1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
5 *
6 * Architecture independence:
7 * Copyright (c) 2005, Bull S.A.
8 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 */
10
11 /*
12 * Extents support for EXT4
13 *
14 * TODO:
15 * - ext4*_error() should be used in some situations
16 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17 * - smart tree reduction
18 */
19
20 #include <linux/fs.h>
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/iomap.h>
31 #include <linux/sched/mm.h>
32 #include "ext4_jbd2.h"
33 #include "ext4_extents.h"
34 #include "xattr.h"
35
36 #include <trace/events/ext4.h>
37
38 /*
39 * used by extent splitting.
40 */
41 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
42 due to ENOSPC */
43 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
44 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
45
46 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
47 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
48
49 static __le32 ext4_extent_block_csum(struct inode *inode,
50 struct ext4_extent_header *eh)
51 {
52 struct ext4_inode_info *ei = EXT4_I(inode);
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
54 __u32 csum;
55
56 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
57 EXT4_EXTENT_TAIL_OFFSET(eh));
58 return cpu_to_le32(csum);
59 }
60
61 static int ext4_extent_block_csum_verify(struct inode *inode,
62 struct ext4_extent_header *eh)
63 {
64 struct ext4_extent_tail *et;
65
66 if (!ext4_has_metadata_csum(inode->i_sb))
67 return 1;
68
69 et = find_ext4_extent_tail(eh);
70 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
71 return 0;
72 return 1;
73 }
74
75 static void ext4_extent_block_csum_set(struct inode *inode,
76 struct ext4_extent_header *eh)
77 {
78 struct ext4_extent_tail *et;
79
80 if (!ext4_has_metadata_csum(inode->i_sb))
81 return;
82
83 et = find_ext4_extent_tail(eh);
84 et->et_checksum = ext4_extent_block_csum(inode, eh);
85 }
86
87 static int ext4_split_extent_at(handle_t *handle,
88 struct inode *inode,
89 struct ext4_ext_path **ppath,
90 ext4_lblk_t split,
91 int split_flag,
92 int flags);
93
94 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
95 {
96 /*
97 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
98 * moment, get_block can be called only for blocks inside i_size since
99 * page cache has been already dropped and writes are blocked by
100 * i_rwsem. So we can safely drop the i_data_sem here.
101 */
102 BUG_ON(EXT4_JOURNAL(inode) == NULL);
103 ext4_discard_preallocations(inode);
104 up_write(&EXT4_I(inode)->i_data_sem);
105 *dropped = 1;
106 return 0;
107 }
108
109 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
110 {
111 int depth, i;
112
113 if (!path)
114 return;
115 depth = path->p_depth;
116 for (i = 0; i <= depth; i++, path++) {
117 brelse(path->p_bh);
118 path->p_bh = NULL;
119 }
120 }
121
122 void ext4_free_ext_path(struct ext4_ext_path *path)
123 {
124 ext4_ext_drop_refs(path);
125 kfree(path);
126 }
127
128 /*
129 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
130 * transaction with 'restart_cred' credits. The function drops i_data_sem
131 * when restarting transaction and gets it after transaction is restarted.
132 *
133 * The function returns 0 on success, 1 if transaction had to be restarted,
134 * and < 0 in case of fatal error.
135 */
136 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
137 int check_cred, int restart_cred,
138 int revoke_cred)
139 {
140 int ret;
141 int dropped = 0;
142
143 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
144 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
145 if (dropped)
146 down_write(&EXT4_I(inode)->i_data_sem);
147 return ret;
148 }
149
150 /*
151 * could return:
152 * - EROFS
153 * - ENOMEM
154 */
155 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
156 struct ext4_ext_path *path)
157 {
158 int err = 0;
159
160 if (path->p_bh) {
161 /* path points to block */
162 BUFFER_TRACE(path->p_bh, "get_write_access");
163 err = ext4_journal_get_write_access(handle, inode->i_sb,
164 path->p_bh, EXT4_JTR_NONE);
165 /*
166 * The extent buffer's verified bit will be set again in
167 * __ext4_ext_dirty(). We could leave an inconsistent
168 * buffer if the extents updating procudure break off du
169 * to some error happens, force to check it again.
170 */
171 if (!err)
172 clear_buffer_verified(path->p_bh);
173 }
174 /* path points to leaf/index in inode body */
175 /* we use in-core data, no need to protect them */
176 return err;
177 }
178
179 /*
180 * could return:
181 * - EROFS
182 * - ENOMEM
183 * - EIO
184 */
185 static int __ext4_ext_dirty(const char *where, unsigned int line,
186 handle_t *handle, struct inode *inode,
187 struct ext4_ext_path *path)
188 {
189 int err;
190
191 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
192 if (path->p_bh) {
193 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
194 /* path points to block */
195 err = __ext4_handle_dirty_metadata(where, line, handle,
196 inode, path->p_bh);
197 /* Extents updating done, re-set verified flag */
198 if (!err)
199 set_buffer_verified(path->p_bh);
200 } else {
201 /* path points to leaf/index in inode body */
202 err = ext4_mark_inode_dirty(handle, inode);
203 }
204 return err;
205 }
206
207 #define ext4_ext_dirty(handle, inode, path) \
208 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
209
210 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
211 struct ext4_ext_path *path,
212 ext4_lblk_t block)
213 {
214 if (path) {
215 int depth = path->p_depth;
216 struct ext4_extent *ex;
217
218 /*
219 * Try to predict block placement assuming that we are
220 * filling in a file which will eventually be
221 * non-sparse --- i.e., in the case of libbfd writing
222 * an ELF object sections out-of-order but in a way
223 * the eventually results in a contiguous object or
224 * executable file, or some database extending a table
225 * space file. However, this is actually somewhat
226 * non-ideal if we are writing a sparse file such as
227 * qemu or KVM writing a raw image file that is going
228 * to stay fairly sparse, since it will end up
229 * fragmenting the file system's free space. Maybe we
230 * should have some hueristics or some way to allow
231 * userspace to pass a hint to file system,
232 * especially if the latter case turns out to be
233 * common.
234 */
235 ex = path[depth].p_ext;
236 if (ex) {
237 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
238 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
239
240 if (block > ext_block)
241 return ext_pblk + (block - ext_block);
242 else
243 return ext_pblk - (ext_block - block);
244 }
245
246 /* it looks like index is empty;
247 * try to find starting block from index itself */
248 if (path[depth].p_bh)
249 return path[depth].p_bh->b_blocknr;
250 }
251
252 /* OK. use inode's group */
253 return ext4_inode_to_goal_block(inode);
254 }
255
256 /*
257 * Allocation for a meta data block
258 */
259 static ext4_fsblk_t
260 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
261 struct ext4_ext_path *path,
262 struct ext4_extent *ex, int *err, unsigned int flags)
263 {
264 ext4_fsblk_t goal, newblock;
265
266 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
267 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
268 NULL, err);
269 return newblock;
270 }
271
272 static inline int ext4_ext_space_block(struct inode *inode, int check)
273 {
274 int size;
275
276 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
277 / sizeof(struct ext4_extent);
278 #ifdef AGGRESSIVE_TEST
279 if (!check && size > 6)
280 size = 6;
281 #endif
282 return size;
283 }
284
285 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
286 {
287 int size;
288
289 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
290 / sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292 if (!check && size > 5)
293 size = 5;
294 #endif
295 return size;
296 }
297
298 static inline int ext4_ext_space_root(struct inode *inode, int check)
299 {
300 int size;
301
302 size = sizeof(EXT4_I(inode)->i_data);
303 size -= sizeof(struct ext4_extent_header);
304 size /= sizeof(struct ext4_extent);
305 #ifdef AGGRESSIVE_TEST
306 if (!check && size > 3)
307 size = 3;
308 #endif
309 return size;
310 }
311
312 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
313 {
314 int size;
315
316 size = sizeof(EXT4_I(inode)->i_data);
317 size -= sizeof(struct ext4_extent_header);
318 size /= sizeof(struct ext4_extent_idx);
319 #ifdef AGGRESSIVE_TEST
320 if (!check && size > 4)
321 size = 4;
322 #endif
323 return size;
324 }
325
326 static inline int
327 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
328 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
329 int nofail)
330 {
331 struct ext4_ext_path *path = *ppath;
332 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
333 int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
334
335 if (nofail)
336 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
337
338 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
339 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
340 flags);
341 }
342
343 static int
344 ext4_ext_max_entries(struct inode *inode, int depth)
345 {
346 int max;
347
348 if (depth == ext_depth(inode)) {
349 if (depth == 0)
350 max = ext4_ext_space_root(inode, 1);
351 else
352 max = ext4_ext_space_root_idx(inode, 1);
353 } else {
354 if (depth == 0)
355 max = ext4_ext_space_block(inode, 1);
356 else
357 max = ext4_ext_space_block_idx(inode, 1);
358 }
359
360 return max;
361 }
362
363 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
364 {
365 ext4_fsblk_t block = ext4_ext_pblock(ext);
366 int len = ext4_ext_get_actual_len(ext);
367 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
368
369 /*
370 * We allow neither:
371 * - zero length
372 * - overflow/wrap-around
373 */
374 if (lblock + len <= lblock)
375 return 0;
376 return ext4_inode_block_valid(inode, block, len);
377 }
378
379 static int ext4_valid_extent_idx(struct inode *inode,
380 struct ext4_extent_idx *ext_idx)
381 {
382 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
383
384 return ext4_inode_block_valid(inode, block, 1);
385 }
386
387 static int ext4_valid_extent_entries(struct inode *inode,
388 struct ext4_extent_header *eh,
389 ext4_lblk_t lblk, ext4_fsblk_t *pblk,
390 int depth)
391 {
392 unsigned short entries;
393 ext4_lblk_t lblock = 0;
394 ext4_lblk_t cur = 0;
395
396 if (eh->eh_entries == 0)
397 return 1;
398
399 entries = le16_to_cpu(eh->eh_entries);
400
401 if (depth == 0) {
402 /* leaf entries */
403 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
404
405 /*
406 * The logical block in the first entry should equal to
407 * the number in the index block.
408 */
409 if (depth != ext_depth(inode) &&
410 lblk != le32_to_cpu(ext->ee_block))
411 return 0;
412 while (entries) {
413 if (!ext4_valid_extent(inode, ext))
414 return 0;
415
416 /* Check for overlapping extents */
417 lblock = le32_to_cpu(ext->ee_block);
418 if (lblock < cur) {
419 *pblk = ext4_ext_pblock(ext);
420 return 0;
421 }
422 cur = lblock + ext4_ext_get_actual_len(ext);
423 ext++;
424 entries--;
425 }
426 } else {
427 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
428
429 /*
430 * The logical block in the first entry should equal to
431 * the number in the parent index block.
432 */
433 if (depth != ext_depth(inode) &&
434 lblk != le32_to_cpu(ext_idx->ei_block))
435 return 0;
436 while (entries) {
437 if (!ext4_valid_extent_idx(inode, ext_idx))
438 return 0;
439
440 /* Check for overlapping index extents */
441 lblock = le32_to_cpu(ext_idx->ei_block);
442 if (lblock < cur) {
443 *pblk = ext4_idx_pblock(ext_idx);
444 return 0;
445 }
446 ext_idx++;
447 entries--;
448 cur = lblock + 1;
449 }
450 }
451 return 1;
452 }
453
454 static int __ext4_ext_check(const char *function, unsigned int line,
455 struct inode *inode, struct ext4_extent_header *eh,
456 int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
457 {
458 const char *error_msg;
459 int max = 0, err = -EFSCORRUPTED;
460
461 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
462 error_msg = "invalid magic";
463 goto corrupted;
464 }
465 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
466 error_msg = "unexpected eh_depth";
467 goto corrupted;
468 }
469 if (unlikely(eh->eh_max == 0)) {
470 error_msg = "invalid eh_max";
471 goto corrupted;
472 }
473 max = ext4_ext_max_entries(inode, depth);
474 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
475 error_msg = "too large eh_max";
476 goto corrupted;
477 }
478 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
479 error_msg = "invalid eh_entries";
480 goto corrupted;
481 }
482 if (unlikely((eh->eh_entries == 0) && (depth > 0))) {
483 error_msg = "eh_entries is 0 but eh_depth is > 0";
484 goto corrupted;
485 }
486 if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
487 error_msg = "invalid extent entries";
488 goto corrupted;
489 }
490 if (unlikely(depth > 32)) {
491 error_msg = "too large eh_depth";
492 goto corrupted;
493 }
494 /* Verify checksum on non-root extent tree nodes */
495 if (ext_depth(inode) != depth &&
496 !ext4_extent_block_csum_verify(inode, eh)) {
497 error_msg = "extent tree corrupted";
498 err = -EFSBADCRC;
499 goto corrupted;
500 }
501 return 0;
502
503 corrupted:
504 ext4_error_inode_err(inode, function, line, 0, -err,
505 "pblk %llu bad header/extent: %s - magic %x, "
506 "entries %u, max %u(%u), depth %u(%u)",
507 (unsigned long long) pblk, error_msg,
508 le16_to_cpu(eh->eh_magic),
509 le16_to_cpu(eh->eh_entries),
510 le16_to_cpu(eh->eh_max),
511 max, le16_to_cpu(eh->eh_depth), depth);
512 return err;
513 }
514
515 #define ext4_ext_check(inode, eh, depth, pblk) \
516 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
517
518 int ext4_ext_check_inode(struct inode *inode)
519 {
520 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
521 }
522
523 static void ext4_cache_extents(struct inode *inode,
524 struct ext4_extent_header *eh)
525 {
526 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
527 ext4_lblk_t prev = 0;
528 int i;
529
530 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
531 unsigned int status = EXTENT_STATUS_WRITTEN;
532 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
533 int len = ext4_ext_get_actual_len(ex);
534
535 if (prev && (prev != lblk))
536 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
537 EXTENT_STATUS_HOLE);
538
539 if (ext4_ext_is_unwritten(ex))
540 status = EXTENT_STATUS_UNWRITTEN;
541 ext4_es_cache_extent(inode, lblk, len,
542 ext4_ext_pblock(ex), status);
543 prev = lblk + len;
544 }
545 }
546
547 static struct buffer_head *
548 __read_extent_tree_block(const char *function, unsigned int line,
549 struct inode *inode, struct ext4_extent_idx *idx,
550 int depth, int flags)
551 {
552 struct buffer_head *bh;
553 int err;
554 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
555 ext4_fsblk_t pblk;
556
557 if (flags & EXT4_EX_NOFAIL)
558 gfp_flags |= __GFP_NOFAIL;
559
560 pblk = ext4_idx_pblock(idx);
561 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
562 if (unlikely(!bh))
563 return ERR_PTR(-ENOMEM);
564
565 if (!bh_uptodate_or_lock(bh)) {
566 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
567 err = ext4_read_bh(bh, 0, NULL);
568 if (err < 0)
569 goto errout;
570 }
571 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
572 return bh;
573 err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
574 depth, pblk, le32_to_cpu(idx->ei_block));
575 if (err)
576 goto errout;
577 set_buffer_verified(bh);
578 /*
579 * If this is a leaf block, cache all of its entries
580 */
581 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
582 struct ext4_extent_header *eh = ext_block_hdr(bh);
583 ext4_cache_extents(inode, eh);
584 }
585 return bh;
586 errout:
587 put_bh(bh);
588 return ERR_PTR(err);
589
590 }
591
592 #define read_extent_tree_block(inode, idx, depth, flags) \
593 __read_extent_tree_block(__func__, __LINE__, (inode), (idx), \
594 (depth), (flags))
595
596 /*
597 * This function is called to cache a file's extent information in the
598 * extent status tree
599 */
600 int ext4_ext_precache(struct inode *inode)
601 {
602 struct ext4_inode_info *ei = EXT4_I(inode);
603 struct ext4_ext_path *path = NULL;
604 struct buffer_head *bh;
605 int i = 0, depth, ret = 0;
606
607 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
608 return 0; /* not an extent-mapped inode */
609
610 down_read(&ei->i_data_sem);
611 depth = ext_depth(inode);
612
613 /* Don't cache anything if there are no external extent blocks */
614 if (!depth) {
615 up_read(&ei->i_data_sem);
616 return ret;
617 }
618
619 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
620 GFP_NOFS);
621 if (path == NULL) {
622 up_read(&ei->i_data_sem);
623 return -ENOMEM;
624 }
625
626 path[0].p_hdr = ext_inode_hdr(inode);
627 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
628 if (ret)
629 goto out;
630 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
631 while (i >= 0) {
632 /*
633 * If this is a leaf block or we've reached the end of
634 * the index block, go up
635 */
636 if ((i == depth) ||
637 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
638 brelse(path[i].p_bh);
639 path[i].p_bh = NULL;
640 i--;
641 continue;
642 }
643 bh = read_extent_tree_block(inode, path[i].p_idx++,
644 depth - i - 1,
645 EXT4_EX_FORCE_CACHE);
646 if (IS_ERR(bh)) {
647 ret = PTR_ERR(bh);
648 break;
649 }
650 i++;
651 path[i].p_bh = bh;
652 path[i].p_hdr = ext_block_hdr(bh);
653 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
654 }
655 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
656 out:
657 up_read(&ei->i_data_sem);
658 ext4_free_ext_path(path);
659 return ret;
660 }
661
662 #ifdef EXT_DEBUG
663 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
664 {
665 int k, l = path->p_depth;
666
667 ext_debug(inode, "path:");
668 for (k = 0; k <= l; k++, path++) {
669 if (path->p_idx) {
670 ext_debug(inode, " %d->%llu",
671 le32_to_cpu(path->p_idx->ei_block),
672 ext4_idx_pblock(path->p_idx));
673 } else if (path->p_ext) {
674 ext_debug(inode, " %d:[%d]%d:%llu ",
675 le32_to_cpu(path->p_ext->ee_block),
676 ext4_ext_is_unwritten(path->p_ext),
677 ext4_ext_get_actual_len(path->p_ext),
678 ext4_ext_pblock(path->p_ext));
679 } else
680 ext_debug(inode, " []");
681 }
682 ext_debug(inode, "\n");
683 }
684
685 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
686 {
687 int depth = ext_depth(inode);
688 struct ext4_extent_header *eh;
689 struct ext4_extent *ex;
690 int i;
691
692 if (!path)
693 return;
694
695 eh = path[depth].p_hdr;
696 ex = EXT_FIRST_EXTENT(eh);
697
698 ext_debug(inode, "Displaying leaf extents\n");
699
700 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
701 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
702 ext4_ext_is_unwritten(ex),
703 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
704 }
705 ext_debug(inode, "\n");
706 }
707
708 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
709 ext4_fsblk_t newblock, int level)
710 {
711 int depth = ext_depth(inode);
712 struct ext4_extent *ex;
713
714 if (depth != level) {
715 struct ext4_extent_idx *idx;
716 idx = path[level].p_idx;
717 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
718 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
719 level, le32_to_cpu(idx->ei_block),
720 ext4_idx_pblock(idx), newblock);
721 idx++;
722 }
723
724 return;
725 }
726
727 ex = path[depth].p_ext;
728 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
729 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
730 le32_to_cpu(ex->ee_block),
731 ext4_ext_pblock(ex),
732 ext4_ext_is_unwritten(ex),
733 ext4_ext_get_actual_len(ex),
734 newblock);
735 ex++;
736 }
737 }
738
739 #else
740 #define ext4_ext_show_path(inode, path)
741 #define ext4_ext_show_leaf(inode, path)
742 #define ext4_ext_show_move(inode, path, newblock, level)
743 #endif
744
745 /*
746 * ext4_ext_binsearch_idx:
747 * binary search for the closest index of the given block
748 * the header must be checked before calling this
749 */
750 static void
751 ext4_ext_binsearch_idx(struct inode *inode,
752 struct ext4_ext_path *path, ext4_lblk_t block)
753 {
754 struct ext4_extent_header *eh = path->p_hdr;
755 struct ext4_extent_idx *r, *l, *m;
756
757
758 ext_debug(inode, "binsearch for %u(idx): ", block);
759
760 l = EXT_FIRST_INDEX(eh) + 1;
761 r = EXT_LAST_INDEX(eh);
762 while (l <= r) {
763 m = l + (r - l) / 2;
764 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
765 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
766 r, le32_to_cpu(r->ei_block));
767
768 if (block < le32_to_cpu(m->ei_block))
769 r = m - 1;
770 else
771 l = m + 1;
772 }
773
774 path->p_idx = l - 1;
775 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
776 ext4_idx_pblock(path->p_idx));
777
778 #ifdef CHECK_BINSEARCH
779 {
780 struct ext4_extent_idx *chix, *ix;
781 int k;
782
783 chix = ix = EXT_FIRST_INDEX(eh);
784 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
785 if (k != 0 && le32_to_cpu(ix->ei_block) <=
786 le32_to_cpu(ix[-1].ei_block)) {
787 printk(KERN_DEBUG "k=%d, ix=0x%p, "
788 "first=0x%p\n", k,
789 ix, EXT_FIRST_INDEX(eh));
790 printk(KERN_DEBUG "%u <= %u\n",
791 le32_to_cpu(ix->ei_block),
792 le32_to_cpu(ix[-1].ei_block));
793 }
794 BUG_ON(k && le32_to_cpu(ix->ei_block)
795 <= le32_to_cpu(ix[-1].ei_block));
796 if (block < le32_to_cpu(ix->ei_block))
797 break;
798 chix = ix;
799 }
800 BUG_ON(chix != path->p_idx);
801 }
802 #endif
803
804 }
805
806 /*
807 * ext4_ext_binsearch:
808 * binary search for closest extent of the given block
809 * the header must be checked before calling this
810 */
811 static void
812 ext4_ext_binsearch(struct inode *inode,
813 struct ext4_ext_path *path, ext4_lblk_t block)
814 {
815 struct ext4_extent_header *eh = path->p_hdr;
816 struct ext4_extent *r, *l, *m;
817
818 if (eh->eh_entries == 0) {
819 /*
820 * this leaf is empty:
821 * we get such a leaf in split/add case
822 */
823 return;
824 }
825
826 ext_debug(inode, "binsearch for %u: ", block);
827
828 l = EXT_FIRST_EXTENT(eh) + 1;
829 r = EXT_LAST_EXTENT(eh);
830
831 while (l <= r) {
832 m = l + (r - l) / 2;
833 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
834 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
835 r, le32_to_cpu(r->ee_block));
836
837 if (block < le32_to_cpu(m->ee_block))
838 r = m - 1;
839 else
840 l = m + 1;
841 }
842
843 path->p_ext = l - 1;
844 ext_debug(inode, " -> %d:%llu:[%d]%d ",
845 le32_to_cpu(path->p_ext->ee_block),
846 ext4_ext_pblock(path->p_ext),
847 ext4_ext_is_unwritten(path->p_ext),
848 ext4_ext_get_actual_len(path->p_ext));
849
850 #ifdef CHECK_BINSEARCH
851 {
852 struct ext4_extent *chex, *ex;
853 int k;
854
855 chex = ex = EXT_FIRST_EXTENT(eh);
856 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
857 BUG_ON(k && le32_to_cpu(ex->ee_block)
858 <= le32_to_cpu(ex[-1].ee_block));
859 if (block < le32_to_cpu(ex->ee_block))
860 break;
861 chex = ex;
862 }
863 BUG_ON(chex != path->p_ext);
864 }
865 #endif
866
867 }
868
869 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
870 {
871 struct ext4_extent_header *eh;
872
873 eh = ext_inode_hdr(inode);
874 eh->eh_depth = 0;
875 eh->eh_entries = 0;
876 eh->eh_magic = EXT4_EXT_MAGIC;
877 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
878 eh->eh_generation = 0;
879 ext4_mark_inode_dirty(handle, inode);
880 }
881
882 struct ext4_ext_path *
883 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
884 struct ext4_ext_path **orig_path, int flags)
885 {
886 struct ext4_extent_header *eh;
887 struct buffer_head *bh;
888 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
889 short int depth, i, ppos = 0;
890 int ret;
891 gfp_t gfp_flags = GFP_NOFS;
892
893 if (flags & EXT4_EX_NOFAIL)
894 gfp_flags |= __GFP_NOFAIL;
895
896 eh = ext_inode_hdr(inode);
897 depth = ext_depth(inode);
898 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
899 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
900 depth);
901 ret = -EFSCORRUPTED;
902 goto err;
903 }
904
905 if (path) {
906 ext4_ext_drop_refs(path);
907 if (depth > path[0].p_maxdepth) {
908 kfree(path);
909 *orig_path = path = NULL;
910 }
911 }
912 if (!path) {
913 /* account possible depth increase */
914 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
915 gfp_flags);
916 if (unlikely(!path))
917 return ERR_PTR(-ENOMEM);
918 path[0].p_maxdepth = depth + 1;
919 }
920 path[0].p_hdr = eh;
921 path[0].p_bh = NULL;
922
923 i = depth;
924 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
925 ext4_cache_extents(inode, eh);
926 /* walk through the tree */
927 while (i) {
928 ext_debug(inode, "depth %d: num %d, max %d\n",
929 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
930
931 ext4_ext_binsearch_idx(inode, path + ppos, block);
932 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
933 path[ppos].p_depth = i;
934 path[ppos].p_ext = NULL;
935
936 bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
937 if (IS_ERR(bh)) {
938 ret = PTR_ERR(bh);
939 goto err;
940 }
941
942 eh = ext_block_hdr(bh);
943 ppos++;
944 path[ppos].p_bh = bh;
945 path[ppos].p_hdr = eh;
946 }
947
948 path[ppos].p_depth = i;
949 path[ppos].p_ext = NULL;
950 path[ppos].p_idx = NULL;
951
952 /* find extent */
953 ext4_ext_binsearch(inode, path + ppos, block);
954 /* if not an empty leaf */
955 if (path[ppos].p_ext)
956 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
957
958 ext4_ext_show_path(inode, path);
959
960 if (orig_path)
961 *orig_path = path;
962 return path;
963
964 err:
965 ext4_free_ext_path(path);
966 if (orig_path)
967 *orig_path = NULL;
968 return ERR_PTR(ret);
969 }
970
971 /*
972 * ext4_ext_insert_index:
973 * insert new index [@logical;@ptr] into the block at @curp;
974 * check where to insert: before @curp or after @curp
975 */
976 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
977 struct ext4_ext_path *curp,
978 int logical, ext4_fsblk_t ptr)
979 {
980 struct ext4_extent_idx *ix;
981 int len, err;
982
983 err = ext4_ext_get_access(handle, inode, curp);
984 if (err)
985 return err;
986
987 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
988 EXT4_ERROR_INODE(inode,
989 "logical %d == ei_block %d!",
990 logical, le32_to_cpu(curp->p_idx->ei_block));
991 return -EFSCORRUPTED;
992 }
993
994 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
995 >= le16_to_cpu(curp->p_hdr->eh_max))) {
996 EXT4_ERROR_INODE(inode,
997 "eh_entries %d >= eh_max %d!",
998 le16_to_cpu(curp->p_hdr->eh_entries),
999 le16_to_cpu(curp->p_hdr->eh_max));
1000 return -EFSCORRUPTED;
1001 }
1002
1003 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
1004 /* insert after */
1005 ext_debug(inode, "insert new index %d after: %llu\n",
1006 logical, ptr);
1007 ix = curp->p_idx + 1;
1008 } else {
1009 /* insert before */
1010 ext_debug(inode, "insert new index %d before: %llu\n",
1011 logical, ptr);
1012 ix = curp->p_idx;
1013 }
1014
1015 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1016 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1017 return -EFSCORRUPTED;
1018 }
1019
1020 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1021 BUG_ON(len < 0);
1022 if (len > 0) {
1023 ext_debug(inode, "insert new index %d: "
1024 "move %d indices from 0x%p to 0x%p\n",
1025 logical, len, ix, ix + 1);
1026 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1027 }
1028
1029 ix->ei_block = cpu_to_le32(logical);
1030 ext4_idx_store_pblock(ix, ptr);
1031 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1032
1033 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1034 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1035 return -EFSCORRUPTED;
1036 }
1037
1038 err = ext4_ext_dirty(handle, inode, curp);
1039 ext4_std_error(inode->i_sb, err);
1040
1041 return err;
1042 }
1043
1044 /*
1045 * ext4_ext_split:
1046 * inserts new subtree into the path, using free index entry
1047 * at depth @at:
1048 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1049 * - makes decision where to split
1050 * - moves remaining extents and index entries (right to the split point)
1051 * into the newly allocated blocks
1052 * - initializes subtree
1053 */
1054 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1055 unsigned int flags,
1056 struct ext4_ext_path *path,
1057 struct ext4_extent *newext, int at)
1058 {
1059 struct buffer_head *bh = NULL;
1060 int depth = ext_depth(inode);
1061 struct ext4_extent_header *neh;
1062 struct ext4_extent_idx *fidx;
1063 int i = at, k, m, a;
1064 ext4_fsblk_t newblock, oldblock;
1065 __le32 border;
1066 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1067 gfp_t gfp_flags = GFP_NOFS;
1068 int err = 0;
1069 size_t ext_size = 0;
1070
1071 if (flags & EXT4_EX_NOFAIL)
1072 gfp_flags |= __GFP_NOFAIL;
1073
1074 /* make decision: where to split? */
1075 /* FIXME: now decision is simplest: at current extent */
1076
1077 /* if current leaf will be split, then we should use
1078 * border from split point */
1079 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1080 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1081 return -EFSCORRUPTED;
1082 }
1083 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1084 border = path[depth].p_ext[1].ee_block;
1085 ext_debug(inode, "leaf will be split."
