Commit | Line | Data |
---|---|---|
a86c6181 AT |
1 | /* |
2 | * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com | |
3 | * Written by Alex Tomas <alex@clusterfs.com> | |
4 | * | |
5 | * Architecture independence: | |
6 | * Copyright (c) 2005, Bull S.A. | |
7 | * Written by Pierre Peiffer <pierre.peiffer@bull.net> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public Licens | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Extents support for EXT4 | |
25 | * | |
26 | * TODO: | |
27 | * - ext4*_error() should be used in some situations | |
28 | * - analyze all BUG()/BUG_ON(), use -EIO where appropriate | |
29 | * - smart tree reduction | |
30 | */ | |
31 | ||
a86c6181 AT |
32 | #include <linux/fs.h> |
33 | #include <linux/time.h> | |
cd02ff0b | 34 | #include <linux/jbd2.h> |
a86c6181 AT |
35 | #include <linux/highuid.h> |
36 | #include <linux/pagemap.h> | |
37 | #include <linux/quotaops.h> | |
38 | #include <linux/string.h> | |
39 | #include <linux/slab.h> | |
a2df2a63 | 40 | #include <linux/falloc.h> |
a86c6181 | 41 | #include <asm/uaccess.h> |
6873fa0d | 42 | #include <linux/fiemap.h> |
3dcf5451 | 43 | #include "ext4_jbd2.h" |
4a092d73 | 44 | #include "ext4_extents.h" |
f19d5870 | 45 | #include "xattr.h" |
a86c6181 | 46 | |
0562e0ba JZ |
47 | #include <trace/events/ext4.h> |
48 | ||
5f95d21f LC |
49 | /* |
50 | * used by extent splitting. | |
51 | */ | |
52 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
53 | due to ENOSPC */ | |
54 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
55 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
56 | ||
dee1f973 DM |
57 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
58 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
59 | ||
7ac5990d DW |
60 | static __le32 ext4_extent_block_csum(struct inode *inode, |
61 | struct ext4_extent_header *eh) | |
62 | { | |
63 | struct ext4_inode_info *ei = EXT4_I(inode); | |
64 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
65 | __u32 csum; | |
66 | ||
67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
68 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
69 | return cpu_to_le32(csum); | |
70 | } | |
71 | ||
72 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
73 | struct ext4_extent_header *eh) | |
74 | { | |
75 | struct ext4_extent_tail *et; | |
76 | ||
77 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
78 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
79 | return 1; | |
80 | ||
81 | et = find_ext4_extent_tail(eh); | |
82 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
83 | return 0; | |
84 | return 1; | |
85 | } | |
86 | ||
87 | static void ext4_extent_block_csum_set(struct inode *inode, | |
88 | struct ext4_extent_header *eh) | |
89 | { | |
90 | struct ext4_extent_tail *et; | |
91 | ||
92 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
93 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
94 | return; | |
95 | ||
96 | et = find_ext4_extent_tail(eh); | |
97 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
98 | } | |
99 | ||
d583fb87 AH |
100 | static int ext4_split_extent(handle_t *handle, |
101 | struct inode *inode, | |
102 | struct ext4_ext_path *path, | |
103 | struct ext4_map_blocks *map, | |
104 | int split_flag, | |
105 | int flags); | |
106 | ||
5f95d21f LC |
107 | static int ext4_split_extent_at(handle_t *handle, |
108 | struct inode *inode, | |
109 | struct ext4_ext_path *path, | |
110 | ext4_lblk_t split, | |
111 | int split_flag, | |
112 | int flags); | |
113 | ||
91dd8c11 | 114 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 115 | struct extent_status *newes); |
91dd8c11 | 116 | |
487caeef JK |
117 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
118 | struct inode *inode, | |
119 | int needed) | |
a86c6181 AT |
120 | { |
121 | int err; | |
122 | ||
0390131b FM |
123 | if (!ext4_handle_valid(handle)) |
124 | return 0; | |
a86c6181 | 125 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
126 | return 0; |
127 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 128 | if (err <= 0) |
9102e4fa | 129 | return err; |
487caeef | 130 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
131 | if (err == 0) |
132 | err = -EAGAIN; | |
487caeef JK |
133 | |
134 | return err; | |
a86c6181 AT |
135 | } |
136 | ||
137 | /* | |
138 | * could return: | |
139 | * - EROFS | |
140 | * - ENOMEM | |
141 | */ | |
142 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
143 | struct ext4_ext_path *path) | |
144 | { | |
145 | if (path->p_bh) { | |
146 | /* path points to block */ | |
147 | return ext4_journal_get_write_access(handle, path->p_bh); | |
148 | } | |
149 | /* path points to leaf/index in inode body */ | |
150 | /* we use in-core data, no need to protect them */ | |
151 | return 0; | |
152 | } | |
153 | ||
154 | /* | |
155 | * could return: | |
156 | * - EROFS | |
157 | * - ENOMEM | |
158 | * - EIO | |
159 | */ | |
9ea7a0df TT |
160 | #define ext4_ext_dirty(handle, inode, path) \ |
161 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
162 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
163 | handle_t *handle, struct inode *inode, | |
164 | struct ext4_ext_path *path) | |
a86c6181 AT |
165 | { |
166 | int err; | |
167 | if (path->p_bh) { | |
7ac5990d | 168 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 169 | /* path points to block */ |
9ea7a0df TT |
170 | err = __ext4_handle_dirty_metadata(where, line, handle, |
171 | inode, path->p_bh); | |
a86c6181 AT |
172 | } else { |
173 | /* path points to leaf/index in inode body */ | |
174 | err = ext4_mark_inode_dirty(handle, inode); | |
175 | } | |
176 | return err; | |
177 | } | |
178 | ||
f65e6fba | 179 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 180 | struct ext4_ext_path *path, |
725d26d3 | 181 | ext4_lblk_t block) |
a86c6181 | 182 | { |
a86c6181 | 183 | if (path) { |
81fdbb4a | 184 | int depth = path->p_depth; |
a86c6181 | 185 | struct ext4_extent *ex; |
a86c6181 | 186 | |
ad4fb9ca KM |
187 | /* |
188 | * Try to predict block placement assuming that we are | |
189 | * filling in a file which will eventually be | |
190 | * non-sparse --- i.e., in the case of libbfd writing | |
191 | * an ELF object sections out-of-order but in a way | |
192 | * the eventually results in a contiguous object or | |
193 | * executable file, or some database extending a table | |
194 | * space file. However, this is actually somewhat | |
195 | * non-ideal if we are writing a sparse file such as | |
196 | * qemu or KVM writing a raw image file that is going | |
197 | * to stay fairly sparse, since it will end up | |
198 | * fragmenting the file system's free space. Maybe we | |
199 | * should have some hueristics or some way to allow | |
200 | * userspace to pass a hint to file system, | |
b8d6568a | 201 | * especially if the latter case turns out to be |
ad4fb9ca KM |
202 | * common. |
203 | */ | |
7e028976 | 204 | ex = path[depth].p_ext; |
ad4fb9ca KM |
205 | if (ex) { |
206 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
207 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
208 | ||
209 | if (block > ext_block) | |
210 | return ext_pblk + (block - ext_block); | |
211 | else | |
212 | return ext_pblk - (ext_block - block); | |
213 | } | |
a86c6181 | 214 | |
d0d856e8 RD |
215 | /* it looks like index is empty; |
216 | * try to find starting block from index itself */ | |
a86c6181 AT |
217 | if (path[depth].p_bh) |
218 | return path[depth].p_bh->b_blocknr; | |
219 | } | |
220 | ||
221 | /* OK. use inode's group */ | |
f86186b4 | 222 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
223 | } |
224 | ||
654b4908 AK |
225 | /* |
226 | * Allocation for a meta data block | |
227 | */ | |
f65e6fba | 228 | static ext4_fsblk_t |
654b4908 | 229 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 230 | struct ext4_ext_path *path, |
55f020db | 231 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 232 | { |
f65e6fba | 233 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
234 | |
235 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
236 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
237 | NULL, err); | |
a86c6181 AT |
238 | return newblock; |
239 | } | |
240 | ||
55ad63bf | 241 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
242 | { |
243 | int size; | |
244 | ||
245 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
246 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 247 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
248 | if (!check && size > 6) |
249 | size = 6; | |
a86c6181 AT |
250 | #endif |
251 | return size; | |
252 | } | |
253 | ||
55ad63bf | 254 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
255 | { |
256 | int size; | |
257 | ||
258 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
259 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 260 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
261 | if (!check && size > 5) |
262 | size = 5; | |
a86c6181 AT |
263 | #endif |
264 | return size; | |
265 | } | |
266 | ||
55ad63bf | 267 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
268 | { |
269 | int size; | |
270 | ||
271 | size = sizeof(EXT4_I(inode)->i_data); | |
272 | size -= sizeof(struct ext4_extent_header); | |
273 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 274 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
275 | if (!check && size > 3) |
276 | size = 3; | |
a86c6181 AT |
277 | #endif |
278 | return size; | |
279 | } | |
280 | ||
55ad63bf | 281 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
282 | { |
283 | int size; | |
284 | ||
285 | size = sizeof(EXT4_I(inode)->i_data); | |
286 | size -= sizeof(struct ext4_extent_header); | |
287 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 288 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
289 | if (!check && size > 4) |
290 | size = 4; | |
a86c6181 AT |
291 | #endif |
292 | return size; | |
293 | } | |
294 | ||
d2a17637 MC |
295 | /* |
296 | * Calculate the number of metadata blocks needed | |
297 | * to allocate @blocks | |
298 | * Worse case is one block per extent | |
299 | */ | |
01f49d0b | 300 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 301 | { |
9d0be502 | 302 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 303 | int idxs; |
d2a17637 | 304 | |
9d0be502 TT |
305 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
306 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
307 | |
308 | /* | |
9d0be502 TT |
309 | * If the new delayed allocation block is contiguous with the |
310 | * previous da block, it can share index blocks with the | |
311 | * previous block, so we only need to allocate a new index | |
312 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
313 | * an additional index block, and at ldxs**3 blocks, yet | |
314 | * another index blocks. | |
d2a17637 | 315 | */ |
9d0be502 TT |
316 | if (ei->i_da_metadata_calc_len && |
317 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
318 | int num = 0; |
319 | ||
9d0be502 TT |
320 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
321 | num++; | |
322 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
323 | num++; | |
324 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
325 | num++; | |
326 | ei->i_da_metadata_calc_len = 0; | |
327 | } else | |
328 | ei->i_da_metadata_calc_len++; | |
329 | ei->i_da_metadata_calc_last_lblock++; | |
330 | return num; | |
331 | } | |
d2a17637 | 332 | |
9d0be502 TT |
333 | /* |
334 | * In the worst case we need a new set of index blocks at | |
335 | * every level of the inode's extent tree. | |
336 | */ | |
337 | ei->i_da_metadata_calc_len = 1; | |
338 | ei->i_da_metadata_calc_last_lblock = lblock; | |
339 | return ext_depth(inode) + 1; | |
d2a17637 MC |
340 | } |
341 | ||
c29c0ae7 AT |
342 | static int |
343 | ext4_ext_max_entries(struct inode *inode, int depth) | |
344 | { | |
345 | int max; | |
346 | ||
347 | if (depth == ext_depth(inode)) { | |
348 | if (depth == 0) | |
55ad63bf | 349 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 350 | else |
55ad63bf | 351 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
352 | } else { |
353 | if (depth == 0) | |
55ad63bf | 354 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 355 | else |
55ad63bf | 356 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
357 | } |
358 | ||
359 | return max; | |
360 | } | |
361 | ||
56b19868 AK |
362 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
363 | { | |
bf89d16f | 364 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 365 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 366 | |
31d4f3a2 TT |
367 | if (len == 0) |
368 | return 0; | |
6fd058f7 | 369 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
370 | } |
371 | ||
372 | static int ext4_valid_extent_idx(struct inode *inode, | |
373 | struct ext4_extent_idx *ext_idx) | |
374 | { | |
bf89d16f | 375 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 376 | |
6fd058f7 | 377 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
378 | } |
379 | ||
380 | static int ext4_valid_extent_entries(struct inode *inode, | |
381 | struct ext4_extent_header *eh, | |
382 | int depth) | |
383 | { | |
56b19868 AK |
384 | unsigned short entries; |
385 | if (eh->eh_entries == 0) | |
386 | return 1; | |
387 | ||
388 | entries = le16_to_cpu(eh->eh_entries); | |
389 | ||
390 | if (depth == 0) { | |
391 | /* leaf entries */ | |
81fdbb4a | 392 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
56b19868 AK |
393 | while (entries) { |
394 | if (!ext4_valid_extent(inode, ext)) | |
395 | return 0; | |
396 | ext++; | |
397 | entries--; | |
398 | } | |
399 | } else { | |
81fdbb4a | 400 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
401 | while (entries) { |
402 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
403 | return 0; | |
404 | ext_idx++; | |
405 | entries--; | |
406 | } | |
407 | } | |
408 | return 1; | |
409 | } | |
410 | ||
c398eda0 TT |
411 | static int __ext4_ext_check(const char *function, unsigned int line, |
412 | struct inode *inode, struct ext4_extent_header *eh, | |
413 | int depth) | |
c29c0ae7 AT |
414 | { |
415 | const char *error_msg; | |
416 | int max = 0; | |
417 | ||
418 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
419 | error_msg = "invalid magic"; | |
420 | goto corrupted; | |
421 | } | |
422 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
423 | error_msg = "unexpected eh_depth"; | |
424 | goto corrupted; | |
425 | } | |
426 | if (unlikely(eh->eh_max == 0)) { | |
427 | error_msg = "invalid eh_max"; | |
428 | goto corrupted; | |
429 | } | |
430 | max = ext4_ext_max_entries(inode, depth); | |
431 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
432 | error_msg = "too large eh_max"; | |
433 | goto corrupted; | |
434 | } | |
435 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
436 | error_msg = "invalid eh_entries"; | |
437 | goto corrupted; | |
438 | } | |
56b19868 AK |
439 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
440 | error_msg = "invalid extent entries"; | |
441 | goto corrupted; | |
442 | } | |
7ac5990d DW |
443 | /* Verify checksum on non-root extent tree nodes */ |
444 | if (ext_depth(inode) != depth && | |
445 | !ext4_extent_block_csum_verify(inode, eh)) { | |
446 | error_msg = "extent tree corrupted"; | |
447 | goto corrupted; | |
448 | } | |
c29c0ae7 AT |
449 | return 0; |
450 | ||
451 | corrupted: | |
c398eda0 | 452 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 453 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 454 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 455 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
456 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
457 | max, le16_to_cpu(eh->eh_depth), depth); | |
458 | ||
459 | return -EIO; | |
460 | } | |
461 | ||
56b19868 | 462 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 463 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 464 | |
7a262f7c AK |
465 | int ext4_ext_check_inode(struct inode *inode) |
466 | { | |
467 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
468 | } | |
469 | ||
f8489128 DW |
470 | static int __ext4_ext_check_block(const char *function, unsigned int line, |
471 | struct inode *inode, | |
472 | struct ext4_extent_header *eh, | |
473 | int depth, | |
474 | struct buffer_head *bh) | |
475 | { | |
476 | int ret; | |
477 | ||
478 | if (buffer_verified(bh)) | |
479 | return 0; | |
480 | ret = ext4_ext_check(inode, eh, depth); | |
481 | if (ret) | |
482 | return ret; | |
483 | set_buffer_verified(bh); | |
484 | return ret; | |
485 | } | |
486 | ||
487 | #define ext4_ext_check_block(inode, eh, depth, bh) \ | |
488 | __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh) | |
489 | ||
a86c6181 AT |
490 | #ifdef EXT_DEBUG |
491 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
492 | { | |
493 | int k, l = path->p_depth; | |
494 | ||
495 | ext_debug("path:"); | |
496 | for (k = 0; k <= l; k++, path++) { | |
497 | if (path->p_idx) { | |
2ae02107 | 498 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 499 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 500 | } else if (path->p_ext) { |
553f9008 | 501 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 502 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 503 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 504 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 505 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
506 | } else |
507 | ext_debug(" []"); | |
508 | } | |
509 | ext_debug("\n"); | |
510 | } | |
511 | ||
512 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
513 | { | |
514 | int depth = ext_depth(inode); | |
515 | struct ext4_extent_header *eh; | |
516 | struct ext4_extent *ex; | |
517 | int i; | |
518 | ||
519 | if (!path) | |
520 | return; | |
521 | ||
522 | eh = path[depth].p_hdr; | |
523 | ex = EXT_FIRST_EXTENT(eh); | |
524 | ||
553f9008 M |
525 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
526 | ||
a86c6181 | 527 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
528 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
529 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 530 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
531 | } |
532 | ext_debug("\n"); | |
533 | } | |
1b16da77 YY |
534 | |
535 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
536 | ext4_fsblk_t newblock, int level) | |
537 | { | |
538 | int depth = ext_depth(inode); | |
539 | struct ext4_extent *ex; | |
540 | ||
541 | if (depth != level) { | |
542 | struct ext4_extent_idx *idx; | |
543 | idx = path[level].p_idx; | |
544 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
545 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
546 | le32_to_cpu(idx->ei_block), | |
547 | ext4_idx_pblock(idx), | |
548 | newblock); | |
549 | idx++; | |
550 | } | |
551 | ||
552 | return; | |
553 | } | |
554 | ||
555 | ex = path[depth].p_ext; | |
556 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
557 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
558 | le32_to_cpu(ex->ee_block), | |
559 | ext4_ext_pblock(ex), | |
560 | ext4_ext_is_uninitialized(ex), | |
561 | ext4_ext_get_actual_len(ex), | |
562 | newblock); | |
563 | ex++; | |
564 | } | |
565 | } | |
566 | ||
a86c6181 | 567 | #else |
af5bc92d TT |
568 | #define ext4_ext_show_path(inode, path) |
569 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 570 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
571 | #endif |
572 | ||
b35905c1 | 573 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
574 | { |
575 | int depth = path->p_depth; | |
576 | int i; | |
577 | ||
578 | for (i = 0; i <= depth; i++, path++) | |
579 | if (path->p_bh) { | |
580 | brelse(path->p_bh); | |
581 | path->p_bh = NULL; | |
582 | } | |
583 | } | |
584 | ||
585 | /* | |
d0d856e8 RD |
586 | * ext4_ext_binsearch_idx: |
587 | * binary search for the closest index of the given block | |
c29c0ae7 | 588 | * the header must be checked before calling this |
a86c6181 AT |
589 | */ |
590 | static void | |
725d26d3 AK |
591 | ext4_ext_binsearch_idx(struct inode *inode, |
592 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
593 | { |
594 | struct ext4_extent_header *eh = path->p_hdr; | |
595 | struct ext4_extent_idx *r, *l, *m; | |
596 | ||
a86c6181 | 597 | |
bba90743 | 598 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
599 | |
600 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 601 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
602 | while (l <= r) { |
603 | m = l + (r - l) / 2; | |
604 | if (block < le32_to_cpu(m->ei_block)) | |
605 | r = m - 1; | |
606 | else | |
607 | l = m + 1; | |
26d535ed DM |
608 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
609 | m, le32_to_cpu(m->ei_block), | |
610 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
611 | } |
612 | ||
613 | path->p_idx = l - 1; | |
4a3c3a51 | 614 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 615 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
616 | |
617 | #ifdef CHECK_BINSEARCH | |
618 | { | |
619 | struct ext4_extent_idx *chix, *ix; | |
620 | int k; | |
621 | ||
622 | chix = ix = EXT_FIRST_INDEX(eh); | |
623 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
624 | if (k != 0 && | |
625 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
626 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
627 | "first=0x%p\n", k, | |
628 | ix, EXT_FIRST_INDEX(eh)); | |
629 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
630 | le32_to_cpu(ix->ei_block), |
631 | le32_to_cpu(ix[-1].ei_block)); | |
632 | } | |
633 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 634 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
635 | if (block < le32_to_cpu(ix->ei_block)) |
636 | break; | |
637 | chix = ix; | |
638 | } | |
639 | BUG_ON(chix != path->p_idx); | |
640 | } | |
641 | #endif | |
642 | ||
643 | } | |
644 | ||
645 | /* | |
d0d856e8 RD |
646 | * ext4_ext_binsearch: |
647 | * binary search for closest extent of the given block | |
c29c0ae7 | 648 | * the header must be checked before calling this |
a86c6181 AT |
649 | */ |
650 | static void | |
725d26d3 AK |
651 | ext4_ext_binsearch(struct inode *inode, |
652 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
653 | { |
654 | struct ext4_extent_header *eh = path->p_hdr; | |
655 | struct ext4_extent *r, *l, *m; | |
656 | ||
a86c6181 AT |
657 | if (eh->eh_entries == 0) { |
658 | /* | |
d0d856e8 RD |
659 | * this leaf is empty: |
660 | * we get such a leaf in split/add case | |
a86c6181 AT |
661 | */ |
662 | return; | |
663 | } | |
664 | ||
bba90743 | 665 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
666 | |
667 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 668 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
669 | |
670 | while (l <= r) { | |
671 | m = l + (r - l) / 2; | |
672 | if (block < le32_to_cpu(m->ee_block)) | |
673 | r = m - 1; | |
674 | else | |
675 | l = m + 1; | |
26d535ed DM |
676 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
677 | m, le32_to_cpu(m->ee_block), | |
678 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
679 | } |
680 | ||
681 | path->p_ext = l - 1; | |
553f9008 | 682 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 683 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 684 | ext4_ext_pblock(path->p_ext), |
