Loading...
1/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
21 */
22
23/*
24 * Extents support for EXT4
25 *
26 * TODO:
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
30 */
31
32#include <linux/fs.h>
33#include <linux/time.h>
34#include <linux/jbd2.h>
35#include <linux/highuid.h>
36#include <linux/pagemap.h>
37#include <linux/quotaops.h>
38#include <linux/string.h>
39#include <linux/slab.h>
40#include <asm/uaccess.h>
41#include <linux/fiemap.h>
42#include <linux/backing-dev.h>
43#include "ext4_jbd2.h"
44#include "ext4_extents.h"
45#include "xattr.h"
46
47#include <trace/events/ext4.h>
48
49/*
50 * used by extent splitting.
51 */
52#define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 due to ENOSPC */
54#define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
55#define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
56
57#define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58#define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59
60static __le32 ext4_extent_block_csum(struct inode *inode,
61 struct ext4_extent_header *eh)
62{
63 struct ext4_inode_info *ei = EXT4_I(inode);
64 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
65 __u32 csum;
66
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
68 EXT4_EXTENT_TAIL_OFFSET(eh));
69 return cpu_to_le32(csum);
70}
71
72static int ext4_extent_block_csum_verify(struct inode *inode,
73 struct ext4_extent_header *eh)
74{
75 struct ext4_extent_tail *et;
76
77 if (!ext4_has_metadata_csum(inode->i_sb))
78 return 1;
79
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
82 return 0;
83 return 1;
84}
85
86static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
88{
89 struct ext4_extent_tail *et;
90
91 if (!ext4_has_metadata_csum(inode->i_sb))
92 return;
93
94 et = find_ext4_extent_tail(eh);
95 et->et_checksum = ext4_extent_block_csum(inode, eh);
96}
97
98static int ext4_split_extent(handle_t *handle,
99 struct inode *inode,
100 struct ext4_ext_path **ppath,
101 struct ext4_map_blocks *map,
102 int split_flag,
103 int flags);
104
105static int ext4_split_extent_at(handle_t *handle,
106 struct inode *inode,
107 struct ext4_ext_path **ppath,
108 ext4_lblk_t split,
109 int split_flag,
110 int flags);
111
112static int ext4_find_delayed_extent(struct inode *inode,
113 struct extent_status *newes);
114
115static int ext4_ext_truncate_extend_restart(handle_t *handle,
116 struct inode *inode,
117 int needed)
118{
119 int err;
120
121 if (!ext4_handle_valid(handle))
122 return 0;
123 if (handle->h_buffer_credits > needed)
124 return 0;
125 err = ext4_journal_extend(handle, needed);
126 if (err <= 0)
127 return err;
128 err = ext4_truncate_restart_trans(handle, inode, needed);
129 if (err == 0)
130 err = -EAGAIN;
131
132 return err;
133}
134
135/*
136 * could return:
137 * - EROFS
138 * - ENOMEM
139 */
140static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
141 struct ext4_ext_path *path)
142{
143 if (path->p_bh) {
144 /* path points to block */
145 BUFFER_TRACE(path->p_bh, "get_write_access");
146 return ext4_journal_get_write_access(handle, path->p_bh);
147 }
148 /* path points to leaf/index in inode body */
149 /* we use in-core data, no need to protect them */
150 return 0;
151}
152
153/*
154 * could return:
155 * - EROFS
156 * - ENOMEM
157 * - EIO
158 */
159int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
160 struct inode *inode, struct ext4_ext_path *path)
161{
162 int err;
163
164 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
165 if (path->p_bh) {
166 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
167 /* path points to block */
168 err = __ext4_handle_dirty_metadata(where, line, handle,
169 inode, path->p_bh);
170 } else {
171 /* path points to leaf/index in inode body */
172 err = ext4_mark_inode_dirty(handle, inode);
173 }
174 return err;
175}
176
177static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
178 struct ext4_ext_path *path,
179 ext4_lblk_t block)
180{
181 if (path) {
182 int depth = path->p_depth;
183 struct ext4_extent *ex;
184
185 /*
186 * Try to predict block placement assuming that we are
187 * filling in a file which will eventually be
188 * non-sparse --- i.e., in the case of libbfd writing
189 * an ELF object sections out-of-order but in a way
190 * the eventually results in a contiguous object or
191 * executable file, or some database extending a table
192 * space file. However, this is actually somewhat
193 * non-ideal if we are writing a sparse file such as
194 * qemu or KVM writing a raw image file that is going
195 * to stay fairly sparse, since it will end up
196 * fragmenting the file system's free space. Maybe we
197 * should have some hueristics or some way to allow
198 * userspace to pass a hint to file system,
199 * especially if the latter case turns out to be
200 * common.
201 */
202 ex = path[depth].p_ext;
203 if (ex) {
204 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
205 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
206
207 if (block > ext_block)
208 return ext_pblk + (block - ext_block);
209 else
210 return ext_pblk - (ext_block - block);
211 }
212
213 /* it looks like index is empty;
214 * try to find starting block from index itself */
215 if (path[depth].p_bh)
216 return path[depth].p_bh->b_blocknr;
217 }
218
219 /* OK. use inode's group */
220 return ext4_inode_to_goal_block(inode);
221}
222
223/*
224 * Allocation for a meta data block
225 */
226static ext4_fsblk_t
227ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
228 struct ext4_ext_path *path,
229 struct ext4_extent *ex, int *err, unsigned int flags)
230{
231 ext4_fsblk_t goal, newblock;
232
233 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
234 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
235 NULL, err);
236 return newblock;
237}
238
239static inline int ext4_ext_space_block(struct inode *inode, int check)
240{
241 int size;
242
243 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
244 / sizeof(struct ext4_extent);
245#ifdef AGGRESSIVE_TEST
246 if (!check && size > 6)
247 size = 6;
248#endif
249 return size;
250}
251
252static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
253{
254 int size;
255
256 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
257 / sizeof(struct ext4_extent_idx);
258#ifdef AGGRESSIVE_TEST
259 if (!check && size > 5)
260 size = 5;
261#endif
262 return size;
263}
264
265static inline int ext4_ext_space_root(struct inode *inode, int check)
266{
267 int size;
268
269 size = sizeof(EXT4_I(inode)->i_data);
270 size -= sizeof(struct ext4_extent_header);
271 size /= sizeof(struct ext4_extent);
272#ifdef AGGRESSIVE_TEST
273 if (!check && size > 3)
274 size = 3;
275#endif
276 return size;
277}
278
279static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
280{
281 int size;
282
283 size = sizeof(EXT4_I(inode)->i_data);
284 size -= sizeof(struct ext4_extent_header);
285 size /= sizeof(struct ext4_extent_idx);
286#ifdef AGGRESSIVE_TEST
287 if (!check && size > 4)
288 size = 4;
289#endif
290 return size;
291}
292
293static inline int
294ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
295 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
296 int nofail)
297{
298 struct ext4_ext_path *path = *ppath;
299 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
300
301 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
302 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
303 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
304 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
305}
306
307/*
308 * Calculate the number of metadata blocks needed
309 * to allocate @blocks
310 * Worse case is one block per extent
311 */
312int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
313{
314 struct ext4_inode_info *ei = EXT4_I(inode);
315 int idxs;
316
317 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
318 / sizeof(struct ext4_extent_idx));
319
320 /*
321 * If the new delayed allocation block is contiguous with the
322 * previous da block, it can share index blocks with the
323 * previous block, so we only need to allocate a new index
324 * block every idxs leaf blocks. At ldxs**2 blocks, we need
325 * an additional index block, and at ldxs**3 blocks, yet
326 * another index blocks.
327 */
328 if (ei->i_da_metadata_calc_len &&
329 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
330 int num = 0;
331
332 if ((ei->i_da_metadata_calc_len % idxs) == 0)
333 num++;
334 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
335 num++;
336 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
337 num++;
338 ei->i_da_metadata_calc_len = 0;
339 } else
340 ei->i_da_metadata_calc_len++;
341 ei->i_da_metadata_calc_last_lblock++;
342 return num;
343 }
344
345 /*
346 * In the worst case we need a new set of index blocks at
347 * every level of the inode's extent tree.
348 */
349 ei->i_da_metadata_calc_len = 1;
350 ei->i_da_metadata_calc_last_lblock = lblock;
351 return ext_depth(inode) + 1;
352}
353
354static int
355ext4_ext_max_entries(struct inode *inode, int depth)
356{
357 int max;
358
359 if (depth == ext_depth(inode)) {
360 if (depth == 0)
361 max = ext4_ext_space_root(inode, 1);
362 else
363 max = ext4_ext_space_root_idx(inode, 1);
364 } else {
365 if (depth == 0)
366 max = ext4_ext_space_block(inode, 1);
367 else
368 max = ext4_ext_space_block_idx(inode, 1);
369 }
370
371 return max;
372}
373
374static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
375{
376 ext4_fsblk_t block = ext4_ext_pblock(ext);
377 int len = ext4_ext_get_actual_len(ext);
378 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
379 ext4_lblk_t last = lblock + len - 1;
380
381 if (len == 0 || lblock > last)
382 return 0;
383 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
384}
385
386static int ext4_valid_extent_idx(struct inode *inode,
387 struct ext4_extent_idx *ext_idx)
388{
389 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
390
391 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
392}
393
394static int ext4_valid_extent_entries(struct inode *inode,
395 struct ext4_extent_header *eh,
396 int depth)
397{
398 unsigned short entries;
399 if (eh->eh_entries == 0)
400 return 1;
401
402 entries = le16_to_cpu(eh->eh_entries);
403
404 if (depth == 0) {
405 /* leaf entries */
406 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
407 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
408 ext4_fsblk_t pblock = 0;
409 ext4_lblk_t lblock = 0;
410 ext4_lblk_t prev = 0;
411 int len = 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 len = ext4_ext_get_actual_len(ext);
419 if ((lblock <= prev) && prev) {
420 pblock = ext4_ext_pblock(ext);
421 es->s_last_error_block = cpu_to_le64(pblock);
422 return 0;
423 }
424 ext++;
425 entries--;
426 prev = lblock + len - 1;
427 }
428 } else {
429 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
430 while (entries) {
431 if (!ext4_valid_extent_idx(inode, ext_idx))
432 return 0;
433 ext_idx++;
434 entries--;
435 }
436 }
437 return 1;
438}
439
440static int __ext4_ext_check(const char *function, unsigned int line,
441 struct inode *inode, struct ext4_extent_header *eh,
442 int depth, ext4_fsblk_t pblk)
443{
444 const char *error_msg;
445 int max = 0, err = -EFSCORRUPTED;
446
447 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
448 error_msg = "invalid magic";
449 goto corrupted;
450 }
451 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
452 error_msg = "unexpected eh_depth";
453 goto corrupted;
454 }
455 if (unlikely(eh->eh_max == 0)) {
456 error_msg = "invalid eh_max";
457 goto corrupted;
458 }
459 max = ext4_ext_max_entries(inode, depth);
460 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
461 error_msg = "too large eh_max";
462 goto corrupted;
463 }
464 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
465 error_msg = "invalid eh_entries";
466 goto corrupted;
467 }
468 if (!ext4_valid_extent_entries(inode, eh, depth)) {
469 error_msg = "invalid extent entries";
470 goto corrupted;
471 }
472 /* Verify checksum on non-root extent tree nodes */
473 if (ext_depth(inode) != depth &&
474 !ext4_extent_block_csum_verify(inode, eh)) {
475 error_msg = "extent tree corrupted";
476 err = -EFSBADCRC;
477 goto corrupted;
478 }
479 return 0;
480
481corrupted:
482 ext4_error_inode(inode, function, line, 0,
483 "pblk %llu bad header/extent: %s - magic %x, "
484 "entries %u, max %u(%u), depth %u(%u)",
485 (unsigned long long) pblk, error_msg,
486 le16_to_cpu(eh->eh_magic),
487 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
488 max, le16_to_cpu(eh->eh_depth), depth);
489 return err;
490}
491
492#define ext4_ext_check(inode, eh, depth, pblk) \
493 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
494
495int ext4_ext_check_inode(struct inode *inode)
496{
497 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
498}
499
500static struct buffer_head *
501__read_extent_tree_block(const char *function, unsigned int line,
502 struct inode *inode, ext4_fsblk_t pblk, int depth,
503 int flags)
504{
505 struct buffer_head *bh;
506 int err;
507
508 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
509 if (unlikely(!bh))
510 return ERR_PTR(-ENOMEM);
511
512 if (!bh_uptodate_or_lock(bh)) {
513 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
514 err = bh_submit_read(bh);
515 if (err < 0)
516 goto errout;
517 }
518 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
519 return bh;
520 err = __ext4_ext_check(function, line, inode,
521 ext_block_hdr(bh), depth, pblk);
522 if (err)
523 goto errout;
524 set_buffer_verified(bh);
525 /*
526 * If this is a leaf block, cache all of its entries
527 */
528 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
529 struct ext4_extent_header *eh = ext_block_hdr(bh);
530 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
531 ext4_lblk_t prev = 0;
532 int i;
533
534 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
535 unsigned int status = EXTENT_STATUS_WRITTEN;
536 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
537 int len = ext4_ext_get_actual_len(ex);
538
539 if (prev && (prev != lblk))
540 ext4_es_cache_extent(inode, prev,
541 lblk - prev, ~0,
542 EXTENT_STATUS_HOLE);
543
544 if (ext4_ext_is_unwritten(ex))
545 status = EXTENT_STATUS_UNWRITTEN;
546 ext4_es_cache_extent(inode, lblk, len,
547 ext4_ext_pblock(ex), status);
548 prev = lblk + len;
549 }
550 }
551 return bh;
552errout:
553 put_bh(bh);
554 return ERR_PTR(err);
555
556}
557
558#define read_extent_tree_block(inode, pblk, depth, flags) \
559 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
560 (depth), (flags))
561
562/*
563 * This function is called to cache a file's extent information in the
564 * extent status tree
565 */
566int ext4_ext_precache(struct inode *inode)
567{
568 struct ext4_inode_info *ei = EXT4_I(inode);
569 struct ext4_ext_path *path = NULL;
570 struct buffer_head *bh;
571 int i = 0, depth, ret = 0;
572
573 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
574 return 0; /* not an extent-mapped inode */
575
576 down_read(&ei->i_data_sem);
577 depth = ext_depth(inode);
578
579 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
580 GFP_NOFS);
581 if (path == NULL) {
582 up_read(&ei->i_data_sem);
583 return -ENOMEM;
584 }
585
586 /* Don't cache anything if there are no external extent blocks */
587 if (depth == 0)
588 goto out;
589 path[0].p_hdr = ext_inode_hdr(inode);
590 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
591 if (ret)
592 goto out;
593 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
594 while (i >= 0) {
595 /*
596 * If this is a leaf block or we've reached the end of
597 * the index block, go up
598 */
599 if ((i == depth) ||
600 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
601 brelse(path[i].p_bh);
602 path[i].p_bh = NULL;
603 i--;
604 continue;
605 }
606 bh = read_extent_tree_block(inode,
607 ext4_idx_pblock(path[i].p_idx++),
608 depth - i - 1,
609 EXT4_EX_FORCE_CACHE);
610 if (IS_ERR(bh)) {
611 ret = PTR_ERR(bh);
612 break;
613 }
614 i++;
615 path[i].p_bh = bh;
616 path[i].p_hdr = ext_block_hdr(bh);
617 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
618 }
619 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
620out:
621 up_read(&ei->i_data_sem);
622 ext4_ext_drop_refs(path);
623 kfree(path);
624 return ret;
625}
626
627#ifdef EXT_DEBUG
628static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
629{
630 int k, l = path->p_depth;
631
632 ext_debug("path:");
633 for (k = 0; k <= l; k++, path++) {
634 if (path->p_idx) {
635 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
636 ext4_idx_pblock(path->p_idx));
637 } else if (path->p_ext) {
638 ext_debug(" %d:[%d]%d:%llu ",
639 le32_to_cpu(path->p_ext->ee_block),
640 ext4_ext_is_unwritten(path->p_ext),
641 ext4_ext_get_actual_len(path->p_ext),
642 ext4_ext_pblock(path->p_ext));
643 } else
644 ext_debug(" []");
645 }
646 ext_debug("\n");
647}
648
649static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
650{
651 int depth = ext_depth(inode);
652 struct ext4_extent_header *eh;
653 struct ext4_extent *ex;
654 int i;
655
656 if (!path)
657 return;
658
659 eh = path[depth].p_hdr;
660 ex = EXT_FIRST_EXTENT(eh);
661
662 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
663
664 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
665 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
666 ext4_ext_is_unwritten(ex),
667 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
668 }
669 ext_debug("\n");
670}
671
672static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
673 ext4_fsblk_t newblock, int level)
674{
675 int depth = ext_depth(inode);
676 struct ext4_extent *ex;
677
678 if (depth != level) {
679 struct ext4_extent_idx *idx;
680 idx = path[level].p_idx;
681 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
682 ext_debug("%d: move %d:%llu in new index %llu\n", level,
683 le32_to_cpu(idx->ei_block),
684 ext4_idx_pblock(idx),
685 newblock);
686 idx++;
687 }
688
689 return;
690 }
691
692 ex = path[depth].p_ext;
693 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
694 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
695 le32_to_cpu(ex->ee_block),
696 ext4_ext_pblock(ex),
697 ext4_ext_is_unwritten(ex),
698 ext4_ext_get_actual_len(ex),
699 newblock);
700 ex++;
701 }
702}
703
704#else
705#define ext4_ext_show_path(inode, path)
706#define ext4_ext_show_leaf(inode, path)
707#define ext4_ext_show_move(inode, path, newblock, level)
708#endif
709
710void ext4_ext_drop_refs(struct ext4_ext_path *path)
711{
712 int depth, i;
713
714 if (!path)
715 return;
716 depth = path->p_depth;
717 for (i = 0; i <= depth; i++, path++)
718 if (path->p_bh) {
719 brelse(path->p_bh);
720 path->p_bh = NULL;
721 }
722}
723
724/*
725 * ext4_ext_binsearch_idx:
726 * binary search for the closest index of the given block
727 * the header must be checked before calling this
728 */
729static void
730ext4_ext_binsearch_idx(struct inode *inode,
731 struct ext4_ext_path *path, ext4_lblk_t block)
732{
733 struct ext4_extent_header *eh = path->p_hdr;
734 struct ext4_extent_idx *r, *l, *m;
735
736
737 ext_debug("binsearch for %u(idx): ", block);
738
739 l = EXT_FIRST_INDEX(eh) + 1;
740 r = EXT_LAST_INDEX(eh);
741 while (l <= r) {
742 m = l + (r - l) / 2;
743 if (block < le32_to_cpu(m->ei_block))
744 r = m - 1;
745 else
746 l = m + 1;
747 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
748 m, le32_to_cpu(m->ei_block),
749 r, le32_to_cpu(r->ei_block));
750 }
751
752 path->p_idx = l - 1;
753 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
754 ext4_idx_pblock(path->p_idx));
755
756#ifdef CHECK_BINSEARCH
757 {
758 struct ext4_extent_idx *chix, *ix;
759 int k;
760
761 chix = ix = EXT_FIRST_INDEX(eh);
762 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
763 if (k != 0 &&
764 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
765 printk(KERN_DEBUG "k=%d, ix=0x%p, "
766 "first=0x%p\n", k,
767 ix, EXT_FIRST_INDEX(eh));
768 printk(KERN_DEBUG "%u <= %u\n",
769 le32_to_cpu(ix->ei_block),
770 le32_to_cpu(ix[-1].ei_block));
771 }
772 BUG_ON(k && le32_to_cpu(ix->ei_block)
773 <= le32_to_cpu(ix[-1].ei_block));
774 if (block < le32_to_cpu(ix->ei_block))
775 break;
776 chix = ix;
777 }
778 BUG_ON(chix != path->p_idx);
779 }
780#endif
781
782}
783
784/*
785 * ext4_ext_binsearch:
786 * binary search for closest extent of the given block
787 * the header must be checked before calling this
788 */
789static void
790ext4_ext_binsearch(struct inode *inode,
791 struct ext4_ext_path *path, ext4_lblk_t block)
792{
793 struct ext4_extent_header *eh = path->p_hdr;
794 struct ext4_extent *r, *l, *m;
795
796 if (eh->eh_entries == 0) {
797 /*
798 * this leaf is empty:
799 * we get such a leaf in split/add case
800 */
801 return;
802 }
803
804 ext_debug("binsearch for %u: ", block);
805
806 l = EXT_FIRST_EXTENT(eh) + 1;
807 r = EXT_LAST_EXTENT(eh);
808
809 while (l <= r) {
810 m = l + (r - l) / 2;
811 if (block < le32_to_cpu(m->ee_block))
812 r = m - 1;
813 else
814 l = m + 1;
815 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
816 m, le32_to_cpu(m->ee_block),
817 r, le32_to_cpu(r->ee_block));
818 }
819
820 path->p_ext = l - 1;
821 ext_debug(" -> %d:%llu:[%d]%d ",
822 le32_to_cpu(path->p_ext->ee_block),
823 ext4_ext_pblock(path->p_ext),
824 ext4_ext_is_unwritten(path->p_ext),
825 ext4_ext_get_actual_len(path->p_ext));
826
827#ifdef CHECK_BINSEARCH
828 {
829 struct ext4_extent *chex, *ex;
830 int k;
831
832 chex = ex = EXT_FIRST_EXTENT(eh);
833 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
834 BUG_ON(k && le32_to_cpu(ex->ee_block)
835 <= le32_to_cpu(ex[-1].ee_block));
836 if (block < le32_to_cpu(ex->ee_block))
837 break;
838 chex = ex;
839 }
840 BUG_ON(chex != path->p_ext);
841 }
842#endif
843
844}
845
846int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
847{
848 struct ext4_extent_header *eh;
849
850 eh = ext_inode_hdr(inode);
851 eh->eh_depth = 0;
852 eh->eh_entries = 0;
853 eh->eh_magic = EXT4_EXT_MAGIC;
854 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
855 ext4_mark_inode_dirty(handle, inode);
856 return 0;
857}
858
859struct ext4_ext_path *
860ext4_find_extent(struct inode *inode, ext4_lblk_t block,
861 struct ext4_ext_path **orig_path, int flags)
862{
863 struct ext4_extent_header *eh;
864 struct buffer_head *bh;
865 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
866 short int depth, i, ppos = 0;
867 int ret;
868
869 eh = ext_inode_hdr(inode);
870 depth = ext_depth(inode);
871
872 if (path) {
873 ext4_ext_drop_refs(path);
874 if (depth > path[0].p_maxdepth) {
875 kfree(path);
876 *orig_path = path = NULL;
877 }
878 }
879 if (!path) {
880 /* account possible depth increase */
881 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
882 GFP_NOFS);
883 if (unlikely(!path))
884 return ERR_PTR(-ENOMEM);
885 path[0].p_maxdepth = depth + 1;
886 }
887 path[0].p_hdr = eh;
888 path[0].p_bh = NULL;
889
890 i = depth;
891 /* walk through the tree */
892 while (i) {
893 ext_debug("depth %d: num %d, max %d\n",
894 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
895
896 ext4_ext_binsearch_idx(inode, path + ppos, block);
897 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
898 path[ppos].p_depth = i;
899 path[ppos].p_ext = NULL;
900
901 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
902 flags);
903 if (IS_ERR(bh)) {
904 ret = PTR_ERR(bh);
905 goto err;
906 }
907
908 eh = ext_block_hdr(bh);
909 ppos++;
910 if (unlikely(ppos > depth)) {
911 put_bh(bh);
912 EXT4_ERROR_INODE(inode,
913 "ppos %d > depth %d", ppos, depth);
914 ret = -EFSCORRUPTED;
915 goto err;
916 }
917 path[ppos].p_bh = bh;
918 path[ppos].p_hdr = eh;
919 }
920
921 path[ppos].p_depth = i;
922 path[ppos].p_ext = NULL;
923 path[ppos].p_idx = NULL;
924
925 /* find extent */
926 ext4_ext_binsearch(inode, path + ppos, block);
927 /* if not an empty leaf */
928 if (path[ppos].p_ext)
929 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
930
931 ext4_ext_show_path(inode, path);
932
933 return path;
934
935err:
936 ext4_ext_drop_refs(path);
937 kfree(path);
938 if (orig_path)
939 *orig_path = NULL;
940 return ERR_PTR(ret);
941}
942
943/*
944 * ext4_ext_insert_index:
945 * insert new index [@logical;@ptr] into the block at @curp;
946 * check where to insert: before @curp or after @curp
947 */
948static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
949 struct ext4_ext_path *curp,
950 int logical, ext4_fsblk_t ptr)
951{
952 struct ext4_extent_idx *ix;
953 int len, err;
954
955 err = ext4_ext_get_access(handle, inode, curp);
956 if (err)
957 return err;
958
959 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
960 EXT4_ERROR_INODE(inode,
961 "logical %d == ei_block %d!",
962 logical, le32_to_cpu(curp->p_idx->ei_block));
963 return -EFSCORRUPTED;
964 }
965
966 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
967 >= le16_to_cpu(curp->p_hdr->eh_max))) {
968 EXT4_ERROR_INODE(inode,
969 "eh_entries %d >= eh_max %d!",
970 le16_to_cpu(curp->p_hdr->eh_entries),
971 le16_to_cpu(curp->p_hdr->eh_max));
972 return -EFSCORRUPTED;
973 }
974
975 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
976 /* insert after */
977 ext_debug("insert new index %d after: %llu\n", logical, ptr);
978 ix = curp->p_idx + 1;
979 } else {
980 /* insert before */
981 ext_debug("insert new index %d before: %llu\n", logical, ptr);
982 ix = curp->p_idx;
983 }
984
985 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
986 BUG_ON(len < 0);
987 if (len > 0) {
988 ext_debug("insert new index %d: "
989 "move %d indices from 0x%p to 0x%p\n",
990 logical, len, ix, ix + 1);
991 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
992 }
993
994 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
995 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
996 return -EFSCORRUPTED;
997 }
998
999 ix->ei_block = cpu_to_le32(logical);
1000 ext4_idx_store_pblock(ix, ptr);
1001 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1002
1003 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1004 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1005 return -EFSCORRUPTED;
1006 }
1007
1008 err = ext4_ext_dirty(handle, inode, curp);
1009 ext4_std_error(inode->i_sb, err);
1010
1011 return err;
1012}
1013
1014/*
1015 * ext4_ext_split:
1016 * inserts new subtree into the path, using free index entry
1017 * at depth @at:
1018 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1019 * - makes decision where to split
1020 * - moves remaining extents and index entries (right to the split point)
1021 * into the newly allocated blocks
1022 * - initializes subtree
1023 */
1024static int ext4_ext_split(handle_t *handle, struct inode *inode,
1025 unsigned int flags,
1026 struct ext4_ext_path *path,
1027 struct ext4_extent *newext, int at)
1028{
1029 struct buffer_head *bh = NULL;
1030 int depth = ext_depth(inode);
1031 struct ext4_extent_header *neh;
1032 struct ext4_extent_idx *fidx;
1033 int i = at, k, m, a;
1034 ext4_fsblk_t newblock, oldblock;
1035 __le32 border;
1036 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1037 int err = 0;
1038
1039 /* make decision: where to split? */
1040 /* FIXME: now decision is simplest: at current extent */
1041
1042 /* if current leaf will be split, then we should use
1043 * border from split point */
1044 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1045 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1046 return -EFSCORRUPTED;
1047 }
1048 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1049 border = path[depth].p_ext[1].ee_block;
1050 ext_debug("leaf will be split."
1051 " next leaf starts at %d\n",
1052 le32_to_cpu(border));
1053 } else {
1054 border = newext->ee_block;
1055 ext_debug("leaf will be added."
1056 " next leaf starts at %d\n",
1057 le32_to_cpu(border));
1058 }
1059
1060 /*
1061 * If error occurs, then we break processing
1062 * and mark filesystem read-only. index won't
1063 * be inserted and tree will be in consistent
1064 * state. Next mount will repair buffers too.
1065 */
1066
1067 /*
1068 * Get array to track all allocated blocks.
1069 * We need this to handle errors and free blocks
1070 * upon them.
1071 */
1072 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1073 if (!ablocks)
1074 return -ENOMEM;
1075
1076 /* allocate all needed blocks */
1077 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1078 for (a = 0; a < depth - at; a++) {
1079 newblock = ext4_ext_new_meta_block(handle, inode, path,
1080 newext, &err, flags);
1081 if (newblock == 0)
1082 goto cleanup;
1083 ablocks[a] = newblock;
1084 }
1085
1086 /* initialize new leaf */
1087 newblock = ablocks[--a];
1088 if (unlikely(newblock == 0)) {
1089 EXT4_ERROR_INODE(inode, "newblock == 0!");
1090 err = -EFSCORRUPTED;
1091 goto cleanup;
1092 }
1093 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1094 if (unlikely(!bh)) {
1095 err = -ENOMEM;
1096 goto cleanup;
1097 }
1098 lock_buffer(bh);
1099
1100 err = ext4_journal_get_create_access(handle, bh);
1101 if (err)
1102 goto cleanup;
1103
1104 neh = ext_block_hdr(bh);
1105 neh->eh_entries = 0;
1106 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1107 neh->eh_magic = EXT4_EXT_MAGIC;
1108 neh->eh_depth = 0;
1109
1110 /* move remainder of path[depth] to the new leaf */
1111 if (unlikely(path[depth].p_hdr->eh_entries !=
1112 path[depth].p_hdr->eh_max)) {
1113 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1114 path[depth].p_hdr->eh_entries,
1115 path[depth].p_hdr->eh_max);
1116 err = -EFSCORRUPTED;
1117 goto cleanup;
1118 }
1119 /* start copy from next extent */
1120 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1121 ext4_ext_show_move(inode, path, newblock, depth);
1122 if (m) {
1123 struct ext4_extent *ex;
1124 ex = EXT_FIRST_EXTENT(neh);
1125 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1126 le16_add_cpu(&neh->eh_entries, m);
1127 }
1128
1129 ext4_extent_block_csum_set(inode, neh);
1130 set_buffer_uptodate(bh);
1131 unlock_buffer(bh);
1132
1133 err = ext4_handle_dirty_metadata(handle, inode, bh);
1134 if (err)
1135 goto cleanup;
1136 brelse(bh);
1137 bh = NULL;
1138
1139 /* correct old leaf */
1140 if (m) {
1141 err = ext4_ext_get_access(handle, inode, path + depth);
1142 if (err)
1143 goto cleanup;
1144 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1145 err = ext4_ext_dirty(handle, inode, path + depth);
1146 if (err)
1147 goto cleanup;
1148
1149 }
1150
1151 /* create intermediate indexes */
1152 k = depth - at - 1;
1153 if (unlikely(k < 0)) {
1154 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1155 err = -EFSCORRUPTED;
1156 goto cleanup;
1157 }
1158 if (k)
1159 ext_debug("create %d intermediate indices\n", k);
1160 /* insert new index into current index block */
1161 /* current depth stored in i var */
1162 i = depth - 1;
1163 while (k--) {
1164 oldblock = newblock;
1165 newblock = ablocks[--a];
1166 bh = sb_getblk(inode->i_sb, newblock);
1167 if (unlikely(!bh)) {
1168 err = -ENOMEM;
1169 goto cleanup;
1170 }
1171 lock_buffer(bh);
1172
1173 err = ext4_journal_get_create_access(handle, bh);
1174 if (err)
1175 goto cleanup;
1176
1177 neh = ext_block_hdr(bh);
1178 neh->eh_entries = cpu_to_le16(1);
1179 neh->eh_magic = EXT4_EXT_MAGIC;
1180 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1181 neh->eh_depth = cpu_to_le16(depth - i);
1182 fidx = EXT_FIRST_INDEX(neh);
1183 fidx->ei_block = border;
1184 ext4_idx_store_pblock(fidx, oldblock);
1185
1186 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1187 i, newblock, le32_to_cpu(border), oldblock);
1188
1189 /* move remainder of path[i] to the new index block */
1190 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1191 EXT_LAST_INDEX(path[i].p_hdr))) {
1192 EXT4_ERROR_INODE(inode,
1193 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1194 le32_to_cpu(path[i].p_ext->ee_block));
1195 err = -EFSCORRUPTED;
1196 goto cleanup;
1197 }
1198 /* start copy indexes */
1199 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1200 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1201 EXT_MAX_INDEX(path[i].p_hdr));
1202 ext4_ext_show_move(inode, path, newblock, i);
1203 if (m) {
1204 memmove(++fidx, path[i].p_idx,
1205 sizeof(struct ext4_extent_idx) * m);
1206 le16_add_cpu(&neh->eh_entries, m);
1207 }
1208 ext4_extent_block_csum_set(inode, neh);
1209 set_buffer_uptodate(bh);
1210 unlock_buffer(bh);
1211
1212 err = ext4_handle_dirty_metadata(handle, inode, bh);
1213 if (err)
1214 goto cleanup;
1215 brelse(bh);
1216 bh = NULL;
1217
1218 /* correct old index */
1219 if (m) {
1220 err = ext4_ext_get_access(handle, inode, path + i);
1221 if (err)
1222 goto cleanup;
1223 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1224 err = ext4_ext_dirty(handle, inode, path + i);
1225 if (err)
1226 goto cleanup;
1227 }
1228
1229 i--;
1230 }
1231
1232 /* insert new index */
1233 err = ext4_ext_insert_index(handle, inode, path + at,
1234 le32_to_cpu(border), newblock);
1235
1236cleanup:
1237 if (bh) {
1238 if (buffer_locked(bh))
1239 unlock_buffer(bh);
1240 brelse(bh);
1241 }
1242
1243 if (err) {
1244 /* free all allocated blocks in error case */
1245 for (i = 0; i < depth; i++) {
1246 if (!ablocks[i])
1247 continue;
1248 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1249 EXT4_FREE_BLOCKS_METADATA);
1250 }
1251 }
1252 kfree(ablocks);
1253
1254 return err;
1255}
1256
1257/*
1258 * ext4_ext_grow_indepth:
1259 * implements tree growing procedure:
1260 * - allocates new block
1261 * - moves top-level data (index block or leaf) into the new block
1262 * - initializes new top-level, creating index that points to the
1263 * just created block
1264 */
1265static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1266 unsigned int flags)
1267{
1268 struct ext4_extent_header *neh;
1269 struct buffer_head *bh;
1270 ext4_fsblk_t newblock, goal = 0;
1271 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1272 int err = 0;
1273
1274 /* Try to prepend new index to old one */
1275 if (ext_depth(inode))
1276 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1277 if (goal > le32_to_cpu(es->s_first_data_block)) {
1278 flags |= EXT4_MB_HINT_TRY_GOAL;
1279 goal--;
1280 } else
1281 goal = ext4_inode_to_goal_block(inode);
1282 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1283 NULL, &err);
1284 if (newblock == 0)
1285 return err;
1286
1287 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1288 if (unlikely(!bh))
1289 return -ENOMEM;
1290 lock_buffer(bh);
1291
1292 err = ext4_journal_get_create_access(handle, bh);
1293 if (err) {
1294 unlock_buffer(bh);
1295 goto out;
1296 }
1297
1298 /* move top-level index/leaf into new block */
1299 memmove(bh->b_data, EXT4_I(inode)->i_data,
1300 sizeof(EXT4_I(inode)->i_data));
1301
1302 /* set size of new block */
1303 neh = ext_block_hdr(bh);
1304 /* old root could have indexes or leaves
1305 * so calculate e_max right way */
1306 if (ext_depth(inode))
1307 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1308 else
1309 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1310 neh->eh_magic = EXT4_EXT_MAGIC;
1311 ext4_extent_block_csum_set(inode, neh);
1312 set_buffer_uptodate(bh);
1313 unlock_buffer(bh);
1314
1315 err = ext4_handle_dirty_metadata(handle, inode, bh);
1316 if (err)
1317 goto out;
1318
1319 /* Update top-level index: num,max,pointer */
1320 neh = ext_inode_hdr(inode);
1321 neh->eh_entries = cpu_to_le16(1);
1322 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1323 if (neh->eh_depth == 0) {
1324 /* Root extent block becomes index block */
1325 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1326 EXT_FIRST_INDEX(neh)->ei_block =
1327 EXT_FIRST_EXTENT(neh)->ee_block;
1328 }
1329 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1330 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1331 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1332 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1333
1334 le16_add_cpu(&neh->eh_depth, 1);
1335 ext4_mark_inode_dirty(handle, inode);
1336out:
1337 brelse(bh);
1338
1339 return err;
1340}
1341
1342/*
1343 * ext4_ext_create_new_leaf:
1344 * finds empty index and adds new leaf.
1345 * if no free index is found, then it requests in-depth growing.