1086 " next leaf starts at %d\n",
1087 le32_to_cpu(border));
1088 } else {
1089 border = newext->ee_block;
1090 ext_debug(inode, "leaf will be added."
1091 " next leaf starts at %d\n",
1092 le32_to_cpu(border));
1093 }
1094
1095 /*
1096 * If error occurs, then we break processing
1097 * and mark filesystem read-only. index won't
1098 * be inserted and tree will be in consistent
1099 * state. Next mount will repair buffers too.
1100 */
1101
1102 /*
1103 * Get array to track all allocated blocks.
1104 * We need this to handle errors and free blocks
1105 * upon them.
1106 */
1107 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1108 if (!ablocks)
1109 return -ENOMEM;
1110
1111 /* allocate all needed blocks */
1112 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1113 for (a = 0; a < depth - at; a++) {
1114 newblock = ext4_ext_new_meta_block(handle, inode, path,
1115 newext, &err, flags);
1116 if (newblock == 0)
1117 goto cleanup;
1118 ablocks[a] = newblock;
1119 }
1120
1121 /* initialize new leaf */
1122 newblock = ablocks[--a];
1123 if (unlikely(newblock == 0)) {
1124 EXT4_ERROR_INODE(inode, "newblock == 0!");
1125 err = -EFSCORRUPTED;
1126 goto cleanup;
1127 }
1128 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1129 if (unlikely(!bh)) {
1130 err = -ENOMEM;
1131 goto cleanup;
1132 }
1133 lock_buffer(bh);
1134
1135 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1136 EXT4_JTR_NONE);
1137 if (err)
1138 goto cleanup;
1139
1140 neh = ext_block_hdr(bh);
1141 neh->eh_entries = 0;
1142 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1143 neh->eh_magic = EXT4_EXT_MAGIC;
1144 neh->eh_depth = 0;
1145 neh->eh_generation = 0;
1146
1147 /* move remainder of path[depth] to the new leaf */
1148 if (unlikely(path[depth].p_hdr->eh_entries !=
1149 path[depth].p_hdr->eh_max)) {
1150 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1151 path[depth].p_hdr->eh_entries,
1152 path[depth].p_hdr->eh_max);
1153 err = -EFSCORRUPTED;
1154 goto cleanup;
1155 }
1156 /* start copy from next extent */
1157 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1158 ext4_ext_show_move(inode, path, newblock, depth);
1159 if (m) {
1160 struct ext4_extent *ex;
1161 ex = EXT_FIRST_EXTENT(neh);
1162 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1163 le16_add_cpu(&neh->eh_entries, m);
1164 }
1165
1166 /* zero out unused area in the extent block */
1167 ext_size = sizeof(struct ext4_extent_header) +
1168 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1169 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1170 ext4_extent_block_csum_set(inode, neh);
1171 set_buffer_uptodate(bh);
1172 unlock_buffer(bh);
1173
1174 err = ext4_handle_dirty_metadata(handle, inode, bh);
1175 if (err)
1176 goto cleanup;
1177 brelse(bh);
1178 bh = NULL;
1179
1180 /* correct old leaf */
1181 if (m) {
1182 err = ext4_ext_get_access(handle, inode, path + depth);
1183 if (err)
1184 goto cleanup;
1185 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1186 err = ext4_ext_dirty(handle, inode, path + depth);
1187 if (err)
1188 goto cleanup;
1189
1190 }
1191
1192 /* create intermediate indexes */
1193 k = depth - at - 1;
1194 if (unlikely(k < 0)) {
1195 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1196 err = -EFSCORRUPTED;
1197 goto cleanup;
1198 }
1199 if (k)
1200 ext_debug(inode, "create %d intermediate indices\n", k);
1201 /* insert new index into current index block */
1202 /* current depth stored in i var */
1203 i = depth - 1;
1204 while (k--) {
1205 oldblock = newblock;
1206 newblock = ablocks[--a];
1207 bh = sb_getblk(inode->i_sb, newblock);
1208 if (unlikely(!bh)) {
1209 err = -ENOMEM;
1210 goto cleanup;
1211 }
1212 lock_buffer(bh);
1213
1214 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1215 EXT4_JTR_NONE);
1216 if (err)
1217 goto cleanup;
1218
1219 neh = ext_block_hdr(bh);
1220 neh->eh_entries = cpu_to_le16(1);
1221 neh->eh_magic = EXT4_EXT_MAGIC;
1222 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1223 neh->eh_depth = cpu_to_le16(depth - i);
1224 neh->eh_generation = 0;
1225 fidx = EXT_FIRST_INDEX(neh);
1226 fidx->ei_block = border;
1227 ext4_idx_store_pblock(fidx, oldblock);
1228
1229 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1230 i, newblock, le32_to_cpu(border), oldblock);
1231
1232 /* move remainder of path[i] to the new index block */
1233 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1234 EXT_LAST_INDEX(path[i].p_hdr))) {
1235 EXT4_ERROR_INODE(inode,
1236 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1237 le32_to_cpu(path[i].p_ext->ee_block));
1238 err = -EFSCORRUPTED;
1239 goto cleanup;
1240 }
1241 /* start copy indexes */
1242 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1243 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1244 EXT_MAX_INDEX(path[i].p_hdr));
1245 ext4_ext_show_move(inode, path, newblock, i);
1246 if (m) {
1247 memmove(++fidx, path[i].p_idx,
1248 sizeof(struct ext4_extent_idx) * m);
1249 le16_add_cpu(&neh->eh_entries, m);
1250 }
1251 /* zero out unused area in the extent block */
1252 ext_size = sizeof(struct ext4_extent_header) +
1253 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1254 memset(bh->b_data + ext_size, 0,
1255 inode->i_sb->s_blocksize - ext_size);
1256 ext4_extent_block_csum_set(inode, neh);
1257 set_buffer_uptodate(bh);
1258 unlock_buffer(bh);
1259
1260 err = ext4_handle_dirty_metadata(handle, inode, bh);
1261 if (err)
1262 goto cleanup;
1263 brelse(bh);
1264 bh = NULL;
1265
1266 /* correct old index */
1267 if (m) {
1268 err = ext4_ext_get_access(handle, inode, path + i);
1269 if (err)
1270 goto cleanup;
1271 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1272 err = ext4_ext_dirty(handle, inode, path + i);
1273 if (err)
1274 goto cleanup;
1275 }
1276
1277 i--;
1278 }
1279
1280 /* insert new index */
1281 err = ext4_ext_insert_index(handle, inode, path + at,
1282 le32_to_cpu(border), newblock);
1283
1284 cleanup:
1285 if (bh) {
1286 if (buffer_locked(bh))
1287 unlock_buffer(bh);
1288 brelse(bh);
1289 }
1290
1291 if (err) {
1292 /* free all allocated blocks in error case */
1293 for (i = 0; i < depth; i++) {
1294 if (!ablocks[i])
1295 continue;
1296 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1297 EXT4_FREE_BLOCKS_METADATA);
1298 }
1299 }
1300 kfree(ablocks);
1301
1302 return err;
1303 }
1304
1305 /*
1306 * ext4_ext_grow_indepth:
1307 * implements tree growing procedure:
1308 * - allocates new block
1309 * - moves top-level data (index block or leaf) into the new block
1310 * - initializes new top-level, creating index that points to the
1311 * just created block
1312 */
1313 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1314 unsigned int flags)
1315 {
1316 struct ext4_extent_header *neh;
1317 struct buffer_head *bh;
1318 ext4_fsblk_t newblock, goal = 0;
1319 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1320 int err = 0;
1321 size_t ext_size = 0;
1322
1323 /* Try to prepend new index to old one */
1324 if (ext_depth(inode))
1325 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1326 if (goal > le32_to_cpu(es->s_first_data_block)) {
1327 flags |= EXT4_MB_HINT_TRY_GOAL;
1328 goal--;
1329 } else
1330 goal = ext4_inode_to_goal_block(inode);
1331 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1332 NULL, &err);
1333 if (newblock == 0)
1334 return err;
1335
1336 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1337 if (unlikely(!bh))
1338 return -ENOMEM;
1339 lock_buffer(bh);
1340
1341 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1342 EXT4_JTR_NONE);
1343 if (err) {
1344 unlock_buffer(bh);
1345 goto out;
1346 }
1347
1348 ext_size = sizeof(EXT4_I(inode)->i_data);
1349 /* move top-level index/leaf into new block */
1350 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1351 /* zero out unused area in the extent block */
1352 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1353
1354 /* set size of new block */
1355 neh = ext_block_hdr(bh);
1356 /* old root could have indexes or leaves
1357 * so calculate e_max right way */
1358 if (ext_depth(inode))
1359 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1360 else
1361 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1362 neh->eh_magic = EXT4_EXT_MAGIC;
1363 ext4_extent_block_csum_set(inode, neh);
1364 set_buffer_uptodate(bh);
1365 set_buffer_verified(bh);
1366 unlock_buffer(bh);
1367
1368 err = ext4_handle_dirty_metadata(handle, inode, bh);
1369 if (err)
1370 goto out;
1371
1372 /* Update top-level index: num,max,pointer */
1373 neh = ext_inode_hdr(inode);
1374 neh->eh_entries = cpu_to_le16(1);
1375 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1376 if (neh->eh_depth == 0) {
1377 /* Root extent block becomes index block */
1378 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1379 EXT_FIRST_INDEX(neh)->ei_block =
1380 EXT_FIRST_EXTENT(neh)->ee_block;
1381 }
1382 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1383 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1384 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1385 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1386
1387 le16_add_cpu(&neh->eh_depth, 1);
1388 err = ext4_mark_inode_dirty(handle, inode);
1389 out:
1390 brelse(bh);
1391
1392 return err;
1393 }
1394
1395 /*
1396 * ext4_ext_create_new_leaf:
1397 * finds empty index and adds new leaf.
1398 * if no free index is found, then it requests in-depth growing.
1399 */
1400 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1401 unsigned int mb_flags,
1402 unsigned int gb_flags,
1403 struct ext4_ext_path **ppath,
1404 struct ext4_extent *newext)
1405 {
1406 struct ext4_ext_path *path = *ppath;
1407 struct ext4_ext_path *curp;
1408 int depth, i, err = 0;
1409
1410 repeat:
1411 i = depth = ext_depth(inode);
1412
1413 /* walk up to the tree and look for free index entry */
1414 curp = path + depth;
1415 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1416 i--;
1417 curp--;
1418 }
1419
1420 /* we use already allocated block for index block,
1421 * so subsequent data blocks should be contiguous */
1422 if (EXT_HAS_FREE_INDEX(curp)) {
1423 /* if we found index with free entry, then use that
1424 * entry: create all needed subtree and add new leaf */
1425 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1426 if (err)
1427 goto out;
1428
1429 /* refill path */
1430 path = ext4_find_extent(inode,
1431 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1432 ppath, gb_flags);
1433 if (IS_ERR(path))
1434 err = PTR_ERR(path);
1435 } else {
1436 /* tree is full, time to grow in depth */
1437 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1438 if (err)
1439 goto out;
1440
1441 /* refill path */
1442 path = ext4_find_extent(inode,
1443 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1444 ppath, gb_flags);
1445 if (IS_ERR(path)) {
1446 err = PTR_ERR(path);
1447 goto out;
1448 }
1449
1450 /*
1451 * only first (depth 0 -> 1) produces free space;
1452 * in all other cases we have to split the grown tree
1453 */
1454 depth = ext_depth(inode);
1455 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1456 /* now we need to split */
1457 goto repeat;
1458 }
1459 }
1460
1461 out:
1462 return err;
1463 }
1464
1465 /*
1466 * search the closest allocated block to the left for *logical
1467 * and returns it at @logical + it's physical address at @phys
1468 * if *logical is the smallest allocated block, the function
1469 * returns 0 at @phys
1470 * return value contains 0 (success) or error code
1471 */
1472 static int ext4_ext_search_left(struct inode *inode,
1473 struct ext4_ext_path *path,
1474 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1475 {
1476 struct ext4_extent_idx *ix;
1477 struct ext4_extent *ex;
1478 int depth, ee_len;
1479
1480 if (unlikely(path == NULL)) {
1481 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1482 return -EFSCORRUPTED;
1483 }
1484 depth = path->p_depth;
1485 *phys = 0;
1486
1487 if (depth == 0 && path->p_ext == NULL)
1488 return 0;
1489
1490 /* usually extent in the path covers blocks smaller
1491 * then *logical, but it can be that extent is the
1492 * first one in the file */
1493
1494 ex = path[depth].p_ext;
1495 ee_len = ext4_ext_get_actual_len(ex);
1496 if (*logical < le32_to_cpu(ex->ee_block)) {
1497 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1498 EXT4_ERROR_INODE(inode,
1499 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1500 *logical, le32_to_cpu(ex->ee_block));
1501 return -EFSCORRUPTED;
1502 }
1503 while (--depth >= 0) {
1504 ix = path[depth].p_idx;
1505 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1506 EXT4_ERROR_INODE(inode,
1507 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1508 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1509 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block),
1510 depth);
1511 return -EFSCORRUPTED;
1512 }
1513 }
1514 return 0;
1515 }
1516
1517 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1518 EXT4_ERROR_INODE(inode,
1519 "logical %d < ee_block %d + ee_len %d!",
1520 *logical, le32_to_cpu(ex->ee_block), ee_len);
1521 return -EFSCORRUPTED;
1522 }
1523
1524 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1525 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1526 return 0;
1527 }
1528
1529 /*
1530 * Search the closest allocated block to the right for *logical
1531 * and returns it at @logical + it's physical address at @phys.
1532 * If not exists, return 0 and @phys is set to 0. We will return
1533 * 1 which means we found an allocated block and ret_ex is valid.
1534 * Or return a (< 0) error code.
1535 */
1536 static int ext4_ext_search_right(struct inode *inode,
1537 struct ext4_ext_path *path,
1538 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1539 struct ext4_extent *ret_ex)
1540 {
1541 struct buffer_head *bh = NULL;
1542 struct ext4_extent_header *eh;
1543 struct ext4_extent_idx *ix;
1544 struct ext4_extent *ex;
1545 int depth; /* Note, NOT eh_depth; depth from top of tree */
1546 int ee_len;
1547
1548 if (unlikely(path == NULL)) {
1549 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1550 return -EFSCORRUPTED;
1551 }
1552 depth = path->p_depth;
1553 *phys = 0;
1554
1555 if (depth == 0 && path->p_ext == NULL)
1556 return 0;
1557
1558 /* usually extent in the path covers blocks smaller
1559 * then *logical, but it can be that extent is the
1560 * first one in the file */
1561
1562 ex = path[depth].p_ext;
1563 ee_len = ext4_ext_get_actual_len(ex);
1564 if (*logical < le32_to_cpu(ex->ee_block)) {
1565 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1566 EXT4_ERROR_INODE(inode,
1567 "first_extent(path[%d].p_hdr) != ex",
1568 depth);
1569 return -EFSCORRUPTED;
1570 }
1571 while (--depth >= 0) {
1572 ix = path[depth].p_idx;
1573 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1574 EXT4_ERROR_INODE(inode,
1575 "ix != EXT_FIRST_INDEX *logical %d!",
1576 *logical);
1577 return -EFSCORRUPTED;
1578 }
1579 }
1580 goto found_extent;
1581 }
1582
1583 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1584 EXT4_ERROR_INODE(inode,
1585 "logical %d < ee_block %d + ee_len %d!",
1586 *logical, le32_to_cpu(ex->ee_block), ee_len);
1587 return -EFSCORRUPTED;
1588 }
1589
1590 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1591 /* next allocated block in this leaf */
1592 ex++;
1593 goto found_extent;
1594 }
1595
1596 /* go up and search for index to the right */
1597 while (--depth >= 0) {
1598 ix = path[depth].p_idx;
1599 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1600 goto got_index;
1601 }
1602
1603 /* we've gone up to the root and found no index to the right */
1604 return 0;
1605
1606 got_index:
1607 /* we've found index to the right, let's
1608 * follow it and find the closest allocated
1609 * block to the right */
1610 ix++;
1611 while (++depth < path->p_depth) {
1612 /* subtract from p_depth to get proper eh_depth */
1613 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1614 if (IS_ERR(bh))
1615 return PTR_ERR(bh);
1616 eh = ext_block_hdr(bh);
1617 ix = EXT_FIRST_INDEX(eh);
1618 put_bh(bh);
1619 }
1620
1621 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1622 if (IS_ERR(bh))
1623 return PTR_ERR(bh);
1624 eh = ext_block_hdr(bh);
1625 ex = EXT_FIRST_EXTENT(eh);
1626 found_extent:
1627 *logical = le32_to_cpu(ex->ee_block);
1628 *phys = ext4_ext_pblock(ex);
1629 if (ret_ex)
1630 *ret_ex = *ex;
1631 if (bh)
1632 put_bh(bh);
1633 return 1;
1634 }
1635
1636 /*
1637 * ext4_ext_next_allocated_block:
1638 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1639 * NOTE: it considers block number from index entry as
1640 * allocated block. Thus, index entries have to be consistent
1641 * with leaves.
1642 */
1643 ext4_lblk_t
1644 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1645 {
1646 int depth;
1647
1648 BUG_ON(path == NULL);
1649 depth = path->p_depth;
1650
1651 if (depth == 0 && path->p_ext == NULL)
1652 return EXT_MAX_BLOCKS;
1653
1654 while (depth >= 0) {
1655 struct ext4_ext_path *p = &path[depth];
1656
1657 if (depth == path->p_depth) {
1658 /* leaf */
1659 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1660 return le32_to_cpu(p->p_ext[1].ee_block);
1661 } else {
1662 /* index */
1663 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1664 return le32_to_cpu(p->p_idx[1].ei_block);
1665 }
1666 depth--;
1667 }
1668
1669 return EXT_MAX_BLOCKS;
1670 }
1671
1672 /*
1673 * ext4_ext_next_leaf_block:
1674 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1675 */
1676 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1677 {
1678 int depth;
1679
1680 BUG_ON(path == NULL);
1681 depth = path->p_depth;
1682
1683 /* zero-tree has no leaf blocks at all */
1684 if (depth == 0)
1685 return EXT_MAX_BLOCKS;
1686
1687 /* go to index block */
1688 depth--;
1689
1690 while (depth >= 0) {
1691 if (path[depth].p_idx !=
1692 EXT_LAST_INDEX(path[depth].p_hdr))
1693 return (ext4_lblk_t)
1694 le32_to_cpu(path[depth].p_idx[1].ei_block);
1695 depth--;
1696 }
1697
1698 return EXT_MAX_BLOCKS;
1699 }
1700
1701 /*
1702 * ext4_ext_correct_indexes:
1703 * if leaf gets modified and modified extent is first in the leaf,
1704 * then we have to correct all indexes above.