553f9008 | 685 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 686 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
687 | |
688 | #ifdef CHECK_BINSEARCH | |
689 | { | |
690 | struct ext4_extent *chex, *ex; | |
691 | int k; | |
692 | ||
693 | chex = ex = EXT_FIRST_EXTENT(eh); | |
694 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
695 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 696 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
697 | if (block < le32_to_cpu(ex->ee_block)) |
698 | break; | |
699 | chex = ex; | |
700 | } | |
701 | BUG_ON(chex != path->p_ext); | |
702 | } | |
703 | #endif | |
704 | ||
705 | } | |
706 | ||
707 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
708 | { | |
709 | struct ext4_extent_header *eh; | |
710 | ||
711 | eh = ext_inode_hdr(inode); | |
712 | eh->eh_depth = 0; | |
713 | eh->eh_entries = 0; | |
714 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 715 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 716 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
717 | return 0; |
718 | } | |
719 | ||
720 | struct ext4_ext_path * | |
725d26d3 AK |
721 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
722 | struct ext4_ext_path *path) | |
a86c6181 AT |
723 | { |
724 | struct ext4_extent_header *eh; | |
725 | struct buffer_head *bh; | |
726 | short int depth, i, ppos = 0, alloc = 0; | |
860d21e2 | 727 | int ret; |
a86c6181 AT |
728 | |
729 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 730 | depth = ext_depth(inode); |
a86c6181 AT |
731 | |
732 | /* account possible depth increase */ | |
733 | if (!path) { | |
5d4958f9 | 734 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
735 | GFP_NOFS); |
736 | if (!path) | |
737 | return ERR_PTR(-ENOMEM); | |
738 | alloc = 1; | |
739 | } | |
a86c6181 | 740 | path[0].p_hdr = eh; |
1973adcb | 741 | path[0].p_bh = NULL; |
a86c6181 | 742 | |
c29c0ae7 | 743 | i = depth; |
a86c6181 AT |
744 | /* walk through the tree */ |
745 | while (i) { | |
746 | ext_debug("depth %d: num %d, max %d\n", | |
747 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 748 | |
a86c6181 | 749 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 750 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
751 | path[ppos].p_depth = i; |
752 | path[ppos].p_ext = NULL; | |
753 | ||
7a262f7c | 754 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
860d21e2 TT |
755 | if (unlikely(!bh)) { |
756 | ret = -ENOMEM; | |
a86c6181 | 757 | goto err; |
860d21e2 | 758 | } |
7a262f7c | 759 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
760 | trace_ext4_ext_load_extent(inode, block, |
761 | path[ppos].p_block); | |
860d21e2 TT |
762 | ret = bh_submit_read(bh); |
763 | if (ret < 0) { | |
7a262f7c AK |
764 | put_bh(bh); |
765 | goto err; | |
766 | } | |
7a262f7c | 767 | } |
a86c6181 AT |
768 | eh = ext_block_hdr(bh); |
769 | ppos++; | |
273df556 FM |
770 | if (unlikely(ppos > depth)) { |
771 | put_bh(bh); | |
772 | EXT4_ERROR_INODE(inode, | |
773 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 774 | ret = -EIO; |
273df556 FM |
775 | goto err; |
776 | } | |
a86c6181 AT |
777 | path[ppos].p_bh = bh; |
778 | path[ppos].p_hdr = eh; | |
779 | i--; | |
780 | ||
860d21e2 TT |
781 | ret = ext4_ext_check_block(inode, eh, i, bh); |
782 | if (ret < 0) | |
a86c6181 AT |
783 | goto err; |
784 | } | |
785 | ||
786 | path[ppos].p_depth = i; | |
a86c6181 AT |
787 | path[ppos].p_ext = NULL; |
788 | path[ppos].p_idx = NULL; | |
789 | ||
a86c6181 AT |
790 | /* find extent */ |
791 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
792 | /* if not an empty leaf */ |
793 | if (path[ppos].p_ext) | |
bf89d16f | 794 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
795 | |
796 | ext4_ext_show_path(inode, path); | |
797 | ||
798 | return path; | |
799 | ||
800 | err: | |
801 | ext4_ext_drop_refs(path); | |
802 | if (alloc) | |
803 | kfree(path); | |
860d21e2 | 804 | return ERR_PTR(ret); |
a86c6181 AT |
805 | } |
806 | ||
807 | /* | |
d0d856e8 RD |
808 | * ext4_ext_insert_index: |
809 | * insert new index [@logical;@ptr] into the block at @curp; | |
810 | * check where to insert: before @curp or after @curp | |
a86c6181 | 811 | */ |
1f109d5a TT |
812 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
813 | struct ext4_ext_path *curp, | |
814 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
815 | { |
816 | struct ext4_extent_idx *ix; | |
817 | int len, err; | |
818 | ||
7e028976 AM |
819 | err = ext4_ext_get_access(handle, inode, curp); |
820 | if (err) | |
a86c6181 AT |
821 | return err; |
822 | ||
273df556 FM |
823 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
824 | EXT4_ERROR_INODE(inode, | |
825 | "logical %d == ei_block %d!", | |
826 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
827 | return -EIO; | |
828 | } | |
d4620315 RD |
829 | |
830 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
831 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
832 | EXT4_ERROR_INODE(inode, | |
833 | "eh_entries %d >= eh_max %d!", | |
834 | le16_to_cpu(curp->p_hdr->eh_entries), | |
835 | le16_to_cpu(curp->p_hdr->eh_max)); | |
836 | return -EIO; | |
837 | } | |
838 | ||
a86c6181 AT |
839 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
840 | /* insert after */ | |
80e675f9 | 841 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
842 | ix = curp->p_idx + 1; |
843 | } else { | |
844 | /* insert before */ | |
80e675f9 | 845 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
846 | ix = curp->p_idx; |
847 | } | |
848 | ||
80e675f9 EG |
849 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
850 | BUG_ON(len < 0); | |
851 | if (len > 0) { | |
852 | ext_debug("insert new index %d: " | |
853 | "move %d indices from 0x%p to 0x%p\n", | |
854 | logical, len, ix, ix + 1); | |
855 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
856 | } | |
857 | ||
f472e026 TM |
858 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
859 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
860 | return -EIO; | |
861 | } | |
862 | ||
a86c6181 | 863 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 864 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 865 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 866 | |
273df556 FM |
867 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
868 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
869 | return -EIO; | |
870 | } | |
a86c6181 AT |
871 | |
872 | err = ext4_ext_dirty(handle, inode, curp); | |
873 | ext4_std_error(inode->i_sb, err); | |
874 | ||
875 | return err; | |
876 | } | |
877 | ||
878 | /* | |
d0d856e8 RD |
879 | * ext4_ext_split: |
880 | * inserts new subtree into the path, using free index entry | |
881 | * at depth @at: | |
882 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
883 | * - makes decision where to split | |
884 | * - moves remaining extents and index entries (right to the split point) | |
885 | * into the newly allocated blocks | |
886 | * - initializes subtree | |
a86c6181 AT |
887 | */ |
888 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
889 | unsigned int flags, |
890 | struct ext4_ext_path *path, | |
891 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
892 | { |
893 | struct buffer_head *bh = NULL; | |
894 | int depth = ext_depth(inode); | |
895 | struct ext4_extent_header *neh; | |
896 | struct ext4_extent_idx *fidx; | |
a86c6181 | 897 | int i = at, k, m, a; |
f65e6fba | 898 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 899 | __le32 border; |
f65e6fba | 900 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
901 | int err = 0; |
902 | ||
903 | /* make decision: where to split? */ | |
d0d856e8 | 904 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 905 | |
d0d856e8 | 906 | /* if current leaf will be split, then we should use |
a86c6181 | 907 | * border from split point */ |
273df556 FM |
908 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
909 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
910 | return -EIO; | |
911 | } | |
a86c6181 AT |
912 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
913 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 914 | ext_debug("leaf will be split." |
a86c6181 | 915 | " next leaf starts at %d\n", |
8c55e204 | 916 | le32_to_cpu(border)); |
a86c6181 AT |
917 | } else { |
918 | border = newext->ee_block; | |
919 | ext_debug("leaf will be added." | |
920 | " next leaf starts at %d\n", | |
8c55e204 | 921 | le32_to_cpu(border)); |
a86c6181 AT |
922 | } |
923 | ||
924 | /* | |
d0d856e8 RD |
925 | * If error occurs, then we break processing |
926 | * and mark filesystem read-only. index won't | |
a86c6181 | 927 | * be inserted and tree will be in consistent |
d0d856e8 | 928 | * state. Next mount will repair buffers too. |
a86c6181 AT |
929 | */ |
930 | ||
931 | /* | |
d0d856e8 RD |
932 | * Get array to track all allocated blocks. |
933 | * We need this to handle errors and free blocks | |
934 | * upon them. | |
a86c6181 | 935 | */ |
5d4958f9 | 936 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
937 | if (!ablocks) |
938 | return -ENOMEM; | |
a86c6181 AT |
939 | |
940 | /* allocate all needed blocks */ | |
941 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
942 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 943 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 944 | newext, &err, flags); |
a86c6181 AT |
945 | if (newblock == 0) |
946 | goto cleanup; | |
947 | ablocks[a] = newblock; | |
948 | } | |
949 | ||
950 | /* initialize new leaf */ | |
951 | newblock = ablocks[--a]; | |
273df556 FM |
952 | if (unlikely(newblock == 0)) { |
953 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
954 | err = -EIO; | |
955 | goto cleanup; | |
956 | } | |
a86c6181 | 957 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 958 | if (unlikely(!bh)) { |
860d21e2 | 959 | err = -ENOMEM; |
a86c6181 AT |
960 | goto cleanup; |
961 | } | |
962 | lock_buffer(bh); | |
963 | ||
7e028976 AM |
964 | err = ext4_journal_get_create_access(handle, bh); |
965 | if (err) | |
a86c6181 AT |
966 | goto cleanup; |
967 | ||
968 | neh = ext_block_hdr(bh); | |
969 | neh->eh_entries = 0; | |
55ad63bf | 970 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
971 | neh->eh_magic = EXT4_EXT_MAGIC; |
972 | neh->eh_depth = 0; | |
a86c6181 | 973 | |
d0d856e8 | 974 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
975 | if (unlikely(path[depth].p_hdr->eh_entries != |
976 | path[depth].p_hdr->eh_max)) { | |
977 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
978 | path[depth].p_hdr->eh_entries, | |
979 | path[depth].p_hdr->eh_max); | |
980 | err = -EIO; | |
981 | goto cleanup; | |
982 | } | |
a86c6181 | 983 | /* start copy from next extent */ |
1b16da77 YY |
984 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
985 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 986 | if (m) { |
1b16da77 YY |
987 | struct ext4_extent *ex; |
988 | ex = EXT_FIRST_EXTENT(neh); | |
989 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 990 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
991 | } |
992 | ||
7ac5990d | 993 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
994 | set_buffer_uptodate(bh); |
995 | unlock_buffer(bh); | |
996 | ||
0390131b | 997 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 998 | if (err) |
a86c6181 AT |
999 | goto cleanup; |
1000 | brelse(bh); | |
1001 | bh = NULL; | |
1002 | ||
1003 | /* correct old leaf */ | |
1004 | if (m) { | |
7e028976 AM |
1005 | err = ext4_ext_get_access(handle, inode, path + depth); |
1006 | if (err) | |
a86c6181 | 1007 | goto cleanup; |
e8546d06 | 1008 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1009 | err = ext4_ext_dirty(handle, inode, path + depth); |
1010 | if (err) | |
a86c6181 AT |
1011 | goto cleanup; |
1012 | ||
1013 | } | |
1014 | ||
1015 | /* create intermediate indexes */ | |
1016 | k = depth - at - 1; | |
273df556 FM |
1017 | if (unlikely(k < 0)) { |
1018 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1019 | err = -EIO; | |
1020 | goto cleanup; | |
1021 | } | |
a86c6181 AT |
1022 | if (k) |
1023 | ext_debug("create %d intermediate indices\n", k); | |
1024 | /* insert new index into current index block */ | |
1025 | /* current depth stored in i var */ | |
1026 | i = depth - 1; | |
1027 | while (k--) { | |
1028 | oldblock = newblock; | |
1029 | newblock = ablocks[--a]; | |
bba90743 | 1030 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1031 | if (unlikely(!bh)) { |
860d21e2 | 1032 | err = -ENOMEM; |
a86c6181 AT |
1033 | goto cleanup; |
1034 | } | |
1035 | lock_buffer(bh); | |
1036 | ||
7e028976 AM |
1037 | err = ext4_journal_get_create_access(handle, bh); |
1038 | if (err) | |
a86c6181 AT |
1039 | goto cleanup; |
1040 | ||
1041 | neh = ext_block_hdr(bh); | |
1042 | neh->eh_entries = cpu_to_le16(1); | |
1043 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1044 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1045 | neh->eh_depth = cpu_to_le16(depth - i); |
1046 | fidx = EXT_FIRST_INDEX(neh); | |
1047 | fidx->ei_block = border; | |
f65e6fba | 1048 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1049 | |
bba90743 ES |
1050 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1051 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1052 | |
1b16da77 | 1053 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1054 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1055 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1056 | EXT4_ERROR_INODE(inode, | |
1057 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1058 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1059 | err = -EIO; | |
1060 | goto cleanup; | |
1061 | } | |
1b16da77 YY |
1062 | /* start copy indexes */ |
1063 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1064 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1065 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1066 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1067 | if (m) { |
1b16da77 | 1068 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1069 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1070 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1071 | } |
7ac5990d | 1072 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1073 | set_buffer_uptodate(bh); |
1074 | unlock_buffer(bh); | |
1075 | ||
0390131b | 1076 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1077 | if (err) |
a86c6181 AT |
1078 | goto cleanup; |
1079 | brelse(bh); | |
1080 | bh = NULL; | |
1081 | ||
1082 | /* correct old index */ | |
1083 | if (m) { | |
1084 | err = ext4_ext_get_access(handle, inode, path + i); | |
1085 | if (err) | |
1086 | goto cleanup; | |
e8546d06 | 1087 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1088 | err = ext4_ext_dirty(handle, inode, path + i); |
1089 | if (err) | |
1090 | goto cleanup; | |
1091 | } | |
1092 | ||
1093 | i--; | |
1094 | } | |
1095 | ||
1096 | /* insert new index */ | |
a86c6181 AT |
1097 | err = ext4_ext_insert_index(handle, inode, path + at, |
1098 | le32_to_cpu(border), newblock); | |
1099 | ||
1100 | cleanup: | |
1101 | if (bh) { | |
1102 | if (buffer_locked(bh)) | |
1103 | unlock_buffer(bh); | |
1104 | brelse(bh); | |
1105 | } | |
1106 | ||
1107 | if (err) { | |
1108 | /* free all allocated blocks in error case */ | |
1109 | for (i = 0; i < depth; i++) { | |
1110 | if (!ablocks[i]) | |
1111 | continue; | |
7dc57615 | 1112 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1113 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1114 | } |
1115 | } | |
1116 | kfree(ablocks); | |
1117 | ||
1118 | return err; | |
1119 | } | |
1120 | ||
1121 | /* | |
d0d856e8 RD |
1122 | * ext4_ext_grow_indepth: |
1123 | * implements tree growing procedure: | |
1124 | * - allocates new block | |
1125 | * - moves top-level data (index block or leaf) into the new block | |
1126 | * - initializes new top-level, creating index that points to the | |
1127 | * just created block | |
a86c6181 AT |
1128 | */ |
1129 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1130 | unsigned int flags, |
55f020db | 1131 | struct ext4_extent *newext) |
a86c6181 | 1132 | { |
a86c6181 | 1133 | struct ext4_extent_header *neh; |
a86c6181 | 1134 | struct buffer_head *bh; |
f65e6fba | 1135 | ext4_fsblk_t newblock; |
a86c6181 AT |
1136 | int err = 0; |
1137 | ||
1939dd84 | 1138 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1139 | newext, &err, flags); |
a86c6181 AT |
1140 | if (newblock == 0) |
1141 | return err; | |
1142 | ||
1143 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1144 | if (unlikely(!bh)) |
860d21e2 | 1145 | return -ENOMEM; |
a86c6181 AT |
1146 | lock_buffer(bh); |
1147 | ||
7e028976 AM |
1148 | err = ext4_journal_get_create_access(handle, bh); |
1149 | if (err) { | |
a86c6181 AT |
1150 | unlock_buffer(bh); |
1151 | goto out; | |
1152 | } | |
1153 | ||
1154 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1155 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1156 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1157 | |
1158 | /* set size of new block */ | |
1159 | neh = ext_block_hdr(bh); | |
1160 | /* old root could have indexes or leaves | |
1161 | * so calculate e_max right way */ | |
1162 | if (ext_depth(inode)) | |
55ad63bf | 1163 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1164 | else |
55ad63bf | 1165 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1166 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1167 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1168 | set_buffer_uptodate(bh); |
1169 | unlock_buffer(bh); | |
1170 | ||
0390131b | 1171 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1172 | if (err) |
a86c6181 AT |
1173 | goto out; |
1174 | ||
1939dd84 | 1175 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1176 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1177 | neh->eh_entries = cpu_to_le16(1); |
1178 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1179 | if (neh->eh_depth == 0) { | |
1180 | /* Root extent block becomes index block */ | |
1181 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1182 | EXT_FIRST_INDEX(neh)->ei_block = | |
1183 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1184 | } | |
2ae02107 | 1185 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1186 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1187 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1188 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1189 | |
ba39ebb6 | 1190 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1191 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1192 | out: |
1193 | brelse(bh); | |
1194 | ||
1195 | return err; | |
1196 | } | |
1197 | ||
1198 | /* | |
d0d856e8 RD |
1199 | * ext4_ext_create_new_leaf: |
1200 | * finds empty index and adds new leaf. | |
1201 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1202 | */ |
1203 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1204 | unsigned int flags, |
1205 | struct ext4_ext_path *path, | |
1206 | struct ext4_extent *newext) | |
a86c6181 AT |
1207 | { |
1208 | struct ext4_ext_path *curp; | |
1209 | int depth, i, err = 0; | |
1210 | ||
1211 | repeat: | |
1212 | i = depth = ext_depth(inode); | |
1213 | ||
1214 | /* walk up to the tree and look for free index entry */ | |
1215 | curp = path + depth; | |
1216 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1217 | i--; | |
1218 | curp--; | |
1219 | } | |
1220 | ||
d0d856e8 RD |
1221 | /* we use already allocated block for index block, |
1222 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1223 | if (EXT_HAS_FREE_INDEX(curp)) { |
1224 | /* if we found index with free entry, then use that | |
1225 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1226 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1227 | if (err) |
1228 | goto out; | |
a86c6181 AT |
1229 | |
1230 | /* refill path */ | |
1231 | ext4_ext_drop_refs(path); | |
1232 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1233 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1234 | path); | |
a86c6181 AT |
1235 | if (IS_ERR(path)) |
1236 | err = PTR_ERR(path); | |
1237 | } else { | |
1238 | /* tree is full, time to grow in depth */ | |
1939dd84 | 1239 | err = ext4_ext_grow_indepth(handle, inode, flags, newext); |
a86c6181 AT |
1240 | if (err) |
1241 | goto out; | |
1242 | ||
1243 | /* refill path */ | |
1244 | ext4_ext_drop_refs(path); | |
1245 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1246 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1247 | path); | |
a86c6181 AT |
1248 | if (IS_ERR(path)) { |
1249 | err = PTR_ERR(path); | |
1250 | goto out; | |
1251 | } | |
1252 | ||
1253 | /* | |
d0d856e8 RD |
1254 | * only first (depth 0 -> 1) produces free space; |
1255 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1256 | */ |
1257 | depth = ext_depth(inode); | |
1258 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1259 | /* now we need to split */ |
a86c6181 AT |
1260 | goto repeat; |
1261 | } | |
1262 | } | |
1263 | ||
1264 | out: | |
1265 | return err; | |
1266 | } | |
1267 | ||
1988b51e AT |
1268 | /* |
1269 | * search the closest allocated block to the left for *logical | |
1270 | * and returns it at @logical + it's physical address at @phys | |
1271 | * if *logical is the smallest allocated block, the function | |
1272 | * returns 0 at @phys | |
1273 | * return value contains 0 (success) or error code | |
1274 | */ | |
1f109d5a TT |
1275 | static int ext4_ext_search_left(struct inode *inode, |
1276 | struct ext4_ext_path *path, | |
1277 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1278 | { |
1279 | struct ext4_extent_idx *ix; | |
1280 | struct ext4_extent *ex; | |
b939e376 | 1281 | int depth, ee_len; |
1988b51e | 1282 | |
273df556 FM |
1283 | if (unlikely(path == NULL)) { |
1284 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1285 | return -EIO; | |
1286 | } | |
1988b51e AT |
1287 | depth = path->p_depth; |
1288 | *phys = 0; | |
1289 | ||
1290 | if (depth == 0 && path->p_ext == NULL) | |
1291 | return 0; | |
1292 | ||
1293 | /* usually extent in the path covers blocks smaller | |
1294 | * then *logical, but it can be that extent is the | |
1295 | * first one in the file */ | |
1296 | ||
1297 | ex = path[depth].p_ext; | |
b939e376 | 1298 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1299 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1300 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1301 | EXT4_ERROR_INODE(inode, | |
1302 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1303 | *logical, le32_to_cpu(ex->ee_block)); | |
1304 | return -EIO; | |
1305 | } | |
1988b51e AT |
1306 | while (--depth >= 0) { |
1307 | ix = path[depth].p_idx; | |
273df556 FM |
1308 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1309 | EXT4_ERROR_INODE(inode, | |
1310 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1311 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1312 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1313 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1314 | depth); |
1315 | return -EIO; | |
1316 | } | |
1988b51e AT |
1317 | } |
1318 | return 0; | |
1319 | } | |
1320 | ||
273df556 FM |
1321 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1322 | EXT4_ERROR_INODE(inode, | |
