |
3900 | */ | |
0edeb71d TM |
3901 | if (io) |
3902 | ext4_set_io_unwritten_flag(inode, io); | |
3903 | else | |
19f5fb7a | 3904 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
a25a4e1a | 3905 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
744692dc | 3906 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3907 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3908 | goto out; |
3909 | } | |
c7064ef1 | 3910 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3911 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3912 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3913 | path); |
58590b06 | 3914 | if (ret >= 0) { |
b436b9be | 3915 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3916 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3917 | path, map->m_len); | |
58590b06 TT |
3918 | } else |
3919 | err = ret; | |
cdee7843 | 3920 | map->m_flags |= EXT4_MAP_MAPPED; |
15cc1767 | 3921 | map->m_pblk = newblock; |
cdee7843 ZL |
3922 | if (allocated > map->m_len) |
3923 | allocated = map->m_len; | |
3924 | map->m_len = allocated; | |
0031462b MC |
3925 | goto out2; |
3926 | } | |
3927 | /* buffered IO case */ | |
3928 | /* | |
3929 | * repeat fallocate creation request | |
3930 | * we already have an unwritten extent | |
3931 | */ | |
a25a4e1a ZL |
3932 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) { |
3933 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
0031462b | 3934 | goto map_out; |
a25a4e1a | 3935 | } |
0031462b MC |
3936 | |
3937 | /* buffered READ or buffered write_begin() lookup */ | |
3938 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3939 | /* | |
3940 | * We have blocks reserved already. We | |
3941 | * return allocated blocks so that delalloc | |
3942 | * won't do block reservation for us. But | |
3943 | * the buffer head will be unmapped so that | |
3944 | * a read from the block returns 0s. | |
3945 | */ | |
e35fd660 | 3946 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3947 | goto out1; |
3948 | } | |
3949 | ||
3950 | /* buffered write, writepage time, convert*/ | |
27dd4385 | 3951 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags); |
a4e5d88b | 3952 | if (ret >= 0) |
b436b9be | 3953 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3954 | out: |
3955 | if (ret <= 0) { | |
3956 | err = ret; | |
3957 | goto out2; | |
3958 | } else | |
3959 | allocated = ret; | |
e35fd660 | 3960 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3961 | /* |
3962 | * if we allocated more blocks than requested | |
3963 | * we need to make sure we unmap the extra block | |
3964 | * allocated. The actual needed block will get | |
3965 | * unmapped later when we find the buffer_head marked | |
3966 | * new. | |
3967 | */ | |
e35fd660 | 3968 | if (allocated > map->m_len) { |
515f41c3 | 3969 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3970 | newblock + map->m_len, |
3971 | allocated - map->m_len); | |
3972 | allocated = map->m_len; | |
515f41c3 | 3973 | } |
3a225670 | 3974 | map->m_len = allocated; |
5f634d06 AK |
3975 | |
3976 | /* | |
3977 | * If we have done fallocate with the offset that is already | |
3978 | * delayed allocated, we would have block reservation | |
3979 | * and quota reservation done in the delayed write path. | |
3980 | * But fallocate would have already updated quota and block | |
3981 | * count for this offset. So cancel these reservation | |
3982 | */ | |
7b415bf6 AK |
3983 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3984 | unsigned int reserved_clusters; | |
3985 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3986 | map->m_lblk, map->m_len); | |
3987 | if (reserved_clusters) | |
3988 | ext4_da_update_reserve_space(inode, | |
3989 | reserved_clusters, | |
3990 | 0); | |
3991 | } | |
5f634d06 | 3992 | |
0031462b | 3993 | map_out: |
e35fd660 | 3994 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3995 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3996 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3997 | map->m_len); | |
3998 | if (err < 0) | |
3999 | goto out2; | |
4000 | } | |
0031462b | 4001 | out1: |
e35fd660 TT |
4002 | if (allocated > map->m_len) |
4003 | allocated = map->m_len; | |
0031462b | 4004 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4005 | map->m_pblk = newblock; |
4006 | map->m_len = allocated; | |
0031462b | 4007 | out2: |
0031462b MC |
4008 | return err ? err : allocated; |
4009 | } | |
58590b06 | 4010 | |
4d33b1ef TT |
4011 | /* |
4012 | * get_implied_cluster_alloc - check to see if the requested | |
4013 | * allocation (in the map structure) overlaps with a cluster already | |
4014 | * allocated in an extent. | |
d8990240 | 4015 | * @sb The filesystem superblock structure |
4d33b1ef TT |
4016 | * @map The requested lblk->pblk mapping |
4017 | * @ex The extent structure which might contain an implied | |
4018 | * cluster allocation | |
4019 | * | |
4020 | * This function is called by ext4_ext_map_blocks() after we failed to | |
4021 | * find blocks that were already in the inode's extent tree. Hence, | |
4022 | * we know that the beginning of the requested region cannot overlap | |
4023 | * the extent from the inode's extent tree. There are three cases we | |
4024 | * want to catch. The first is this case: | |
4025 | * | |
4026 | * |--- cluster # N--| | |
4027 | * |--- extent ---| |---- requested region ---| | |
4028 | * |==========| | |
4029 | * | |
4030 | * The second case that we need to test for is this one: | |
4031 | * | |
4032 | * |--------- cluster # N ----------------| | |
4033 | * |--- requested region --| |------- extent ----| | |
4034 | * |=======================| | |
4035 | * | |
4036 | * The third case is when the requested region lies between two extents | |
4037 | * within the same cluster: | |
4038 | * |------------- cluster # N-------------| | |
4039 | * |----- ex -----| |---- ex_right ----| | |
4040 | * |------ requested region ------| | |
4041 | * |================| | |
4042 | * | |
4043 | * In each of the above cases, we need to set the map->m_pblk and | |
4044 | * map->m_len so it corresponds to the return the extent labelled as | |
4045 | * "|====|" from cluster #N, since it is already in use for data in | |
4046 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
4047 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
4048 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