1705 * TODO: do we need to correct tree in all cases?
1706 */
1707 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1708 struct ext4_ext_path *path)
1709 {
1710 struct ext4_extent_header *eh;
1711 int depth = ext_depth(inode);
1712 struct ext4_extent *ex;
1713 __le32 border;
1714 int k, err = 0;
1715
1716 eh = path[depth].p_hdr;
1717 ex = path[depth].p_ext;
1718
1719 if (unlikely(ex == NULL || eh == NULL)) {
1720 EXT4_ERROR_INODE(inode,
1721 "ex %p == NULL or eh %p == NULL", ex, eh);
1722 return -EFSCORRUPTED;
1723 }
1724
1725 if (depth == 0) {
1726 /* there is no tree at all */
1727 return 0;
1728 }
1729
1730 if (ex != EXT_FIRST_EXTENT(eh)) {
1731 /* we correct tree if first leaf got modified only */
1732 return 0;
1733 }
1734
1735 /*
1736 * TODO: we need correction if border is smaller than current one
1737 */
1738 k = depth - 1;
1739 border = path[depth].p_ext->ee_block;
1740 err = ext4_ext_get_access(handle, inode, path + k);
1741 if (err)
1742 return err;
1743 path[k].p_idx->ei_block = border;
1744 err = ext4_ext_dirty(handle, inode, path + k);
1745 if (err)
1746 return err;
1747
1748 while (k--) {
1749 /* change all left-side indexes */
1750 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1751 break;
1752 err = ext4_ext_get_access(handle, inode, path + k);
1753 if (err)
1754 break;
1755 path[k].p_idx->ei_block = border;
1756 err = ext4_ext_dirty(handle, inode, path + k);
1757 if (err)
1758 break;
1759 }
1760
1761 return err;
1762 }
1763
1764 static int ext4_can_extents_be_merged(struct inode *inode,
1765 struct ext4_extent *ex1,
1766 struct ext4_extent *ex2)
1767 {
1768 unsigned short ext1_ee_len, ext2_ee_len;
1769
1770 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1771 return 0;
1772
1773 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1774 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1775
1776 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1777 le32_to_cpu(ex2->ee_block))
1778 return 0;
1779
1780 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1781 return 0;
1782
1783 if (ext4_ext_is_unwritten(ex1) &&
1784 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1785 return 0;
1786 #ifdef AGGRESSIVE_TEST
1787 if (ext1_ee_len >= 4)
1788 return 0;
1789 #endif
1790
1791 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1792 return 1;
1793 return 0;
1794 }
1795
1796 /*
1797 * This function tries to merge the "ex" extent to the next extent in the tree.
1798 * It always tries to merge towards right. If you want to merge towards
1799 * left, pass "ex - 1" as argument instead of "ex".
1800 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1801 * 1 if they got merged.
1802 */
1803 static int ext4_ext_try_to_merge_right(struct inode *inode,
1804 struct ext4_ext_path *path,
1805 struct ext4_extent *ex)
1806 {
1807 struct ext4_extent_header *eh;
1808 unsigned int depth, len;
1809 int merge_done = 0, unwritten;
1810
1811 depth = ext_depth(inode);
1812 BUG_ON(path[depth].p_hdr == NULL);
1813 eh = path[depth].p_hdr;
1814
1815 while (ex < EXT_LAST_EXTENT(eh)) {
1816 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1817 break;
1818 /* merge with next extent! */
1819 unwritten = ext4_ext_is_unwritten(ex);
1820 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1821 + ext4_ext_get_actual_len(ex + 1));
1822 if (unwritten)
1823 ext4_ext_mark_unwritten(ex);
1824
1825 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1826 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1827 * sizeof(struct ext4_extent);
1828 memmove(ex + 1, ex + 2, len);
1829 }
1830 le16_add_cpu(&eh->eh_entries, -1);
1831 merge_done = 1;
1832 WARN_ON(eh->eh_entries == 0);
1833 if (!eh->eh_entries)
1834 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1835 }
1836
1837 return merge_done;
1838 }
1839
1840 /*
1841 * This function does a very simple check to see if we can collapse
1842 * an extent tree with a single extent tree leaf block into the inode.
1843 */
1844 static void ext4_ext_try_to_merge_up(handle_t *handle,
1845 struct inode *inode,
1846 struct ext4_ext_path *path)
1847 {
1848 size_t s;
1849 unsigned max_root = ext4_ext_space_root(inode, 0);
1850 ext4_fsblk_t blk;
1851
1852 if ((path[0].p_depth != 1) ||
1853 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1854 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1855 return;
1856
1857 /*
1858 * We need to modify the block allocation bitmap and the block
1859 * group descriptor to release the extent tree block. If we
1860 * can't get the journal credits, give up.
1861 */
1862 if (ext4_journal_extend(handle, 2,
1863 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1864 return;
1865
1866 /*
1867 * Copy the extent data up to the inode
1868 */
1869 blk = ext4_idx_pblock(path[0].p_idx);
1870 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1871 sizeof(struct ext4_extent_idx);
1872 s += sizeof(struct ext4_extent_header);
1873
1874 path[1].p_maxdepth = path[0].p_maxdepth;
1875 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1876 path[0].p_depth = 0;
1877 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1878 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1879 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1880
1881 brelse(path[1].p_bh);
1882 path[1].p_bh = NULL;
1883 ext4_free_blocks(handle, inode, NULL, blk, 1,
1884 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1885 }
1886
1887 /*
1888 * This function tries to merge the @ex extent to neighbours in the tree, then
1889 * tries to collapse the extent tree into the inode.
1890 */
1891 static void ext4_ext_try_to_merge(handle_t *handle,
1892 struct inode *inode,
1893 struct ext4_ext_path *path,
1894 struct ext4_extent *ex)
1895 {
1896 struct ext4_extent_header *eh;
1897 unsigned int depth;
1898 int merge_done = 0;
1899
1900 depth = ext_depth(inode);
1901 BUG_ON(path[depth].p_hdr == NULL);
1902 eh = path[depth].p_hdr;
1903
1904 if (ex > EXT_FIRST_EXTENT(eh))
1905 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1906
1907 if (!merge_done)
1908 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1909
1910 ext4_ext_try_to_merge_up(handle, inode, path);
1911 }
1912
1913 /*
1914 * check if a portion of the "newext" extent overlaps with an
1915 * existing extent.
1916 *
1917 * If there is an overlap discovered, it updates the length of the newext
1918 * such that there will be no overlap, and then returns 1.
1919 * If there is no overlap found, it returns 0.
1920 */
1921 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1922 struct inode *inode,
1923 struct ext4_extent *newext,
1924 struct ext4_ext_path *path)
1925 {
1926 ext4_lblk_t b1, b2;
1927 unsigned int depth, len1;
1928 unsigned int ret = 0;
1929
1930 b1 = le32_to_cpu(newext->ee_block);
1931 len1 = ext4_ext_get_actual_len(newext);
1932 depth = ext_depth(inode);
1933 if (!path[depth].p_ext)
1934 goto out;
1935 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1936
1937 /*
1938 * get the next allocated block if the extent in the path
1939 * is before the requested block(s)
1940 */
1941 if (b2 < b1) {
1942 b2 = ext4_ext_next_allocated_block(path);
1943 if (b2 == EXT_MAX_BLOCKS)
1944 goto out;
1945 b2 = EXT4_LBLK_CMASK(sbi, b2);
1946 }
1947
1948 /* check for wrap through zero on extent logical start block*/
1949 if (b1 + len1 < b1) {
1950 len1 = EXT_MAX_BLOCKS - b1;
1951 newext->ee_len = cpu_to_le16(len1);
1952 ret = 1;
1953 }
1954
1955 /* check for overlap */
1956 if (b1 + len1 > b2) {
1957 newext->ee_len = cpu_to_le16(b2 - b1);
1958 ret = 1;
1959 }
1960 out:
1961 return ret;
1962 }
1963
1964 /*
1965 * ext4_ext_insert_extent:
1966 * tries to merge requested extent into the existing extent or
1967 * inserts requested extent as new one into the tree,
1968 * creating new leaf in the no-space case.
1969 */
1970 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1971 struct ext4_ext_path **ppath,
1972 struct ext4_extent *newext, int gb_flags)
1973 {
1974 struct ext4_ext_path *path = *ppath;
1975 struct ext4_extent_header *eh;
1976 struct ext4_extent *ex, *fex;
1977 struct ext4_extent *nearex; /* nearest extent */
1978 struct ext4_ext_path *npath = NULL;
1979 int depth, len, err;
1980 ext4_lblk_t next;
1981 int mb_flags = 0, unwritten;
1982
1983 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1984 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1985 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1986 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1987 return -EFSCORRUPTED;
1988 }
1989 depth = ext_depth(inode);
1990 ex = path[depth].p_ext;
1991 eh = path[depth].p_hdr;
1992 if (unlikely(path[depth].p_hdr == NULL)) {
1993 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1994 return -EFSCORRUPTED;
1995 }
1996
1997 /* try to insert block into found extent and return */
1998 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1999
2000 /*
2001 * Try to see whether we should rather test the extent on
2002 * right from ex, or from the left of ex. This is because
2003 * ext4_find_extent() can return either extent on the
2004 * left, or on the right from the searched position. This
2005 * will make merging more effective.
2006 */
2007 if (ex < EXT_LAST_EXTENT(eh) &&
2008 (le32_to_cpu(ex->ee_block) +
2009 ext4_ext_get_actual_len(ex) <
2010 le32_to_cpu(newext->ee_block))) {
2011 ex += 1;
2012 goto prepend;
2013 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2014 (le32_to_cpu(newext->ee_block) +
2015 ext4_ext_get_actual_len(newext) <
2016 le32_to_cpu(ex->ee_block)))
2017 ex -= 1;
2018
2019 /* Try to append newex to the ex */
2020 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2021 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2022 "(from %llu)\n",
2023 ext4_ext_is_unwritten(newext),
2024 ext4_ext_get_actual_len(newext),
2025 le32_to_cpu(ex->ee_block),
2026 ext4_ext_is_unwritten(ex),
2027 ext4_ext_get_actual_len(ex),
2028 ext4_ext_pblock(ex));
2029 err = ext4_ext_get_access(handle, inode,
2030 path + depth);
2031 if (err)
2032 return err;
2033 unwritten = ext4_ext_is_unwritten(ex);
2034 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2035 + ext4_ext_get_actual_len(newext));
2036 if (unwritten)
2037 ext4_ext_mark_unwritten(ex);
2038 nearex = ex;
2039 goto merge;
2040 }
2041
2042 prepend:
2043 /* Try to prepend newex to the ex */
2044 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2045 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2046 "(from %llu)\n",
2047 le32_to_cpu(newext->ee_block),
2048 ext4_ext_is_unwritten(newext),
2049 ext4_ext_get_actual_len(newext),
2050 le32_to_cpu(ex->ee_block),
2051 ext4_ext_is_unwritten(ex),
2052 ext4_ext_get_actual_len(ex),
2053 ext4_ext_pblock(ex));
2054 err = ext4_ext_get_access(handle, inode,
2055 path + depth);
2056 if (err)
2057 return err;
2058
2059 unwritten = ext4_ext_is_unwritten(ex);
2060 ex->ee_block = newext->ee_block;
2061 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2062 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2063 + ext4_ext_get_actual_len(newext));
2064 if (unwritten)
2065 ext4_ext_mark_unwritten(ex);
2066 nearex = ex;
2067 goto merge;
2068 }
2069 }
2070
2071 depth = ext_depth(inode);
2072 eh = path[depth].p_hdr;
2073 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2074 goto has_space;
2075
2076 /* probably next leaf has space for us? */
2077 fex = EXT_LAST_EXTENT(eh);
2078 next = EXT_MAX_BLOCKS;
2079 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2080 next = ext4_ext_next_leaf_block(path);
2081 if (next != EXT_MAX_BLOCKS) {
2082 ext_debug(inode, "next leaf block - %u\n", next);
2083 BUG_ON(npath != NULL);
2084 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2085 if (IS_ERR(npath))
2086 return PTR_ERR(npath);
2087 BUG_ON(npath->p_depth != path->p_depth);
2088 eh = npath[depth].p_hdr;
2089 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2090 ext_debug(inode, "next leaf isn't full(%d)\n",
2091 le16_to_cpu(eh->eh_entries));
2092 path = npath;
2093 goto has_space;
2094 }
2095 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2096 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2097 }
2098
2099 /*
2100 * There is no free space in the found leaf.
2101 * We're gonna add a new leaf in the tree.
2102 */
2103 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2104 mb_flags |= EXT4_MB_USE_RESERVED;
2105 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2106 ppath, newext);
2107 if (err)
2108 goto cleanup;
2109 path = *ppath;
2110 depth = ext_depth(inode);
2111 eh = path[depth].p_hdr;
2112
2113 has_space:
2114 nearex = path[depth].p_ext;
2115
2116 err = ext4_ext_get_access(handle, inode, path + depth);
2117 if (err)
2118 goto cleanup;
2119
2120 if (!nearex) {
2121 /* there is no extent in this leaf, create first one */
2122 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2123 le32_to_cpu(newext->ee_block),
2124 ext4_ext_pblock(newext),
2125 ext4_ext_is_unwritten(newext),
2126 ext4_ext_get_actual_len(newext));
2127 nearex = EXT_FIRST_EXTENT(eh);
2128 } else {
2129 if (le32_to_cpu(newext->ee_block)
2130 > le32_to_cpu(nearex->ee_block)) {
2131 /* Insert after */
2132 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2133 "nearest %p\n",
2134 le32_to_cpu(newext->ee_block),
2135 ext4_ext_pblock(newext),
2136 ext4_ext_is_unwritten(newext),
2137 ext4_ext_get_actual_len(newext),
2138 nearex);
2139 nearex++;
2140 } else {
2141 /* Insert before */
2142 BUG_ON(newext->ee_block == nearex->ee_block);
2143 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2144 "nearest %p\n",
2145 le32_to_cpu(newext->ee_block),
2146 ext4_ext_pblock(newext),
2147 ext4_ext_is_unwritten(newext),
2148 ext4_ext_get_actual_len(newext),
2149 nearex);
2150 }
2151 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2152 if (len > 0) {
2153 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2154 "move %d extents from 0x%p to 0x%p\n",
2155 le32_to_cpu(newext->ee_block),
2156 ext4_ext_pblock(newext),
2157 ext4_ext_is_unwritten(newext),
2158 ext4_ext_get_actual_len(newext),
2159 len, nearex, nearex + 1);
2160 memmove(nearex + 1, nearex,
2161 len * sizeof(struct ext4_extent));
2162 }
2163 }
2164
2165 le16_add_cpu(&eh->eh_entries, 1);
2166 path[depth].p_ext = nearex;
2167 nearex->ee_block = newext->ee_block;
2168 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2169 nearex->ee_len = newext->ee_len;
2170
2171 merge:
2172 /* try to merge extents */
2173 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2174 ext4_ext_try_to_merge(handle, inode, path, nearex);
2175
2176
2177 /* time to correct all indexes above */
2178 err = ext4_ext_correct_indexes(handle, inode, path);
2179 if (err)
2180 goto cleanup;
2181
2182 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2183
2184 cleanup:
2185 ext4_free_ext_path(npath);
2186 return err;
2187 }
2188
2189 static int ext4_fill_es_cache_info(struct inode *inode,
2190 ext4_lblk_t block, ext4_lblk_t num,
2191 struct fiemap_extent_info *fieinfo)
2192 {
2193 ext4_lblk_t next, end = block + num - 1;
2194 struct extent_status es;
2195 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2196 unsigned int flags;
2197 int err;
2198
2199 while (block <= end) {
2200 next = 0;
2201 flags = 0;
2202 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2203 break;
2204 if (ext4_es_is_unwritten(&es))
2205 flags |= FIEMAP_EXTENT_UNWRITTEN;
2206 if (ext4_es_is_delayed(&es))
2207 flags |= (FIEMAP_EXTENT_DELALLOC |
2208 FIEMAP_EXTENT_UNKNOWN);
2209 if (ext4_es_is_hole(&es))
2210 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2211 if (next == 0)
2212 flags |= FIEMAP_EXTENT_LAST;
2213 if (flags & (FIEMAP_EXTENT_DELALLOC|
2214 EXT4_FIEMAP_EXTENT_HOLE))
2215 es.es_pblk = 0;
2216 else
2217 es.es_pblk = ext4_es_pblock(&es);
2218 err = fiemap_fill_next_extent(fieinfo,
2219 (__u64)es.es_lblk << blksize_bits,
2220 (__u64)es.es_pblk << blksize_bits,
2221 (__u64)es.es_len << blksize_bits,
2222 flags);
2223 if (next == 0)
2224 break;
2225 block = next;
2226 if (err < 0)
2227 return err;
2228 if (err == 1)
2229 return 0;
2230 }
2231 return 0;
2232 }
2233
2234
2235 /*
2236 * ext4_ext_find_hole - find hole around given block according to the given path
2237 * @inode: inode we lookup in
2238 * @path: path in extent tree to @lblk
2239 * @lblk: pointer to logical block around which we want to determine hole
2240 *
2241 * Determine hole length (and start if easily possible) around given logical
2242 * block. We don't try too hard to find the beginning of the hole but @path
2243 * actually points to extent before @lblk, we provide it.
2244 *
2245 * The function returns the length of a hole starting at @lblk. We update @lblk
2246 * to the beginning of the hole if we managed to find it.
2247 */
2248 static ext4_lblk_t ext4_ext_find_hole(struct inode *inode,
2249 struct ext4_ext_path *path,
2250 ext4_lblk_t *lblk)
2251 {
2252 int depth = ext_depth(inode);
2253 struct ext4_extent *ex;
2254 ext4_lblk_t len;
2255
2256 ex = path[depth].p_ext;
2257 if (ex == NULL) {
2258 /* there is no extent yet, so gap is [0;-] */
2259 *lblk = 0;
2260 len = EXT_MAX_BLOCKS;
2261 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2262 len = le32_to_cpu(ex->ee_block) - *lblk;
2263 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2264 + ext4_ext_get_actual_len(ex)) {
2265 ext4_lblk_t next;
2266
2267 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2268 next = ext4_ext_next_allocated_block(path);
2269 BUG_ON(next == *lblk);
2270 len = next - *lblk;
2271 } else {
2272 BUG();
2273 }
2274 return len;
2275 }
2276
2277 /*
2278 * ext4_ext_rm_idx:
2279 * removes index from the index block.
2280 */
2281 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2282 struct ext4_ext_path *path, int depth)
2283 {
2284 int err;
2285 ext4_fsblk_t leaf;
2286
2287 /* free index block */
2288 depth--;
2289 path = path + depth;
2290 leaf = ext4_idx_pblock(path->p_idx);
2291 if (unlikely(path->p_hdr->eh_entries == 0)) {
2292 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2293 return -EFSCORRUPTED;
2294 }
2295 err = ext4_ext_get_access(handle, inode, path);
2296 if (err)
2297 return err;
2298
2299 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2300 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2301 len *= sizeof(struct ext4_extent_idx);
2302 memmove(path->p_idx, path->p_idx + 1, len);
2303 }
2304
2305 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2306 err = ext4_ext_dirty(handle, inode, path);
2307 if (err)
2308 return err;
2309 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2310 trace_ext4_ext_rm_idx(inode, leaf);
2311
2312 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2313 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2314
2315 while (--depth >= 0) {
2316 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2317 break;
2318 path--;
2319 err = ext4_ext_get_access(handle, inode, path);
2320 if (err)
2321 break;
2322 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2323 err = ext4_ext_dirty(handle, inode, path);
2324 if (err)
2325 break;
2326 }
2327 return err;
2328 }
2329
2330 /*
2331 * ext4_ext_calc_credits_for_single_extent:
2332 * This routine returns max. credits that needed to insert an extent
2333 * to the extent tree.
2334 * When pass the actual path, the caller should calculate credits
2335 * under i_data_sem.
2336 */
2337 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2338 struct ext4_ext_path *path)
2339 {
2340 if (path) {
2341 int depth = ext_depth(inode);
2342 int ret = 0;
2343
2344 /* probably there is space in leaf? */
2345 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2346 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2347
2348 /*
2349 * There are some space in the leaf tree, no
2350 * need to account for leaf block credit
2351 *
2352 * bitmaps and block group descriptor blocks
2353 * and other metadata blocks still need to be
2354 * accounted.
2355 */
2356 /* 1 bitmap, 1 block group descriptor */
2357 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2358 return ret;
2359 }
2360 }
2361
2362 return ext4_chunk_trans_blocks(inode, nrblocks);
2363 }
2364
2365 /*
2366 * How many index/leaf blocks need to change/allocate to add @extents extents?
2367 *
2368 * If we add a single extent, then in the worse case, each tree level
2369 * index/leaf need to be changed in case of the tree split.
2370 *
2371 * If more extents are inserted, they could cause the whole tree split more
2372 * than once, but this is really rare.
2373 */
2374 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2375 {
2376 int index;
2377 int depth;
2378
2379 /* If we are converting the inline data, only one is needed here. */
2380 if (ext4_has_inline_data(inode))
2381 return 1;
2382
2383 depth = ext_depth(inode);
2384
2385 if (extents <= 1)
2386 index = depth * 2;
2387 else
2388 index = depth * 3;
2389
2390 return index;
2391 }
2392
2393 static inline int get_default_free_blocks_flags(struct inode *inode)
2394 {
2395 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2396 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2397 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2398 else if (ext4_should_journal_data(inode))
2399 return EXT4_FREE_BLOCKS_FORGET;
2400 return 0;
2401 }
2402
2403 /*
2404 * ext4_rereserve_cluster - increment the reserved cluster count when
2405 * freeing a cluster with a pending reservation
2406 *
2407 * @inode - file containing the cluster
2408 * @lblk - logical block in cluster to be reserved
2409 *
2410 * Increments the reserved cluster count and adjusts quota in a bigalloc
2411 * file system when freeing a partial cluster containing at least one
2412 * delayed and unwritten block. A partial cluster meeting that
2413 * requirement will have a pending reservation. If so, the
2414 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2415 * defer reserved and allocated space accounting to a subsequent call
2416 * to this function.