1346 */
1347static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1348 unsigned int mb_flags,
1349 unsigned int gb_flags,
1350 struct ext4_ext_path **ppath,
1351 struct ext4_extent *newext)
1352{
1353 struct ext4_ext_path *path = *ppath;
1354 struct ext4_ext_path *curp;
1355 int depth, i, err = 0;
1356
1357repeat:
1358 i = depth = ext_depth(inode);
1359
1360 /* walk up to the tree and look for free index entry */
1361 curp = path + depth;
1362 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1363 i--;
1364 curp--;
1365 }
1366
1367 /* we use already allocated block for index block,
1368 * so subsequent data blocks should be contiguous */
1369 if (EXT_HAS_FREE_INDEX(curp)) {
1370 /* if we found index with free entry, then use that
1371 * entry: create all needed subtree and add new leaf */
1372 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1373 if (err)
1374 goto out;
1375
1376 /* refill path */
1377 path = ext4_find_extent(inode,
1378 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1379 ppath, gb_flags);
1380 if (IS_ERR(path))
1381 err = PTR_ERR(path);
1382 } else {
1383 /* tree is full, time to grow in depth */
1384 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1385 if (err)
1386 goto out;
1387
1388 /* refill path */
1389 path = ext4_find_extent(inode,
1390 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1391 ppath, gb_flags);
1392 if (IS_ERR(path)) {
1393 err = PTR_ERR(path);
1394 goto out;
1395 }
1396
1397 /*
1398 * only first (depth 0 -> 1) produces free space;
1399 * in all other cases we have to split the grown tree
1400 */
1401 depth = ext_depth(inode);
1402 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1403 /* now we need to split */
1404 goto repeat;
1405 }
1406 }
1407
1408out:
1409 return err;
1410}
1411
1412/*
1413 * search the closest allocated block to the left for *logical
1414 * and returns it at @logical + it's physical address at @phys
1415 * if *logical is the smallest allocated block, the function
1416 * returns 0 at @phys
1417 * return value contains 0 (success) or error code
1418 */
1419static int ext4_ext_search_left(struct inode *inode,
1420 struct ext4_ext_path *path,
1421 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1422{
1423 struct ext4_extent_idx *ix;
1424 struct ext4_extent *ex;
1425 int depth, ee_len;
1426
1427 if (unlikely(path == NULL)) {
1428 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1429 return -EFSCORRUPTED;
1430 }
1431 depth = path->p_depth;
1432 *phys = 0;
1433
1434 if (depth == 0 && path->p_ext == NULL)
1435 return 0;
1436
1437 /* usually extent in the path covers blocks smaller
1438 * then *logical, but it can be that extent is the
1439 * first one in the file */
1440
1441 ex = path[depth].p_ext;
1442 ee_len = ext4_ext_get_actual_len(ex);
1443 if (*logical < le32_to_cpu(ex->ee_block)) {
1444 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1445 EXT4_ERROR_INODE(inode,
1446 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1447 *logical, le32_to_cpu(ex->ee_block));
1448 return -EFSCORRUPTED;
1449 }
1450 while (--depth >= 0) {
1451 ix = path[depth].p_idx;
1452 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1453 EXT4_ERROR_INODE(inode,
1454 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1455 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1456 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1457 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1458 depth);
1459 return -EFSCORRUPTED;
1460 }
1461 }
1462 return 0;
1463 }
1464
1465 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1466 EXT4_ERROR_INODE(inode,
1467 "logical %d < ee_block %d + ee_len %d!",
1468 *logical, le32_to_cpu(ex->ee_block), ee_len);
1469 return -EFSCORRUPTED;
1470 }
1471
1472 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1473 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1474 return 0;
1475}
1476
1477/*
1478 * search the closest allocated block to the right for *logical
1479 * and returns it at @logical + it's physical address at @phys
1480 * if *logical is the largest allocated block, the function
1481 * returns 0 at @phys
1482 * return value contains 0 (success) or error code
1483 */
1484static int ext4_ext_search_right(struct inode *inode,
1485 struct ext4_ext_path *path,
1486 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1487 struct ext4_extent **ret_ex)
1488{
1489 struct buffer_head *bh = NULL;
1490 struct ext4_extent_header *eh;
1491 struct ext4_extent_idx *ix;
1492 struct ext4_extent *ex;
1493 ext4_fsblk_t block;
1494 int depth; /* Note, NOT eh_depth; depth from top of tree */
1495 int ee_len;
1496
1497 if (unlikely(path == NULL)) {
1498 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1499 return -EFSCORRUPTED;
1500 }
1501 depth = path->p_depth;
1502 *phys = 0;
1503
1504 if (depth == 0 && path->p_ext == NULL)
1505 return 0;
1506
1507 /* usually extent in the path covers blocks smaller
1508 * then *logical, but it can be that extent is the
1509 * first one in the file */
1510
1511 ex = path[depth].p_ext;
1512 ee_len = ext4_ext_get_actual_len(ex);
1513 if (*logical < le32_to_cpu(ex->ee_block)) {
1514 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1515 EXT4_ERROR_INODE(inode,
1516 "first_extent(path[%d].p_hdr) != ex",
1517 depth);
1518 return -EFSCORRUPTED;
1519 }
1520 while (--depth >= 0) {
1521 ix = path[depth].p_idx;
1522 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1523 EXT4_ERROR_INODE(inode,
1524 "ix != EXT_FIRST_INDEX *logical %d!",
1525 *logical);
1526 return -EFSCORRUPTED;
1527 }
1528 }
1529 goto found_extent;
1530 }
1531
1532 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1533 EXT4_ERROR_INODE(inode,
1534 "logical %d < ee_block %d + ee_len %d!",
1535 *logical, le32_to_cpu(ex->ee_block), ee_len);
1536 return -EFSCORRUPTED;
1537 }
1538
1539 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1540 /* next allocated block in this leaf */
1541 ex++;
1542 goto found_extent;
1543 }
1544
1545 /* go up and search for index to the right */
1546 while (--depth >= 0) {
1547 ix = path[depth].p_idx;
1548 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1549 goto got_index;
1550 }
1551
1552 /* we've gone up to the root and found no index to the right */
1553 return 0;
1554
1555got_index:
1556 /* we've found index to the right, let's
1557 * follow it and find the closest allocated
1558 * block to the right */
1559 ix++;
1560 block = ext4_idx_pblock(ix);
1561 while (++depth < path->p_depth) {
1562 /* subtract from p_depth to get proper eh_depth */
1563 bh = read_extent_tree_block(inode, block,
1564 path->p_depth - depth, 0);
1565 if (IS_ERR(bh))
1566 return PTR_ERR(bh);
1567 eh = ext_block_hdr(bh);
1568 ix = EXT_FIRST_INDEX(eh);
1569 block = ext4_idx_pblock(ix);
1570 put_bh(bh);
1571 }
1572
1573 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1574 if (IS_ERR(bh))
1575 return PTR_ERR(bh);
1576 eh = ext_block_hdr(bh);
1577 ex = EXT_FIRST_EXTENT(eh);
1578found_extent:
1579 *logical = le32_to_cpu(ex->ee_block);
1580 *phys = ext4_ext_pblock(ex);
1581 *ret_ex = ex;
1582 if (bh)
1583 put_bh(bh);
1584 return 0;
1585}
1586
1587/*
1588 * ext4_ext_next_allocated_block:
1589 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1590 * NOTE: it considers block number from index entry as
1591 * allocated block. Thus, index entries have to be consistent
1592 * with leaves.
1593 */
1594ext4_lblk_t
1595ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1596{
1597 int depth;
1598
1599 BUG_ON(path == NULL);
1600 depth = path->p_depth;
1601
1602 if (depth == 0 && path->p_ext == NULL)
1603 return EXT_MAX_BLOCKS;
1604
1605 while (depth >= 0) {
1606 if (depth == path->p_depth) {
1607 /* leaf */
1608 if (path[depth].p_ext &&
1609 path[depth].p_ext !=
1610 EXT_LAST_EXTENT(path[depth].p_hdr))
1611 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1612 } else {
1613 /* index */
1614 if (path[depth].p_idx !=
1615 EXT_LAST_INDEX(path[depth].p_hdr))
1616 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1617 }
1618 depth--;
1619 }
1620
1621 return EXT_MAX_BLOCKS;
1622}
1623
1624/*
1625 * ext4_ext_next_leaf_block:
1626 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1627 */
1628static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1629{
1630 int depth;
1631
1632 BUG_ON(path == NULL);
1633 depth = path->p_depth;
1634
1635 /* zero-tree has no leaf blocks at all */
1636 if (depth == 0)
1637 return EXT_MAX_BLOCKS;
1638
1639 /* go to index block */
1640 depth--;
1641
1642 while (depth >= 0) {
1643 if (path[depth].p_idx !=
1644 EXT_LAST_INDEX(path[depth].p_hdr))
1645 return (ext4_lblk_t)
1646 le32_to_cpu(path[depth].p_idx[1].ei_block);
1647 depth--;
1648 }
1649
1650 return EXT_MAX_BLOCKS;
1651}
1652
1653/*
1654 * ext4_ext_correct_indexes:
1655 * if leaf gets modified and modified extent is first in the leaf,
1656 * then we have to correct all indexes above.
1657 * TODO: do we need to correct tree in all cases?
1658 */
1659static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1660 struct ext4_ext_path *path)
1661{
1662 struct ext4_extent_header *eh;
1663 int depth = ext_depth(inode);
1664 struct ext4_extent *ex;
1665 __le32 border;
1666 int k, err = 0;
1667
1668 eh = path[depth].p_hdr;
1669 ex = path[depth].p_ext;
1670
1671 if (unlikely(ex == NULL || eh == NULL)) {
1672 EXT4_ERROR_INODE(inode,
1673 "ex %p == NULL or eh %p == NULL", ex, eh);
1674 return -EFSCORRUPTED;
1675 }
1676
1677 if (depth == 0) {
1678 /* there is no tree at all */
1679 return 0;
1680 }
1681
1682 if (ex != EXT_FIRST_EXTENT(eh)) {
1683 /* we correct tree if first leaf got modified only */
1684 return 0;
1685 }
1686
1687 /*
1688 * TODO: we need correction if border is smaller than current one
1689 */
1690 k = depth - 1;
1691 border = path[depth].p_ext->ee_block;
1692 err = ext4_ext_get_access(handle, inode, path + k);
1693 if (err)
1694 return err;
1695 path[k].p_idx->ei_block = border;
1696 err = ext4_ext_dirty(handle, inode, path + k);
1697 if (err)
1698 return err;
1699
1700 while (k--) {
1701 /* change all left-side indexes */
1702 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1703 break;
1704 err = ext4_ext_get_access(handle, inode, path + k);
1705 if (err)
1706 break;
1707 path[k].p_idx->ei_block = border;
1708 err = ext4_ext_dirty(handle, inode, path + k);
1709 if (err)
1710 break;
1711 }
1712
1713 return err;
1714}
1715
1716int
1717ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1718 struct ext4_extent *ex2)
1719{
1720 unsigned short ext1_ee_len, ext2_ee_len;
1721
1722 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1723 return 0;
1724
1725 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1726 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1727
1728 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1729 le32_to_cpu(ex2->ee_block))
1730 return 0;
1731
1732 /*
1733 * To allow future support for preallocated extents to be added
1734 * as an RO_COMPAT feature, refuse to merge to extents if
1735 * this can result in the top bit of ee_len being set.
1736 */
1737 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1738 return 0;
1739 /*
1740 * The check for IO to unwritten extent is somewhat racy as we
1741 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1742 * dropping i_data_sem. But reserved blocks should save us in that
1743 * case.
1744 */
1745 if (ext4_ext_is_unwritten(ex1) &&
1746 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1747 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1748 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1749 return 0;
1750#ifdef AGGRESSIVE_TEST
1751 if (ext1_ee_len >= 4)
1752 return 0;
1753#endif
1754
1755 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1756 return 1;
1757 return 0;
1758}
1759
1760/*
1761 * This function tries to merge the "ex" extent to the next extent in the tree.
1762 * It always tries to merge towards right. If you want to merge towards
1763 * left, pass "ex - 1" as argument instead of "ex".
1764 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1765 * 1 if they got merged.
1766 */
1767static int ext4_ext_try_to_merge_right(struct inode *inode,
1768 struct ext4_ext_path *path,
1769 struct ext4_extent *ex)
1770{
1771 struct ext4_extent_header *eh;
1772 unsigned int depth, len;
1773 int merge_done = 0, unwritten;
1774
1775 depth = ext_depth(inode);
1776 BUG_ON(path[depth].p_hdr == NULL);
1777 eh = path[depth].p_hdr;
1778
1779 while (ex < EXT_LAST_EXTENT(eh)) {
1780 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1781 break;
1782 /* merge with next extent! */
1783 unwritten = ext4_ext_is_unwritten(ex);
1784 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1785 + ext4_ext_get_actual_len(ex + 1));
1786 if (unwritten)
1787 ext4_ext_mark_unwritten(ex);
1788
1789 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1790 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1791 * sizeof(struct ext4_extent);
1792 memmove(ex + 1, ex + 2, len);
1793 }
1794 le16_add_cpu(&eh->eh_entries, -1);
1795 merge_done = 1;
1796 WARN_ON(eh->eh_entries == 0);
1797 if (!eh->eh_entries)
1798 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1799 }
1800
1801 return merge_done;
1802}
1803
1804/*
1805 * This function does a very simple check to see if we can collapse
1806 * an extent tree with a single extent tree leaf block into the inode.
1807 */
1808static void ext4_ext_try_to_merge_up(handle_t *handle,
1809 struct inode *inode,
1810 struct ext4_ext_path *path)
1811{
1812 size_t s;
1813 unsigned max_root = ext4_ext_space_root(inode, 0);
1814 ext4_fsblk_t blk;
1815
1816 if ((path[0].p_depth != 1) ||
1817 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1818 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1819 return;
1820
1821 /*
1822 * We need to modify the block allocation bitmap and the block
1823 * group descriptor to release the extent tree block. If we
1824 * can't get the journal credits, give up.
1825 */
1826 if (ext4_journal_extend(handle, 2))
1827 return;
1828
1829 /*
1830 * Copy the extent data up to the inode
1831 */
1832 blk = ext4_idx_pblock(path[0].p_idx);
1833 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1834 sizeof(struct ext4_extent_idx);
1835 s += sizeof(struct ext4_extent_header);
1836
1837 path[1].p_maxdepth = path[0].p_maxdepth;
1838 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1839 path[0].p_depth = 0;
1840 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1841 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1842 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1843
1844 brelse(path[1].p_bh);
1845 ext4_free_blocks(handle, inode, NULL, blk, 1,
1846 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1847}
1848
1849/*
1850 * This function tries to merge the @ex extent to neighbours in the tree.
1851 * return 1 if merge left else 0.
1852 */
1853static void ext4_ext_try_to_merge(handle_t *handle,
1854 struct inode *inode,
1855 struct ext4_ext_path *path,
1856 struct ext4_extent *ex) {
1857 struct ext4_extent_header *eh;
1858 unsigned int depth;
1859 int merge_done = 0;
1860
1861 depth = ext_depth(inode);
1862 BUG_ON(path[depth].p_hdr == NULL);
1863 eh = path[depth].p_hdr;
1864
1865 if (ex > EXT_FIRST_EXTENT(eh))
1866 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1867
1868 if (!merge_done)
1869 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1870
1871 ext4_ext_try_to_merge_up(handle, inode, path);
1872}
1873
1874/*
1875 * check if a portion of the "newext" extent overlaps with an
1876 * existing extent.
1877 *
1878 * If there is an overlap discovered, it updates the length of the newext
1879 * such that there will be no overlap, and then returns 1.
1880 * If there is no overlap found, it returns 0.
1881 */
1882static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1883 struct inode *inode,
1884 struct ext4_extent *newext,
1885 struct ext4_ext_path *path)
1886{
1887 ext4_lblk_t b1, b2;
1888 unsigned int depth, len1;
1889 unsigned int ret = 0;
1890
1891 b1 = le32_to_cpu(newext->ee_block);
1892 len1 = ext4_ext_get_actual_len(newext);
1893 depth = ext_depth(inode);
1894 if (!path[depth].p_ext)
1895 goto out;
1896 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1897
1898 /*
1899 * get the next allocated block if the extent in the path
1900 * is before the requested block(s)
1901 */
1902 if (b2 < b1) {
1903 b2 = ext4_ext_next_allocated_block(path);
1904 if (b2 == EXT_MAX_BLOCKS)
1905 goto out;
1906 b2 = EXT4_LBLK_CMASK(sbi, b2);
1907 }
1908
1909 /* check for wrap through zero on extent logical start block*/
1910 if (b1 + len1 < b1) {
1911 len1 = EXT_MAX_BLOCKS - b1;
1912 newext->ee_len = cpu_to_le16(len1);
1913 ret = 1;
1914 }
1915
1916 /* check for overlap */
1917 if (b1 + len1 > b2) {
1918 newext->ee_len = cpu_to_le16(b2 - b1);
1919 ret = 1;
1920 }
1921out:
1922 return ret;
1923}
1924
1925/*
1926 * ext4_ext_insert_extent:
1927 * tries to merge requsted extent into the existing extent or
1928 * inserts requested extent as new one into the tree,
1929 * creating new leaf in the no-space case.
1930 */
1931int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1932 struct ext4_ext_path **ppath,
1933 struct ext4_extent *newext, int gb_flags)
1934{
1935 struct ext4_ext_path *path = *ppath;
1936 struct ext4_extent_header *eh;
1937 struct ext4_extent *ex, *fex;
1938 struct ext4_extent *nearex; /* nearest extent */
1939 struct ext4_ext_path *npath = NULL;
1940 int depth, len, err;
1941 ext4_lblk_t next;
1942 int mb_flags = 0, unwritten;
1943
1944 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1945 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1946 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1947 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1948 return -EFSCORRUPTED;
1949 }
1950 depth = ext_depth(inode);
1951 ex = path[depth].p_ext;
1952 eh = path[depth].p_hdr;
1953 if (unlikely(path[depth].p_hdr == NULL)) {
1954 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1955 return -EFSCORRUPTED;
1956 }
1957
1958 /* try to insert block into found extent and return */
1959 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1960
1961 /*
1962 * Try to see whether we should rather test the extent on
1963 * right from ex, or from the left of ex. This is because
1964 * ext4_find_extent() can return either extent on the
1965 * left, or on the right from the searched position. This
1966 * will make merging more effective.
1967 */
1968 if (ex < EXT_LAST_EXTENT(eh) &&
1969 (le32_to_cpu(ex->ee_block) +
1970 ext4_ext_get_actual_len(ex) <
1971 le32_to_cpu(newext->ee_block))) {
1972 ex += 1;
1973 goto prepend;
1974 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1975 (le32_to_cpu(newext->ee_block) +
1976 ext4_ext_get_actual_len(newext) <
1977 le32_to_cpu(ex->ee_block)))
1978 ex -= 1;
1979
1980 /* Try to append newex to the ex */
1981 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1982 ext_debug("append [%d]%d block to %u:[%d]%d"
1983 "(from %llu)\n",
1984 ext4_ext_is_unwritten(newext),
1985 ext4_ext_get_actual_len(newext),
1986 le32_to_cpu(ex->ee_block),
1987 ext4_ext_is_unwritten(ex),
1988 ext4_ext_get_actual_len(ex),
1989 ext4_ext_pblock(ex));
1990 err = ext4_ext_get_access(handle, inode,
1991 path + depth);
1992 if (err)
1993 return err;
1994 unwritten = ext4_ext_is_unwritten(ex);
1995 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1996 + ext4_ext_get_actual_len(newext));
1997 if (unwritten)
1998 ext4_ext_mark_unwritten(ex);
1999 eh = path[depth].p_hdr;
2000 nearex = ex;
2001 goto merge;
2002 }
2003
2004prepend:
2005 /* Try to prepend newex to the ex */
2006 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2007 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2008 "(from %llu)\n",
2009 le32_to_cpu(newext->ee_block),
2010 ext4_ext_is_unwritten(newext),
2011 ext4_ext_get_actual_len(newext),
2012 le32_to_cpu(ex->ee_block),
2013 ext4_ext_is_unwritten(ex),
2014 ext4_ext_get_actual_len(ex),
2015 ext4_ext_pblock(ex));
2016 err = ext4_ext_get_access(handle, inode,
2017 path + depth);
2018 if (err)
2019 return err;
2020
2021 unwritten = ext4_ext_is_unwritten(ex);
2022 ex->ee_block = newext->ee_block;
2023 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2024 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2025 + ext4_ext_get_actual_len(newext));
2026 if (unwritten)
2027 ext4_ext_mark_unwritten(ex);
2028 eh = path[depth].p_hdr;
2029 nearex = ex;
2030 goto merge;
2031 }
2032 }
2033
2034 depth = ext_depth(inode);
2035 eh = path[depth].p_hdr;
2036 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2037 goto has_space;
2038
2039 /* probably next leaf has space for us? */
2040 fex = EXT_LAST_EXTENT(eh);
2041 next = EXT_MAX_BLOCKS;
2042 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2043 next = ext4_ext_next_leaf_block(path);
2044 if (next != EXT_MAX_BLOCKS) {
2045 ext_debug("next leaf block - %u\n", next);
2046 BUG_ON(npath != NULL);
2047 npath = ext4_find_extent(inode, next, NULL, 0);
2048 if (IS_ERR(npath))
2049 return PTR_ERR(npath);
2050 BUG_ON(npath->p_depth != path->p_depth);
2051 eh = npath[depth].p_hdr;
2052 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2053 ext_debug("next leaf isn't full(%d)\n",
2054 le16_to_cpu(eh->eh_entries));
2055 path = npath;
2056 goto has_space;
2057 }
2058 ext_debug("next leaf has no free space(%d,%d)\n",
2059 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2060 }
2061
2062 /*
2063 * There is no free space in the found leaf.
2064 * We're gonna add a new leaf in the tree.
2065 */
2066 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2067 mb_flags |= EXT4_MB_USE_RESERVED;
2068 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2069 ppath, newext);
2070 if (err)
2071 goto cleanup;
2072 depth = ext_depth(inode);
2073 eh = path[depth].p_hdr;
2074
2075has_space:
2076 nearex = path[depth].p_ext;
2077
2078 err = ext4_ext_get_access(handle, inode, path + depth);
2079 if (err)
2080 goto cleanup;
2081
2082 if (!nearex) {
2083 /* there is no extent in this leaf, create first one */
2084 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2085 le32_to_cpu(newext->ee_block),
2086 ext4_ext_pblock(newext),
2087 ext4_ext_is_unwritten(newext),
2088 ext4_ext_get_actual_len(newext));
2089 nearex = EXT_FIRST_EXTENT(eh);
2090 } else {
2091 if (le32_to_cpu(newext->ee_block)
2092 > le32_to_cpu(nearex->ee_block)) {
2093 /* Insert after */
2094 ext_debug("insert %u:%llu:[%d]%d before: "
2095 "nearest %p\n",
2096 le32_to_cpu(newext->ee_block),
2097 ext4_ext_pblock(newext),
2098 ext4_ext_is_unwritten(newext),
2099 ext4_ext_get_actual_len(newext),
2100 nearex);
2101 nearex++;
2102 } else {
2103 /* Insert before */
2104 BUG_ON(newext->ee_block == nearex->ee_block);
2105 ext_debug("insert %u:%llu:[%d]%d after: "
2106 "nearest %p\n",
2107 le32_to_cpu(newext->ee_block),
2108 ext4_ext_pblock(newext),
2109 ext4_ext_is_unwritten(newext),
2110 ext4_ext_get_actual_len(newext),
2111 nearex);
2112 }
2113 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2114 if (len > 0) {
2115 ext_debug("insert %u:%llu:[%d]%d: "
2116 "move %d extents from 0x%p to 0x%p\n",
2117 le32_to_cpu(newext->ee_block),
2118 ext4_ext_pblock(newext),
2119 ext4_ext_is_unwritten(newext),
2120 ext4_ext_get_actual_len(newext),
2121 len, nearex, nearex + 1);
2122 memmove(nearex + 1, nearex,
2123 len * sizeof(struct ext4_extent));
2124 }
2125 }
2126
2127 le16_add_cpu(&eh->eh_entries, 1);
2128 path[depth].p_ext = nearex;
2129 nearex->ee_block = newext->ee_block;
2130 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2131 nearex->ee_len = newext->ee_len;
2132
2133merge:
2134 /* try to merge extents */
2135 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2136 ext4_ext_try_to_merge(handle, inode, path, nearex);
2137
2138
2139 /* time to correct all indexes above */
2140 err = ext4_ext_correct_indexes(handle, inode, path);
2141 if (err)
2142 goto cleanup;
2143
2144 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2145
2146cleanup:
2147 ext4_ext_drop_refs(npath);
2148 kfree(npath);
2149 return err;
2150}
2151
2152static int ext4_fill_fiemap_extents(struct inode *inode,
2153 ext4_lblk_t block, ext4_lblk_t num,
2154 struct fiemap_extent_info *fieinfo)
2155{
2156 struct ext4_ext_path *path = NULL;
2157 struct ext4_extent *ex;
2158 struct extent_status es;
2159 ext4_lblk_t next, next_del, start = 0, end = 0;
2160 ext4_lblk_t last = block + num;
2161 int exists, depth = 0, err = 0;
2162 unsigned int flags = 0;
2163 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2164
2165 while (block < last && block != EXT_MAX_BLOCKS) {
2166 num = last - block;
2167 /* find extent for this block */
2168 down_read(&EXT4_I(inode)->i_data_sem);
2169
2170 path = ext4_find_extent(inode, block, &path, 0);
2171 if (IS_ERR(path)) {
2172 up_read(&EXT4_I(inode)->i_data_sem);
2173 err = PTR_ERR(path);
2174 path = NULL;
2175 break;
2176 }
2177
2178 depth = ext_depth(inode);
2179 if (unlikely(path[depth].p_hdr == NULL)) {
2180 up_read(&EXT4_I(inode)->i_data_sem);
2181 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2182 err = -EFSCORRUPTED;
2183 break;
2184 }
2185 ex = path[depth].p_ext;
2186 next = ext4_ext_next_allocated_block(path);
2187
2188 flags = 0;
2189 exists = 0;
2190 if (!ex) {
2191 /* there is no extent yet, so try to allocate
2192 * all requested space */
2193 start = block;
2194 end = block + num;
2195 } else if (le32_to_cpu(ex->ee_block) > block) {
2196 /* need to allocate space before found extent */
2197 start = block;
2198 end = le32_to_cpu(ex->ee_block);
2199 if (block + num < end)
2200 end = block + num;
2201 } else if (block >= le32_to_cpu(ex->ee_block)
2202 + ext4_ext_get_actual_len(ex)) {
2203 /* need to allocate space after found extent */
2204 start = block;
2205 end = block + num;
2206 if (end >= next)
2207 end = next;
2208 } else if (block >= le32_to_cpu(ex->ee_block)) {
2209 /*
2210 * some part of requested space is covered
2211 * by found extent
2212 */
2213 start = block;
2214 end = le32_to_cpu(ex->ee_block)
2215 + ext4_ext_get_actual_len(ex);
2216 if (block + num < end)
2217 end = block + num;
2218 exists = 1;
2219 } else {
2220 BUG();
2221 }
2222 BUG_ON(end <= start);
2223
2224 if (!exists) {
2225 es.es_lblk = start;
2226 es.es_len = end - start;
2227 es.es_pblk = 0;
2228 } else {
2229 es.es_lblk = le32_to_cpu(ex->ee_block);
2230 es.es_len = ext4_ext_get_actual_len(ex);
2231 es.es_pblk = ext4_ext_pblock(ex);
2232 if (ext4_ext_is_unwritten(ex))
2233 flags |= FIEMAP_EXTENT_UNWRITTEN;
2234 }
2235
2236 /*
2237 * Find delayed extent and update es accordingly. We call
2238 * it even in !exists case to find out whether es is the
2239 * last existing extent or not.
2240 */
2241 next_del = ext4_find_delayed_extent(inode, &es);
2242 if (!exists && next_del) {
2243 exists = 1;
2244 flags |= (FIEMAP_EXTENT_DELALLOC |
2245 FIEMAP_EXTENT_UNKNOWN);
2246 }
2247 up_read(&EXT4_I(inode)->i_data_sem);
2248
2249 if (unlikely(es.es_len == 0)) {
2250 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2251 err = -EFSCORRUPTED;
2252 break;
2253 }
2254
2255 /*
2256 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2257 * we need to check next == EXT_MAX_BLOCKS because it is
2258 * possible that an extent is with unwritten and delayed
2259 * status due to when an extent is delayed allocated and
2260 * is allocated by fallocate status tree will track both of
2261 * them in a extent.
2262 *
2263 * So we could return a unwritten and delayed extent, and
2264 * its block is equal to 'next'.
2265 */
2266 if (next == next_del && next == EXT_MAX_BLOCKS) {
2267 flags |= FIEMAP_EXTENT_LAST;
2268 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2269 next != EXT_MAX_BLOCKS)) {
2270 EXT4_ERROR_INODE(inode,
2271 "next extent == %u, next "
2272 "delalloc extent = %u",
2273 next, next_del);
2274 err = -EFSCORRUPTED;
2275 break;
2276 }
2277 }
2278
2279 if (exists) {
2280 err = fiemap_fill_next_extent(fieinfo,
2281 (__u64)es.es_lblk << blksize_bits,
2282 (__u64)es.es_pblk << blksize_bits,
2283 (__u64)es.es_len << blksize_bits,
2284 flags);
2285 if (err < 0)
2286 break;
2287 if (err == 1) {
2288 err = 0;
2289 break;
2290 }
2291 }
2292
2293 block = es.es_lblk + es.es_len;
2294 }
2295
2296 ext4_ext_drop_refs(path);
2297 kfree(path);
2298 return err;
2299}
2300
2301/*
2302 * ext4_ext_determine_hole - determine hole around given block
2303 * @inode: inode we lookup in
2304 * @path: path in extent tree to @lblk
2305 * @lblk: pointer to logical block around which we want to determine hole
2306 *
2307 * Determine hole length (and start if easily possible) around given logical
2308 * block. We don't try too hard to find the beginning of the hole but @path
2309 * actually points to extent before @lblk, we provide it.
2310 *
2311 * The function returns the length of a hole starting at @lblk. We update @lblk
2312 * to the beginning of the hole if we managed to find it.
2313 */
2314static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2315 struct ext4_ext_path *path,
2316 ext4_lblk_t *lblk)
2317{
2318 int depth = ext_depth(inode);
2319 struct ext4_extent *ex;
2320 ext4_lblk_t len;
2321
2322 ex = path[depth].p_ext;
2323 if (ex == NULL) {
2324 /* there is no extent yet, so gap is [0;-] */
2325 *lblk = 0;
2326 len = EXT_MAX_BLOCKS;
2327 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2328 len = le32_to_cpu(ex->ee_block) - *lblk;
2329 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2330 + ext4_ext_get_actual_len(ex)) {
2331 ext4_lblk_t next;
2332
2333 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2334 next = ext4_ext_next_allocated_block(path);
2335 BUG_ON(next == *lblk);
2336 len = next - *lblk;
2337 } else {
2338 BUG();
2339 }
2340 return len;
2341}
2342
2343/*
2344 * ext4_ext_put_gap_in_cache:
2345 * calculate boundaries of the gap that the requested block fits into
2346 * and cache this gap
2347 */
2348static void
2349ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2350 ext4_lblk_t hole_len)
2351{
2352 struct extent_status es;
2353
2354 ext4_es_find_delayed_extent_range(inode, hole_start,
2355 hole_start + hole_len - 1, &es);
2356 if (es.es_len) {
2357 /* There's delayed extent containing lblock? */
2358 if (es.es_lblk <= hole_start)
2359 return;
2360 hole_len = min(es.es_lblk - hole_start, hole_len);
2361 }
2362 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2363 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2364 EXTENT_STATUS_HOLE);
2365}
2366
2367/*
2368 * ext4_ext_rm_idx:
2369 * removes index from the index block.
2370 */
2371static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2372 struct ext4_ext_path *path, int depth)
2373{
2374 int err;
2375 ext4_fsblk_t leaf;
2376
2377 /* free index block */
2378 depth--;
2379 path = path + depth;
2380 leaf = ext4_idx_pblock(path->p_idx);
2381 if (unlikely(path->p_hdr->eh_entries == 0)) {
2382 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2383 return -EFSCORRUPTED;
2384 }
2385 err = ext4_ext_get_access(handle, inode, path);
2386 if (err)
2387 return err;
2388
2389 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2390 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2391 len *= sizeof(struct ext4_extent_idx);
2392 memmove(path->p_idx, path->p_idx + 1, len);
2393 }
2394
2395 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2396 err = ext4_ext_dirty(handle, inode, path);
2397 if (err)
2398 return err;
2399 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2400 trace_ext4_ext_rm_idx(inode, leaf);
2401
2402 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2403 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2404
2405 while (--depth >= 0) {
2406 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2407 break;
2408 path--;
2409 err = ext4_ext_get_access(handle, inode, path);
2410 if (err)
2411 break;
2412 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2413 err = ext4_ext_dirty(handle, inode, path);
2414 if (err)
2415 break;
2416 }
2417 return err;
2418}
2419
2420/*
2421 * ext4_ext_calc_credits_for_single_extent:
2422 * This routine returns max. credits that needed to insert an extent
2423 * to the extent tree.
2424 * When pass the actual path, the caller should calculate credits
2425 * under i_data_sem.
2426 */
2427int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2428 struct ext4_ext_path *path)
2429{
2430 if (path) {
2431 int depth = ext_depth(inode);
2432 int ret = 0;
2433
2434 /* probably there is space in leaf? */
2435 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2436 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2437
2438 /*
2439 * There are some space in the leaf tree, no
2440 * need to account for leaf block credit
2441 *
2442 * bitmaps and block group descriptor blocks
2443 * and other metadata blocks still need to be
2444 * accounted.
2445 */
2446 /* 1 bitmap, 1 block group descriptor */
2447 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2448 return ret;
2449 }
2450 }
2451
2452 return ext4_chunk_trans_blocks(inode, nrblocks);
2453}
2454
2455/*
2456 * How many index/leaf blocks need to change/allocate to add @extents extents?
2457 *
2458 * If we add a single extent, then in the worse case, each tree level
2459 * index/leaf need to be changed in case of the tree split.
2460 *
2461 * If more extents are inserted, they could cause the whole tree split more
2462 * than once, but this is really rare.
2463 */
2464int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2465{
2466 int index;
2467 int depth;
2468
2469 /* If we are converting the inline data, only one is needed here. */
2470 if (ext4_has_inline_data(inode))
2471 return 1;
2472
2473 depth = ext_depth(inode);
2474
2475 if (extents <= 1)
2476 index = depth * 2;
2477 else
2478 index = depth * 3;
2479
2480 return index;
2481}
2482
2483static inline int get_default_free_blocks_flags(struct inode *inode)
2484{
2485 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2486 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2487 else if (ext4_should_journal_data(inode))
2488 return EXT4_FREE_BLOCKS_FORGET;
2489 return 0;
2490}
2491
2492static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2493 struct ext4_extent *ex,
2494 long long *partial_cluster,
2495 ext4_lblk_t from, ext4_lblk_t to)
2496{
2497 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2498 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2499 ext4_fsblk_t pblk;
2500 int flags = get_default_free_blocks_flags(inode);
2501
2502 /*
2503 * For bigalloc file systems, we never free a partial cluster
2504 * at the beginning of the extent. Instead, we make a note
2505 * that we tried freeing the cluster, and check to see if we
2506 * need to free it on a subsequent call to ext4_remove_blocks,
2507 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2508 */
2509 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2510
2511 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2512 /*
2513 * If we have a partial cluster, and it's different from the
2514 * cluster of the last block, we need to explicitly free the
2515 * partial cluster here.
2516 */
2517 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2518 if (*partial_cluster > 0 &&
2519 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2520 ext4_free_blocks(handle, inode, NULL,
2521 EXT4_C2B(sbi, *partial_cluster),
2522 sbi->s_cluster_ratio, flags);
2523 *partial_cluster = 0;
2524 }
2525
2526#ifdef EXTENTS_STATS
2527 {
2528 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2529 spin_lock(&sbi->s_ext_stats_lock);
2530 sbi->s_ext_blocks += ee_len;
2531 sbi->s_ext_extents++;
2532 if (ee_len < sbi->s_ext_min)
2533 sbi->s_ext_min = ee_len;
2534 if (ee_len > sbi->s_ext_max)
2535 sbi->s_ext_max = ee_len;
2536 if (ext_depth(inode) > sbi->s_depth_max)
2537 sbi->s_depth_max = ext_depth(inode);
2538 spin_unlock(&sbi->s_ext_stats_lock);
2539 }
2540#endif
2541 if (from >= le32_to_cpu(ex->ee_block)
2542 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2543 /* tail removal */
2544 ext4_lblk_t num;
2545 long long first_cluster;
2546
2547 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2548 pblk = ext4_ext_pblock(ex) + ee_len - num;
2549 /*
2550 * Usually we want to free partial cluster at the end of the
2551 * extent, except for the situation when the cluster is still
2552 * used by any other extent (partial_cluster is negative).
2553 */
2554 if (*partial_cluster < 0 &&
2555 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2556 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2557
2558 ext_debug("free last %u blocks starting %llu partial %lld\n",
2559 num, pblk, *partial_cluster);
2560 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2561 /*
2562 * If the block range to be freed didn't start at the
2563 * beginning of a cluster, and we removed the entire
2564 * extent and the cluster is not used by any other extent,
2565 * save the partial cluster here, since we might need to
2566 * delete if we determine that the truncate or punch hole
2567 * operation has removed all of the blocks in the cluster.
2568 * If that cluster is used by another extent, preserve its
2569 * negative value so it isn't freed later on.
2570 *
2571 * If the whole extent wasn't freed, we've reached the
2572 * start of the truncated/punched region and have finished
2573 * removing blocks. If there's a partial cluster here it's
2574 * shared with the remainder of the extent and is no longer
2575 * a candidate for removal.
2576 */
2577 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2578 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2579 if (first_cluster != -*partial_cluster)
2580 *partial_cluster = first_cluster;
2581 } else {
2582 *partial_cluster = 0;
2583 }
2584 } else
2585 ext4_error(sbi->s_sb, "strange request: removal(2) "
2586 "%u-%u from %u:%u\n",
2587 from, to, le32_to_cpu(ex->ee_block), ee_len);
2588 return 0;
2589}
2590
2591
2592/*
2593 * ext4_ext_rm_leaf() Removes the extents associated with the
2594 * blocks appearing between "start" and "end". Both "start"
2595 * and "end" must appear in the same extent or EIO is returned.
2596 *
2597 * @handle: The journal handle
2598 * @inode: The files inode
2599 * @path: The path to the leaf
2600 * @partial_cluster: The cluster which we'll have to free if all extents
2601 * has been released from it. However, if this value is
2602 * negative, it's a cluster just to the right of the
2603 * punched region and it must not be freed.
2604 * @start: The first block to remove
2605 * @end: The last block to remove
2606 */
2607static int
2608ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2609 struct ext4_ext_path *path,
2610 long long *partial_cluster,
2611 ext4_lblk_t start, ext4_lblk_t end)
2612{
2613 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2614 int err = 0, correct_index = 0;
2615 int depth = ext_depth(inode), credits;
2616 struct ext4_extent_header *eh;
2617 ext4_lblk_t a, b;
2618 unsigned num;
2619 ext4_lblk_t ex_ee_block;
2620 unsigned short ex_ee_len;
2621 unsigned unwritten = 0;
2622 struct ext4_extent *ex;
2623 ext4_fsblk_t pblk;
2624
2625 /* the header must be checked already in ext4_ext_remove_space() */
2626 ext_debug("truncate since %u in leaf to %u\n", start, end);
2627 if (!path[depth].p_hdr)
2628 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2629 eh = path[depth].p_hdr;
2630 if (unlikely(path[depth].p_hdr == NULL)) {
2631 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2632 return -EFSCORRUPTED;
2633 }
2634 /* find where to start removing */
2635 ex = path[depth].p_ext;
2636 if (!ex)
2637 ex = EXT_LAST_EXTENT(eh);
2638
2639 ex_ee_block = le32_to_cpu(ex->ee_block);
2640 ex_ee_len = ext4_ext_get_actual_len(ex);
2641
2642 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2643
2644 while (ex >= EXT_FIRST_EXTENT(eh) &&
2645 ex_ee_block + ex_ee_len > start) {
2646
2647 if (ext4_ext_is_unwritten(ex))
2648 unwritten = 1;
2649 else
2650 unwritten = 0;
2651
2652 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2653 unwritten, ex_ee_len);
2654 path[depth].p_ext = ex;
2655
2656 a = ex_ee_block > start ? ex_ee_block : start;
2657 b = ex_ee_block+ex_ee_len - 1 < end ?
2658 ex_ee_block+ex_ee_len - 1 : end;
2659
2660 ext_debug(" border %u:%u\n", a, b);
2661
2662 /* If this extent is beyond the end of the hole, skip it */
2663 if (end < ex_ee_block) {
2664 /*
2665 * We're going to skip this extent and move to another,
2666 * so note that its first cluster is in use to avoid
2667 * freeing it when removing blocks. Eventually, the
2668 * right edge of the truncated/punched region will
2669 * be just to the left.