4049 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
4050 | * by calling ext4_mb_new_blocks(). | |
4051 | */ | |
d8990240 | 4052 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
4053 | struct ext4_map_blocks *map, |
4054 | struct ext4_extent *ex, | |
4055 | struct ext4_ext_path *path) | |
4056 | { | |
d8990240 | 4057 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
f5a44db5 | 4058 | ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef | 4059 | ext4_lblk_t ex_cluster_start, ex_cluster_end; |
14d7f3ef | 4060 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
4061 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
4062 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
4063 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
4064 | ||
4065 | /* The extent passed in that we are trying to match */ | |
4066 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
4067 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
4068 | ||
4069 | /* The requested region passed into ext4_map_blocks() */ | |
4070 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
4071 | |
4072 | if ((rr_cluster_start == ex_cluster_end) || | |
4073 | (rr_cluster_start == ex_cluster_start)) { | |
4074 | if (rr_cluster_start == ex_cluster_end) | |
4075 | ee_start += ee_len - 1; | |
f5a44db5 | 4076 | map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset; |
4d33b1ef TT |
4077 | map->m_len = min(map->m_len, |
4078 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
4079 | /* | |
4080 | * Check for and handle this case: | |
4081 | * | |
4082 | * |--------- cluster # N-------------| | |
4083 | * |------- extent ----| | |
4084 | * |--- requested region ---| | |
4085 | * |===========| | |
4086 | */ | |
4087 | ||
4088 | if (map->m_lblk < ee_block) | |
4089 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
4090 | ||
4091 | /* | |
4092 | * Check for the case where there is already another allocated | |
4093 | * block to the right of 'ex' but before the end of the cluster. | |
4094 | * | |
4095 | * |------------- cluster # N-------------| | |
4096 | * |----- ex -----| |---- ex_right ----| | |
4097 | * |------ requested region ------| | |
4098 | * |================| | |
4099 | */ | |
4100 | if (map->m_lblk > ee_block) { | |
4101 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
4102 | map->m_len = min(map->m_len, next - map->m_lblk); | |
4103 | } | |
d8990240 AK |
4104 | |
4105 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
4106 | return 1; |
4107 | } | |
d8990240 AK |
4108 | |
4109 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
4110 | return 0; |
4111 | } | |
4112 | ||
4113 | ||
c278bfec | 4114 | /* |
f5ab0d1f MC |
4115 | * Block allocation/map/preallocation routine for extents based files |
4116 | * | |
4117 | * | |
c278bfec | 4118 | * Need to be called with |
0e855ac8 AK |
4119 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
4120 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
4121 | * |
4122 | * return > 0, number of of blocks already mapped/allocated | |
4123 | * if create == 0 and these are pre-allocated blocks | |
4124 | * buffer head is unmapped | |
4125 | * otherwise blocks are mapped | |
4126 | * | |
4127 | * return = 0, if plain look up failed (blocks have not been allocated) | |
4128 | * buffer head is unmapped | |
4129 | * | |
4130 | * return < 0, error case. | |
c278bfec | 4131 | */ |
e35fd660 TT |
4132 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
4133 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
4134 | { |
4135 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
4136 | struct ext4_extent newex, *ex, *ex2; |
4137 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 4138 | ext4_fsblk_t newblock = 0; |
ce37c429 | 4139 | int free_on_err = 0, err = 0, depth, ret; |
4d33b1ef | 4140 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 4141 | unsigned int allocated_clusters = 0; |
c9de560d | 4142 | struct ext4_allocation_request ar; |
f45ee3a1 | 4143 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 4144 | ext4_lblk_t cluster_offset; |
82e54229 | 4145 | int set_unwritten = 0; |
a86c6181 | 4146 | |
84fe3bef | 4147 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4148 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4149 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4150 | |
a86c6181 | 4151 | /* find extent for this block */ |
107a7bd3 | 4152 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0); |
a86c6181 AT |
4153 | if (IS_ERR(path)) { |
4154 | err = PTR_ERR(path); | |
4155 | path = NULL; | |
4156 | goto out2; | |
4157 | } | |
4158 | ||
4159 | depth = ext_depth(inode); | |
4160 | ||
4161 | /* | |
d0d856e8 RD |
4162 | * consistent leaf must not be empty; |
4163 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
4164 | * this is why assert can't be put in ext4_ext_find_extent() |
4165 | */ | |
273df556 FM |
4166 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4167 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4168 | "lblock: %lu, depth: %d pblock %lld", |
4169 | (unsigned long) map->m_lblk, depth, | |
4170 | path[depth].p_block); | |
034fb4c9 SP |
4171 | err = -EIO; |
4172 | goto out2; | |
4173 | } | |
a86c6181 | 4174 | |
7e028976 AM |
4175 | ex = path[depth].p_ext; |
4176 | if (ex) { | |
725d26d3 | 4177 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4178 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4179 | unsigned short ee_len; |
471d4011 SB |
4180 | |
4181 | /* | |
471d4011 | 4182 | * Uninitialized extents are treated as holes, except that |
56055d3a | 4183 | * we split out initialized portions during a write. |
471d4011 | 4184 | */ |
a2df2a63 | 4185 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4186 | |
4187 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4188 | ||
d0d856e8 | 4189 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4190 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4191 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4192 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4193 | allocated = ee_len - (map->m_lblk - ee_block); |
4194 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4195 | ee_block, ee_len, newblock); | |
56055d3a | 4196 | |
69eb33dc | 4197 | if (!ext4_ext_is_uninitialized(ex)) |
7877191c | 4198 | goto out; |
69eb33dc | 4199 | |
ce37c429 | 4200 | ret = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
4201 | handle, inode, map, path, flags, |