1323 | "logical %d < ee_block %d + ee_len %d!", | |
1324 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1325 | return -EIO; | |
1326 | } | |
1988b51e | 1327 | |
b939e376 | 1328 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1329 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1330 | return 0; |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * search the closest allocated block to the right for *logical | |
1335 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1336 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1337 | * returns 0 at @phys |
1338 | * return value contains 0 (success) or error code | |
1339 | */ | |
1f109d5a TT |
1340 | static int ext4_ext_search_right(struct inode *inode, |
1341 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1342 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1343 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1344 | { |
1345 | struct buffer_head *bh = NULL; | |
1346 | struct ext4_extent_header *eh; | |
1347 | struct ext4_extent_idx *ix; | |
1348 | struct ext4_extent *ex; | |
1349 | ext4_fsblk_t block; | |
395a87bf ES |
1350 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1351 | int ee_len; | |
1988b51e | 1352 | |
273df556 FM |
1353 | if (unlikely(path == NULL)) { |
1354 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1355 | return -EIO; | |
1356 | } | |
1988b51e AT |
1357 | depth = path->p_depth; |
1358 | *phys = 0; | |
1359 | ||
1360 | if (depth == 0 && path->p_ext == NULL) | |
1361 | return 0; | |
1362 | ||
1363 | /* usually extent in the path covers blocks smaller | |
1364 | * then *logical, but it can be that extent is the | |
1365 | * first one in the file */ | |
1366 | ||
1367 | ex = path[depth].p_ext; | |
b939e376 | 1368 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1369 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1370 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1371 | EXT4_ERROR_INODE(inode, | |
1372 | "first_extent(path[%d].p_hdr) != ex", | |
1373 | depth); | |
1374 | return -EIO; | |
1375 | } | |
1988b51e AT |
1376 | while (--depth >= 0) { |
1377 | ix = path[depth].p_idx; | |
273df556 FM |
1378 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1379 | EXT4_ERROR_INODE(inode, | |
1380 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1381 | *logical); | |
1382 | return -EIO; | |
1383 | } | |
1988b51e | 1384 | } |
4d33b1ef | 1385 | goto found_extent; |
1988b51e AT |
1386 | } |
1387 | ||
273df556 FM |
1388 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1389 | EXT4_ERROR_INODE(inode, | |
1390 | "logical %d < ee_block %d + ee_len %d!", | |
1391 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1392 | return -EIO; | |
1393 | } | |
1988b51e AT |
1394 | |
1395 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1396 | /* next allocated block in this leaf */ | |
1397 | ex++; | |
4d33b1ef | 1398 | goto found_extent; |
1988b51e AT |
1399 | } |
1400 | ||
1401 | /* go up and search for index to the right */ | |
1402 | while (--depth >= 0) { | |
1403 | ix = path[depth].p_idx; | |
1404 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1405 | goto got_index; |
1988b51e AT |
1406 | } |
1407 | ||
25f1ee3a WF |
1408 | /* we've gone up to the root and found no index to the right */ |
1409 | return 0; | |
1988b51e | 1410 | |
25f1ee3a | 1411 | got_index: |
1988b51e AT |
1412 | /* we've found index to the right, let's |
1413 | * follow it and find the closest allocated | |
1414 | * block to the right */ | |
1415 | ix++; | |
bf89d16f | 1416 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1417 | while (++depth < path->p_depth) { |
1418 | bh = sb_bread(inode->i_sb, block); | |
1419 | if (bh == NULL) | |
1420 | return -EIO; | |
1421 | eh = ext_block_hdr(bh); | |
395a87bf | 1422 | /* subtract from p_depth to get proper eh_depth */ |
f8489128 DW |
1423 | if (ext4_ext_check_block(inode, eh, |
1424 | path->p_depth - depth, bh)) { | |
1988b51e AT |
1425 | put_bh(bh); |
1426 | return -EIO; | |
1427 | } | |
1428 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1429 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1430 | put_bh(bh); |
1431 | } | |
1432 | ||
1433 | bh = sb_bread(inode->i_sb, block); | |
1434 | if (bh == NULL) | |
1435 | return -EIO; | |
1436 | eh = ext_block_hdr(bh); | |
f8489128 | 1437 | if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) { |
1988b51e AT |
1438 | put_bh(bh); |
1439 | return -EIO; | |
1440 | } | |
1441 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1442 | found_extent: |
1988b51e | 1443 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1444 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1445 | *ret_ex = ex; |
1446 | if (bh) | |
1447 | put_bh(bh); | |
1988b51e | 1448 | return 0; |
1988b51e AT |
1449 | } |
1450 | ||
a86c6181 | 1451 | /* |
d0d856e8 | 1452 | * ext4_ext_next_allocated_block: |
f17722f9 | 1453 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1454 | * NOTE: it considers block number from index entry as |
1455 | * allocated block. Thus, index entries have to be consistent | |
1456 | * with leaves. | |
a86c6181 | 1457 | */ |
725d26d3 | 1458 | static ext4_lblk_t |
a86c6181 AT |
1459 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1460 | { | |
1461 | int depth; | |
1462 | ||
1463 | BUG_ON(path == NULL); | |
1464 | depth = path->p_depth; | |
1465 | ||
1466 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1467 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1468 | |
1469 | while (depth >= 0) { | |
1470 | if (depth == path->p_depth) { | |
1471 | /* leaf */ | |
6f8ff537 CW |
1472 | if (path[depth].p_ext && |
1473 | path[depth].p_ext != | |
a86c6181 AT |
1474 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1475 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1476 | } else { | |
1477 | /* index */ | |
1478 | if (path[depth].p_idx != | |
1479 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1480 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1481 | } | |
1482 | depth--; | |
1483 | } | |
1484 | ||
f17722f9 | 1485 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1486 | } |
1487 | ||
1488 | /* | |
d0d856e8 | 1489 | * ext4_ext_next_leaf_block: |
f17722f9 | 1490 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1491 | */ |
5718789d | 1492 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1493 | { |
1494 | int depth; | |
1495 | ||
1496 | BUG_ON(path == NULL); | |
1497 | depth = path->p_depth; | |
1498 | ||
1499 | /* zero-tree has no leaf blocks at all */ | |
1500 | if (depth == 0) | |
f17722f9 | 1501 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1502 | |
1503 | /* go to index block */ | |
1504 | depth--; | |
1505 | ||
1506 | while (depth >= 0) { | |
1507 | if (path[depth].p_idx != | |
1508 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1509 | return (ext4_lblk_t) |
1510 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1511 | depth--; |
1512 | } | |
1513 | ||
f17722f9 | 1514 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1515 | } |
1516 | ||
1517 | /* | |
d0d856e8 RD |
1518 | * ext4_ext_correct_indexes: |
1519 | * if leaf gets modified and modified extent is first in the leaf, | |
1520 | * then we have to correct all indexes above. | |
a86c6181 AT |
1521 | * TODO: do we need to correct tree in all cases? |
1522 | */ | |
1d03ec98 | 1523 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1524 | struct ext4_ext_path *path) |
1525 | { | |
1526 | struct ext4_extent_header *eh; | |
1527 | int depth = ext_depth(inode); | |
1528 | struct ext4_extent *ex; | |
1529 | __le32 border; | |
1530 | int k, err = 0; | |
1531 | ||
1532 | eh = path[depth].p_hdr; | |
1533 | ex = path[depth].p_ext; | |
273df556 FM |
1534 | |
1535 | if (unlikely(ex == NULL || eh == NULL)) { | |
1536 | EXT4_ERROR_INODE(inode, | |
1537 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1538 | return -EIO; | |
1539 | } | |
a86c6181 AT |
1540 | |
1541 | if (depth == 0) { | |
1542 | /* there is no tree at all */ | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1546 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1547 | /* we correct tree if first leaf got modified only */ | |
1548 | return 0; | |
1549 | } | |
1550 | ||
1551 | /* | |
d0d856e8 | 1552 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1553 | */ |
1554 | k = depth - 1; | |
1555 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1556 | err = ext4_ext_get_access(handle, inode, path + k); |
1557 | if (err) | |
a86c6181 AT |
1558 | return err; |
1559 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1560 | err = ext4_ext_dirty(handle, inode, path + k); |
1561 | if (err) | |
a86c6181 AT |
1562 | return err; |
1563 | ||
1564 | while (k--) { | |
1565 | /* change all left-side indexes */ | |
1566 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1567 | break; | |
7e028976 AM |
1568 | err = ext4_ext_get_access(handle, inode, path + k); |
1569 | if (err) | |
a86c6181 AT |
1570 | break; |
1571 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1572 | err = ext4_ext_dirty(handle, inode, path + k); |
1573 | if (err) | |
a86c6181 AT |
1574 | break; |
1575 | } | |
1576 | ||
1577 | return err; | |
1578 | } | |
1579 | ||
748de673 | 1580 | int |
a86c6181 AT |
1581 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1582 | struct ext4_extent *ex2) | |
1583 | { | |
749269fa | 1584 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1585 | |
1586 | /* | |
ec22ba8e DM |
1587 | * Make sure that both extents are initialized. We don't merge |
1588 | * uninitialized extents so that we can be sure that end_io code has | |
1589 | * the extent that was written properly split out and conversion to | |
1590 | * initialized is trivial. | |
a2df2a63 | 1591 | */ |
ec22ba8e | 1592 | if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2)) |
a2df2a63 AA |
1593 | return 0; |
1594 | ||
749269fa AA |
1595 | if (ext4_ext_is_uninitialized(ex1)) |
1596 | max_len = EXT_UNINIT_MAX_LEN; | |
1597 | else | |
1598 | max_len = EXT_INIT_MAX_LEN; | |
1599 | ||
a2df2a63 AA |
1600 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1601 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1602 | ||
1603 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1604 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1605 | return 0; |
1606 | ||
471d4011 SB |
1607 | /* |
1608 | * To allow future support for preallocated extents to be added | |
1609 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1610 | * this can result in the top bit of ee_len being set. |
471d4011 | 1611 | */ |
749269fa | 1612 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1613 | return 0; |
bbf2f9fb | 1614 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1615 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1616 | return 0; |
1617 | #endif | |
1618 | ||
bf89d16f | 1619 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1620 | return 1; |
1621 | return 0; | |
1622 | } | |
1623 | ||
56055d3a AA |
1624 | /* |
1625 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1626 | * It always tries to merge towards right. If you want to merge towards | |
1627 | * left, pass "ex - 1" as argument instead of "ex". | |
1628 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1629 | * 1 if they got merged. | |
1630 | */ | |
197217a5 | 1631 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1632 | struct ext4_ext_path *path, |
1633 | struct ext4_extent *ex) | |
56055d3a AA |
1634 | { |
1635 | struct ext4_extent_header *eh; | |
1636 | unsigned int depth, len; | |
1637 | int merge_done = 0; | |
1638 | int uninitialized = 0; | |
1639 | ||
1640 | depth = ext_depth(inode); | |
1641 | BUG_ON(path[depth].p_hdr == NULL); | |
1642 | eh = path[depth].p_hdr; | |
1643 | ||
1644 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1645 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1646 | break; | |
1647 | /* merge with next extent! */ | |
1648 | if (ext4_ext_is_uninitialized(ex)) | |
1649 | uninitialized = 1; | |
1650 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1651 | + ext4_ext_get_actual_len(ex + 1)); | |
1652 | if (uninitialized) | |
1653 | ext4_ext_mark_uninitialized(ex); | |
1654 | ||
1655 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1656 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1657 | * sizeof(struct ext4_extent); | |
1658 | memmove(ex + 1, ex + 2, len); | |
1659 | } | |
e8546d06 | 1660 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1661 | merge_done = 1; |
1662 | WARN_ON(eh->eh_entries == 0); | |
1663 | if (!eh->eh_entries) | |
24676da4 | 1664 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1665 | } |
1666 | ||
1667 | return merge_done; | |
1668 | } | |
1669 | ||
ecb94f5f TT |
1670 | /* |
1671 | * This function does a very simple check to see if we can collapse | |
1672 | * an extent tree with a single extent tree leaf block into the inode. | |
1673 | */ | |
1674 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1675 | struct inode *inode, | |
1676 | struct ext4_ext_path *path) | |
1677 | { | |
1678 | size_t s; | |
1679 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1680 | ext4_fsblk_t blk; | |
1681 | ||
1682 | if ((path[0].p_depth != 1) || | |
1683 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1684 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1685 | return; | |
1686 | ||
1687 | /* | |
1688 | * We need to modify the block allocation bitmap and the block | |
1689 | * group descriptor to release the extent tree block. If we | |
1690 | * can't get the journal credits, give up. | |
1691 | */ | |
1692 | if (ext4_journal_extend(handle, 2)) | |
1693 | return; | |
1694 | ||
1695 | /* | |
1696 | * Copy the extent data up to the inode | |
1697 | */ | |
1698 | blk = ext4_idx_pblock(path[0].p_idx); | |
1699 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1700 | sizeof(struct ext4_extent_idx); | |
1701 | s += sizeof(struct ext4_extent_header); | |
1702 | ||
1703 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1704 | path[0].p_depth = 0; | |
1705 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1706 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1707 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1708 | ||
1709 | brelse(path[1].p_bh); | |
1710 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
1711 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); | |
1712 | } | |
1713 | ||
197217a5 YY |
1714 | /* |
1715 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1716 | * return 1 if merge left else 0. | |
1717 | */ | |
ecb94f5f TT |
1718 | static void ext4_ext_try_to_merge(handle_t *handle, |
1719 | struct inode *inode, | |
197217a5 YY |
1720 | struct ext4_ext_path *path, |
1721 | struct ext4_extent *ex) { | |
1722 | struct ext4_extent_header *eh; | |
1723 | unsigned int depth; | |
1724 | int merge_done = 0; | |
197217a5 YY |
1725 | |
1726 | depth = ext_depth(inode); | |
1727 | BUG_ON(path[depth].p_hdr == NULL); | |
1728 | eh = path[depth].p_hdr; | |
1729 | ||
1730 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1731 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1732 | ||
1733 | if (!merge_done) | |
ecb94f5f | 1734 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1735 | |
ecb94f5f | 1736 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1737 | } |
1738 | ||
25d14f98 AA |
1739 | /* |
1740 | * check if a portion of the "newext" extent overlaps with an | |
1741 | * existing extent. | |
1742 | * | |
1743 | * If there is an overlap discovered, it updates the length of the newext | |
1744 | * such that there will be no overlap, and then returns 1. | |
1745 | * If there is no overlap found, it returns 0. | |
1746 | */ | |
4d33b1ef TT |
1747 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1748 | struct inode *inode, | |
1f109d5a TT |
1749 | struct ext4_extent *newext, |
1750 | struct ext4_ext_path *path) | |
25d14f98 | 1751 | { |
725d26d3 | 1752 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1753 | unsigned int depth, len1; |
1754 | unsigned int ret = 0; | |
1755 | ||
1756 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1757 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1758 | depth = ext_depth(inode); |
1759 | if (!path[depth].p_ext) | |
1760 | goto out; | |
1761 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1762 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1763 | |
1764 | /* | |
1765 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1766 | * is before the requested block(s) |
25d14f98 AA |
1767 | */ |
1768 | if (b2 < b1) { | |
1769 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1770 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1771 | goto out; |
4d33b1ef | 1772 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1773 | } |
1774 | ||
725d26d3 | 1775 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1776 | if (b1 + len1 < b1) { |
f17722f9 | 1777 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1778 | newext->ee_len = cpu_to_le16(len1); |
1779 | ret = 1; | |
1780 | } | |
1781 | ||
1782 | /* check for overlap */ | |
1783 | if (b1 + len1 > b2) { | |
1784 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1785 | ret = 1; | |
1786 | } | |
1787 | out: | |
1788 | return ret; | |
1789 | } | |
1790 | ||
a86c6181 | 1791 | /* |
d0d856e8 RD |
1792 | * ext4_ext_insert_extent: |
1793 | * tries to merge requsted extent into the existing extent or | |
1794 | * inserts requested extent as new one into the tree, | |
1795 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1796 | */ |
1797 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1798 | struct ext4_ext_path *path, | |
0031462b | 1799 | struct ext4_extent *newext, int flag) |
a86c6181 | 1800 | { |
af5bc92d | 1801 | struct ext4_extent_header *eh; |
a86c6181 AT |
1802 | struct ext4_extent *ex, *fex; |
1803 | struct ext4_extent *nearex; /* nearest extent */ | |
1804 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1805 | int depth, len, err; |
1806 | ext4_lblk_t next; | |
a2df2a63 | 1807 | unsigned uninitialized = 0; |
55f020db | 1808 | int flags = 0; |
a86c6181 | 1809 | |
273df556 FM |
1810 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1811 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1812 | return -EIO; | |
1813 | } | |
a86c6181 AT |
1814 | depth = ext_depth(inode); |
1815 | ex = path[depth].p_ext; | |
273df556 FM |
1816 | if (unlikely(path[depth].p_hdr == NULL)) { |
1817 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1818 | return -EIO; | |
1819 | } | |
a86c6181 AT |
1820 | |
1821 | /* try to insert block into found extent and return */ | |
744692dc | 1822 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO) |
0031462b | 1823 | && ext4_can_extents_be_merged(inode, ex, newext)) { |
32de6756 | 1824 | ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n", |
bf89d16f TT |
1825 | ext4_ext_is_uninitialized(newext), |
1826 | ext4_ext_get_actual_len(newext), | |
1827 | le32_to_cpu(ex->ee_block), | |
1828 | ext4_ext_is_uninitialized(ex), | |
1829 | ext4_ext_get_actual_len(ex), | |
1830 | ext4_ext_pblock(ex)); | |
7e028976 AM |
1831 | err = ext4_ext_get_access(handle, inode, path + depth); |
1832 | if (err) | |
a86c6181 | 1833 | return err; |
a2df2a63 AA |
1834 | |
1835 | /* | |
1836 | * ext4_can_extents_be_merged should have checked that either | |
1837 | * both extents are uninitialized, or both aren't. Thus we | |
1838 | * need to check only one of them here. | |
1839 | */ | |
1840 | if (ext4_ext_is_uninitialized(ex)) | |
1841 | uninitialized = 1; | |
1842 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1843 | + ext4_ext_get_actual_len(newext)); | |
1844 | if (uninitialized) | |
1845 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1846 | eh = path[depth].p_hdr; |
1847 | nearex = ex; | |
1848 | goto merge; | |
1849 | } | |
1850 | ||
a86c6181 AT |
1851 | depth = ext_depth(inode); |
1852 | eh = path[depth].p_hdr; | |
1853 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1854 | goto has_space; | |
1855 | ||
1856 | /* probably next leaf has space for us? */ | |
1857 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1858 | next = EXT_MAX_BLOCKS; |
1859 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1860 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1861 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1862 | ext_debug("next leaf block - %u\n", next); |
a86c6181 AT |
1863 | BUG_ON(npath != NULL); |
1864 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1865 | if (IS_ERR(npath)) | |
1866 | return PTR_ERR(npath); | |
1867 | BUG_ON(npath->p_depth != path->p_depth); | |
1868 | eh = npath[depth].p_hdr; | |
1869 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1870 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1871 | le16_to_cpu(eh->eh_entries)); |
1872 | path = npath; | |
ffb505ff | 1873 | goto has_space; |
a86c6181 AT |
1874 | } |
1875 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1876 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1877 | } | |
1878 | ||
1879 | /* | |
d0d856e8 RD |
1880 | * There is no free space in the found leaf. |
1881 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1882 | */ |
55f020db AH |
1883 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1884 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1885 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1886 | if (err) |
1887 | goto cleanup; | |
1888 | depth = ext_depth(inode); | |
1889 | eh = path[depth].p_hdr; | |
1890 | ||
1891 | has_space: | |
1892 | nearex = path[depth].p_ext; | |
1893 | ||
7e028976 AM |
1894 | err = ext4_ext_get_access(handle, inode, path + depth); |
1895 | if (err) | |
a86c6181 AT |
1896 | goto cleanup; |
1897 | ||
1898 | if (!nearex) { | |
1899 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1900 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1901 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1902 | ext4_ext_pblock(newext), |
553f9008 | 1903 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1904 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1905 | nearex = EXT_FIRST_EXTENT(eh); |
1906 | } else { | |
1907 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1908 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1909 | /* Insert after */ |
32de6756 YY |
1910 | ext_debug("insert %u:%llu:[%d]%d before: " |
1911 | "nearest %p\n", | |
80e675f9 EG |
1912 | le32_to_cpu(newext->ee_block), |
1913 | ext4_ext_pblock(newext), | |
1914 | ext4_ext_is_uninitialized(newext), | |
1915 | ext4_ext_get_actual_len(newext), | |
1916 | nearex); | |
1917 | nearex++; | |
1918 | } else { | |
1919 | /* Insert before */ | |
1920 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
1921 | ext_debug("insert %u:%llu:[%d]%d after: " |
1922 | "nearest %p\n", | |
8c55e204 | 1923 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1924 | ext4_ext_pblock(newext), |
553f9008 | 1925 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1926 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
1927 | nearex); |
1928 | } | |
1929 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
1930 | if (len > 0) { | |
32de6756 | 1931 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
1932 | "move %d extents from 0x%p to 0x%p\n", |
1933 | le32_to_cpu(newext->ee_block), | |