2417 */
2418 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2419 {
2420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2421 struct ext4_inode_info *ei = EXT4_I(inode);
2422
2423 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2424
2425 spin_lock(&ei->i_block_reservation_lock);
2426 ei->i_reserved_data_blocks++;
2427 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2428 spin_unlock(&ei->i_block_reservation_lock);
2429
2430 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2431 ext4_remove_pending(inode, lblk);
2432 }
2433
2434 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2435 struct ext4_extent *ex,
2436 struct partial_cluster *partial,
2437 ext4_lblk_t from, ext4_lblk_t to)
2438 {
2439 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2440 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2441 ext4_fsblk_t last_pblk, pblk;
2442 ext4_lblk_t num;
2443 int flags;
2444
2445 /* only extent tail removal is allowed */
2446 if (from < le32_to_cpu(ex->ee_block) ||
2447 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2448 ext4_error(sbi->s_sb,
2449 "strange request: removal(2) %u-%u from %u:%u",
2450 from, to, le32_to_cpu(ex->ee_block), ee_len);
2451 return 0;
2452 }
2453
2454 #ifdef EXTENTS_STATS
2455 spin_lock(&sbi->s_ext_stats_lock);
2456 sbi->s_ext_blocks += ee_len;
2457 sbi->s_ext_extents++;
2458 if (ee_len < sbi->s_ext_min)
2459 sbi->s_ext_min = ee_len;
2460 if (ee_len > sbi->s_ext_max)
2461 sbi->s_ext_max = ee_len;
2462 if (ext_depth(inode) > sbi->s_depth_max)
2463 sbi->s_depth_max = ext_depth(inode);
2464 spin_unlock(&sbi->s_ext_stats_lock);
2465 #endif
2466
2467 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2468
2469 /*
2470 * if we have a partial cluster, and it's different from the
2471 * cluster of the last block in the extent, we free it
2472 */
2473 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2474
2475 if (partial->state != initial &&
2476 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2477 if (partial->state == tofree) {
2478 flags = get_default_free_blocks_flags(inode);
2479 if (ext4_is_pending(inode, partial->lblk))
2480 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2481 ext4_free_blocks(handle, inode, NULL,
2482 EXT4_C2B(sbi, partial->pclu),
2483 sbi->s_cluster_ratio, flags);
2484 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2485 ext4_rereserve_cluster(inode, partial->lblk);
2486 }
2487 partial->state = initial;
2488 }
2489
2490 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2491 pblk = ext4_ext_pblock(ex) + ee_len - num;
2492
2493 /*
2494 * We free the partial cluster at the end of the extent (if any),
2495 * unless the cluster is used by another extent (partial_cluster
2496 * state is nofree). If a partial cluster exists here, it must be
2497 * shared with the last block in the extent.
2498 */
2499 flags = get_default_free_blocks_flags(inode);
2500
2501 /* partial, left end cluster aligned, right end unaligned */
2502 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2503 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2504 (partial->state != nofree)) {
2505 if (ext4_is_pending(inode, to))
2506 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2507 ext4_free_blocks(handle, inode, NULL,
2508 EXT4_PBLK_CMASK(sbi, last_pblk),
2509 sbi->s_cluster_ratio, flags);
2510 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2511 ext4_rereserve_cluster(inode, to);
2512 partial->state = initial;
2513 flags = get_default_free_blocks_flags(inode);
2514 }
2515
2516 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2517
2518 /*
2519 * For bigalloc file systems, we never free a partial cluster
2520 * at the beginning of the extent. Instead, we check to see if we
2521 * need to free it on a subsequent call to ext4_remove_blocks,
2522 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2523 */
2524 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2525 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2526
2527 /* reset the partial cluster if we've freed past it */
2528 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2529 partial->state = initial;
2530
2531 /*
2532 * If we've freed the entire extent but the beginning is not left
2533 * cluster aligned and is not marked as ineligible for freeing we
2534 * record the partial cluster at the beginning of the extent. It
2535 * wasn't freed by the preceding ext4_free_blocks() call, and we
2536 * need to look farther to the left to determine if it's to be freed
2537 * (not shared with another extent). Else, reset the partial
2538 * cluster - we're either done freeing or the beginning of the
2539 * extent is left cluster aligned.
2540 */
2541 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2542 if (partial->state == initial) {
2543 partial->pclu = EXT4_B2C(sbi, pblk);
2544 partial->lblk = from;
2545 partial->state = tofree;
2546 }
2547 } else {
2548 partial->state = initial;
2549 }
2550
2551 return 0;
2552 }
2553
2554 /*
2555 * ext4_ext_rm_leaf() Removes the extents associated with the
2556 * blocks appearing between "start" and "end". Both "start"
2557 * and "end" must appear in the same extent or EIO is returned.
2558 *
2559 * @handle: The journal handle
2560 * @inode: The files inode
2561 * @path: The path to the leaf
2562 * @partial_cluster: The cluster which we'll have to free if all extents
2563 * has been released from it. However, if this value is
2564 * negative, it's a cluster just to the right of the
2565 * punched region and it must not be freed.
2566 * @start: The first block to remove
2567 * @end: The last block to remove
2568 */
2569 static int
2570 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2571 struct ext4_ext_path *path,
2572 struct partial_cluster *partial,
2573 ext4_lblk_t start, ext4_lblk_t end)
2574 {
2575 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2576 int err = 0, correct_index = 0;
2577 int depth = ext_depth(inode), credits, revoke_credits;
2578 struct ext4_extent_header *eh;
2579 ext4_lblk_t a, b;
2580 unsigned num;
2581 ext4_lblk_t ex_ee_block;
2582 unsigned short ex_ee_len;
2583 unsigned unwritten = 0;
2584 struct ext4_extent *ex;
2585 ext4_fsblk_t pblk;
2586
2587 /* the header must be checked already in ext4_ext_remove_space() */
2588 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2589 if (!path[depth].p_hdr)
2590 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2591 eh = path[depth].p_hdr;
2592 if (unlikely(path[depth].p_hdr == NULL)) {
2593 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2594 return -EFSCORRUPTED;
2595 }
2596 /* find where to start removing */
2597 ex = path[depth].p_ext;
2598 if (!ex)
2599 ex = EXT_LAST_EXTENT(eh);
2600
2601 ex_ee_block = le32_to_cpu(ex->ee_block);
2602 ex_ee_len = ext4_ext_get_actual_len(ex);
2603
2604 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2605
2606 while (ex >= EXT_FIRST_EXTENT(eh) &&
2607 ex_ee_block + ex_ee_len > start) {
2608
2609 if (ext4_ext_is_unwritten(ex))
2610 unwritten = 1;
2611 else
2612 unwritten = 0;
2613
2614 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2615 unwritten, ex_ee_len);
2616 path[depth].p_ext = ex;
2617
2618 a = max(ex_ee_block, start);
2619 b = min(ex_ee_block + ex_ee_len - 1, end);
2620
2621 ext_debug(inode, " border %u:%u\n", a, b);
2622
2623 /* If this extent is beyond the end of the hole, skip it */
2624 if (end < ex_ee_block) {
2625 /*
2626 * We're going to skip this extent and move to another,
2627 * so note that its first cluster is in use to avoid
2628 * freeing it when removing blocks. Eventually, the
2629 * right edge of the truncated/punched region will
2630 * be just to the left.
2631 */
2632 if (sbi->s_cluster_ratio > 1) {
2633 pblk = ext4_ext_pblock(ex);
2634 partial->pclu = EXT4_B2C(sbi, pblk);
2635 partial->state = nofree;
2636 }
2637 ex--;
2638 ex_ee_block = le32_to_cpu(ex->ee_block);
2639 ex_ee_len = ext4_ext_get_actual_len(ex);
2640 continue;
2641 } else if (b != ex_ee_block + ex_ee_len - 1) {
2642 EXT4_ERROR_INODE(inode,
2643 "can not handle truncate %u:%u "
2644 "on extent %u:%u",
2645 start, end, ex_ee_block,
2646 ex_ee_block + ex_ee_len - 1);
2647 err = -EFSCORRUPTED;
2648 goto out;
2649 } else if (a != ex_ee_block) {
2650 /* remove tail of the extent */
2651 num = a - ex_ee_block;
2652 } else {
2653 /* remove whole extent: excellent! */
2654 num = 0;
2655 }
2656 /*
2657 * 3 for leaf, sb, and inode plus 2 (bmap and group
2658 * descriptor) for each block group; assume two block
2659 * groups plus ex_ee_len/blocks_per_block_group for
2660 * the worst case
2661 */
2662 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2663 if (ex == EXT_FIRST_EXTENT(eh)) {
2664 correct_index = 1;
2665 credits += (ext_depth(inode)) + 1;
2666 }
2667 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2668 /*
2669 * We may end up freeing some index blocks and data from the
2670 * punched range. Note that partial clusters are accounted for
2671 * by ext4_free_data_revoke_credits().
2672 */
2673 revoke_credits =
2674 ext4_free_metadata_revoke_credits(inode->i_sb,
2675 ext_depth(inode)) +
2676 ext4_free_data_revoke_credits(inode, b - a + 1);
2677
2678 err = ext4_datasem_ensure_credits(handle, inode, credits,
2679 credits, revoke_credits);
2680 if (err) {
2681 if (err > 0)
2682 err = -EAGAIN;
2683 goto out;
2684 }
2685
2686 err = ext4_ext_get_access(handle, inode, path + depth);
2687 if (err)
2688 goto out;
2689
2690 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2691 if (err)
2692 goto out;
2693
2694 if (num == 0)
2695 /* this extent is removed; mark slot entirely unused */
2696 ext4_ext_store_pblock(ex, 0);
2697
2698 ex->ee_len = cpu_to_le16(num);
2699 /*
2700 * Do not mark unwritten if all the blocks in the
2701 * extent have been removed.
2702 */
2703 if (unwritten && num)
2704 ext4_ext_mark_unwritten(ex);
2705 /*
2706 * If the extent was completely released,
2707 * we need to remove it from the leaf
2708 */
2709 if (num == 0) {
2710 if (end != EXT_MAX_BLOCKS - 1) {
2711 /*
2712 * For hole punching, we need to scoot all the
2713 * extents up when an extent is removed so that
2714 * we dont have blank extents in the middle
2715 */
2716 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2717 sizeof(struct ext4_extent));
2718
2719 /* Now get rid of the one at the end */
2720 memset(EXT_LAST_EXTENT(eh), 0,
2721 sizeof(struct ext4_extent));
2722 }
2723 le16_add_cpu(&eh->eh_entries, -1);
2724 }
2725
2726 err = ext4_ext_dirty(handle, inode, path + depth);
2727 if (err)
2728 goto out;
2729
2730 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2731 ext4_ext_pblock(ex));
2732 ex--;
2733 ex_ee_block = le32_to_cpu(ex->ee_block);
2734 ex_ee_len = ext4_ext_get_actual_len(ex);
2735 }
2736
2737 if (correct_index && eh->eh_entries)
2738 err = ext4_ext_correct_indexes(handle, inode, path);
2739
2740 /*
2741 * If there's a partial cluster and at least one extent remains in
2742 * the leaf, free the partial cluster if it isn't shared with the
2743 * current extent. If it is shared with the current extent
2744 * we reset the partial cluster because we've reached the start of the
2745 * truncated/punched region and we're done removing blocks.
2746 */
2747 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2748 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2749 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2750 int flags = get_default_free_blocks_flags(inode);
2751
2752 if (ext4_is_pending(inode, partial->lblk))
2753 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2754 ext4_free_blocks(handle, inode, NULL,
2755 EXT4_C2B(sbi, partial->pclu),
2756 sbi->s_cluster_ratio, flags);
2757 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2758 ext4_rereserve_cluster(inode, partial->lblk);
2759 }
2760 partial->state = initial;
2761 }
2762
2763 /* if this leaf is free, then we should
2764 * remove it from index block above */
2765 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2766 err = ext4_ext_rm_idx(handle, inode, path, depth);
2767
2768 out:
2769 return err;
2770 }
2771
2772 /*
2773 * ext4_ext_more_to_rm:
2774 * returns 1 if current index has to be freed (even partial)
2775 */
2776 static int
2777 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2778 {
2779 BUG_ON(path->p_idx == NULL);
2780
2781 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2782 return 0;
2783
2784 /*
2785 * if truncate on deeper level happened, it wasn't partial,
2786 * so we have to consider current index for truncation
2787 */
2788 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2789 return 0;
2790 return 1;
2791 }
2792
2793 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2794 ext4_lblk_t end)
2795 {
2796 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2797 int depth = ext_depth(inode);
2798 struct ext4_ext_path *path = NULL;
2799 struct partial_cluster partial;
2800 handle_t *handle;
2801 int i = 0, err = 0;
2802
2803 partial.pclu = 0;
2804 partial.lblk = 0;
2805 partial.state = initial;
2806
2807 ext_debug(inode, "truncate since %u to %u\n", start, end);
2808
2809 /* probably first extent we're gonna free will be last in block */
2810 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2811 depth + 1,
2812 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2813 if (IS_ERR(handle))
2814 return PTR_ERR(handle);
2815
2816 again:
2817 trace_ext4_ext_remove_space(inode, start, end, depth);
2818
2819 /*
2820 * Check if we are removing extents inside the extent tree. If that
2821 * is the case, we are going to punch a hole inside the extent tree
2822 * so we have to check whether we need to split the extent covering
2823 * the last block to remove so we can easily remove the part of it
2824 * in ext4_ext_rm_leaf().
2825 */
2826 if (end < EXT_MAX_BLOCKS - 1) {
2827 struct ext4_extent *ex;
2828 ext4_lblk_t ee_block, ex_end, lblk;
2829 ext4_fsblk_t pblk;
2830
2831 /* find extent for or closest extent to this block */
2832 path = ext4_find_extent(inode, end, NULL,
2833 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2834 if (IS_ERR(path)) {
2835 ext4_journal_stop(handle);
2836 return PTR_ERR(path);
2837 }
2838 depth = ext_depth(inode);
2839 /* Leaf not may not exist only if inode has no blocks at all */
2840 ex = path[depth].p_ext;
2841 if (!ex) {
2842 if (depth) {
2843 EXT4_ERROR_INODE(inode,
2844 "path[%d].p_hdr == NULL",
2845 depth);
2846 err = -EFSCORRUPTED;
2847 }
2848 goto out;
2849 }
2850
2851 ee_block = le32_to_cpu(ex->ee_block);
2852 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2853
2854 /*
2855 * See if the last block is inside the extent, if so split
2856 * the extent at 'end' block so we can easily remove the
2857 * tail of the first part of the split extent in
2858 * ext4_ext_rm_leaf().
2859 */
2860 if (end >= ee_block && end < ex_end) {
2861
2862 /*
2863 * If we're going to split the extent, note that
2864 * the cluster containing the block after 'end' is
2865 * in use to avoid freeing it when removing blocks.
2866 */
2867 if (sbi->s_cluster_ratio > 1) {
2868 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2869 partial.pclu = EXT4_B2C(sbi, pblk);
2870 partial.state = nofree;
2871 }
2872
2873 /*
2874 * Split the extent in two so that 'end' is the last
2875 * block in the first new extent. Also we should not
2876 * fail removing space due to ENOSPC so try to use
2877 * reserved block if that happens.
2878 */
2879 err = ext4_force_split_extent_at(handle, inode, &path,
2880 end + 1, 1);
2881 if (err < 0)
2882 goto out;
2883
2884 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2885 partial.state == initial) {
2886 /*
2887 * If we're punching, there's an extent to the right.
2888 * If the partial cluster hasn't been set, set it to
2889 * that extent's first cluster and its state to nofree
2890 * so it won't be freed should it contain blocks to be
2891 * removed. If it's already set (tofree/nofree), we're
2892 * retrying and keep the original partial cluster info
2893 * so a cluster marked tofree as a result of earlier
2894 * extent removal is not lost.
2895 */
2896 lblk = ex_end + 1;
2897 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2898 NULL);
2899 if (err < 0)
2900 goto out;
2901 if (pblk) {
2902 partial.pclu = EXT4_B2C(sbi, pblk);
2903 partial.state = nofree;
2904 }
2905 }
2906 }
2907 /*
2908 * We start scanning from right side, freeing all the blocks
2909 * after i_size and walking into the tree depth-wise.
2910 */
2911 depth = ext_depth(inode);
2912 if (path) {
2913 int k = i = depth;
2914 while (--k > 0)
2915 path[k].p_block =
2916 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2917 } else {
2918 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2919 GFP_NOFS | __GFP_NOFAIL);
2920 if (path == NULL) {
2921 ext4_journal_stop(handle);
2922 return -ENOMEM;
2923 }
2924 path[0].p_maxdepth = path[0].p_depth = depth;
2925 path[0].p_hdr = ext_inode_hdr(inode);
2926 i = 0;
2927
2928 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2929 err = -EFSCORRUPTED;
2930 goto out;
2931 }
2932 }
2933 err = 0;
2934
2935 while (i >= 0 && err == 0) {
2936 if (i == depth) {
2937 /* this is leaf block */
2938 err = ext4_ext_rm_leaf(handle, inode, path,
2939 &partial, start, end);
2940 /* root level has p_bh == NULL, brelse() eats this */
2941 brelse(path[i].p_bh);
2942 path[i].p_bh = NULL;
2943 i--;
2944 continue;
2945 }
2946
2947 /* this is index block */
2948 if (!path[i].p_hdr) {
2949 ext_debug(inode, "initialize header\n");
2950 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2951 }
2952
2953 if (!path[i].p_idx) {
2954 /* this level hasn't been touched yet */
2955 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2956 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2957 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2958 path[i].p_hdr,
2959 le16_to_cpu(path[i].p_hdr->eh_entries));
2960 } else {
2961 /* we were already here, see at next index */
2962 path[i].p_idx--;
2963 }
2964
2965 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2966 i, EXT_FIRST_INDEX(path[i].p_hdr),
2967 path[i].p_idx);
2968 if (ext4_ext_more_to_rm(path + i)) {
2969 struct buffer_head *bh;
2970 /* go to the next level */
2971 ext_debug(inode, "move to level %d (block %llu)\n",
2972 i + 1, ext4_idx_pblock(path[i].p_idx));
2973 memset(path + i + 1, 0, sizeof(*path));
2974 bh = read_extent_tree_block(inode, path[i].p_idx,
2975 depth - i - 1,
2976 EXT4_EX_NOCACHE);
2977 if (IS_ERR(bh)) {
2978 /* should we reset i_size? */
2979 err = PTR_ERR(bh);
2980 break;
2981 }
2982 /* Yield here to deal with large extent trees.
2983 * Should be a no-op if we did IO above. */
2984 cond_resched();
2985 if (WARN_ON(i + 1 > depth)) {
2986 err = -EFSCORRUPTED;
2987 break;
2988 }
2989 path[i + 1].p_bh = bh;
2990
2991 /* save actual number of indexes since this
2992 * number is changed at the next iteration */
2993 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2994 i++;
2995 } else {
2996 /* we finished processing this index, go up */
2997 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2998 /* index is empty, remove it;
2999 * handle must be already prepared by the
3000 * truncatei_leaf() */
3001 err = ext4_ext_rm_idx(handle, inode, path, i);
3002 }
3003 /* root level has p_bh == NULL, brelse() eats this */
3004 brelse(path[i].p_bh);
3005 path[i].p_bh = NULL;
3006 i--;
3007 ext_debug(inode, "return to level %d\n", i);
3008 }
3009 }
3010
3011 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3012 path->p_hdr->eh_entries);
3013
3014 /*
3015 * if there's a partial cluster and we have removed the first extent
3016 * in the file, then we also free the partial cluster, if any
3017 */
3018 if (partial.state == tofree && err == 0) {
3019 int flags = get_default_free_blocks_flags(inode);
3020
3021 if (ext4_is_pending(inode, partial.lblk))
3022 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3023 ext4_free_blocks(handle, inode, NULL,
3024 EXT4_C2B(sbi, partial.pclu),
3025 sbi->s_cluster_ratio, flags);
3026 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3027 ext4_rereserve_cluster(inode, partial.lblk);
3028 partial.state = initial;
3029 }
3030
3031 /* TODO: flexible tree reduction should be here */
3032 if (path->p_hdr->eh_entries == 0) {
3033 /*
3034 * truncate to zero freed all the tree,
3035 * so we need to correct eh_depth
3036 */
3037 err = ext4_ext_get_access(handle, inode, path);
3038 if (err == 0) {
3039 ext_inode_hdr(inode)->eh_depth = 0;
3040 ext_inode_hdr(inode)->eh_max =
3041 cpu_to_le16(ext4_ext_space_root(inode, 0));
3042 err = ext4_ext_dirty(handle, inode, path);
3043 }
3044 }
3045 out:
3046 ext4_free_ext_path(path);
3047 path = NULL;
3048 if (err == -EAGAIN)
3049 goto again;
3050 ext4_journal_stop(handle);
3051
3052 return err;
3053 }
3054
3055 /*
3056 * called at mount time
3057 */
3058 void ext4_ext_init(struct super_block *sb)
3059 {
3060 /*
3061 * possible initialization would be here
3062 */
3063
3064 if (ext4_has_feature_extents(sb)) {
3065 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3066 printk(KERN_INFO "EXT4-fs: file extents enabled"
3067 #ifdef AGGRESSIVE_TEST
3068 ", aggressive tests"
3069 #endif
3070 #ifdef CHECK_BINSEARCH
3071 ", check binsearch"
3072 #endif
3073 #ifdef EXTENTS_STATS
3074 ", stats"
3075 #endif
3076 "\n");
3077 #endif
3078 #ifdef EXTENTS_STATS
3079 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3080 EXT4_SB(sb)->s_ext_min = 1 << 30;
3081 EXT4_SB(sb)->s_ext_max = 0;
3082 #endif
3083 }
3084 }
3085
3086 /*
3087 * called at umount time
3088 */
3089 void ext4_ext_release(struct super_block *sb)
3090 {
3091 if (!ext4_has_feature_extents(sb))
3092 return;
3093
3094 #ifdef EXTENTS_STATS
3095 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3096 struct ext4_sb_info *sbi = EXT4_SB(sb);
3097 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3098 sbi->s_ext_blocks, sbi->s_ext_extents,
3099 sbi->s_ext_blocks / sbi->s_ext_extents);
3100 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3101 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3102 }
3103 #endif
3104 }
3105
3106 static void ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3107 {
3108 ext4_lblk_t ee_block;
3109 ext4_fsblk_t ee_pblock;
3110 unsigned int ee_len;
3111
3112 ee_block = le32_to_cpu(ex->ee_block);
3113 ee_len = ext4_ext_get_actual_len(ex);
3114 ee_pblock = ext4_ext_pblock(ex);
3115
3116 if (ee_len == 0)
3117 return;
3118
3119 ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3120 EXTENT_STATUS_WRITTEN);
3121 }
3122
3123 /* FIXME!! we need to try to merge to left or right after zero-out */
3124 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3125 {
3126 ext4_fsblk_t ee_pblock;
3127 unsigned int ee_len;
3128
3129 ee_len = ext4_ext_get_actual_len(ex);
3130 ee_pblock = ext4_ext_pblock(ex);
3131 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3132 ee_len);
3133 }
3134
3135 /*
3136 * ext4_split_extent_at() splits an extent at given block.
3137 *
3138 * @handle: the journal handle
3139 * @inode: the file inode
3140 * @path: the path to the extent
3141 * @split: the logical block where the extent is splitted.
3142 * @split_flags: indicates if the extent could be zeroout if split fails, and
3143 * the states(init or unwritten) of new extents.
3144 * @flags: flags used to insert new extent to extent tree.
3145 *
3146 *
3147 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3148 * of which are determined by split_flag.
3149 *
3150 * There are two cases:
3151 * a> the extent are splitted into two extent.
3152 * b> split is not needed, and just mark the extent.
3153 *
3154 * return 0 on success.
3155 */
3156 static int ext4_split_extent_at(handle_t *handle,
3157 struct inode *inode,
3158 struct ext4_ext_path **ppath,
3159 ext4_lblk_t split,
3160 int split_flag,
3161 int flags)
3162 {
3163 struct ext4_ext_path *path = *ppath;
3164 ext4_fsblk_t newblock;
3165 ext4_lblk_t ee_block;
3166 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3167 struct ext4_extent *ex2 = NULL;
3168 unsigned int ee_len, depth;
3169 int err = 0;
3170
3171 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3172 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3173
3174 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3175
3176 ext4_ext_show_leaf(inode, path);
3177
3178 depth = ext_depth(inode);
3179 ex = path[depth].p_ext;
3180 ee_block = le32_to_cpu(ex->ee_block);
3181 ee_len = ext4_ext_get_actual_len(ex);
3182 newblock = split - ee_block + ext4_ext_pblock(ex);
3183
3184 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3185 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3186 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3187 EXT4_EXT_MARK_UNWRIT1 |
3188 EXT4_EXT_MARK_UNWRIT2));
3189
3190 err = ext4_ext_get_access(handle, inode, path + depth);
3191 if (err)
3192 goto out;
3193
3194 if (split == ee_block) {
3195 /*
3196 * case b: block @split is the block that the extent begins with
3197 * then we just change the state of the extent, and splitting
3198 * is not needed.