4202 | allocated, newblock); | |
ce37c429 EW |
4203 | if (ret < 0) |
4204 | err = ret; | |
4205 | else | |
4206 | allocated = ret; | |
31cf0f2c | 4207 | goto out2; |
a86c6181 AT |
4208 | } |
4209 | } | |
4210 | ||
7b415bf6 | 4211 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 4212 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
4213 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4214 | ||
a86c6181 | 4215 | /* |
d0d856e8 | 4216 | * requested block isn't allocated yet; |
a86c6181 AT |
4217 | * we couldn't try to create block if create flag is zero |
4218 | */ | |
c2177057 | 4219 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4220 | /* |
4221 | * put just found gap into cache to speed up | |
4222 | * subsequent requests | |
4223 | */ | |
d100eef2 ZL |
4224 | if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0) |
4225 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); | |
a86c6181 AT |
4226 | goto out2; |
4227 | } | |
4d33b1ef | 4228 | |
a86c6181 | 4229 | /* |
c2ea3fde | 4230 | * Okay, we need to do block allocation. |
63f57933 | 4231 | */ |
7b415bf6 | 4232 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef | 4233 | newex.ee_block = cpu_to_le32(map->m_lblk); |
d0abafac | 4234 | cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4235 | |
4236 | /* | |
4237 | * If we are doing bigalloc, check to see if the extent returned | |
4238 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4239 | */ | |
4240 | if (cluster_offset && ex && | |
d8990240 | 4241 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4242 | ar.len = allocated = map->m_len; |
4243 | newblock = map->m_pblk; | |
7b415bf6 | 4244 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4245 | goto got_allocated_blocks; |
4246 | } | |
a86c6181 | 4247 | |
c9de560d | 4248 | /* find neighbour allocated blocks */ |
e35fd660 | 4249 | ar.lleft = map->m_lblk; |
c9de560d AT |
4250 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4251 | if (err) | |
4252 | goto out2; | |
e35fd660 | 4253 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4254 | ex2 = NULL; |
4255 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4256 | if (err) |
4257 | goto out2; | |
25d14f98 | 4258 | |
4d33b1ef TT |
4259 | /* Check if the extent after searching to the right implies a |
4260 | * cluster we can use. */ | |
4261 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4262 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4263 | ar.len = allocated = map->m_len; |
4264 | newblock = map->m_pblk; | |
7b415bf6 | 4265 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4266 | goto got_allocated_blocks; |
4267 | } | |
4268 | ||
749269fa AA |
4269 | /* |
4270 | * See if request is beyond maximum number of blocks we can have in | |
4271 | * a single extent. For an initialized extent this limit is | |
4272 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4273 | * EXT_UNINIT_MAX_LEN. | |
4274 | */ | |
e35fd660 | 4275 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4276 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4277 | map->m_len = EXT_INIT_MAX_LEN; |
4278 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4279 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4280 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4281 | |
e35fd660 | 4282 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4283 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4284 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4285 | if (err) |
b939e376 | 4286 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4287 | else |
e35fd660 | 4288 | allocated = map->m_len; |
c9de560d AT |
4289 | |
4290 | /* allocate new block */ | |
4291 | ar.inode = inode; | |
e35fd660 TT |
4292 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4293 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4294 | /* |
4295 | * We calculate the offset from the beginning of the cluster | |
4296 | * for the logical block number, since when we allocate a | |
4297 | * physical cluster, the physical block should start at the | |
4298 | * same offset from the beginning of the cluster. This is | |
4299 | * needed so that future calls to get_implied_cluster_alloc() | |
4300 | * work correctly. | |
4301 | */ | |
f5a44db5 | 4302 | offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4303 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); |
4304 | ar.goal -= offset; | |
4305 | ar.logical -= offset; | |
c9de560d AT |
4306 | if (S_ISREG(inode->i_mode)) |
4307 | ar.flags = EXT4_MB_HINT_DATA; | |
4308 | else | |
4309 | /* disable in-core preallocation for non-regular files */ | |
4310 | ar.flags = 0; | |
556b27ab VH |
4311 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4312 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4313 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4314 | if (!newblock) |
4315 | goto out2; | |
84fe3bef | 4316 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4317 | ar.goal, newblock, allocated); |
4d33b1ef | 4318 | free_on_err = 1; |
7b415bf6 | 4319 | allocated_clusters = ar.len; |
4d33b1ef TT |
4320 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4321 | if (ar.len > allocated) | |
4322 | ar.len = allocated; | |
a86c6181 | 4323 | |
4d33b1ef | 4324 | got_allocated_blocks: |
a86c6181 | 4325 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4326 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4327 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4328 | /* Mark uninitialized */ |
4329 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4330 | ext4_ext_mark_uninitialized(&newex); |
a25a4e1a | 4331 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4332 | /* |
744692dc | 4333 | * io_end structure was created for every IO write to an |
25985edc | 4334 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4335 | * here we flag the IO that really needs the conversion. |
5f524950 | 4336 | * For non asycn direct IO case, flag the inode state |
25985edc | 4337 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4338 | */ |
82e54229 DM |
4339 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4340 | set_unwritten = 1; | |
744692dc | 4341 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4342 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4343 | } |
c8d46e41 | 4344 | |
a4e5d88b DM |
4345 | err = 0; |