1934 | ext4_ext_pblock(newext), | |
1935 | ext4_ext_is_uninitialized(newext), | |
1936 | ext4_ext_get_actual_len(newext), | |
1937 | len, nearex, nearex + 1); | |
1938 | memmove(nearex + 1, nearex, | |
1939 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 1940 | } |
a86c6181 AT |
1941 | } |
1942 | ||
e8546d06 | 1943 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 1944 | path[depth].p_ext = nearex; |
a86c6181 | 1945 | nearex->ee_block = newext->ee_block; |
bf89d16f | 1946 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 1947 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1948 | |
1949 | merge: | |
e7bcf823 | 1950 | /* try to merge extents */ |
744692dc | 1951 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 1952 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 1953 | |
a86c6181 AT |
1954 | |
1955 | /* time to correct all indexes above */ | |
1956 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1957 | if (err) | |
1958 | goto cleanup; | |
1959 | ||
ecb94f5f | 1960 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
1961 | |
1962 | cleanup: | |
1963 | if (npath) { | |
1964 | ext4_ext_drop_refs(npath); | |
1965 | kfree(npath); | |
1966 | } | |
a86c6181 AT |
1967 | return err; |
1968 | } | |
1969 | ||
91dd8c11 LC |
1970 | static int ext4_fill_fiemap_extents(struct inode *inode, |
1971 | ext4_lblk_t block, ext4_lblk_t num, | |
1972 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
1973 | { |
1974 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 1975 | struct ext4_extent *ex; |
69eb33dc | 1976 | struct extent_status es; |
91dd8c11 | 1977 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 1978 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
1979 | int exists, depth = 0, err = 0; |
1980 | unsigned int flags = 0; | |
1981 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 1982 | |
f17722f9 | 1983 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
1984 | num = last - block; |
1985 | /* find extent for this block */ | |
fab3a549 | 1986 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
1987 | |
1988 | if (path && ext_depth(inode) != depth) { | |
1989 | /* depth was changed. we have to realloc path */ | |
1990 | kfree(path); | |
1991 | path = NULL; | |
1992 | } | |
1993 | ||
6873fa0d ES |
1994 | path = ext4_ext_find_extent(inode, block, path); |
1995 | if (IS_ERR(path)) { | |
91dd8c11 | 1996 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
1997 | err = PTR_ERR(path); |
1998 | path = NULL; | |
1999 | break; | |
2000 | } | |
2001 | ||
2002 | depth = ext_depth(inode); | |
273df556 | 2003 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2004 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2005 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2006 | err = -EIO; | |
2007 | break; | |
2008 | } | |
6873fa0d ES |
2009 | ex = path[depth].p_ext; |
2010 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2011 | ext4_ext_drop_refs(path); |
6873fa0d | 2012 | |
91dd8c11 | 2013 | flags = 0; |
6873fa0d ES |
2014 | exists = 0; |
2015 | if (!ex) { | |
2016 | /* there is no extent yet, so try to allocate | |
2017 | * all requested space */ | |
2018 | start = block; | |
2019 | end = block + num; | |
2020 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2021 | /* need to allocate space before found extent */ | |
2022 | start = block; | |
2023 | end = le32_to_cpu(ex->ee_block); | |
2024 | if (block + num < end) | |
2025 | end = block + num; | |
2026 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2027 | + ext4_ext_get_actual_len(ex)) { | |
2028 | /* need to allocate space after found extent */ | |
2029 | start = block; | |
2030 | end = block + num; | |
2031 | if (end >= next) | |
2032 | end = next; | |
2033 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2034 | /* | |
2035 | * some part of requested space is covered | |
2036 | * by found extent | |
2037 | */ | |
2038 | start = block; | |
2039 | end = le32_to_cpu(ex->ee_block) | |
2040 | + ext4_ext_get_actual_len(ex); | |
2041 | if (block + num < end) | |
2042 | end = block + num; | |
2043 | exists = 1; | |
2044 | } else { | |
2045 | BUG(); | |
2046 | } | |
2047 | BUG_ON(end <= start); | |
2048 | ||
2049 | if (!exists) { | |
69eb33dc ZL |
2050 | es.es_lblk = start; |
2051 | es.es_len = end - start; | |
2052 | es.es_pblk = 0; | |
6873fa0d | 2053 | } else { |
69eb33dc ZL |
2054 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2055 | es.es_len = ext4_ext_get_actual_len(ex); | |
2056 | es.es_pblk = ext4_ext_pblock(ex); | |
91dd8c11 LC |
2057 | if (ext4_ext_is_uninitialized(ex)) |
2058 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
6873fa0d ES |
2059 | } |
2060 | ||
91dd8c11 | 2061 | /* |
69eb33dc ZL |
2062 | * Find delayed extent and update es accordingly. We call |
2063 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2064 | * last existing extent or not. |
2065 | */ | |
69eb33dc | 2066 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2067 | if (!exists && next_del) { |
2068 | exists = 1; | |
2069 | flags |= FIEMAP_EXTENT_DELALLOC; | |
2070 | } | |
2071 | up_read(&EXT4_I(inode)->i_data_sem); | |
2072 | ||
69eb33dc ZL |
2073 | if (unlikely(es.es_len == 0)) { |
2074 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
273df556 FM |
2075 | err = -EIO; |
2076 | break; | |
2077 | } | |
6873fa0d | 2078 | |
f7fec032 ZL |
2079 | /* |
2080 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2081 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2082 | * possible that an extent is with unwritten and delayed | |
2083 | * status due to when an extent is delayed allocated and | |
2084 | * is allocated by fallocate status tree will track both of | |
2085 | * them in a extent. | |
2086 | * | |
2087 | * So we could return a unwritten and delayed extent, and | |
2088 | * its block is equal to 'next'. | |
2089 | */ | |
2090 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2091 | flags |= FIEMAP_EXTENT_LAST; |
2092 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2093 | next != EXT_MAX_BLOCKS)) { | |
2094 | EXT4_ERROR_INODE(inode, | |
2095 | "next extent == %u, next " | |
2096 | "delalloc extent = %u", | |
2097 | next, next_del); | |
2098 | err = -EIO; | |
2099 | break; | |
2100 | } | |
6873fa0d ES |
2101 | } |
2102 | ||
91dd8c11 LC |
2103 | if (exists) { |
2104 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2105 | (__u64)es.es_lblk << blksize_bits, |
2106 | (__u64)es.es_pblk << blksize_bits, | |
2107 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2108 | flags); |
2109 | if (err < 0) | |
2110 | break; | |
2111 | if (err == 1) { | |
2112 | err = 0; | |
2113 | break; | |
2114 | } | |
6873fa0d ES |
2115 | } |
2116 | ||
69eb33dc | 2117 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2118 | } |
2119 | ||
2120 | if (path) { | |
2121 | ext4_ext_drop_refs(path); | |
2122 | kfree(path); | |
2123 | } | |
2124 | ||
2125 | return err; | |
2126 | } | |
2127 | ||
a86c6181 | 2128 | /* |
d0d856e8 RD |
2129 | * ext4_ext_put_gap_in_cache: |
2130 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2131 | * and cache this gap |
2132 | */ | |
09b88252 | 2133 | static void |
a86c6181 | 2134 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2135 | ext4_lblk_t block) |
a86c6181 AT |
2136 | { |
2137 | int depth = ext_depth(inode); | |
725d26d3 AK |
2138 | unsigned long len; |
2139 | ext4_lblk_t lblock; | |
a86c6181 AT |
2140 | struct ext4_extent *ex; |
2141 | ||
2142 | ex = path[depth].p_ext; | |
2143 | if (ex == NULL) { | |
69eb33dc ZL |
2144 | /* |
2145 | * there is no extent yet, so gap is [0;-] and we | |
2146 | * don't cache it | |
2147 | */ | |
a86c6181 AT |
2148 | ext_debug("cache gap(whole file):"); |
2149 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2150 | lblock = block; | |
2151 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2152 | ext_debug("cache gap(before): %u [%u:%u]", |
2153 | block, | |
2154 | le32_to_cpu(ex->ee_block), | |
2155 | ext4_ext_get_actual_len(ex)); | |
d100eef2 ZL |
2156 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2157 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2158 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2159 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2160 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2161 | ext4_lblk_t next; |
8c55e204 | 2162 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2163 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2164 | |
2165 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2166 | ext_debug("cache gap(after): [%u:%u] %u", |
2167 | le32_to_cpu(ex->ee_block), | |
2168 | ext4_ext_get_actual_len(ex), | |
2169 | block); | |
725d26d3 AK |
2170 | BUG_ON(next == lblock); |
2171 | len = next - lblock; | |
d100eef2 ZL |
2172 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2173 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2174 | EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2175 | } else { |
2176 | lblock = len = 0; | |
2177 | BUG(); | |
2178 | } | |
2179 | ||
bba90743 | 2180 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
2181 | } |
2182 | ||
2183 | /* | |
d0d856e8 RD |
2184 | * ext4_ext_rm_idx: |
2185 | * removes index from the index block. | |
a86c6181 | 2186 | */ |
1d03ec98 | 2187 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2188 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2189 | { |
a86c6181 | 2190 | int err; |
f65e6fba | 2191 | ext4_fsblk_t leaf; |
a86c6181 AT |
2192 | |
2193 | /* free index block */ | |
c36575e6 FL |
2194 | depth--; |
2195 | path = path + depth; | |
bf89d16f | 2196 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2197 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2198 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2199 | return -EIO; | |
2200 | } | |
7e028976 AM |
2201 | err = ext4_ext_get_access(handle, inode, path); |
2202 | if (err) | |
a86c6181 | 2203 | return err; |
0e1147b0 RD |
2204 | |
2205 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2206 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2207 | len *= sizeof(struct ext4_extent_idx); | |
2208 | memmove(path->p_idx, path->p_idx + 1, len); | |
2209 | } | |
2210 | ||
e8546d06 | 2211 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2212 | err = ext4_ext_dirty(handle, inode, path); |
2213 | if (err) | |
a86c6181 | 2214 | return err; |
2ae02107 | 2215 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2216 | trace_ext4_ext_rm_idx(inode, leaf); |
2217 | ||
7dc57615 | 2218 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2219 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2220 | |
2221 | while (--depth >= 0) { | |
2222 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2223 | break; | |
2224 | path--; | |
2225 | err = ext4_ext_get_access(handle, inode, path); | |
2226 | if (err) | |
2227 | break; | |
2228 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2229 | err = ext4_ext_dirty(handle, inode, path); | |
2230 | if (err) | |
2231 | break; | |
2232 | } | |
a86c6181 AT |
2233 | return err; |
2234 | } | |
2235 | ||
2236 | /* | |
ee12b630 MC |
2237 | * ext4_ext_calc_credits_for_single_extent: |
2238 | * This routine returns max. credits that needed to insert an extent | |
2239 | * to the extent tree. | |
2240 | * When pass the actual path, the caller should calculate credits | |
2241 | * under i_data_sem. | |
a86c6181 | 2242 | */ |
525f4ed8 | 2243 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2244 | struct ext4_ext_path *path) |
2245 | { | |
a86c6181 | 2246 | if (path) { |
ee12b630 | 2247 | int depth = ext_depth(inode); |
f3bd1f3f | 2248 | int ret = 0; |
ee12b630 | 2249 | |
a86c6181 | 2250 | /* probably there is space in leaf? */ |
a86c6181 | 2251 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2252 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2253 | |
ee12b630 MC |
2254 | /* |
2255 | * There are some space in the leaf tree, no | |
2256 | * need to account for leaf block credit | |
2257 | * | |
2258 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2259 | * and other metadata blocks still need to be |
ee12b630 MC |
2260 | * accounted. |
2261 | */ | |
525f4ed8 | 2262 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2263 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2264 | return ret; |
ee12b630 MC |
2265 | } |
2266 | } | |
a86c6181 | 2267 | |
525f4ed8 | 2268 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2269 | } |
a86c6181 | 2270 | |
ee12b630 MC |
2271 | /* |
2272 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2273 | * | |
2274 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2275 | * in the worse case, each tree level index/leaf need to be changed | |
2276 | * if the tree split due to insert a new extent, then the old tree | |
2277 | * index/leaf need to be updated too | |
2278 | * | |
2279 | * If the nrblocks are discontiguous, they could cause | |
2280 | * the whole tree split more than once, but this is really rare. | |
2281 | */ | |
525f4ed8 | 2282 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2283 | { |
2284 | int index; | |
f19d5870 TM |
2285 | int depth; |
2286 | ||
2287 | /* If we are converting the inline data, only one is needed here. */ | |
2288 | if (ext4_has_inline_data(inode)) | |
2289 | return 1; | |
2290 | ||
2291 | depth = ext_depth(inode); | |
a86c6181 | 2292 | |
ee12b630 MC |
2293 | if (chunk) |
2294 | index = depth * 2; | |
2295 | else | |
2296 | index = depth * 3; | |
a86c6181 | 2297 | |
ee12b630 | 2298 | return index; |
a86c6181 AT |
2299 | } |
2300 | ||
2301 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2302 | struct ext4_extent *ex, |
2303 | ext4_fsblk_t *partial_cluster, | |
2304 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2305 | { |
0aa06000 | 2306 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2307 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2308 | ext4_fsblk_t pblk; |
18888cf0 | 2309 | int flags = 0; |
a86c6181 | 2310 | |
c9de560d | 2311 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
18888cf0 AS |
2312 | flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2313 | else if (ext4_should_journal_data(inode)) | |
2314 | flags |= EXT4_FREE_BLOCKS_FORGET; | |
2315 | ||
0aa06000 TT |
2316 | /* |
2317 | * For bigalloc file systems, we never free a partial cluster | |
2318 | * at the beginning of the extent. Instead, we make a note | |
2319 | * that we tried freeing the cluster, and check to see if we | |
2320 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2321 | * or at the end of the ext4_truncate() operation. | |
2322 | */ | |
2323 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2324 | ||
d8990240 | 2325 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2326 | /* |
2327 | * If we have a partial cluster, and it's different from the | |
2328 | * cluster of the last block, we need to explicitly free the | |
2329 | * partial cluster here. | |
2330 | */ | |
2331 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2332 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2333 | ext4_free_blocks(handle, inode, NULL, | |
2334 | EXT4_C2B(sbi, *partial_cluster), | |
2335 | sbi->s_cluster_ratio, flags); | |
2336 | *partial_cluster = 0; | |
2337 | } | |
2338 | ||
a86c6181 AT |
2339 | #ifdef EXTENTS_STATS |
2340 | { | |
2341 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2342 | spin_lock(&sbi->s_ext_stats_lock); |
2343 | sbi->s_ext_blocks += ee_len; | |
2344 | sbi->s_ext_extents++; | |
2345 | if (ee_len < sbi->s_ext_min) | |
2346 | sbi->s_ext_min = ee_len; | |
2347 | if (ee_len > sbi->s_ext_max) | |
2348 | sbi->s_ext_max = ee_len; | |
2349 | if (ext_depth(inode) > sbi->s_depth_max) | |
2350 | sbi->s_depth_max = ext_depth(inode); | |
2351 | spin_unlock(&sbi->s_ext_stats_lock); | |
2352 | } | |
2353 | #endif | |
2354 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2355 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2356 | /* tail removal */ |
725d26d3 | 2357 | ext4_lblk_t num; |
725d26d3 | 2358 | |
a2df2a63 | 2359 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2360 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2361 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2362 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2363 | /* | |
2364 | * If the block range to be freed didn't start at the | |
2365 | * beginning of a cluster, and we removed the entire | |
2366 | * extent, save the partial cluster here, since we | |
2367 | * might need to delete if we determine that the | |
2368 | * truncate operation has removed all of the blocks in | |
2369 | * the cluster. | |
2370 | */ | |
2371 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2372 | (ee_len == num)) | |
2373 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2374 | else | |
2375 | *partial_cluster = 0; | |
a86c6181 | 2376 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2377 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2378 | /* head removal */ |
2379 | ext4_lblk_t num; | |
2380 | ext4_fsblk_t start; | |
2381 | ||
2382 | num = to - from; | |
2383 | start = ext4_ext_pblock(ex); | |
2384 | ||
2385 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
ee90d57e | 2386 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
d583fb87 | 2387 | |
a86c6181 | 2388 | } else { |
725d26d3 AK |
2389 | printk(KERN_INFO "strange request: removal(2) " |
2390 | "%u-%u from %u:%u\n", | |
2391 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2392 | } |
2393 | return 0; | |
2394 | } | |
2395 | ||
d583fb87 AH |
2396 | |
2397 | /* | |
2398 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2399 | * blocks appearing between "start" and "end", and splits the extents | |
2400 | * if "start" and "end" appear in the same extent | |
2401 | * | |
2402 | * @handle: The journal handle | |
2403 | * @inode: The files inode | |
2404 | * @path: The path to the leaf | |
2405 | * @start: The first block to remove | |
2406 | * @end: The last block to remove | |
2407 | */ | |
a86c6181 AT |
2408 | static int |
2409 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2410 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2411 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2412 | { |
0aa06000 | 2413 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2414 | int err = 0, correct_index = 0; |
2415 | int depth = ext_depth(inode), credits; | |
2416 | struct ext4_extent_header *eh; | |
750c9c47 | 2417 | ext4_lblk_t a, b; |
725d26d3 AK |
2418 | unsigned num; |
2419 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2420 | unsigned short ex_ee_len; |
a2df2a63 | 2421 | unsigned uninitialized = 0; |
a86c6181 AT |
2422 | struct ext4_extent *ex; |
2423 | ||
c29c0ae7 | 2424 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2425 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2426 | if (!path[depth].p_hdr) |
2427 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2428 | eh = path[depth].p_hdr; | |
273df556 FM |
2429 | if (unlikely(path[depth].p_hdr == NULL)) { |
2430 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2431 | return -EIO; | |
2432 | } | |
a86c6181 AT |
2433 | /* find where to start removing */ |
2434 | ex = EXT_LAST_EXTENT(eh); | |
2435 | ||
2436 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2437 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2438 | |
d8990240 AK |
2439 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2440 | ||
a86c6181 AT |
2441 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2442 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2443 | |
2444 | if (ext4_ext_is_uninitialized(ex)) | |
2445 | uninitialized = 1; | |
2446 | else | |
2447 | uninitialized = 0; | |
2448 | ||
553f9008 M |
2449 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2450 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2451 | path[depth].p_ext = ex; |
2452 | ||
2453 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2454 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2455 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2456 | |
2457 | ext_debug(" border %u:%u\n", a, b); | |
2458 | ||
d583fb87 | 2459 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2460 | if (end < ex_ee_block) { |
d583fb87 AH |
2461 | ex--; |
2462 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2463 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2464 | continue; | |
750c9c47 | 2465 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2466 | EXT4_ERROR_INODE(inode, |
2467 | "can not handle truncate %u:%u " | |
2468 | "on extent %u:%u", | |
2469 | start, end, ex_ee_block, | |
2470 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2471 | err = -EIO; |
2472 | goto out; | |
a86c6181 AT |
2473 | } else if (a != ex_ee_block) { |
2474 | /* remove tail of the extent */ | |
750c9c47 | 2475 | num = a - ex_ee_block; |
a86c6181 AT |
2476 | } else { |
2477 | /* remove whole extent: excellent! */ | |
a86c6181 | 2478 | num = 0; |
a86c6181 | 2479 | } |
34071da7 TT |
2480 | /* |
2481 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2482 | * descriptor) for each block group; assume two block | |
2483 | * groups plus ex_ee_len/blocks_per_block_group for | |
2484 | * the worst case | |
2485 | */ | |
2486 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2487 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2488 | correct_index = 1; | |
2489 | credits += (ext_depth(inode)) + 1; | |
2490 | } | |
5aca07eb | 2491 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2492 | |
487caeef | 2493 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2494 | if (err) |
a86c6181 | 2495 | goto out; |
a86c6181 AT |
2496 | |
2497 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2498 | if (err) | |
2499 | goto out; | |
2500 | ||
0aa06000 TT |
2501 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2502 | a, b); | |
a86c6181 AT |
2503 | if (err) |
2504 | goto out; | |
2505 | ||
750c9c47 | 2506 | if (num == 0) |
d0d856e8 | 2507 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2508 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2509 | |
a86c6181 | 2510 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2511 | /* |
2512 | * Do not mark uninitialized if all the blocks in the | |
2513 | * extent have been removed. | |
2514 | */ | |
2515 | if (uninitialized && num) | |
a2df2a63 | 2516 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2517 | /* |
2518 | * If the extent was completely released, | |
2519 | * we need to remove it from the leaf | |
2520 | */ | |
2521 | if (num == 0) { | |
f17722f9 | 2522 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2523 | /* |
2524 | * For hole punching, we need to scoot all the | |
2525 | * extents up when an extent is removed so that | |
2526 | * we dont have blank extents in the middle | |
2527 | */ | |
2528 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2529 | sizeof(struct ext4_extent)); | |
2530 | ||
2531 | /* Now get rid of the one at the end */ | |
2532 | memset(EXT_LAST_EXTENT(eh), 0, | |
2533 | sizeof(struct ext4_extent)); | |
2534 | } | |
2535 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2536 | } else |
2537 | *partial_cluster = 0; | |
d583fb87 | 2538 | |
750c9c47 DM |
2539 | err = ext4_ext_dirty(handle, inode, path + depth); |
2540 | if (err) | |