2670 */
2671 if (sbi->s_cluster_ratio > 1) {
2672 pblk = ext4_ext_pblock(ex);
2673 *partial_cluster =
2674 -(long long) EXT4_B2C(sbi, pblk);
2675 }
2676 ex--;
2677 ex_ee_block = le32_to_cpu(ex->ee_block);
2678 ex_ee_len = ext4_ext_get_actual_len(ex);
2679 continue;
2680 } else if (b != ex_ee_block + ex_ee_len - 1) {
2681 EXT4_ERROR_INODE(inode,
2682 "can not handle truncate %u:%u "
2683 "on extent %u:%u",
2684 start, end, ex_ee_block,
2685 ex_ee_block + ex_ee_len - 1);
2686 err = -EFSCORRUPTED;
2687 goto out;
2688 } else if (a != ex_ee_block) {
2689 /* remove tail of the extent */
2690 num = a - ex_ee_block;
2691 } else {
2692 /* remove whole extent: excellent! */
2693 num = 0;
2694 }
2695 /*
2696 * 3 for leaf, sb, and inode plus 2 (bmap and group
2697 * descriptor) for each block group; assume two block
2698 * groups plus ex_ee_len/blocks_per_block_group for
2699 * the worst case
2700 */
2701 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2702 if (ex == EXT_FIRST_EXTENT(eh)) {
2703 correct_index = 1;
2704 credits += (ext_depth(inode)) + 1;
2705 }
2706 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2707
2708 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2709 if (err)
2710 goto out;
2711
2712 err = ext4_ext_get_access(handle, inode, path + depth);
2713 if (err)
2714 goto out;
2715
2716 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2717 a, b);
2718 if (err)
2719 goto out;
2720
2721 if (num == 0)
2722 /* this extent is removed; mark slot entirely unused */
2723 ext4_ext_store_pblock(ex, 0);
2724
2725 ex->ee_len = cpu_to_le16(num);
2726 /*
2727 * Do not mark unwritten if all the blocks in the
2728 * extent have been removed.
2729 */
2730 if (unwritten && num)
2731 ext4_ext_mark_unwritten(ex);
2732 /*
2733 * If the extent was completely released,
2734 * we need to remove it from the leaf
2735 */
2736 if (num == 0) {
2737 if (end != EXT_MAX_BLOCKS - 1) {
2738 /*
2739 * For hole punching, we need to scoot all the
2740 * extents up when an extent is removed so that
2741 * we dont have blank extents in the middle
2742 */
2743 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2744 sizeof(struct ext4_extent));
2745
2746 /* Now get rid of the one at the end */
2747 memset(EXT_LAST_EXTENT(eh), 0,
2748 sizeof(struct ext4_extent));
2749 }
2750 le16_add_cpu(&eh->eh_entries, -1);
2751 }
2752
2753 err = ext4_ext_dirty(handle, inode, path + depth);
2754 if (err)
2755 goto out;
2756
2757 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2758 ext4_ext_pblock(ex));
2759 ex--;
2760 ex_ee_block = le32_to_cpu(ex->ee_block);
2761 ex_ee_len = ext4_ext_get_actual_len(ex);
2762 }
2763
2764 if (correct_index && eh->eh_entries)
2765 err = ext4_ext_correct_indexes(handle, inode, path);
2766
2767 /*
2768 * If there's a partial cluster and at least one extent remains in
2769 * the leaf, free the partial cluster if it isn't shared with the
2770 * current extent. If it is shared with the current extent
2771 * we zero partial_cluster because we've reached the start of the
2772 * truncated/punched region and we're done removing blocks.
2773 */
2774 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2775 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2776 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2777 ext4_free_blocks(handle, inode, NULL,
2778 EXT4_C2B(sbi, *partial_cluster),
2779 sbi->s_cluster_ratio,
2780 get_default_free_blocks_flags(inode));
2781 }
2782 *partial_cluster = 0;
2783 }
2784
2785 /* if this leaf is free, then we should
2786 * remove it from index block above */
2787 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2788 err = ext4_ext_rm_idx(handle, inode, path, depth);
2789
2790out:
2791 return err;
2792}
2793
2794/*
2795 * ext4_ext_more_to_rm:
2796 * returns 1 if current index has to be freed (even partial)
2797 */
2798static int
2799ext4_ext_more_to_rm(struct ext4_ext_path *path)
2800{
2801 BUG_ON(path->p_idx == NULL);
2802
2803 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2804 return 0;
2805
2806 /*
2807 * if truncate on deeper level happened, it wasn't partial,
2808 * so we have to consider current index for truncation
2809 */
2810 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2811 return 0;
2812 return 1;
2813}
2814
2815int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2816 ext4_lblk_t end)
2817{
2818 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2819 int depth = ext_depth(inode);
2820 struct ext4_ext_path *path = NULL;
2821 long long partial_cluster = 0;
2822 handle_t *handle;
2823 int i = 0, err = 0;
2824
2825 ext_debug("truncate since %u to %u\n", start, end);
2826
2827 /* probably first extent we're gonna free will be last in block */
2828 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2829 if (IS_ERR(handle))
2830 return PTR_ERR(handle);
2831
2832again:
2833 trace_ext4_ext_remove_space(inode, start, end, depth);
2834
2835 /*
2836 * Check if we are removing extents inside the extent tree. If that
2837 * is the case, we are going to punch a hole inside the extent tree
2838 * so we have to check whether we need to split the extent covering
2839 * the last block to remove so we can easily remove the part of it
2840 * in ext4_ext_rm_leaf().
2841 */
2842 if (end < EXT_MAX_BLOCKS - 1) {
2843 struct ext4_extent *ex;
2844 ext4_lblk_t ee_block, ex_end, lblk;
2845 ext4_fsblk_t pblk;
2846
2847 /* find extent for or closest extent to this block */
2848 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2849 if (IS_ERR(path)) {
2850 ext4_journal_stop(handle);
2851 return PTR_ERR(path);
2852 }
2853 depth = ext_depth(inode);
2854 /* Leaf not may not exist only if inode has no blocks at all */
2855 ex = path[depth].p_ext;
2856 if (!ex) {
2857 if (depth) {
2858 EXT4_ERROR_INODE(inode,
2859 "path[%d].p_hdr == NULL",
2860 depth);
2861 err = -EFSCORRUPTED;
2862 }
2863 goto out;
2864 }
2865
2866 ee_block = le32_to_cpu(ex->ee_block);
2867 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2868
2869 /*
2870 * See if the last block is inside the extent, if so split
2871 * the extent at 'end' block so we can easily remove the
2872 * tail of the first part of the split extent in
2873 * ext4_ext_rm_leaf().
2874 */
2875 if (end >= ee_block && end < ex_end) {
2876
2877 /*
2878 * If we're going to split the extent, note that
2879 * the cluster containing the block after 'end' is
2880 * in use to avoid freeing it when removing blocks.
2881 */
2882 if (sbi->s_cluster_ratio > 1) {
2883 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2884 partial_cluster =
2885 -(long long) EXT4_B2C(sbi, pblk);
2886 }
2887
2888 /*
2889 * Split the extent in two so that 'end' is the last
2890 * block in the first new extent. Also we should not
2891 * fail removing space due to ENOSPC so try to use
2892 * reserved block if that happens.
2893 */
2894 err = ext4_force_split_extent_at(handle, inode, &path,
2895 end + 1, 1);
2896 if (err < 0)
2897 goto out;
2898
2899 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2900 /*
2901 * If there's an extent to the right its first cluster
2902 * contains the immediate right boundary of the
2903 * truncated/punched region. Set partial_cluster to
2904 * its negative value so it won't be freed if shared
2905 * with the current extent. The end < ee_block case
2906 * is handled in ext4_ext_rm_leaf().
2907 */
2908 lblk = ex_end + 1;
2909 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2910 &ex);
2911 if (err)
2912 goto out;
2913 if (pblk)
2914 partial_cluster =
2915 -(long long) EXT4_B2C(sbi, pblk);
2916 }
2917 }
2918 /*
2919 * We start scanning from right side, freeing all the blocks
2920 * after i_size and walking into the tree depth-wise.
2921 */
2922 depth = ext_depth(inode);
2923 if (path) {
2924 int k = i = depth;
2925 while (--k > 0)
2926 path[k].p_block =
2927 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2928 } else {
2929 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2930 GFP_NOFS);
2931 if (path == NULL) {
2932 ext4_journal_stop(handle);
2933 return -ENOMEM;
2934 }
2935 path[0].p_maxdepth = path[0].p_depth = depth;
2936 path[0].p_hdr = ext_inode_hdr(inode);
2937 i = 0;
2938
2939 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2940 err = -EFSCORRUPTED;
2941 goto out;
2942 }
2943 }
2944 err = 0;
2945
2946 while (i >= 0 && err == 0) {
2947 if (i == depth) {
2948 /* this is leaf block */
2949 err = ext4_ext_rm_leaf(handle, inode, path,
2950 &partial_cluster, start,
2951 end);
2952 /* root level has p_bh == NULL, brelse() eats this */
2953 brelse(path[i].p_bh);
2954 path[i].p_bh = NULL;
2955 i--;
2956 continue;
2957 }
2958
2959 /* this is index block */
2960 if (!path[i].p_hdr) {
2961 ext_debug("initialize header\n");
2962 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2963 }
2964
2965 if (!path[i].p_idx) {
2966 /* this level hasn't been touched yet */
2967 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2968 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2969 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2970 path[i].p_hdr,
2971 le16_to_cpu(path[i].p_hdr->eh_entries));
2972 } else {
2973 /* we were already here, see at next index */
2974 path[i].p_idx--;
2975 }
2976
2977 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2978 i, EXT_FIRST_INDEX(path[i].p_hdr),
2979 path[i].p_idx);
2980 if (ext4_ext_more_to_rm(path + i)) {
2981 struct buffer_head *bh;
2982 /* go to the next level */
2983 ext_debug("move to level %d (block %llu)\n",
2984 i + 1, ext4_idx_pblock(path[i].p_idx));
2985 memset(path + i + 1, 0, sizeof(*path));
2986 bh = read_extent_tree_block(inode,
2987 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2988 EXT4_EX_NOCACHE);
2989 if (IS_ERR(bh)) {
2990 /* should we reset i_size? */
2991 err = PTR_ERR(bh);
2992 break;
2993 }
2994 /* Yield here to deal with large extent trees.
2995 * Should be a no-op if we did IO above. */
2996 cond_resched();
2997 if (WARN_ON(i + 1 > depth)) {
2998 err = -EFSCORRUPTED;
2999 break;
3000 }
3001 path[i + 1].p_bh = bh;
3002
3003 /* save actual number of indexes since this
3004 * number is changed at the next iteration */
3005 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3006 i++;
3007 } else {
3008 /* we finished processing this index, go up */
3009 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3010 /* index is empty, remove it;
3011 * handle must be already prepared by the
3012 * truncatei_leaf() */
3013 err = ext4_ext_rm_idx(handle, inode, path, i);
3014 }
3015 /* root level has p_bh == NULL, brelse() eats this */
3016 brelse(path[i].p_bh);
3017 path[i].p_bh = NULL;
3018 i--;
3019 ext_debug("return to level %d\n", i);
3020 }
3021 }
3022
3023 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3024 partial_cluster, path->p_hdr->eh_entries);
3025
3026 /*
3027 * If we still have something in the partial cluster and we have removed
3028 * even the first extent, then we should free the blocks in the partial
3029 * cluster as well. (This code will only run when there are no leaves
3030 * to the immediate left of the truncated/punched region.)
3031 */
3032 if (partial_cluster > 0 && err == 0) {
3033 /* don't zero partial_cluster since it's not used afterwards */
3034 ext4_free_blocks(handle, inode, NULL,
3035 EXT4_C2B(sbi, partial_cluster),
3036 sbi->s_cluster_ratio,
3037 get_default_free_blocks_flags(inode));
3038 }
3039
3040 /* TODO: flexible tree reduction should be here */
3041 if (path->p_hdr->eh_entries == 0) {
3042 /*
3043 * truncate to zero freed all the tree,
3044 * so we need to correct eh_depth
3045 */
3046 err = ext4_ext_get_access(handle, inode, path);
3047 if (err == 0) {
3048 ext_inode_hdr(inode)->eh_depth = 0;
3049 ext_inode_hdr(inode)->eh_max =
3050 cpu_to_le16(ext4_ext_space_root(inode, 0));
3051 err = ext4_ext_dirty(handle, inode, path);
3052 }
3053 }
3054out:
3055 ext4_ext_drop_refs(path);
3056 kfree(path);
3057 path = NULL;
3058 if (err == -EAGAIN)
3059 goto again;
3060 ext4_journal_stop(handle);
3061
3062 return err;
3063}
3064
3065/*
3066 * called at mount time
3067 */
3068void ext4_ext_init(struct super_block *sb)
3069{
3070 /*
3071 * possible initialization would be here
3072 */
3073
3074 if (ext4_has_feature_extents(sb)) {
3075#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3076 printk(KERN_INFO "EXT4-fs: file extents enabled"
3077#ifdef AGGRESSIVE_TEST
3078 ", aggressive tests"
3079#endif
3080#ifdef CHECK_BINSEARCH
3081 ", check binsearch"
3082#endif
3083#ifdef EXTENTS_STATS
3084 ", stats"
3085#endif
3086 "\n");
3087#endif
3088#ifdef EXTENTS_STATS
3089 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3090 EXT4_SB(sb)->s_ext_min = 1 << 30;
3091 EXT4_SB(sb)->s_ext_max = 0;
3092#endif
3093 }
3094}
3095
3096/*
3097 * called at umount time
3098 */
3099void ext4_ext_release(struct super_block *sb)
3100{
3101 if (!ext4_has_feature_extents(sb))
3102 return;
3103
3104#ifdef EXTENTS_STATS
3105 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3106 struct ext4_sb_info *sbi = EXT4_SB(sb);
3107 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3108 sbi->s_ext_blocks, sbi->s_ext_extents,
3109 sbi->s_ext_blocks / sbi->s_ext_extents);
3110 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3111 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3112 }
3113#endif
3114}
3115
3116static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3117{
3118 ext4_lblk_t ee_block;
3119 ext4_fsblk_t ee_pblock;
3120 unsigned int ee_len;
3121
3122 ee_block = le32_to_cpu(ex->ee_block);
3123 ee_len = ext4_ext_get_actual_len(ex);
3124 ee_pblock = ext4_ext_pblock(ex);
3125
3126 if (ee_len == 0)
3127 return 0;
3128
3129 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3130 EXTENT_STATUS_WRITTEN);
3131}
3132
3133/* FIXME!! we need to try to merge to left or right after zero-out */
3134static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3135{
3136 ext4_fsblk_t ee_pblock;
3137 unsigned int ee_len;
3138
3139 ee_len = ext4_ext_get_actual_len(ex);
3140 ee_pblock = ext4_ext_pblock(ex);
3141 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3142 ee_len);
3143}
3144
3145/*
3146 * ext4_split_extent_at() splits an extent at given block.
3147 *
3148 * @handle: the journal handle
3149 * @inode: the file inode
3150 * @path: the path to the extent
3151 * @split: the logical block where the extent is splitted.
3152 * @split_flags: indicates if the extent could be zeroout if split fails, and
3153 * the states(init or unwritten) of new extents.
3154 * @flags: flags used to insert new extent to extent tree.
3155 *
3156 *
3157 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3158 * of which are deterimined by split_flag.
3159 *
3160 * There are two cases:
3161 * a> the extent are splitted into two extent.
3162 * b> split is not needed, and just mark the extent.
3163 *
3164 * return 0 on success.
3165 */
3166static int ext4_split_extent_at(handle_t *handle,
3167 struct inode *inode,
3168 struct ext4_ext_path **ppath,
3169 ext4_lblk_t split,
3170 int split_flag,
3171 int flags)
3172{
3173 struct ext4_ext_path *path = *ppath;
3174 ext4_fsblk_t newblock;
3175 ext4_lblk_t ee_block;
3176 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3177 struct ext4_extent *ex2 = NULL;
3178 unsigned int ee_len, depth;
3179 int err = 0;
3180
3181 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3182 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3183
3184 ext_debug("ext4_split_extents_at: inode %lu, logical"
3185 "block %llu\n", inode->i_ino, (unsigned long long)split);
3186
3187 ext4_ext_show_leaf(inode, path);
3188
3189 depth = ext_depth(inode);
3190 ex = path[depth].p_ext;
3191 ee_block = le32_to_cpu(ex->ee_block);
3192 ee_len = ext4_ext_get_actual_len(ex);
3193 newblock = split - ee_block + ext4_ext_pblock(ex);
3194
3195 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3196 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3197 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3198 EXT4_EXT_MARK_UNWRIT1 |
3199 EXT4_EXT_MARK_UNWRIT2));
3200
3201 err = ext4_ext_get_access(handle, inode, path + depth);
3202 if (err)
3203 goto out;
3204
3205 if (split == ee_block) {
3206 /*
3207 * case b: block @split is the block that the extent begins with
3208 * then we just change the state of the extent, and splitting
3209 * is not needed.
3210 */
3211 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3212 ext4_ext_mark_unwritten(ex);
3213 else
3214 ext4_ext_mark_initialized(ex);
3215
3216 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3217 ext4_ext_try_to_merge(handle, inode, path, ex);
3218
3219 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3220 goto out;
3221 }
3222
3223 /* case a */
3224 memcpy(&orig_ex, ex, sizeof(orig_ex));
3225 ex->ee_len = cpu_to_le16(split - ee_block);
3226 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3227 ext4_ext_mark_unwritten(ex);
3228
3229 /*
3230 * path may lead to new leaf, not to original leaf any more
3231 * after ext4_ext_insert_extent() returns,
3232 */
3233 err = ext4_ext_dirty(handle, inode, path + depth);
3234 if (err)
3235 goto fix_extent_len;
3236
3237 ex2 = &newex;
3238 ex2->ee_block = cpu_to_le32(split);
3239 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3240 ext4_ext_store_pblock(ex2, newblock);
3241 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3242 ext4_ext_mark_unwritten(ex2);
3243
3244 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3245 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3246 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3247 if (split_flag & EXT4_EXT_DATA_VALID1) {
3248 err = ext4_ext_zeroout(inode, ex2);
3249 zero_ex.ee_block = ex2->ee_block;
3250 zero_ex.ee_len = cpu_to_le16(
3251 ext4_ext_get_actual_len(ex2));
3252 ext4_ext_store_pblock(&zero_ex,
3253 ext4_ext_pblock(ex2));
3254 } else {
3255 err = ext4_ext_zeroout(inode, ex);
3256 zero_ex.ee_block = ex->ee_block;
3257 zero_ex.ee_len = cpu_to_le16(
3258 ext4_ext_get_actual_len(ex));
3259 ext4_ext_store_pblock(&zero_ex,
3260 ext4_ext_pblock(ex));
3261 }
3262 } else {
3263 err = ext4_ext_zeroout(inode, &orig_ex);
3264 zero_ex.ee_block = orig_ex.ee_block;
3265 zero_ex.ee_len = cpu_to_le16(
3266 ext4_ext_get_actual_len(&orig_ex));
3267 ext4_ext_store_pblock(&zero_ex,
3268 ext4_ext_pblock(&orig_ex));
3269 }
3270
3271 if (err)
3272 goto fix_extent_len;
3273 /* update the extent length and mark as initialized */
3274 ex->ee_len = cpu_to_le16(ee_len);
3275 ext4_ext_try_to_merge(handle, inode, path, ex);
3276 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3277 if (err)
3278 goto fix_extent_len;
3279
3280 /* update extent status tree */
3281 err = ext4_zeroout_es(inode, &zero_ex);
3282
3283 goto out;
3284 } else if (err)
3285 goto fix_extent_len;
3286
3287out:
3288 ext4_ext_show_leaf(inode, path);
3289 return err;
3290
3291fix_extent_len:
3292 ex->ee_len = orig_ex.ee_len;
3293 ext4_ext_dirty(handle, inode, path + path->p_depth);
3294 return err;
3295}
3296
3297/*
3298 * ext4_split_extents() splits an extent and mark extent which is covered
3299 * by @map as split_flags indicates
3300 *
3301 * It may result in splitting the extent into multiple extents (up to three)
3302 * There are three possibilities:
3303 * a> There is no split required
3304 * b> Splits in two extents: Split is happening at either end of the extent
3305 * c> Splits in three extents: Somone is splitting in middle of the extent
3306 *
3307 */
3308static int ext4_split_extent(handle_t *handle,
3309 struct inode *inode,
3310 struct ext4_ext_path **ppath,
3311 struct ext4_map_blocks *map,
3312 int split_flag,
3313 int flags)
3314{
3315 struct ext4_ext_path *path = *ppath;
3316 ext4_lblk_t ee_block;
3317 struct ext4_extent *ex;
3318 unsigned int ee_len, depth;
3319 int err = 0;
3320 int unwritten;
3321 int split_flag1, flags1;
3322 int allocated = map->m_len;
3323
3324 depth = ext_depth(inode);
3325 ex = path[depth].p_ext;
3326 ee_block = le32_to_cpu(ex->ee_block);
3327 ee_len = ext4_ext_get_actual_len(ex);
3328 unwritten = ext4_ext_is_unwritten(ex);
3329
3330 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3331 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3332 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3333 if (unwritten)
3334 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3335 EXT4_EXT_MARK_UNWRIT2;
3336 if (split_flag & EXT4_EXT_DATA_VALID2)
3337 split_flag1 |= EXT4_EXT_DATA_VALID1;
3338 err = ext4_split_extent_at(handle, inode, ppath,
3339 map->m_lblk + map->m_len, split_flag1, flags1);
3340 if (err)
3341 goto out;
3342 } else {
3343 allocated = ee_len - (map->m_lblk - ee_block);
3344 }
3345 /*
3346 * Update path is required because previous ext4_split_extent_at() may
3347 * result in split of original leaf or extent zeroout.
3348 */
3349 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3350 if (IS_ERR(path))
3351 return PTR_ERR(path);
3352 depth = ext_depth(inode);
3353 ex = path[depth].p_ext;
3354 if (!ex) {
3355 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3356 (unsigned long) map->m_lblk);
3357 return -EFSCORRUPTED;
3358 }
3359 unwritten = ext4_ext_is_unwritten(ex);
3360 split_flag1 = 0;
3361
3362 if (map->m_lblk >= ee_block) {
3363 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3364 if (unwritten) {
3365 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3366 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3367 EXT4_EXT_MARK_UNWRIT2);
3368 }
3369 err = ext4_split_extent_at(handle, inode, ppath,
3370 map->m_lblk, split_flag1, flags);
3371 if (err)
3372 goto out;
3373 }
3374
3375 ext4_ext_show_leaf(inode, path);
3376out:
3377 return err ? err : allocated;
3378}
3379
3380/*
3381 * This function is called by ext4_ext_map_blocks() if someone tries to write
3382 * to an unwritten extent. It may result in splitting the unwritten
3383 * extent into multiple extents (up to three - one initialized and two
3384 * unwritten).
3385 * There are three possibilities:
3386 * a> There is no split required: Entire extent should be initialized
3387 * b> Splits in two extents: Write is happening at either end of the extent
3388 * c> Splits in three extents: Somone is writing in middle of the extent
3389 *
3390 * Pre-conditions:
3391 * - The extent pointed to by 'path' is unwritten.
3392 * - The extent pointed to by 'path' contains a superset
3393 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3394 *
3395 * Post-conditions on success:
3396 * - the returned value is the number of blocks beyond map->l_lblk
3397 * that are allocated and initialized.
3398 * It is guaranteed to be >= map->m_len.
3399 */
3400static int ext4_ext_convert_to_initialized(handle_t *handle,
3401 struct inode *inode,
3402 struct ext4_map_blocks *map,
3403 struct ext4_ext_path **ppath,
3404 int flags)
3405{
3406 struct ext4_ext_path *path = *ppath;
3407 struct ext4_sb_info *sbi;
3408 struct ext4_extent_header *eh;
3409 struct ext4_map_blocks split_map;
3410 struct ext4_extent zero_ex;
3411 struct ext4_extent *ex, *abut_ex;
3412 ext4_lblk_t ee_block, eof_block;
3413 unsigned int ee_len, depth, map_len = map->m_len;
3414 int allocated = 0, max_zeroout = 0;
3415 int err = 0;
3416 int split_flag = 0;
3417
3418 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3419 "block %llu, max_blocks %u\n", inode->i_ino,
3420 (unsigned long long)map->m_lblk, map_len);
3421
3422 sbi = EXT4_SB(inode->i_sb);
3423 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3424 inode->i_sb->s_blocksize_bits;
3425 if (eof_block < map->m_lblk + map_len)
3426 eof_block = map->m_lblk + map_len;
3427
3428 depth = ext_depth(inode);
3429 eh = path[depth].p_hdr;
3430 ex = path[depth].p_ext;
3431 ee_block = le32_to_cpu(ex->ee_block);
3432 ee_len = ext4_ext_get_actual_len(ex);
3433 zero_ex.ee_len = 0;
3434
3435 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3436
3437 /* Pre-conditions */
3438 BUG_ON(!ext4_ext_is_unwritten(ex));
3439 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3440
3441 /*
3442 * Attempt to transfer newly initialized blocks from the currently
3443 * unwritten extent to its neighbor. This is much cheaper
3444 * than an insertion followed by a merge as those involve costly
3445 * memmove() calls. Transferring to the left is the common case in
3446 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3447 * followed by append writes.
3448 *
3449 * Limitations of the current logic:
3450 * - L1: we do not deal with writes covering the whole extent.
3451 * This would require removing the extent if the transfer
3452 * is possible.
3453 * - L2: we only attempt to merge with an extent stored in the
3454 * same extent tree node.
3455 */
3456 if ((map->m_lblk == ee_block) &&
3457 /* See if we can merge left */
3458 (map_len < ee_len) && /*L1*/
3459 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3460 ext4_lblk_t prev_lblk;
3461 ext4_fsblk_t prev_pblk, ee_pblk;
3462 unsigned int prev_len;
3463
3464 abut_ex = ex - 1;
3465 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3466 prev_len = ext4_ext_get_actual_len(abut_ex);
3467 prev_pblk = ext4_ext_pblock(abut_ex);
3468 ee_pblk = ext4_ext_pblock(ex);
3469
3470 /*
3471 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3472 * upon those conditions:
3473 * - C1: abut_ex is initialized,
3474 * - C2: abut_ex is logically abutting ex,
3475 * - C3: abut_ex is physically abutting ex,
3476 * - C4: abut_ex can receive the additional blocks without
3477 * overflowing the (initialized) length limit.
3478 */
3479 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3480 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3481 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3482 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3483 err = ext4_ext_get_access(handle, inode, path + depth);
3484 if (err)
3485 goto out;
3486
3487 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3488 map, ex, abut_ex);
3489
3490 /* Shift the start of ex by 'map_len' blocks */
3491 ex->ee_block = cpu_to_le32(ee_block + map_len);
3492 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3493 ex->ee_len = cpu_to_le16(ee_len - map_len);
3494 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3495
3496 /* Extend abut_ex by 'map_len' blocks */
3497 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3498
3499 /* Result: number of initialized blocks past m_lblk */
3500 allocated = map_len;
3501 }
3502 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3503 (map_len < ee_len) && /*L1*/
3504 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3505 /* See if we can merge right */
3506 ext4_lblk_t next_lblk;
3507 ext4_fsblk_t next_pblk, ee_pblk;
3508 unsigned int next_len;
3509
3510 abut_ex = ex + 1;
3511 next_lblk = le32_to_cpu(abut_ex->ee_block);
3512 next_len = ext4_ext_get_actual_len(abut_ex);
3513 next_pblk = ext4_ext_pblock(abut_ex);
3514 ee_pblk = ext4_ext_pblock(ex);
3515
3516 /*
3517 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3518 * upon those conditions:
3519 * - C1: abut_ex is initialized,
3520 * - C2: abut_ex is logically abutting ex,
3521 * - C3: abut_ex is physically abutting ex,
3522 * - C4: abut_ex can receive the additional blocks without
3523 * overflowing the (initialized) length limit.
3524 */
3525 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3526 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3527 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3528 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3529 err = ext4_ext_get_access(handle, inode, path + depth);
3530 if (err)
3531 goto out;
3532
3533 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3534 map, ex, abut_ex);
3535
3536 /* Shift the start of abut_ex by 'map_len' blocks */
3537 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3538 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3539 ex->ee_len = cpu_to_le16(ee_len - map_len);
3540 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3541
3542 /* Extend abut_ex by 'map_len' blocks */
3543 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3544
3545 /* Result: number of initialized blocks past m_lblk */
3546 allocated = map_len;
3547 }
3548 }
3549 if (allocated) {
3550 /* Mark the block containing both extents as dirty */
3551 ext4_ext_dirty(handle, inode, path + depth);
3552
3553 /* Update path to point to the right extent */
3554 path[depth].p_ext = abut_ex;
3555 goto out;
3556 } else
3557 allocated = ee_len - (map->m_lblk - ee_block);
3558
3559 WARN_ON(map->m_lblk < ee_block);
3560 /*
3561 * It is safe to convert extent to initialized via explicit
3562 * zeroout only if extent is fully inside i_size or new_size.
3563 */
3564 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3565
3566 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3567 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3568 (inode->i_sb->s_blocksize_bits - 10);
3569
3570 if (ext4_encrypted_inode(inode))
3571 max_zeroout = 0;
3572
3573 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3574 if (max_zeroout && (ee_len <= max_zeroout)) {
3575 err = ext4_ext_zeroout(inode, ex);
3576 if (err)
3577 goto out;
3578 zero_ex.ee_block = ex->ee_block;
3579 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3580 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3581
3582 err = ext4_ext_get_access(handle, inode, path + depth);
3583 if (err)
3584 goto out;
3585 ext4_ext_mark_initialized(ex);
3586 ext4_ext_try_to_merge(handle, inode, path, ex);
3587 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3588 goto out;
3589 }
3590
3591 /*
3592 * four cases:
3593 * 1. split the extent into three extents.
3594 * 2. split the extent into two extents, zeroout the first half.
3595 * 3. split the extent into two extents, zeroout the second half.
3596 * 4. split the extent into two extents with out zeroout.
3597 */
3598 split_map.m_lblk = map->m_lblk;
3599 split_map.m_len = map->m_len;
3600
3601 if (max_zeroout && (allocated > map->m_len)) {
3602 if (allocated <= max_zeroout) {
3603 /* case 3 */
3604 zero_ex.ee_block =
3605 cpu_to_le32(map->m_lblk);
3606 zero_ex.ee_len = cpu_to_le16(allocated);
3607 ext4_ext_store_pblock(&zero_ex,
3608 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3609 err = ext4_ext_zeroout(inode, &zero_ex);
3610 if (err)
3611 goto out;
3612 split_map.m_lblk = map->m_lblk;
3613 split_map.m_len = allocated;
3614 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3615 /* case 2 */
3616 if (map->m_lblk != ee_block) {
3617 zero_ex.ee_block = ex->ee_block;
3618 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3619 ee_block);
3620 ext4_ext_store_pblock(&zero_ex,
3621 ext4_ext_pblock(ex));
3622 err = ext4_ext_zeroout(inode, &zero_ex);
3623 if (err)
3624 goto out;
3625 }
3626
3627 split_map.m_lblk = ee_block;
3628 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3629 allocated = map->m_len;
3630 }
3631 }
3632
3633 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3634 flags);
3635 if (err > 0)
3636 err = 0;
3637out:
3638 /* If we have gotten a failure, don't zero out status tree */
3639 if (!err)
3640 err = ext4_zeroout_es(inode, &zero_ex);
3641 return err ? err : allocated;
3642}
3643
3644/*
3645 * This function is called by ext4_ext_map_blocks() from
3646 * ext4_get_blocks_dio_write() when DIO to write
3647 * to an unwritten extent.
3648 *
3649 * Writing to an unwritten extent may result in splitting the unwritten
3650 * extent into multiple initialized/unwritten extents (up to three)
3651 * There are three possibilities:
3652 * a> There is no split required: Entire extent should be unwritten
3653 * b> Splits in two extents: Write is happening at either end of the extent
3654 * c> Splits in three extents: Somone is writing in middle of the extent
3655 *
3656 * This works the same way in the case of initialized -> unwritten conversion.
3657 *
3658 * One of more index blocks maybe needed if the extent tree grow after
3659 * the unwritten extent split. To prevent ENOSPC occur at the IO
3660 * complete, we need to split the unwritten extent before DIO submit
3661 * the IO. The unwritten extent called at this time will be split
3662 * into three unwritten extent(at most). After IO complete, the part
3663 * being filled will be convert to initialized by the end_io callback function
3664 * via ext4_convert_unwritten_extents().
3665 *
3666 * Returns the size of unwritten extent to be written on success.
3667 */
3668static int ext4_split_convert_extents(handle_t *handle,
3669 struct inode *inode,
3670 struct ext4_map_blocks *map,
3671 struct ext4_ext_path **ppath,
3672 int flags)
3673{
3674 struct ext4_ext_path *path = *ppath;
3675 ext4_lblk_t eof_block;
3676 ext4_lblk_t ee_block;
3677 struct ext4_extent *ex;
3678 unsigned int ee_len;
3679 int split_flag = 0, depth;
3680
3681 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3682 __func__, inode->i_ino,
3683 (unsigned long long)map->m_lblk, map->m_len);
3684
3685 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3686 inode->i_sb->s_blocksize_bits;
3687 if (eof_block < map->m_lblk + map->m_len)
3688 eof_block = map->m_lblk + map->m_len;
3689 /*
3690 * It is safe to convert extent to initialized via explicit
3691 * zeroout only if extent is fully insde i_size or new_size.
3692 */
3693 depth = ext_depth(inode);
3694 ex = path[depth].p_ext;
3695 ee_block = le32_to_cpu(ex->ee_block);
3696 ee_len = ext4_ext_get_actual_len(ex);
3697
3698 /* Convert to unwritten */
3699 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3700 split_flag |= EXT4_EXT_DATA_VALID1;
3701 /* Convert to initialized */
3702 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3703 split_flag |= ee_block + ee_len <= eof_block ?
3704 EXT4_EXT_MAY_ZEROOUT : 0;
3705 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3706 }
3707 flags |= EXT4_GET_BLOCKS_PRE_IO;
3708 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3709}
3710
3711static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3712 struct inode *inode,
3713 struct ext4_map_blocks *map,
3714 struct ext4_ext_path **ppath)
3715{
3716 struct ext4_ext_path *path = *ppath;
3717 struct ext4_extent *ex;
3718 ext4_lblk_t ee_block;
3719 unsigned int ee_len;
3720 int depth;
3721 int err = 0;
3722
3723 depth = ext_depth(inode);
3724 ex = path[depth].p_ext;
3725 ee_block = le32_to_cpu(ex->ee_block);
3726 ee_len = ext4_ext_get_actual_len(ex);
3727
3728 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3729 "block %llu, max_blocks %u\n", inode->i_ino,
3730 (unsigned long long)ee_block, ee_len);
3731
3732 /* If extent is larger than requested it is a clear sign that we still
3733 * have some extent state machine issues left. So extent_split is still
3734 * required.
3735 * TODO: Once all related issues will be fixed this situation should be
3736 * illegal.
3737 */
3738 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3739#ifdef EXT4_DEBUG
3740 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3741 " len %u; IO logical block %llu, len %u\n",
3742 inode->i_ino, (unsigned long long)ee_block, ee_len,
3743 (unsigned long long)map->m_lblk, map->m_len);
3744#endif
3745 err = ext4_split_convert_extents(handle, inode, map, ppath,
3746 EXT4_GET_BLOCKS_CONVERT);
3747 if (err < 0)
3748 return err;
3749 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3750 if (IS_ERR(path))
3751 return PTR_ERR(path);
3752 depth = ext_depth(inode);
3753 ex = path[depth].p_ext;
3754 }
3755
3756 err = ext4_ext_get_access(handle, inode, path + depth);
3757 if (err)
3758 goto out;
3759 /* first mark the extent as initialized */
3760 ext4_ext_mark_initialized(ex);
3761
3762 /* note: ext4_ext_correct_indexes() isn't needed here because
3763 * borders are not changed
3764 */
3765 ext4_ext_try_to_merge(handle, inode, path, ex);
3766
3767 /* Mark modified extent as dirty */
3768 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3769out:
3770 ext4_ext_show_leaf(inode, path);
3771 return err;
3772}
3773
3774static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3775 sector_t block, int count)
3776{
3777 int i;
3778 for (i = 0; i < count; i++)
3779 unmap_underlying_metadata(bdev, block + i);
3780}
3781
3782/*
3783 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3784 */
3785static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3786 ext4_lblk_t lblk,
3787 struct ext4_ext_path *path,
3788 unsigned int len)
3789{
3790 int i, depth;
3791 struct ext4_extent_header *eh;
3792 struct ext4_extent *last_ex;
3793
3794 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3795 return 0;
3796
3797 depth = ext_depth(inode);
3798 eh = path[depth].p_hdr;
3799
3800 /*
3801 * We're going to remove EOFBLOCKS_FL entirely in future so we
3802 * do not care for this case anymore. Simply remove the flag
3803 * if there are no extents.
3804 */
3805 if (unlikely(!eh->eh_entries))
3806 goto out;
3807 last_ex = EXT_LAST_EXTENT(eh);
3808 /*
3809 * We should clear the EOFBLOCKS_FL flag if we are writing the
3810 * last block in the last extent in the file. We test this by
3811 * first checking to see if the caller to
3812 * ext4_ext_get_blocks() was interested in the last block (or
3813 * a block beyond the last block) in the current extent. If
3814 * this turns out to be false, we can bail out from this
3815 * function immediately.
3816 */
3817 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3818 ext4_ext_get_actual_len(last_ex))
3819 return 0;
3820 /*
3821 * If the caller does appear to be planning to write at or
3822 * beyond the end of the current extent, we then test to see
3823 * if the current extent is the last extent in the file, by
3824 * checking to make sure it was reached via the rightmost node
3825 * at each level of the tree.
3826 */
3827 for (i = depth-1; i >= 0; i--)
3828 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3829 return 0;
3830out:
3831 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3832 return ext4_mark_inode_dirty(handle, inode);
3833}
3834
3835/**
3836 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3837 *
3838 * Return 1 if there is a delalloc block in the range, otherwise 0.
3839 */
3840int ext4_find_delalloc_range(struct inode *inode,
3841 ext4_lblk_t lblk_start,
3842 ext4_lblk_t lblk_end)
3843{
3844 struct extent_status es;
3845
3846 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3847 if (es.es_len == 0)
3848 return 0; /* there is no delay extent in this tree */
3849 else if (es.es_lblk <= lblk_start &&
3850 lblk_start < es.es_lblk + es.es_len)
3851 return 1;
3852 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3853 return 1;
3854 else
3855 return 0;
3856}
3857
3858int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3859{
3860 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3861 ext4_lblk_t lblk_start, lblk_end;
3862 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3863 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3864
3865 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3866}
3867
3868/**
3869 * Determines how many complete clusters (out of those specified by the 'map')
3870 * are under delalloc and were reserved quota for.
3871 * This function is called when we are writing out the blocks that were
3872 * originally written with their allocation delayed, but then the space was
3873 * allocated using fallocate() before the delayed allocation could be resolved.
3874 * The cases to look for are:
3875 * ('=' indicated delayed allocated blocks
3876 * '-' indicates non-delayed allocated blocks)
3877 * (a) partial clusters towards beginning and/or end outside of allocated range
3878 * are not delalloc'ed.
3879 * Ex:
3880 * |----c---=|====c====|====c====|===-c----|
3881 * |++++++ allocated ++++++|
3882 * ==> 4 complete clusters in above example
3883 *
3884 * (b) partial cluster (outside of allocated range) towards either end is
3885 * marked for delayed allocation. In this case, we will exclude that
3886 * cluster.