3199 */
3200 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3201 ext4_ext_mark_unwritten(ex);
3202 else
3203 ext4_ext_mark_initialized(ex);
3204
3205 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3206 ext4_ext_try_to_merge(handle, inode, path, ex);
3207
3208 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3209 goto out;
3210 }
3211
3212 /* case a */
3213 memcpy(&orig_ex, ex, sizeof(orig_ex));
3214 ex->ee_len = cpu_to_le16(split - ee_block);
3215 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3216 ext4_ext_mark_unwritten(ex);
3217
3218 /*
3219 * path may lead to new leaf, not to original leaf any more
3220 * after ext4_ext_insert_extent() returns,
3221 */
3222 err = ext4_ext_dirty(handle, inode, path + depth);
3223 if (err)
3224 goto fix_extent_len;
3225
3226 ex2 = &newex;
3227 ex2->ee_block = cpu_to_le32(split);
3228 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3229 ext4_ext_store_pblock(ex2, newblock);
3230 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3231 ext4_ext_mark_unwritten(ex2);
3232
3233 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3234 if (err != -ENOSPC && err != -EDQUOT && err != -ENOMEM)
3235 goto out;
3236
3237 /*
3238 * Update path is required because previous ext4_ext_insert_extent()
3239 * may have freed or reallocated the path. Using EXT4_EX_NOFAIL
3240 * guarantees that ext4_find_extent() will not return -ENOMEM,
3241 * otherwise -ENOMEM will cause a retry in do_writepages(), and a
3242 * WARN_ON may be triggered in ext4_da_update_reserve_space() due to
3243 * an incorrect ee_len causing the i_reserved_data_blocks exception.
3244 */
3245 path = ext4_find_extent(inode, ee_block, ppath,
3246 flags | EXT4_EX_NOFAIL);
3247 if (IS_ERR(path)) {
3248 EXT4_ERROR_INODE(inode, "Failed split extent on %u, err %ld",
3249 split, PTR_ERR(path));
3250 return PTR_ERR(path);
3251 }
3252 depth = ext_depth(inode);
3253 ex = path[depth].p_ext;
3254
3255 if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3256 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3257 if (split_flag & EXT4_EXT_DATA_VALID1) {
3258 err = ext4_ext_zeroout(inode, ex2);
3259 zero_ex.ee_block = ex2->ee_block;
3260 zero_ex.ee_len = cpu_to_le16(
3261 ext4_ext_get_actual_len(ex2));
3262 ext4_ext_store_pblock(&zero_ex,
3263 ext4_ext_pblock(ex2));
3264 } else {
3265 err = ext4_ext_zeroout(inode, ex);
3266 zero_ex.ee_block = ex->ee_block;
3267 zero_ex.ee_len = cpu_to_le16(
3268 ext4_ext_get_actual_len(ex));
3269 ext4_ext_store_pblock(&zero_ex,
3270 ext4_ext_pblock(ex));
3271 }
3272 } else {
3273 err = ext4_ext_zeroout(inode, &orig_ex);
3274 zero_ex.ee_block = orig_ex.ee_block;
3275 zero_ex.ee_len = cpu_to_le16(
3276 ext4_ext_get_actual_len(&orig_ex));
3277 ext4_ext_store_pblock(&zero_ex,
3278 ext4_ext_pblock(&orig_ex));
3279 }
3280
3281 if (!err) {
3282 /* update the extent length and mark as initialized */
3283 ex->ee_len = cpu_to_le16(ee_len);
3284 ext4_ext_try_to_merge(handle, inode, path, ex);
3285 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3286 if (!err)
3287 /* update extent status tree */
3288 ext4_zeroout_es(inode, &zero_ex);
3289 /* If we failed at this point, we don't know in which
3290 * state the extent tree exactly is so don't try to fix
3291 * length of the original extent as it may do even more
3292 * damage.
3293 */
3294 goto out;
3295 }
3296 }
3297
3298 fix_extent_len:
3299 ex->ee_len = orig_ex.ee_len;
3300 /*
3301 * Ignore ext4_ext_dirty return value since we are already in error path
3302 * and err is a non-zero error code.
3303 */
3304 ext4_ext_dirty(handle, inode, path + path->p_depth);
3305 return err;
3306 out:
3307 ext4_ext_show_leaf(inode, *ppath);
3308 return err;
3309 }
3310
3311 /*
3312 * ext4_split_extent() splits an extent and mark extent which is covered
3313 * by @map as split_flags indicates
3314 *
3315 * It may result in splitting the extent into multiple extents (up to three)
3316 * There are three possibilities:
3317 * a> There is no split required
3318 * b> Splits in two extents: Split is happening at either end of the extent
3319 * c> Splits in three extents: Somone is splitting in middle of the extent
3320 *
3321 */
3322 static int ext4_split_extent(handle_t *handle,
3323 struct inode *inode,
3324 struct ext4_ext_path **ppath,
3325 struct ext4_map_blocks *map,
3326 int split_flag,
3327 int flags)
3328 {
3329 struct ext4_ext_path *path = *ppath;
3330 ext4_lblk_t ee_block;
3331 struct ext4_extent *ex;
3332 unsigned int ee_len, depth;
3333 int err = 0;
3334 int unwritten;
3335 int split_flag1, flags1;
3336 int allocated = map->m_len;
3337
3338 depth = ext_depth(inode);
3339 ex = path[depth].p_ext;
3340 ee_block = le32_to_cpu(ex->ee_block);
3341 ee_len = ext4_ext_get_actual_len(ex);
3342 unwritten = ext4_ext_is_unwritten(ex);
3343
3344 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3345 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3346 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3347 if (unwritten)
3348 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3349 EXT4_EXT_MARK_UNWRIT2;
3350 if (split_flag & EXT4_EXT_DATA_VALID2)
3351 split_flag1 |= EXT4_EXT_DATA_VALID1;
3352 err = ext4_split_extent_at(handle, inode, ppath,
3353 map->m_lblk + map->m_len, split_flag1, flags1);
3354 if (err)
3355 goto out;
3356 } else {
3357 allocated = ee_len - (map->m_lblk - ee_block);
3358 }
3359 /*
3360 * Update path is required because previous ext4_split_extent_at() may
3361 * result in split of original leaf or extent zeroout.
3362 */
3363 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3364 if (IS_ERR(path))
3365 return PTR_ERR(path);
3366 depth = ext_depth(inode);
3367 ex = path[depth].p_ext;
3368 if (!ex) {
3369 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3370 (unsigned long) map->m_lblk);
3371 return -EFSCORRUPTED;
3372 }
3373 unwritten = ext4_ext_is_unwritten(ex);
3374
3375 if (map->m_lblk >= ee_block) {
3376 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3377 if (unwritten) {
3378 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3379 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3380 EXT4_EXT_MARK_UNWRIT2);
3381 }
3382 err = ext4_split_extent_at(handle, inode, ppath,
3383 map->m_lblk, split_flag1, flags);
3384 if (err)
3385 goto out;
3386 }
3387
3388 ext4_ext_show_leaf(inode, *ppath);
3389 out:
3390 return err ? err : allocated;
3391 }
3392
3393 /*
3394 * This function is called by ext4_ext_map_blocks() if someone tries to write
3395 * to an unwritten extent. It may result in splitting the unwritten
3396 * extent into multiple extents (up to three - one initialized and two
3397 * unwritten).
3398 * There are three possibilities:
3399 * a> There is no split required: Entire extent should be initialized
3400 * b> Splits in two extents: Write is happening at either end of the extent
3401 * c> Splits in three extents: Somone is writing in middle of the extent
3402 *
3403 * Pre-conditions:
3404 * - The extent pointed to by 'path' is unwritten.
3405 * - The extent pointed to by 'path' contains a superset
3406 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3407 *
3408 * Post-conditions on success:
3409 * - the returned value is the number of blocks beyond map->l_lblk
3410 * that are allocated and initialized.
3411 * It is guaranteed to be >= map->m_len.
3412 */
3413 static int ext4_ext_convert_to_initialized(handle_t *handle,
3414 struct inode *inode,
3415 struct ext4_map_blocks *map,
3416 struct ext4_ext_path **ppath,
3417 int flags)
3418 {
3419 struct ext4_ext_path *path = *ppath;
3420 struct ext4_sb_info *sbi;
3421 struct ext4_extent_header *eh;
3422 struct ext4_map_blocks split_map;
3423 struct ext4_extent zero_ex1, zero_ex2;
3424 struct ext4_extent *ex, *abut_ex;
3425 ext4_lblk_t ee_block, eof_block;
3426 unsigned int ee_len, depth, map_len = map->m_len;
3427 int err = 0;
3428 int split_flag = EXT4_EXT_DATA_VALID2;
3429 int allocated = 0;
3430 unsigned int max_zeroout = 0;
3431
3432 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3433 (unsigned long long)map->m_lblk, map_len);
3434
3435 sbi = EXT4_SB(inode->i_sb);
3436 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3437 >> inode->i_sb->s_blocksize_bits;
3438 if (eof_block < map->m_lblk + map_len)
3439 eof_block = map->m_lblk + map_len;
3440
3441 depth = ext_depth(inode);
3442 eh = path[depth].p_hdr;
3443 ex = path[depth].p_ext;
3444 ee_block = le32_to_cpu(ex->ee_block);
3445 ee_len = ext4_ext_get_actual_len(ex);
3446 zero_ex1.ee_len = 0;
3447 zero_ex2.ee_len = 0;
3448
3449 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3450
3451 /* Pre-conditions */
3452 BUG_ON(!ext4_ext_is_unwritten(ex));
3453 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3454
3455 /*
3456 * Attempt to transfer newly initialized blocks from the currently
3457 * unwritten extent to its neighbor. This is much cheaper
3458 * than an insertion followed by a merge as those involve costly
3459 * memmove() calls. Transferring to the left is the common case in
3460 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3461 * followed by append writes.
3462 *
3463 * Limitations of the current logic:
3464 * - L1: we do not deal with writes covering the whole extent.
3465 * This would require removing the extent if the transfer
3466 * is possible.
3467 * - L2: we only attempt to merge with an extent stored in the
3468 * same extent tree node.
3469 */
3470 if ((map->m_lblk == ee_block) &&
3471 /* See if we can merge left */
3472 (map_len < ee_len) && /*L1*/
3473 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3474 ext4_lblk_t prev_lblk;
3475 ext4_fsblk_t prev_pblk, ee_pblk;
3476 unsigned int prev_len;
3477
3478 abut_ex = ex - 1;
3479 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3480 prev_len = ext4_ext_get_actual_len(abut_ex);
3481 prev_pblk = ext4_ext_pblock(abut_ex);
3482 ee_pblk = ext4_ext_pblock(ex);
3483
3484 /*
3485 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3486 * upon those conditions:
3487 * - C1: abut_ex is initialized,
3488 * - C2: abut_ex is logically abutting ex,
3489 * - C3: abut_ex is physically abutting ex,
3490 * - C4: abut_ex can receive the additional blocks without
3491 * overflowing the (initialized) length limit.
3492 */
3493 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3494 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3495 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3496 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3497 err = ext4_ext_get_access(handle, inode, path + depth);
3498 if (err)
3499 goto out;
3500
3501 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3502 map, ex, abut_ex);
3503
3504 /* Shift the start of ex by 'map_len' blocks */
3505 ex->ee_block = cpu_to_le32(ee_block + map_len);
3506 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3507 ex->ee_len = cpu_to_le16(ee_len - map_len);
3508 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3509
3510 /* Extend abut_ex by 'map_len' blocks */
3511 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3512
3513 /* Result: number of initialized blocks past m_lblk */
3514 allocated = map_len;
3515 }
3516 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3517 (map_len < ee_len) && /*L1*/
3518 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3519 /* See if we can merge right */
3520 ext4_lblk_t next_lblk;
3521 ext4_fsblk_t next_pblk, ee_pblk;
3522 unsigned int next_len;
3523
3524 abut_ex = ex + 1;
3525 next_lblk = le32_to_cpu(abut_ex->ee_block);
3526 next_len = ext4_ext_get_actual_len(abut_ex);
3527 next_pblk = ext4_ext_pblock(abut_ex);
3528 ee_pblk = ext4_ext_pblock(ex);
3529
3530 /*
3531 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3532 * upon those conditions:
3533 * - C1: abut_ex is initialized,
3534 * - C2: abut_ex is logically abutting ex,
3535 * - C3: abut_ex is physically abutting ex,
3536 * - C4: abut_ex can receive the additional blocks without
3537 * overflowing the (initialized) length limit.
3538 */
3539 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3540 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3541 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3542 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3543 err = ext4_ext_get_access(handle, inode, path + depth);
3544 if (err)
3545 goto out;
3546
3547 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3548 map, ex, abut_ex);
3549
3550 /* Shift the start of abut_ex by 'map_len' blocks */
3551 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3552 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3553 ex->ee_len = cpu_to_le16(ee_len - map_len);
3554 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3555
3556 /* Extend abut_ex by 'map_len' blocks */
3557 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3558
3559 /* Result: number of initialized blocks past m_lblk */
3560 allocated = map_len;
3561 }
3562 }
3563 if (allocated) {
3564 /* Mark the block containing both extents as dirty */
3565 err = ext4_ext_dirty(handle, inode, path + depth);
3566
3567 /* Update path to point to the right extent */
3568 path[depth].p_ext = abut_ex;
3569 goto out;
3570 } else
3571 allocated = ee_len - (map->m_lblk - ee_block);
3572
3573 WARN_ON(map->m_lblk < ee_block);
3574 /*
3575 * It is safe to convert extent to initialized via explicit
3576 * zeroout only if extent is fully inside i_size or new_size.
3577 */
3578 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3579
3580 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3581 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3582 (inode->i_sb->s_blocksize_bits - 10);
3583
3584 /*
3585 * five cases:
3586 * 1. split the extent into three extents.
3587 * 2. split the extent into two extents, zeroout the head of the first
3588 * extent.
3589 * 3. split the extent into two extents, zeroout the tail of the second
3590 * extent.
3591 * 4. split the extent into two extents with out zeroout.
3592 * 5. no splitting needed, just possibly zeroout the head and / or the
3593 * tail of the extent.
3594 */
3595 split_map.m_lblk = map->m_lblk;
3596 split_map.m_len = map->m_len;
3597
3598 if (max_zeroout && (allocated > split_map.m_len)) {
3599 if (allocated <= max_zeroout) {
3600 /* case 3 or 5 */
3601 zero_ex1.ee_block =
3602 cpu_to_le32(split_map.m_lblk +
3603 split_map.m_len);
3604 zero_ex1.ee_len =
3605 cpu_to_le16(allocated - split_map.m_len);
3606 ext4_ext_store_pblock(&zero_ex1,
3607 ext4_ext_pblock(ex) + split_map.m_lblk +
3608 split_map.m_len - ee_block);
3609 err = ext4_ext_zeroout(inode, &zero_ex1);
3610 if (err)
3611 goto fallback;
3612 split_map.m_len = allocated;
3613 }
3614 if (split_map.m_lblk - ee_block + split_map.m_len <
3615 max_zeroout) {
3616 /* case 2 or 5 */
3617 if (split_map.m_lblk != ee_block) {
3618 zero_ex2.ee_block = ex->ee_block;
3619 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3620 ee_block);
3621 ext4_ext_store_pblock(&zero_ex2,
3622 ext4_ext_pblock(ex));
3623 err = ext4_ext_zeroout(inode, &zero_ex2);
3624 if (err)
3625 goto fallback;
3626 }
3627
3628 split_map.m_len += split_map.m_lblk - ee_block;
3629 split_map.m_lblk = ee_block;
3630 allocated = map->m_len;
3631 }
3632 }
3633
3634 fallback:
3635 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3636 flags);
3637 if (err > 0)
3638 err = 0;
3639 out:
3640 /* If we have gotten a failure, don't zero out status tree */
3641 if (!err) {
3642 ext4_zeroout_es(inode, &zero_ex1);
3643 ext4_zeroout_es(inode, &zero_ex2);
3644 }
3645 return err ? err : allocated;
3646 }
3647
3648 /*
3649 * This function is called by ext4_ext_map_blocks() from
3650 * ext4_get_blocks_dio_write() when DIO to write
3651 * to an unwritten extent.
3652 *
3653 * Writing to an unwritten extent may result in splitting the unwritten
3654 * extent into multiple initialized/unwritten extents (up to three)
3655 * There are three possibilities:
3656 * a> There is no split required: Entire extent should be unwritten
3657 * b> Splits in two extents: Write is happening at either end of the extent
3658 * c> Splits in three extents: Somone is writing in middle of the extent
3659 *
3660 * This works the same way in the case of initialized -> unwritten conversion.
3661 *
3662 * One of more index blocks maybe needed if the extent tree grow after
3663 * the unwritten extent split. To prevent ENOSPC occur at the IO
3664 * complete, we need to split the unwritten extent before DIO submit
3665 * the IO. The unwritten extent called at this time will be split
3666 * into three unwritten extent(at most). After IO complete, the part
3667 * being filled will be convert to initialized by the end_io callback function
3668 * via ext4_convert_unwritten_extents().
3669 *
3670 * Returns the size of unwritten extent to be written on success.
3671 */
3672 static int ext4_split_convert_extents(handle_t *handle,
3673 struct inode *inode,
3674 struct ext4_map_blocks *map,
3675 struct ext4_ext_path **ppath,
3676 int flags)
3677 {
3678 struct ext4_ext_path *path = *ppath;
3679 ext4_lblk_t eof_block;
3680 ext4_lblk_t ee_block;
3681 struct ext4_extent *ex;
3682 unsigned int ee_len;
3683 int split_flag = 0, depth;
3684
3685 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3686 (unsigned long long)map->m_lblk, map->m_len);
3687
3688 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3689 >> inode->i_sb->s_blocksize_bits;
3690 if (eof_block < map->m_lblk + map->m_len)
3691 eof_block = map->m_lblk + map->m_len;
3692 /*
3693 * It is safe to convert extent to initialized via explicit
3694 * zeroout only if extent is fully inside i_size or new_size.
3695 */
3696 depth = ext_depth(inode);
3697 ex = path[depth].p_ext;
3698 ee_block = le32_to_cpu(ex->ee_block);
3699 ee_len = ext4_ext_get_actual_len(ex);
3700
3701 /* Convert to unwritten */
3702 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3703 split_flag |= EXT4_EXT_DATA_VALID1;
3704 /* Convert to initialized */
3705 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3706 split_flag |= ee_block + ee_len <= eof_block ?
3707 EXT4_EXT_MAY_ZEROOUT : 0;
3708 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3709 }
3710 flags |= EXT4_GET_BLOCKS_PRE_IO;
3711 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3712 }
3713
3714 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3715 struct inode *inode,
3716 struct ext4_map_blocks *map,
3717 struct ext4_ext_path **ppath)
3718 {
3719 struct ext4_ext_path *path = *ppath;
3720 struct ext4_extent *ex;
3721 ext4_lblk_t ee_block;
3722 unsigned int ee_len;
3723 int depth;
3724 int err = 0;
3725
3726 depth = ext_depth(inode);
3727 ex = path[depth].p_ext;
3728 ee_block = le32_to_cpu(ex->ee_block);
3729 ee_len = ext4_ext_get_actual_len(ex);
3730
3731 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3732 (unsigned long long)ee_block, ee_len);
3733
3734 /* If extent is larger than requested it is a clear sign that we still
3735 * have some extent state machine issues left. So extent_split is still
3736 * required.
3737 * TODO: Once all related issues will be fixed this situation should be
3738 * illegal.
3739 */
3740 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3741 #ifdef CONFIG_EXT4_DEBUG
3742 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3743 " len %u; IO logical block %llu, len %u",
3744 inode->i_ino, (unsigned long long)ee_block, ee_len,
3745 (unsigned long long)map->m_lblk, map->m_len);
3746 #endif
3747 err = ext4_split_convert_extents(handle, inode, map, ppath,
3748 EXT4_GET_BLOCKS_CONVERT);
3749 if (err < 0)
3750 return err;
3751 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3752 if (IS_ERR(path))
3753 return PTR_ERR(path);
3754 depth = ext_depth(inode);
3755 ex = path[depth].p_ext;
3756 }
3757
3758 err = ext4_ext_get_access(handle, inode, path + depth);
3759 if (err)
3760 goto out;
3761 /* first mark the extent as initialized */
3762 ext4_ext_mark_initialized(ex);
3763
3764 /* note: ext4_ext_correct_indexes() isn't needed here because
3765 * borders are not changed
3766 */
3767 ext4_ext_try_to_merge(handle, inode, path, ex);
3768
3769 /* Mark modified extent as dirty */
3770 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3771 out:
3772 ext4_ext_show_leaf(inode, path);
3773 return err;
3774 }
3775
3776 static int
3777 convert_initialized_extent(handle_t *handle, struct inode *inode,
3778 struct ext4_map_blocks *map,
3779 struct ext4_ext_path **ppath,
3780 unsigned int *allocated)
3781 {
3782 struct ext4_ext_path *path = *ppath;
3783 struct ext4_extent *ex;
3784 ext4_lblk_t ee_block;
3785 unsigned int ee_len;
3786 int depth;
3787 int err = 0;
3788
3789 /*
3790 * Make sure that the extent is no bigger than we support with
3791 * unwritten extent
3792 */
3793 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3794 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3795
3796 depth = ext_depth(inode);
3797 ex = path[depth].p_ext;
3798 ee_block = le32_to_cpu(ex->ee_block);
3799 ee_len = ext4_ext_get_actual_len(ex);
3800
3801 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3802 (unsigned long long)ee_block, ee_len);
3803
3804 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3805 err = ext4_split_convert_extents(handle, inode, map, ppath,
3806 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3807 if (err < 0)
3808 return err;
3809 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3810 if (IS_ERR(path))
3811 return PTR_ERR(path);
3812 depth = ext_depth(inode);
3813 ex = path[depth].p_ext;
3814 if (!ex) {
3815 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3816 (unsigned long) map->m_lblk);
3817 return -EFSCORRUPTED;
3818 }
3819 }
3820
3821 err = ext4_ext_get_access(handle, inode, path + depth);
3822 if (err)
3823 return err;
3824 /* first mark the extent as unwritten */
3825 ext4_ext_mark_unwritten(ex);
3826
3827 /* note: ext4_ext_correct_indexes() isn't needed here because
3828 * borders are not changed
3829 */
3830 ext4_ext_try_to_merge(handle, inode, path, ex);
3831
3832 /* Mark modified extent as dirty */
3833 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3834 if (err)
3835 return err;
3836 ext4_ext_show_leaf(inode, path);
3837
3838 ext4_update_inode_fsync_trans(handle, inode, 1);
3839
3840 map->m_flags |= EXT4_MAP_UNWRITTEN;
3841 if (*allocated > map->m_len)
3842 *allocated = map->m_len;
3843 map->m_len = *allocated;
3844 return 0;
3845 }
3846
3847 static int
3848 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3849 struct ext4_map_blocks *map,
3850 struct ext4_ext_path **ppath, int flags,
3851 unsigned int allocated, ext4_fsblk_t newblock)
3852 {
3853 int ret = 0;
3854 int err = 0;
3855
3856 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3857 (unsigned long long)map->m_lblk, map->m_len, flags,
3858 allocated);
3859 ext4_ext_show_leaf(inode, *ppath);
3860
3861 /*
3862 * When writing into unwritten space, we should not fail to
3863 * allocate metadata blocks for the new extent block if needed.