2541 | goto out; | |
2542 | ||
bf52c6f7 | 2543 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2544 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2545 | ex--; |
2546 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2547 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2548 | } |
2549 | ||
2550 | if (correct_index && eh->eh_entries) | |
2551 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2552 | ||
0aa06000 TT |
2553 | /* |
2554 | * If there is still a entry in the leaf node, check to see if | |
2555 | * it references the partial cluster. This is the only place | |
2556 | * where it could; if it doesn't, we can free the cluster. | |
2557 | */ | |
2558 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2559 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2560 | *partial_cluster)) { | |
2561 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2562 | ||
2563 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2564 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2565 | ||
2566 | ext4_free_blocks(handle, inode, NULL, | |
2567 | EXT4_C2B(sbi, *partial_cluster), | |
2568 | sbi->s_cluster_ratio, flags); | |
2569 | *partial_cluster = 0; | |
2570 | } | |
2571 | ||
a86c6181 AT |
2572 | /* if this leaf is free, then we should |
2573 | * remove it from index block above */ | |
2574 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2575 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2576 | |
2577 | out: | |
2578 | return err; | |
2579 | } | |
2580 | ||
2581 | /* | |
d0d856e8 RD |
2582 | * ext4_ext_more_to_rm: |
2583 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2584 | */ |
09b88252 | 2585 | static int |
a86c6181 AT |
2586 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2587 | { | |
2588 | BUG_ON(path->p_idx == NULL); | |
2589 | ||
2590 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2591 | return 0; | |
2592 | ||
2593 | /* | |
d0d856e8 | 2594 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2595 | * so we have to consider current index for truncation |
2596 | */ | |
2597 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2598 | return 0; | |
2599 | return 1; | |
2600 | } | |
2601 | ||
5f95d21f LC |
2602 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2603 | ext4_lblk_t end) | |
a86c6181 AT |
2604 | { |
2605 | struct super_block *sb = inode->i_sb; | |
2606 | int depth = ext_depth(inode); | |
968dee77 | 2607 | struct ext4_ext_path *path = NULL; |
0aa06000 | 2608 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2609 | handle_t *handle; |
6f2080e6 | 2610 | int i = 0, err = 0; |
a86c6181 | 2611 | |
5f95d21f | 2612 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2613 | |
2614 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2615 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2616 | if (IS_ERR(handle)) |
2617 | return PTR_ERR(handle); | |
2618 | ||
0617b83f | 2619 | again: |
d8990240 AK |
2620 | trace_ext4_ext_remove_space(inode, start, depth); |
2621 | ||
5f95d21f LC |
2622 | /* |
2623 | * Check if we are removing extents inside the extent tree. If that | |
2624 | * is the case, we are going to punch a hole inside the extent tree | |
2625 | * so we have to check whether we need to split the extent covering | |
2626 | * the last block to remove so we can easily remove the part of it | |
2627 | * in ext4_ext_rm_leaf(). | |
2628 | */ | |
2629 | if (end < EXT_MAX_BLOCKS - 1) { | |
2630 | struct ext4_extent *ex; | |
2631 | ext4_lblk_t ee_block; | |
2632 | ||
2633 | /* find extent for this block */ | |
2634 | path = ext4_ext_find_extent(inode, end, NULL); | |
2635 | if (IS_ERR(path)) { | |
2636 | ext4_journal_stop(handle); | |
2637 | return PTR_ERR(path); | |
2638 | } | |
2639 | depth = ext_depth(inode); | |
6f2080e6 | 2640 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2641 | ex = path[depth].p_ext; |
968dee77 | 2642 | if (!ex) { |
6f2080e6 DM |
2643 | if (depth) { |
2644 | EXT4_ERROR_INODE(inode, | |
2645 | "path[%d].p_hdr == NULL", | |
2646 | depth); | |
2647 | err = -EIO; | |
2648 | } | |
2649 | goto out; | |
968dee77 | 2650 | } |
5f95d21f LC |
2651 | |
2652 | ee_block = le32_to_cpu(ex->ee_block); | |
2653 | ||
2654 | /* | |
2655 | * See if the last block is inside the extent, if so split | |
2656 | * the extent at 'end' block so we can easily remove the | |
2657 | * tail of the first part of the split extent in | |
2658 | * ext4_ext_rm_leaf(). | |
2659 | */ | |
2660 | if (end >= ee_block && | |
2661 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2662 | int split_flag = 0; | |
2663 | ||
2664 | if (ext4_ext_is_uninitialized(ex)) | |
2665 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2666 | EXT4_EXT_MARK_UNINIT2; | |
2667 | ||
2668 | /* | |
2669 | * Split the extent in two so that 'end' is the last | |
2670 | * block in the first new extent | |
2671 | */ | |
2672 | err = ext4_split_extent_at(handle, inode, path, | |
2673 | end + 1, split_flag, | |
2674 | EXT4_GET_BLOCKS_PRE_IO | | |
2675 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
2676 | ||
2677 | if (err < 0) | |
2678 | goto out; | |
2679 | } | |
5f95d21f | 2680 | } |
a86c6181 | 2681 | /* |
d0d856e8 RD |
2682 | * We start scanning from right side, freeing all the blocks |
2683 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2684 | */ |
0617b83f | 2685 | depth = ext_depth(inode); |
968dee77 AS |
2686 | if (path) { |
2687 | int k = i = depth; | |
2688 | while (--k > 0) | |
2689 | path[k].p_block = | |
2690 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2691 | } else { | |
2692 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2693 | GFP_NOFS); | |
2694 | if (path == NULL) { | |
2695 | ext4_journal_stop(handle); | |
2696 | return -ENOMEM; | |
2697 | } | |
2698 | path[0].p_depth = depth; | |
2699 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2700 | i = 0; |
5f95d21f | 2701 | |
968dee77 AS |
2702 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
2703 | err = -EIO; | |
2704 | goto out; | |
2705 | } | |
a86c6181 | 2706 | } |
968dee77 | 2707 | err = 0; |
a86c6181 AT |
2708 | |
2709 | while (i >= 0 && err == 0) { | |
2710 | if (i == depth) { | |
2711 | /* this is leaf block */ | |
d583fb87 | 2712 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2713 | &partial_cluster, start, |
5f95d21f | 2714 | end); |
d0d856e8 | 2715 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2716 | brelse(path[i].p_bh); |
2717 | path[i].p_bh = NULL; | |
2718 | i--; | |
2719 | continue; | |
2720 | } | |
2721 | ||
2722 | /* this is index block */ | |
2723 | if (!path[i].p_hdr) { | |
2724 | ext_debug("initialize header\n"); | |
2725 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2726 | } |
2727 | ||
a86c6181 | 2728 | if (!path[i].p_idx) { |
d0d856e8 | 2729 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2730 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2731 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2732 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2733 | path[i].p_hdr, | |
2734 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2735 | } else { | |
d0d856e8 | 2736 | /* we were already here, see at next index */ |
a86c6181 AT |
2737 | path[i].p_idx--; |
2738 | } | |
2739 | ||
2740 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2741 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2742 | path[i].p_idx); | |
2743 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2744 | struct buffer_head *bh; |
a86c6181 | 2745 | /* go to the next level */ |
2ae02107 | 2746 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2747 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2748 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2749 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2750 | if (!bh) { |
a86c6181 AT |
2751 | /* should we reset i_size? */ |
2752 | err = -EIO; | |
2753 | break; | |
2754 | } | |
c29c0ae7 AT |
2755 | if (WARN_ON(i + 1 > depth)) { |
2756 | err = -EIO; | |
2757 | break; | |
2758 | } | |
f8489128 DW |
2759 | if (ext4_ext_check_block(inode, ext_block_hdr(bh), |
2760 | depth - i - 1, bh)) { | |
c29c0ae7 AT |
2761 | err = -EIO; |
2762 | break; | |
2763 | } | |
2764 | path[i + 1].p_bh = bh; | |
a86c6181 | 2765 | |
d0d856e8 RD |
2766 | /* save actual number of indexes since this |
2767 | * number is changed at the next iteration */ | |
a86c6181 AT |
2768 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2769 | i++; | |
2770 | } else { | |
d0d856e8 | 2771 | /* we finished processing this index, go up */ |
a86c6181 | 2772 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2773 | /* index is empty, remove it; |
a86c6181 AT |
2774 | * handle must be already prepared by the |
2775 | * truncatei_leaf() */ | |
c36575e6 | 2776 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2777 | } |
d0d856e8 | 2778 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2779 | brelse(path[i].p_bh); |
2780 | path[i].p_bh = NULL; | |
2781 | i--; | |
2782 | ext_debug("return to level %d\n", i); | |
2783 | } | |
2784 | } | |
2785 | ||
d8990240 AK |
2786 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2787 | path->p_hdr->eh_entries); | |
2788 | ||
7b415bf6 AK |
2789 | /* If we still have something in the partial cluster and we have removed |
2790 | * even the first extent, then we should free the blocks in the partial | |
2791 | * cluster as well. */ | |
2792 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2793 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2794 | ||
2795 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2796 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2797 | ||
2798 | ext4_free_blocks(handle, inode, NULL, | |
2799 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2800 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2801 | partial_cluster = 0; | |
2802 | } | |
2803 | ||
a86c6181 AT |
2804 | /* TODO: flexible tree reduction should be here */ |
2805 | if (path->p_hdr->eh_entries == 0) { | |
2806 | /* | |
d0d856e8 RD |
2807 | * truncate to zero freed all the tree, |
2808 | * so we need to correct eh_depth | |
a86c6181 AT |
2809 | */ |
2810 | err = ext4_ext_get_access(handle, inode, path); | |
2811 | if (err == 0) { | |
2812 | ext_inode_hdr(inode)->eh_depth = 0; | |
2813 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2814 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2815 | err = ext4_ext_dirty(handle, inode, path); |
2816 | } | |
2817 | } | |
2818 | out: | |
a86c6181 AT |
2819 | ext4_ext_drop_refs(path); |
2820 | kfree(path); | |
968dee77 AS |
2821 | if (err == -EAGAIN) { |
2822 | path = NULL; | |
0617b83f | 2823 | goto again; |
968dee77 | 2824 | } |
a86c6181 AT |
2825 | ext4_journal_stop(handle); |
2826 | ||
2827 | return err; | |
2828 | } | |
2829 | ||
2830 | /* | |
2831 | * called at mount time | |
2832 | */ | |
2833 | void ext4_ext_init(struct super_block *sb) | |
2834 | { | |
2835 | /* | |
2836 | * possible initialization would be here | |
2837 | */ | |
2838 | ||
83982b6f | 2839 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2840 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2841 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2842 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2843 | ", aggressive tests" |
a86c6181 AT |
2844 | #endif |
2845 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2846 | ", check binsearch" |
a86c6181 AT |
2847 | #endif |
2848 | #ifdef EXTENTS_STATS | |
92b97816 | 2849 | ", stats" |
a86c6181 | 2850 | #endif |
92b97816 | 2851 | "\n"); |
90576c0b | 2852 | #endif |
a86c6181 AT |
2853 | #ifdef EXTENTS_STATS |
2854 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2855 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2856 | EXT4_SB(sb)->s_ext_max = 0; | |
2857 | #endif | |
2858 | } | |
2859 | } | |
2860 | ||
2861 | /* | |
2862 | * called at umount time | |
2863 | */ | |
2864 | void ext4_ext_release(struct super_block *sb) | |
2865 | { | |
83982b6f | 2866 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2867 | return; |
2868 | ||
2869 | #ifdef EXTENTS_STATS | |
2870 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2871 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2872 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2873 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2874 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2875 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2876 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2877 | } | |
2878 | #endif | |
2879 | } | |
2880 | ||
093a088b AK |
2881 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2882 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2883 | { | |
2407518d LC |
2884 | ext4_fsblk_t ee_pblock; |
2885 | unsigned int ee_len; | |
b720303d | 2886 | int ret; |
093a088b | 2887 | |
093a088b | 2888 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2889 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2890 | |
a107e5a3 | 2891 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2892 | if (ret > 0) |
2893 | ret = 0; | |
093a088b | 2894 | |
2407518d | 2895 | return ret; |
093a088b AK |
2896 | } |
2897 | ||
47ea3bb5 YY |
2898 | /* |
2899 | * ext4_split_extent_at() splits an extent at given block. | |
2900 | * | |
2901 | * @handle: the journal handle | |
2902 | * @inode: the file inode | |
2903 | * @path: the path to the extent | |
2904 | * @split: the logical block where the extent is splitted. | |
2905 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2906 | * the states(init or uninit) of new extents. | |
2907 | * @flags: flags used to insert new extent to extent tree. | |
2908 | * | |
2909 | * | |
2910 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2911 | * of which are deterimined by split_flag. | |
2912 | * | |
2913 | * There are two cases: | |
2914 | * a> the extent are splitted into two extent. | |
2915 | * b> split is not needed, and just mark the extent. | |
2916 | * | |
2917 | * return 0 on success. | |
2918 | */ | |
2919 | static int ext4_split_extent_at(handle_t *handle, | |
2920 | struct inode *inode, | |
2921 | struct ext4_ext_path *path, | |
2922 | ext4_lblk_t split, | |
2923 | int split_flag, | |
2924 | int flags) | |
2925 | { | |
2926 | ext4_fsblk_t newblock; | |
2927 | ext4_lblk_t ee_block; | |
adb23551 | 2928 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
2929 | struct ext4_extent *ex2 = NULL; |
2930 | unsigned int ee_len, depth; | |
2931 | int err = 0; | |
2932 | ||
dee1f973 DM |
2933 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
2934 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
2935 | ||
47ea3bb5 YY |
2936 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
2937 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
2938 | ||
2939 | ext4_ext_show_leaf(inode, path); | |
2940 | ||
2941 | depth = ext_depth(inode); | |
2942 | ex = path[depth].p_ext; | |
2943 | ee_block = le32_to_cpu(ex->ee_block); | |
2944 | ee_len = ext4_ext_get_actual_len(ex); | |
2945 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
2946 | ||
2947 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
357b66fd DM |
2948 | BUG_ON(!ext4_ext_is_uninitialized(ex) && |
2949 | split_flag & (EXT4_EXT_MAY_ZEROOUT | | |
2950 | EXT4_EXT_MARK_UNINIT1 | | |
2951 | EXT4_EXT_MARK_UNINIT2)); | |
47ea3bb5 YY |
2952 | |
2953 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2954 | if (err) | |
2955 | goto out; | |
2956 | ||
2957 | if (split == ee_block) { | |
2958 | /* | |
2959 | * case b: block @split is the block that the extent begins with | |
2960 | * then we just change the state of the extent, and splitting | |
2961 | * is not needed. | |
2962 | */ | |
2963 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2964 | ext4_ext_mark_uninitialized(ex); | |
2965 | else | |
2966 | ext4_ext_mark_initialized(ex); | |
2967 | ||
2968 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 2969 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 2970 | |
ecb94f5f | 2971 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
2972 | goto out; |
2973 | } | |
2974 | ||
2975 | /* case a */ | |
2976 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
2977 | ex->ee_len = cpu_to_le16(split - ee_block); | |
2978 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
2979 | ext4_ext_mark_uninitialized(ex); | |
2980 | ||
2981 | /* | |
2982 | * path may lead to new leaf, not to original leaf any more | |
2983 | * after ext4_ext_insert_extent() returns, | |
2984 | */ | |
2985 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2986 | if (err) | |
2987 | goto fix_extent_len; | |
2988 | ||
2989 | ex2 = &newex; | |
2990 | ex2->ee_block = cpu_to_le32(split); | |
2991 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
2992 | ext4_ext_store_pblock(ex2, newblock); | |
2993 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
2994 | ext4_ext_mark_uninitialized(ex2); | |
2995 | ||
2996 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
2997 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 | 2998 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 2999 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3000 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 ZL |
3001 | zero_ex.ee_block = ex2->ee_block; |
3002 | zero_ex.ee_len = ext4_ext_get_actual_len(ex2); | |
3003 | ext4_ext_store_pblock(&zero_ex, | |
3004 | ext4_ext_pblock(ex2)); | |
3005 | } else { | |
dee1f973 | 3006 | err = ext4_ext_zeroout(inode, ex); |
adb23551 ZL |
3007 | zero_ex.ee_block = ex->ee_block; |
3008 | zero_ex.ee_len = ext4_ext_get_actual_len(ex); | |
3009 | ext4_ext_store_pblock(&zero_ex, | |
3010 | ext4_ext_pblock(ex)); | |
3011 | } | |
3012 | } else { | |
dee1f973 | 3013 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 ZL |
3014 | zero_ex.ee_block = orig_ex.ee_block; |
3015 | zero_ex.ee_len = ext4_ext_get_actual_len(&orig_ex); | |
3016 | ext4_ext_store_pblock(&zero_ex, | |
3017 | ext4_ext_pblock(&orig_ex)); | |
3018 | } | |
dee1f973 | 3019 | |
47ea3bb5 YY |
3020 | if (err) |
3021 | goto fix_extent_len; | |
3022 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3023 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3024 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3025 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3026 | if (err) |
3027 | goto fix_extent_len; | |
3028 | ||
3029 | /* update extent status tree */ | |
3030 | err = ext4_es_zeroout(inode, &zero_ex); | |
3031 | ||
47ea3bb5 YY |
3032 | goto out; |
3033 | } else if (err) | |
3034 | goto fix_extent_len; | |
3035 | ||
3036 | out: | |
3037 | ext4_ext_show_leaf(inode, path); | |
3038 | return err; | |
3039 | ||
3040 | fix_extent_len: | |
3041 | ex->ee_len = orig_ex.ee_len; | |
3042 | ext4_ext_dirty(handle, inode, path + depth); | |
3043 | return err; | |
3044 | } | |
3045 | ||
3046 | /* | |
3047 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3048 | * by @map as split_flags indicates | |
3049 | * | |
3050 | * It may result in splitting the extent into multiple extents (upto three) | |
3051 | * There are three possibilities: | |
3052 | * a> There is no split required | |
3053 | * b> Splits in two extents: Split is happening at either end of the extent | |
3054 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3055 | * | |
3056 | */ | |
3057 | static int ext4_split_extent(handle_t *handle, | |
3058 | struct inode *inode, | |
3059 | struct ext4_ext_path *path, | |
3060 | struct ext4_map_blocks *map, | |
3061 | int split_flag, | |
3062 | int flags) | |
3063 | { | |
3064 | ext4_lblk_t ee_block; | |
3065 | struct ext4_extent *ex; | |
3066 | unsigned int ee_len, depth; | |
3067 | int err = 0; | |
3068 | int uninitialized; | |
3069 | int split_flag1, flags1; | |
3a225670 | 3070 | int allocated = map->m_len; |
47ea3bb5 YY |
3071 | |
3072 | depth = ext_depth(inode); | |
3073 | ex = path[depth].p_ext; | |
3074 | ee_block = le32_to_cpu(ex->ee_block); | |
3075 | ee_len = ext4_ext_get_actual_len(ex); | |
3076 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3077 | ||
3078 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3079 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3080 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3081 | if (uninitialized) | |
3082 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3083 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3084 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3085 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3086 | err = ext4_split_extent_at(handle, inode, path, |
3087 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3088 | if (err) |
3089 | goto out; | |
3a225670 ZL |
3090 | } else { |
3091 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3092 | } |
357b66fd DM |
3093 | /* |
3094 | * Update path is required because previous ext4_split_extent_at() may | |
3095 | * result in split of original leaf or extent zeroout. | |
3096 | */ | |
47ea3bb5 YY |
3097 | ext4_ext_drop_refs(path); |
3098 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3099 | if (IS_ERR(path)) | |
3100 | return PTR_ERR(path); | |
357b66fd DM |
3101 | depth = ext_depth(inode); |
3102 | ex = path[depth].p_ext; | |
3103 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3104 | split_flag1 = 0; | |
47ea3bb5 YY |
3105 | |
3106 | if (map->m_lblk >= ee_block) { | |
357b66fd DM |
3107 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
3108 | if (uninitialized) { | |
47ea3bb5 | 3109 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; |
357b66fd DM |
3110 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3111 | EXT4_EXT_MARK_UNINIT2); | |
3112 | } | |
47ea3bb5 YY |
3113 | err = ext4_split_extent_at(handle, inode, path, |
3114 | map->m_lblk, split_flag1, flags); | |
3115 | if (err) | |
3116 | goto out; | |
3117 | } | |
3118 | ||
3119 | ext4_ext_show_leaf(inode, path); | |
3120 | out: | |
3a225670 | 3121 | return err ? err : allocated; |
47ea3bb5 YY |
3122 | } |
3123 | ||
56055d3a | 3124 | /* |
e35fd660 | 3125 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3126 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3127 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3128 | * uninitialized). |
3129 | * There are three possibilities: | |
3130 | * a> There is no split required: Entire extent should be initialized | |
3131 | * b> Splits in two extents: Write is happening at either end of the extent | |
3132 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3133 | * |
3134 | * Pre-conditions: | |
3135 | * - The extent pointed to by 'path' is uninitialized. | |
3136 | * - The extent pointed to by 'path' contains a superset | |
3137 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3138 | * | |
3139 | * Post-conditions on success: | |
3140 | * - the returned value is the number of blocks beyond map->l_lblk | |
3141 | * that are allocated and initialized. | |
3142 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3143 | */ |
725d26d3 | 3144 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3145 | struct inode *inode, |
3146 | struct ext4_map_blocks *map, | |
3147 | struct ext4_ext_path *path) | |
56055d3a | 3148 | { |
67a5da56 | 3149 | struct ext4_sb_info *sbi; |
6f91bc5f | 3150 | struct ext4_extent_header *eh; |
667eff35 YY |
3151 | struct ext4_map_blocks split_map; |
3152 | struct ext4_extent zero_ex; | |