3887 * Ex:
3888 * |----====c========|========c========|
3889 * |++++++ allocated ++++++|
3890 * ==> 1 complete clusters in above example
3891 *
3892 * Ex:
3893 * |================c================|
3894 * |++++++ allocated ++++++|
3895 * ==> 0 complete clusters in above example
3896 *
3897 * The ext4_da_update_reserve_space will be called only if we
3898 * determine here that there were some "entire" clusters that span
3899 * this 'allocated' range.
3900 * In the non-bigalloc case, this function will just end up returning num_blks
3901 * without ever calling ext4_find_delalloc_range.
3902 */
3903static unsigned int
3904get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3905 unsigned int num_blks)
3906{
3907 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3908 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3909 ext4_lblk_t lblk_from, lblk_to, c_offset;
3910 unsigned int allocated_clusters = 0;
3911
3912 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3913 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3914
3915 /* max possible clusters for this allocation */
3916 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3917
3918 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3919
3920 /* Check towards left side */
3921 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3922 if (c_offset) {
3923 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3924 lblk_to = lblk_from + c_offset - 1;
3925
3926 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3927 allocated_clusters--;
3928 }
3929
3930 /* Now check towards right. */
3931 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3932 if (allocated_clusters && c_offset) {
3933 lblk_from = lblk_start + num_blks;
3934 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3935
3936 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3937 allocated_clusters--;
3938 }
3939
3940 return allocated_clusters;
3941}
3942
3943static int
3944convert_initialized_extent(handle_t *handle, struct inode *inode,
3945 struct ext4_map_blocks *map,
3946 struct ext4_ext_path **ppath,
3947 unsigned int allocated)
3948{
3949 struct ext4_ext_path *path = *ppath;
3950 struct ext4_extent *ex;
3951 ext4_lblk_t ee_block;
3952 unsigned int ee_len;
3953 int depth;
3954 int err = 0;
3955
3956 /*
3957 * Make sure that the extent is no bigger than we support with
3958 * unwritten extent
3959 */
3960 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3961 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3962
3963 depth = ext_depth(inode);
3964 ex = path[depth].p_ext;
3965 ee_block = le32_to_cpu(ex->ee_block);
3966 ee_len = ext4_ext_get_actual_len(ex);
3967
3968 ext_debug("%s: inode %lu, logical"
3969 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3970 (unsigned long long)ee_block, ee_len);
3971
3972 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3973 err = ext4_split_convert_extents(handle, inode, map, ppath,
3974 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3975 if (err < 0)
3976 return err;
3977 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3978 if (IS_ERR(path))
3979 return PTR_ERR(path);
3980 depth = ext_depth(inode);
3981 ex = path[depth].p_ext;
3982 if (!ex) {
3983 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3984 (unsigned long) map->m_lblk);
3985 return -EFSCORRUPTED;
3986 }
3987 }
3988
3989 err = ext4_ext_get_access(handle, inode, path + depth);
3990 if (err)
3991 return err;
3992 /* first mark the extent as unwritten */
3993 ext4_ext_mark_unwritten(ex);
3994
3995 /* note: ext4_ext_correct_indexes() isn't needed here because
3996 * borders are not changed
3997 */
3998 ext4_ext_try_to_merge(handle, inode, path, ex);
3999
4000 /* Mark modified extent as dirty */
4001 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4002 if (err)
4003 return err;
4004 ext4_ext_show_leaf(inode, path);
4005
4006 ext4_update_inode_fsync_trans(handle, inode, 1);
4007 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4008 if (err)
4009 return err;
4010 map->m_flags |= EXT4_MAP_UNWRITTEN;
4011 if (allocated > map->m_len)
4012 allocated = map->m_len;
4013 map->m_len = allocated;
4014 return allocated;
4015}
4016
4017static int
4018ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4019 struct ext4_map_blocks *map,
4020 struct ext4_ext_path **ppath, int flags,
4021 unsigned int allocated, ext4_fsblk_t newblock)
4022{
4023 struct ext4_ext_path *path = *ppath;
4024 int ret = 0;
4025 int err = 0;
4026
4027 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4028 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4029 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4030 flags, allocated);
4031 ext4_ext_show_leaf(inode, path);
4032
4033 /*
4034 * When writing into unwritten space, we should not fail to
4035 * allocate metadata blocks for the new extent block if needed.
4036 */
4037 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4038
4039 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4040 allocated, newblock);
4041
4042 /* get_block() before submit the IO, split the extent */
4043 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4044 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4045 flags | EXT4_GET_BLOCKS_CONVERT);
4046 if (ret <= 0)
4047 goto out;
4048 map->m_flags |= EXT4_MAP_UNWRITTEN;
4049 goto out;
4050 }
4051 /* IO end_io complete, convert the filled extent to written */
4052 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4053 if (flags & EXT4_GET_BLOCKS_ZERO) {
4054 if (allocated > map->m_len)
4055 allocated = map->m_len;
4056 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4057 allocated);
4058 if (err < 0)
4059 goto out2;
4060 }
4061 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4062 ppath);
4063 if (ret >= 0) {
4064 ext4_update_inode_fsync_trans(handle, inode, 1);
4065 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4066 path, map->m_len);
4067 } else
4068 err = ret;
4069 map->m_flags |= EXT4_MAP_MAPPED;
4070 map->m_pblk = newblock;
4071 if (allocated > map->m_len)
4072 allocated = map->m_len;
4073 map->m_len = allocated;
4074 goto out2;
4075 }
4076 /* buffered IO case */
4077 /*
4078 * repeat fallocate creation request
4079 * we already have an unwritten extent
4080 */
4081 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4082 map->m_flags |= EXT4_MAP_UNWRITTEN;
4083 goto map_out;
4084 }
4085
4086 /* buffered READ or buffered write_begin() lookup */
4087 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4088 /*
4089 * We have blocks reserved already. We
4090 * return allocated blocks so that delalloc
4091 * won't do block reservation for us. But
4092 * the buffer head will be unmapped so that
4093 * a read from the block returns 0s.
4094 */
4095 map->m_flags |= EXT4_MAP_UNWRITTEN;
4096 goto out1;
4097 }
4098
4099 /* buffered write, writepage time, convert*/
4100 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4101 if (ret >= 0)
4102 ext4_update_inode_fsync_trans(handle, inode, 1);
4103out:
4104 if (ret <= 0) {
4105 err = ret;
4106 goto out2;
4107 } else
4108 allocated = ret;
4109 map->m_flags |= EXT4_MAP_NEW;
4110 /*
4111 * if we allocated more blocks than requested
4112 * we need to make sure we unmap the extra block
4113 * allocated. The actual needed block will get
4114 * unmapped later when we find the buffer_head marked
4115 * new.
4116 */
4117 if (allocated > map->m_len) {
4118 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4119 newblock + map->m_len,
4120 allocated - map->m_len);
4121 allocated = map->m_len;
4122 }
4123 map->m_len = allocated;
4124
4125 /*
4126 * If we have done fallocate with the offset that is already
4127 * delayed allocated, we would have block reservation
4128 * and quota reservation done in the delayed write path.
4129 * But fallocate would have already updated quota and block
4130 * count for this offset. So cancel these reservation
4131 */
4132 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4133 unsigned int reserved_clusters;
4134 reserved_clusters = get_reserved_cluster_alloc(inode,
4135 map->m_lblk, map->m_len);
4136 if (reserved_clusters)
4137 ext4_da_update_reserve_space(inode,
4138 reserved_clusters,
4139 0);
4140 }
4141
4142map_out:
4143 map->m_flags |= EXT4_MAP_MAPPED;
4144 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4145 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4146 map->m_len);
4147 if (err < 0)
4148 goto out2;
4149 }
4150out1:
4151 if (allocated > map->m_len)
4152 allocated = map->m_len;
4153 ext4_ext_show_leaf(inode, path);
4154 map->m_pblk = newblock;
4155 map->m_len = allocated;
4156out2:
4157 return err ? err : allocated;
4158}
4159
4160/*
4161 * get_implied_cluster_alloc - check to see if the requested
4162 * allocation (in the map structure) overlaps with a cluster already
4163 * allocated in an extent.
4164 * @sb The filesystem superblock structure
4165 * @map The requested lblk->pblk mapping
4166 * @ex The extent structure which might contain an implied
4167 * cluster allocation
4168 *
4169 * This function is called by ext4_ext_map_blocks() after we failed to
4170 * find blocks that were already in the inode's extent tree. Hence,
4171 * we know that the beginning of the requested region cannot overlap
4172 * the extent from the inode's extent tree. There are three cases we
4173 * want to catch. The first is this case:
4174 *
4175 * |--- cluster # N--|
4176 * |--- extent ---| |---- requested region ---|
4177 * |==========|
4178 *
4179 * The second case that we need to test for is this one:
4180 *
4181 * |--------- cluster # N ----------------|
4182 * |--- requested region --| |------- extent ----|
4183 * |=======================|
4184 *
4185 * The third case is when the requested region lies between two extents
4186 * within the same cluster:
4187 * |------------- cluster # N-------------|
4188 * |----- ex -----| |---- ex_right ----|
4189 * |------ requested region ------|
4190 * |================|
4191 *
4192 * In each of the above cases, we need to set the map->m_pblk and
4193 * map->m_len so it corresponds to the return the extent labelled as
4194 * "|====|" from cluster #N, since it is already in use for data in
4195 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4196 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4197 * as a new "allocated" block region. Otherwise, we will return 0 and
4198 * ext4_ext_map_blocks() will then allocate one or more new clusters
4199 * by calling ext4_mb_new_blocks().
4200 */
4201static int get_implied_cluster_alloc(struct super_block *sb,
4202 struct ext4_map_blocks *map,
4203 struct ext4_extent *ex,
4204 struct ext4_ext_path *path)
4205{
4206 struct ext4_sb_info *sbi = EXT4_SB(sb);
4207 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4208 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4209 ext4_lblk_t rr_cluster_start;
4210 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4211 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4212 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4213
4214 /* The extent passed in that we are trying to match */
4215 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4216 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4217
4218 /* The requested region passed into ext4_map_blocks() */
4219 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4220
4221 if ((rr_cluster_start == ex_cluster_end) ||
4222 (rr_cluster_start == ex_cluster_start)) {
4223 if (rr_cluster_start == ex_cluster_end)
4224 ee_start += ee_len - 1;
4225 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4226 map->m_len = min(map->m_len,
4227 (unsigned) sbi->s_cluster_ratio - c_offset);
4228 /*
4229 * Check for and handle this case:
4230 *
4231 * |--------- cluster # N-------------|
4232 * |------- extent ----|
4233 * |--- requested region ---|
4234 * |===========|
4235 */
4236
4237 if (map->m_lblk < ee_block)
4238 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4239
4240 /*
4241 * Check for the case where there is already another allocated
4242 * block to the right of 'ex' but before the end of the cluster.
4243 *
4244 * |------------- cluster # N-------------|
4245 * |----- ex -----| |---- ex_right ----|
4246 * |------ requested region ------|
4247 * |================|
4248 */
4249 if (map->m_lblk > ee_block) {
4250 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4251 map->m_len = min(map->m_len, next - map->m_lblk);
4252 }
4253
4254 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4255 return 1;
4256 }
4257
4258 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4259 return 0;
4260}
4261
4262
4263/*
4264 * Block allocation/map/preallocation routine for extents based files
4265 *
4266 *
4267 * Need to be called with
4268 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4269 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4270 *
4271 * return > 0, number of of blocks already mapped/allocated
4272 * if create == 0 and these are pre-allocated blocks
4273 * buffer head is unmapped
4274 * otherwise blocks are mapped
4275 *
4276 * return = 0, if plain look up failed (blocks have not been allocated)
4277 * buffer head is unmapped
4278 *
4279 * return < 0, error case.
4280 */
4281int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4282 struct ext4_map_blocks *map, int flags)
4283{
4284 struct ext4_ext_path *path = NULL;
4285 struct ext4_extent newex, *ex, *ex2;
4286 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4287 ext4_fsblk_t newblock = 0;
4288 int free_on_err = 0, err = 0, depth, ret;
4289 unsigned int allocated = 0, offset = 0;
4290 unsigned int allocated_clusters = 0;
4291 struct ext4_allocation_request ar;
4292 ext4_lblk_t cluster_offset;
4293 bool map_from_cluster = false;
4294
4295 ext_debug("blocks %u/%u requested for inode %lu\n",
4296 map->m_lblk, map->m_len, inode->i_ino);
4297 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4298
4299 /* find extent for this block */
4300 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4301 if (IS_ERR(path)) {
4302 err = PTR_ERR(path);
4303 path = NULL;
4304 goto out2;
4305 }
4306
4307 depth = ext_depth(inode);
4308
4309 /*
4310 * consistent leaf must not be empty;
4311 * this situation is possible, though, _during_ tree modification;
4312 * this is why assert can't be put in ext4_find_extent()
4313 */
4314 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4315 EXT4_ERROR_INODE(inode, "bad extent address "
4316 "lblock: %lu, depth: %d pblock %lld",
4317 (unsigned long) map->m_lblk, depth,
4318 path[depth].p_block);
4319 err = -EFSCORRUPTED;
4320 goto out2;
4321 }
4322
4323 ex = path[depth].p_ext;
4324 if (ex) {
4325 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4326 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4327 unsigned short ee_len;
4328
4329
4330 /*
4331 * unwritten extents are treated as holes, except that
4332 * we split out initialized portions during a write.
4333 */
4334 ee_len = ext4_ext_get_actual_len(ex);
4335
4336 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4337
4338 /* if found extent covers block, simply return it */
4339 if (in_range(map->m_lblk, ee_block, ee_len)) {
4340 newblock = map->m_lblk - ee_block + ee_start;
4341 /* number of remaining blocks in the extent */
4342 allocated = ee_len - (map->m_lblk - ee_block);
4343 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4344 ee_block, ee_len, newblock);
4345
4346 /*
4347 * If the extent is initialized check whether the
4348 * caller wants to convert it to unwritten.
4349 */
4350 if ((!ext4_ext_is_unwritten(ex)) &&
4351 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4352 allocated = convert_initialized_extent(
4353 handle, inode, map, &path,
4354 allocated);
4355 goto out2;
4356 } else if (!ext4_ext_is_unwritten(ex))
4357 goto out;
4358
4359 ret = ext4_ext_handle_unwritten_extents(
4360 handle, inode, map, &path, flags,
4361 allocated, newblock);
4362 if (ret < 0)
4363 err = ret;
4364 else
4365 allocated = ret;
4366 goto out2;
4367 }
4368 }
4369
4370 /*
4371 * requested block isn't allocated yet;
4372 * we couldn't try to create block if create flag is zero
4373 */
4374 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4375 ext4_lblk_t hole_start, hole_len;
4376
4377 hole_start = map->m_lblk;
4378 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4379 /*
4380 * put just found gap into cache to speed up
4381 * subsequent requests
4382 */
4383 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4384
4385 /* Update hole_len to reflect hole size after map->m_lblk */
4386 if (hole_start != map->m_lblk)
4387 hole_len -= map->m_lblk - hole_start;
4388 map->m_pblk = 0;
4389 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4390
4391 goto out2;
4392 }
4393
4394 /*
4395 * Okay, we need to do block allocation.
4396 */
4397 newex.ee_block = cpu_to_le32(map->m_lblk);
4398 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4399
4400 /*
4401 * If we are doing bigalloc, check to see if the extent returned
4402 * by ext4_find_extent() implies a cluster we can use.
4403 */
4404 if (cluster_offset && ex &&
4405 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4406 ar.len = allocated = map->m_len;
4407 newblock = map->m_pblk;
4408 map_from_cluster = true;
4409 goto got_allocated_blocks;
4410 }
4411
4412 /* find neighbour allocated blocks */
4413 ar.lleft = map->m_lblk;
4414 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4415 if (err)
4416 goto out2;
4417 ar.lright = map->m_lblk;
4418 ex2 = NULL;
4419 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4420 if (err)
4421 goto out2;
4422
4423 /* Check if the extent after searching to the right implies a
4424 * cluster we can use. */
4425 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4426 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4427 ar.len = allocated = map->m_len;
4428 newblock = map->m_pblk;
4429 map_from_cluster = true;
4430 goto got_allocated_blocks;
4431 }
4432
4433 /*
4434 * See if request is beyond maximum number of blocks we can have in
4435 * a single extent. For an initialized extent this limit is
4436 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4437 * EXT_UNWRITTEN_MAX_LEN.
4438 */
4439 if (map->m_len > EXT_INIT_MAX_LEN &&
4440 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4441 map->m_len = EXT_INIT_MAX_LEN;
4442 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4443 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4444 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4445
4446 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4447 newex.ee_len = cpu_to_le16(map->m_len);
4448 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4449 if (err)
4450 allocated = ext4_ext_get_actual_len(&newex);
4451 else
4452 allocated = map->m_len;
4453
4454 /* allocate new block */
4455 ar.inode = inode;
4456 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4457 ar.logical = map->m_lblk;
4458 /*
4459 * We calculate the offset from the beginning of the cluster
4460 * for the logical block number, since when we allocate a
4461 * physical cluster, the physical block should start at the
4462 * same offset from the beginning of the cluster. This is
4463 * needed so that future calls to get_implied_cluster_alloc()
4464 * work correctly.
4465 */
4466 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4467 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4468 ar.goal -= offset;
4469 ar.logical -= offset;
4470 if (S_ISREG(inode->i_mode))
4471 ar.flags = EXT4_MB_HINT_DATA;
4472 else
4473 /* disable in-core preallocation for non-regular files */
4474 ar.flags = 0;
4475 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4476 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4477 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4478 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4479 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4480 ar.flags |= EXT4_MB_USE_RESERVED;
4481 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4482 if (!newblock)
4483 goto out2;
4484 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4485 ar.goal, newblock, allocated);
4486 free_on_err = 1;
4487 allocated_clusters = ar.len;
4488 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4489 if (ar.len > allocated)
4490 ar.len = allocated;
4491
4492got_allocated_blocks:
4493 /* try to insert new extent into found leaf and return */
4494 ext4_ext_store_pblock(&newex, newblock + offset);
4495 newex.ee_len = cpu_to_le16(ar.len);
4496 /* Mark unwritten */
4497 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4498 ext4_ext_mark_unwritten(&newex);
4499 map->m_flags |= EXT4_MAP_UNWRITTEN;
4500 }
4501
4502 err = 0;
4503 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4504 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4505 path, ar.len);
4506 if (!err)
4507 err = ext4_ext_insert_extent(handle, inode, &path,
4508 &newex, flags);
4509
4510 if (err && free_on_err) {
4511 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4512 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4513 /* free data blocks we just allocated */
4514 /* not a good idea to call discard here directly,
4515 * but otherwise we'd need to call it every free() */
4516 ext4_discard_preallocations(inode);
4517 ext4_free_blocks(handle, inode, NULL, newblock,
4518 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4519 goto out2;
4520 }
4521
4522 /* previous routine could use block we allocated */
4523 newblock = ext4_ext_pblock(&newex);
4524 allocated = ext4_ext_get_actual_len(&newex);
4525 if (allocated > map->m_len)
4526 allocated = map->m_len;
4527 map->m_flags |= EXT4_MAP_NEW;
4528
4529 /*
4530 * Update reserved blocks/metadata blocks after successful
4531 * block allocation which had been deferred till now.
4532 */
4533 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4534 unsigned int reserved_clusters;
4535 /*
4536 * Check how many clusters we had reserved this allocated range
4537 */
4538 reserved_clusters = get_reserved_cluster_alloc(inode,
4539 map->m_lblk, allocated);
4540 if (!map_from_cluster) {
4541 BUG_ON(allocated_clusters < reserved_clusters);
4542 if (reserved_clusters < allocated_clusters) {
4543 struct ext4_inode_info *ei = EXT4_I(inode);
4544 int reservation = allocated_clusters -
4545 reserved_clusters;
4546 /*
4547 * It seems we claimed few clusters outside of
4548 * the range of this allocation. We should give
4549 * it back to the reservation pool. This can
4550 * happen in the following case:
4551 *
4552 * * Suppose s_cluster_ratio is 4 (i.e., each
4553 * cluster has 4 blocks. Thus, the clusters
4554 * are [0-3],[4-7],[8-11]...
4555 * * First comes delayed allocation write for
4556 * logical blocks 10 & 11. Since there were no
4557 * previous delayed allocated blocks in the
4558 * range [8-11], we would reserve 1 cluster
4559 * for this write.
4560 * * Next comes write for logical blocks 3 to 8.
4561 * In this case, we will reserve 2 clusters
4562 * (for [0-3] and [4-7]; and not for [8-11] as
4563 * that range has a delayed allocated blocks.
4564 * Thus total reserved clusters now becomes 3.
4565 * * Now, during the delayed allocation writeout
4566 * time, we will first write blocks [3-8] and
4567 * allocate 3 clusters for writing these
4568 * blocks. Also, we would claim all these
4569 * three clusters above.
4570 * * Now when we come here to writeout the
4571 * blocks [10-11], we would expect to claim
4572 * the reservation of 1 cluster we had made
4573 * (and we would claim it since there are no
4574 * more delayed allocated blocks in the range
4575 * [8-11]. But our reserved cluster count had
4576 * already gone to 0.
4577 *
4578 * Thus, at the step 4 above when we determine
4579 * that there are still some unwritten delayed
4580 * allocated blocks outside of our current
4581 * block range, we should increment the
4582 * reserved clusters count so that when the
4583 * remaining blocks finally gets written, we
4584 * could claim them.
4585 */
4586 dquot_reserve_block(inode,
4587 EXT4_C2B(sbi, reservation));
4588 spin_lock(&ei->i_block_reservation_lock);
4589 ei->i_reserved_data_blocks += reservation;
4590 spin_unlock(&ei->i_block_reservation_lock);
4591 }
4592 /*
4593 * We will claim quota for all newly allocated blocks.
4594 * We're updating the reserved space *after* the
4595 * correction above so we do not accidentally free
4596 * all the metadata reservation because we might
4597 * actually need it later on.
4598 */
4599 ext4_da_update_reserve_space(inode, allocated_clusters,
4600 1);
4601 }
4602 }
4603
4604 /*
4605 * Cache the extent and update transaction to commit on fdatasync only
4606 * when it is _not_ an unwritten extent.
4607 */
4608 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4609 ext4_update_inode_fsync_trans(handle, inode, 1);
4610 else
4611 ext4_update_inode_fsync_trans(handle, inode, 0);
4612out:
4613 if (allocated > map->m_len)
4614 allocated = map->m_len;
4615 ext4_ext_show_leaf(inode, path);
4616 map->m_flags |= EXT4_MAP_MAPPED;
4617 map->m_pblk = newblock;
4618 map->m_len = allocated;
4619out2:
4620 ext4_ext_drop_refs(path);
4621 kfree(path);
4622
4623 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4624 err ? err : allocated);
4625 return err ? err : allocated;
4626}
4627
4628void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4629{
4630 struct super_block *sb = inode->i_sb;
4631 ext4_lblk_t last_block;
4632 int err = 0;
4633
4634 /*
4635 * TODO: optimization is possible here.
4636 * Probably we need not scan at all,
4637 * because page truncation is enough.
4638 */
4639
4640 /* we have to know where to truncate from in crash case */
4641 EXT4_I(inode)->i_disksize = inode->i_size;
4642 ext4_mark_inode_dirty(handle, inode);
4643
4644 last_block = (inode->i_size + sb->s_blocksize - 1)
4645 >> EXT4_BLOCK_SIZE_BITS(sb);
4646retry:
4647 err = ext4_es_remove_extent(inode, last_block,
4648 EXT_MAX_BLOCKS - last_block);
4649 if (err == -ENOMEM) {
4650 cond_resched();
4651 congestion_wait(BLK_RW_ASYNC, HZ/50);
4652 goto retry;
4653 }
4654 if (err) {
4655 ext4_std_error(inode->i_sb, err);
4656 return;
4657 }
4658 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4659 ext4_std_error(inode->i_sb, err);
4660}
4661
4662static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4663 ext4_lblk_t len, loff_t new_size,
4664 int flags, int mode)
4665{
4666 struct inode *inode = file_inode(file);
4667 handle_t *handle;
4668 int ret = 0;
4669 int ret2 = 0;
4670 int retries = 0;
4671 int depth = 0;
4672 struct ext4_map_blocks map;
4673 unsigned int credits;
4674 loff_t epos;
4675
4676 map.m_lblk = offset;
4677 map.m_len = len;
4678 /*
4679 * Don't normalize the request if it can fit in one extent so
4680 * that it doesn't get unnecessarily split into multiple
4681 * extents.
4682 */
4683 if (len <= EXT_UNWRITTEN_MAX_LEN)
4684 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4685
4686 /*
4687 * credits to insert 1 extent into extent tree
4688 */
4689 credits = ext4_chunk_trans_blocks(inode, len);
4690 /*
4691 * We can only call ext_depth() on extent based inodes
4692 */
4693 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4694 depth = ext_depth(inode);
4695 else
4696 depth = -1;
4697
4698retry:
4699 while (ret >= 0 && len) {
4700 /*
4701 * Recalculate credits when extent tree depth changes.
4702 */
4703 if (depth >= 0 && depth != ext_depth(inode)) {
4704 credits = ext4_chunk_trans_blocks(inode, len);
4705 depth = ext_depth(inode);
4706 }
4707
4708 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4709 credits);
4710 if (IS_ERR(handle)) {
4711 ret = PTR_ERR(handle);
4712 break;
4713 }
4714 ret = ext4_map_blocks(handle, inode, &map, flags);
4715 if (ret <= 0) {
4716 ext4_debug("inode #%lu: block %u: len %u: "
4717 "ext4_ext_map_blocks returned %d",
4718 inode->i_ino, map.m_lblk,
4719 map.m_len, ret);
4720 ext4_mark_inode_dirty(handle, inode);
4721 ret2 = ext4_journal_stop(handle);
4722 break;
4723 }
4724 map.m_lblk += ret;
4725 map.m_len = len = len - ret;
4726 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4727 inode->i_ctime = ext4_current_time(inode);
4728 if (new_size) {
4729 if (epos > new_size)
4730 epos = new_size;
4731 if (ext4_update_inode_size(inode, epos) & 0x1)
4732 inode->i_mtime = inode->i_ctime;
4733 } else {
4734 if (epos > inode->i_size)
4735 ext4_set_inode_flag(inode,
4736 EXT4_INODE_EOFBLOCKS);
4737 }
4738 ext4_mark_inode_dirty(handle, inode);
4739 ret2 = ext4_journal_stop(handle);
4740 if (ret2)
4741 break;
4742 }
4743 if (ret == -ENOSPC &&
4744 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4745 ret = 0;
4746 goto retry;
4747 }
4748
4749 return ret > 0 ? ret2 : ret;
4750}
4751
4752static long ext4_zero_range(struct file *file, loff_t offset,
4753 loff_t len, int mode)
4754{
4755 struct inode *inode = file_inode(file);
4756 handle_t *handle = NULL;
4757 unsigned int max_blocks;
4758 loff_t new_size = 0;
4759 int ret = 0;
4760 int flags;
4761 int credits;
4762 int partial_begin, partial_end;
4763 loff_t start, end;
4764 ext4_lblk_t lblk;
4765 unsigned int blkbits = inode->i_blkbits;
4766
4767 trace_ext4_zero_range(inode, offset, len, mode);
4768
4769 if (!S_ISREG(inode->i_mode))
4770 return -EINVAL;
4771
4772 /* Call ext4_force_commit to flush all data in case of data=journal. */
4773 if (ext4_should_journal_data(inode)) {
4774 ret = ext4_force_commit(inode->i_sb);
4775 if (ret)
4776 return ret;
4777 }
4778
4779 /*
4780 * Round up offset. This is not fallocate, we neet to zero out
4781 * blocks, so convert interior block aligned part of the range to
4782 * unwritten and possibly manually zero out unaligned parts of the
4783 * range.
4784 */
4785 start = round_up(offset, 1 << blkbits);
4786 end = round_down((offset + len), 1 << blkbits);
4787
4788 if (start < offset || end > offset + len)
4789 return -EINVAL;
4790 partial_begin = offset & ((1 << blkbits) - 1);
4791 partial_end = (offset + len) & ((1 << blkbits) - 1);
4792
4793 lblk = start >> blkbits;
4794 max_blocks = (end >> blkbits);
4795 if (max_blocks < lblk)
4796 max_blocks = 0;
4797 else
4798 max_blocks -= lblk;
4799
4800 inode_lock(inode);
4801
4802 /*
4803 * Indirect files do not support unwritten extnets
4804 */
4805 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4806 ret = -EOPNOTSUPP;
4807 goto out_mutex;
4808 }
4809
4810 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4811 offset + len > i_size_read(inode)) {
4812 new_size = offset + len;
4813 ret = inode_newsize_ok(inode, new_size);
4814 if (ret)
4815 goto out_mutex;
4816 }
4817
4818 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4819 if (mode & FALLOC_FL_KEEP_SIZE)
4820 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4821
4822 /* Wait all existing dio workers, newcomers will block on i_mutex */
4823 ext4_inode_block_unlocked_dio(inode);
4824 inode_dio_wait(inode);
4825
4826 /* Preallocate the range including the unaligned edges */
4827 if (partial_begin || partial_end) {
4828 ret = ext4_alloc_file_blocks(file,
4829 round_down(offset, 1 << blkbits) >> blkbits,
4830 (round_up((offset + len), 1 << blkbits) -
4831 round_down(offset, 1 << blkbits)) >> blkbits,
4832 new_size, flags, mode);
4833 if (ret)
4834 goto out_dio;
4835
4836 }
4837
4838 /* Zero range excluding the unaligned edges */
4839 if (max_blocks > 0) {
4840 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4841 EXT4_EX_NOCACHE);
4842
4843 /*
4844 * Prevent page faults from reinstantiating pages we have
4845 * released from page cache.
4846 */
4847 down_write(&EXT4_I(inode)->i_mmap_sem);
4848 ret = ext4_update_disksize_before_punch(inode, offset, len);
4849 if (ret) {
4850 up_write(&EXT4_I(inode)->i_mmap_sem);
4851 goto out_dio;
4852 }
4853 /* Now release the pages and zero block aligned part of pages */
4854 truncate_pagecache_range(inode, start, end - 1);
4855 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4856
4857 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4858 flags, mode);
4859 up_write(&EXT4_I(inode)->i_mmap_sem);
4860 if (ret)
4861 goto out_dio;
4862 }
4863 if (!partial_begin && !partial_end)
4864 goto out_dio;
4865
4866 /*
4867 * In worst case we have to writeout two nonadjacent unwritten
4868 * blocks and update the inode
4869 */
4870 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4871 if (ext4_should_journal_data(inode))
4872 credits += 2;
4873 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4874 if (IS_ERR(handle)) {
4875 ret = PTR_ERR(handle);
4876 ext4_std_error(inode->i_sb, ret);
4877 goto out_dio;
4878 }
4879
4880 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4881 if (new_size) {
4882 ext4_update_inode_size(inode, new_size);
4883 } else {
4884 /*
4885 * Mark that we allocate beyond EOF so the subsequent truncate
4886 * can proceed even if the new size is the same as i_size.
4887 */
4888 if ((offset + len) > i_size_read(inode))
4889 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4890 }
4891 ext4_mark_inode_dirty(handle, inode);
4892
4893 /* Zero out partial block at the edges of the range */
4894 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4895
4896 if (file->f_flags & O_SYNC)
4897 ext4_handle_sync(handle);
4898
4899 ext4_journal_stop(handle);
4900out_dio:
4901 ext4_inode_resume_unlocked_dio(inode);
4902out_mutex:
4903 inode_unlock(inode);
4904 return ret;
4905}
4906
4907/*
4908 * preallocate space for a file. This implements ext4's fallocate file
4909 * operation, which gets called from sys_fallocate system call.
4910 * For block-mapped files, posix_fallocate should fall back to the method
4911 * of writing zeroes to the required new blocks (the same behavior which is
4912 * expected for file systems which do not support fallocate() system call).
4913 */
4914long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4915{
4916 struct inode *inode = file_inode(file);
4917 loff_t new_size = 0;
4918 unsigned int max_blocks;
4919 int ret = 0;
4920 int flags;
4921 ext4_lblk_t lblk;
4922 unsigned int blkbits = inode->i_blkbits;
4923
4924 /*
4925 * Encrypted inodes can't handle collapse range or insert
4926 * range since we would need to re-encrypt blocks with a
4927 * different IV or XTS tweak (which are based on the logical
4928 * block number).
4929 *
4930 * XXX It's not clear why zero range isn't working, but we'll
4931 * leave it disabled for encrypted inodes for now. This is a
4932 * bug we should fix....
4933 */
4934 if (ext4_encrypted_inode(inode) &&
4935 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4936 FALLOC_FL_ZERO_RANGE)))
4937 return -EOPNOTSUPP;
4938
4939 /* Return error if mode is not supported */
4940 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4941 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4942 FALLOC_FL_INSERT_RANGE))
4943 return -EOPNOTSUPP;
4944
4945 if (mode & FALLOC_FL_PUNCH_HOLE)
4946 return ext4_punch_hole(inode, offset, len);
4947
4948 ret = ext4_convert_inline_data(inode);
4949 if (ret)
4950 return ret;
4951
4952 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4953 return ext4_collapse_range(inode, offset, len);
4954
4955 if (mode & FALLOC_FL_INSERT_RANGE)
4956 return ext4_insert_range(inode, offset, len);
4957
4958 if (mode & FALLOC_FL_ZERO_RANGE)
4959 return ext4_zero_range(file, offset, len, mode);
4960
4961 trace_ext4_fallocate_enter(inode, offset, len, mode);
4962 lblk = offset >> blkbits;
4963 /*
4964 * We can't just convert len to max_blocks because
4965 * If blocksize = 4096 offset = 3072 and len = 2048
4966 */
4967 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4968 - lblk;
4969
4970 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4971 if (mode & FALLOC_FL_KEEP_SIZE)
4972 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4973
4974 inode_lock(inode);
4975
4976 /*
4977 * We only support preallocation for extent-based files only
4978 */
4979 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4980 ret = -EOPNOTSUPP;
4981 goto out;
4982 }
4983
4984 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4985 offset + len > i_size_read(inode)) {
4986 new_size = offset + len;
4987 ret = inode_newsize_ok(inode, new_size);
4988 if (ret)
4989 goto out;
4990 }
4991
4992 /* Wait all existing dio workers, newcomers will block on i_mutex */
4993 ext4_inode_block_unlocked_dio(inode);
4994 inode_dio_wait(inode);
4995
4996 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4997 flags, mode);
4998 ext4_inode_resume_unlocked_dio(inode);
4999 if (ret)
5000 goto out;
5001
5002 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
5003 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5004 EXT4_I(inode)->i_sync_tid);
5005 }
5006out:
5007 inode_unlock(inode);
5008 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
5009 return ret;
5010}
5011
5012/*
5013 * This function convert a range of blocks to written extents
5014 * The caller of this function will pass the start offset and the size.
5015 * all unwritten extents within this range will be converted to
5016 * written extents.
5017 *
5018 * This function is called from the direct IO end io call back
5019 * function, to convert the fallocated extents after IO is completed.
5020 * Returns 0 on success.
5021 */
5022int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5023 loff_t offset, ssize_t len)
5024{
5025 unsigned int max_blocks;
5026 int ret = 0;
5027 int ret2 = 0;
5028 struct ext4_map_blocks map;
5029 unsigned int credits, blkbits = inode->i_blkbits;
5030
5031 map.m_lblk = offset >> blkbits;
5032 /*
5033 * We can't just convert len to max_blocks because
5034 * If blocksize = 4096 offset = 3072 and len = 2048
5035 */
5036 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5037 map.m_lblk);
5038 /*
5039 * This is somewhat ugly but the idea is clear: When transaction is
5040 * reserved, everything goes into it. Otherwise we rather start several
5041 * smaller transactions for conversion of each extent separately.
5042 */
5043 if (handle) {
5044 handle = ext4_journal_start_reserved(handle,
5045 EXT4_HT_EXT_CONVERT);
5046 if (IS_ERR(handle))
5047 return PTR_ERR(handle);
5048 credits = 0;
5049 } else {
5050 /*
5051 * credits to insert 1 extent into extent tree
5052 */
5053 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5054 }
5055 while (ret >= 0 && ret < max_blocks) {
5056 map.m_lblk += ret;
5057 map.m_len = (max_blocks -= ret);
5058 if (credits) {
5059 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5060 credits);
5061 if (IS_ERR(handle)) {
5062 ret = PTR_ERR(handle);
5063 break;
5064 }
5065 }
5066 ret = ext4_map_blocks(handle, inode, &map,
5067 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5068 if (ret <= 0)
5069 ext4_warning(inode->i_sb,
5070 "inode #%lu: block %u: len %u: "
5071 "ext4_ext_map_blocks returned %d",
5072 inode->i_ino, map.m_lblk,
5073 map.m_len, ret);
5074 ext4_mark_inode_dirty(handle, inode);
5075 if (credits)
5076 ret2 = ext4_journal_stop(handle);
5077 if (ret <= 0 || ret2)
5078 break;
5079 }
5080 if (!credits)
5081 ret2 = ext4_journal_stop(handle);
5082 return ret > 0 ? ret2 : ret;
5083}
5084
5085/*
5086 * If newes is not existing extent (newes->ec_pblk equals zero) find
5087 * delayed extent at start of newes and update newes accordingly and
5088 * return start of the next delayed extent.
5089 *
5090 * If newes is existing extent (newes->ec_pblk is not equal zero)
5091 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5092 * extent found. Leave newes unmodified.
5093 */
5094static int ext4_find_delayed_extent(struct inode *inode,
5095 struct extent_status *newes)
5096{
5097 struct extent_status es;
5098 ext4_lblk_t block, next_del;
5099
5100 if (newes->es_pblk == 0) {
5101 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5102 newes->es_lblk + newes->es_len - 1, &es);
5103
5104 /*
5105 * No extent in extent-tree contains block @newes->es_pblk,
5106 * then the block may stay in 1)a hole or 2)delayed-extent.