3864 */
3865 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3866
3867 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3868 allocated, newblock);
3869
3870 /* get_block() before submitting IO, split the extent */
3871 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3872 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3873 flags | EXT4_GET_BLOCKS_CONVERT);
3874 if (ret < 0) {
3875 err = ret;
3876 goto out2;
3877 }
3878 /*
3879 * shouldn't get a 0 return when splitting an extent unless
3880 * m_len is 0 (bug) or extent has been corrupted
3881 */
3882 if (unlikely(ret == 0)) {
3883 EXT4_ERROR_INODE(inode,
3884 "unexpected ret == 0, m_len = %u",
3885 map->m_len);
3886 err = -EFSCORRUPTED;
3887 goto out2;
3888 }
3889 map->m_flags |= EXT4_MAP_UNWRITTEN;
3890 goto out;
3891 }
3892 /* IO end_io complete, convert the filled extent to written */
3893 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3894 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3895 ppath);
3896 if (err < 0)
3897 goto out2;
3898 ext4_update_inode_fsync_trans(handle, inode, 1);
3899 goto map_out;
3900 }
3901 /* buffered IO cases */
3902 /*
3903 * repeat fallocate creation request
3904 * we already have an unwritten extent
3905 */
3906 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3907 map->m_flags |= EXT4_MAP_UNWRITTEN;
3908 goto map_out;
3909 }
3910
3911 /* buffered READ or buffered write_begin() lookup */
3912 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3913 /*
3914 * We have blocks reserved already. We
3915 * return allocated blocks so that delalloc
3916 * won't do block reservation for us. But
3917 * the buffer head will be unmapped so that
3918 * a read from the block returns 0s.
3919 */
3920 map->m_flags |= EXT4_MAP_UNWRITTEN;
3921 goto out1;
3922 }
3923
3924 /*
3925 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3926 * For buffered writes, at writepage time, etc. Convert a
3927 * discovered unwritten extent to written.
3928 */
3929 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3930 if (ret < 0) {
3931 err = ret;
3932 goto out2;
3933 }
3934 ext4_update_inode_fsync_trans(handle, inode, 1);
3935 /*
3936 * shouldn't get a 0 return when converting an unwritten extent
3937 * unless m_len is 0 (bug) or extent has been corrupted
3938 */
3939 if (unlikely(ret == 0)) {
3940 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3941 map->m_len);
3942 err = -EFSCORRUPTED;
3943 goto out2;
3944 }
3945
3946 out:
3947 allocated = ret;
3948 map->m_flags |= EXT4_MAP_NEW;
3949 map_out:
3950 map->m_flags |= EXT4_MAP_MAPPED;
3951 out1:
3952 map->m_pblk = newblock;
3953 if (allocated > map->m_len)
3954 allocated = map->m_len;
3955 map->m_len = allocated;
3956 ext4_ext_show_leaf(inode, *ppath);
3957 out2:
3958 return err ? err : allocated;
3959 }
3960
3961 /*
3962 * get_implied_cluster_alloc - check to see if the requested
3963 * allocation (in the map structure) overlaps with a cluster already
3964 * allocated in an extent.
3965 * @sb The filesystem superblock structure
3966 * @map The requested lblk->pblk mapping
3967 * @ex The extent structure which might contain an implied
3968 * cluster allocation
3969 *
3970 * This function is called by ext4_ext_map_blocks() after we failed to
3971 * find blocks that were already in the inode's extent tree. Hence,
3972 * we know that the beginning of the requested region cannot overlap
3973 * the extent from the inode's extent tree. There are three cases we
3974 * want to catch. The first is this case:
3975 *
3976 * |--- cluster # N--|
3977 * |--- extent ---| |---- requested region ---|
3978 * |==========|
3979 *
3980 * The second case that we need to test for is this one:
3981 *
3982 * |--------- cluster # N ----------------|
3983 * |--- requested region --| |------- extent ----|
3984 * |=======================|
3985 *
3986 * The third case is when the requested region lies between two extents
3987 * within the same cluster:
3988 * |------------- cluster # N-------------|
3989 * |----- ex -----| |---- ex_right ----|
3990 * |------ requested region ------|
3991 * |================|
3992 *
3993 * In each of the above cases, we need to set the map->m_pblk and
3994 * map->m_len so it corresponds to the return the extent labelled as
3995 * "|====|" from cluster #N, since it is already in use for data in
3996 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3997 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3998 * as a new "allocated" block region. Otherwise, we will return 0 and
3999 * ext4_ext_map_blocks() will then allocate one or more new clusters
4000 * by calling ext4_mb_new_blocks().
4001 */
4002 static int get_implied_cluster_alloc(struct super_block *sb,
4003 struct ext4_map_blocks *map,
4004 struct ext4_extent *ex,
4005 struct ext4_ext_path *path)
4006 {
4007 struct ext4_sb_info *sbi = EXT4_SB(sb);
4008 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4009 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4010 ext4_lblk_t rr_cluster_start;
4011 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4012 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4013 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4014
4015 /* The extent passed in that we are trying to match */
4016 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4017 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4018
4019 /* The requested region passed into ext4_map_blocks() */
4020 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4021
4022 if ((rr_cluster_start == ex_cluster_end) ||
4023 (rr_cluster_start == ex_cluster_start)) {
4024 if (rr_cluster_start == ex_cluster_end)
4025 ee_start += ee_len - 1;
4026 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4027 map->m_len = min(map->m_len,
4028 (unsigned) sbi->s_cluster_ratio - c_offset);
4029 /*
4030 * Check for and handle this case:
4031 *
4032 * |--------- cluster # N-------------|
4033 * |------- extent ----|
4034 * |--- requested region ---|
4035 * |===========|
4036 */
4037
4038 if (map->m_lblk < ee_block)
4039 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4040
4041 /*
4042 * Check for the case where there is already another allocated
4043 * block to the right of 'ex' but before the end of the cluster.
4044 *
4045 * |------------- cluster # N-------------|
4046 * |----- ex -----| |---- ex_right ----|
4047 * |------ requested region ------|
4048 * |================|
4049 */
4050 if (map->m_lblk > ee_block) {
4051 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4052 map->m_len = min(map->m_len, next - map->m_lblk);
4053 }
4054
4055 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4056 return 1;
4057 }
4058
4059 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4060 return 0;
4061 }
4062
4063 /*
4064 * Determine hole length around the given logical block, first try to
4065 * locate and expand the hole from the given @path, and then adjust it
4066 * if it's partially or completely converted to delayed extents, insert
4067 * it into the extent cache tree if it's indeed a hole, finally return
4068 * the length of the determined extent.
4069 */
4070 static ext4_lblk_t ext4_ext_determine_insert_hole(struct inode *inode,
4071 struct ext4_ext_path *path,
4072 ext4_lblk_t lblk)
4073 {
4074 ext4_lblk_t hole_start, len;
4075 struct extent_status es;
4076
4077 hole_start = lblk;
4078 len = ext4_ext_find_hole(inode, path, &hole_start);
4079 again:
4080 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
4081 hole_start + len - 1, &es);
4082 if (!es.es_len)
4083 goto insert_hole;
4084
4085 /*
4086 * There's a delalloc extent in the hole, handle it if the delalloc
4087 * extent is in front of, behind and straddle the queried range.
4088 */
4089 if (lblk >= es.es_lblk + es.es_len) {
4090 /*
4091 * The delalloc extent is in front of the queried range,
4092 * find again from the queried start block.
4093 */
4094 len -= lblk - hole_start;
4095 hole_start = lblk;
4096 goto again;
4097 } else if (in_range(lblk, es.es_lblk, es.es_len)) {
4098 /*
4099 * The delalloc extent containing lblk, it must have been
4100 * added after ext4_map_blocks() checked the extent status
4101 * tree so we are not holding i_rwsem and delalloc info is
4102 * only stabilized by i_data_sem we are going to release
4103 * soon. Don't modify the extent status tree and report
4104 * extent as a hole, just adjust the length to the delalloc
4105 * extent's after lblk.
4106 */
4107 len = es.es_lblk + es.es_len - lblk;
4108 return len;
4109 } else {
4110 /*
4111 * The delalloc extent is partially or completely behind
4112 * the queried range, update hole length until the
4113 * beginning of the delalloc extent.
4114 */
4115 len = min(es.es_lblk - hole_start, len);
4116 }
4117
4118 insert_hole:
4119 /* Put just found gap into cache to speed up subsequent requests */
4120 ext_debug(inode, " -> %u:%u\n", hole_start, len);
4121 ext4_es_insert_extent(inode, hole_start, len, ~0, EXTENT_STATUS_HOLE);
4122
4123 /* Update hole_len to reflect hole size after lblk */
4124 if (hole_start != lblk)
4125 len -= lblk - hole_start;
4126
4127 return len;
4128 }
4129
4130 /*
4131 * Block allocation/map/preallocation routine for extents based files
4132 *
4133 *
4134 * Need to be called with
4135 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4136 * (ie, flags is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4137 *
4138 * return > 0, number of blocks already mapped/allocated
4139 * if flags doesn't contain EXT4_GET_BLOCKS_CREATE and these are pre-allocated blocks
4140 * buffer head is unmapped
4141 * otherwise blocks are mapped
4142 *
4143 * return = 0, if plain look up failed (blocks have not been allocated)
4144 * buffer head is unmapped
4145 *
4146 * return < 0, error case.
4147 */
4148 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4149 struct ext4_map_blocks *map, int flags)
4150 {
4151 struct ext4_ext_path *path = NULL;
4152 struct ext4_extent newex, *ex, ex2;
4153 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4154 ext4_fsblk_t newblock = 0, pblk;
4155 int err = 0, depth, ret;
4156 unsigned int allocated = 0, offset = 0;
4157 unsigned int allocated_clusters = 0;
4158 struct ext4_allocation_request ar;
4159 ext4_lblk_t cluster_offset;
4160
4161 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4162 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4163
4164 /* find extent for this block */
4165 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4166 if (IS_ERR(path)) {
4167 err = PTR_ERR(path);
4168 path = NULL;
4169 goto out;
4170 }
4171
4172 depth = ext_depth(inode);
4173
4174 /*
4175 * consistent leaf must not be empty;
4176 * this situation is possible, though, _during_ tree modification;
4177 * this is why assert can't be put in ext4_find_extent()
4178 */
4179 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4180 EXT4_ERROR_INODE(inode, "bad extent address "
4181 "lblock: %lu, depth: %d pblock %lld",
4182 (unsigned long) map->m_lblk, depth,
4183 path[depth].p_block);
4184 err = -EFSCORRUPTED;
4185 goto out;
4186 }
4187
4188 ex = path[depth].p_ext;
4189 if (ex) {
4190 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4191 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4192 unsigned short ee_len;
4193
4194
4195 /*
4196 * unwritten extents are treated as holes, except that
4197 * we split out initialized portions during a write.
4198 */
4199 ee_len = ext4_ext_get_actual_len(ex);
4200
4201 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4202
4203 /* if found extent covers block, simply return it */
4204 if (in_range(map->m_lblk, ee_block, ee_len)) {
4205 newblock = map->m_lblk - ee_block + ee_start;
4206 /* number of remaining blocks in the extent */
4207 allocated = ee_len - (map->m_lblk - ee_block);
4208 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4209 map->m_lblk, ee_block, ee_len, newblock);
4210
4211 /*
4212 * If the extent is initialized check whether the
4213 * caller wants to convert it to unwritten.
4214 */
4215 if ((!ext4_ext_is_unwritten(ex)) &&
4216 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4217 err = convert_initialized_extent(handle,
4218 inode, map, &path, &allocated);
4219 goto out;
4220 } else if (!ext4_ext_is_unwritten(ex)) {
4221 map->m_flags |= EXT4_MAP_MAPPED;
4222 map->m_pblk = newblock;
4223 if (allocated > map->m_len)
4224 allocated = map->m_len;
4225 map->m_len = allocated;
4226 ext4_ext_show_leaf(inode, path);
4227 goto out;
4228 }
4229
4230 ret = ext4_ext_handle_unwritten_extents(
4231 handle, inode, map, &path, flags,
4232 allocated, newblock);
4233 if (ret < 0)
4234 err = ret;
4235 else
4236 allocated = ret;
4237 goto out;
4238 }
4239 }
4240
4241 /*
4242 * requested block isn't allocated yet;
4243 * we couldn't try to create block if flags doesn't contain EXT4_GET_BLOCKS_CREATE
4244 */
4245 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4246 ext4_lblk_t len;
4247
4248 len = ext4_ext_determine_insert_hole(inode, path, map->m_lblk);
4249
4250 map->m_pblk = 0;
4251 map->m_len = min_t(unsigned int, map->m_len, len);
4252 goto out;
4253 }
4254
4255 /*
4256 * Okay, we need to do block allocation.
4257 */
4258 newex.ee_block = cpu_to_le32(map->m_lblk);
4259 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4260
4261 /*
4262 * If we are doing bigalloc, check to see if the extent returned
4263 * by ext4_find_extent() implies a cluster we can use.
4264 */
4265 if (cluster_offset && ex &&
4266 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4267 ar.len = allocated = map->m_len;
4268 newblock = map->m_pblk;
4269 goto got_allocated_blocks;
4270 }
4271
4272 /* find neighbour allocated blocks */
4273 ar.lleft = map->m_lblk;
4274 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4275 if (err)
4276 goto out;
4277 ar.lright = map->m_lblk;
4278 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4279 if (err < 0)
4280 goto out;
4281
4282 /* Check if the extent after searching to the right implies a
4283 * cluster we can use. */
4284 if ((sbi->s_cluster_ratio > 1) && err &&
4285 get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4286 ar.len = allocated = map->m_len;
4287 newblock = map->m_pblk;
4288 goto got_allocated_blocks;
4289 }
4290
4291 /*
4292 * See if request is beyond maximum number of blocks we can have in
4293 * a single extent. For an initialized extent this limit is
4294 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4295 * EXT_UNWRITTEN_MAX_LEN.
4296 */
4297 if (map->m_len > EXT_INIT_MAX_LEN &&
4298 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4299 map->m_len = EXT_INIT_MAX_LEN;
4300 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4301 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4302 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4303
4304 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4305 newex.ee_len = cpu_to_le16(map->m_len);
4306 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4307 if (err)
4308 allocated = ext4_ext_get_actual_len(&newex);
4309 else
4310 allocated = map->m_len;
4311
4312 /* allocate new block */
4313 ar.inode = inode;
4314 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4315 ar.logical = map->m_lblk;
4316 /*
4317 * We calculate the offset from the beginning of the cluster
4318 * for the logical block number, since when we allocate a
4319 * physical cluster, the physical block should start at the
4320 * same offset from the beginning of the cluster. This is
4321 * needed so that future calls to get_implied_cluster_alloc()
4322 * work correctly.
4323 */
4324 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4325 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4326 ar.goal -= offset;
4327 ar.logical -= offset;
4328 if (S_ISREG(inode->i_mode))
4329 ar.flags = EXT4_MB_HINT_DATA;
4330 else
4331 /* disable in-core preallocation for non-regular files */
4332 ar.flags = 0;
4333 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4334 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4335 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4336 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4337 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4338 ar.flags |= EXT4_MB_USE_RESERVED;
4339 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4340 if (!newblock)
4341 goto out;
4342 allocated_clusters = ar.len;
4343 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4344 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4345 ar.goal, newblock, ar.len, allocated);
4346 if (ar.len > allocated)
4347 ar.len = allocated;
4348
4349 got_allocated_blocks:
4350 /* try to insert new extent into found leaf and return */
4351 pblk = newblock + offset;
4352 ext4_ext_store_pblock(&newex, pblk);
4353 newex.ee_len = cpu_to_le16(ar.len);
4354 /* Mark unwritten */
4355 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4356 ext4_ext_mark_unwritten(&newex);
4357 map->m_flags |= EXT4_MAP_UNWRITTEN;
4358 }
4359
4360 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4361 if (err) {
4362 if (allocated_clusters) {
4363 int fb_flags = 0;
4364
4365 /*
4366 * free data blocks we just allocated.
4367 * not a good idea to call discard here directly,
4368 * but otherwise we'd need to call it every free().
4369 */
4370 ext4_discard_preallocations(inode);
4371 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4372 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4373 ext4_free_blocks(handle, inode, NULL, newblock,
4374 EXT4_C2B(sbi, allocated_clusters),
4375 fb_flags);
4376 }
4377 goto out;
4378 }
4379
4380 /*
4381 * Reduce the reserved cluster count to reflect successful deferred
4382 * allocation of delayed allocated clusters or direct allocation of
4383 * clusters discovered to be delayed allocated. Once allocated, a
4384 * cluster is not included in the reserved count.
4385 */
4386 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4387 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4388 /*
4389 * When allocating delayed allocated clusters, simply
4390 * reduce the reserved cluster count and claim quota
4391 */
4392 ext4_da_update_reserve_space(inode, allocated_clusters,
4393 1);
4394 } else {
4395 ext4_lblk_t lblk, len;
4396 unsigned int n;
4397
4398 /*
4399 * When allocating non-delayed allocated clusters
4400 * (from fallocate, filemap, DIO, or clusters
4401 * allocated when delalloc has been disabled by
4402 * ext4_nonda_switch), reduce the reserved cluster
4403 * count by the number of allocated clusters that
4404 * have previously been delayed allocated. Quota
4405 * has been claimed by ext4_mb_new_blocks() above,
4406 * so release the quota reservations made for any
4407 * previously delayed allocated clusters.
4408 */
4409 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4410 len = allocated_clusters << sbi->s_cluster_bits;
4411 n = ext4_es_delayed_clu(inode, lblk, len);
4412 if (n > 0)
4413 ext4_da_update_reserve_space(inode, (int) n, 0);
4414 }
4415 }
4416
4417 /*
4418 * Cache the extent and update transaction to commit on fdatasync only
4419 * when it is _not_ an unwritten extent.
4420 */
4421 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4422 ext4_update_inode_fsync_trans(handle, inode, 1);
4423 else
4424 ext4_update_inode_fsync_trans(handle, inode, 0);
4425
4426 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4427 map->m_pblk = pblk;
4428 map->m_len = ar.len;
4429 allocated = map->m_len;
4430 ext4_ext_show_leaf(inode, path);
4431 out:
4432 ext4_free_ext_path(path);
4433
4434 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4435 err ? err : allocated);
4436 return err ? err : allocated;
4437 }
4438
4439 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4440 {
4441 struct super_block *sb = inode->i_sb;
4442 ext4_lblk_t last_block;
4443 int err = 0;
4444
4445 /*
4446 * TODO: optimization is possible here.
4447 * Probably we need not scan at all,
4448 * because page truncation is enough.
4449 */
4450
4451 /* we have to know where to truncate from in crash case */
4452 EXT4_I(inode)->i_disksize = inode->i_size;
4453 err = ext4_mark_inode_dirty(handle, inode);
4454 if (err)
4455 return err;
4456
4457 last_block = (inode->i_size + sb->s_blocksize - 1)
4458 >> EXT4_BLOCK_SIZE_BITS(sb);
4459 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
4460
4461 retry_remove_space:
4462 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4463 if (err == -ENOMEM) {
4464 memalloc_retry_wait(GFP_ATOMIC);
4465 goto retry_remove_space;
4466 }
4467 return err;
4468 }
4469
4470 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4471 ext4_lblk_t len, loff_t new_size,
4472 int flags)
4473 {
4474 struct inode *inode = file_inode(file);
4475 handle_t *handle;
4476 int ret = 0, ret2 = 0, ret3 = 0;
4477 int retries = 0;
4478 int depth = 0;
4479 struct ext4_map_blocks map;
4480 unsigned int credits;
4481 loff_t epos;
4482
4483 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4484 map.m_lblk = offset;
4485 map.m_len = len;
4486 /*
4487 * Don't normalize the request if it can fit in one extent so
4488 * that it doesn't get unnecessarily split into multiple
4489 * extents.
4490 */
4491 if (len <= EXT_UNWRITTEN_MAX_LEN)
4492 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4493
4494 /*
4495 * credits to insert 1 extent into extent tree
4496 */
4497 credits = ext4_chunk_trans_blocks(inode, len);
4498 depth = ext_depth(inode);
4499
4500 retry:
4501 while (len) {
4502 /*
4503 * Recalculate credits when extent tree depth changes.
4504 */
4505 if (depth != ext_depth(inode)) {
4506 credits = ext4_chunk_trans_blocks(inode, len);
4507 depth = ext_depth(inode);
4508 }
4509
4510 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4511 credits);
4512 if (IS_ERR(handle)) {
4513 ret = PTR_ERR(handle);
4514 break;
4515 }
4516 ret = ext4_map_blocks(handle, inode, &map, flags);
4517 if (ret <= 0) {
4518 ext4_debug("inode #%lu: block %u: len %u: "
4519 "ext4_ext_map_blocks returned %d",
4520 inode->i_ino, map.m_lblk,
4521 map.m_len, ret);
4522 ext4_mark_inode_dirty(handle, inode);
4523 ext4_journal_stop(handle);
4524 break;
4525 }
4526 /*
4527 * allow a full retry cycle for any remaining allocations
4528 */
4529 retries = 0;
4530 map.m_lblk += ret;
4531 map.m_len = len = len - ret;
4532 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4533 inode_set_ctime_current(inode);
4534 if (new_size) {
4535 if (epos > new_size)
4536 epos = new_size;
4537 if (ext4_update_inode_size(inode, epos) & 0x1)
4538 inode_set_mtime_to_ts(inode,
4539 inode_get_ctime(inode));
4540 }
4541 ret2 = ext4_mark_inode_dirty(handle, inode);
4542 ext4_update_inode_fsync_trans(handle, inode, 1);
4543 ret3 = ext4_journal_stop(handle);
4544 ret2 = ret3 ? ret3 : ret2;
4545 if (unlikely(ret2))
4546 break;
4547 }
4548 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4549 goto retry;
4550
4551 return ret > 0 ? ret2 : ret;
4552 }
4553
4554 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4555
4556 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4557
4558 static long ext4_zero_range(struct file *file, loff_t offset,
4559 loff_t len, int mode)
4560 {
4561 struct inode *inode = file_inode(file);
4562 struct address_space *mapping = file->f_mapping;
4563 handle_t *handle = NULL;
4564 unsigned int max_blocks;
4565 loff_t new_size = 0;
4566 int ret = 0;
4567 int flags;
4568 int credits;
4569 int partial_begin, partial_end;
4570 loff_t start, end;
4571 ext4_lblk_t lblk;
4572 unsigned int blkbits = inode->i_blkbits;
4573
4574 trace_ext4_zero_range(inode, offset, len, mode);
4575
4576 /*
4577 * Round up offset. This is not fallocate, we need to zero out
4578 * blocks, so convert interior block aligned part of the range to
4579 * unwritten and possibly manually zero out unaligned parts of the
4580 * range. Here, start and partial_begin are inclusive, end and
4581 * partial_end are exclusive.
4582 */
4583 start = round_up(offset, 1 << blkbits);
4584 end = round_down((offset + len), 1 << blkbits);
4585
4586 if (start < offset || end > offset + len)
4587 return -EINVAL;
4588 partial_begin = offset & ((1 << blkbits) - 1);
4589 partial_end = (offset + len) & ((1 << blkbits) - 1);
4590
4591 lblk = start >> blkbits;
4592 max_blocks = (end >> blkbits);
4593 if (max_blocks < lblk)
4594 max_blocks = 0;
4595 else
4596 max_blocks -= lblk;
4597
4598 inode_lock(inode);
4599
4600 /*
4601 * Indirect files do not support unwritten extents
4602 */
4603 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4604 ret = -EOPNOTSUPP;
4605 goto out_mutex;
4606 }
4607
4608 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4609 (offset + len > inode->i_size ||
4610 offset + len > EXT4_I(inode)->i_disksize)) {
4611 new_size = offset + len;
4612 ret = inode_newsize_ok(inode, new_size);
4613 if (ret)
4614 goto out_mutex;
4615 }
4616
4617 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4618
4619 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4620 inode_dio_wait(inode);
4621
4622 ret = file_modified(file);
4623 if (ret)
4624 goto out_mutex;
4625
4626 /* Preallocate the range including the unaligned edges */
4627 if (partial_begin || partial_end) {
4628 ret = ext4_alloc_file_blocks(file,
4629 round_down(offset, 1 << blkbits) >> blkbits,
4630 (round_up((offset + len), 1 << blkbits) -
4631 round_down(offset, 1 << blkbits)) >> blkbits,
4632 new_size, flags);
4633 if (ret)
4634 goto out_mutex;
4635
4636 }
4637
4638 /* Zero range excluding the unaligned edges */
4639 if (max_blocks > 0) {
4640 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4641 EXT4_EX_NOCACHE);
4642
4643 /*
4644 * Prevent page faults from reinstantiating pages we have
4645 * released from page cache.