3153 | struct ext4_extent *ex; | |
21ca087a | 3154 | ext4_lblk_t ee_block, eof_block; |
f85b287a | 3155 | unsigned int ee_len, depth; |
67a5da56 | 3156 | int allocated, max_zeroout = 0; |
56055d3a | 3157 | int err = 0; |
667eff35 | 3158 | int split_flag = 0; |
21ca087a DM |
3159 | |
3160 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3161 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3162 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a | 3163 | |
67a5da56 | 3164 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3165 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3166 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3167 | if (eof_block < map->m_lblk + map->m_len) |
3168 | eof_block = map->m_lblk + map->m_len; | |
56055d3a AA |
3169 | |
3170 | depth = ext_depth(inode); | |
6f91bc5f | 3171 | eh = path[depth].p_hdr; |
56055d3a AA |
3172 | ex = path[depth].p_ext; |
3173 | ee_block = le32_to_cpu(ex->ee_block); | |
3174 | ee_len = ext4_ext_get_actual_len(ex); | |
e35fd660 | 3175 | allocated = ee_len - (map->m_lblk - ee_block); |
adb23551 | 3176 | zero_ex.ee_len = 0; |
56055d3a | 3177 | |
6f91bc5f EG |
3178 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3179 | ||
3180 | /* Pre-conditions */ | |
3181 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3182 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3183 | |
3184 | /* | |
3185 | * Attempt to transfer newly initialized blocks from the currently | |
3186 | * uninitialized extent to its left neighbor. This is much cheaper | |
3187 | * than an insertion followed by a merge as those involve costly | |
3188 | * memmove() calls. This is the common case in steady state for | |
3189 | * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append | |
3190 | * writes. | |
3191 | * | |
3192 | * Limitations of the current logic: | |
3193 | * - L1: we only deal with writes at the start of the extent. | |
3194 | * The approach could be extended to writes at the end | |
3195 | * of the extent but this scenario was deemed less common. | |
3196 | * - L2: we do not deal with writes covering the whole extent. | |
3197 | * This would require removing the extent if the transfer | |
3198 | * is possible. | |
3199 | * - L3: we only attempt to merge with an extent stored in the | |
3200 | * same extent tree node. | |
3201 | */ | |
3202 | if ((map->m_lblk == ee_block) && /*L1*/ | |
3203 | (map->m_len < ee_len) && /*L2*/ | |
3204 | (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/ | |
3205 | struct ext4_extent *prev_ex; | |
3206 | ext4_lblk_t prev_lblk; | |
3207 | ext4_fsblk_t prev_pblk, ee_pblk; | |
3208 | unsigned int prev_len, write_len; | |
3209 | ||
3210 | prev_ex = ex - 1; | |
3211 | prev_lblk = le32_to_cpu(prev_ex->ee_block); | |
3212 | prev_len = ext4_ext_get_actual_len(prev_ex); | |
3213 | prev_pblk = ext4_ext_pblock(prev_ex); | |
3214 | ee_pblk = ext4_ext_pblock(ex); | |
3215 | write_len = map->m_len; | |
3216 | ||
3217 | /* | |
3218 | * A transfer of blocks from 'ex' to 'prev_ex' is allowed | |
3219 | * upon those conditions: | |
3220 | * - C1: prev_ex is initialized, | |
3221 | * - C2: prev_ex is logically abutting ex, | |
3222 | * - C3: prev_ex is physically abutting ex, | |
3223 | * - C4: prev_ex can receive the additional blocks without | |
3224 | * overflowing the (initialized) length limit. | |
3225 | */ | |
3226 | if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/ | |
3227 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ | |
3228 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
3229 | (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/ | |
3230 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3231 | if (err) | |
3232 | goto out; | |
3233 | ||
3234 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3235 | map, ex, prev_ex); | |
3236 | ||
3237 | /* Shift the start of ex by 'write_len' blocks */ | |
3238 | ex->ee_block = cpu_to_le32(ee_block + write_len); | |
3239 | ext4_ext_store_pblock(ex, ee_pblk + write_len); | |
3240 | ex->ee_len = cpu_to_le16(ee_len - write_len); | |
3241 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3242 | ||
3243 | /* Extend prev_ex by 'write_len' blocks */ | |
3244 | prev_ex->ee_len = cpu_to_le16(prev_len + write_len); | |
3245 | ||
3246 | /* Mark the block containing both extents as dirty */ | |
3247 | ext4_ext_dirty(handle, inode, path + depth); | |
3248 | ||
3249 | /* Update path to point to the right extent */ | |
3250 | path[depth].p_ext = prev_ex; | |
3251 | ||
3252 | /* Result: number of initialized blocks past m_lblk */ | |
3253 | allocated = write_len; | |
3254 | goto out; | |
3255 | } | |
3256 | } | |
3257 | ||
667eff35 | 3258 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3259 | /* |
3260 | * It is safe to convert extent to initialized via explicit | |
3261 | * zeroout only if extent is fully insde i_size or new_size. | |
3262 | */ | |
667eff35 | 3263 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3264 | |
67a5da56 ZL |
3265 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3266 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
3267 | inode->i_sb->s_blocksize_bits; | |
3268 | ||
3269 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3270 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3271 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3272 | if (err) |
d03856bd | 3273 | goto out; |
adb23551 ZL |
3274 | zero_ex.ee_block = ex->ee_block; |
3275 | zero_ex.ee_len = ext4_ext_get_actual_len(ex); | |
3276 | ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex)); | |
d03856bd AK |
3277 | |
3278 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3279 | if (err) | |
3280 | goto out; | |
667eff35 | 3281 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3282 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3283 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3284 | goto out; |
56055d3a | 3285 | } |
667eff35 | 3286 | |
56055d3a | 3287 | /* |
667eff35 YY |
3288 | * four cases: |
3289 | * 1. split the extent into three extents. | |
3290 | * 2. split the extent into two extents, zeroout the first half. | |
3291 | * 3. split the extent into two extents, zeroout the second half. | |
3292 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3293 | */ |
667eff35 YY |
3294 | split_map.m_lblk = map->m_lblk; |
3295 | split_map.m_len = map->m_len; | |
3296 | ||
67a5da56 ZL |
3297 | if (max_zeroout && (allocated > map->m_len)) { |
3298 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3299 | /* case 3 */ |
3300 | zero_ex.ee_block = | |
9b940f8e AH |
3301 | cpu_to_le32(map->m_lblk); |
3302 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3303 | ext4_ext_store_pblock(&zero_ex, |
3304 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3305 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3306 | if (err) |
3307 | goto out; | |
667eff35 YY |
3308 | split_map.m_lblk = map->m_lblk; |
3309 | split_map.m_len = allocated; | |
67a5da56 | 3310 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3311 | /* case 2 */ |
3312 | if (map->m_lblk != ee_block) { | |
3313 | zero_ex.ee_block = ex->ee_block; | |
3314 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3315 | ee_block); | |
3316 | ext4_ext_store_pblock(&zero_ex, | |
3317 | ext4_ext_pblock(ex)); | |
3318 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3319 | if (err) | |
3320 | goto out; | |
3321 | } | |
3322 | ||
667eff35 | 3323 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3324 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3325 | allocated = map->m_len; | |
56055d3a AA |
3326 | } |
3327 | } | |
667eff35 YY |
3328 | |
3329 | allocated = ext4_split_extent(handle, inode, path, | |
67a5da56 | 3330 | &split_map, split_flag, 0); |
667eff35 YY |
3331 | if (allocated < 0) |
3332 | err = allocated; | |
3333 | ||
56055d3a | 3334 | out: |
adb23551 ZL |
3335 | /* If we have gotten a failure, don't zero out status tree */ |
3336 | if (!err) | |
3337 | err = ext4_es_zeroout(inode, &zero_ex); | |
56055d3a AA |
3338 | return err ? err : allocated; |
3339 | } | |
3340 | ||
0031462b | 3341 | /* |
e35fd660 | 3342 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3343 | * ext4_get_blocks_dio_write() when DIO to write |
3344 | * to an uninitialized extent. | |
3345 | * | |
fd018fe8 | 3346 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3347 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3348 | * There are three possibilities: |
3349 | * a> There is no split required: Entire extent should be uninitialized | |
3350 | * b> Splits in two extents: Write is happening at either end of the extent | |
3351 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3352 | * | |
3353 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3354 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3355 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3356 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3357 | * into three uninitialized extent(at most). After IO complete, the part |
3358 | * being filled will be convert to initialized by the end_io callback function | |
3359 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3360 | * |
3361 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3362 | */ |
3363 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3364 | struct inode *inode, | |
e35fd660 | 3365 | struct ext4_map_blocks *map, |
0031462b | 3366 | struct ext4_ext_path *path, |
0031462b MC |
3367 | int flags) |
3368 | { | |
667eff35 YY |
3369 | ext4_lblk_t eof_block; |
3370 | ext4_lblk_t ee_block; | |
3371 | struct ext4_extent *ex; | |
3372 | unsigned int ee_len; | |
3373 | int split_flag = 0, depth; | |
21ca087a DM |
3374 | |
3375 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3376 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3377 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3378 | |
3379 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3380 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3381 | if (eof_block < map->m_lblk + map->m_len) |
3382 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3383 | /* |
3384 | * It is safe to convert extent to initialized via explicit | |
3385 | * zeroout only if extent is fully insde i_size or new_size. | |
3386 | */ | |
667eff35 YY |
3387 | depth = ext_depth(inode); |
3388 | ex = path[depth].p_ext; | |
3389 | ee_block = le32_to_cpu(ex->ee_block); | |
3390 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3391 | |
667eff35 YY |
3392 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3393 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3394 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3395 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3396 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3397 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3398 | } |
197217a5 | 3399 | |
c7064ef1 | 3400 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3401 | struct inode *inode, |
3402 | struct ext4_map_blocks *map, | |
3403 | struct ext4_ext_path *path) | |
0031462b MC |
3404 | { |
3405 | struct ext4_extent *ex; | |
dee1f973 DM |
3406 | ext4_lblk_t ee_block; |
3407 | unsigned int ee_len; | |
0031462b MC |
3408 | int depth; |
3409 | int err = 0; | |
0031462b MC |
3410 | |
3411 | depth = ext_depth(inode); | |
0031462b | 3412 | ex = path[depth].p_ext; |
dee1f973 DM |
3413 | ee_block = le32_to_cpu(ex->ee_block); |
3414 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3415 | |
197217a5 YY |
3416 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3417 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3418 | (unsigned long long)ee_block, ee_len); |
3419 | ||
ff95ec22 DM |
3420 | /* If extent is larger than requested it is a clear sign that we still |
3421 | * have some extent state machine issues left. So extent_split is still | |
3422 | * required. | |
3423 | * TODO: Once all related issues will be fixed this situation should be | |
3424 | * illegal. | |
3425 | */ | |
dee1f973 | 3426 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ff95ec22 DM |
3427 | #ifdef EXT4_DEBUG |
3428 | ext4_warning("Inode (%ld) finished: extent logical block %llu," | |
3429 | " len %u; IO logical block %llu, len %u\n", | |
3430 | inode->i_ino, (unsigned long long)ee_block, ee_len, | |
3431 | (unsigned long long)map->m_lblk, map->m_len); | |
3432 | #endif | |
dee1f973 DM |
3433 | err = ext4_split_unwritten_extents(handle, inode, map, path, |
3434 | EXT4_GET_BLOCKS_CONVERT); | |
3435 | if (err < 0) | |
3436 | goto out; | |
3437 | ext4_ext_drop_refs(path); | |
3438 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3439 | if (IS_ERR(path)) { | |
3440 | err = PTR_ERR(path); | |
3441 | goto out; | |
3442 | } | |
3443 | depth = ext_depth(inode); | |
3444 | ex = path[depth].p_ext; | |
3445 | } | |
197217a5 | 3446 | |
0031462b MC |
3447 | err = ext4_ext_get_access(handle, inode, path + depth); |
3448 | if (err) | |
3449 | goto out; | |
3450 | /* first mark the extent as initialized */ | |
3451 | ext4_ext_mark_initialized(ex); | |
3452 | ||
197217a5 YY |
3453 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3454 | * borders are not changed | |
0031462b | 3455 | */ |
ecb94f5f | 3456 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3457 | |
0031462b | 3458 | /* Mark modified extent as dirty */ |
ecb94f5f | 3459 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3460 | out: |
3461 | ext4_ext_show_leaf(inode, path); | |
3462 | return err; | |
3463 | } | |
3464 | ||
515f41c3 AK |
3465 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3466 | sector_t block, int count) | |
3467 | { | |
3468 | int i; | |
3469 | for (i = 0; i < count; i++) | |
3470 | unmap_underlying_metadata(bdev, block + i); | |
3471 | } | |
3472 | ||
58590b06 TT |
3473 | /* |
3474 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3475 | */ | |
3476 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3477 | ext4_lblk_t lblk, |
58590b06 TT |
3478 | struct ext4_ext_path *path, |
3479 | unsigned int len) | |
3480 | { | |
3481 | int i, depth; | |
3482 | struct ext4_extent_header *eh; | |
65922cb5 | 3483 | struct ext4_extent *last_ex; |
58590b06 TT |
3484 | |
3485 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3486 | return 0; | |
3487 | ||
3488 | depth = ext_depth(inode); | |
3489 | eh = path[depth].p_hdr; | |
58590b06 | 3490 | |
afcff5d8 LC |
3491 | /* |
3492 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3493 | * do not care for this case anymore. Simply remove the flag | |
3494 | * if there are no extents. | |
3495 | */ | |
3496 | if (unlikely(!eh->eh_entries)) | |
3497 | goto out; | |
58590b06 TT |
3498 | last_ex = EXT_LAST_EXTENT(eh); |
3499 | /* | |
3500 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3501 | * last block in the last extent in the file. We test this by | |
3502 | * first checking to see if the caller to | |
3503 | * ext4_ext_get_blocks() was interested in the last block (or | |
3504 | * a block beyond the last block) in the current extent. If | |
3505 | * this turns out to be false, we can bail out from this | |
3506 | * function immediately. | |
3507 | */ | |
d002ebf1 | 3508 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3509 | ext4_ext_get_actual_len(last_ex)) |
3510 | return 0; | |
3511 | /* | |
3512 | * If the caller does appear to be planning to write at or | |
3513 | * beyond the end of the current extent, we then test to see | |
3514 | * if the current extent is the last extent in the file, by | |
3515 | * checking to make sure it was reached via the rightmost node | |
3516 | * at each level of the tree. | |
3517 | */ | |
3518 | for (i = depth-1; i >= 0; i--) | |
3519 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3520 | return 0; | |
afcff5d8 | 3521 | out: |
58590b06 TT |
3522 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3523 | return ext4_mark_inode_dirty(handle, inode); | |
3524 | } | |
3525 | ||
7b415bf6 AK |
3526 | /** |
3527 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3528 | * | |
7d1b1fbc | 3529 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3530 | */ |
f7fec032 ZL |
3531 | int ext4_find_delalloc_range(struct inode *inode, |
3532 | ext4_lblk_t lblk_start, | |
3533 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3534 | { |
7d1b1fbc | 3535 | struct extent_status es; |
7b415bf6 | 3536 | |
be401363 | 3537 | ext4_es_find_delayed_extent(inode, lblk_start, &es); |
06b0c886 | 3538 | if (es.es_len == 0) |
7d1b1fbc | 3539 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3540 | else if (es.es_lblk <= lblk_start && |
3541 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3542 | return 1; |
06b0c886 | 3543 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3544 | return 1; |
7b415bf6 | 3545 | else |
7d1b1fbc | 3546 | return 0; |
7b415bf6 AK |
3547 | } |
3548 | ||
7d1b1fbc | 3549 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3550 | { |
3551 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3552 | ext4_lblk_t lblk_start, lblk_end; | |
3553 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3554 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3555 | ||
7d1b1fbc | 3556 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3557 | } |
3558 | ||
3559 | /** | |
3560 | * Determines how many complete clusters (out of those specified by the 'map') | |
3561 | * are under delalloc and were reserved quota for. | |
3562 | * This function is called when we are writing out the blocks that were | |
3563 | * originally written with their allocation delayed, but then the space was | |
3564 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3565 | * The cases to look for are: | |
3566 | * ('=' indicated delayed allocated blocks | |
3567 | * '-' indicates non-delayed allocated blocks) | |
3568 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3569 | * are not delalloc'ed. | |
3570 | * Ex: | |
3571 | * |----c---=|====c====|====c====|===-c----| | |
3572 | * |++++++ allocated ++++++| | |
3573 | * ==> 4 complete clusters in above example | |
3574 | * | |
3575 | * (b) partial cluster (outside of allocated range) towards either end is | |
3576 | * marked for delayed allocation. In this case, we will exclude that | |
3577 | * cluster. | |
3578 | * Ex: | |
3579 | * |----====c========|========c========| | |
3580 | * |++++++ allocated ++++++| | |
3581 | * ==> 1 complete clusters in above example | |
3582 | * | |
3583 | * Ex: | |
3584 | * |================c================| | |
3585 | * |++++++ allocated ++++++| | |
3586 | * ==> 0 complete clusters in above example | |
3587 | * | |
3588 | * The ext4_da_update_reserve_space will be called only if we | |
3589 | * determine here that there were some "entire" clusters that span | |
3590 | * this 'allocated' range. | |
3591 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3592 | * without ever calling ext4_find_delalloc_range. | |
3593 | */ | |
3594 | static unsigned int | |
3595 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3596 | unsigned int num_blks) | |
3597 | { | |
3598 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3599 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3600 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3601 | unsigned int allocated_clusters = 0; | |
3602 | ||
3603 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3604 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3605 | ||
3606 | /* max possible clusters for this allocation */ | |
3607 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3608 | ||
d8990240 AK |
3609 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3610 | ||
7b415bf6 AK |
3611 | /* Check towards left side */ |
3612 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3613 | if (c_offset) { | |
3614 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3615 | lblk_to = lblk_from + c_offset - 1; | |
3616 | ||
7d1b1fbc | 3617 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3618 | allocated_clusters--; |
3619 | } | |
3620 | ||
3621 | /* Now check towards right. */ | |
3622 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3623 | if (allocated_clusters && c_offset) { | |
3624 | lblk_from = lblk_start + num_blks; | |
3625 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3626 | ||
7d1b1fbc | 3627 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3628 | allocated_clusters--; |
3629 | } | |
3630 | ||
3631 | return allocated_clusters; | |
3632 | } | |
3633 | ||
0031462b MC |
3634 | static int |
3635 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3636 | struct ext4_map_blocks *map, |
0031462b | 3637 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3638 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3639 | { |
3640 | int ret = 0; | |
3641 | int err = 0; | |
f45ee3a1 | 3642 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3643 | |
88635ca2 ZL |
3644 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3645 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3646 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3647 | flags, allocated); |
3648 | ext4_ext_show_leaf(inode, path); | |
3649 | ||
b5645534 ZL |
3650 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3651 | allocated, newblock); | |
d8990240 | 3652 | |
c7064ef1 | 3653 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3654 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3655 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3656 | path, flags); | |
82e54229 DM |
3657 | if (ret <= 0) |
3658 | goto out; | |
5f524950 M |
3659 | /* |
3660 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3661 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3662 | * completed |
3663 | */ | |
0edeb71d TM |
3664 | if (io) |
3665 | ext4_set_io_unwritten_flag(inode, io); | |
3666 | else | |
19f5fb7a | 3667 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
a25a4e1a | 3668 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
744692dc | 3669 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3670 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3671 | goto out; |
3672 | } | |
c7064ef1 | 3673 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3674 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3675 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3676 | path); |
58590b06 | 3677 | if (ret >= 0) { |
b436b9be | 3678 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3679 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3680 | path, map->m_len); | |
58590b06 TT |
3681 | } else |
3682 | err = ret; | |
cdee7843 ZL |
3683 | map->m_flags |= EXT4_MAP_MAPPED; |
3684 | if (allocated > map->m_len) | |
3685 | allocated = map->m_len; | |
3686 | map->m_len = allocated; | |
0031462b MC |
3687 | goto out2; |
3688 | } | |
3689 | /* buffered IO case */ | |
3690 | /* | |
3691 | * repeat fallocate creation request | |
3692 | * we already have an unwritten extent | |
3693 | */ | |
a25a4e1a ZL |
3694 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) { |
3695 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
0031462b | 3696 | goto map_out; |
a25a4e1a | 3697 | } |
0031462b MC |
3698 | |
3699 | /* buffered READ or buffered write_begin() lookup */ | |
3700 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3701 | /* | |
3702 | * We have blocks reserved already. We | |