5107 */
5108 if (es.es_len == 0)
5109 /* A hole found. */
5110 return 0;
5111
5112 if (es.es_lblk > newes->es_lblk) {
5113 /* A hole found. */
5114 newes->es_len = min(es.es_lblk - newes->es_lblk,
5115 newes->es_len);
5116 return 0;
5117 }
5118
5119 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5120 }
5121
5122 block = newes->es_lblk + newes->es_len;
5123 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5124 if (es.es_len == 0)
5125 next_del = EXT_MAX_BLOCKS;
5126 else
5127 next_del = es.es_lblk;
5128
5129 return next_del;
5130}
5131/* fiemap flags we can handle specified here */
5132#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5133
5134static int ext4_xattr_fiemap(struct inode *inode,
5135 struct fiemap_extent_info *fieinfo)
5136{
5137 __u64 physical = 0;
5138 __u64 length;
5139 __u32 flags = FIEMAP_EXTENT_LAST;
5140 int blockbits = inode->i_sb->s_blocksize_bits;
5141 int error = 0;
5142
5143 /* in-inode? */
5144 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5145 struct ext4_iloc iloc;
5146 int offset; /* offset of xattr in inode */
5147
5148 error = ext4_get_inode_loc(inode, &iloc);
5149 if (error)
5150 return error;
5151 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5152 offset = EXT4_GOOD_OLD_INODE_SIZE +
5153 EXT4_I(inode)->i_extra_isize;
5154 physical += offset;
5155 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5156 flags |= FIEMAP_EXTENT_DATA_INLINE;
5157 brelse(iloc.bh);
5158 } else { /* external block */
5159 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5160 length = inode->i_sb->s_blocksize;
5161 }
5162
5163 if (physical)
5164 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5165 length, flags);
5166 return (error < 0 ? error : 0);
5167}
5168
5169int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5170 __u64 start, __u64 len)
5171{
5172 ext4_lblk_t start_blk;
5173 int error = 0;
5174
5175 if (ext4_has_inline_data(inode)) {
5176 int has_inline = 1;
5177
5178 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5179 start, len);
5180
5181 if (has_inline)
5182 return error;
5183 }
5184
5185 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5186 error = ext4_ext_precache(inode);
5187 if (error)
5188 return error;
5189 }
5190
5191 /* fallback to generic here if not in extents fmt */
5192 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5193 return generic_block_fiemap(inode, fieinfo, start, len,
5194 ext4_get_block);
5195
5196 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5197 return -EBADR;
5198
5199 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5200 error = ext4_xattr_fiemap(inode, fieinfo);
5201 } else {
5202 ext4_lblk_t len_blks;
5203 __u64 last_blk;
5204
5205 start_blk = start >> inode->i_sb->s_blocksize_bits;
5206 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5207 if (last_blk >= EXT_MAX_BLOCKS)
5208 last_blk = EXT_MAX_BLOCKS-1;
5209 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5210
5211 /*
5212 * Walk the extent tree gathering extent information
5213 * and pushing extents back to the user.
5214 */
5215 error = ext4_fill_fiemap_extents(inode, start_blk,
5216 len_blks, fieinfo);
5217 }
5218 return error;
5219}
5220
5221/*
5222 * ext4_access_path:
5223 * Function to access the path buffer for marking it dirty.
5224 * It also checks if there are sufficient credits left in the journal handle
5225 * to update path.
5226 */
5227static int
5228ext4_access_path(handle_t *handle, struct inode *inode,
5229 struct ext4_ext_path *path)
5230{
5231 int credits, err;
5232
5233 if (!ext4_handle_valid(handle))
5234 return 0;
5235
5236 /*
5237 * Check if need to extend journal credits
5238 * 3 for leaf, sb, and inode plus 2 (bmap and group
5239 * descriptor) for each block group; assume two block
5240 * groups
5241 */
5242 if (handle->h_buffer_credits < 7) {
5243 credits = ext4_writepage_trans_blocks(inode);
5244 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5245 /* EAGAIN is success */
5246 if (err && err != -EAGAIN)
5247 return err;
5248 }
5249
5250 err = ext4_ext_get_access(handle, inode, path);
5251 return err;
5252}
5253
5254/*
5255 * ext4_ext_shift_path_extents:
5256 * Shift the extents of a path structure lying between path[depth].p_ext
5257 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5258 * if it is right shift or left shift operation.
5259 */
5260static int
5261ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5262 struct inode *inode, handle_t *handle,
5263 enum SHIFT_DIRECTION SHIFT)
5264{
5265 int depth, err = 0;
5266 struct ext4_extent *ex_start, *ex_last;
5267 bool update = 0;
5268 depth = path->p_depth;
5269
5270 while (depth >= 0) {
5271 if (depth == path->p_depth) {
5272 ex_start = path[depth].p_ext;
5273 if (!ex_start)
5274 return -EFSCORRUPTED;
5275
5276 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5277
5278 err = ext4_access_path(handle, inode, path + depth);
5279 if (err)
5280 goto out;
5281
5282 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5283 update = 1;
5284
5285 while (ex_start <= ex_last) {
5286 if (SHIFT == SHIFT_LEFT) {
5287 le32_add_cpu(&ex_start->ee_block,
5288 -shift);
5289 /* Try to merge to the left. */
5290 if ((ex_start >
5291 EXT_FIRST_EXTENT(path[depth].p_hdr))
5292 &&
5293 ext4_ext_try_to_merge_right(inode,
5294 path, ex_start - 1))
5295 ex_last--;
5296 else
5297 ex_start++;
5298 } else {
5299 le32_add_cpu(&ex_last->ee_block, shift);
5300 ext4_ext_try_to_merge_right(inode, path,
5301 ex_last);
5302 ex_last--;
5303 }
5304 }
5305 err = ext4_ext_dirty(handle, inode, path + depth);
5306 if (err)
5307 goto out;
5308
5309 if (--depth < 0 || !update)
5310 break;
5311 }
5312
5313 /* Update index too */
5314 err = ext4_access_path(handle, inode, path + depth);
5315 if (err)
5316 goto out;
5317
5318 if (SHIFT == SHIFT_LEFT)
5319 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5320 else
5321 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5322 err = ext4_ext_dirty(handle, inode, path + depth);
5323 if (err)
5324 goto out;
5325
5326 /* we are done if current index is not a starting index */
5327 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5328 break;
5329
5330 depth--;
5331 }
5332
5333out:
5334 return err;
5335}
5336
5337/*
5338 * ext4_ext_shift_extents:
5339 * All the extents which lies in the range from @start to the last allocated
5340 * block for the @inode are shifted either towards left or right (depending
5341 * upon @SHIFT) by @shift blocks.
5342 * On success, 0 is returned, error otherwise.
5343 */
5344static int
5345ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5346 ext4_lblk_t start, ext4_lblk_t shift,
5347 enum SHIFT_DIRECTION SHIFT)
5348{
5349 struct ext4_ext_path *path;
5350 int ret = 0, depth;
5351 struct ext4_extent *extent;
5352 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5353
5354 /* Let path point to the last extent */
5355 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5356 if (IS_ERR(path))
5357 return PTR_ERR(path);
5358
5359 depth = path->p_depth;
5360 extent = path[depth].p_ext;
5361 if (!extent)
5362 goto out;
5363
5364 stop = le32_to_cpu(extent->ee_block) +
5365 ext4_ext_get_actual_len(extent);
5366
5367 /*
5368 * In case of left shift, Don't start shifting extents until we make
5369 * sure the hole is big enough to accommodate the shift.
5370 */
5371 if (SHIFT == SHIFT_LEFT) {
5372 path = ext4_find_extent(inode, start - 1, &path, 0);
5373 if (IS_ERR(path))
5374 return PTR_ERR(path);
5375 depth = path->p_depth;
5376 extent = path[depth].p_ext;
5377 if (extent) {
5378 ex_start = le32_to_cpu(extent->ee_block);
5379 ex_end = le32_to_cpu(extent->ee_block) +
5380 ext4_ext_get_actual_len(extent);
5381 } else {
5382 ex_start = 0;
5383 ex_end = 0;
5384 }
5385
5386 if ((start == ex_start && shift > ex_start) ||
5387 (shift > start - ex_end)) {
5388 ext4_ext_drop_refs(path);
5389 kfree(path);
5390 return -EINVAL;
5391 }
5392 }
5393
5394 /*
5395 * In case of left shift, iterator points to start and it is increased
5396 * till we reach stop. In case of right shift, iterator points to stop
5397 * and it is decreased till we reach start.
5398 */
5399 if (SHIFT == SHIFT_LEFT)
5400 iterator = &start;
5401 else
5402 iterator = &stop;
5403
5404 /* Its safe to start updating extents */
5405 while (start < stop) {
5406 path = ext4_find_extent(inode, *iterator, &path, 0);
5407 if (IS_ERR(path))
5408 return PTR_ERR(path);
5409 depth = path->p_depth;
5410 extent = path[depth].p_ext;
5411 if (!extent) {
5412 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5413 (unsigned long) *iterator);
5414 return -EFSCORRUPTED;
5415 }
5416 if (SHIFT == SHIFT_LEFT && *iterator >
5417 le32_to_cpu(extent->ee_block)) {
5418 /* Hole, move to the next extent */
5419 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5420 path[depth].p_ext++;
5421 } else {
5422 *iterator = ext4_ext_next_allocated_block(path);
5423 continue;
5424 }
5425 }
5426
5427 if (SHIFT == SHIFT_LEFT) {
5428 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5429 *iterator = le32_to_cpu(extent->ee_block) +
5430 ext4_ext_get_actual_len(extent);
5431 } else {
5432 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5433 *iterator = le32_to_cpu(extent->ee_block) > 0 ?
5434 le32_to_cpu(extent->ee_block) - 1 : 0;
5435 /* Update path extent in case we need to stop */
5436 while (le32_to_cpu(extent->ee_block) < start)
5437 extent++;
5438 path[depth].p_ext = extent;
5439 }
5440 ret = ext4_ext_shift_path_extents(path, shift, inode,
5441 handle, SHIFT);
5442 if (ret)
5443 break;
5444 }
5445out:
5446 ext4_ext_drop_refs(path);
5447 kfree(path);
5448 return ret;
5449}
5450
5451/*
5452 * ext4_collapse_range:
5453 * This implements the fallocate's collapse range functionality for ext4
5454 * Returns: 0 and non-zero on error.
5455 */
5456int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5457{
5458 struct super_block *sb = inode->i_sb;
5459 ext4_lblk_t punch_start, punch_stop;
5460 handle_t *handle;
5461 unsigned int credits;
5462 loff_t new_size, ioffset;
5463 int ret;
5464
5465 /*
5466 * We need to test this early because xfstests assumes that a
5467 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5468 * system does not support collapse range.
5469 */
5470 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5471 return -EOPNOTSUPP;
5472
5473 /* Collapse range works only on fs block size aligned offsets. */
5474 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5475 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5476 return -EINVAL;
5477
5478 if (!S_ISREG(inode->i_mode))
5479 return -EINVAL;
5480
5481 trace_ext4_collapse_range(inode, offset, len);
5482
5483 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5484 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5485
5486 /* Call ext4_force_commit to flush all data in case of data=journal. */
5487 if (ext4_should_journal_data(inode)) {
5488 ret = ext4_force_commit(inode->i_sb);
5489 if (ret)
5490 return ret;
5491 }
5492
5493 inode_lock(inode);
5494 /*
5495 * There is no need to overlap collapse range with EOF, in which case
5496 * it is effectively a truncate operation
5497 */
5498 if (offset + len >= i_size_read(inode)) {
5499 ret = -EINVAL;
5500 goto out_mutex;
5501 }
5502
5503 /* Currently just for extent based files */
5504 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5505 ret = -EOPNOTSUPP;
5506 goto out_mutex;
5507 }
5508
5509 /* Wait for existing dio to complete */
5510 ext4_inode_block_unlocked_dio(inode);
5511 inode_dio_wait(inode);
5512
5513 /*
5514 * Prevent page faults from reinstantiating pages we have released from
5515 * page cache.
5516 */
5517 down_write(&EXT4_I(inode)->i_mmap_sem);
5518 /*
5519 * Need to round down offset to be aligned with page size boundary
5520 * for page size > block size.
5521 */
5522 ioffset = round_down(offset, PAGE_SIZE);
5523 /*
5524 * Write tail of the last page before removed range since it will get
5525 * removed from the page cache below.
5526 */
5527 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5528 if (ret)
5529 goto out_mmap;
5530 /*
5531 * Write data that will be shifted to preserve them when discarding
5532 * page cache below. We are also protected from pages becoming dirty
5533 * by i_mmap_sem.
5534 */
5535 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5536 LLONG_MAX);
5537 if (ret)
5538 goto out_mmap;
5539 truncate_pagecache(inode, ioffset);
5540
5541 credits = ext4_writepage_trans_blocks(inode);
5542 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5543 if (IS_ERR(handle)) {
5544 ret = PTR_ERR(handle);
5545 goto out_mmap;
5546 }
5547
5548 down_write(&EXT4_I(inode)->i_data_sem);
5549 ext4_discard_preallocations(inode);
5550
5551 ret = ext4_es_remove_extent(inode, punch_start,
5552 EXT_MAX_BLOCKS - punch_start);
5553 if (ret) {
5554 up_write(&EXT4_I(inode)->i_data_sem);
5555 goto out_stop;
5556 }
5557
5558 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5559 if (ret) {
5560 up_write(&EXT4_I(inode)->i_data_sem);
5561 goto out_stop;
5562 }
5563 ext4_discard_preallocations(inode);
5564
5565 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5566 punch_stop - punch_start, SHIFT_LEFT);
5567 if (ret) {
5568 up_write(&EXT4_I(inode)->i_data_sem);
5569 goto out_stop;
5570 }
5571
5572 new_size = i_size_read(inode) - len;
5573 i_size_write(inode, new_size);
5574 EXT4_I(inode)->i_disksize = new_size;
5575
5576 up_write(&EXT4_I(inode)->i_data_sem);
5577 if (IS_SYNC(inode))
5578 ext4_handle_sync(handle);
5579 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5580 ext4_mark_inode_dirty(handle, inode);
5581
5582out_stop:
5583 ext4_journal_stop(handle);
5584out_mmap:
5585 up_write(&EXT4_I(inode)->i_mmap_sem);
5586 ext4_inode_resume_unlocked_dio(inode);
5587out_mutex:
5588 inode_unlock(inode);
5589 return ret;
5590}
5591
5592/*
5593 * ext4_insert_range:
5594 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5595 * The data blocks starting from @offset to the EOF are shifted by @len
5596 * towards right to create a hole in the @inode. Inode size is increased
5597 * by len bytes.
5598 * Returns 0 on success, error otherwise.
5599 */
5600int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5601{
5602 struct super_block *sb = inode->i_sb;
5603 handle_t *handle;
5604 struct ext4_ext_path *path;
5605 struct ext4_extent *extent;
5606 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5607 unsigned int credits, ee_len;
5608 int ret = 0, depth, split_flag = 0;
5609 loff_t ioffset;
5610
5611 /*
5612 * We need to test this early because xfstests assumes that an
5613 * insert range of (0, 1) will return EOPNOTSUPP if the file
5614 * system does not support insert range.
5615 */
5616 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5617 return -EOPNOTSUPP;
5618
5619 /* Insert range works only on fs block size aligned offsets. */
5620 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5621 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5622 return -EINVAL;
5623
5624 if (!S_ISREG(inode->i_mode))
5625 return -EOPNOTSUPP;
5626
5627 trace_ext4_insert_range(inode, offset, len);
5628
5629 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5630 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5631
5632 /* Call ext4_force_commit to flush all data in case of data=journal */
5633 if (ext4_should_journal_data(inode)) {
5634 ret = ext4_force_commit(inode->i_sb);
5635 if (ret)
5636 return ret;
5637 }
5638
5639 inode_lock(inode);
5640 /* Currently just for extent based files */
5641 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5642 ret = -EOPNOTSUPP;
5643 goto out_mutex;
5644 }
5645
5646 /* Check for wrap through zero */
5647 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5648 ret = -EFBIG;
5649 goto out_mutex;
5650 }
5651
5652 /* Offset should be less than i_size */
5653 if (offset >= i_size_read(inode)) {
5654 ret = -EINVAL;
5655 goto out_mutex;
5656 }
5657
5658 /* Wait for existing dio to complete */
5659 ext4_inode_block_unlocked_dio(inode);
5660 inode_dio_wait(inode);
5661
5662 /*
5663 * Prevent page faults from reinstantiating pages we have released from
5664 * page cache.
5665 */
5666 down_write(&EXT4_I(inode)->i_mmap_sem);
5667 /*
5668 * Need to round down to align start offset to page size boundary
5669 * for page size > block size.
5670 */
5671 ioffset = round_down(offset, PAGE_SIZE);
5672 /* Write out all dirty pages */
5673 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5674 LLONG_MAX);
5675 if (ret)
5676 goto out_mmap;
5677 truncate_pagecache(inode, ioffset);
5678
5679 credits = ext4_writepage_trans_blocks(inode);
5680 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5681 if (IS_ERR(handle)) {
5682 ret = PTR_ERR(handle);
5683 goto out_mmap;
5684 }
5685
5686 /* Expand file to avoid data loss if there is error while shifting */
5687 inode->i_size += len;
5688 EXT4_I(inode)->i_disksize += len;
5689 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5690 ret = ext4_mark_inode_dirty(handle, inode);
5691 if (ret)
5692 goto out_stop;
5693
5694 down_write(&EXT4_I(inode)->i_data_sem);
5695 ext4_discard_preallocations(inode);
5696
5697 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5698 if (IS_ERR(path)) {
5699 up_write(&EXT4_I(inode)->i_data_sem);
5700 goto out_stop;
5701 }
5702
5703 depth = ext_depth(inode);
5704 extent = path[depth].p_ext;
5705 if (extent) {
5706 ee_start_lblk = le32_to_cpu(extent->ee_block);
5707 ee_len = ext4_ext_get_actual_len(extent);
5708
5709 /*
5710 * If offset_lblk is not the starting block of extent, split
5711 * the extent @offset_lblk
5712 */
5713 if ((offset_lblk > ee_start_lblk) &&
5714 (offset_lblk < (ee_start_lblk + ee_len))) {
5715 if (ext4_ext_is_unwritten(extent))
5716 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5717 EXT4_EXT_MARK_UNWRIT2;
5718 ret = ext4_split_extent_at(handle, inode, &path,
5719 offset_lblk, split_flag,
5720 EXT4_EX_NOCACHE |
5721 EXT4_GET_BLOCKS_PRE_IO |
5722 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5723 }
5724
5725 ext4_ext_drop_refs(path);
5726 kfree(path);
5727 if (ret < 0) {
5728 up_write(&EXT4_I(inode)->i_data_sem);
5729 goto out_stop;
5730 }
5731 }
5732
5733 ret = ext4_es_remove_extent(inode, offset_lblk,
5734 EXT_MAX_BLOCKS - offset_lblk);
5735 if (ret) {
5736 up_write(&EXT4_I(inode)->i_data_sem);
5737 goto out_stop;
5738 }
5739
5740 /*
5741 * if offset_lblk lies in a hole which is at start of file, use
5742 * ee_start_lblk to shift extents
5743 */
5744 ret = ext4_ext_shift_extents(inode, handle,
5745 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5746 len_lblk, SHIFT_RIGHT);
5747
5748 up_write(&EXT4_I(inode)->i_data_sem);
5749 if (IS_SYNC(inode))
5750 ext4_handle_sync(handle);
5751
5752out_stop:
5753 ext4_journal_stop(handle);
5754out_mmap:
5755 up_write(&EXT4_I(inode)->i_mmap_sem);
5756 ext4_inode_resume_unlocked_dio(inode);
5757out_mutex:
5758 inode_unlock(inode);
5759 return ret;
5760}
5761
5762/**
5763 * ext4_swap_extents - Swap extents between two inodes
5764 *
5765 * @inode1: First inode
5766 * @inode2: Second inode
5767 * @lblk1: Start block for first inode
5768 * @lblk2: Start block for second inode
5769 * @count: Number of blocks to swap
5770 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5771 * @erp: Pointer to save error value
5772 *
5773 * This helper routine does exactly what is promise "swap extents". All other
5774 * stuff such as page-cache locking consistency, bh mapping consistency or
5775 * extent's data copying must be performed by caller.
5776 * Locking:
5777 * i_mutex is held for both inodes
5778 * i_data_sem is locked for write for both inodes
5779 * Assumptions:
5780 * All pages from requested range are locked for both inodes
5781 */
5782int
5783ext4_swap_extents(handle_t *handle, struct inode *inode1,
5784 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5785 ext4_lblk_t count, int unwritten, int *erp)
5786{
5787 struct ext4_ext_path *path1 = NULL;
5788 struct ext4_ext_path *path2 = NULL;
5789 int replaced_count = 0;
5790
5791 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5792 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5793 BUG_ON(!inode_is_locked(inode1));
5794 BUG_ON(!inode_is_locked(inode2));
5795
5796 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5797 if (unlikely(*erp))
5798 return 0;
5799 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5800 if (unlikely(*erp))
5801 return 0;
5802
5803 while (count) {
5804 struct ext4_extent *ex1, *ex2, tmp_ex;
5805 ext4_lblk_t e1_blk, e2_blk;
5806 int e1_len, e2_len, len;
5807 int split = 0;
5808
5809 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5810 if (IS_ERR(path1)) {
5811 *erp = PTR_ERR(path1);
5812 path1 = NULL;
5813 finish:
5814 count = 0;
5815 goto repeat;
5816 }
5817 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5818 if (IS_ERR(path2)) {
5819 *erp = PTR_ERR(path2);
5820 path2 = NULL;
5821 goto finish;
5822 }
5823 ex1 = path1[path1->p_depth].p_ext;
5824 ex2 = path2[path2->p_depth].p_ext;
5825 /* Do we have somthing to swap ? */
5826 if (unlikely(!ex2 || !ex1))
5827 goto finish;
5828
5829 e1_blk = le32_to_cpu(ex1->ee_block);
5830 e2_blk = le32_to_cpu(ex2->ee_block);
5831 e1_len = ext4_ext_get_actual_len(ex1);
5832 e2_len = ext4_ext_get_actual_len(ex2);
5833
5834 /* Hole handling */
5835 if (!in_range(lblk1, e1_blk, e1_len) ||
5836 !in_range(lblk2, e2_blk, e2_len)) {
5837 ext4_lblk_t next1, next2;
5838
5839 /* if hole after extent, then go to next extent */
5840 next1 = ext4_ext_next_allocated_block(path1);
5841 next2 = ext4_ext_next_allocated_block(path2);
5842 /* If hole before extent, then shift to that extent */
5843 if (e1_blk > lblk1)
5844 next1 = e1_blk;
5845 if (e2_blk > lblk2)
5846 next2 = e1_blk;
5847 /* Do we have something to swap */
5848 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5849 goto finish;
5850 /* Move to the rightest boundary */
5851 len = next1 - lblk1;
5852 if (len < next2 - lblk2)
5853 len = next2 - lblk2;
5854 if (len > count)
5855 len = count;
5856 lblk1 += len;
5857 lblk2 += len;
5858 count -= len;
5859 goto repeat;
5860 }
5861
5862 /* Prepare left boundary */
5863 if (e1_blk < lblk1) {
5864 split = 1;
5865 *erp = ext4_force_split_extent_at(handle, inode1,
5866 &path1, lblk1, 0);
5867 if (unlikely(*erp))
5868 goto finish;
5869 }
5870 if (e2_blk < lblk2) {
5871 split = 1;
5872 *erp = ext4_force_split_extent_at(handle, inode2,
5873 &path2, lblk2, 0);
5874 if (unlikely(*erp))
5875 goto finish;
5876 }
5877 /* ext4_split_extent_at() may result in leaf extent split,
5878 * path must to be revalidated. */
5879 if (split)
5880 goto repeat;
5881
5882 /* Prepare right boundary */
5883 len = count;
5884 if (len > e1_blk + e1_len - lblk1)
5885 len = e1_blk + e1_len - lblk1;
5886 if (len > e2_blk + e2_len - lblk2)
5887 len = e2_blk + e2_len - lblk2;
5888
5889 if (len != e1_len) {
5890 split = 1;
5891 *erp = ext4_force_split_extent_at(handle, inode1,
5892 &path1, lblk1 + len, 0);
5893 if (unlikely(*erp))
5894 goto finish;
5895 }
5896 if (len != e2_len) {
5897 split = 1;
5898 *erp = ext4_force_split_extent_at(handle, inode2,
5899 &path2, lblk2 + len, 0);
5900 if (*erp)
5901 goto finish;
5902 }
5903 /* ext4_split_extent_at() may result in leaf extent split,
5904 * path must to be revalidated. */
5905 if (split)
5906 goto repeat;
5907
5908 BUG_ON(e2_len != e1_len);
5909 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5910 if (unlikely(*erp))
5911 goto finish;
5912 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5913 if (unlikely(*erp))
5914 goto finish;
5915
5916 /* Both extents are fully inside boundaries. Swap it now */
5917 tmp_ex = *ex1;
5918 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5919 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5920 ex1->ee_len = cpu_to_le16(e2_len);
5921 ex2->ee_len = cpu_to_le16(e1_len);
5922 if (unwritten)
5923 ext4_ext_mark_unwritten(ex2);
5924 if (ext4_ext_is_unwritten(&tmp_ex))
5925 ext4_ext_mark_unwritten(ex1);
5926
5927 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5928 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5929 *erp = ext4_ext_dirty(handle, inode2, path2 +
5930 path2->p_depth);
5931 if (unlikely(*erp))
5932 goto finish;
5933 *erp = ext4_ext_dirty(handle, inode1, path1 +
5934 path1->p_depth);
5935 /*
5936 * Looks scarry ah..? second inode already points to new blocks,
5937 * and it was successfully dirtied. But luckily error may happen
5938 * only due to journal error, so full transaction will be
5939 * aborted anyway.
5940 */
5941 if (unlikely(*erp))
5942 goto finish;
5943 lblk1 += len;
5944 lblk2 += len;
5945 replaced_count += len;
5946 count -= len;
5947
5948 repeat:
5949 ext4_ext_drop_refs(path1);
5950 kfree(path1);
5951 ext4_ext_drop_refs(path2);
5952 kfree(path2);
5953 path1 = path2 = NULL;
5954 }
5955 return replaced_count;
5956}
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/backing-dev.h>
31#include <linux/iomap.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
49static __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
61static 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
75static 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
87static 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
94static 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_mutex. So we can safely drop the i_data_sem here.
101 */
102 BUG_ON(EXT4_JOURNAL(inode) == NULL);
103 ext4_discard_preallocations(inode, 0);
104 up_write(&EXT4_I(inode)->i_data_sem);
105 *dropped = 1;
106 return 0;
107}
108
109/*
110 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
111 * transaction with 'restart_cred' credits. The function drops i_data_sem
112 * when restarting transaction and gets it after transaction is restarted.
113 *
114 * The function returns 0 on success, 1 if transaction had to be restarted,
115 * and < 0 in case of fatal error.
116 */
117int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
118 int check_cred, int restart_cred,
119 int revoke_cred)
120{
121 int ret;
122 int dropped = 0;
123
124 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
125 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
126 if (dropped)
127 down_write(&EXT4_I(inode)->i_data_sem);
128 return ret;
129}
130
131/*
132 * could return:
133 * - EROFS
134 * - ENOMEM
135 */
136static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
137 struct ext4_ext_path *path)
138{
139 if (path->p_bh) {
140 /* path points to block */
141 BUFFER_TRACE(path->p_bh, "get_write_access");
142 return ext4_journal_get_write_access(handle, path->p_bh);
143 }
144 /* path points to leaf/index in inode body */
145 /* we use in-core data, no need to protect them */
146 return 0;
147}
148
149/*
150 * could return:
151 * - EROFS
152 * - ENOMEM
153 * - EIO
154 */
155static int __ext4_ext_dirty(const char *where, unsigned int line,
156 handle_t *handle, struct inode *inode,
157 struct ext4_ext_path *path)
158{
159 int err;
160
161 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
162 if (path->p_bh) {
163 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
164 /* path points to block */
165 err = __ext4_handle_dirty_metadata(where, line, handle,
166 inode, path->p_bh);
167 } else {
168 /* path points to leaf/index in inode body */
169 err = ext4_mark_inode_dirty(handle, inode);
170 }
171 return err;
172}
173
174#define ext4_ext_dirty(handle, inode, path) \
175 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
176
177static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
178 struct ext4_ext_path *path,
179 ext4_lblk_t block)
180{
181 if (path) {
182 int depth = path->p_depth;
183 struct ext4_extent *ex;
184
185 /*
186 * Try to predict block placement assuming that we are
187 * filling in a file which will eventually be
188 * non-sparse --- i.e., in the case of libbfd writing
189 * an ELF object sections out-of-order but in a way
190 * the eventually results in a contiguous object or
191 * executable file, or some database extending a table
192 * space file. However, this is actually somewhat
193 * non-ideal if we are writing a sparse file such as
194 * qemu or KVM writing a raw image file that is going
195 * to stay fairly sparse, since it will end up
196 * fragmenting the file system's free space. Maybe we
197 * should have some hueristics or some way to allow
198 * userspace to pass a hint to file system,
199 * especially if the latter case turns out to be
200 * common.