4646 */
4647 filemap_invalidate_lock(mapping);
4648
4649 ret = ext4_break_layouts(inode);
4650 if (ret) {
4651 filemap_invalidate_unlock(mapping);
4652 goto out_mutex;
4653 }
4654
4655 ret = ext4_update_disksize_before_punch(inode, offset, len);
4656 if (ret) {
4657 filemap_invalidate_unlock(mapping);
4658 goto out_mutex;
4659 }
4660
4661 /*
4662 * For journalled data we need to write (and checkpoint) pages
4663 * before discarding page cache to avoid inconsitent data on
4664 * disk in case of crash before zeroing trans is committed.
4665 */
4666 if (ext4_should_journal_data(inode)) {
4667 ret = filemap_write_and_wait_range(mapping, start,
4668 end - 1);
4669 if (ret) {
4670 filemap_invalidate_unlock(mapping);
4671 goto out_mutex;
4672 }
4673 }
4674
4675 /* Now release the pages and zero block aligned part of pages */
4676 truncate_pagecache_range(inode, start, end - 1);
4677 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4678
4679 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4680 flags);
4681 filemap_invalidate_unlock(mapping);
4682 if (ret)
4683 goto out_mutex;
4684 }
4685 if (!partial_begin && !partial_end)
4686 goto out_mutex;
4687
4688 /*
4689 * In worst case we have to writeout two nonadjacent unwritten
4690 * blocks and update the inode
4691 */
4692 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4693 if (ext4_should_journal_data(inode))
4694 credits += 2;
4695 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4696 if (IS_ERR(handle)) {
4697 ret = PTR_ERR(handle);
4698 ext4_std_error(inode->i_sb, ret);
4699 goto out_mutex;
4700 }
4701
4702 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4703 if (new_size)
4704 ext4_update_inode_size(inode, new_size);
4705 ret = ext4_mark_inode_dirty(handle, inode);
4706 if (unlikely(ret))
4707 goto out_handle;
4708 /* Zero out partial block at the edges of the range */
4709 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4710 if (ret >= 0)
4711 ext4_update_inode_fsync_trans(handle, inode, 1);
4712
4713 if (file->f_flags & O_SYNC)
4714 ext4_handle_sync(handle);
4715
4716 out_handle:
4717 ext4_journal_stop(handle);
4718 out_mutex:
4719 inode_unlock(inode);
4720 return ret;
4721 }
4722
4723 /*
4724 * preallocate space for a file. This implements ext4's fallocate file
4725 * operation, which gets called from sys_fallocate system call.
4726 * For block-mapped files, posix_fallocate should fall back to the method
4727 * of writing zeroes to the required new blocks (the same behavior which is
4728 * expected for file systems which do not support fallocate() system call).
4729 */
4730 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4731 {
4732 struct inode *inode = file_inode(file);
4733 loff_t new_size = 0;
4734 unsigned int max_blocks;
4735 int ret = 0;
4736 int flags;
4737 ext4_lblk_t lblk;
4738 unsigned int blkbits = inode->i_blkbits;
4739
4740 /*
4741 * Encrypted inodes can't handle collapse range or insert
4742 * range since we would need to re-encrypt blocks with a
4743 * different IV or XTS tweak (which are based on the logical
4744 * block number).
4745 */
4746 if (IS_ENCRYPTED(inode) &&
4747 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4748 return -EOPNOTSUPP;
4749
4750 /* Return error if mode is not supported */
4751 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4752 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4753 FALLOC_FL_INSERT_RANGE))
4754 return -EOPNOTSUPP;
4755
4756 inode_lock(inode);
4757 ret = ext4_convert_inline_data(inode);
4758 inode_unlock(inode);
4759 if (ret)
4760 goto exit;
4761
4762 if (mode & FALLOC_FL_PUNCH_HOLE) {
4763 ret = ext4_punch_hole(file, offset, len);
4764 goto exit;
4765 }
4766
4767 if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4768 ret = ext4_collapse_range(file, offset, len);
4769 goto exit;
4770 }
4771
4772 if (mode & FALLOC_FL_INSERT_RANGE) {
4773 ret = ext4_insert_range(file, offset, len);
4774 goto exit;
4775 }
4776
4777 if (mode & FALLOC_FL_ZERO_RANGE) {
4778 ret = ext4_zero_range(file, offset, len, mode);
4779 goto exit;
4780 }
4781 trace_ext4_fallocate_enter(inode, offset, len, mode);
4782 lblk = offset >> blkbits;
4783
4784 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4785 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4786
4787 inode_lock(inode);
4788
4789 /*
4790 * We only support preallocation for extent-based files only
4791 */
4792 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4793 ret = -EOPNOTSUPP;
4794 goto out;
4795 }
4796
4797 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4798 (offset + len > inode->i_size ||
4799 offset + len > EXT4_I(inode)->i_disksize)) {
4800 new_size = offset + len;
4801 ret = inode_newsize_ok(inode, new_size);
4802 if (ret)
4803 goto out;
4804 }
4805
4806 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4807 inode_dio_wait(inode);
4808
4809 ret = file_modified(file);
4810 if (ret)
4811 goto out;
4812
4813 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4814 if (ret)
4815 goto out;
4816
4817 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4818 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4819 EXT4_I(inode)->i_sync_tid);
4820 }
4821 out:
4822 inode_unlock(inode);
4823 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4824 exit:
4825 return ret;
4826 }
4827
4828 /*
4829 * This function convert a range of blocks to written extents
4830 * The caller of this function will pass the start offset and the size.
4831 * all unwritten extents within this range will be converted to
4832 * written extents.
4833 *
4834 * This function is called from the direct IO end io call back
4835 * function, to convert the fallocated extents after IO is completed.
4836 * Returns 0 on success.
4837 */
4838 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4839 loff_t offset, ssize_t len)
4840 {
4841 unsigned int max_blocks;
4842 int ret = 0, ret2 = 0, ret3 = 0;
4843 struct ext4_map_blocks map;
4844 unsigned int blkbits = inode->i_blkbits;
4845 unsigned int credits = 0;
4846
4847 map.m_lblk = offset >> blkbits;
4848 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4849
4850 if (!handle) {
4851 /*
4852 * credits to insert 1 extent into extent tree
4853 */
4854 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4855 }
4856 while (ret >= 0 && ret < max_blocks) {
4857 map.m_lblk += ret;
4858 map.m_len = (max_blocks -= ret);
4859 if (credits) {
4860 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4861 credits);
4862 if (IS_ERR(handle)) {
4863 ret = PTR_ERR(handle);
4864 break;
4865 }
4866 }
4867 ret = ext4_map_blocks(handle, inode, &map,
4868 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4869 if (ret <= 0)
4870 ext4_warning(inode->i_sb,
4871 "inode #%lu: block %u: len %u: "
4872 "ext4_ext_map_blocks returned %d",
4873 inode->i_ino, map.m_lblk,
4874 map.m_len, ret);
4875 ret2 = ext4_mark_inode_dirty(handle, inode);
4876 if (credits) {
4877 ret3 = ext4_journal_stop(handle);
4878 if (unlikely(ret3))
4879 ret2 = ret3;
4880 }
4881
4882 if (ret <= 0 || ret2)
4883 break;
4884 }
4885 return ret > 0 ? ret2 : ret;
4886 }
4887
4888 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4889 {
4890 int ret = 0, err = 0;
4891 struct ext4_io_end_vec *io_end_vec;
4892
4893 /*
4894 * This is somewhat ugly but the idea is clear: When transaction is
4895 * reserved, everything goes into it. Otherwise we rather start several
4896 * smaller transactions for conversion of each extent separately.
4897 */
4898 if (handle) {
4899 handle = ext4_journal_start_reserved(handle,
4900 EXT4_HT_EXT_CONVERT);
4901 if (IS_ERR(handle))
4902 return PTR_ERR(handle);
4903 }
4904
4905 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4906 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4907 io_end_vec->offset,
4908 io_end_vec->size);
4909 if (ret)
4910 break;
4911 }
4912
4913 if (handle)
4914 err = ext4_journal_stop(handle);
4915
4916 return ret < 0 ? ret : err;
4917 }
4918
4919 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4920 {
4921 __u64 physical = 0;
4922 __u64 length = 0;
4923 int blockbits = inode->i_sb->s_blocksize_bits;
4924 int error = 0;
4925 u16 iomap_type;
4926
4927 /* in-inode? */
4928 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4929 struct ext4_iloc iloc;
4930 int offset; /* offset of xattr in inode */
4931
4932 error = ext4_get_inode_loc(inode, &iloc);
4933 if (error)
4934 return error;
4935 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4936 offset = EXT4_GOOD_OLD_INODE_SIZE +
4937 EXT4_I(inode)->i_extra_isize;
4938 physical += offset;
4939 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4940 brelse(iloc.bh);
4941 iomap_type = IOMAP_INLINE;
4942 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4943 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4944 length = inode->i_sb->s_blocksize;
4945 iomap_type = IOMAP_MAPPED;
4946 } else {
4947 /* no in-inode or external block for xattr, so return -ENOENT */
4948 error = -ENOENT;
4949 goto out;
4950 }
4951
4952 iomap->addr = physical;
4953 iomap->offset = 0;
4954 iomap->length = length;
4955 iomap->type = iomap_type;
4956 iomap->flags = 0;
4957 out:
4958 return error;
4959 }
4960
4961 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4962 loff_t length, unsigned flags,
4963 struct iomap *iomap, struct iomap *srcmap)
4964 {
4965 int error;
4966
4967 error = ext4_iomap_xattr_fiemap(inode, iomap);
4968 if (error == 0 && (offset >= iomap->length))
4969 error = -ENOENT;
4970 return error;
4971 }
4972
4973 static const struct iomap_ops ext4_iomap_xattr_ops = {
4974 .iomap_begin = ext4_iomap_xattr_begin,
4975 };
4976
4977 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4978 {
4979 u64 maxbytes;
4980
4981 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4982 maxbytes = inode->i_sb->s_maxbytes;
4983 else
4984 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4985
4986 if (*len == 0)
4987 return -EINVAL;
4988 if (start > maxbytes)
4989 return -EFBIG;
4990
4991 /*
4992 * Shrink request scope to what the fs can actually handle.
4993 */
4994 if (*len > maxbytes || (maxbytes - *len) < start)
4995 *len = maxbytes - start;
4996 return 0;
4997 }
4998
4999 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5000 u64 start, u64 len)
5001 {
5002 int error = 0;
5003
5004 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5005 error = ext4_ext_precache(inode);
5006 if (error)
5007 return error;
5008 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5009 }
5010
5011 /*
5012 * For bitmap files the maximum size limit could be smaller than
5013 * s_maxbytes, so check len here manually instead of just relying on the
5014 * generic check.
5015 */
5016 error = ext4_fiemap_check_ranges(inode, start, &len);
5017 if (error)
5018 return error;
5019
5020 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5021 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
5022 return iomap_fiemap(inode, fieinfo, start, len,
5023 &ext4_iomap_xattr_ops);
5024 }
5025
5026 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
5027 }
5028
5029 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5030 __u64 start, __u64 len)
5031 {
5032 ext4_lblk_t start_blk, len_blks;
5033 __u64 last_blk;
5034 int error = 0;
5035
5036 if (ext4_has_inline_data(inode)) {
5037 int has_inline;
5038
5039 down_read(&EXT4_I(inode)->xattr_sem);
5040 has_inline = ext4_has_inline_data(inode);
5041 up_read(&EXT4_I(inode)->xattr_sem);
5042 if (has_inline)
5043 return 0;
5044 }
5045
5046 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5047 error = ext4_ext_precache(inode);
5048 if (error)
5049 return error;
5050 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5051 }
5052
5053 error = fiemap_prep(inode, fieinfo, start, &len, 0);
5054 if (error)
5055 return error;
5056
5057 error = ext4_fiemap_check_ranges(inode, start, &len);
5058 if (error)
5059 return error;
5060
5061 start_blk = start >> inode->i_sb->s_blocksize_bits;
5062 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5063 if (last_blk >= EXT_MAX_BLOCKS)
5064 last_blk = EXT_MAX_BLOCKS-1;
5065 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5066
5067 /*
5068 * Walk the extent tree gathering extent information
5069 * and pushing extents back to the user.
5070 */
5071 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5072 }
5073
5074 /*
5075 * ext4_ext_shift_path_extents:
5076 * Shift the extents of a path structure lying between path[depth].p_ext
5077 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5078 * if it is right shift or left shift operation.
5079 */
5080 static int
5081 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5082 struct inode *inode, handle_t *handle,
5083 enum SHIFT_DIRECTION SHIFT)
5084 {
5085 int depth, err = 0;
5086 struct ext4_extent *ex_start, *ex_last;
5087 bool update = false;
5088 int credits, restart_credits;
5089 depth = path->p_depth;
5090
5091 while (depth >= 0) {
5092 if (depth == path->p_depth) {
5093 ex_start = path[depth].p_ext;
5094 if (!ex_start)
5095 return -EFSCORRUPTED;
5096
5097 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5098 /* leaf + sb + inode */
5099 credits = 3;
5100 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5101 update = true;
5102 /* extent tree + sb + inode */
5103 credits = depth + 2;
5104 }
5105
5106 restart_credits = ext4_writepage_trans_blocks(inode);
5107 err = ext4_datasem_ensure_credits(handle, inode, credits,
5108 restart_credits, 0);
5109 if (err) {
5110 if (err > 0)
5111 err = -EAGAIN;
5112 goto out;
5113 }
5114
5115 err = ext4_ext_get_access(handle, inode, path + depth);
5116 if (err)
5117 goto out;
5118
5119 while (ex_start <= ex_last) {
5120 if (SHIFT == SHIFT_LEFT) {
5121 le32_add_cpu(&ex_start->ee_block,
5122 -shift);
5123 /* Try to merge to the left. */
5124 if ((ex_start >
5125 EXT_FIRST_EXTENT(path[depth].p_hdr))
5126 &&
5127 ext4_ext_try_to_merge_right(inode,
5128 path, ex_start - 1))
5129 ex_last--;
5130 else
5131 ex_start++;
5132 } else {
5133 le32_add_cpu(&ex_last->ee_block, shift);
5134 ext4_ext_try_to_merge_right(inode, path,
5135 ex_last);
5136 ex_last--;
5137 }
5138 }
5139 err = ext4_ext_dirty(handle, inode, path + depth);
5140 if (err)
5141 goto out;
5142
5143 if (--depth < 0 || !update)
5144 break;
5145 }
5146
5147 /* Update index too */
5148 err = ext4_ext_get_access(handle, inode, path + depth);
5149 if (err)
5150 goto out;
5151
5152 if (SHIFT == SHIFT_LEFT)
5153 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5154 else
5155 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5156 err = ext4_ext_dirty(handle, inode, path + depth);
5157 if (err)
5158 goto out;
5159
5160 /* we are done if current index is not a starting index */
5161 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5162 break;
5163
5164 depth--;
5165 }
5166
5167 out:
5168 return err;
5169 }
5170
5171 /*
5172 * ext4_ext_shift_extents:
5173 * All the extents which lies in the range from @start to the last allocated
5174 * block for the @inode are shifted either towards left or right (depending
5175 * upon @SHIFT) by @shift blocks.
5176 * On success, 0 is returned, error otherwise.
5177 */
5178 static int
5179 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5180 ext4_lblk_t start, ext4_lblk_t shift,
5181 enum SHIFT_DIRECTION SHIFT)
5182 {
5183 struct ext4_ext_path *path;
5184 int ret = 0, depth;
5185 struct ext4_extent *extent;
5186 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5187 ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5188
5189 /* Let path point to the last extent */
5190 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5191 EXT4_EX_NOCACHE);
5192 if (IS_ERR(path))
5193 return PTR_ERR(path);
5194
5195 depth = path->p_depth;
5196 extent = path[depth].p_ext;
5197 if (!extent)
5198 goto out;
5199
5200 stop = le32_to_cpu(extent->ee_block);
5201
5202 /*
5203 * For left shifts, make sure the hole on the left is big enough to
5204 * accommodate the shift. For right shifts, make sure the last extent
5205 * won't be shifted beyond EXT_MAX_BLOCKS.
5206 */
5207 if (SHIFT == SHIFT_LEFT) {
5208 path = ext4_find_extent(inode, start - 1, &path,
5209 EXT4_EX_NOCACHE);
5210 if (IS_ERR(path))
5211 return PTR_ERR(path);
5212 depth = path->p_depth;
5213 extent = path[depth].p_ext;
5214 if (extent) {
5215 ex_start = le32_to_cpu(extent->ee_block);
5216 ex_end = le32_to_cpu(extent->ee_block) +
5217 ext4_ext_get_actual_len(extent);
5218 } else {
5219 ex_start = 0;
5220 ex_end = 0;
5221 }
5222
5223 if ((start == ex_start && shift > ex_start) ||
5224 (shift > start - ex_end)) {
5225 ret = -EINVAL;
5226 goto out;
5227 }
5228 } else {
5229 if (shift > EXT_MAX_BLOCKS -
5230 (stop + ext4_ext_get_actual_len(extent))) {
5231 ret = -EINVAL;
5232 goto out;
5233 }
5234 }
5235
5236 /*
5237 * In case of left shift, iterator points to start and it is increased
5238 * till we reach stop. In case of right shift, iterator points to stop
5239 * and it is decreased till we reach start.
5240 */
5241 again:
5242 ret = 0;
5243 if (SHIFT == SHIFT_LEFT)
5244 iterator = &start;
5245 else
5246 iterator = &stop;
5247
5248 if (tmp != EXT_MAX_BLOCKS)
5249 *iterator = tmp;
5250
5251 /*
5252 * Its safe to start updating extents. Start and stop are unsigned, so
5253 * in case of right shift if extent with 0 block is reached, iterator
5254 * becomes NULL to indicate the end of the loop.
5255 */
5256 while (iterator && start <= stop) {
5257 path = ext4_find_extent(inode, *iterator, &path,
5258 EXT4_EX_NOCACHE);
5259 if (IS_ERR(path))
5260 return PTR_ERR(path);
5261 depth = path->p_depth;
5262 extent = path[depth].p_ext;
5263 if (!extent) {
5264 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5265 (unsigned long) *iterator);
5266 return -EFSCORRUPTED;
5267 }
5268 if (SHIFT == SHIFT_LEFT && *iterator >
5269 le32_to_cpu(extent->ee_block)) {
5270 /* Hole, move to the next extent */
5271 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5272 path[depth].p_ext++;
5273 } else {
5274 *iterator = ext4_ext_next_allocated_block(path);
5275 continue;
5276 }
5277 }
5278
5279 tmp = *iterator;
5280 if (SHIFT == SHIFT_LEFT) {
5281 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5282 *iterator = le32_to_cpu(extent->ee_block) +
5283 ext4_ext_get_actual_len(extent);
5284 } else {
5285 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5286 if (le32_to_cpu(extent->ee_block) > start)
5287 *iterator = le32_to_cpu(extent->ee_block) - 1;
5288 else if (le32_to_cpu(extent->ee_block) == start)
5289 iterator = NULL;
5290 else {
5291 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5292 while (le32_to_cpu(extent->ee_block) >= start)
5293 extent--;
5294
5295 if (extent == EXT_LAST_EXTENT(path[depth].p_hdr))
5296 break;
5297
5298 extent++;
5299 iterator = NULL;
5300 }
5301 path[depth].p_ext = extent;
5302 }
5303 ret = ext4_ext_shift_path_extents(path, shift, inode,
5304 handle, SHIFT);
5305 /* iterator can be NULL which means we should break */
5306 if (ret == -EAGAIN)
5307 goto again;
5308 if (ret)
5309 break;
5310 }
5311 out:
5312 ext4_free_ext_path(path);
5313 return ret;
5314 }
5315
5316 /*
5317 * ext4_collapse_range:
5318 * This implements the fallocate's collapse range functionality for ext4
5319 * Returns: 0 and non-zero on error.
5320 */
5321 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5322 {
5323 struct inode *inode = file_inode(file);
5324 struct super_block *sb = inode->i_sb;
5325 struct address_space *mapping = inode->i_mapping;
5326 ext4_lblk_t punch_start, punch_stop;
5327 handle_t *handle;
5328 unsigned int credits;
5329 loff_t new_size, ioffset;
5330 int ret;
5331
5332 /*
5333 * We need to test this early because xfstests assumes that a
5334 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5335 * system does not support collapse range.
5336 */
5337 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5338 return -EOPNOTSUPP;
5339
5340 /* Collapse range works only on fs cluster size aligned regions. */
5341 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5342 return -EINVAL;
5343
5344 trace_ext4_collapse_range(inode, offset, len);
5345
5346 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5347 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5348
5349 inode_lock(inode);
5350 /*
5351 * There is no need to overlap collapse range with EOF, in which case
5352 * it is effectively a truncate operation
5353 */
5354 if (offset + len >= inode->i_size) {
5355 ret = -EINVAL;
5356 goto out_mutex;
5357 }
5358
5359 /* Currently just for extent based files */
5360 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5361 ret = -EOPNOTSUPP;
5362 goto out_mutex;
5363 }
5364
5365 /* Wait for existing dio to complete */
5366 inode_dio_wait(inode);
5367
5368 ret = file_modified(file);
5369 if (ret)
5370 goto out_mutex;
5371
5372 /*
5373 * Prevent page faults from reinstantiating pages we have released from
5374 * page cache.
5375 */
5376 filemap_invalidate_lock(mapping);
5377
5378 ret = ext4_break_layouts(inode);
5379 if (ret)
5380 goto out_mmap;
5381
5382 /*
5383 * Need to round down offset to be aligned with page size boundary
5384 * for page size > block size.
5385 */
5386 ioffset = round_down(offset, PAGE_SIZE);
5387 /*
5388 * Write tail of the last page before removed range since it will get
5389 * removed from the page cache below.
5390 */
5391 ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5392 if (ret)
5393 goto out_mmap;
5394 /*
5395 * Write data that will be shifted to preserve them when discarding
5396 * page cache below. We are also protected from pages becoming dirty
5397 * by i_rwsem and invalidate_lock.
5398 */
5399 ret = filemap_write_and_wait_range(mapping, offset + len,
5400 LLONG_MAX);
5401 if (ret)
5402 goto out_mmap;
5403 truncate_pagecache(inode, ioffset);
5404
5405 credits = ext4_writepage_trans_blocks(inode);
5406 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5407 if (IS_ERR(handle)) {
5408 ret = PTR_ERR(handle);
5409 goto out_mmap;
5410 }
5411 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5412
5413 down_write(&EXT4_I(inode)->i_data_sem);
5414 ext4_discard_preallocations(inode);
5415 ext4_es_remove_extent(inode, punch_start, EXT_MAX_BLOCKS - punch_start);
5416
5417 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5418 if (ret) {
5419 up_write(&EXT4_I(inode)->i_data_sem);
5420 goto out_stop;
5421 }
5422 ext4_discard_preallocations(inode);
5423
5424 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5425 punch_stop - punch_start, SHIFT_LEFT);
5426 if (ret) {
5427 up_write(&EXT4_I(inode)->i_data_sem);
5428 goto out_stop;
5429 }
5430
5431 new_size = inode->i_size - len;
5432 i_size_write(inode, new_size);
5433 EXT4_I(inode)->i_disksize = new_size;
5434
5435 up_write(&EXT4_I(inode)->i_data_sem);
5436 if (IS_SYNC(inode))
5437 ext4_handle_sync(handle);
5438 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
5439 ret = ext4_mark_inode_dirty(handle, inode);
5440 ext4_update_inode_fsync_trans(handle, inode, 1);
5441
5442 out_stop:
5443 ext4_journal_stop(handle);
5444 out_mmap:
5445 filemap_invalidate_unlock(mapping);
5446 out_mutex:
5447 inode_unlock(inode);
5448 return ret;
5449 }
5450
5451 /*
5452 * ext4_insert_range:
5453 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5454 * The data blocks starting from @offset to the EOF are shifted by @len
5455 * towards right to create a hole in the @inode. Inode size is increased
5456 * by len bytes.