3703 | * return allocated blocks so that delalloc | |
3704 | * won't do block reservation for us. But | |
3705 | * the buffer head will be unmapped so that | |
3706 | * a read from the block returns 0s. | |
3707 | */ | |
e35fd660 | 3708 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3709 | goto out1; |
3710 | } | |
3711 | ||
3712 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3713 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
a4e5d88b | 3714 | if (ret >= 0) |
b436b9be | 3715 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3716 | out: |
3717 | if (ret <= 0) { | |
3718 | err = ret; | |
3719 | goto out2; | |
3720 | } else | |
3721 | allocated = ret; | |
e35fd660 | 3722 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3723 | /* |
3724 | * if we allocated more blocks than requested | |
3725 | * we need to make sure we unmap the extra block | |
3726 | * allocated. The actual needed block will get | |
3727 | * unmapped later when we find the buffer_head marked | |
3728 | * new. | |
3729 | */ | |
e35fd660 | 3730 | if (allocated > map->m_len) { |
515f41c3 | 3731 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3732 | newblock + map->m_len, |
3733 | allocated - map->m_len); | |
3734 | allocated = map->m_len; | |
515f41c3 | 3735 | } |
3a225670 | 3736 | map->m_len = allocated; |
5f634d06 AK |
3737 | |
3738 | /* | |
3739 | * If we have done fallocate with the offset that is already | |
3740 | * delayed allocated, we would have block reservation | |
3741 | * and quota reservation done in the delayed write path. | |
3742 | * But fallocate would have already updated quota and block | |
3743 | * count for this offset. So cancel these reservation | |
3744 | */ | |
7b415bf6 AK |
3745 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3746 | unsigned int reserved_clusters; | |
3747 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3748 | map->m_lblk, map->m_len); | |
3749 | if (reserved_clusters) | |
3750 | ext4_da_update_reserve_space(inode, | |
3751 | reserved_clusters, | |
3752 | 0); | |
3753 | } | |
5f634d06 | 3754 | |
0031462b | 3755 | map_out: |
e35fd660 | 3756 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3757 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3758 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3759 | map->m_len); | |
3760 | if (err < 0) | |
3761 | goto out2; | |
3762 | } | |
0031462b | 3763 | out1: |
e35fd660 TT |
3764 | if (allocated > map->m_len) |
3765 | allocated = map->m_len; | |
0031462b | 3766 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3767 | map->m_pblk = newblock; |
3768 | map->m_len = allocated; | |
0031462b MC |
3769 | out2: |
3770 | if (path) { | |
3771 | ext4_ext_drop_refs(path); | |
3772 | kfree(path); | |
3773 | } | |
3774 | return err ? err : allocated; | |
3775 | } | |
58590b06 | 3776 | |
4d33b1ef TT |
3777 | /* |
3778 | * get_implied_cluster_alloc - check to see if the requested | |
3779 | * allocation (in the map structure) overlaps with a cluster already | |
3780 | * allocated in an extent. | |
d8990240 | 3781 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3782 | * @map The requested lblk->pblk mapping |
3783 | * @ex The extent structure which might contain an implied | |
3784 | * cluster allocation | |
3785 | * | |
3786 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3787 | * find blocks that were already in the inode's extent tree. Hence, | |
3788 | * we know that the beginning of the requested region cannot overlap | |
3789 | * the extent from the inode's extent tree. There are three cases we | |
3790 | * want to catch. The first is this case: | |
3791 | * | |
3792 | * |--- cluster # N--| | |
3793 | * |--- extent ---| |---- requested region ---| | |
3794 | * |==========| | |
3795 | * | |
3796 | * The second case that we need to test for is this one: | |
3797 | * | |
3798 | * |--------- cluster # N ----------------| | |
3799 | * |--- requested region --| |------- extent ----| | |
3800 | * |=======================| | |
3801 | * | |
3802 | * The third case is when the requested region lies between two extents | |
3803 | * within the same cluster: | |
3804 | * |------------- cluster # N-------------| | |
3805 | * |----- ex -----| |---- ex_right ----| | |
3806 | * |------ requested region ------| | |
3807 | * |================| | |
3808 | * | |
3809 | * In each of the above cases, we need to set the map->m_pblk and | |
3810 | * map->m_len so it corresponds to the return the extent labelled as | |
3811 | * "|====|" from cluster #N, since it is already in use for data in | |
3812 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3813 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3814 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3815 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3816 | * by calling ext4_mb_new_blocks(). | |
3817 | */ | |
d8990240 | 3818 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3819 | struct ext4_map_blocks *map, |
3820 | struct ext4_extent *ex, | |
3821 | struct ext4_ext_path *path) | |
3822 | { | |
d8990240 | 3823 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3824 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3825 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3826 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3827 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3828 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3829 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3830 | ||
3831 | /* The extent passed in that we are trying to match */ | |
3832 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3833 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3834 | ||
3835 | /* The requested region passed into ext4_map_blocks() */ | |
3836 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3837 | |
3838 | if ((rr_cluster_start == ex_cluster_end) || | |
3839 | (rr_cluster_start == ex_cluster_start)) { | |
3840 | if (rr_cluster_start == ex_cluster_end) | |
3841 | ee_start += ee_len - 1; | |
3842 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3843 | c_offset; | |
3844 | map->m_len = min(map->m_len, | |
3845 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3846 | /* | |
3847 | * Check for and handle this case: | |
3848 | * | |
3849 | * |--------- cluster # N-------------| | |
3850 | * |------- extent ----| | |
3851 | * |--- requested region ---| | |
3852 | * |===========| | |
3853 | */ | |
3854 | ||
3855 | if (map->m_lblk < ee_block) | |
3856 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3857 | ||
3858 | /* | |
3859 | * Check for the case where there is already another allocated | |
3860 | * block to the right of 'ex' but before the end of the cluster. | |
3861 | * | |
3862 | * |------------- cluster # N-------------| | |
3863 | * |----- ex -----| |---- ex_right ----| | |
3864 | * |------ requested region ------| | |
3865 | * |================| | |
3866 | */ | |
3867 | if (map->m_lblk > ee_block) { | |
3868 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3869 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3870 | } | |
d8990240 AK |
3871 | |
3872 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3873 | return 1; |
3874 | } | |
d8990240 AK |
3875 | |
3876 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3877 | return 0; |
3878 | } | |
3879 | ||
3880 | ||
c278bfec | 3881 | /* |
f5ab0d1f MC |
3882 | * Block allocation/map/preallocation routine for extents based files |
3883 | * | |
3884 | * | |
c278bfec | 3885 | * Need to be called with |
0e855ac8 AK |
3886 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3887 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3888 | * |
3889 | * return > 0, number of of blocks already mapped/allocated | |
3890 | * if create == 0 and these are pre-allocated blocks | |
3891 | * buffer head is unmapped | |
3892 | * otherwise blocks are mapped | |
3893 | * | |
3894 | * return = 0, if plain look up failed (blocks have not been allocated) | |
3895 | * buffer head is unmapped | |
3896 | * | |
3897 | * return < 0, error case. | |
c278bfec | 3898 | */ |
e35fd660 TT |
3899 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
3900 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
3901 | { |
3902 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
3903 | struct ext4_extent newex, *ex, *ex2; |
3904 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 3905 | ext4_fsblk_t newblock = 0; |
37794732 | 3906 | int free_on_err = 0, err = 0, depth; |
4d33b1ef | 3907 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 3908 | unsigned int allocated_clusters = 0; |
c9de560d | 3909 | struct ext4_allocation_request ar; |
f45ee3a1 | 3910 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 3911 | ext4_lblk_t cluster_offset; |
82e54229 | 3912 | int set_unwritten = 0; |
a86c6181 | 3913 | |
84fe3bef | 3914 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 3915 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 3916 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 3917 | |
a86c6181 | 3918 | /* find extent for this block */ |
e35fd660 | 3919 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
3920 | if (IS_ERR(path)) { |
3921 | err = PTR_ERR(path); | |
3922 | path = NULL; | |
3923 | goto out2; | |
3924 | } | |
3925 | ||
3926 | depth = ext_depth(inode); | |
3927 | ||
3928 | /* | |
d0d856e8 RD |
3929 | * consistent leaf must not be empty; |
3930 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
3931 | * this is why assert can't be put in ext4_ext_find_extent() |
3932 | */ | |
273df556 FM |
3933 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
3934 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
3935 | "lblock: %lu, depth: %d pblock %lld", |
3936 | (unsigned long) map->m_lblk, depth, | |
3937 | path[depth].p_block); | |
034fb4c9 SP |
3938 | err = -EIO; |
3939 | goto out2; | |
3940 | } | |
a86c6181 | 3941 | |
7e028976 AM |
3942 | ex = path[depth].p_ext; |
3943 | if (ex) { | |
725d26d3 | 3944 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 3945 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 3946 | unsigned short ee_len; |
471d4011 SB |
3947 | |
3948 | /* | |
471d4011 | 3949 | * Uninitialized extents are treated as holes, except that |
56055d3a | 3950 | * we split out initialized portions during a write. |
471d4011 | 3951 | */ |
a2df2a63 | 3952 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
3953 | |
3954 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
3955 | ||
d0d856e8 | 3956 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
3957 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
3958 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 3959 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
3960 | allocated = ee_len - (map->m_lblk - ee_block); |
3961 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
3962 | ee_block, ee_len, newblock); | |
56055d3a | 3963 | |
69eb33dc | 3964 | if (!ext4_ext_is_uninitialized(ex)) |
7877191c | 3965 | goto out; |
69eb33dc | 3966 | |
37794732 | 3967 | allocated = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
3968 | handle, inode, map, path, flags, |
3969 | allocated, newblock); | |
37794732 | 3970 | goto out3; |
a86c6181 AT |
3971 | } |
3972 | } | |
3973 | ||
7b415bf6 | 3974 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 3975 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
3976 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
3977 | ||
a86c6181 | 3978 | /* |
d0d856e8 | 3979 | * requested block isn't allocated yet; |
a86c6181 AT |
3980 | * we couldn't try to create block if create flag is zero |
3981 | */ | |
c2177057 | 3982 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
3983 | /* |
3984 | * put just found gap into cache to speed up | |
3985 | * subsequent requests | |
3986 | */ | |
d100eef2 ZL |
3987 | if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0) |
3988 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); | |
a86c6181 AT |
3989 | goto out2; |
3990 | } | |
4d33b1ef | 3991 | |
a86c6181 | 3992 | /* |
c2ea3fde | 3993 | * Okay, we need to do block allocation. |
63f57933 | 3994 | */ |
7b415bf6 | 3995 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
3996 | newex.ee_block = cpu_to_le32(map->m_lblk); |
3997 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
3998 | ||
3999 | /* | |
4000 | * If we are doing bigalloc, check to see if the extent returned | |
4001 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4002 | */ | |
4003 | if (cluster_offset && ex && | |
d8990240 | 4004 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4005 | ar.len = allocated = map->m_len; |
4006 | newblock = map->m_pblk; | |
7b415bf6 | 4007 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4008 | goto got_allocated_blocks; |
4009 | } | |
a86c6181 | 4010 | |
c9de560d | 4011 | /* find neighbour allocated blocks */ |
e35fd660 | 4012 | ar.lleft = map->m_lblk; |
c9de560d AT |
4013 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4014 | if (err) | |
4015 | goto out2; | |
e35fd660 | 4016 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4017 | ex2 = NULL; |
4018 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4019 | if (err) |
4020 | goto out2; | |
25d14f98 | 4021 | |
4d33b1ef TT |
4022 | /* Check if the extent after searching to the right implies a |
4023 | * cluster we can use. */ | |
4024 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4025 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4026 | ar.len = allocated = map->m_len; |
4027 | newblock = map->m_pblk; | |
7b415bf6 | 4028 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4029 | goto got_allocated_blocks; |
4030 | } | |
4031 | ||
749269fa AA |
4032 | /* |
4033 | * See if request is beyond maximum number of blocks we can have in | |
4034 | * a single extent. For an initialized extent this limit is | |
4035 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4036 | * EXT_UNINIT_MAX_LEN. | |
4037 | */ | |
e35fd660 | 4038 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4039 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4040 | map->m_len = EXT_INIT_MAX_LEN; |
4041 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4042 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4043 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4044 | |
e35fd660 | 4045 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4046 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4047 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4048 | if (err) |
b939e376 | 4049 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4050 | else |
e35fd660 | 4051 | allocated = map->m_len; |
c9de560d AT |
4052 | |
4053 | /* allocate new block */ | |
4054 | ar.inode = inode; | |
e35fd660 TT |
4055 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4056 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4057 | /* |
4058 | * We calculate the offset from the beginning of the cluster | |
4059 | * for the logical block number, since when we allocate a | |
4060 | * physical cluster, the physical block should start at the | |
4061 | * same offset from the beginning of the cluster. This is | |
4062 | * needed so that future calls to get_implied_cluster_alloc() | |
4063 | * work correctly. | |
4064 | */ | |
4065 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4066 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4067 | ar.goal -= offset; | |
4068 | ar.logical -= offset; | |
c9de560d AT |
4069 | if (S_ISREG(inode->i_mode)) |
4070 | ar.flags = EXT4_MB_HINT_DATA; | |
4071 | else | |
4072 | /* disable in-core preallocation for non-regular files */ | |
4073 | ar.flags = 0; | |
556b27ab VH |
4074 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4075 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4076 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4077 | if (!newblock) |
4078 | goto out2; | |
84fe3bef | 4079 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4080 | ar.goal, newblock, allocated); |
4d33b1ef | 4081 | free_on_err = 1; |
7b415bf6 | 4082 | allocated_clusters = ar.len; |
4d33b1ef TT |
4083 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4084 | if (ar.len > allocated) | |
4085 | ar.len = allocated; | |
a86c6181 | 4086 | |
4d33b1ef | 4087 | got_allocated_blocks: |
a86c6181 | 4088 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4089 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4090 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4091 | /* Mark uninitialized */ |
4092 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4093 | ext4_ext_mark_uninitialized(&newex); |
a25a4e1a | 4094 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4095 | /* |
744692dc | 4096 | * io_end structure was created for every IO write to an |
25985edc | 4097 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4098 | * here we flag the IO that really needs the conversion. |
5f524950 | 4099 | * For non asycn direct IO case, flag the inode state |
25985edc | 4100 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4101 | */ |
82e54229 DM |
4102 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4103 | set_unwritten = 1; | |
744692dc | 4104 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4105 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4106 | } |
c8d46e41 | 4107 | |
a4e5d88b DM |
4108 | err = 0; |
4109 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4110 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4111 | path, ar.len); | |
575a1d4b JZ |
4112 | if (!err) |
4113 | err = ext4_ext_insert_extent(handle, inode, path, | |
4114 | &newex, flags); | |
82e54229 DM |
4115 | |
4116 | if (!err && set_unwritten) { | |
4117 | if (io) | |
4118 | ext4_set_io_unwritten_flag(inode, io); | |
4119 | else | |
4120 | ext4_set_inode_state(inode, | |
4121 | EXT4_STATE_DIO_UNWRITTEN); | |
4122 | } | |
4123 | ||
4d33b1ef | 4124 | if (err && free_on_err) { |
7132de74 MP |
4125 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4126 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4127 | /* free data blocks we just allocated */ |
c9de560d AT |
4128 | /* not a good idea to call discard here directly, |
4129 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4130 | ext4_discard_preallocations(inode); |
7dc57615 | 4131 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4132 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4133 | goto out2; |
315054f0 | 4134 | } |
a86c6181 | 4135 | |
a86c6181 | 4136 | /* previous routine could use block we allocated */ |
bf89d16f | 4137 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4138 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4139 | if (allocated > map->m_len) |
4140 | allocated = map->m_len; | |
4141 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4142 | |
5f634d06 AK |
4143 | /* |
4144 | * Update reserved blocks/metadata blocks after successful | |
4145 | * block allocation which had been deferred till now. | |
4146 | */ | |
7b415bf6 | 4147 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4148 | unsigned int reserved_clusters; |
7b415bf6 | 4149 | /* |
81fdbb4a | 4150 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4151 | */ |
4152 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4153 | map->m_lblk, allocated); | |
4154 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4155 | if (reserved_clusters) { | |
4156 | /* | |
4157 | * We have clusters reserved for this range. | |
4158 | * But since we are not doing actual allocation | |
4159 | * and are simply using blocks from previously | |
4160 | * allocated cluster, we should release the | |
4161 | * reservation and not claim quota. | |
4162 | */ | |
4163 | ext4_da_update_reserve_space(inode, | |
4164 | reserved_clusters, 0); | |
4165 | } | |
4166 | } else { | |
4167 | BUG_ON(allocated_clusters < reserved_clusters); | |
4168 | /* We will claim quota for all newly allocated blocks.*/ | |
4169 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4170 | 1); | |
4171 | if (reserved_clusters < allocated_clusters) { | |
5356f261 | 4172 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4173 | int reservation = allocated_clusters - |
4174 | reserved_clusters; | |
4175 | /* | |
4176 | * It seems we claimed few clusters outside of | |
4177 | * the range of this allocation. We should give | |
4178 | * it back to the reservation pool. This can | |
4179 | * happen in the following case: | |
4180 | * | |
4181 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4182 | * cluster has 4 blocks. Thus, the clusters | |
4183 | * are [0-3],[4-7],[8-11]... | |
4184 | * * First comes delayed allocation write for | |
4185 | * logical blocks 10 & 11. Since there were no | |
4186 | * previous delayed allocated blocks in the | |
4187 | * range [8-11], we would reserve 1 cluster | |
4188 | * for this write. | |
4189 | * * Next comes write for logical blocks 3 to 8. | |
4190 | * In this case, we will reserve 2 clusters | |
4191 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4192 | * that range has a delayed allocated blocks. | |
4193 | * Thus total reserved clusters now becomes 3. | |
4194 | * * Now, during the delayed allocation writeout | |
4195 | * time, we will first write blocks [3-8] and | |
4196 | * allocate 3 clusters for writing these | |
4197 | * blocks. Also, we would claim all these | |
4198 | * three clusters above. | |
4199 | * * Now when we come here to writeout the | |
4200 | * blocks [10-11], we would expect to claim | |
4201 | * the reservation of 1 cluster we had made | |
4202 | * (and we would claim it since there are no | |
4203 | * more delayed allocated blocks in the range | |
4204 | * [8-11]. But our reserved cluster count had | |
4205 | * already gone to 0. | |
4206 | * | |
4207 | * Thus, at the step 4 above when we determine | |
4208 | * that there are still some unwritten delayed | |
4209 | * allocated blocks outside of our current | |
4210 | * block range, we should increment the | |
4211 | * reserved clusters count so that when the | |
4212 | * remaining blocks finally gets written, we | |
4213 | * could claim them. | |
4214 | */ | |
5356f261 AK |
4215 | dquot_reserve_block(inode, |
4216 | EXT4_C2B(sbi, reservation)); | |
4217 | spin_lock(&ei->i_block_reservation_lock); | |
4218 | ei->i_reserved_data_blocks += reservation; | |
4219 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 AK |
4220 | } |
4221 | } | |
4222 | } | |
5f634d06 | 4223 | |
b436b9be JK |
4224 | /* |
4225 | * Cache the extent and update transaction to commit on fdatasync only | |
4226 | * when it is _not_ an uninitialized extent. | |
4227 | */ | |
69eb33dc | 4228 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) |
b436b9be | 4229 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4230 | else |
b436b9be | 4231 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4232 | out: |
e35fd660 TT |
4233 | if (allocated > map->m_len) |
4234 | allocated = map->m_len; | |
a86c6181 | 4235 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4236 | map->m_flags |= EXT4_MAP_MAPPED; |
4237 | map->m_pblk = newblock; | |
4238 | map->m_len = allocated; | |
a86c6181 AT |
4239 | out2: |
4240 | if (path) { | |
4241 | ext4_ext_drop_refs(path); | |
4242 | kfree(path); | |
4243 | } | |
e861304b | 4244 | |
37794732 | 4245 | out3: |