4346 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4347 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4348 | path, ar.len); | |
575a1d4b JZ |
4349 | if (!err) |
4350 | err = ext4_ext_insert_extent(handle, inode, path, | |
4351 | &newex, flags); | |
82e54229 DM |
4352 | |
4353 | if (!err && set_unwritten) { | |
4354 | if (io) | |
4355 | ext4_set_io_unwritten_flag(inode, io); | |
4356 | else | |
4357 | ext4_set_inode_state(inode, | |
4358 | EXT4_STATE_DIO_UNWRITTEN); | |
4359 | } | |
4360 | ||
4d33b1ef | 4361 | if (err && free_on_err) { |
7132de74 MP |
4362 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4363 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4364 | /* free data blocks we just allocated */ |
c9de560d AT |
4365 | /* not a good idea to call discard here directly, |
4366 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4367 | ext4_discard_preallocations(inode); |
c8e15130 TT |
4368 | ext4_free_blocks(handle, inode, NULL, newblock, |
4369 | EXT4_C2B(sbi, allocated_clusters), fb_flags); | |
a86c6181 | 4370 | goto out2; |
315054f0 | 4371 | } |
a86c6181 | 4372 | |
a86c6181 | 4373 | /* previous routine could use block we allocated */ |
bf89d16f | 4374 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4375 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4376 | if (allocated > map->m_len) |
4377 | allocated = map->m_len; | |
4378 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4379 | |
5f634d06 AK |
4380 | /* |
4381 | * Update reserved blocks/metadata blocks after successful | |
4382 | * block allocation which had been deferred till now. | |
4383 | */ | |
7b415bf6 | 4384 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4385 | unsigned int reserved_clusters; |
7b415bf6 | 4386 | /* |
81fdbb4a | 4387 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4388 | */ |
4389 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4390 | map->m_lblk, allocated); | |
4391 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4392 | if (reserved_clusters) { | |
4393 | /* | |
4394 | * We have clusters reserved for this range. | |
4395 | * But since we are not doing actual allocation | |
4396 | * and are simply using blocks from previously | |
4397 | * allocated cluster, we should release the | |
4398 | * reservation and not claim quota. | |
4399 | */ | |
4400 | ext4_da_update_reserve_space(inode, | |
4401 | reserved_clusters, 0); | |
4402 | } | |
4403 | } else { | |
4404 | BUG_ON(allocated_clusters < reserved_clusters); | |
7b415bf6 | 4405 | if (reserved_clusters < allocated_clusters) { |
5356f261 | 4406 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4407 | int reservation = allocated_clusters - |
4408 | reserved_clusters; | |
4409 | /* | |
4410 | * It seems we claimed few clusters outside of | |
4411 | * the range of this allocation. We should give | |
4412 | * it back to the reservation pool. This can | |
4413 | * happen in the following case: | |
4414 | * | |
4415 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4416 | * cluster has 4 blocks. Thus, the clusters | |
4417 | * are [0-3],[4-7],[8-11]... | |
4418 | * * First comes delayed allocation write for | |
4419 | * logical blocks 10 & 11. Since there were no | |
4420 | * previous delayed allocated blocks in the | |
4421 | * range [8-11], we would reserve 1 cluster | |
4422 | * for this write. | |
4423 | * * Next comes write for logical blocks 3 to 8. | |
4424 | * In this case, we will reserve 2 clusters | |
4425 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4426 | * that range has a delayed allocated blocks. | |
4427 | * Thus total reserved clusters now becomes 3. | |
4428 | * * Now, during the delayed allocation writeout | |
4429 | * time, we will first write blocks [3-8] and | |
4430 | * allocate 3 clusters for writing these | |
4431 | * blocks. Also, we would claim all these | |
4432 | * three clusters above. | |
4433 | * * Now when we come here to writeout the | |
4434 | * blocks [10-11], we would expect to claim | |
4435 | * the reservation of 1 cluster we had made | |
4436 | * (and we would claim it since there are no | |
4437 | * more delayed allocated blocks in the range | |
4438 | * [8-11]. But our reserved cluster count had | |
4439 | * already gone to 0. | |
4440 | * | |
4441 | * Thus, at the step 4 above when we determine | |
4442 | * that there are still some unwritten delayed | |
4443 | * allocated blocks outside of our current | |
4444 | * block range, we should increment the | |
4445 | * reserved clusters count so that when the | |
4446 | * remaining blocks finally gets written, we | |
4447 | * could claim them. | |
4448 | */ | |
5356f261 AK |
4449 | dquot_reserve_block(inode, |
4450 | EXT4_C2B(sbi, reservation)); | |
4451 | spin_lock(&ei->i_block_reservation_lock); | |
4452 | ei->i_reserved_data_blocks += reservation; | |
4453 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 | 4454 | } |
232ec872 LC |
4455 | /* |
4456 | * We will claim quota for all newly allocated blocks. | |
4457 | * We're updating the reserved space *after* the | |
4458 | * correction above so we do not accidentally free | |
4459 | * all the metadata reservation because we might | |
4460 | * actually need it later on. | |
4461 | */ | |
4462 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4463 | 1); | |
7b415bf6 AK |
4464 | } |
4465 | } | |
5f634d06 | 4466 | |
b436b9be JK |
4467 | /* |
4468 | * Cache the extent and update transaction to commit on fdatasync only | |
4469 | * when it is _not_ an uninitialized extent. | |
4470 | */ | |
69eb33dc | 4471 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) |
b436b9be | 4472 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4473 | else |
b436b9be | 4474 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4475 | out: |
e35fd660 TT |
4476 | if (allocated > map->m_len) |
4477 | allocated = map->m_len; | |
a86c6181 | 4478 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4479 | map->m_flags |= EXT4_MAP_MAPPED; |
4480 | map->m_pblk = newblock; | |
4481 | map->m_len = allocated; | |
a86c6181 AT |
4482 | out2: |
4483 | if (path) { | |
4484 | ext4_ext_drop_refs(path); | |
4485 | kfree(path); | |
4486 | } | |
e861304b | 4487 | |
63b99968 TT |
4488 | trace_ext4_ext_map_blocks_exit(inode, flags, map, |
4489 | err ? err : allocated); | |
4490 | ext4_es_lru_add(inode); | |
7877191c | 4491 | return err ? err : allocated; |
a86c6181 AT |
4492 | } |
4493 | ||