201 */
202 ex = path[depth].p_ext;
203 if (ex) {
204 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
205 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
206
207 if (block > ext_block)
208 return ext_pblk + (block - ext_block);
209 else
210 return ext_pblk - (ext_block - block);
211 }
212
213 /* it looks like index is empty;
214 * try to find starting block from index itself */
215 if (path[depth].p_bh)
216 return path[depth].p_bh->b_blocknr;
217 }
218
219 /* OK. use inode's group */
220 return ext4_inode_to_goal_block(inode);
221}
222
223/*
224 * Allocation for a meta data block
225 */
226static ext4_fsblk_t
227ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
228 struct ext4_ext_path *path,
229 struct ext4_extent *ex, int *err, unsigned int flags)
230{
231 ext4_fsblk_t goal, newblock;
232
233 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
234 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
235 NULL, err);
236 return newblock;
237}
238
239static inline int ext4_ext_space_block(struct inode *inode, int check)
240{
241 int size;
242
243 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
244 / sizeof(struct ext4_extent);
245#ifdef AGGRESSIVE_TEST
246 if (!check && size > 6)
247 size = 6;
248#endif
249 return size;
250}
251
252static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
253{
254 int size;
255
256 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
257 / sizeof(struct ext4_extent_idx);
258#ifdef AGGRESSIVE_TEST
259 if (!check && size > 5)
260 size = 5;
261#endif
262 return size;
263}
264
265static inline int ext4_ext_space_root(struct inode *inode, int check)
266{
267 int size;
268
269 size = sizeof(EXT4_I(inode)->i_data);
270 size -= sizeof(struct ext4_extent_header);
271 size /= sizeof(struct ext4_extent);
272#ifdef AGGRESSIVE_TEST
273 if (!check && size > 3)
274 size = 3;
275#endif
276 return size;
277}
278
279static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
280{
281 int size;
282
283 size = sizeof(EXT4_I(inode)->i_data);
284 size -= sizeof(struct ext4_extent_header);
285 size /= sizeof(struct ext4_extent_idx);
286#ifdef AGGRESSIVE_TEST
287 if (!check && size > 4)
288 size = 4;
289#endif
290 return size;
291}
292
293static inline int
294ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
295 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
296 int nofail)
297{
298 struct ext4_ext_path *path = *ppath;
299 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
300 int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
301
302 if (nofail)
303 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
304
305 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
306 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
307 flags);
308}
309
310static int
311ext4_ext_max_entries(struct inode *inode, int depth)
312{
313 int max;
314
315 if (depth == ext_depth(inode)) {
316 if (depth == 0)
317 max = ext4_ext_space_root(inode, 1);
318 else
319 max = ext4_ext_space_root_idx(inode, 1);
320 } else {
321 if (depth == 0)
322 max = ext4_ext_space_block(inode, 1);
323 else
324 max = ext4_ext_space_block_idx(inode, 1);
325 }
326
327 return max;
328}
329
330static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
331{
332 ext4_fsblk_t block = ext4_ext_pblock(ext);
333 int len = ext4_ext_get_actual_len(ext);
334 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
335
336 /*
337 * We allow neither:
338 * - zero length
339 * - overflow/wrap-around
340 */
341 if (lblock + len <= lblock)
342 return 0;
343 return ext4_inode_block_valid(inode, block, len);
344}
345
346static int ext4_valid_extent_idx(struct inode *inode,
347 struct ext4_extent_idx *ext_idx)
348{
349 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
350
351 return ext4_inode_block_valid(inode, block, 1);
352}
353
354static int ext4_valid_extent_entries(struct inode *inode,
355 struct ext4_extent_header *eh,
356 ext4_fsblk_t *pblk, int depth)
357{
358 unsigned short entries;
359 if (eh->eh_entries == 0)
360 return 1;
361
362 entries = le16_to_cpu(eh->eh_entries);
363
364 if (depth == 0) {
365 /* leaf entries */
366 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
367 ext4_lblk_t lblock = 0;
368 ext4_lblk_t prev = 0;
369 int len = 0;
370 while (entries) {
371 if (!ext4_valid_extent(inode, ext))
372 return 0;
373
374 /* Check for overlapping extents */
375 lblock = le32_to_cpu(ext->ee_block);
376 len = ext4_ext_get_actual_len(ext);
377 if ((lblock <= prev) && prev) {
378 *pblk = ext4_ext_pblock(ext);
379 return 0;
380 }
381 ext++;
382 entries--;
383 prev = lblock + len - 1;
384 }
385 } else {
386 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
387 while (entries) {
388 if (!ext4_valid_extent_idx(inode, ext_idx))
389 return 0;
390 ext_idx++;
391 entries--;
392 }
393 }
394 return 1;
395}
396
397static int __ext4_ext_check(const char *function, unsigned int line,
398 struct inode *inode, struct ext4_extent_header *eh,
399 int depth, ext4_fsblk_t pblk)
400{
401 const char *error_msg;
402 int max = 0, err = -EFSCORRUPTED;
403
404 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
405 error_msg = "invalid magic";
406 goto corrupted;
407 }
408 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
409 error_msg = "unexpected eh_depth";
410 goto corrupted;
411 }
412 if (unlikely(eh->eh_max == 0)) {
413 error_msg = "invalid eh_max";
414 goto corrupted;
415 }
416 max = ext4_ext_max_entries(inode, depth);
417 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
418 error_msg = "too large eh_max";
419 goto corrupted;
420 }
421 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
422 error_msg = "invalid eh_entries";
423 goto corrupted;
424 }
425 if (!ext4_valid_extent_entries(inode, eh, &pblk, depth)) {
426 error_msg = "invalid extent entries";
427 goto corrupted;
428 }
429 if (unlikely(depth > 32)) {
430 error_msg = "too large eh_depth";
431 goto corrupted;
432 }
433 /* Verify checksum on non-root extent tree nodes */
434 if (ext_depth(inode) != depth &&
435 !ext4_extent_block_csum_verify(inode, eh)) {
436 error_msg = "extent tree corrupted";
437 err = -EFSBADCRC;
438 goto corrupted;
439 }
440 return 0;
441
442corrupted:
443 ext4_error_inode_err(inode, function, line, 0, -err,
444 "pblk %llu bad header/extent: %s - magic %x, "
445 "entries %u, max %u(%u), depth %u(%u)",
446 (unsigned long long) pblk, error_msg,
447 le16_to_cpu(eh->eh_magic),
448 le16_to_cpu(eh->eh_entries),
449 le16_to_cpu(eh->eh_max),
450 max, le16_to_cpu(eh->eh_depth), depth);
451 return err;
452}
453
454#define ext4_ext_check(inode, eh, depth, pblk) \
455 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
456
457int ext4_ext_check_inode(struct inode *inode)
458{
459 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
460}
461
462static void ext4_cache_extents(struct inode *inode,
463 struct ext4_extent_header *eh)
464{
465 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
466 ext4_lblk_t prev = 0;
467 int i;
468
469 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
470 unsigned int status = EXTENT_STATUS_WRITTEN;
471 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
472 int len = ext4_ext_get_actual_len(ex);
473
474 if (prev && (prev != lblk))
475 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
476 EXTENT_STATUS_HOLE);
477
478 if (ext4_ext_is_unwritten(ex))
479 status = EXTENT_STATUS_UNWRITTEN;
480 ext4_es_cache_extent(inode, lblk, len,
481 ext4_ext_pblock(ex), status);
482 prev = lblk + len;
483 }
484}
485
486static struct buffer_head *
487__read_extent_tree_block(const char *function, unsigned int line,
488 struct inode *inode, ext4_fsblk_t pblk, int depth,
489 int flags)
490{
491 struct buffer_head *bh;
492 int err;
493 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
494
495 if (flags & EXT4_EX_NOFAIL)
496 gfp_flags |= __GFP_NOFAIL;
497
498 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
499 if (unlikely(!bh))
500 return ERR_PTR(-ENOMEM);
501
502 if (!bh_uptodate_or_lock(bh)) {
503 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
504 err = bh_submit_read(bh);
505 if (err < 0)
506 goto errout;
507 }
508 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
509 return bh;
510 err = __ext4_ext_check(function, line, inode,
511 ext_block_hdr(bh), depth, pblk);
512 if (err)
513 goto errout;
514 set_buffer_verified(bh);
515 /*
516 * If this is a leaf block, cache all of its entries
517 */
518 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
519 struct ext4_extent_header *eh = ext_block_hdr(bh);
520 ext4_cache_extents(inode, eh);
521 }
522 return bh;
523errout:
524 put_bh(bh);
525 return ERR_PTR(err);
526
527}
528
529#define read_extent_tree_block(inode, pblk, depth, flags) \
530 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
531 (depth), (flags))
532
533/*
534 * This function is called to cache a file's extent information in the
535 * extent status tree
536 */
537int ext4_ext_precache(struct inode *inode)
538{
539 struct ext4_inode_info *ei = EXT4_I(inode);
540 struct ext4_ext_path *path = NULL;
541 struct buffer_head *bh;
542 int i = 0, depth, ret = 0;
543
544 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
545 return 0; /* not an extent-mapped inode */
546
547 down_read(&ei->i_data_sem);
548 depth = ext_depth(inode);
549
550 /* Don't cache anything if there are no external extent blocks */
551 if (!depth) {
552 up_read(&ei->i_data_sem);
553 return ret;
554 }
555
556 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
557 GFP_NOFS);
558 if (path == NULL) {
559 up_read(&ei->i_data_sem);
560 return -ENOMEM;
561 }
562
563 path[0].p_hdr = ext_inode_hdr(inode);
564 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
565 if (ret)
566 goto out;
567 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
568 while (i >= 0) {
569 /*
570 * If this is a leaf block or we've reached the end of
571 * the index block, go up
572 */
573 if ((i == depth) ||
574 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
575 brelse(path[i].p_bh);
576 path[i].p_bh = NULL;
577 i--;
578 continue;
579 }
580 bh = read_extent_tree_block(inode,
581 ext4_idx_pblock(path[i].p_idx++),
582 depth - i - 1,
583 EXT4_EX_FORCE_CACHE);
584 if (IS_ERR(bh)) {
585 ret = PTR_ERR(bh);
586 break;
587 }
588 i++;
589 path[i].p_bh = bh;
590 path[i].p_hdr = ext_block_hdr(bh);
591 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
592 }
593 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
594out:
595 up_read(&ei->i_data_sem);
596 ext4_ext_drop_refs(path);
597 kfree(path);
598 return ret;
599}
600
601#ifdef EXT_DEBUG
602static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
603{
604 int k, l = path->p_depth;
605
606 ext_debug(inode, "path:");
607 for (k = 0; k <= l; k++, path++) {
608 if (path->p_idx) {
609 ext_debug(inode, " %d->%llu",
610 le32_to_cpu(path->p_idx->ei_block),
611 ext4_idx_pblock(path->p_idx));
612 } else if (path->p_ext) {
613 ext_debug(inode, " %d:[%d]%d:%llu ",
614 le32_to_cpu(path->p_ext->ee_block),
615 ext4_ext_is_unwritten(path->p_ext),
616 ext4_ext_get_actual_len(path->p_ext),
617 ext4_ext_pblock(path->p_ext));
618 } else
619 ext_debug(inode, " []");
620 }
621 ext_debug(inode, "\n");
622}
623
624static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
625{
626 int depth = ext_depth(inode);
627 struct ext4_extent_header *eh;
628 struct ext4_extent *ex;
629 int i;
630
631 if (!path)
632 return;
633
634 eh = path[depth].p_hdr;
635 ex = EXT_FIRST_EXTENT(eh);
636
637 ext_debug(inode, "Displaying leaf extents\n");
638
639 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
640 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
641 ext4_ext_is_unwritten(ex),
642 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
643 }
644 ext_debug(inode, "\n");
645}
646
647static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
648 ext4_fsblk_t newblock, int level)
649{
650 int depth = ext_depth(inode);
651 struct ext4_extent *ex;
652
653 if (depth != level) {
654 struct ext4_extent_idx *idx;
655 idx = path[level].p_idx;
656 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
657 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
658 level, le32_to_cpu(idx->ei_block),
659 ext4_idx_pblock(idx), newblock);
660 idx++;
661 }
662
663 return;
664 }
665
666 ex = path[depth].p_ext;
667 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
668 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
669 le32_to_cpu(ex->ee_block),
670 ext4_ext_pblock(ex),
671 ext4_ext_is_unwritten(ex),
672 ext4_ext_get_actual_len(ex),
673 newblock);
674 ex++;
675 }
676}
677
678#else
679#define ext4_ext_show_path(inode, path)
680#define ext4_ext_show_leaf(inode, path)
681#define ext4_ext_show_move(inode, path, newblock, level)
682#endif
683
684void ext4_ext_drop_refs(struct ext4_ext_path *path)
685{
686 int depth, i;
687
688 if (!path)
689 return;
690 depth = path->p_depth;
691 for (i = 0; i <= depth; i++, path++) {
692 brelse(path->p_bh);
693 path->p_bh = NULL;
694 }
695}
696
697/*
698 * ext4_ext_binsearch_idx:
699 * binary search for the closest index of the given block
700 * the header must be checked before calling this
701 */
702static void
703ext4_ext_binsearch_idx(struct inode *inode,
704 struct ext4_ext_path *path, ext4_lblk_t block)
705{
706 struct ext4_extent_header *eh = path->p_hdr;
707 struct ext4_extent_idx *r, *l, *m;
708
709
710 ext_debug(inode, "binsearch for %u(idx): ", block);
711
712 l = EXT_FIRST_INDEX(eh) + 1;
713 r = EXT_LAST_INDEX(eh);
714 while (l <= r) {
715 m = l + (r - l) / 2;
716 if (block < le32_to_cpu(m->ei_block))
717 r = m - 1;
718 else
719 l = m + 1;
720 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
721 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
722 r, le32_to_cpu(r->ei_block));
723 }
724
725 path->p_idx = l - 1;
726 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
727 ext4_idx_pblock(path->p_idx));
728
729#ifdef CHECK_BINSEARCH
730 {
731 struct ext4_extent_idx *chix, *ix;
732 int k;
733
734 chix = ix = EXT_FIRST_INDEX(eh);
735 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
736 if (k != 0 && le32_to_cpu(ix->ei_block) <=
737 le32_to_cpu(ix[-1].ei_block)) {
738 printk(KERN_DEBUG "k=%d, ix=0x%p, "
739 "first=0x%p\n", k,
740 ix, EXT_FIRST_INDEX(eh));
741 printk(KERN_DEBUG "%u <= %u\n",
742 le32_to_cpu(ix->ei_block),
743 le32_to_cpu(ix[-1].ei_block));
744 }
745 BUG_ON(k && le32_to_cpu(ix->ei_block)
746 <= le32_to_cpu(ix[-1].ei_block));
747 if (block < le32_to_cpu(ix->ei_block))
748 break;
749 chix = ix;
750 }
751 BUG_ON(chix != path->p_idx);
752 }
753#endif
754
755}
756
757/*
758 * ext4_ext_binsearch:
759 * binary search for closest extent of the given block
760 * the header must be checked before calling this
761 */
762static void
763ext4_ext_binsearch(struct inode *inode,
764 struct ext4_ext_path *path, ext4_lblk_t block)
765{
766 struct ext4_extent_header *eh = path->p_hdr;
767 struct ext4_extent *r, *l, *m;
768
769 if (eh->eh_entries == 0) {
770 /*
771 * this leaf is empty:
772 * we get such a leaf in split/add case
773 */
774 return;
775 }
776
777 ext_debug(inode, "binsearch for %u: ", block);
778
779 l = EXT_FIRST_EXTENT(eh) + 1;
780 r = EXT_LAST_EXTENT(eh);
781
782 while (l <= r) {
783 m = l + (r - l) / 2;
784 if (block < le32_to_cpu(m->ee_block))
785 r = m - 1;
786 else
787 l = m + 1;
788 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
789 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
790 r, le32_to_cpu(r->ee_block));
791 }
792
793 path->p_ext = l - 1;
794 ext_debug(inode, " -> %d:%llu:[%d]%d ",
795 le32_to_cpu(path->p_ext->ee_block),
796 ext4_ext_pblock(path->p_ext),
797 ext4_ext_is_unwritten(path->p_ext),
798 ext4_ext_get_actual_len(path->p_ext));
799
800#ifdef CHECK_BINSEARCH
801 {
802 struct ext4_extent *chex, *ex;
803 int k;
804
805 chex = ex = EXT_FIRST_EXTENT(eh);
806 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
807 BUG_ON(k && le32_to_cpu(ex->ee_block)
808 <= le32_to_cpu(ex[-1].ee_block));
809 if (block < le32_to_cpu(ex->ee_block))
810 break;
811 chex = ex;
812 }
813 BUG_ON(chex != path->p_ext);
814 }
815#endif
816
817}
818
819void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
820{
821 struct ext4_extent_header *eh;
822
823 eh = ext_inode_hdr(inode);
824 eh->eh_depth = 0;
825 eh->eh_entries = 0;
826 eh->eh_magic = EXT4_EXT_MAGIC;
827 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
828 ext4_mark_inode_dirty(handle, inode);
829}
830
831struct ext4_ext_path *
832ext4_find_extent(struct inode *inode, ext4_lblk_t block,
833 struct ext4_ext_path **orig_path, int flags)
834{
835 struct ext4_extent_header *eh;
836 struct buffer_head *bh;
837 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
838 short int depth, i, ppos = 0;
839 int ret;
840 gfp_t gfp_flags = GFP_NOFS;
841
842 if (flags & EXT4_EX_NOFAIL)
843 gfp_flags |= __GFP_NOFAIL;
844
845 eh = ext_inode_hdr(inode);
846 depth = ext_depth(inode);
847 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
848 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
849 depth);
850 ret = -EFSCORRUPTED;
851 goto err;
852 }
853
854 if (path) {
855 ext4_ext_drop_refs(path);
856 if (depth > path[0].p_maxdepth) {
857 kfree(path);
858 *orig_path = path = NULL;
859 }
860 }
861 if (!path) {
862 /* account possible depth increase */
863 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
864 gfp_flags);
865 if (unlikely(!path))
866 return ERR_PTR(-ENOMEM);
867 path[0].p_maxdepth = depth + 1;
868 }
869 path[0].p_hdr = eh;
870 path[0].p_bh = NULL;
871
872 i = depth;
873 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
874 ext4_cache_extents(inode, eh);
875 /* walk through the tree */
876 while (i) {
877 ext_debug(inode, "depth %d: num %d, max %d\n",
878 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
879
880 ext4_ext_binsearch_idx(inode, path + ppos, block);
881 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
882 path[ppos].p_depth = i;
883 path[ppos].p_ext = NULL;
884
885 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
886 flags);
887 if (IS_ERR(bh)) {
888 ret = PTR_ERR(bh);
889 goto err;
890 }
891
892 eh = ext_block_hdr(bh);
893 ppos++;
894 path[ppos].p_bh = bh;
895 path[ppos].p_hdr = eh;
896 }
897
898 path[ppos].p_depth = i;
899 path[ppos].p_ext = NULL;
900 path[ppos].p_idx = NULL;
901
902 /* find extent */
903 ext4_ext_binsearch(inode, path + ppos, block);
904 /* if not an empty leaf */
905 if (path[ppos].p_ext)
906 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
907
908 ext4_ext_show_path(inode, path);
909
910 return path;
911
912err:
913 ext4_ext_drop_refs(path);
914 kfree(path);
915 if (orig_path)
916 *orig_path = NULL;
917 return ERR_PTR(ret);
918}
919
920/*
921 * ext4_ext_insert_index:
922 * insert new index [@logical;@ptr] into the block at @curp;
923 * check where to insert: before @curp or after @curp
924 */
925static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
926 struct ext4_ext_path *curp,
927 int logical, ext4_fsblk_t ptr)
928{
929 struct ext4_extent_idx *ix;
930 int len, err;
931
932 err = ext4_ext_get_access(handle, inode, curp);
933 if (err)
934 return err;
935
936 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
937 EXT4_ERROR_INODE(inode,
938 "logical %d == ei_block %d!",
939 logical, le32_to_cpu(curp->p_idx->ei_block));
940 return -EFSCORRUPTED;
941 }
942
943 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
944 >= le16_to_cpu(curp->p_hdr->eh_max))) {
945 EXT4_ERROR_INODE(inode,
946 "eh_entries %d >= eh_max %d!",
947 le16_to_cpu(curp->p_hdr->eh_entries),
948 le16_to_cpu(curp->p_hdr->eh_max));
949 return -EFSCORRUPTED;
950 }
951
952 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
953 /* insert after */
954 ext_debug(inode, "insert new index %d after: %llu\n",
955 logical, ptr);
956 ix = curp->p_idx + 1;
957 } else {
958 /* insert before */
959 ext_debug(inode, "insert new index %d before: %llu\n",
960 logical, ptr);
961 ix = curp->p_idx;
962 }
963
964 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
965 BUG_ON(len < 0);
966 if (len > 0) {
967 ext_debug(inode, "insert new index %d: "
968 "move %d indices from 0x%p to 0x%p\n",
969 logical, len, ix, ix + 1);
970 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
971 }
972
973 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
974 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
975 return -EFSCORRUPTED;
976 }
977
978 ix->ei_block = cpu_to_le32(logical);
979 ext4_idx_store_pblock(ix, ptr);
980 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
981
982 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
983 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
984 return -EFSCORRUPTED;
985 }
986
987 err = ext4_ext_dirty(handle, inode, curp);
988 ext4_std_error(inode->i_sb, err);
989
990 return err;
991}
992
993/*
994 * ext4_ext_split:
995 * inserts new subtree into the path, using free index entry
996 * at depth @at:
997 * - allocates all needed blocks (new leaf and all intermediate index blocks)
998 * - makes decision where to split
999 * - moves remaining extents and index entries (right to the split point)
1000 * into the newly allocated blocks
1001 * - initializes subtree
1002 */
1003static int ext4_ext_split(handle_t *handle, struct inode *inode,
1004 unsigned int flags,
1005 struct ext4_ext_path *path,
1006 struct ext4_extent *newext, int at)
1007{
1008 struct buffer_head *bh = NULL;
1009 int depth = ext_depth(inode);
1010 struct ext4_extent_header *neh;
1011 struct ext4_extent_idx *fidx;
1012 int i = at, k, m, a;
1013 ext4_fsblk_t newblock, oldblock;
1014 __le32 border;
1015 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1016 gfp_t gfp_flags = GFP_NOFS;
1017 int err = 0;
1018 size_t ext_size = 0;
1019
1020 if (flags & EXT4_EX_NOFAIL)
1021 gfp_flags |= __GFP_NOFAIL;
1022
1023 /* make decision: where to split? */
1024 /* FIXME: now decision is simplest: at current extent */
1025
1026 /* if current leaf will be split, then we should use
1027 * border from split point */
1028 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1029 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1030 return -EFSCORRUPTED;
1031 }
1032 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1033 border = path[depth].p_ext[1].ee_block;
1034 ext_debug(inode, "leaf will be split."
1035 " next leaf starts at %d\n",
1036 le32_to_cpu(border));
1037 } else {
1038 border = newext->ee_block;
1039 ext_debug(inode, "leaf will be added."
1040 " next leaf starts at %d\n",
1041 le32_to_cpu(border));
1042 }
1043
1044 /*
1045 * If error occurs, then we break processing
1046 * and mark filesystem read-only. index won't
1047 * be inserted and tree will be in consistent
1048 * state. Next mount will repair buffers too.
1049 */
1050
1051 /*
1052 * Get array to track all allocated blocks.
1053 * We need this to handle errors and free blocks
1054 * upon them.
1055 */
1056 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1057 if (!ablocks)
1058 return -ENOMEM;
1059
1060 /* allocate all needed blocks */
1061 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1062 for (a = 0; a < depth - at; a++) {
1063 newblock = ext4_ext_new_meta_block(handle, inode, path,
1064 newext, &err, flags);
1065 if (newblock == 0)
1066 goto cleanup;
1067 ablocks[a] = newblock;
1068 }
1069
1070 /* initialize new leaf */
1071 newblock = ablocks[--a];
1072 if (unlikely(newblock == 0)) {
1073 EXT4_ERROR_INODE(inode, "newblock == 0!");
1074 err = -EFSCORRUPTED;
1075 goto cleanup;
1076 }
1077 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1078 if (unlikely(!bh)) {
1079 err = -ENOMEM;
1080 goto cleanup;
1081 }
1082 lock_buffer(bh);
1083
1084 err = ext4_journal_get_create_access(handle, bh);
1085 if (err)
1086 goto cleanup;
1087
1088 neh = ext_block_hdr(bh);
1089 neh->eh_entries = 0;
1090 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1091 neh->eh_magic = EXT4_EXT_MAGIC;
1092 neh->eh_depth = 0;
1093
1094 /* move remainder of path[depth] to the new leaf */
1095 if (unlikely(path[depth].p_hdr->eh_entries !=
1096 path[depth].p_hdr->eh_max)) {
1097 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1098 path[depth].p_hdr->eh_entries,
1099 path[depth].p_hdr->eh_max);
1100 err = -EFSCORRUPTED;
1101 goto cleanup;
1102 }
1103 /* start copy from next extent */
1104 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1105 ext4_ext_show_move(inode, path, newblock, depth);
1106 if (m) {
1107 struct ext4_extent *ex;
1108 ex = EXT_FIRST_EXTENT(neh);
1109 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1110 le16_add_cpu(&neh->eh_entries, m);
1111 }
1112
1113 /* zero out unused area in the extent block */
1114 ext_size = sizeof(struct ext4_extent_header) +
1115 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1116 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1117 ext4_extent_block_csum_set(inode, neh);
1118 set_buffer_uptodate(bh);
1119 unlock_buffer(bh);
1120
1121 err = ext4_handle_dirty_metadata(handle, inode, bh);
1122 if (err)
1123 goto cleanup;
1124 brelse(bh);
1125 bh = NULL;
1126
1127 /* correct old leaf */
1128 if (m) {
1129 err = ext4_ext_get_access(handle, inode, path + depth);
1130 if (err)
1131 goto cleanup;
1132 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1133 err = ext4_ext_dirty(handle, inode, path + depth);
1134 if (err)
1135 goto cleanup;
1136
1137 }
1138
1139 /* create intermediate indexes */
1140 k = depth - at - 1;
1141 if (unlikely(k < 0)) {
1142 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1143 err = -EFSCORRUPTED;
1144 goto cleanup;
1145 }
1146 if (k)
1147 ext_debug(inode, "create %d intermediate indices\n", k);
1148 /* insert new index into current index block */
1149 /* current depth stored in i var */
1150 i = depth - 1;
1151 while (k--) {
1152 oldblock = newblock;
1153 newblock = ablocks[--a];
1154 bh = sb_getblk(inode->i_sb, newblock);
1155 if (unlikely(!bh)) {
1156 err = -ENOMEM;
1157 goto cleanup;
1158 }
1159 lock_buffer(bh);
1160
1161 err = ext4_journal_get_create_access(handle, bh);
1162 if (err)
1163 goto cleanup;
1164
1165 neh = ext_block_hdr(bh);
1166 neh->eh_entries = cpu_to_le16(1);
1167 neh->eh_magic = EXT4_EXT_MAGIC;
1168 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1169 neh->eh_depth = cpu_to_le16(depth - i);
1170 fidx = EXT_FIRST_INDEX(neh);
1171 fidx->ei_block = border;
1172 ext4_idx_store_pblock(fidx, oldblock);
1173
1174 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1175 i, newblock, le32_to_cpu(border), oldblock);
1176
1177 /* move remainder of path[i] to the new index block */
1178 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1179 EXT_LAST_INDEX(path[i].p_hdr))) {
1180 EXT4_ERROR_INODE(inode,
1181 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1182 le32_to_cpu(path[i].p_ext->ee_block));
1183 err = -EFSCORRUPTED;
1184 goto cleanup;
1185 }
1186 /* start copy indexes */
1187 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1188 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1189 EXT_MAX_INDEX(path[i].p_hdr));
1190 ext4_ext_show_move(inode, path, newblock, i);
1191 if (m) {
1192 memmove(++fidx, path[i].p_idx,
1193 sizeof(struct ext4_extent_idx) * m);
1194 le16_add_cpu(&neh->eh_entries, m);
1195 }
1196 /* zero out unused area in the extent block */
1197 ext_size = sizeof(struct ext4_extent_header) +
1198 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1199 memset(bh->b_data + ext_size, 0,
1200 inode->i_sb->s_blocksize - ext_size);
1201 ext4_extent_block_csum_set(inode, neh);
1202 set_buffer_uptodate(bh);
1203 unlock_buffer(bh);
1204
1205 err = ext4_handle_dirty_metadata(handle, inode, bh);
1206 if (err)
1207 goto cleanup;
1208 brelse(bh);
1209 bh = NULL;
1210
1211 /* correct old index */
1212 if (m) {
1213 err = ext4_ext_get_access(handle, inode, path + i);
1214 if (err)
1215 goto cleanup;
1216 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1217 err = ext4_ext_dirty(handle, inode, path + i);
1218 if (err)
1219 goto cleanup;
1220 }
1221
1222 i--;
1223 }
1224
1225 /* insert new index */
1226 err = ext4_ext_insert_index(handle, inode, path + at,
1227 le32_to_cpu(border), newblock);
1228
1229cleanup:
1230 if (bh) {
1231 if (buffer_locked(bh))
1232 unlock_buffer(bh);
1233 brelse(bh);
1234 }
1235
1236 if (err) {
1237 /* free all allocated blocks in error case */
1238 for (i = 0; i < depth; i++) {
1239 if (!ablocks[i])
1240 continue;
1241 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1242 EXT4_FREE_BLOCKS_METADATA);
1243 }
1244 }
1245 kfree(ablocks);
1246
1247 return err;
1248}
1249
1250/*
1251 * ext4_ext_grow_indepth:
1252 * implements tree growing procedure:
1253 * - allocates new block
1254 * - moves top-level data (index block or leaf) into the new block
1255 * - initializes new top-level, creating index that points to the
1256 * just created block
1257 */
1258static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1259 unsigned int flags)
1260{
1261 struct ext4_extent_header *neh;
1262 struct buffer_head *bh;
1263 ext4_fsblk_t newblock, goal = 0;
1264 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1265 int err = 0;
1266 size_t ext_size = 0;
1267
1268 /* Try to prepend new index to old one */
1269 if (ext_depth(inode))
1270 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1271 if (goal > le32_to_cpu(es->s_first_data_block)) {
1272 flags |= EXT4_MB_HINT_TRY_GOAL;
1273 goal--;
1274 } else
1275 goal = ext4_inode_to_goal_block(inode);
1276 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1277 NULL, &err);
1278 if (newblock == 0)
1279 return err;
1280
1281 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1282 if (unlikely(!bh))
1283 return -ENOMEM;
1284 lock_buffer(bh);
1285
1286 err = ext4_journal_get_create_access(handle, bh);
1287 if (err) {
1288 unlock_buffer(bh);
1289 goto out;
1290 }
1291
1292 ext_size = sizeof(EXT4_I(inode)->i_data);
1293 /* move top-level index/leaf into new block */
1294 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1295 /* zero out unused area in the extent block */
1296 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1297
1298 /* set size of new block */
1299 neh = ext_block_hdr(bh);
1300 /* old root could have indexes or leaves
1301 * so calculate e_max right way */
1302 if (ext_depth(inode))
1303 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1304 else
1305 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1306 neh->eh_magic = EXT4_EXT_MAGIC;
1307 ext4_extent_block_csum_set(inode, neh);
1308 set_buffer_uptodate(bh);
1309 unlock_buffer(bh);
1310
1311 err = ext4_handle_dirty_metadata(handle, inode, bh);
1312 if (err)
1313 goto out;
1314
1315 /* Update top-level index: num,max,pointer */
1316 neh = ext_inode_hdr(inode);
1317 neh->eh_entries = cpu_to_le16(1);
1318 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1319 if (neh->eh_depth == 0) {
1320 /* Root extent block becomes index block */
1321 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1322 EXT_FIRST_INDEX(neh)->ei_block =
1323 EXT_FIRST_EXTENT(neh)->ee_block;
1324 }
1325 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1326 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1327 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1328 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1329
1330 le16_add_cpu(&neh->eh_depth, 1);
1331 err = ext4_mark_inode_dirty(handle, inode);
1332out:
1333 brelse(bh);
1334
1335 return err;
1336}
1337
1338/*
1339 * ext4_ext_create_new_leaf:
1340 * finds empty index and adds new leaf.
1341 * if no free index is found, then it requests in-depth growing.
1342 */
1343static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1344 unsigned int mb_flags,
1345 unsigned int gb_flags,
1346 struct ext4_ext_path **ppath,
1347 struct ext4_extent *newext)
1348{
1349 struct ext4_ext_path *path = *ppath;
1350 struct ext4_ext_path *curp;
1351 int depth, i, err = 0;
1352
1353repeat:
1354 i = depth = ext_depth(inode);
1355
1356 /* walk up to the tree and look for free index entry */
1357 curp = path + depth;
1358 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1359 i--;
1360 curp--;
1361 }
1362
1363 /* we use already allocated block for index block,
1364 * so subsequent data blocks should be contiguous */
1365 if (EXT_HAS_FREE_INDEX(curp)) {
1366 /* if we found index with free entry, then use that
1367 * entry: create all needed subtree and add new leaf */
1368 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1369 if (err)
1370 goto out;
1371
1372 /* refill path */
1373 path = ext4_find_extent(inode,
1374 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1375 ppath, gb_flags);
1376 if (IS_ERR(path))
1377 err = PTR_ERR(path);
1378 } else {
1379 /* tree is full, time to grow in depth */
1380 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1381 if (err)
1382 goto out;
1383
1384 /* refill path */
1385 path = ext4_find_extent(inode,
1386 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1387 ppath, gb_flags);
1388 if (IS_ERR(path)) {
1389 err = PTR_ERR(path);
1390 goto out;
1391 }
1392
1393 /*
1394 * only first (depth 0 -> 1) produces free space;
1395 * in all other cases we have to split the grown tree
1396 */
1397 depth = ext_depth(inode);
1398 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1399 /* now we need to split */
1400 goto repeat;
1401 }
1402 }
1403
1404out:
1405 return err;
1406}
1407
1408/*
1409 * search the closest allocated block to the left for *logical
1410 * and returns it at @logical + it's physical address at @phys
1411 * if *logical is the smallest allocated block, the function
1412 * returns 0 at @phys
1413 * return value contains 0 (success) or error code
1414 */
1415static int ext4_ext_search_left(struct inode *inode,
1416 struct ext4_ext_path *path,
1417 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1418{
1419 struct ext4_extent_idx *ix;
1420 struct ext4_extent *ex;
1421 int depth, ee_len;
1422
1423 if (unlikely(path == NULL)) {
1424 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1425 return -EFSCORRUPTED;
1426 }
1427 depth = path->p_depth;
1428 *phys = 0;
1429
1430 if (depth == 0 && path->p_ext == NULL)
1431 return 0;
1432
1433 /* usually extent in the path covers blocks smaller
1434 * then *logical, but it can be that extent is the
1435 * first one in the file */
1436
1437 ex = path[depth].p_ext;
1438 ee_len = ext4_ext_get_actual_len(ex);
1439 if (*logical < le32_to_cpu(ex->ee_block)) {
1440 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1441 EXT4_ERROR_INODE(inode,
1442 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1443 *logical, le32_to_cpu(ex->ee_block));
1444 return -EFSCORRUPTED;
1445 }
1446 while (--depth >= 0) {
1447 ix = path[depth].p_idx;
1448 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1449 EXT4_ERROR_INODE(inode,
1450 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1451 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1452 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1453 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1454 depth);
1455 return -EFSCORRUPTED;
1456 }
1457 }
1458 return 0;
1459 }
1460
1461 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1462 EXT4_ERROR_INODE(inode,
1463 "logical %d < ee_block %d + ee_len %d!",
1464 *logical, le32_to_cpu(ex->ee_block), ee_len);
1465 return -EFSCORRUPTED;
1466 }
1467
1468 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1469 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1470 return 0;
1471}
1472
1473/*
1474 * search the closest allocated block to the right for *logical
1475 * and returns it at @logical + it's physical address at @phys
1476 * if *logical is the largest allocated block, the function
1477 * returns 0 at @phys
1478 * return value contains 0 (success) or error code
1479 */
1480static int ext4_ext_search_right(struct inode *inode,
1481 struct ext4_ext_path *path,
1482 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1483 struct ext4_extent **ret_ex)
1484{
1485 struct buffer_head *bh = NULL;
1486 struct ext4_extent_header *eh;
1487 struct ext4_extent_idx *ix;
1488 struct ext4_extent *ex;
1489 ext4_fsblk_t block;
1490 int depth; /* Note, NOT eh_depth; depth from top of tree */
1491 int ee_len;
1492
1493 if (unlikely(path == NULL)) {
1494 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1495 return -EFSCORRUPTED;
1496 }
1497 depth = path->p_depth;
1498 *phys = 0;
1499
1500 if (depth == 0 && path->p_ext == NULL)
1501 return 0;
1502
1503 /* usually extent in the path covers blocks smaller
1504 * then *logical, but it can be that extent is the
1505 * first one in the file */
1506
1507 ex = path[depth].p_ext;
1508 ee_len = ext4_ext_get_actual_len(ex);
1509 if (*logical < le32_to_cpu(ex->ee_block)) {
1510 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1511 EXT4_ERROR_INODE(inode,
1512 "first_extent(path[%d].p_hdr) != ex",
1513 depth);
1514 return -EFSCORRUPTED;
1515 }
1516 while (--depth >= 0) {
1517 ix = path[depth].p_idx;
1518 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1519 EXT4_ERROR_INODE(inode,
1520 "ix != EXT_FIRST_INDEX *logical %d!",
1521 *logical);
1522 return -EFSCORRUPTED;
1523 }
1524 }
1525 goto found_extent;
1526 }
1527
1528 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1529 EXT4_ERROR_INODE(inode,
1530 "logical %d < ee_block %d + ee_len %d!",
1531 *logical, le32_to_cpu(ex->ee_block), ee_len);
1532 return -EFSCORRUPTED;
1533 }
1534
1535 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1536 /* next allocated block in this leaf */
1537 ex++;
1538 goto found_extent;
1539 }
1540
1541 /* go up and search for index to the right */
1542 while (--depth >= 0) {
1543 ix = path[depth].p_idx;
1544 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1545 goto got_index;
1546 }
1547
1548 /* we've gone up to the root and found no index to the right */
1549 return 0;
1550
1551got_index:
1552 /* we've found index to the right, let's
1553 * follow it and find the closest allocated
1554 * block to the right */
1555 ix++;
1556 block = ext4_idx_pblock(ix);
1557 while (++depth < path->p_depth) {
1558 /* subtract from p_depth to get proper eh_depth */
1559 bh = read_extent_tree_block(inode, block,
1560 path->p_depth - depth, 0);
1561 if (IS_ERR(bh))
1562 return PTR_ERR(bh);
1563 eh = ext_block_hdr(bh);
1564 ix = EXT_FIRST_INDEX(eh);
1565 block = ext4_idx_pblock(ix);
1566 put_bh(bh);
1567 }
1568
1569 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1570 if (IS_ERR(bh))
1571 return PTR_ERR(bh);
1572 eh = ext_block_hdr(bh);
1573 ex = EXT_FIRST_EXTENT(eh);
1574found_extent:
1575 *logical = le32_to_cpu(ex->ee_block);
1576 *phys = ext4_ext_pblock(ex);
1577 *ret_ex = ex;
1578 if (bh)
1579 put_bh(bh);
1580 return 0;
1581}
1582
1583/*
1584 * ext4_ext_next_allocated_block:
1585 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1586 * NOTE: it considers block number from index entry as
1587 * allocated block. Thus, index entries have to be consistent
1588 * with leaves.
1589 */
1590ext4_lblk_t
1591ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1592{
1593 int depth;
1594
1595 BUG_ON(path == NULL);
1596 depth = path->p_depth;
1597
1598 if (depth == 0 && path->p_ext == NULL)
1599 return EXT_MAX_BLOCKS;
1600
1601 while (depth >= 0) {
1602 struct ext4_ext_path *p = &path[depth];
1603
1604 if (depth == path->p_depth) {
1605 /* leaf */
1606 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1607 return le32_to_cpu(p->p_ext[1].ee_block);
1608 } else {
1609 /* index */
1610 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1611 return le32_to_cpu(p->p_idx[1].ei_block);
1612 }
1613 depth--;
1614 }
1615
1616 return EXT_MAX_BLOCKS;
1617}
1618
1619/*
1620 * ext4_ext_next_leaf_block:
1621 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1622 */
1623static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1624{
1625 int depth;
1626
1627 BUG_ON(path == NULL);
1628 depth = path->p_depth;
1629
1630 /* zero-tree has no leaf blocks at all */
1631 if (depth == 0)
1632 return EXT_MAX_BLOCKS;
1633
1634 /* go to index block */
1635 depth--;
1636
1637 while (depth >= 0) {
1638 if (path[depth].p_idx !=
1639 EXT_LAST_INDEX(path[depth].p_hdr))
1640 return (ext4_lblk_t)
1641 le32_to_cpu(path[depth].p_idx[1].ei_block);
1642 depth--;
1643 }
1644
1645 return EXT_MAX_BLOCKS;
1646}
1647
1648/*
1649 * ext4_ext_correct_indexes:
1650 * if leaf gets modified and modified extent is first in the leaf,
1651 * then we have to correct all indexes above.
1652 * TODO: do we need to correct tree in all cases?
1653 */
1654static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1655 struct ext4_ext_path *path)
1656{
1657 struct ext4_extent_header *eh;
1658 int depth = ext_depth(inode);
1659 struct ext4_extent *ex;
1660 __le32 border;
1661 int k, err = 0;
1662
1663 eh = path[depth].p_hdr;
1664 ex = path[depth].p_ext;
1665
1666 if (unlikely(ex == NULL || eh == NULL)) {
1667 EXT4_ERROR_INODE(inode,
1668 "ex %p == NULL or eh %p == NULL", ex, eh);
1669 return -EFSCORRUPTED;
1670 }
1671
1672 if (depth == 0) {
1673 /* there is no tree at all */
1674 return 0;
1675 }
1676
1677 if (ex != EXT_FIRST_EXTENT(eh)) {
1678 /* we correct tree if first leaf got modified only */
1679 return 0;
1680 }
1681
1682 /*
1683 * TODO: we need correction if border is smaller than current one
1684 */
1685 k = depth - 1;
1686 border = path[depth].p_ext->ee_block;
1687 err = ext4_ext_get_access(handle, inode, path + k);
1688 if (err)
1689 return err;
1690 path[k].p_idx->ei_block = border;
1691 err = ext4_ext_dirty(handle, inode, path + k);
1692 if (err)
1693 return err;
1694
1695 while (k--) {
1696 /* change all left-side indexes */
1697 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1698 break;
1699 err = ext4_ext_get_access(handle, inode, path + k);
1700 if (err)
1701 break;
1702 path[k].p_idx->ei_block = border;
1703 err = ext4_ext_dirty(handle, inode, path + k);
1704 if (err)
1705 break;
1706 }
1707
1708 return err;
1709}
1710
1711static int ext4_can_extents_be_merged(struct inode *inode,
1712 struct ext4_extent *ex1,
1713 struct ext4_extent *ex2)
1714{
1715 unsigned short ext1_ee_len, ext2_ee_len;
1716
1717 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1718 return 0;
1719
1720 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1721 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1722
1723 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1724 le32_to_cpu(ex2->ee_block))
1725 return 0;
1726
1727 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1728 return 0;
1729
1730 if (ext4_ext_is_unwritten(ex1) &&
1731 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1732 return 0;
1733#ifdef AGGRESSIVE_TEST
1734 if (ext1_ee_len >= 4)
1735 return 0;
1736#endif
1737
1738 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1739 return 1;
1740 return 0;
1741}
1742
1743/*
1744 * This function tries to merge the "ex" extent to the next extent in the tree.
1745 * It always tries to merge towards right. If you want to merge towards
1746 * left, pass "ex - 1" as argument instead of "ex".
1747 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1748 * 1 if they got merged.
1749 */
1750static int ext4_ext_try_to_merge_right(struct inode *inode,
1751 struct ext4_ext_path *path,
1752 struct ext4_extent *ex)
1753{
1754 struct ext4_extent_header *eh;
1755 unsigned int depth, len;
1756 int merge_done = 0, unwritten;
1757
1758 depth = ext_depth(inode);
1759 BUG_ON(path[depth].p_hdr == NULL);
1760 eh = path[depth].p_hdr;
1761
1762 while (ex < EXT_LAST_EXTENT(eh)) {
1763 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1764 break;
1765 /* merge with next extent! */
1766 unwritten = ext4_ext_is_unwritten(ex);
1767 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1768 + ext4_ext_get_actual_len(ex + 1));
1769 if (unwritten)
1770 ext4_ext_mark_unwritten(ex);
1771
1772 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1773 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1774 * sizeof(struct ext4_extent);
1775 memmove(ex + 1, ex + 2, len);
1776 }
1777 le16_add_cpu(&eh->eh_entries, -1);
1778 merge_done = 1;
1779 WARN_ON(eh->eh_entries == 0);
1780 if (!eh->eh_entries)
1781 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1782 }
1783
1784 return merge_done;
1785}
1786
1787/*
1788 * This function does a very simple check to see if we can collapse
1789 * an extent tree with a single extent tree leaf block into the inode.
1790 */
1791static void ext4_ext_try_to_merge_up(handle_t *handle,
1792 struct inode *inode,
1793 struct ext4_ext_path *path)
1794{
1795 size_t s;
1796 unsigned max_root = ext4_ext_space_root(inode, 0);
1797 ext4_fsblk_t blk;
1798
1799 if ((path[0].p_depth != 1) ||
1800 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1801 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1802 return;
1803
1804 /*
1805 * We need to modify the block allocation bitmap and the block
1806 * group descriptor to release the extent tree block. If we
1807 * can't get the journal credits, give up.
1808 */
1809 if (ext4_journal_extend(handle, 2,
1810 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1811 return;
1812
1813 /*
1814 * Copy the extent data up to the inode
1815 */
1816 blk = ext4_idx_pblock(path[0].p_idx);
1817 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1818 sizeof(struct ext4_extent_idx);
1819 s += sizeof(struct ext4_extent_header);
1820
1821 path[1].p_maxdepth = path[0].p_maxdepth;
1822 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1823 path[0].p_depth = 0;
1824 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1825 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1826 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1827
1828 brelse(path[1].p_bh);
1829 ext4_free_blocks(handle, inode, NULL, blk, 1,
1830 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1831}
1832
1833/*
1834 * This function tries to merge the @ex extent to neighbours in the tree, then
1835 * tries to collapse the extent tree into the inode.
1836 */
1837static void ext4_ext_try_to_merge(handle_t *handle,
1838 struct inode *inode,
1839 struct ext4_ext_path *path,
1840 struct ext4_extent *ex)
1841{
1842 struct ext4_extent_header *eh;
1843 unsigned int depth;
1844 int merge_done = 0;
1845
1846 depth = ext_depth(inode);
1847 BUG_ON(path[depth].p_hdr == NULL);
1848 eh = path[depth].p_hdr;
1849
1850 if (ex > EXT_FIRST_EXTENT(eh))
1851 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1852
1853 if (!merge_done)
1854 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1855
1856 ext4_ext_try_to_merge_up(handle, inode, path);
1857}
1858
1859/*
1860 * check if a portion of the "newext" extent overlaps with an
1861 * existing extent.
1862 *
1863 * If there is an overlap discovered, it updates the length of the newext
1864 * such that there will be no overlap, and then returns 1.
1865 * If there is no overlap found, it returns 0.
1866 */
1867static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1868 struct inode *inode,
1869 struct ext4_extent *newext,
1870 struct ext4_ext_path *path)
1871{
1872 ext4_lblk_t b1, b2;
1873 unsigned int depth, len1;
1874 unsigned int ret = 0;
1875
1876 b1 = le32_to_cpu(newext->ee_block);
1877 len1 = ext4_ext_get_actual_len(newext);
1878 depth = ext_depth(inode);
1879 if (!path[depth].p_ext)
1880 goto out;
1881 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1882
1883 /*
1884 * get the next allocated block if the extent in the path
1885 * is before the requested block(s)
1886 */
1887 if (b2 < b1) {
1888 b2 = ext4_ext_next_allocated_block(path);
1889 if (b2 == EXT_MAX_BLOCKS)
1890 goto out;
1891 b2 = EXT4_LBLK_CMASK(sbi, b2);
1892 }
1893
1894 /* check for wrap through zero on extent logical start block*/
1895 if (b1 + len1 < b1) {
1896 len1 = EXT_MAX_BLOCKS - b1;
1897 newext->ee_len = cpu_to_le16(len1);
1898 ret = 1;
1899 }
1900
1901 /* check for overlap */
1902 if (b1 + len1 > b2) {
1903 newext->ee_len = cpu_to_le16(b2 - b1);
1904 ret = 1;
1905 }
1906out:
1907 return ret;
1908}
1909
1910/*
1911 * ext4_ext_insert_extent:
1912 * tries to merge requested extent into the existing extent or
1913 * inserts requested extent as new one into the tree,
1914 * creating new leaf in the no-space case.
1915 */
1916int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1917 struct ext4_ext_path **ppath,
1918 struct ext4_extent *newext, int gb_flags)
1919{
1920 struct ext4_ext_path *path = *ppath;
1921 struct ext4_extent_header *eh;
1922 struct ext4_extent *ex, *fex;
1923 struct ext4_extent *nearex; /* nearest extent */
1924 struct ext4_ext_path *npath = NULL;
1925 int depth, len, err;
1926 ext4_lblk_t next;
1927 int mb_flags = 0, unwritten;
1928
1929 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1930 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1931 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1932 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1933 return -EFSCORRUPTED;
1934 }
1935 depth = ext_depth(inode);
1936 ex = path[depth].p_ext;
1937 eh = path[depth].p_hdr;
1938 if (unlikely(path[depth].p_hdr == NULL)) {
1939 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1940 return -EFSCORRUPTED;
1941 }
1942
1943 /* try to insert block into found extent and return */
1944 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1945
1946 /*
1947 * Try to see whether we should rather test the extent on
1948 * right from ex, or from the left of ex. This is because
1949 * ext4_find_extent() can return either extent on the
1950 * left, or on the right from the searched position. This
1951 * will make merging more effective.