19b303d8 | 4246 | trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated); |
e7b319e3 | 4247 | |
7877191c | 4248 | return err ? err : allocated; |
a86c6181 AT |
4249 | } |
4250 | ||
cf108bca | 4251 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
4252 | { |
4253 | struct address_space *mapping = inode->i_mapping; | |
4254 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 4255 | ext4_lblk_t last_block; |
a86c6181 | 4256 | handle_t *handle; |
189e868f | 4257 | loff_t page_len; |
a86c6181 AT |
4258 | int err = 0; |
4259 | ||
3889fd57 JZ |
4260 | /* |
4261 | * finish any pending end_io work so we won't run the risk of | |
4262 | * converting any truncated blocks to initialized later | |
4263 | */ | |
c278531d | 4264 | ext4_flush_unwritten_io(inode); |
3889fd57 | 4265 | |
a86c6181 AT |
4266 | /* |
4267 | * probably first extent we're gonna free will be last in block | |
4268 | */ | |
f3bd1f3f | 4269 | err = ext4_writepage_trans_blocks(inode); |
9924a92a | 4270 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, err); |
cf108bca | 4271 | if (IS_ERR(handle)) |
a86c6181 | 4272 | return; |
a86c6181 | 4273 | |
189e868f AH |
4274 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
4275 | page_len = PAGE_CACHE_SIZE - | |
4276 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4277 | ||
4278 | err = ext4_discard_partial_page_buffers(handle, | |
4279 | mapping, inode->i_size, page_len, 0); | |
4280 | ||
4281 | if (err) | |
4282 | goto out_stop; | |
4283 | } | |
a86c6181 | 4284 | |
9ddfc3dc JK |
4285 | if (ext4_orphan_add(handle, inode)) |
4286 | goto out_stop; | |
4287 | ||
0e855ac8 | 4288 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 | 4289 | |
c2ea3fde | 4290 | ext4_discard_preallocations(inode); |
c9de560d | 4291 | |
a86c6181 | 4292 | /* |
d0d856e8 RD |
4293 | * TODO: optimization is possible here. |
4294 | * Probably we need not scan at all, | |
4295 | * because page truncation is enough. | |
a86c6181 | 4296 | */ |
a86c6181 AT |
4297 | |
4298 | /* we have to know where to truncate from in crash case */ | |
4299 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4300 | ext4_mark_inode_dirty(handle, inode); | |
4301 | ||
4302 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4303 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
51865fda ZL |
4304 | err = ext4_es_remove_extent(inode, last_block, |
4305 | EXT_MAX_BLOCKS - last_block); | |
5f95d21f | 4306 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
a86c6181 AT |
4307 | |
4308 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
4309 | * transaction synchronous. |
4310 | */ | |
a86c6181 | 4311 | if (IS_SYNC(inode)) |
0390131b | 4312 | ext4_handle_sync(handle); |
a86c6181 | 4313 | |
9ddfc3dc | 4314 | up_write(&EXT4_I(inode)->i_data_sem); |
f6d2f6b3 EG |
4315 | |
4316 | out_stop: | |
a86c6181 | 4317 | /* |
d0d856e8 | 4318 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
4319 | * then we need to clear up the orphan record which we created above. |
4320 | * However, if this was a real unlink then we were called by | |
4321 | * ext4_delete_inode(), and we allow that function to clean up the | |
4322 | * orphan info for us. | |
4323 | */ | |
4324 | if (inode->i_nlink) | |
4325 | ext4_orphan_del(handle, inode); | |
4326 | ||
ef737728 SR |
4327 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4328 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
4329 | ext4_journal_stop(handle); |
4330 | } | |
4331 | ||
fd28784a AK |
4332 | static void ext4_falloc_update_inode(struct inode *inode, |
4333 | int mode, loff_t new_size, int update_ctime) | |
4334 | { | |
4335 | struct timespec now; | |
4336 | ||
4337 | if (update_ctime) { | |
4338 | now = current_fs_time(inode->i_sb); | |
4339 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4340 | inode->i_ctime = now; | |
4341 | } | |
4342 | /* | |
4343 | * Update only when preallocation was requested beyond | |
4344 | * the file size. | |
4345 | */ | |
cf17fea6 AK |
4346 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4347 | if (new_size > i_size_read(inode)) | |
4348 | i_size_write(inode, new_size); | |
4349 | if (new_size > EXT4_I(inode)->i_disksize) | |
4350 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4351 | } else { |
4352 | /* | |
4353 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4354 | * can proceed even if the new size is the same as i_size. | |
4355 | */ | |
4356 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4357 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4358 | } |
4359 | ||
4360 | } | |
4361 | ||
a2df2a63 | 4362 | /* |
2fe17c10 | 4363 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4364 | * operation, which gets called from sys_fallocate system call. |
4365 | * For block-mapped files, posix_fallocate should fall back to the method | |
4366 | * of writing zeroes to the required new blocks (the same behavior which is | |
4367 | * expected for file systems which do not support fallocate() system call). | |
4368 | */ | |
2fe17c10 | 4369 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4370 | { |
2fe17c10 | 4371 | struct inode *inode = file->f_path.dentry->d_inode; |
a2df2a63 | 4372 | handle_t *handle; |
fd28784a | 4373 | loff_t new_size; |
498e5f24 | 4374 | unsigned int max_blocks; |
a2df2a63 AA |
4375 | int ret = 0; |
4376 | int ret2 = 0; | |
4377 | int retries = 0; | |
a4e5d88b | 4378 | int flags; |
2ed88685 | 4379 | struct ext4_map_blocks map; |
a2df2a63 AA |
4380 | unsigned int credits, blkbits = inode->i_blkbits; |
4381 | ||
a4bb6b64 AH |
4382 | /* Return error if mode is not supported */ |
4383 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4384 | return -EOPNOTSUPP; | |
4385 | ||
4386 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4387 | return ext4_punch_hole(file, offset, len); | |
4388 | ||
0c8d414f TM |
4389 | ret = ext4_convert_inline_data(inode); |
4390 | if (ret) | |
4391 | return ret; | |
4392 | ||
8bad6fc8 ZL |
4393 | /* |
4394 | * currently supporting (pre)allocate mode for extent-based | |
4395 | * files _only_ | |
4396 | */ | |
4397 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4398 | return -EOPNOTSUPP; | |
4399 | ||
0562e0ba | 4400 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4401 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4402 | /* |
4403 | * We can't just convert len to max_blocks because | |
4404 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4405 | */ | |
a2df2a63 | 4406 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4407 | - map.m_lblk; |
a2df2a63 | 4408 | /* |
f3bd1f3f | 4409 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4410 | */ |
f3bd1f3f | 4411 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4412 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4413 | ret = inode_newsize_ok(inode, (len + offset)); |
4414 | if (ret) { | |
4415 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4416 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4417 | return ret; |
4418 | } | |
3c6fe770 | 4419 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4420 | if (mode & FALLOC_FL_KEEP_SIZE) |
4421 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4422 | /* |
4423 | * Don't normalize the request if it can fit in one extent so | |
4424 | * that it doesn't get unnecessarily split into multiple | |
4425 | * extents. | |
4426 | */ | |
4427 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4428 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f | 4429 | |
a2df2a63 AA |
4430 | retry: |
4431 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4432 | map.m_lblk = map.m_lblk + ret; |
4433 | map.m_len = max_blocks = max_blocks - ret; | |
9924a92a TT |
4434 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4435 | credits); | |
a2df2a63 AA |
4436 | if (IS_ERR(handle)) { |
4437 | ret = PTR_ERR(handle); | |
4438 | break; | |
4439 | } | |
a4e5d88b | 4440 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4441 | if (ret <= 0) { |
2c98615d | 4442 | #ifdef EXT4FS_DEBUG |
b06acd38 LC |
4443 | ext4_warning(inode->i_sb, |
4444 | "inode #%lu: block %u: len %u: " | |
4445 | "ext4_ext_map_blocks returned %d", | |
4446 | inode->i_ino, map.m_lblk, | |
4447 | map.m_len, ret); | |
2c98615d | 4448 | #endif |
a2df2a63 AA |
4449 | ext4_mark_inode_dirty(handle, inode); |
4450 | ret2 = ext4_journal_stop(handle); | |
4451 | break; | |
4452 | } | |
2ed88685 | 4453 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4454 | blkbits) >> blkbits)) |
4455 | new_size = offset + len; | |
4456 | else | |
29ae07b7 | 4457 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4458 | |
fd28784a | 4459 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4460 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4461 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4462 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4463 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4464 | ret2 = ext4_journal_stop(handle); |
4465 | if (ret2) | |
4466 | break; | |
4467 | } | |
fd28784a AK |
4468 | if (ret == -ENOSPC && |
4469 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4470 | ret = 0; | |
a2df2a63 | 4471 | goto retry; |
a2df2a63 | 4472 | } |
55bd725a | 4473 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4474 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4475 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4476 | return ret > 0 ? ret2 : ret; |
4477 | } | |
6873fa0d | 4478 | |
0031462b MC |
4479 | /* |
4480 | * This function convert a range of blocks to written extents | |
4481 | * The caller of this function will pass the start offset and the size. | |
4482 | * all unwritten extents within this range will be converted to | |
4483 | * written extents. | |
4484 | * | |
4485 | * This function is called from the direct IO end io call back | |
4486 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4487 | * Returns 0 on success. |
0031462b MC |
4488 | */ |
4489 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4490 | ssize_t len) |
0031462b MC |
4491 | { |
4492 | handle_t *handle; | |
0031462b MC |
4493 | unsigned int max_blocks; |
4494 | int ret = 0; | |
4495 | int ret2 = 0; | |
2ed88685 | 4496 | struct ext4_map_blocks map; |
0031462b MC |
4497 | unsigned int credits, blkbits = inode->i_blkbits; |
4498 | ||
2ed88685 | 4499 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4500 | /* |
4501 | * We can't just convert len to max_blocks because | |
4502 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4503 | */ | |
2ed88685 TT |
4504 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4505 | map.m_lblk); | |
0031462b MC |
4506 | /* |
4507 | * credits to insert 1 extent into extent tree | |
4508 | */ | |
4509 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4510 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4511 | map.m_lblk += ret; |
4512 | map.m_len = (max_blocks -= ret); | |
9924a92a | 4513 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits); |
0031462b MC |
4514 | if (IS_ERR(handle)) { |
4515 | ret = PTR_ERR(handle); | |
4516 | break; | |
4517 | } | |
2ed88685 | 4518 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4519 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4520 | if (ret <= 0) |
4521 | ext4_warning(inode->i_sb, | |
4522 | "inode #%lu: block %u: len %u: " | |
4523 | "ext4_ext_map_blocks returned %d", | |
4524 | inode->i_ino, map.m_lblk, | |
4525 | map.m_len, ret); | |
0031462b MC |
4526 | ext4_mark_inode_dirty(handle, inode); |
4527 | ret2 = ext4_journal_stop(handle); | |
4528 | if (ret <= 0 || ret2 ) | |
4529 | break; | |
4530 | } | |
4531 | return ret > 0 ? ret2 : ret; | |
4532 | } | |
6d9c85eb | 4533 | |
6873fa0d | 4534 | /* |
69eb33dc ZL |
4535 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
4536 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
4537 | * return start of the next delayed extent. |
4538 | * | |
69eb33dc | 4539 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 4540 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 4541 | * extent found. Leave newes unmodified. |
6873fa0d | 4542 | */ |
91dd8c11 | 4543 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 4544 | struct extent_status *newes) |
6873fa0d | 4545 | { |
b3aff3e3 | 4546 | struct extent_status es; |
be401363 | 4547 | ext4_lblk_t block, next_del; |
6873fa0d | 4548 | |
69eb33dc | 4549 | ext4_es_find_delayed_extent(inode, newes->es_lblk, &es); |
6873fa0d | 4550 | |
69eb33dc | 4551 | if (newes->es_pblk == 0) { |
6d9c85eb | 4552 | /* |
69eb33dc | 4553 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 4554 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 4555 | */ |
06b0c886 | 4556 | if (es.es_len == 0) |
b3aff3e3 | 4557 | /* A hole found. */ |
91dd8c11 | 4558 | return 0; |
b3aff3e3 | 4559 | |
69eb33dc | 4560 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 4561 | /* A hole found. */ |
69eb33dc ZL |
4562 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
4563 | newes->es_len); | |
91dd8c11 | 4564 | return 0; |
6873fa0d | 4565 | } |
6d9c85eb | 4566 | |
69eb33dc | 4567 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
4568 | } |
4569 | ||
69eb33dc | 4570 | block = newes->es_lblk + newes->es_len; |
be401363 ZL |
4571 | ext4_es_find_delayed_extent(inode, block, &es); |
4572 | if (es.es_len == 0) | |
4573 | next_del = EXT_MAX_BLOCKS; | |
4574 | else | |
4575 | next_del = es.es_lblk; | |
4576 | ||
91dd8c11 | 4577 | return next_del; |
6873fa0d | 4578 | } |
6873fa0d ES |
4579 | /* fiemap flags we can handle specified here */ |
4580 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4581 | ||
3a06d778 AK |
4582 | static int ext4_xattr_fiemap(struct inode *inode, |
4583 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4584 | { |
4585 | __u64 physical = 0; | |
4586 | __u64 length; | |
4587 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4588 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4589 | int error = 0; | |
4590 | ||
4591 | /* in-inode? */ | |
19f5fb7a | 4592 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4593 | struct ext4_iloc iloc; |
4594 | int offset; /* offset of xattr in inode */ | |
4595 | ||
4596 | error = ext4_get_inode_loc(inode, &iloc); | |
4597 | if (error) | |
4598 | return error; | |
4599 | physical = iloc.bh->b_blocknr << blockbits; | |
4600 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4601 | EXT4_I(inode)->i_extra_isize; | |
4602 | physical += offset; | |
4603 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4604 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4605 | brelse(iloc.bh); |
6873fa0d ES |
4606 | } else { /* external block */ |
4607 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4608 | length = inode->i_sb->s_blocksize; | |
4609 | } | |
4610 | ||
4611 | if (physical) | |
4612 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4613 | length, flags); | |
4614 | return (error < 0 ? error : 0); | |
4615 | } | |
4616 | ||
a4bb6b64 AH |
4617 | /* |
4618 | * ext4_ext_punch_hole | |
4619 | * | |
4620 | * Punches a hole of "length" bytes in a file starting | |
4621 | * at byte "offset" | |
4622 | * | |
4623 | * @inode: The inode of the file to punch a hole in | |
4624 | * @offset: The starting byte offset of the hole | |
4625 | * @length: The length of the hole | |
4626 | * | |
4627 | * Returns the number of blocks removed or negative on err | |
4628 | */ | |
4629 | int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length) | |
4630 | { | |
4631 | struct inode *inode = file->f_path.dentry->d_inode; | |
4632 | struct super_block *sb = inode->i_sb; | |
5f95d21f | 4633 | ext4_lblk_t first_block, stop_block; |
a4bb6b64 | 4634 | struct address_space *mapping = inode->i_mapping; |
a4bb6b64 | 4635 | handle_t *handle; |
ba06208a AH |
4636 | loff_t first_page, last_page, page_len; |
4637 | loff_t first_page_offset, last_page_offset; | |
5f95d21f | 4638 | int credits, err = 0; |
a4bb6b64 | 4639 | |
02d262df DM |
4640 | /* |
4641 | * Write out all dirty pages to avoid race conditions | |
4642 | * Then release them. | |
4643 | */ | |
4644 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4645 | err = filemap_write_and_wait_range(mapping, | |
4646 | offset, offset + length - 1); | |
4647 | ||
4648 | if (err) | |
4649 | return err; | |
4650 | } | |
4651 | ||
4652 | mutex_lock(&inode->i_mutex); | |
4653 | /* It's not possible punch hole on append only file */ | |
4654 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
4655 | err = -EPERM; | |
4656 | goto out_mutex; | |
4657 | } | |
4658 | if (IS_SWAPFILE(inode)) { | |
4659 | err = -ETXTBSY; | |
4660 | goto out_mutex; | |
4661 | } | |
4662 | ||
2be4751b AH |
4663 | /* No need to punch hole beyond i_size */ |
4664 | if (offset >= inode->i_size) | |
02d262df | 4665 | goto out_mutex; |
2be4751b AH |
4666 | |
4667 | /* | |
4668 | * If the hole extends beyond i_size, set the hole | |
4669 | * to end after the page that contains i_size | |
4670 | */ | |
4671 | if (offset + length > inode->i_size) { | |
4672 | length = inode->i_size + | |
4673 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
4674 | offset; | |
4675 | } | |
4676 | ||
a4bb6b64 AH |
4677 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
4678 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
4679 | ||
4680 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
4681 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
4682 | ||
a4bb6b64 AH |
4683 | /* Now release the pages */ |
4684 | if (last_page_offset > first_page_offset) { | |
5e44f8c3 HD |
4685 | truncate_pagecache_range(inode, first_page_offset, |
4686 | last_page_offset - 1); | |
a4bb6b64 AH |
4687 | } |
4688 | ||
02d262df DM |
4689 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
4690 | ext4_inode_block_unlocked_dio(inode); | |
c278531d | 4691 | err = ext4_flush_unwritten_io(inode); |
28a535f9 | 4692 | if (err) |
02d262df | 4693 | goto out_dio; |
c278531d | 4694 | inode_dio_wait(inode); |
a4bb6b64 AH |
4695 | |
4696 | credits = ext4_writepage_trans_blocks(inode); | |
9924a92a | 4697 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
02d262df DM |
4698 | if (IS_ERR(handle)) { |
4699 | err = PTR_ERR(handle); | |
4700 | goto out_dio; | |
4701 | } | |
a4bb6b64 | 4702 | |
a4bb6b64 AH |
4703 | |
4704 | /* | |
ba06208a AH |
4705 | * Now we need to zero out the non-page-aligned data in the |
4706 | * pages at the start and tail of the hole, and unmap the buffer | |
4707 | * heads for the block aligned regions of the page that were | |
4708 | * completely zeroed. | |
a4bb6b64 | 4709 | */ |
ba06208a AH |
4710 | if (first_page > last_page) { |
4711 | /* | |
4712 | * If the file space being truncated is contained within a page | |
4713 | * just zero out and unmap the middle of that page | |
4714 | */ | |
4715 | err = ext4_discard_partial_page_buffers(handle, | |
4716 | mapping, offset, length, 0); | |
4717 | ||
4718 | if (err) | |
4719 | goto out; | |
4720 | } else { | |
4721 | /* | |
4722 | * zero out and unmap the partial page that contains | |
4723 | * the start of the hole | |
4724 | */ | |
4725 | page_len = first_page_offset - offset; | |
4726 | if (page_len > 0) { | |
4727 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4728 | offset, page_len, 0); | |
4729 | if (err) | |
4730 | goto out; | |
4731 | } | |
4732 | ||
4733 | /* | |
4734 | * zero out and unmap the partial page that contains | |
4735 | * the end of the hole | |
4736 | */ | |
4737 | page_len = offset + length - last_page_offset; | |
4738 | if (page_len > 0) { | |
4739 | err = ext4_discard_partial_page_buffers(handle, mapping, | |
4740 | last_page_offset, page_len, 0); | |
4741 | if (err) | |
4742 | goto out; | |
a4bb6b64 AH |
4743 | } |
4744 | } | |
4745 | ||
2be4751b AH |
4746 | /* |
4747 | * If i_size is contained in the last page, we need to | |
4748 | * unmap and zero the partial page after i_size | |
4749 | */ | |
4750 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
4751 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
4752 | ||
4753 | page_len = PAGE_CACHE_SIZE - | |
4754 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
4755 | ||
4756 | if (page_len > 0) { | |
4757 | err = ext4_discard_partial_page_buffers(handle, | |
4758 | mapping, inode->i_size, page_len, 0); | |
4759 | ||
4760 | if (err) | |
4761 | goto out; | |
4762 | } | |
4763 | } | |
4764 | ||
5f95d21f LC |
4765 | first_block = (offset + sb->s_blocksize - 1) >> |
4766 | EXT4_BLOCK_SIZE_BITS(sb); | |
4767 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4768 | ||
a4bb6b64 | 4769 | /* If there are no blocks to remove, return now */ |
5f95d21f | 4770 | if (first_block >= stop_block) |
a4bb6b64 AH |
4771 | goto out; |
4772 | ||
4773 | down_write(&EXT4_I(inode)->i_data_sem); | |
a4bb6b64 AH |
4774 | ext4_discard_preallocations(inode); |
4775 | ||
51865fda ZL |
4776 | err = ext4_es_remove_extent(inode, first_block, |
4777 | stop_block - first_block); | |
5f95d21f | 4778 | err = ext4_ext_remove_space(inode, first_block, stop_block - 1); |
a4bb6b64 | 4779 | |
5f95d21f | 4780 | ext4_discard_preallocations(inode); |
a4bb6b64 AH |
4781 | |
4782 | if (IS_SYNC(inode)) | |
4783 | ext4_handle_sync(handle); | |
4784 | ||
4785 | up_write(&EXT4_I(inode)->i_data_sem); | |
4786 | ||
4787 | out: | |
a4bb6b64 AH |
4788 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4789 | ext4_mark_inode_dirty(handle, inode); | |
4790 | ext4_journal_stop(handle); | |
02d262df DM |
4791 | out_dio: |
4792 | ext4_inode_resume_unlocked_dio(inode); | |
4793 | out_mutex: | |
4794 | mutex_unlock(&inode->i_mutex); | |
a4bb6b64 AH |
4795 | return err; |
4796 | } | |
91dd8c11 | 4797 | |
6873fa0d ES |
4798 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4799 | __u64 start, __u64 len) | |
4800 | { | |
4801 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4802 | int error = 0; |
4803 | ||
94191985 TM |
4804 | if (ext4_has_inline_data(inode)) { |
4805 | int has_inline = 1; | |
4806 | ||
4807 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline); | |
4808 | ||
4809 | if (has_inline) | |
4810 | return error; | |
4811 | } | |
4812 | ||
6873fa0d | 4813 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 4814 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4815 | return generic_block_fiemap(inode, fieinfo, start, len, |
4816 | ext4_get_block); | |
4817 | ||
4818 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4819 | return -EBADR; | |
4820 | ||
4821 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4822 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4823 | } else { | |
aca92ff6 LM |
4824 | ext4_lblk_t len_blks; |
4825 | __u64 last_blk; | |
4826 | ||
6873fa0d | 4827 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4828 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4829 | if (last_blk >= EXT_MAX_BLOCKS) |
4830 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4831 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4832 | |
4833 | /* | |
91dd8c11 LC |
4834 | * Walk the extent tree gathering extent information |
4835 | * and pushing extents back to the user. | |
6873fa0d | 4836 | */ |
91dd8c11 LC |
4837 | error = ext4_fill_fiemap_extents(inode, start_blk, |
4838 | len_blks, fieinfo); | |
6873fa0d ES |
4839 | } |
4840 | ||
4841 | return error; | |
4842 | } |