819c4920 | 4494 | void ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4495 | { |
a86c6181 | 4496 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4497 | ext4_lblk_t last_block; |
a86c6181 AT |
4498 | int err = 0; |
4499 | ||
a86c6181 | 4500 | /* |
d0d856e8 RD |
4501 | * TODO: optimization is possible here. |
4502 | * Probably we need not scan at all, | |
4503 | * because page truncation is enough. | |
a86c6181 | 4504 | */ |
a86c6181 AT |
4505 | |
4506 | /* we have to know where to truncate from in crash case */ | |
4507 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4508 | ext4_mark_inode_dirty(handle, inode); | |
4509 | ||
4510 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4511 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
8acd5e9b | 4512 | retry: |
51865fda ZL |
4513 | err = ext4_es_remove_extent(inode, last_block, |
4514 | EXT_MAX_BLOCKS - last_block); | |
94eec0fc | 4515 | if (err == -ENOMEM) { |
8acd5e9b TT |
4516 | cond_resched(); |
4517 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
4518 | goto retry; | |
4519 | } | |
4520 | if (err) { | |
4521 | ext4_std_error(inode->i_sb, err); | |
4522 | return; | |
4523 | } | |
5f95d21f | 4524 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
8acd5e9b | 4525 | ext4_std_error(inode->i_sb, err); |
a86c6181 AT |
4526 | } |
4527 | ||
fd28784a AK |
4528 | static void ext4_falloc_update_inode(struct inode *inode, |
4529 | int mode, loff_t new_size, int update_ctime) | |
4530 | { | |
4531 | struct timespec now; | |
4532 | ||
4533 | if (update_ctime) { | |
4534 | now = current_fs_time(inode->i_sb); | |
4535 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4536 | inode->i_ctime = now; | |
4537 | } | |
4538 | /* | |
4539 | * Update only when preallocation was requested beyond | |
4540 | * the file size. | |
4541 | */ | |
cf17fea6 AK |
4542 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4543 | if (new_size > i_size_read(inode)) | |
4544 | i_size_write(inode, new_size); | |
4545 | if (new_size > EXT4_I(inode)->i_disksize) | |
4546 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4547 | } else { |
4548 | /* | |
4549 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4550 | * can proceed even if the new size is the same as i_size. | |
4551 | */ | |
4552 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4553 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4554 | } |
4555 | ||
4556 | } | |
4557 | ||
a2df2a63 | 4558 | /* |
2fe17c10 | 4559 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4560 | * operation, which gets called from sys_fallocate system call. |
4561 | * For block-mapped files, posix_fallocate should fall back to the method | |
4562 | * of writing zeroes to the required new blocks (the same behavior which is | |
4563 | * expected for file systems which do not support fallocate() system call). | |
4564 | */ | |
2fe17c10 | 4565 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4566 | { |
496ad9aa | 4567 | struct inode *inode = file_inode(file); |
a2df2a63 | 4568 | handle_t *handle; |
fd28784a | 4569 | loff_t new_size; |
498e5f24 | 4570 | unsigned int max_blocks; |
a2df2a63 AA |
4571 | int ret = 0; |
4572 | int ret2 = 0; | |
4573 | int retries = 0; | |
a4e5d88b | 4574 | int flags; |
2ed88685 | 4575 | struct ext4_map_blocks map; |
a2df2a63 AA |
4576 | unsigned int credits, blkbits = inode->i_blkbits; |
4577 | ||
a4bb6b64 | 4578 | /* Return error if mode is not supported */ |
9eb79482 NJ |
4579 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
4580 | FALLOC_FL_COLLAPSE_RANGE)) | |
a4bb6b64 AH |
4581 | return -EOPNOTSUPP; |
4582 | ||
4583 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
aeb2817a | 4584 | return ext4_punch_hole(inode, offset, len); |
a4bb6b64 | 4585 | |
9eb79482 NJ |
4586 | if (mode & FALLOC_FL_COLLAPSE_RANGE) |
4587 | return ext4_collapse_range(inode, offset, len); | |
4588 | ||
0c8d414f TM |
4589 | ret = ext4_convert_inline_data(inode); |
4590 | if (ret) | |
4591 | return ret; | |
4592 | ||
8bad6fc8 ZL |
4593 | /* |
4594 | * currently supporting (pre)allocate mode for extent-based | |
4595 | * files _only_ | |
4596 | */ | |
4597 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4598 | return -EOPNOTSUPP; | |
4599 | ||
0562e0ba | 4600 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4601 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4602 | /* |
4603 | * We can't just convert len to max_blocks because | |
4604 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4605 | */ | |
a2df2a63 | 4606 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4607 | - map.m_lblk; |
a2df2a63 | 4608 | /* |
f3bd1f3f | 4609 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4610 | */ |
f3bd1f3f | 4611 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4612 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4613 | ret = inode_newsize_ok(inode, (len + offset)); |
4614 | if (ret) { | |
4615 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4616 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4617 | return ret; |
4618 | } | |
3c6fe770 | 4619 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4620 | if (mode & FALLOC_FL_KEEP_SIZE) |
4621 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4622 | /* |
4623 | * Don't normalize the request if it can fit in one extent so | |
4624 | * that it doesn't get unnecessarily split into multiple | |
4625 | * extents. | |
4626 | */ | |
4627 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4628 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f | 4629 | |
a2df2a63 AA |
4630 | retry: |
4631 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4632 | map.m_lblk = map.m_lblk + ret; |
4633 | map.m_len = max_blocks = max_blocks - ret; | |
9924a92a TT |
4634 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4635 | credits); | |
a2df2a63 AA |
4636 | if (IS_ERR(handle)) { |
4637 | ret = PTR_ERR(handle); | |
4638 | break; | |
4639 | } | |
a4e5d88b | 4640 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4641 | if (ret <= 0) { |
2c98615d | 4642 | #ifdef EXT4FS_DEBUG |
b06acd38 LC |
4643 | ext4_warning(inode->i_sb, |
4644 | "inode #%lu: block %u: len %u: " | |
4645 | "ext4_ext_map_blocks returned %d", | |
4646 | inode->i_ino, map.m_lblk, | |
4647 | map.m_len, ret); | |
2c98615d | 4648 | #endif |
a2df2a63 AA |
4649 | ext4_mark_inode_dirty(handle, inode); |
4650 | ret2 = ext4_journal_stop(handle); | |
4651 | break; | |
4652 | } | |