1952 */
1953 if (ex < EXT_LAST_EXTENT(eh) &&
1954 (le32_to_cpu(ex->ee_block) +
1955 ext4_ext_get_actual_len(ex) <
1956 le32_to_cpu(newext->ee_block))) {
1957 ex += 1;
1958 goto prepend;
1959 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1960 (le32_to_cpu(newext->ee_block) +
1961 ext4_ext_get_actual_len(newext) <
1962 le32_to_cpu(ex->ee_block)))
1963 ex -= 1;
1964
1965 /* Try to append newex to the ex */
1966 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1967 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
1968 "(from %llu)\n",
1969 ext4_ext_is_unwritten(newext),
1970 ext4_ext_get_actual_len(newext),
1971 le32_to_cpu(ex->ee_block),
1972 ext4_ext_is_unwritten(ex),
1973 ext4_ext_get_actual_len(ex),
1974 ext4_ext_pblock(ex));
1975 err = ext4_ext_get_access(handle, inode,
1976 path + depth);
1977 if (err)
1978 return err;
1979 unwritten = ext4_ext_is_unwritten(ex);
1980 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1981 + ext4_ext_get_actual_len(newext));
1982 if (unwritten)
1983 ext4_ext_mark_unwritten(ex);
1984 eh = path[depth].p_hdr;
1985 nearex = ex;
1986 goto merge;
1987 }
1988
1989prepend:
1990 /* Try to prepend newex to the ex */
1991 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1992 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
1993 "(from %llu)\n",
1994 le32_to_cpu(newext->ee_block),
1995 ext4_ext_is_unwritten(newext),
1996 ext4_ext_get_actual_len(newext),
1997 le32_to_cpu(ex->ee_block),
1998 ext4_ext_is_unwritten(ex),
1999 ext4_ext_get_actual_len(ex),
2000 ext4_ext_pblock(ex));
2001 err = ext4_ext_get_access(handle, inode,
2002 path + depth);
2003 if (err)
2004 return err;
2005
2006 unwritten = ext4_ext_is_unwritten(ex);
2007 ex->ee_block = newext->ee_block;
2008 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2009 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2010 + ext4_ext_get_actual_len(newext));
2011 if (unwritten)
2012 ext4_ext_mark_unwritten(ex);
2013 eh = path[depth].p_hdr;
2014 nearex = ex;
2015 goto merge;
2016 }
2017 }
2018
2019 depth = ext_depth(inode);
2020 eh = path[depth].p_hdr;
2021 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2022 goto has_space;
2023
2024 /* probably next leaf has space for us? */
2025 fex = EXT_LAST_EXTENT(eh);
2026 next = EXT_MAX_BLOCKS;
2027 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2028 next = ext4_ext_next_leaf_block(path);
2029 if (next != EXT_MAX_BLOCKS) {
2030 ext_debug(inode, "next leaf block - %u\n", next);
2031 BUG_ON(npath != NULL);
2032 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2033 if (IS_ERR(npath))
2034 return PTR_ERR(npath);
2035 BUG_ON(npath->p_depth != path->p_depth);
2036 eh = npath[depth].p_hdr;
2037 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2038 ext_debug(inode, "next leaf isn't full(%d)\n",
2039 le16_to_cpu(eh->eh_entries));
2040 path = npath;
2041 goto has_space;
2042 }
2043 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2044 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2045 }
2046
2047 /*
2048 * There is no free space in the found leaf.
2049 * We're gonna add a new leaf in the tree.
2050 */
2051 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2052 mb_flags |= EXT4_MB_USE_RESERVED;
2053 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2054 ppath, newext);
2055 if (err)
2056 goto cleanup;
2057 depth = ext_depth(inode);
2058 eh = path[depth].p_hdr;
2059
2060has_space:
2061 nearex = path[depth].p_ext;
2062
2063 err = ext4_ext_get_access(handle, inode, path + depth);
2064 if (err)
2065 goto cleanup;
2066
2067 if (!nearex) {
2068 /* there is no extent in this leaf, create first one */
2069 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2070 le32_to_cpu(newext->ee_block),
2071 ext4_ext_pblock(newext),
2072 ext4_ext_is_unwritten(newext),
2073 ext4_ext_get_actual_len(newext));
2074 nearex = EXT_FIRST_EXTENT(eh);
2075 } else {
2076 if (le32_to_cpu(newext->ee_block)
2077 > le32_to_cpu(nearex->ee_block)) {
2078 /* Insert after */
2079 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2080 "nearest %p\n",
2081 le32_to_cpu(newext->ee_block),
2082 ext4_ext_pblock(newext),
2083 ext4_ext_is_unwritten(newext),
2084 ext4_ext_get_actual_len(newext),
2085 nearex);
2086 nearex++;
2087 } else {
2088 /* Insert before */
2089 BUG_ON(newext->ee_block == nearex->ee_block);
2090 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2091 "nearest %p\n",
2092 le32_to_cpu(newext->ee_block),
2093 ext4_ext_pblock(newext),
2094 ext4_ext_is_unwritten(newext),
2095 ext4_ext_get_actual_len(newext),
2096 nearex);
2097 }
2098 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2099 if (len > 0) {
2100 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2101 "move %d extents from 0x%p to 0x%p\n",
2102 le32_to_cpu(newext->ee_block),
2103 ext4_ext_pblock(newext),
2104 ext4_ext_is_unwritten(newext),
2105 ext4_ext_get_actual_len(newext),
2106 len, nearex, nearex + 1);
2107 memmove(nearex + 1, nearex,
2108 len * sizeof(struct ext4_extent));
2109 }
2110 }
2111
2112 le16_add_cpu(&eh->eh_entries, 1);
2113 path[depth].p_ext = nearex;
2114 nearex->ee_block = newext->ee_block;
2115 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2116 nearex->ee_len = newext->ee_len;
2117
2118merge:
2119 /* try to merge extents */
2120 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2121 ext4_ext_try_to_merge(handle, inode, path, nearex);
2122
2123
2124 /* time to correct all indexes above */
2125 err = ext4_ext_correct_indexes(handle, inode, path);
2126 if (err)
2127 goto cleanup;
2128
2129 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2130
2131cleanup:
2132 ext4_ext_drop_refs(npath);
2133 kfree(npath);
2134 return err;
2135}
2136
2137static int ext4_fill_es_cache_info(struct inode *inode,
2138 ext4_lblk_t block, ext4_lblk_t num,
2139 struct fiemap_extent_info *fieinfo)
2140{
2141 ext4_lblk_t next, end = block + num - 1;
2142 struct extent_status es;
2143 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2144 unsigned int flags;
2145 int err;
2146
2147 while (block <= end) {
2148 next = 0;
2149 flags = 0;
2150 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2151 break;
2152 if (ext4_es_is_unwritten(&es))
2153 flags |= FIEMAP_EXTENT_UNWRITTEN;
2154 if (ext4_es_is_delayed(&es))
2155 flags |= (FIEMAP_EXTENT_DELALLOC |
2156 FIEMAP_EXTENT_UNKNOWN);
2157 if (ext4_es_is_hole(&es))
2158 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2159 if (next == 0)
2160 flags |= FIEMAP_EXTENT_LAST;
2161 if (flags & (FIEMAP_EXTENT_DELALLOC|
2162 EXT4_FIEMAP_EXTENT_HOLE))
2163 es.es_pblk = 0;
2164 else
2165 es.es_pblk = ext4_es_pblock(&es);
2166 err = fiemap_fill_next_extent(fieinfo,
2167 (__u64)es.es_lblk << blksize_bits,
2168 (__u64)es.es_pblk << blksize_bits,
2169 (__u64)es.es_len << blksize_bits,
2170 flags);
2171 if (next == 0)
2172 break;
2173 block = next;
2174 if (err < 0)
2175 return err;
2176 if (err == 1)
2177 return 0;
2178 }
2179 return 0;
2180}
2181
2182
2183/*
2184 * ext4_ext_determine_hole - determine hole around given block
2185 * @inode: inode we lookup in
2186 * @path: path in extent tree to @lblk
2187 * @lblk: pointer to logical block around which we want to determine hole
2188 *
2189 * Determine hole length (and start if easily possible) around given logical
2190 * block. We don't try too hard to find the beginning of the hole but @path
2191 * actually points to extent before @lblk, we provide it.
2192 *
2193 * The function returns the length of a hole starting at @lblk. We update @lblk
2194 * to the beginning of the hole if we managed to find it.
2195 */
2196static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2197 struct ext4_ext_path *path,
2198 ext4_lblk_t *lblk)
2199{
2200 int depth = ext_depth(inode);
2201 struct ext4_extent *ex;
2202 ext4_lblk_t len;
2203
2204 ex = path[depth].p_ext;
2205 if (ex == NULL) {
2206 /* there is no extent yet, so gap is [0;-] */
2207 *lblk = 0;
2208 len = EXT_MAX_BLOCKS;
2209 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2210 len = le32_to_cpu(ex->ee_block) - *lblk;
2211 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2212 + ext4_ext_get_actual_len(ex)) {
2213 ext4_lblk_t next;
2214
2215 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2216 next = ext4_ext_next_allocated_block(path);
2217 BUG_ON(next == *lblk);
2218 len = next - *lblk;
2219 } else {
2220 BUG();
2221 }
2222 return len;
2223}
2224
2225/*
2226 * ext4_ext_put_gap_in_cache:
2227 * calculate boundaries of the gap that the requested block fits into
2228 * and cache this gap
2229 */
2230static void
2231ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2232 ext4_lblk_t hole_len)
2233{
2234 struct extent_status es;
2235
2236 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2237 hole_start + hole_len - 1, &es);
2238 if (es.es_len) {
2239 /* There's delayed extent containing lblock? */
2240 if (es.es_lblk <= hole_start)
2241 return;
2242 hole_len = min(es.es_lblk - hole_start, hole_len);
2243 }
2244 ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2245 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2246 EXTENT_STATUS_HOLE);
2247}
2248
2249/*
2250 * ext4_ext_rm_idx:
2251 * removes index from the index block.
2252 */
2253static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2254 struct ext4_ext_path *path, int depth)
2255{
2256 int err;
2257 ext4_fsblk_t leaf;
2258
2259 /* free index block */
2260 depth--;
2261 path = path + depth;
2262 leaf = ext4_idx_pblock(path->p_idx);
2263 if (unlikely(path->p_hdr->eh_entries == 0)) {
2264 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2265 return -EFSCORRUPTED;
2266 }
2267 err = ext4_ext_get_access(handle, inode, path);
2268 if (err)
2269 return err;
2270
2271 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2272 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2273 len *= sizeof(struct ext4_extent_idx);
2274 memmove(path->p_idx, path->p_idx + 1, len);
2275 }
2276
2277 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2278 err = ext4_ext_dirty(handle, inode, path);
2279 if (err)
2280 return err;
2281 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2282 trace_ext4_ext_rm_idx(inode, leaf);
2283
2284 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2285 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2286
2287 while (--depth >= 0) {
2288 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2289 break;
2290 path--;
2291 err = ext4_ext_get_access(handle, inode, path);
2292 if (err)
2293 break;
2294 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2295 err = ext4_ext_dirty(handle, inode, path);
2296 if (err)
2297 break;
2298 }
2299 return err;
2300}
2301
2302/*
2303 * ext4_ext_calc_credits_for_single_extent:
2304 * This routine returns max. credits that needed to insert an extent
2305 * to the extent tree.
2306 * When pass the actual path, the caller should calculate credits
2307 * under i_data_sem.
2308 */
2309int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2310 struct ext4_ext_path *path)
2311{
2312 if (path) {
2313 int depth = ext_depth(inode);
2314 int ret = 0;
2315
2316 /* probably there is space in leaf? */
2317 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2318 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2319
2320 /*
2321 * There are some space in the leaf tree, no
2322 * need to account for leaf block credit
2323 *
2324 * bitmaps and block group descriptor blocks
2325 * and other metadata blocks still need to be
2326 * accounted.
2327 */
2328 /* 1 bitmap, 1 block group descriptor */
2329 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2330 return ret;
2331 }
2332 }
2333
2334 return ext4_chunk_trans_blocks(inode, nrblocks);
2335}
2336
2337/*
2338 * How many index/leaf blocks need to change/allocate to add @extents extents?
2339 *
2340 * If we add a single extent, then in the worse case, each tree level
2341 * index/leaf need to be changed in case of the tree split.
2342 *
2343 * If more extents are inserted, they could cause the whole tree split more
2344 * than once, but this is really rare.
2345 */
2346int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2347{
2348 int index;
2349 int depth;
2350
2351 /* If we are converting the inline data, only one is needed here. */
2352 if (ext4_has_inline_data(inode))
2353 return 1;
2354
2355 depth = ext_depth(inode);
2356
2357 if (extents <= 1)
2358 index = depth * 2;
2359 else
2360 index = depth * 3;
2361
2362 return index;
2363}
2364
2365static inline int get_default_free_blocks_flags(struct inode *inode)
2366{
2367 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2368 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2369 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2370 else if (ext4_should_journal_data(inode))
2371 return EXT4_FREE_BLOCKS_FORGET;
2372 return 0;
2373}
2374
2375/*
2376 * ext4_rereserve_cluster - increment the reserved cluster count when
2377 * freeing a cluster with a pending reservation
2378 *
2379 * @inode - file containing the cluster
2380 * @lblk - logical block in cluster to be reserved
2381 *
2382 * Increments the reserved cluster count and adjusts quota in a bigalloc
2383 * file system when freeing a partial cluster containing at least one
2384 * delayed and unwritten block. A partial cluster meeting that
2385 * requirement will have a pending reservation. If so, the
2386 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2387 * defer reserved and allocated space accounting to a subsequent call
2388 * to this function.
2389 */
2390static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2391{
2392 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2393 struct ext4_inode_info *ei = EXT4_I(inode);
2394
2395 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2396
2397 spin_lock(&ei->i_block_reservation_lock);
2398 ei->i_reserved_data_blocks++;
2399 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2400 spin_unlock(&ei->i_block_reservation_lock);
2401
2402 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2403 ext4_remove_pending(inode, lblk);
2404}
2405
2406static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2407 struct ext4_extent *ex,
2408 struct partial_cluster *partial,
2409 ext4_lblk_t from, ext4_lblk_t to)
2410{
2411 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2412 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2413 ext4_fsblk_t last_pblk, pblk;
2414 ext4_lblk_t num;
2415 int flags;
2416
2417 /* only extent tail removal is allowed */
2418 if (from < le32_to_cpu(ex->ee_block) ||
2419 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2420 ext4_error(sbi->s_sb,
2421 "strange request: removal(2) %u-%u from %u:%u",
2422 from, to, le32_to_cpu(ex->ee_block), ee_len);
2423 return 0;
2424 }
2425
2426#ifdef EXTENTS_STATS
2427 spin_lock(&sbi->s_ext_stats_lock);
2428 sbi->s_ext_blocks += ee_len;
2429 sbi->s_ext_extents++;
2430 if (ee_len < sbi->s_ext_min)
2431 sbi->s_ext_min = ee_len;
2432 if (ee_len > sbi->s_ext_max)
2433 sbi->s_ext_max = ee_len;
2434 if (ext_depth(inode) > sbi->s_depth_max)
2435 sbi->s_depth_max = ext_depth(inode);
2436 spin_unlock(&sbi->s_ext_stats_lock);
2437#endif
2438
2439 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2440
2441 /*
2442 * if we have a partial cluster, and it's different from the
2443 * cluster of the last block in the extent, we free it
2444 */
2445 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2446
2447 if (partial->state != initial &&
2448 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2449 if (partial->state == tofree) {
2450 flags = get_default_free_blocks_flags(inode);
2451 if (ext4_is_pending(inode, partial->lblk))
2452 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2453 ext4_free_blocks(handle, inode, NULL,
2454 EXT4_C2B(sbi, partial->pclu),
2455 sbi->s_cluster_ratio, flags);
2456 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2457 ext4_rereserve_cluster(inode, partial->lblk);
2458 }
2459 partial->state = initial;
2460 }
2461
2462 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2463 pblk = ext4_ext_pblock(ex) + ee_len - num;
2464
2465 /*
2466 * We free the partial cluster at the end of the extent (if any),
2467 * unless the cluster is used by another extent (partial_cluster
2468 * state is nofree). If a partial cluster exists here, it must be
2469 * shared with the last block in the extent.
2470 */
2471 flags = get_default_free_blocks_flags(inode);
2472
2473 /* partial, left end cluster aligned, right end unaligned */
2474 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2475 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2476 (partial->state != nofree)) {
2477 if (ext4_is_pending(inode, to))
2478 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2479 ext4_free_blocks(handle, inode, NULL,
2480 EXT4_PBLK_CMASK(sbi, last_pblk),
2481 sbi->s_cluster_ratio, flags);
2482 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2483 ext4_rereserve_cluster(inode, to);
2484 partial->state = initial;
2485 flags = get_default_free_blocks_flags(inode);
2486 }
2487
2488 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2489
2490 /*
2491 * For bigalloc file systems, we never free a partial cluster
2492 * at the beginning of the extent. Instead, we check to see if we
2493 * need to free it on a subsequent call to ext4_remove_blocks,
2494 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2495 */
2496 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2497 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2498
2499 /* reset the partial cluster if we've freed past it */
2500 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2501 partial->state = initial;
2502
2503 /*
2504 * If we've freed the entire extent but the beginning is not left
2505 * cluster aligned and is not marked as ineligible for freeing we
2506 * record the partial cluster at the beginning of the extent. It
2507 * wasn't freed by the preceding ext4_free_blocks() call, and we
2508 * need to look farther to the left to determine if it's to be freed
2509 * (not shared with another extent). Else, reset the partial
2510 * cluster - we're either done freeing or the beginning of the
2511 * extent is left cluster aligned.
2512 */
2513 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2514 if (partial->state == initial) {
2515 partial->pclu = EXT4_B2C(sbi, pblk);
2516 partial->lblk = from;
2517 partial->state = tofree;
2518 }
2519 } else {
2520 partial->state = initial;
2521 }
2522
2523 return 0;
2524}
2525
2526/*
2527 * ext4_ext_rm_leaf() Removes the extents associated with the
2528 * blocks appearing between "start" and "end". Both "start"
2529 * and "end" must appear in the same extent or EIO is returned.
2530 *
2531 * @handle: The journal handle
2532 * @inode: The files inode
2533 * @path: The path to the leaf
2534 * @partial_cluster: The cluster which we'll have to free if all extents
2535 * has been released from it. However, if this value is
2536 * negative, it's a cluster just to the right of the
2537 * punched region and it must not be freed.
2538 * @start: The first block to remove
2539 * @end: The last block to remove
2540 */
2541static int
2542ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2543 struct ext4_ext_path *path,
2544 struct partial_cluster *partial,
2545 ext4_lblk_t start, ext4_lblk_t end)
2546{
2547 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2548 int err = 0, correct_index = 0;
2549 int depth = ext_depth(inode), credits, revoke_credits;
2550 struct ext4_extent_header *eh;
2551 ext4_lblk_t a, b;
2552 unsigned num;
2553 ext4_lblk_t ex_ee_block;
2554 unsigned short ex_ee_len;
2555 unsigned unwritten = 0;
2556 struct ext4_extent *ex;
2557 ext4_fsblk_t pblk;
2558
2559 /* the header must be checked already in ext4_ext_remove_space() */
2560 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2561 if (!path[depth].p_hdr)
2562 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2563 eh = path[depth].p_hdr;
2564 if (unlikely(path[depth].p_hdr == NULL)) {
2565 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2566 return -EFSCORRUPTED;
2567 }
2568 /* find where to start removing */
2569 ex = path[depth].p_ext;
2570 if (!ex)
2571 ex = EXT_LAST_EXTENT(eh);
2572
2573 ex_ee_block = le32_to_cpu(ex->ee_block);
2574 ex_ee_len = ext4_ext_get_actual_len(ex);
2575
2576 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2577
2578 while (ex >= EXT_FIRST_EXTENT(eh) &&
2579 ex_ee_block + ex_ee_len > start) {
2580
2581 if (ext4_ext_is_unwritten(ex))
2582 unwritten = 1;
2583 else
2584 unwritten = 0;
2585
2586 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2587 unwritten, ex_ee_len);
2588 path[depth].p_ext = ex;
2589
2590 a = ex_ee_block > start ? ex_ee_block : start;
2591 b = ex_ee_block+ex_ee_len - 1 < end ?
2592 ex_ee_block+ex_ee_len - 1 : end;
2593
2594 ext_debug(inode, " border %u:%u\n", a, b);
2595
2596 /* If this extent is beyond the end of the hole, skip it */
2597 if (end < ex_ee_block) {
2598 /*
2599 * We're going to skip this extent and move to another,
2600 * so note that its first cluster is in use to avoid
2601 * freeing it when removing blocks. Eventually, the
2602 * right edge of the truncated/punched region will
2603 * be just to the left.
2604 */
2605 if (sbi->s_cluster_ratio > 1) {
2606 pblk = ext4_ext_pblock(ex);
2607 partial->pclu = EXT4_B2C(sbi, pblk);
2608 partial->state = nofree;
2609 }
2610 ex--;
2611 ex_ee_block = le32_to_cpu(ex->ee_block);
2612 ex_ee_len = ext4_ext_get_actual_len(ex);
2613 continue;
2614 } else if (b != ex_ee_block + ex_ee_len - 1) {
2615 EXT4_ERROR_INODE(inode,
2616 "can not handle truncate %u:%u "
2617 "on extent %u:%u",
2618 start, end, ex_ee_block,
2619 ex_ee_block + ex_ee_len - 1);
2620 err = -EFSCORRUPTED;
2621 goto out;
2622 } else if (a != ex_ee_block) {
2623 /* remove tail of the extent */
2624 num = a - ex_ee_block;
2625 } else {
2626 /* remove whole extent: excellent! */
2627 num = 0;
2628 }
2629 /*
2630 * 3 for leaf, sb, and inode plus 2 (bmap and group
2631 * descriptor) for each block group; assume two block
2632 * groups plus ex_ee_len/blocks_per_block_group for
2633 * the worst case
2634 */
2635 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2636 if (ex == EXT_FIRST_EXTENT(eh)) {
2637 correct_index = 1;
2638 credits += (ext_depth(inode)) + 1;
2639 }
2640 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2641 /*
2642 * We may end up freeing some index blocks and data from the
2643 * punched range. Note that partial clusters are accounted for
2644 * by ext4_free_data_revoke_credits().
2645 */
2646 revoke_credits =
2647 ext4_free_metadata_revoke_credits(inode->i_sb,
2648 ext_depth(inode)) +
2649 ext4_free_data_revoke_credits(inode, b - a + 1);
2650
2651 err = ext4_datasem_ensure_credits(handle, inode, credits,
2652 credits, revoke_credits);
2653 if (err) {
2654 if (err > 0)
2655 err = -EAGAIN;
2656 goto out;
2657 }
2658
2659 err = ext4_ext_get_access(handle, inode, path + depth);
2660 if (err)
2661 goto out;
2662
2663 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2664 if (err)
2665 goto out;
2666
2667 if (num == 0)
2668 /* this extent is removed; mark slot entirely unused */
2669 ext4_ext_store_pblock(ex, 0);
2670
2671 ex->ee_len = cpu_to_le16(num);
2672 /*
2673 * Do not mark unwritten if all the blocks in the
2674 * extent have been removed.
2675 */
2676 if (unwritten && num)
2677 ext4_ext_mark_unwritten(ex);
2678 /*
2679 * If the extent was completely released,
2680 * we need to remove it from the leaf
2681 */
2682 if (num == 0) {
2683 if (end != EXT_MAX_BLOCKS - 1) {
2684 /*
2685 * For hole punching, we need to scoot all the
2686 * extents up when an extent is removed so that
2687 * we dont have blank extents in the middle
2688 */
2689 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2690 sizeof(struct ext4_extent));
2691
2692 /* Now get rid of the one at the end */
2693 memset(EXT_LAST_EXTENT(eh), 0,
2694 sizeof(struct ext4_extent));
2695 }
2696 le16_add_cpu(&eh->eh_entries, -1);
2697 }
2698
2699 err = ext4_ext_dirty(handle, inode, path + depth);
2700 if (err)
2701 goto out;
2702
2703 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2704 ext4_ext_pblock(ex));
2705 ex--;
2706 ex_ee_block = le32_to_cpu(ex->ee_block);
2707 ex_ee_len = ext4_ext_get_actual_len(ex);
2708 }
2709
2710 if (correct_index && eh->eh_entries)
2711 err = ext4_ext_correct_indexes(handle, inode, path);
2712
2713 /*
2714 * If there's a partial cluster and at least one extent remains in
2715 * the leaf, free the partial cluster if it isn't shared with the
2716 * current extent. If it is shared with the current extent
2717 * we reset the partial cluster because we've reached the start of the
2718 * truncated/punched region and we're done removing blocks.
2719 */
2720 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2721 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2722 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2723 int flags = get_default_free_blocks_flags(inode);
2724
2725 if (ext4_is_pending(inode, partial->lblk))
2726 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2727 ext4_free_blocks(handle, inode, NULL,
2728 EXT4_C2B(sbi, partial->pclu),
2729 sbi->s_cluster_ratio, flags);
2730 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2731 ext4_rereserve_cluster(inode, partial->lblk);
2732 }
2733 partial->state = initial;
2734 }
2735
2736 /* if this leaf is free, then we should
2737 * remove it from index block above */
2738 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2739 err = ext4_ext_rm_idx(handle, inode, path, depth);
2740
2741out:
2742 return err;
2743}
2744
2745/*
2746 * ext4_ext_more_to_rm:
2747 * returns 1 if current index has to be freed (even partial)
2748 */
2749static int
2750ext4_ext_more_to_rm(struct ext4_ext_path *path)
2751{
2752 BUG_ON(path->p_idx == NULL);
2753
2754 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2755 return 0;
2756
2757 /*
2758 * if truncate on deeper level happened, it wasn't partial,
2759 * so we have to consider current index for truncation
2760 */
2761 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2762 return 0;
2763 return 1;
2764}
2765
2766int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2767 ext4_lblk_t end)
2768{
2769 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2770 int depth = ext_depth(inode);
2771 struct ext4_ext_path *path = NULL;
2772 struct partial_cluster partial;
2773 handle_t *handle;
2774 int i = 0, err = 0;
2775
2776 partial.pclu = 0;
2777 partial.lblk = 0;
2778 partial.state = initial;
2779
2780 ext_debug(inode, "truncate since %u to %u\n", start, end);
2781
2782 /* probably first extent we're gonna free will be last in block */
2783 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2784 depth + 1,
2785 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2786 if (IS_ERR(handle))
2787 return PTR_ERR(handle);
2788
2789again:
2790 trace_ext4_ext_remove_space(inode, start, end, depth);
2791
2792 /*
2793 * Check if we are removing extents inside the extent tree. If that
2794 * is the case, we are going to punch a hole inside the extent tree
2795 * so we have to check whether we need to split the extent covering
2796 * the last block to remove so we can easily remove the part of it
2797 * in ext4_ext_rm_leaf().
2798 */
2799 if (end < EXT_MAX_BLOCKS - 1) {
2800 struct ext4_extent *ex;
2801 ext4_lblk_t ee_block, ex_end, lblk;
2802 ext4_fsblk_t pblk;
2803
2804 /* find extent for or closest extent to this block */
2805 path = ext4_find_extent(inode, end, NULL,
2806 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2807 if (IS_ERR(path)) {
2808 ext4_journal_stop(handle);
2809 return PTR_ERR(path);
2810 }
2811 depth = ext_depth(inode);
2812 /* Leaf not may not exist only if inode has no blocks at all */
2813 ex = path[depth].p_ext;
2814 if (!ex) {
2815 if (depth) {
2816 EXT4_ERROR_INODE(inode,
2817 "path[%d].p_hdr == NULL",
2818 depth);
2819 err = -EFSCORRUPTED;
2820 }
2821 goto out;
2822 }
2823
2824 ee_block = le32_to_cpu(ex->ee_block);
2825 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2826
2827 /*
2828 * See if the last block is inside the extent, if so split
2829 * the extent at 'end' block so we can easily remove the
2830 * tail of the first part of the split extent in
2831 * ext4_ext_rm_leaf().
2832 */
2833 if (end >= ee_block && end < ex_end) {
2834
2835 /*
2836 * If we're going to split the extent, note that
2837 * the cluster containing the block after 'end' is
2838 * in use to avoid freeing it when removing blocks.
2839 */
2840 if (sbi->s_cluster_ratio > 1) {
2841 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2842 partial.pclu = EXT4_B2C(sbi, pblk);
2843 partial.state = nofree;
2844 }
2845
2846 /*
2847 * Split the extent in two so that 'end' is the last
2848 * block in the first new extent. Also we should not
2849 * fail removing space due to ENOSPC so try to use
2850 * reserved block if that happens.
2851 */
2852 err = ext4_force_split_extent_at(handle, inode, &path,
2853 end + 1, 1);
2854 if (err < 0)
2855 goto out;
2856
2857 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2858 partial.state == initial) {
2859 /*
2860 * If we're punching, there's an extent to the right.
2861 * If the partial cluster hasn't been set, set it to
2862 * that extent's first cluster and its state to nofree
2863 * so it won't be freed should it contain blocks to be
2864 * removed. If it's already set (tofree/nofree), we're
2865 * retrying and keep the original partial cluster info
2866 * so a cluster marked tofree as a result of earlier
2867 * extent removal is not lost.
2868 */
2869 lblk = ex_end + 1;
2870 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2871 &ex);
2872 if (err)
2873 goto out;
2874 if (pblk) {
2875 partial.pclu = EXT4_B2C(sbi, pblk);
2876 partial.state = nofree;
2877 }
2878 }
2879 }
2880 /*
2881 * We start scanning from right side, freeing all the blocks
2882 * after i_size and walking into the tree depth-wise.
2883 */
2884 depth = ext_depth(inode);
2885 if (path) {
2886 int k = i = depth;
2887 while (--k > 0)
2888 path[k].p_block =
2889 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2890 } else {
2891 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2892 GFP_NOFS | __GFP_NOFAIL);
2893 if (path == NULL) {
2894 ext4_journal_stop(handle);
2895 return -ENOMEM;
2896 }
2897 path[0].p_maxdepth = path[0].p_depth = depth;
2898 path[0].p_hdr = ext_inode_hdr(inode);
2899 i = 0;
2900
2901 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2902 err = -EFSCORRUPTED;
2903 goto out;
2904 }
2905 }
2906 err = 0;
2907
2908 while (i >= 0 && err == 0) {
2909 if (i == depth) {
2910 /* this is leaf block */
2911 err = ext4_ext_rm_leaf(handle, inode, path,
2912 &partial, start, end);
2913 /* root level has p_bh == NULL, brelse() eats this */
2914 brelse(path[i].p_bh);
2915 path[i].p_bh = NULL;
2916 i--;
2917 continue;
2918 }
2919
2920 /* this is index block */
2921 if (!path[i].p_hdr) {
2922 ext_debug(inode, "initialize header\n");
2923 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2924 }
2925
2926 if (!path[i].p_idx) {
2927 /* this level hasn't been touched yet */
2928 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2929 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2930 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2931 path[i].p_hdr,
2932 le16_to_cpu(path[i].p_hdr->eh_entries));
2933 } else {
2934 /* we were already here, see at next index */
2935 path[i].p_idx--;
2936 }
2937
2938 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2939 i, EXT_FIRST_INDEX(path[i].p_hdr),
2940 path[i].p_idx);
2941 if (ext4_ext_more_to_rm(path + i)) {
2942 struct buffer_head *bh;
2943 /* go to the next level */
2944 ext_debug(inode, "move to level %d (block %llu)\n",
2945 i + 1, ext4_idx_pblock(path[i].p_idx));
2946 memset(path + i + 1, 0, sizeof(*path));
2947 bh = read_extent_tree_block(inode,
2948 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2949 EXT4_EX_NOCACHE);
2950 if (IS_ERR(bh)) {
2951 /* should we reset i_size? */
2952 err = PTR_ERR(bh);
2953 break;
2954 }
2955 /* Yield here to deal with large extent trees.
2956 * Should be a no-op if we did IO above. */
2957 cond_resched();
2958 if (WARN_ON(i + 1 > depth)) {
2959 err = -EFSCORRUPTED;
2960 break;
2961 }
2962 path[i + 1].p_bh = bh;
2963
2964 /* save actual number of indexes since this
2965 * number is changed at the next iteration */
2966 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2967 i++;
2968 } else {
2969 /* we finished processing this index, go up */
2970 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2971 /* index is empty, remove it;
2972 * handle must be already prepared by the
2973 * truncatei_leaf() */
2974 err = ext4_ext_rm_idx(handle, inode, path, i);
2975 }
2976 /* root level has p_bh == NULL, brelse() eats this */
2977 brelse(path[i].p_bh);
2978 path[i].p_bh = NULL;
2979 i--;
2980 ext_debug(inode, "return to level %d\n", i);
2981 }
2982 }
2983
2984 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
2985 path->p_hdr->eh_entries);
2986
2987 /*
2988 * if there's a partial cluster and we have removed the first extent
2989 * in the file, then we also free the partial cluster, if any
2990 */
2991 if (partial.state == tofree && err == 0) {
2992 int flags = get_default_free_blocks_flags(inode);
2993
2994 if (ext4_is_pending(inode, partial.lblk))
2995 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2996 ext4_free_blocks(handle, inode, NULL,
2997 EXT4_C2B(sbi, partial.pclu),
2998 sbi->s_cluster_ratio, flags);
2999 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3000 ext4_rereserve_cluster(inode, partial.lblk);
3001 partial.state = initial;
3002 }
3003
3004 /* TODO: flexible tree reduction should be here */
3005 if (path->p_hdr->eh_entries == 0) {
3006 /*
3007 * truncate to zero freed all the tree,
3008 * so we need to correct eh_depth
3009 */
3010 err = ext4_ext_get_access(handle, inode, path);
3011 if (err == 0) {
3012 ext_inode_hdr(inode)->eh_depth = 0;
3013 ext_inode_hdr(inode)->eh_max =
3014 cpu_to_le16(ext4_ext_space_root(inode, 0));
3015 err = ext4_ext_dirty(handle, inode, path);
3016 }
3017 }
3018out:
3019 ext4_ext_drop_refs(path);
3020 kfree(path);
3021 path = NULL;
3022 if (err == -EAGAIN)
3023 goto again;
3024 ext4_journal_stop(handle);
3025
3026 return err;
3027}
3028
3029/*
3030 * called at mount time
3031 */
3032void ext4_ext_init(struct super_block *sb)
3033{
3034 /*
3035 * possible initialization would be here
3036 */
3037
3038 if (ext4_has_feature_extents(sb)) {
3039#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3040 printk(KERN_INFO "EXT4-fs: file extents enabled"
3041#ifdef AGGRESSIVE_TEST
3042 ", aggressive tests"
3043#endif
3044#ifdef CHECK_BINSEARCH
3045 ", check binsearch"
3046#endif
3047#ifdef EXTENTS_STATS
3048 ", stats"
3049#endif
3050 "\n");
3051#endif
3052#ifdef EXTENTS_STATS
3053 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3054 EXT4_SB(sb)->s_ext_min = 1 << 30;
3055 EXT4_SB(sb)->s_ext_max = 0;
3056#endif
3057 }
3058}
3059
3060/*
3061 * called at umount time
3062 */
3063void ext4_ext_release(struct super_block *sb)
3064{
3065 if (!ext4_has_feature_extents(sb))
3066 return;
3067
3068#ifdef EXTENTS_STATS
3069 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3070 struct ext4_sb_info *sbi = EXT4_SB(sb);
3071 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3072 sbi->s_ext_blocks, sbi->s_ext_extents,
3073 sbi->s_ext_blocks / sbi->s_ext_extents);
3074 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3075 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3076 }
3077#endif
3078}
3079
3080static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3081{
3082 ext4_lblk_t ee_block;
3083 ext4_fsblk_t ee_pblock;
3084 unsigned int ee_len;
3085
3086 ee_block = le32_to_cpu(ex->ee_block);
3087 ee_len = ext4_ext_get_actual_len(ex);
3088 ee_pblock = ext4_ext_pblock(ex);
3089
3090 if (ee_len == 0)
3091 return 0;
3092
3093 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3094 EXTENT_STATUS_WRITTEN);
3095}
3096
3097/* FIXME!! we need to try to merge to left or right after zero-out */
3098static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3099{
3100 ext4_fsblk_t ee_pblock;
3101 unsigned int ee_len;
3102
3103 ee_len = ext4_ext_get_actual_len(ex);
3104 ee_pblock = ext4_ext_pblock(ex);
3105 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3106 ee_len);
3107}
3108
3109/*
3110 * ext4_split_extent_at() splits an extent at given block.
3111 *
3112 * @handle: the journal handle
3113 * @inode: the file inode
3114 * @path: the path to the extent
3115 * @split: the logical block where the extent is splitted.
3116 * @split_flags: indicates if the extent could be zeroout if split fails, and
3117 * the states(init or unwritten) of new extents.
3118 * @flags: flags used to insert new extent to extent tree.
3119 *
3120 *
3121 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3122 * of which are determined by split_flag.
3123 *
3124 * There are two cases:
3125 * a> the extent are splitted into two extent.
3126 * b> split is not needed, and just mark the extent.
3127 *
3128 * return 0 on success.
3129 */
3130static int ext4_split_extent_at(handle_t *handle,
3131 struct inode *inode,
3132 struct ext4_ext_path **ppath,
3133 ext4_lblk_t split,
3134 int split_flag,
3135 int flags)
3136{
3137 struct ext4_ext_path *path = *ppath;
3138 ext4_fsblk_t newblock;
3139 ext4_lblk_t ee_block;
3140 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3141 struct ext4_extent *ex2 = NULL;
3142 unsigned int ee_len, depth;
3143 int err = 0;
3144
3145 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3146 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3147
3148 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3149
3150 ext4_ext_show_leaf(inode, path);
3151
3152 depth = ext_depth(inode);
3153 ex = path[depth].p_ext;
3154 ee_block = le32_to_cpu(ex->ee_block);
3155 ee_len = ext4_ext_get_actual_len(ex);
3156 newblock = split - ee_block + ext4_ext_pblock(ex);
3157
3158 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3159 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3160 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3161 EXT4_EXT_MARK_UNWRIT1 |
3162 EXT4_EXT_MARK_UNWRIT2));
3163
3164 err = ext4_ext_get_access(handle, inode, path + depth);
3165 if (err)
3166 goto out;
3167
3168 if (split == ee_block) {
3169 /*
3170 * case b: block @split is the block that the extent begins with
3171 * then we just change the state of the extent, and splitting
3172 * is not needed.