5457 * Returns 0 on success, error otherwise.
5458 */
5459 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5460 {
5461 struct inode *inode = file_inode(file);
5462 struct super_block *sb = inode->i_sb;
5463 struct address_space *mapping = inode->i_mapping;
5464 handle_t *handle;
5465 struct ext4_ext_path *path;
5466 struct ext4_extent *extent;
5467 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5468 unsigned int credits, ee_len;
5469 int ret = 0, depth, split_flag = 0;
5470 loff_t ioffset;
5471
5472 /*
5473 * We need to test this early because xfstests assumes that an
5474 * insert range of (0, 1) will return EOPNOTSUPP if the file
5475 * system does not support insert range.
5476 */
5477 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5478 return -EOPNOTSUPP;
5479
5480 /* Insert range works only on fs cluster size aligned regions. */
5481 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5482 return -EINVAL;
5483
5484 trace_ext4_insert_range(inode, offset, len);
5485
5486 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5487 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5488
5489 inode_lock(inode);
5490 /* Currently just for extent based files */
5491 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5492 ret = -EOPNOTSUPP;
5493 goto out_mutex;
5494 }
5495
5496 /* Check whether the maximum file size would be exceeded */
5497 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5498 ret = -EFBIG;
5499 goto out_mutex;
5500 }
5501
5502 /* Offset must be less than i_size */
5503 if (offset >= inode->i_size) {
5504 ret = -EINVAL;
5505 goto out_mutex;
5506 }
5507
5508 /* Wait for existing dio to complete */
5509 inode_dio_wait(inode);
5510
5511 ret = file_modified(file);
5512 if (ret)
5513 goto out_mutex;
5514
5515 /*
5516 * Prevent page faults from reinstantiating pages we have released from
5517 * page cache.
5518 */
5519 filemap_invalidate_lock(mapping);
5520
5521 ret = ext4_break_layouts(inode);
5522 if (ret)
5523 goto out_mmap;
5524
5525 /*
5526 * Need to round down to align start offset to page size boundary
5527 * for page size > block size.
5528 */
5529 ioffset = round_down(offset, PAGE_SIZE);
5530 /* Write out all dirty pages */
5531 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5532 LLONG_MAX);
5533 if (ret)
5534 goto out_mmap;
5535 truncate_pagecache(inode, ioffset);
5536
5537 credits = ext4_writepage_trans_blocks(inode);
5538 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5539 if (IS_ERR(handle)) {
5540 ret = PTR_ERR(handle);
5541 goto out_mmap;
5542 }
5543 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5544
5545 /* Expand file to avoid data loss if there is error while shifting */
5546 inode->i_size += len;
5547 EXT4_I(inode)->i_disksize += len;
5548 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
5549 ret = ext4_mark_inode_dirty(handle, inode);
5550 if (ret)
5551 goto out_stop;
5552
5553 down_write(&EXT4_I(inode)->i_data_sem);
5554 ext4_discard_preallocations(inode);
5555
5556 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5557 if (IS_ERR(path)) {
5558 up_write(&EXT4_I(inode)->i_data_sem);
5559 ret = PTR_ERR(path);
5560 goto out_stop;
5561 }
5562
5563 depth = ext_depth(inode);
5564 extent = path[depth].p_ext;
5565 if (extent) {
5566 ee_start_lblk = le32_to_cpu(extent->ee_block);
5567 ee_len = ext4_ext_get_actual_len(extent);
5568
5569 /*
5570 * If offset_lblk is not the starting block of extent, split
5571 * the extent @offset_lblk
5572 */
5573 if ((offset_lblk > ee_start_lblk) &&
5574 (offset_lblk < (ee_start_lblk + ee_len))) {
5575 if (ext4_ext_is_unwritten(extent))
5576 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5577 EXT4_EXT_MARK_UNWRIT2;
5578 ret = ext4_split_extent_at(handle, inode, &path,
5579 offset_lblk, split_flag,
5580 EXT4_EX_NOCACHE |
5581 EXT4_GET_BLOCKS_PRE_IO |
5582 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5583 }
5584
5585 ext4_free_ext_path(path);
5586 if (ret < 0) {
5587 up_write(&EXT4_I(inode)->i_data_sem);
5588 goto out_stop;
5589 }
5590 } else {
5591 ext4_free_ext_path(path);
5592 }
5593
5594 ext4_es_remove_extent(inode, offset_lblk, EXT_MAX_BLOCKS - offset_lblk);
5595
5596 /*
5597 * if offset_lblk lies in a hole which is at start of file, use
5598 * ee_start_lblk to shift extents
5599 */
5600 ret = ext4_ext_shift_extents(inode, handle,
5601 max(ee_start_lblk, offset_lblk), len_lblk, SHIFT_RIGHT);
5602
5603 up_write(&EXT4_I(inode)->i_data_sem);
5604 if (IS_SYNC(inode))
5605 ext4_handle_sync(handle);
5606 if (ret >= 0)
5607 ext4_update_inode_fsync_trans(handle, inode, 1);
5608
5609 out_stop:
5610 ext4_journal_stop(handle);
5611 out_mmap:
5612 filemap_invalidate_unlock(mapping);
5613 out_mutex:
5614 inode_unlock(inode);
5615 return ret;
5616 }
5617
5618 /**
5619 * ext4_swap_extents() - Swap extents between two inodes
5620 * @handle: handle for this transaction
5621 * @inode1: First inode
5622 * @inode2: Second inode
5623 * @lblk1: Start block for first inode
5624 * @lblk2: Start block for second inode
5625 * @count: Number of blocks to swap
5626 * @unwritten: Mark second inode's extents as unwritten after swap
5627 * @erp: Pointer to save error value
5628 *
5629 * This helper routine does exactly what is promise "swap extents". All other
5630 * stuff such as page-cache locking consistency, bh mapping consistency or
5631 * extent's data copying must be performed by caller.
5632 * Locking:
5633 * i_rwsem is held for both inodes
5634 * i_data_sem is locked for write for both inodes
5635 * Assumptions:
5636 * All pages from requested range are locked for both inodes
5637 */
5638 int
5639 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5640 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5641 ext4_lblk_t count, int unwritten, int *erp)
5642 {
5643 struct ext4_ext_path *path1 = NULL;
5644 struct ext4_ext_path *path2 = NULL;
5645 int replaced_count = 0;
5646
5647 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5648 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5649 BUG_ON(!inode_is_locked(inode1));
5650 BUG_ON(!inode_is_locked(inode2));
5651
5652 ext4_es_remove_extent(inode1, lblk1, count);
5653 ext4_es_remove_extent(inode2, lblk2, count);
5654
5655 while (count) {
5656 struct ext4_extent *ex1, *ex2, tmp_ex;
5657 ext4_lblk_t e1_blk, e2_blk;
5658 int e1_len, e2_len, len;
5659 int split = 0;
5660
5661 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5662 if (IS_ERR(path1)) {
5663 *erp = PTR_ERR(path1);
5664 path1 = NULL;
5665 finish:
5666 count = 0;
5667 goto repeat;
5668 }
5669 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5670 if (IS_ERR(path2)) {
5671 *erp = PTR_ERR(path2);
5672 path2 = NULL;
5673 goto finish;
5674 }
5675 ex1 = path1[path1->p_depth].p_ext;
5676 ex2 = path2[path2->p_depth].p_ext;
5677 /* Do we have something to swap ? */
5678 if (unlikely(!ex2 || !ex1))
5679 goto finish;
5680
5681 e1_blk = le32_to_cpu(ex1->ee_block);
5682 e2_blk = le32_to_cpu(ex2->ee_block);
5683 e1_len = ext4_ext_get_actual_len(ex1);
5684 e2_len = ext4_ext_get_actual_len(ex2);
5685
5686 /* Hole handling */
5687 if (!in_range(lblk1, e1_blk, e1_len) ||
5688 !in_range(lblk2, e2_blk, e2_len)) {
5689 ext4_lblk_t next1, next2;
5690
5691 /* if hole after extent, then go to next extent */
5692 next1 = ext4_ext_next_allocated_block(path1);
5693 next2 = ext4_ext_next_allocated_block(path2);
5694 /* If hole before extent, then shift to that extent */
5695 if (e1_blk > lblk1)
5696 next1 = e1_blk;
5697 if (e2_blk > lblk2)
5698 next2 = e2_blk;
5699 /* Do we have something to swap */
5700 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5701 goto finish;
5702 /* Move to the rightest boundary */
5703 len = next1 - lblk1;
5704 if (len < next2 - lblk2)
5705 len = next2 - lblk2;
5706 if (len > count)
5707 len = count;
5708 lblk1 += len;
5709 lblk2 += len;
5710 count -= len;
5711 goto repeat;
5712 }
5713
5714 /* Prepare left boundary */
5715 if (e1_blk < lblk1) {
5716 split = 1;
5717 *erp = ext4_force_split_extent_at(handle, inode1,
5718 &path1, lblk1, 0);
5719 if (unlikely(*erp))
5720 goto finish;
5721 }
5722 if (e2_blk < lblk2) {
5723 split = 1;
5724 *erp = ext4_force_split_extent_at(handle, inode2,
5725 &path2, lblk2, 0);
5726 if (unlikely(*erp))
5727 goto finish;
5728 }
5729 /* ext4_split_extent_at() may result in leaf extent split,
5730 * path must to be revalidated. */
5731 if (split)
5732 goto repeat;
5733
5734 /* Prepare right boundary */
5735 len = count;
5736 if (len > e1_blk + e1_len - lblk1)
5737 len = e1_blk + e1_len - lblk1;
5738 if (len > e2_blk + e2_len - lblk2)
5739 len = e2_blk + e2_len - lblk2;
5740
5741 if (len != e1_len) {
5742 split = 1;
5743 *erp = ext4_force_split_extent_at(handle, inode1,
5744 &path1, lblk1 + len, 0);
5745 if (unlikely(*erp))
5746 goto finish;
5747 }
5748 if (len != e2_len) {
5749 split = 1;
5750 *erp = ext4_force_split_extent_at(handle, inode2,
5751 &path2, lblk2 + len, 0);
5752 if (*erp)
5753 goto finish;
5754 }
5755 /* ext4_split_extent_at() may result in leaf extent split,
5756 * path must to be revalidated. */
5757 if (split)
5758 goto repeat;
5759
5760 BUG_ON(e2_len != e1_len);
5761 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5762 if (unlikely(*erp))
5763 goto finish;
5764 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5765 if (unlikely(*erp))
5766 goto finish;
5767
5768 /* Both extents are fully inside boundaries. Swap it now */
5769 tmp_ex = *ex1;
5770 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5771 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5772 ex1->ee_len = cpu_to_le16(e2_len);
5773 ex2->ee_len = cpu_to_le16(e1_len);
5774 if (unwritten)
5775 ext4_ext_mark_unwritten(ex2);
5776 if (ext4_ext_is_unwritten(&tmp_ex))
5777 ext4_ext_mark_unwritten(ex1);
5778
5779 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5780 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5781 *erp = ext4_ext_dirty(handle, inode2, path2 +
5782 path2->p_depth);
5783 if (unlikely(*erp))
5784 goto finish;
5785 *erp = ext4_ext_dirty(handle, inode1, path1 +
5786 path1->p_depth);
5787 /*
5788 * Looks scarry ah..? second inode already points to new blocks,
5789 * and it was successfully dirtied. But luckily error may happen
5790 * only due to journal error, so full transaction will be
5791 * aborted anyway.
5792 */
5793 if (unlikely(*erp))
5794 goto finish;
5795 lblk1 += len;
5796 lblk2 += len;
5797 replaced_count += len;
5798 count -= len;
5799
5800 repeat:
5801 ext4_free_ext_path(path1);
5802 ext4_free_ext_path(path2);
5803 path1 = path2 = NULL;
5804 }
5805 return replaced_count;
5806 }
5807
5808 /*
5809 * ext4_clu_mapped - determine whether any block in a logical cluster has
5810 * been mapped to a physical cluster
5811 *
5812 * @inode - file containing the logical cluster
5813 * @lclu - logical cluster of interest
5814 *
5815 * Returns 1 if any block in the logical cluster is mapped, signifying
5816 * that a physical cluster has been allocated for it. Otherwise,
5817 * returns 0. Can also return negative error codes. Derived from
5818 * ext4_ext_map_blocks().
5819 */
5820 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5821 {
5822 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5823 struct ext4_ext_path *path;
5824 int depth, mapped = 0, err = 0;
5825 struct ext4_extent *extent;
5826 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5827
5828 /*
5829 * if data can be stored inline, the logical cluster isn't
5830 * mapped - no physical clusters have been allocated, and the
5831 * file has no extents
5832 */
5833 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) ||
5834 ext4_has_inline_data(inode))
5835 return 0;
5836
5837 /* search for the extent closest to the first block in the cluster */
5838 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5839 if (IS_ERR(path)) {
5840 err = PTR_ERR(path);
5841 path = NULL;
5842 goto out;
5843 }
5844
5845 depth = ext_depth(inode);
5846
5847 /*
5848 * A consistent leaf must not be empty. This situation is possible,
5849 * though, _during_ tree modification, and it's why an assert can't
5850 * be put in ext4_find_extent().
5851 */
5852 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5853 EXT4_ERROR_INODE(inode,
5854 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5855 (unsigned long) EXT4_C2B(sbi, lclu),
5856 depth, path[depth].p_block);
5857 err = -EFSCORRUPTED;
5858 goto out;
5859 }
5860
5861 extent = path[depth].p_ext;
5862
5863 /* can't be mapped if the extent tree is empty */
5864 if (extent == NULL)
5865 goto out;
5866
5867 first_lblk = le32_to_cpu(extent->ee_block);
5868 first_lclu = EXT4_B2C(sbi, first_lblk);
5869
5870 /*
5871 * Three possible outcomes at this point - found extent spanning
5872 * the target cluster, to the left of the target cluster, or to the
5873 * right of the target cluster. The first two cases are handled here.
5874 * The last case indicates the target cluster is not mapped.
5875 */
5876 if (lclu >= first_lclu) {
5877 last_lclu = EXT4_B2C(sbi, first_lblk +
5878 ext4_ext_get_actual_len(extent) - 1);
5879 if (lclu <= last_lclu) {
5880 mapped = 1;
5881 } else {
5882 first_lblk = ext4_ext_next_allocated_block(path);
5883 first_lclu = EXT4_B2C(sbi, first_lblk);
5884 if (lclu == first_lclu)
5885 mapped = 1;
5886 }
5887 }
5888
5889 out:
5890 ext4_free_ext_path(path);
5891
5892 return err ? err : mapped;
5893 }
5894
5895 /*
5896 * Updates physical block address and unwritten status of extent
5897 * starting at lblk start and of len. If such an extent doesn't exist,
5898 * this function splits the extent tree appropriately to create an
5899 * extent like this. This function is called in the fast commit
5900 * replay path. Returns 0 on success and error on failure.
5901 */
5902 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5903 int len, int unwritten, ext4_fsblk_t pblk)
5904 {
5905 struct ext4_ext_path *path;
5906 struct ext4_extent *ex;
5907 int ret;
5908
5909 path = ext4_find_extent(inode, start, NULL, 0);
5910 if (IS_ERR(path))
5911 return PTR_ERR(path);
5912 ex = path[path->p_depth].p_ext;
5913 if (!ex) {
5914 ret = -EFSCORRUPTED;
5915 goto out;
5916 }
5917
5918 if (le32_to_cpu(ex->ee_block) != start ||
5919 ext4_ext_get_actual_len(ex) != len) {
5920 /* We need to split this extent to match our extent first */
5921 down_write(&EXT4_I(inode)->i_data_sem);
5922 ret = ext4_force_split_extent_at(NULL, inode, &path, start, 1);
5923 up_write(&EXT4_I(inode)->i_data_sem);
5924 if (ret)
5925 goto out;
5926
5927 path = ext4_find_extent(inode, start, &path, 0);
5928 if (IS_ERR(path))
5929 return PTR_ERR(path);
5930 ex = path[path->p_depth].p_ext;
5931 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5932
5933 if (ext4_ext_get_actual_len(ex) != len) {
5934 down_write(&EXT4_I(inode)->i_data_sem);
5935 ret = ext4_force_split_extent_at(NULL, inode, &path,
5936 start + len, 1);
5937 up_write(&EXT4_I(inode)->i_data_sem);
5938 if (ret)
5939 goto out;
5940
5941 path = ext4_find_extent(inode, start, &path, 0);
5942 if (IS_ERR(path))
5943 return PTR_ERR(path);
5944 ex = path[path->p_depth].p_ext;
5945 }
5946 }
5947 if (unwritten)
5948 ext4_ext_mark_unwritten(ex);
5949 else
5950 ext4_ext_mark_initialized(ex);
5951 ext4_ext_store_pblock(ex, pblk);
5952 down_write(&EXT4_I(inode)->i_data_sem);
5953 ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5954 up_write(&EXT4_I(inode)->i_data_sem);
5955 out:
5956 ext4_free_ext_path(path);
5957 ext4_mark_inode_dirty(NULL, inode);
5958 return ret;
5959 }
5960
5961 /* Try to shrink the extent tree */
5962 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5963 {
5964 struct ext4_ext_path *path = NULL;
5965 struct ext4_extent *ex;
5966 ext4_lblk_t old_cur, cur = 0;
5967
5968 while (cur < end) {
5969 path = ext4_find_extent(inode, cur, NULL, 0);
5970 if (IS_ERR(path))
5971 return;
5972 ex = path[path->p_depth].p_ext;
5973 if (!ex) {
5974 ext4_free_ext_path(path);
5975 ext4_mark_inode_dirty(NULL, inode);
5976 return;
5977 }
5978 old_cur = cur;
5979 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5980 if (cur <= old_cur)
5981 cur = old_cur + 1;
5982 ext4_ext_try_to_merge(NULL, inode, path, ex);
5983 down_write(&EXT4_I(inode)->i_data_sem);
5984 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5985 up_write(&EXT4_I(inode)->i_data_sem);
5986 ext4_mark_inode_dirty(NULL, inode);
5987 ext4_free_ext_path(path);
5988 }
5989 }
5990
5991 /* Check if *cur is a hole and if it is, skip it */
5992 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5993 {
5994 int ret;
5995 struct ext4_map_blocks map;
5996
5997 map.m_lblk = *cur;
5998 map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5999
6000 ret = ext4_map_blocks(NULL, inode, &map, 0);
6001 if (ret < 0)
6002 return ret;
6003 if (ret != 0)
6004 return 0;
6005 *cur = *cur + map.m_len;
6006 return 0;
6007 }
6008
6009 /* Count number of blocks used by this inode and update i_blocks */
6010 int ext4_ext_replay_set_iblocks(struct inode *inode)
6011 {
6012 struct ext4_ext_path *path = NULL, *path2 = NULL;
6013 struct ext4_extent *ex;
6014 ext4_lblk_t cur = 0, end;
6015 int numblks = 0, i, ret = 0;
6016 ext4_fsblk_t cmp1, cmp2;
6017 struct ext4_map_blocks map;
6018
6019 /* Determin the size of the file first */
6020 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6021 EXT4_EX_NOCACHE);
6022 if (IS_ERR(path))
6023 return PTR_ERR(path);
6024 ex = path[path->p_depth].p_ext;
6025 if (!ex) {
6026 ext4_free_ext_path(path);
6027 goto out;
6028 }
6029 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6030 ext4_free_ext_path(path);
6031
6032 /* Count the number of data blocks */
6033 cur = 0;
6034 while (cur < end) {
6035 map.m_lblk = cur;
6036 map.m_len = end - cur;
6037 ret = ext4_map_blocks(NULL, inode, &map, 0);
6038 if (ret < 0)
6039 break;
6040 if (ret > 0)
6041 numblks += ret;
6042 cur = cur + map.m_len;
6043 }
6044
6045 /*
6046 * Count the number of extent tree blocks. We do it by looking up
6047 * two successive extents and determining the difference between
6048 * their paths. When path is different for 2 successive extents
6049 * we compare the blocks in the path at each level and increment
6050 * iblocks by total number of differences found.
6051 */
6052 cur = 0;
6053 ret = skip_hole(inode, &cur);
6054 if (ret < 0)
6055 goto out;
6056 path = ext4_find_extent(inode, cur, NULL, 0);
6057 if (IS_ERR(path))
6058 goto out;
6059 numblks += path->p_depth;
6060 ext4_free_ext_path(path);
6061 while (cur < end) {
6062 path = ext4_find_extent(inode, cur, NULL, 0);
6063 if (IS_ERR(path))
6064 break;
6065 ex = path[path->p_depth].p_ext;
6066 if (!ex) {
6067 ext4_free_ext_path(path);
6068 return 0;
6069 }
6070 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6071 ext4_ext_get_actual_len(ex));
6072 ret = skip_hole(inode, &cur);
6073 if (ret < 0) {
6074 ext4_free_ext_path(path);
6075 break;
6076 }
6077 path2 = ext4_find_extent(inode, cur, NULL, 0);
6078 if (IS_ERR(path2)) {
6079 ext4_free_ext_path(path);
6080 break;
6081 }
6082 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6083 cmp1 = cmp2 = 0;
6084 if (i <= path->p_depth)
6085 cmp1 = path[i].p_bh ?
6086 path[i].p_bh->b_blocknr : 0;
6087 if (i <= path2->p_depth)
6088 cmp2 = path2[i].p_bh ?
6089 path2[i].p_bh->b_blocknr : 0;
6090 if (cmp1 != cmp2 && cmp2 != 0)
6091 numblks++;
6092 }
6093 ext4_free_ext_path(path);
6094 ext4_free_ext_path(path2);
6095 }
6096
6097 out:
6098 inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6099 ext4_mark_inode_dirty(NULL, inode);
6100 return 0;
6101 }
6102
6103 int ext4_ext_clear_bb(struct inode *inode)
6104 {
6105 struct ext4_ext_path *path = NULL;
6106 struct ext4_extent *ex;
6107 ext4_lblk_t cur = 0, end;
6108 int j, ret = 0;
6109 struct ext4_map_blocks map;
6110
6111 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6112 return 0;
6113
6114 /* Determin the size of the file first */
6115 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6116 EXT4_EX_NOCACHE);
6117 if (IS_ERR(path))
6118 return PTR_ERR(path);
6119 ex = path[path->p_depth].p_ext;
6120 if (!ex) {
6121 ext4_free_ext_path(path);
6122 return 0;
6123 }
6124 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6125 ext4_free_ext_path(path);
6126
6127 cur = 0;
6128 while (cur < end) {
6129 map.m_lblk = cur;
6130 map.m_len = end - cur;
6131 ret = ext4_map_blocks(NULL, inode, &map, 0);
6132 if (ret < 0)
6133 break;
6134 if (ret > 0) {
6135 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6136 if (!IS_ERR_OR_NULL(path)) {
6137 for (j = 0; j < path->p_depth; j++) {
6138
6139 ext4_mb_mark_bb(inode->i_sb,
6140 path[j].p_block, 1, false);
6141 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6142 0, path[j].p_block, 1, 1);
6143 }
6144 ext4_free_ext_path(path);
6145 }
6146 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, false);
6147 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6148 map.m_lblk, map.m_pblk, map.m_len, 1);
6149 }
6150 cur = cur + map.m_len;
6151 }
6152
6153 return 0;
6154 }
6155
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.