2ed88685 | 4653 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4654 | blkbits) >> blkbits)) |
4655 | new_size = offset + len; | |
4656 | else | |
29ae07b7 | 4657 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4658 | |
fd28784a | 4659 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4660 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4661 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4662 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4663 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4664 | ret2 = ext4_journal_stop(handle); |
4665 | if (ret2) | |
4666 | break; | |
4667 | } | |
fd28784a AK |
4668 | if (ret == -ENOSPC && |
4669 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4670 | ret = 0; | |
a2df2a63 | 4671 | goto retry; |
a2df2a63 | 4672 | } |
55bd725a | 4673 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4674 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4675 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4676 | return ret > 0 ? ret2 : ret; |
4677 | } | |
6873fa0d | 4678 | |
0031462b MC |
4679 | /* |
4680 | * This function convert a range of blocks to written extents | |
4681 | * The caller of this function will pass the start offset and the size. | |
4682 | * all unwritten extents within this range will be converted to | |
4683 | * written extents. | |
4684 | * | |
4685 | * This function is called from the direct IO end io call back | |
4686 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4687 | * Returns 0 on success. |
0031462b | 4688 | */ |
6b523df4 JK |
4689 | int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
4690 | loff_t offset, ssize_t len) | |
0031462b | 4691 | { |
0031462b MC |
4692 | unsigned int max_blocks; |
4693 | int ret = 0; | |
4694 | int ret2 = 0; | |
2ed88685 | 4695 | struct ext4_map_blocks map; |
0031462b MC |
4696 | unsigned int credits, blkbits = inode->i_blkbits; |
4697 | ||
2ed88685 | 4698 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4699 | /* |
4700 | * We can't just convert len to max_blocks because | |
4701 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4702 | */ | |
2ed88685 TT |
4703 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4704 | map.m_lblk); | |
0031462b | 4705 | /* |
6b523df4 JK |
4706 | * This is somewhat ugly but the idea is clear: When transaction is |
4707 | * reserved, everything goes into it. Otherwise we rather start several | |
4708 | * smaller transactions for conversion of each extent separately. | |
0031462b | 4709 | */ |
6b523df4 JK |
4710 | if (handle) { |
4711 | handle = ext4_journal_start_reserved(handle, | |
4712 | EXT4_HT_EXT_CONVERT); | |
4713 | if (IS_ERR(handle)) | |
4714 | return PTR_ERR(handle); | |
4715 | credits = 0; | |
4716 | } else { | |
4717 | /* | |
4718 | * credits to insert 1 extent into extent tree | |
4719 | */ | |
4720 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4721 | } | |
0031462b | 4722 | while (ret >= 0 && ret < max_blocks) { |
2ed88685 TT |
4723 | map.m_lblk += ret; |
4724 | map.m_len = (max_blocks -= ret); | |
6b523df4 JK |
4725 | if (credits) { |
4726 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
4727 | credits); | |
4728 | if (IS_ERR(handle)) { | |
4729 | ret = PTR_ERR(handle); | |
4730 | break; | |
4731 | } | |
0031462b | 4732 | } |
2ed88685 | 4733 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4734 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4735 | if (ret <= 0) |
4736 | ext4_warning(inode->i_sb, | |
4737 | "inode #%lu: block %u: len %u: " | |
4738 | "ext4_ext_map_blocks returned %d", | |
4739 | inode->i_ino, map.m_lblk, | |
4740 | map.m_len, ret); | |
0031462b | 4741 | ext4_mark_inode_dirty(handle, inode); |
6b523df4 JK |
4742 | if (credits) |
4743 | ret2 = ext4_journal_stop(handle); | |
4744 | if (ret <= 0 || ret2) | |
0031462b MC |
4745 | break; |
4746 | } | |
6b523df4 JK |
4747 | if (!credits) |
4748 | ret2 = ext4_journal_stop(handle); | |
0031462b MC |
4749 | return ret > 0 ? ret2 : ret; |
4750 | } | |
6d9c85eb | 4751 | |
6873fa0d | 4752 | /* |
69eb33dc ZL |
4753 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
4754 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
4755 | * return start of the next delayed extent. |
4756 | * | |
69eb33dc | 4757 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 4758 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 4759 | * extent found. Leave newes unmodified. |
6873fa0d | 4760 | */ |
91dd8c11 | 4761 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 4762 | struct extent_status *newes) |
6873fa0d | 4763 | { |
b3aff3e3 | 4764 | struct extent_status es; |
be401363 | 4765 | ext4_lblk_t block, next_del; |
6873fa0d | 4766 | |
69eb33dc | 4767 | if (newes->es_pblk == 0) { |
e30b5dca YZ |
4768 | ext4_es_find_delayed_extent_range(inode, newes->es_lblk, |
4769 | newes->es_lblk + newes->es_len - 1, &es); | |
4770 | ||
6d9c85eb | 4771 | /* |
69eb33dc | 4772 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 4773 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 4774 | */ |
06b0c886 | 4775 | if (es.es_len == 0) |
b3aff3e3 | 4776 | /* A hole found. */ |
91dd8c11 | 4777 | return 0; |
b3aff3e3 | 4778 | |
69eb33dc | 4779 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 4780 | /* A hole found. */ |
69eb33dc ZL |
4781 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
4782 | newes->es_len); | |
91dd8c11 | 4783 | return 0; |
6873fa0d | 4784 | } |
6d9c85eb | 4785 | |
69eb33dc | 4786 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
4787 | } |
4788 | ||
69eb33dc | 4789 | block = newes->es_lblk + newes->es_len; |
e30b5dca | 4790 | ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es); |
be401363 ZL |
4791 | if (es.es_len == 0) |
4792 | next_del = EXT_MAX_BLOCKS; | |
4793 | else | |
4794 | next_del = es.es_lblk; | |
4795 | ||
91dd8c11 | 4796 | return next_del; |
6873fa0d | 4797 | } |
6873fa0d ES |
4798 | /* fiemap flags we can handle specified here */ |
4799 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4800 | ||
3a06d778 AK |
4801 | static int ext4_xattr_fiemap(struct inode *inode, |
4802 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4803 | { |
4804 | __u64 physical = 0; | |
4805 | __u64 length; | |
4806 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4807 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4808 | int error = 0; | |
4809 | ||
4810 | /* in-inode? */ | |
19f5fb7a | 4811 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4812 | struct ext4_iloc iloc; |
4813 | int offset; /* offset of xattr in inode */ | |
4814 | ||
4815 | error = ext4_get_inode_loc(inode, &iloc); | |
4816 | if (error) | |
4817 | return error; | |
a60697f4 | 4818 | physical = (__u64)iloc.bh->b_blocknr << blockbits; |
6873fa0d ES |
4819 | offset = EXT4_GOOD_OLD_INODE_SIZE + |
4820 | EXT4_I(inode)->i_extra_isize; | |
4821 | physical += offset; | |
4822 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4823 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4824 | brelse(iloc.bh); |
6873fa0d | 4825 | } else { /* external block */ |
a60697f4 | 4826 | physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
6873fa0d ES |
4827 | length = inode->i_sb->s_blocksize; |
4828 | } | |
4829 | ||
4830 | if (physical) | |
4831 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4832 | length, flags); | |
4833 | return (error < 0 ? error : 0); | |
4834 | } | |
4835 | ||
4836 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
4837 | __u64 start, __u64 len) | |
4838 | { | |
4839 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4840 | int error = 0; |
4841 | ||
94191985 TM |
4842 | if (ext4_has_inline_data(inode)) { |
4843 | int has_inline = 1; | |
4844 | ||
4845 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline); | |
4846 | ||
4847 | if (has_inline) | |
4848 | return error; | |
4849 | } | |
4850 | ||
7869a4a6 TT |
4851 | if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
4852 | error = ext4_ext_precache(inode); | |
4853 | if (error) | |
4854 | return error; | |
4855 | } | |
4856 | ||
6873fa0d | 4857 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 4858 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4859 | return generic_block_fiemap(inode, fieinfo, start, len, |
4860 | ext4_get_block); | |
4861 | ||
4862 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4863 | return -EBADR; | |
4864 | ||
4865 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4866 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4867 | } else { | |
aca92ff6 LM |
4868 | ext4_lblk_t len_blks; |
4869 | __u64 last_blk; | |
4870 | ||
6873fa0d | 4871 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4872 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4873 | if (last_blk >= EXT_MAX_BLOCKS) |
4874 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4875 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4876 | |
4877 | /* | |
91dd8c11 LC |
4878 | * Walk the extent tree gathering extent information |
4879 | * and pushing extents back to the user. | |
6873fa0d | 4880 | */ |
91dd8c11 LC |
4881 | error = ext4_fill_fiemap_extents(inode, start_blk, |
4882 | len_blks, fieinfo); | |
6873fa0d | 4883 | } |
107a7bd3 | 4884 | ext4_es_lru_add(inode); |
6873fa0d ES |
4885 | return error; |
4886 | } | |
9eb79482 NJ |
4887 | |
4888 | /* | |
4889 | * ext4_access_path: | |
4890 | * Function to access the path buffer for marking it dirty. | |
4891 | * It also checks if there are sufficient credits left in the journal handle | |
4892 | * to update path. | |
4893 | */ | |
4894 | static int | |
4895 | ext4_access_path(handle_t *handle, struct inode *inode, | |
4896 | struct ext4_ext_path *path) | |
4897 | { | |
4898 | int credits, err; | |
4899 | ||
4900 | if (!ext4_handle_valid(handle)) | |
4901 | return 0; | |
4902 | ||
4903 | /* | |
4904 | * Check if need to extend journal credits | |
4905 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
4906 | * descriptor) for each block group; assume two block | |
4907 | * groups | |
4908 | */ | |
4909 | if (handle->h_buffer_credits < 7) { | |
4910 | credits = ext4_writepage_trans_blocks(inode); | |
4911 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); | |
4912 | /* EAGAIN is success */ | |
4913 | if (err && err != -EAGAIN) | |
4914 | return err; | |
4915 | } | |
4916 | ||
4917 | err = ext4_ext_get_access(handle, inode, path); | |
4918 | return err; | |
4919 | } | |
4920 | ||
4921 | /* | |
4922 | * ext4_ext_shift_path_extents: | |
4923 | * Shift the extents of a path structure lying between path[depth].p_ext | |
4924 | * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift | |
4925 | * from starting block for each extent. | |
4926 | */ | |
4927 | static int | |
4928 | ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift, | |
4929 | struct inode *inode, handle_t *handle, | |
4930 | ext4_lblk_t *start) | |
4931 | { | |
4932 | int depth, err = 0; | |
4933 | struct ext4_extent *ex_start, *ex_last; | |
4934 | bool update = 0; | |
4935 | depth = path->p_depth; | |
4936 | ||
4937 | while (depth >= 0) { | |
4938 | if (depth == path->p_depth) { | |
4939 | ex_start = path[depth].p_ext; | |
4940 | if (!ex_start) | |
4941 | return -EIO; | |
4942 | ||
4943 | ex_last = EXT_LAST_EXTENT(path[depth].p_hdr); | |
4944 | if (!ex_last) | |
4945 | return -EIO; | |
4946 | ||
4947 | err = ext4_access_path(handle, inode, path + depth); | |
4948 | if (err) | |
4949 | goto out; | |
4950 | ||
4951 | if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
4952 | update = 1; | |
4953 | ||
4954 | *start = ex_last->ee_block + | |
4955 | ext4_ext_get_actual_len(ex_last); | |
4956 | ||
4957 | while (ex_start <= ex_last) { | |
4958 | ex_start->ee_block -= shift; | |
4959 | if (ex_start > | |
4960 | EXT_FIRST_EXTENT(path[depth].p_hdr)) { | |
4961 | if (ext4_ext_try_to_merge_right(inode, | |
4962 | path, ex_start - 1)) | |
4963 | ex_last--; | |
4964 | } | |
4965 | ex_start++; | |
4966 | } | |
4967 | err = ext4_ext_dirty(handle, inode, path + depth); | |
4968 | if (err) | |
4969 | goto out; | |
4970 | ||
4971 | if (--depth < 0 || !update) | |
4972 | break; | |
4973 | } | |
4974 | ||
4975 | /* Update index too */ | |
4976 | err = ext4_access_path(handle, inode, path + depth); | |
4977 | if (err) | |
4978 | goto out; | |
4979 | ||
4980 | path[depth].p_idx->ei_block -= shift; | |
4981 | err = ext4_ext_dirty(handle, inode, path + depth); | |
4982 | if (err) | |
4983 | goto out; | |
4984 | ||
4985 | /* we are done if current index is not a starting index */ | |
4986 | if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr)) | |
4987 | break; | |
4988 | ||
4989 | depth--; | |
4990 | } | |
4991 | ||
4992 | out: | |
4993 | return err; | |
4994 | } | |
4995 | ||
4996 | /* | |
4997 | * ext4_ext_shift_extents: | |
4998 | * All the extents which lies in the range from start to the last allocated | |
4999 | * block for the file are shifted downwards by shift blocks. | |
5000 | * On success, 0 is returned, error otherwise. | |
5001 | */ | |
5002 | static int | |
5003 | ext4_ext_shift_extents(struct inode *inode, handle_t *handle, | |
5004 | ext4_lblk_t start, ext4_lblk_t shift) | |
5005 | { | |
5006 | struct ext4_ext_path *path; | |
5007 | int ret = 0, depth; | |
5008 | struct ext4_extent *extent; | |
5009 | ext4_lblk_t stop_block, current_block; | |
5010 | ext4_lblk_t ex_start, ex_end; | |
5011 | ||
5012 | /* Let path point to the last extent */ | |
5013 | path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0); | |
5014 | if (IS_ERR(path)) | |
5015 | return PTR_ERR(path); | |
5016 | ||
5017 | depth = path->p_depth; | |
5018 | extent = path[depth].p_ext; | |
5019 | if (!extent) { | |
5020 | ext4_ext_drop_refs(path); | |
5021 | kfree(path); | |
5022 | return ret; | |
5023 | } | |
5024 | ||
5025 | stop_block = extent->ee_block + ext4_ext_get_actual_len(extent); | |
5026 | ext4_ext_drop_refs(path); | |
5027 | kfree(path); | |
5028 | ||
5029 | /* Nothing to shift, if hole is at the end of file */ | |
5030 | if (start >= stop_block) | |
5031 | return ret; | |
5032 | ||
5033 | /* | |
5034 | * Don't start shifting extents until we make sure the hole is big | |
5035 | * enough to accomodate the shift. | |
5036 | */ | |
5037 | path = ext4_ext_find_extent(inode, start - 1, NULL, 0); | |
5038 | depth = path->p_depth; | |
5039 | extent = path[depth].p_ext; | |
5040 | ex_start = extent->ee_block; | |
5041 | ex_end = extent->ee_block + ext4_ext_get_actual_len(extent); | |
5042 | ext4_ext_drop_refs(path); | |
5043 | kfree(path); | |
5044 | ||
5045 | if ((start == ex_start && shift > ex_start) || | |
5046 | (shift > start - ex_end)) | |
5047 | return -EINVAL; | |
5048 | ||
5049 | /* Its safe to start updating extents */ | |
5050 | while (start < stop_block) { | |
5051 | path = ext4_ext_find_extent(inode, start, NULL, 0); | |
5052 | if (IS_ERR(path)) | |
5053 | return PTR_ERR(path); | |
5054 | depth = path->p_depth; | |
5055 | extent = path[depth].p_ext; | |
5056 | current_block = extent->ee_block; | |
5057 | if (start > current_block) { | |
5058 | /* Hole, move to the next extent */ | |
5059 | ret = mext_next_extent(inode, path, &extent); | |
5060 | if (ret != 0) { | |
5061 | ext4_ext_drop_refs(path); | |
5062 | kfree(path); | |
5063 | if (ret == 1) | |
5064 | ret = 0; | |
5065 | break; | |
5066 | } | |
5067 | } | |
5068 | ret = ext4_ext_shift_path_extents(path, shift, inode, | |
5069 | handle, &start); | |
5070 | ext4_ext_drop_refs(path); | |
5071 | kfree(path); | |
5072 | if (ret) | |
5073 | break; | |
5074 | } | |
5075 | ||
5076 | return ret; | |
5077 | } | |
5078 | ||
5079 | /* | |
5080 | * ext4_collapse_range: | |
5081 | * This implements the fallocate's collapse range functionality for ext4 | |
5082 | * Returns: 0 and non-zero on error. | |
5083 | */ | |
5084 | int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len) | |
5085 | { | |
5086 | struct super_block *sb = inode->i_sb; | |
5087 | ext4_lblk_t punch_start, punch_stop; | |
5088 | handle_t *handle; | |
5089 | unsigned int credits; | |
5090 | loff_t new_size; | |
5091 | int ret; | |
5092 | ||
5093 | BUG_ON(offset + len > i_size_read(inode)); | |
5094 | ||
5095 | /* Collapse range works only on fs block size aligned offsets. */ | |
5096 | if (offset & (EXT4_BLOCK_SIZE(sb) - 1) || | |
5097 | len & (EXT4_BLOCK_SIZE(sb) - 1)) | |
5098 | return -EINVAL; | |
5099 | ||
5100 | if (!S_ISREG(inode->i_mode)) | |
5101 | return -EOPNOTSUPP; | |
5102 | ||
5103 | trace_ext4_collapse_range(inode, offset, len); | |
5104 | ||
5105 | punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5106 | punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb); | |
5107 | ||
5108 | /* Write out all dirty pages */ | |
5109 | ret = filemap_write_and_wait_range(inode->i_mapping, offset, -1); | |
5110 | if (ret) | |
5111 | return ret; | |
5112 | ||
5113 | /* Take mutex lock */ | |
5114 | mutex_lock(&inode->i_mutex); | |
5115 | ||
5116 | /* It's not possible punch hole on append only file */ | |
5117 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
5118 | ret = -EPERM; | |
5119 | goto out_mutex; | |
5120 | } | |
5121 | ||
5122 | if (IS_SWAPFILE(inode)) { | |
5123 | ret = -ETXTBSY; | |
5124 | goto out_mutex; | |
5125 | } | |
5126 | ||
5127 | /* Currently just for extent based files */ | |
5128 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5129 | ret = -EOPNOTSUPP; | |
5130 | goto out_mutex; | |
5131 | } | |
5132 | ||
5133 | truncate_pagecache_range(inode, offset, -1); | |
5134 | ||
5135 | /* Wait for existing dio to complete */ | |
5136 | ext4_inode_block_unlocked_dio(inode); | |
5137 | inode_dio_wait(inode); | |
5138 | ||
5139 | credits = ext4_writepage_trans_blocks(inode); | |
5140 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5141 | if (IS_ERR(handle)) { | |
5142 | ret = PTR_ERR(handle); | |
5143 | goto out_dio; | |
5144 | } | |
5145 | ||
5146 | down_write(&EXT4_I(inode)->i_data_sem); | |
5147 | ext4_discard_preallocations(inode); | |
5148 | ||
5149 | ret = ext4_es_remove_extent(inode, punch_start, | |
5150 | EXT_MAX_BLOCKS - punch_start - 1); | |
5151 | if (ret) { | |
5152 | up_write(&EXT4_I(inode)->i_data_sem); | |
5153 | goto out_stop; | |
5154 | } | |
5155 | ||
5156 | ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1); | |
5157 | if (ret) { | |
5158 | up_write(&EXT4_I(inode)->i_data_sem); | |
5159 | goto out_stop; | |
5160 | } | |
5161 | ||
5162 | ret = ext4_ext_shift_extents(inode, handle, punch_stop, | |
5163 | punch_stop - punch_start); | |
5164 | if (ret) { | |
5165 | up_write(&EXT4_I(inode)->i_data_sem); | |
5166 | goto out_stop; | |
5167 | } | |
5168 | ||
5169 | new_size = i_size_read(inode) - len; | |
5170 | truncate_setsize(inode, new_size); | |
5171 | EXT4_I(inode)->i_disksize = new_size; | |
5172 | ||
5173 | ext4_discard_preallocations(inode); | |
5174 | up_write(&EXT4_I(inode)->i_data_sem); | |
5175 | if (IS_SYNC(inode)) | |
5176 | ext4_handle_sync(handle); | |
5177 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
5178 | ext4_mark_inode_dirty(handle, inode); | |
5179 | ||
5180 | out_stop: | |
5181 | ext4_journal_stop(handle); | |
5182 | out_dio: | |
5183 | ext4_inode_resume_unlocked_dio(inode); | |
5184 | out_mutex: | |
5185 | mutex_unlock(&inode->i_mutex); | |
5186 | return ret; | |
5187 | } |