3173 */
3174 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3175 ext4_ext_mark_unwritten(ex);
3176 else
3177 ext4_ext_mark_initialized(ex);
3178
3179 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3180 ext4_ext_try_to_merge(handle, inode, path, ex);
3181
3182 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3183 goto out;
3184 }
3185
3186 /* case a */
3187 memcpy(&orig_ex, ex, sizeof(orig_ex));
3188 ex->ee_len = cpu_to_le16(split - ee_block);
3189 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3190 ext4_ext_mark_unwritten(ex);
3191
3192 /*
3193 * path may lead to new leaf, not to original leaf any more
3194 * after ext4_ext_insert_extent() returns,
3195 */
3196 err = ext4_ext_dirty(handle, inode, path + depth);
3197 if (err)
3198 goto fix_extent_len;
3199
3200 ex2 = &newex;
3201 ex2->ee_block = cpu_to_le32(split);
3202 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3203 ext4_ext_store_pblock(ex2, newblock);
3204 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3205 ext4_ext_mark_unwritten(ex2);
3206
3207 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3208 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3209 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3210 if (split_flag & EXT4_EXT_DATA_VALID1) {
3211 err = ext4_ext_zeroout(inode, ex2);
3212 zero_ex.ee_block = ex2->ee_block;
3213 zero_ex.ee_len = cpu_to_le16(
3214 ext4_ext_get_actual_len(ex2));
3215 ext4_ext_store_pblock(&zero_ex,
3216 ext4_ext_pblock(ex2));
3217 } else {
3218 err = ext4_ext_zeroout(inode, ex);
3219 zero_ex.ee_block = ex->ee_block;
3220 zero_ex.ee_len = cpu_to_le16(
3221 ext4_ext_get_actual_len(ex));
3222 ext4_ext_store_pblock(&zero_ex,
3223 ext4_ext_pblock(ex));
3224 }
3225 } else {
3226 err = ext4_ext_zeroout(inode, &orig_ex);
3227 zero_ex.ee_block = orig_ex.ee_block;
3228 zero_ex.ee_len = cpu_to_le16(
3229 ext4_ext_get_actual_len(&orig_ex));
3230 ext4_ext_store_pblock(&zero_ex,
3231 ext4_ext_pblock(&orig_ex));
3232 }
3233
3234 if (err)
3235 goto fix_extent_len;
3236 /* update the extent length and mark as initialized */
3237 ex->ee_len = cpu_to_le16(ee_len);
3238 ext4_ext_try_to_merge(handle, inode, path, ex);
3239 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3240 if (err)
3241 goto fix_extent_len;
3242
3243 /* update extent status tree */
3244 err = ext4_zeroout_es(inode, &zero_ex);
3245
3246 goto out;
3247 } else if (err)
3248 goto fix_extent_len;
3249
3250out:
3251 ext4_ext_show_leaf(inode, path);
3252 return err;
3253
3254fix_extent_len:
3255 ex->ee_len = orig_ex.ee_len;
3256 /*
3257 * Ignore ext4_ext_dirty return value since we are already in error path
3258 * and err is a non-zero error code.
3259 */
3260 ext4_ext_dirty(handle, inode, path + path->p_depth);
3261 return err;
3262}
3263
3264/*
3265 * ext4_split_extents() splits an extent and mark extent which is covered
3266 * by @map as split_flags indicates
3267 *
3268 * It may result in splitting the extent into multiple extents (up to three)
3269 * There are three possibilities:
3270 * a> There is no split required
3271 * b> Splits in two extents: Split is happening at either end of the extent
3272 * c> Splits in three extents: Somone is splitting in middle of the extent
3273 *
3274 */
3275static int ext4_split_extent(handle_t *handle,
3276 struct inode *inode,
3277 struct ext4_ext_path **ppath,
3278 struct ext4_map_blocks *map,
3279 int split_flag,
3280 int flags)
3281{
3282 struct ext4_ext_path *path = *ppath;
3283 ext4_lblk_t ee_block;
3284 struct ext4_extent *ex;
3285 unsigned int ee_len, depth;
3286 int err = 0;
3287 int unwritten;
3288 int split_flag1, flags1;
3289 int allocated = map->m_len;
3290
3291 depth = ext_depth(inode);
3292 ex = path[depth].p_ext;
3293 ee_block = le32_to_cpu(ex->ee_block);
3294 ee_len = ext4_ext_get_actual_len(ex);
3295 unwritten = ext4_ext_is_unwritten(ex);
3296
3297 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3298 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3299 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3300 if (unwritten)
3301 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3302 EXT4_EXT_MARK_UNWRIT2;
3303 if (split_flag & EXT4_EXT_DATA_VALID2)
3304 split_flag1 |= EXT4_EXT_DATA_VALID1;
3305 err = ext4_split_extent_at(handle, inode, ppath,
3306 map->m_lblk + map->m_len, split_flag1, flags1);
3307 if (err)
3308 goto out;
3309 } else {
3310 allocated = ee_len - (map->m_lblk - ee_block);
3311 }
3312 /*
3313 * Update path is required because previous ext4_split_extent_at() may
3314 * result in split of original leaf or extent zeroout.
3315 */
3316 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3317 if (IS_ERR(path))
3318 return PTR_ERR(path);
3319 depth = ext_depth(inode);
3320 ex = path[depth].p_ext;
3321 if (!ex) {
3322 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3323 (unsigned long) map->m_lblk);
3324 return -EFSCORRUPTED;
3325 }
3326 unwritten = ext4_ext_is_unwritten(ex);
3327 split_flag1 = 0;
3328
3329 if (map->m_lblk >= ee_block) {
3330 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3331 if (unwritten) {
3332 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3333 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3334 EXT4_EXT_MARK_UNWRIT2);
3335 }
3336 err = ext4_split_extent_at(handle, inode, ppath,
3337 map->m_lblk, split_flag1, flags);
3338 if (err)
3339 goto out;
3340 }
3341
3342 ext4_ext_show_leaf(inode, path);
3343out:
3344 return err ? err : allocated;
3345}
3346
3347/*
3348 * This function is called by ext4_ext_map_blocks() if someone tries to write
3349 * to an unwritten extent. It may result in splitting the unwritten
3350 * extent into multiple extents (up to three - one initialized and two
3351 * unwritten).
3352 * There are three possibilities:
3353 * a> There is no split required: Entire extent should be initialized
3354 * b> Splits in two extents: Write is happening at either end of the extent
3355 * c> Splits in three extents: Somone is writing in middle of the extent
3356 *
3357 * Pre-conditions:
3358 * - The extent pointed to by 'path' is unwritten.
3359 * - The extent pointed to by 'path' contains a superset
3360 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3361 *
3362 * Post-conditions on success:
3363 * - the returned value is the number of blocks beyond map->l_lblk
3364 * that are allocated and initialized.
3365 * It is guaranteed to be >= map->m_len.
3366 */
3367static int ext4_ext_convert_to_initialized(handle_t *handle,
3368 struct inode *inode,
3369 struct ext4_map_blocks *map,
3370 struct ext4_ext_path **ppath,
3371 int flags)
3372{
3373 struct ext4_ext_path *path = *ppath;
3374 struct ext4_sb_info *sbi;
3375 struct ext4_extent_header *eh;
3376 struct ext4_map_blocks split_map;
3377 struct ext4_extent zero_ex1, zero_ex2;
3378 struct ext4_extent *ex, *abut_ex;
3379 ext4_lblk_t ee_block, eof_block;
3380 unsigned int ee_len, depth, map_len = map->m_len;
3381 int allocated = 0, max_zeroout = 0;
3382 int err = 0;
3383 int split_flag = EXT4_EXT_DATA_VALID2;
3384
3385 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3386 (unsigned long long)map->m_lblk, map_len);
3387
3388 sbi = EXT4_SB(inode->i_sb);
3389 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3390 >> inode->i_sb->s_blocksize_bits;
3391 if (eof_block < map->m_lblk + map_len)
3392 eof_block = map->m_lblk + map_len;
3393
3394 depth = ext_depth(inode);
3395 eh = path[depth].p_hdr;
3396 ex = path[depth].p_ext;
3397 ee_block = le32_to_cpu(ex->ee_block);
3398 ee_len = ext4_ext_get_actual_len(ex);
3399 zero_ex1.ee_len = 0;
3400 zero_ex2.ee_len = 0;
3401
3402 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3403
3404 /* Pre-conditions */
3405 BUG_ON(!ext4_ext_is_unwritten(ex));
3406 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3407
3408 /*
3409 * Attempt to transfer newly initialized blocks from the currently
3410 * unwritten extent to its neighbor. This is much cheaper
3411 * than an insertion followed by a merge as those involve costly
3412 * memmove() calls. Transferring to the left is the common case in
3413 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3414 * followed by append writes.
3415 *
3416 * Limitations of the current logic:
3417 * - L1: we do not deal with writes covering the whole extent.
3418 * This would require removing the extent if the transfer
3419 * is possible.
3420 * - L2: we only attempt to merge with an extent stored in the
3421 * same extent tree node.
3422 */
3423 if ((map->m_lblk == ee_block) &&
3424 /* See if we can merge left */
3425 (map_len < ee_len) && /*L1*/
3426 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3427 ext4_lblk_t prev_lblk;
3428 ext4_fsblk_t prev_pblk, ee_pblk;
3429 unsigned int prev_len;
3430
3431 abut_ex = ex - 1;
3432 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3433 prev_len = ext4_ext_get_actual_len(abut_ex);
3434 prev_pblk = ext4_ext_pblock(abut_ex);
3435 ee_pblk = ext4_ext_pblock(ex);
3436
3437 /*
3438 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3439 * upon those conditions:
3440 * - C1: abut_ex is initialized,
3441 * - C2: abut_ex is logically abutting ex,
3442 * - C3: abut_ex is physically abutting ex,
3443 * - C4: abut_ex can receive the additional blocks without
3444 * overflowing the (initialized) length limit.
3445 */
3446 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3447 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3448 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3449 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3450 err = ext4_ext_get_access(handle, inode, path + depth);
3451 if (err)
3452 goto out;
3453
3454 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3455 map, ex, abut_ex);
3456
3457 /* Shift the start of ex by 'map_len' blocks */
3458 ex->ee_block = cpu_to_le32(ee_block + map_len);
3459 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3460 ex->ee_len = cpu_to_le16(ee_len - map_len);
3461 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3462
3463 /* Extend abut_ex by 'map_len' blocks */
3464 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3465
3466 /* Result: number of initialized blocks past m_lblk */
3467 allocated = map_len;
3468 }
3469 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3470 (map_len < ee_len) && /*L1*/
3471 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3472 /* See if we can merge right */
3473 ext4_lblk_t next_lblk;
3474 ext4_fsblk_t next_pblk, ee_pblk;
3475 unsigned int next_len;
3476
3477 abut_ex = ex + 1;
3478 next_lblk = le32_to_cpu(abut_ex->ee_block);
3479 next_len = ext4_ext_get_actual_len(abut_ex);
3480 next_pblk = ext4_ext_pblock(abut_ex);
3481 ee_pblk = ext4_ext_pblock(ex);
3482
3483 /*
3484 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3485 * upon those conditions:
3486 * - C1: abut_ex is initialized,
3487 * - C2: abut_ex is logically abutting ex,
3488 * - C3: abut_ex is physically abutting ex,
3489 * - C4: abut_ex can receive the additional blocks without
3490 * overflowing the (initialized) length limit.
3491 */
3492 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3493 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3494 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3495 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3496 err = ext4_ext_get_access(handle, inode, path + depth);
3497 if (err)
3498 goto out;
3499
3500 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3501 map, ex, abut_ex);
3502
3503 /* Shift the start of abut_ex by 'map_len' blocks */
3504 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3505 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3506 ex->ee_len = cpu_to_le16(ee_len - map_len);
3507 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3508
3509 /* Extend abut_ex by 'map_len' blocks */
3510 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3511
3512 /* Result: number of initialized blocks past m_lblk */
3513 allocated = map_len;
3514 }
3515 }
3516 if (allocated) {
3517 /* Mark the block containing both extents as dirty */
3518 err = ext4_ext_dirty(handle, inode, path + depth);
3519
3520 /* Update path to point to the right extent */
3521 path[depth].p_ext = abut_ex;
3522 goto out;
3523 } else
3524 allocated = ee_len - (map->m_lblk - ee_block);
3525
3526 WARN_ON(map->m_lblk < ee_block);
3527 /*
3528 * It is safe to convert extent to initialized via explicit
3529 * zeroout only if extent is fully inside i_size or new_size.
3530 */
3531 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3532
3533 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3534 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3535 (inode->i_sb->s_blocksize_bits - 10);
3536
3537 /*
3538 * five cases:
3539 * 1. split the extent into three extents.
3540 * 2. split the extent into two extents, zeroout the head of the first
3541 * extent.
3542 * 3. split the extent into two extents, zeroout the tail of the second
3543 * extent.
3544 * 4. split the extent into two extents with out zeroout.
3545 * 5. no splitting needed, just possibly zeroout the head and / or the
3546 * tail of the extent.
3547 */
3548 split_map.m_lblk = map->m_lblk;
3549 split_map.m_len = map->m_len;
3550
3551 if (max_zeroout && (allocated > split_map.m_len)) {
3552 if (allocated <= max_zeroout) {
3553 /* case 3 or 5 */
3554 zero_ex1.ee_block =
3555 cpu_to_le32(split_map.m_lblk +
3556 split_map.m_len);
3557 zero_ex1.ee_len =
3558 cpu_to_le16(allocated - split_map.m_len);
3559 ext4_ext_store_pblock(&zero_ex1,
3560 ext4_ext_pblock(ex) + split_map.m_lblk +
3561 split_map.m_len - ee_block);
3562 err = ext4_ext_zeroout(inode, &zero_ex1);
3563 if (err)
3564 goto out;
3565 split_map.m_len = allocated;
3566 }
3567 if (split_map.m_lblk - ee_block + split_map.m_len <
3568 max_zeroout) {
3569 /* case 2 or 5 */
3570 if (split_map.m_lblk != ee_block) {
3571 zero_ex2.ee_block = ex->ee_block;
3572 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3573 ee_block);
3574 ext4_ext_store_pblock(&zero_ex2,
3575 ext4_ext_pblock(ex));
3576 err = ext4_ext_zeroout(inode, &zero_ex2);
3577 if (err)
3578 goto out;
3579 }
3580
3581 split_map.m_len += split_map.m_lblk - ee_block;
3582 split_map.m_lblk = ee_block;
3583 allocated = map->m_len;
3584 }
3585 }
3586
3587 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3588 flags);
3589 if (err > 0)
3590 err = 0;
3591out:
3592 /* If we have gotten a failure, don't zero out status tree */
3593 if (!err) {
3594 err = ext4_zeroout_es(inode, &zero_ex1);
3595 if (!err)
3596 err = ext4_zeroout_es(inode, &zero_ex2);
3597 }
3598 return err ? err : allocated;
3599}
3600
3601/*
3602 * This function is called by ext4_ext_map_blocks() from
3603 * ext4_get_blocks_dio_write() when DIO to write
3604 * to an unwritten extent.
3605 *
3606 * Writing to an unwritten extent may result in splitting the unwritten
3607 * extent into multiple initialized/unwritten extents (up to three)
3608 * There are three possibilities:
3609 * a> There is no split required: Entire extent should be unwritten
3610 * b> Splits in two extents: Write is happening at either end of the extent
3611 * c> Splits in three extents: Somone is writing in middle of the extent
3612 *
3613 * This works the same way in the case of initialized -> unwritten conversion.
3614 *
3615 * One of more index blocks maybe needed if the extent tree grow after
3616 * the unwritten extent split. To prevent ENOSPC occur at the IO
3617 * complete, we need to split the unwritten extent before DIO submit
3618 * the IO. The unwritten extent called at this time will be split
3619 * into three unwritten extent(at most). After IO complete, the part
3620 * being filled will be convert to initialized by the end_io callback function
3621 * via ext4_convert_unwritten_extents().
3622 *
3623 * Returns the size of unwritten extent to be written on success.
3624 */
3625static int ext4_split_convert_extents(handle_t *handle,
3626 struct inode *inode,
3627 struct ext4_map_blocks *map,
3628 struct ext4_ext_path **ppath,
3629 int flags)
3630{
3631 struct ext4_ext_path *path = *ppath;
3632 ext4_lblk_t eof_block;
3633 ext4_lblk_t ee_block;
3634 struct ext4_extent *ex;
3635 unsigned int ee_len;
3636 int split_flag = 0, depth;
3637
3638 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3639 (unsigned long long)map->m_lblk, map->m_len);
3640
3641 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3642 >> inode->i_sb->s_blocksize_bits;
3643 if (eof_block < map->m_lblk + map->m_len)
3644 eof_block = map->m_lblk + map->m_len;
3645 /*
3646 * It is safe to convert extent to initialized via explicit
3647 * zeroout only if extent is fully inside i_size or new_size.
3648 */
3649 depth = ext_depth(inode);
3650 ex = path[depth].p_ext;
3651 ee_block = le32_to_cpu(ex->ee_block);
3652 ee_len = ext4_ext_get_actual_len(ex);
3653
3654 /* Convert to unwritten */
3655 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3656 split_flag |= EXT4_EXT_DATA_VALID1;
3657 /* Convert to initialized */
3658 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3659 split_flag |= ee_block + ee_len <= eof_block ?
3660 EXT4_EXT_MAY_ZEROOUT : 0;
3661 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3662 }
3663 flags |= EXT4_GET_BLOCKS_PRE_IO;
3664 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3665}
3666
3667static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3668 struct inode *inode,
3669 struct ext4_map_blocks *map,
3670 struct ext4_ext_path **ppath)
3671{
3672 struct ext4_ext_path *path = *ppath;
3673 struct ext4_extent *ex;
3674 ext4_lblk_t ee_block;
3675 unsigned int ee_len;
3676 int depth;
3677 int err = 0;
3678
3679 depth = ext_depth(inode);
3680 ex = path[depth].p_ext;
3681 ee_block = le32_to_cpu(ex->ee_block);
3682 ee_len = ext4_ext_get_actual_len(ex);
3683
3684 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3685 (unsigned long long)ee_block, ee_len);
3686
3687 /* If extent is larger than requested it is a clear sign that we still
3688 * have some extent state machine issues left. So extent_split is still
3689 * required.
3690 * TODO: Once all related issues will be fixed this situation should be
3691 * illegal.
3692 */
3693 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3694#ifdef CONFIG_EXT4_DEBUG
3695 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3696 " len %u; IO logical block %llu, len %u",
3697 inode->i_ino, (unsigned long long)ee_block, ee_len,
3698 (unsigned long long)map->m_lblk, map->m_len);
3699#endif
3700 err = ext4_split_convert_extents(handle, inode, map, ppath,
3701 EXT4_GET_BLOCKS_CONVERT);
3702 if (err < 0)
3703 return err;
3704 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3705 if (IS_ERR(path))
3706 return PTR_ERR(path);
3707 depth = ext_depth(inode);
3708 ex = path[depth].p_ext;
3709 }
3710
3711 err = ext4_ext_get_access(handle, inode, path + depth);
3712 if (err)
3713 goto out;
3714 /* first mark the extent as initialized */
3715 ext4_ext_mark_initialized(ex);
3716
3717 /* note: ext4_ext_correct_indexes() isn't needed here because
3718 * borders are not changed
3719 */
3720 ext4_ext_try_to_merge(handle, inode, path, ex);
3721
3722 /* Mark modified extent as dirty */
3723 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3724out:
3725 ext4_ext_show_leaf(inode, path);
3726 return err;
3727}
3728
3729static int
3730convert_initialized_extent(handle_t *handle, struct inode *inode,
3731 struct ext4_map_blocks *map,
3732 struct ext4_ext_path **ppath,
3733 unsigned int *allocated)
3734{
3735 struct ext4_ext_path *path = *ppath;
3736 struct ext4_extent *ex;
3737 ext4_lblk_t ee_block;
3738 unsigned int ee_len;
3739 int depth;
3740 int err = 0;
3741
3742 /*
3743 * Make sure that the extent is no bigger than we support with
3744 * unwritten extent
3745 */
3746 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3747 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3748
3749 depth = ext_depth(inode);
3750 ex = path[depth].p_ext;
3751 ee_block = le32_to_cpu(ex->ee_block);
3752 ee_len = ext4_ext_get_actual_len(ex);
3753
3754 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3755 (unsigned long long)ee_block, ee_len);
3756
3757 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3758 err = ext4_split_convert_extents(handle, inode, map, ppath,
3759 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3760 if (err < 0)
3761 return err;
3762 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3763 if (IS_ERR(path))
3764 return PTR_ERR(path);
3765 depth = ext_depth(inode);
3766 ex = path[depth].p_ext;
3767 if (!ex) {
3768 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3769 (unsigned long) map->m_lblk);
3770 return -EFSCORRUPTED;
3771 }
3772 }
3773
3774 err = ext4_ext_get_access(handle, inode, path + depth);
3775 if (err)
3776 return err;
3777 /* first mark the extent as unwritten */
3778 ext4_ext_mark_unwritten(ex);
3779
3780 /* note: ext4_ext_correct_indexes() isn't needed here because
3781 * borders are not changed
3782 */
3783 ext4_ext_try_to_merge(handle, inode, path, ex);
3784
3785 /* Mark modified extent as dirty */
3786 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3787 if (err)
3788 return err;
3789 ext4_ext_show_leaf(inode, path);
3790
3791 ext4_update_inode_fsync_trans(handle, inode, 1);
3792
3793 map->m_flags |= EXT4_MAP_UNWRITTEN;
3794 if (*allocated > map->m_len)
3795 *allocated = map->m_len;
3796 map->m_len = *allocated;
3797 return 0;
3798}
3799
3800static int
3801ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3802 struct ext4_map_blocks *map,
3803 struct ext4_ext_path **ppath, int flags,
3804 unsigned int allocated, ext4_fsblk_t newblock)
3805{
3806 struct ext4_ext_path __maybe_unused *path = *ppath;
3807 int ret = 0;
3808 int err = 0;
3809
3810 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3811 (unsigned long long)map->m_lblk, map->m_len, flags,
3812 allocated);
3813 ext4_ext_show_leaf(inode, path);
3814
3815 /*
3816 * When writing into unwritten space, we should not fail to
3817 * allocate metadata blocks for the new extent block if needed.
3818 */
3819 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3820
3821 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3822 allocated, newblock);
3823
3824 /* get_block() before submitting IO, split the extent */
3825 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3826 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3827 flags | EXT4_GET_BLOCKS_CONVERT);
3828 if (ret < 0) {
3829 err = ret;
3830 goto out2;
3831 }
3832 /*
3833 * shouldn't get a 0 return when splitting an extent unless
3834 * m_len is 0 (bug) or extent has been corrupted
3835 */
3836 if (unlikely(ret == 0)) {
3837 EXT4_ERROR_INODE(inode,
3838 "unexpected ret == 0, m_len = %u",
3839 map->m_len);
3840 err = -EFSCORRUPTED;
3841 goto out2;
3842 }
3843 map->m_flags |= EXT4_MAP_UNWRITTEN;
3844 goto out;
3845 }
3846 /* IO end_io complete, convert the filled extent to written */
3847 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3848 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3849 ppath);
3850 if (err < 0)
3851 goto out2;
3852 ext4_update_inode_fsync_trans(handle, inode, 1);
3853 goto map_out;
3854 }
3855 /* buffered IO cases */
3856 /*
3857 * repeat fallocate creation request
3858 * we already have an unwritten extent
3859 */
3860 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3861 map->m_flags |= EXT4_MAP_UNWRITTEN;
3862 goto map_out;
3863 }
3864
3865 /* buffered READ or buffered write_begin() lookup */
3866 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3867 /*
3868 * We have blocks reserved already. We
3869 * return allocated blocks so that delalloc
3870 * won't do block reservation for us. But
3871 * the buffer head will be unmapped so that
3872 * a read from the block returns 0s.
3873 */
3874 map->m_flags |= EXT4_MAP_UNWRITTEN;
3875 goto out1;
3876 }
3877
3878 /*
3879 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3880 * For buffered writes, at writepage time, etc. Convert a
3881 * discovered unwritten extent to written.
3882 */
3883 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3884 if (ret < 0) {
3885 err = ret;
3886 goto out2;
3887 }
3888 ext4_update_inode_fsync_trans(handle, inode, 1);
3889 /*
3890 * shouldn't get a 0 return when converting an unwritten extent
3891 * unless m_len is 0 (bug) or extent has been corrupted
3892 */
3893 if (unlikely(ret == 0)) {
3894 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3895 map->m_len);
3896 err = -EFSCORRUPTED;
3897 goto out2;
3898 }
3899
3900out:
3901 allocated = ret;
3902 map->m_flags |= EXT4_MAP_NEW;
3903map_out:
3904 map->m_flags |= EXT4_MAP_MAPPED;
3905out1:
3906 map->m_pblk = newblock;
3907 if (allocated > map->m_len)
3908 allocated = map->m_len;
3909 map->m_len = allocated;
3910 ext4_ext_show_leaf(inode, path);
3911out2:
3912 return err ? err : allocated;
3913}
3914
3915/*
3916 * get_implied_cluster_alloc - check to see if the requested
3917 * allocation (in the map structure) overlaps with a cluster already
3918 * allocated in an extent.
3919 * @sb The filesystem superblock structure
3920 * @map The requested lblk->pblk mapping
3921 * @ex The extent structure which might contain an implied
3922 * cluster allocation
3923 *
3924 * This function is called by ext4_ext_map_blocks() after we failed to
3925 * find blocks that were already in the inode's extent tree. Hence,
3926 * we know that the beginning of the requested region cannot overlap
3927 * the extent from the inode's extent tree. There are three cases we
3928 * want to catch. The first is this case:
3929 *
3930 * |--- cluster # N--|
3931 * |--- extent ---| |---- requested region ---|
3932 * |==========|
3933 *
3934 * The second case that we need to test for is this one:
3935 *
3936 * |--------- cluster # N ----------------|
3937 * |--- requested region --| |------- extent ----|
3938 * |=======================|
3939 *
3940 * The third case is when the requested region lies between two extents
3941 * within the same cluster:
3942 * |------------- cluster # N-------------|
3943 * |----- ex -----| |---- ex_right ----|
3944 * |------ requested region ------|
3945 * |================|
3946 *
3947 * In each of the above cases, we need to set the map->m_pblk and
3948 * map->m_len so it corresponds to the return the extent labelled as
3949 * "|====|" from cluster #N, since it is already in use for data in
3950 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3951 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3952 * as a new "allocated" block region. Otherwise, we will return 0 and
3953 * ext4_ext_map_blocks() will then allocate one or more new clusters
3954 * by calling ext4_mb_new_blocks().
3955 */
3956static int get_implied_cluster_alloc(struct super_block *sb,
3957 struct ext4_map_blocks *map,
3958 struct ext4_extent *ex,
3959 struct ext4_ext_path *path)
3960{
3961 struct ext4_sb_info *sbi = EXT4_SB(sb);
3962 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3963 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3964 ext4_lblk_t rr_cluster_start;
3965 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3966 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3967 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3968
3969 /* The extent passed in that we are trying to match */
3970 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3971 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3972
3973 /* The requested region passed into ext4_map_blocks() */
3974 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3975
3976 if ((rr_cluster_start == ex_cluster_end) ||
3977 (rr_cluster_start == ex_cluster_start)) {
3978 if (rr_cluster_start == ex_cluster_end)
3979 ee_start += ee_len - 1;
3980 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
3981 map->m_len = min(map->m_len,
3982 (unsigned) sbi->s_cluster_ratio - c_offset);
3983 /*
3984 * Check for and handle this case:
3985 *
3986 * |--------- cluster # N-------------|
3987 * |------- extent ----|
3988 * |--- requested region ---|
3989 * |===========|
3990 */
3991
3992 if (map->m_lblk < ee_block)
3993 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3994
3995 /*
3996 * Check for the case where there is already another allocated
3997 * block to the right of 'ex' but before the end of the cluster.
3998 *
3999 * |------------- cluster # N-------------|
4000 * |----- ex -----| |---- ex_right ----|
4001 * |------ requested region ------|
4002 * |================|
4003 */
4004 if (map->m_lblk > ee_block) {
4005 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4006 map->m_len = min(map->m_len, next - map->m_lblk);
4007 }
4008
4009 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4010 return 1;
4011 }
4012
4013 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4014 return 0;
4015}
4016
4017
4018/*
4019 * Block allocation/map/preallocation routine for extents based files
4020 *
4021 *
4022 * Need to be called with
4023 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4024 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4025 *
4026 * return > 0, number of of blocks already mapped/allocated
4027 * if create == 0 and these are pre-allocated blocks
4028 * buffer head is unmapped
4029 * otherwise blocks are mapped
4030 *
4031 * return = 0, if plain look up failed (blocks have not been allocated)
4032 * buffer head is unmapped
4033 *
4034 * return < 0, error case.
4035 */
4036int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4037 struct ext4_map_blocks *map, int flags)
4038{
4039 struct ext4_ext_path *path = NULL;
4040 struct ext4_extent newex, *ex, *ex2;
4041 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4042 ext4_fsblk_t newblock = 0, pblk;
4043 int err = 0, depth, ret;
4044 unsigned int allocated = 0, offset = 0;
4045 unsigned int allocated_clusters = 0;
4046 struct ext4_allocation_request ar;
4047 ext4_lblk_t cluster_offset;
4048
4049 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4050 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4051
4052 /* find extent for this block */
4053 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4054 if (IS_ERR(path)) {
4055 err = PTR_ERR(path);
4056 path = NULL;
4057 goto out;
4058 }
4059
4060 depth = ext_depth(inode);
4061
4062 /*
4063 * consistent leaf must not be empty;
4064 * this situation is possible, though, _during_ tree modification;
4065 * this is why assert can't be put in ext4_find_extent()
4066 */
4067 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4068 EXT4_ERROR_INODE(inode, "bad extent address "
4069 "lblock: %lu, depth: %d pblock %lld",
4070 (unsigned long) map->m_lblk, depth,
4071 path[depth].p_block);
4072 err = -EFSCORRUPTED;
4073 goto out;
4074 }
4075
4076 ex = path[depth].p_ext;
4077 if (ex) {
4078 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4079 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4080 unsigned short ee_len;
4081
4082
4083 /*
4084 * unwritten extents are treated as holes, except that
4085 * we split out initialized portions during a write.
4086 */
4087 ee_len = ext4_ext_get_actual_len(ex);
4088
4089 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4090
4091 /* if found extent covers block, simply return it */
4092 if (in_range(map->m_lblk, ee_block, ee_len)) {
4093 newblock = map->m_lblk - ee_block + ee_start;
4094 /* number of remaining blocks in the extent */
4095 allocated = ee_len - (map->m_lblk - ee_block);
4096 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4097 map->m_lblk, ee_block, ee_len, newblock);
4098
4099 /*
4100 * If the extent is initialized check whether the
4101 * caller wants to convert it to unwritten.
4102 */
4103 if ((!ext4_ext_is_unwritten(ex)) &&
4104 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4105 err = convert_initialized_extent(handle,
4106 inode, map, &path, &allocated);
4107 goto out;
4108 } else if (!ext4_ext_is_unwritten(ex)) {
4109 map->m_flags |= EXT4_MAP_MAPPED;
4110 map->m_pblk = newblock;
4111 if (allocated > map->m_len)
4112 allocated = map->m_len;
4113 map->m_len = allocated;
4114 ext4_ext_show_leaf(inode, path);
4115 goto out;
4116 }
4117
4118 ret = ext4_ext_handle_unwritten_extents(
4119 handle, inode, map, &path, flags,
4120 allocated, newblock);
4121 if (ret < 0)
4122 err = ret;
4123 else
4124 allocated = ret;
4125 goto out;
4126 }
4127 }
4128
4129 /*
4130 * requested block isn't allocated yet;
4131 * we couldn't try to create block if create flag is zero
4132 */
4133 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4134 ext4_lblk_t hole_start, hole_len;
4135
4136 hole_start = map->m_lblk;
4137 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4138 /*
4139 * put just found gap into cache to speed up
4140 * subsequent requests
4141 */
4142 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4143
4144 /* Update hole_len to reflect hole size after map->m_lblk */
4145 if (hole_start != map->m_lblk)
4146 hole_len -= map->m_lblk - hole_start;
4147 map->m_pblk = 0;
4148 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4149
4150 goto out;
4151 }
4152
4153 /*
4154 * Okay, we need to do block allocation.
4155 */
4156 newex.ee_block = cpu_to_le32(map->m_lblk);
4157 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4158
4159 /*
4160 * If we are doing bigalloc, check to see if the extent returned
4161 * by ext4_find_extent() implies a cluster we can use.
4162 */
4163 if (cluster_offset && ex &&
4164 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4165 ar.len = allocated = map->m_len;
4166 newblock = map->m_pblk;
4167 goto got_allocated_blocks;
4168 }
4169
4170 /* find neighbour allocated blocks */
4171 ar.lleft = map->m_lblk;
4172 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4173 if (err)
4174 goto out;
4175 ar.lright = map->m_lblk;
4176 ex2 = NULL;
4177 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4178 if (err)
4179 goto out;
4180
4181 /* Check if the extent after searching to the right implies a
4182 * cluster we can use. */
4183 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4184 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4185 ar.len = allocated = map->m_len;
4186 newblock = map->m_pblk;
4187 goto got_allocated_blocks;
4188 }
4189
4190 /*
4191 * See if request is beyond maximum number of blocks we can have in
4192 * a single extent. For an initialized extent this limit is
4193 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4194 * EXT_UNWRITTEN_MAX_LEN.
4195 */
4196 if (map->m_len > EXT_INIT_MAX_LEN &&
4197 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4198 map->m_len = EXT_INIT_MAX_LEN;
4199 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4200 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4201 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4202
4203 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4204 newex.ee_len = cpu_to_le16(map->m_len);
4205 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4206 if (err)
4207 allocated = ext4_ext_get_actual_len(&newex);
4208 else
4209 allocated = map->m_len;
4210
4211 /* allocate new block */
4212 ar.inode = inode;
4213 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4214 ar.logical = map->m_lblk;
4215 /*
4216 * We calculate the offset from the beginning of the cluster
4217 * for the logical block number, since when we allocate a
4218 * physical cluster, the physical block should start at the
4219 * same offset from the beginning of the cluster. This is
4220 * needed so that future calls to get_implied_cluster_alloc()
4221 * work correctly.
4222 */
4223 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4224 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4225 ar.goal -= offset;
4226 ar.logical -= offset;
4227 if (S_ISREG(inode->i_mode))
4228 ar.flags = EXT4_MB_HINT_DATA;
4229 else
4230 /* disable in-core preallocation for non-regular files */
4231 ar.flags = 0;
4232 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4233 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4234 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4235 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4236 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4237 ar.flags |= EXT4_MB_USE_RESERVED;
4238 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4239 if (!newblock)
4240 goto out;
4241 allocated_clusters = ar.len;
4242 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4243 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4244 ar.goal, newblock, ar.len, allocated);
4245 if (ar.len > allocated)
4246 ar.len = allocated;
4247
4248got_allocated_blocks:
4249 /* try to insert new extent into found leaf and return */
4250 pblk = newblock + offset;
4251 ext4_ext_store_pblock(&newex, pblk);
4252 newex.ee_len = cpu_to_le16(ar.len);
4253 /* Mark unwritten */
4254 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4255 ext4_ext_mark_unwritten(&newex);
4256 map->m_flags |= EXT4_MAP_UNWRITTEN;
4257 }
4258
4259 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4260 if (err) {
4261 if (allocated_clusters) {
4262 int fb_flags = 0;
4263
4264 /*
4265 * free data blocks we just allocated.
4266 * not a good idea to call discard here directly,
4267 * but otherwise we'd need to call it every free().
4268 */
4269 ext4_discard_preallocations(inode, 0);
4270 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4271 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4272 ext4_free_blocks(handle, inode, NULL, newblock,
4273 EXT4_C2B(sbi, allocated_clusters),
4274 fb_flags);
4275 }
4276 goto out;
4277 }
4278
4279 /*
4280 * Reduce the reserved cluster count to reflect successful deferred
4281 * allocation of delayed allocated clusters or direct allocation of
4282 * clusters discovered to be delayed allocated. Once allocated, a
4283 * cluster is not included in the reserved count.
4284 */
4285 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4286 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4287 /*
4288 * When allocating delayed allocated clusters, simply
4289 * reduce the reserved cluster count and claim quota
4290 */
4291 ext4_da_update_reserve_space(inode, allocated_clusters,
4292 1);
4293 } else {
4294 ext4_lblk_t lblk, len;
4295 unsigned int n;
4296
4297 /*
4298 * When allocating non-delayed allocated clusters
4299 * (from fallocate, filemap, DIO, or clusters
4300 * allocated when delalloc has been disabled by
4301 * ext4_nonda_switch), reduce the reserved cluster
4302 * count by the number of allocated clusters that
4303 * have previously been delayed allocated. Quota
4304 * has been claimed by ext4_mb_new_blocks() above,
4305 * so release the quota reservations made for any
4306 * previously delayed allocated clusters.
4307 */
4308 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4309 len = allocated_clusters << sbi->s_cluster_bits;
4310 n = ext4_es_delayed_clu(inode, lblk, len);
4311 if (n > 0)
4312 ext4_da_update_reserve_space(inode, (int) n, 0);
4313 }
4314 }
4315
4316 /*
4317 * Cache the extent and update transaction to commit on fdatasync only
4318 * when it is _not_ an unwritten extent.
4319 */
4320 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4321 ext4_update_inode_fsync_trans(handle, inode, 1);
4322 else
4323 ext4_update_inode_fsync_trans(handle, inode, 0);
4324
4325 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4326 map->m_pblk = pblk;
4327 map->m_len = ar.len;
4328 allocated = map->m_len;
4329 ext4_ext_show_leaf(inode, path);
4330
4331out:
4332 ext4_ext_drop_refs(path);
4333 kfree(path);
4334
4335 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4336 err ? err : allocated);
4337 return err ? err : allocated;
4338}
4339
4340int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4341{
4342 struct super_block *sb = inode->i_sb;
4343 ext4_lblk_t last_block;
4344 int err = 0;
4345
4346 /*
4347 * TODO: optimization is possible here.
4348 * Probably we need not scan at all,
4349 * because page truncation is enough.
4350 */
4351
4352 /* we have to know where to truncate from in crash case */
4353 EXT4_I(inode)->i_disksize = inode->i_size;
4354 err = ext4_mark_inode_dirty(handle, inode);
4355 if (err)
4356 return err;
4357
4358 last_block = (inode->i_size + sb->s_blocksize - 1)
4359 >> EXT4_BLOCK_SIZE_BITS(sb);
4360retry:
4361 err = ext4_es_remove_extent(inode, last_block,
4362 EXT_MAX_BLOCKS - last_block);
4363 if (err == -ENOMEM) {
4364 cond_resched();
4365 congestion_wait(BLK_RW_ASYNC, HZ/50);
4366 goto retry;
4367 }
4368 if (err)
4369 return err;
4370retry_remove_space:
4371 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4372 if (err == -ENOMEM) {
4373 cond_resched();
4374 congestion_wait(BLK_RW_ASYNC, HZ/50);
4375 goto retry_remove_space;
4376 }
4377 return err;
4378}
4379
4380static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4381 ext4_lblk_t len, loff_t new_size,
4382 int flags)
4383{
4384 struct inode *inode = file_inode(file);
4385 handle_t *handle;
4386 int ret = 0;
4387 int ret2 = 0, ret3 = 0;
4388 int retries = 0;
4389 int depth = 0;
4390 struct ext4_map_blocks map;
4391 unsigned int credits;
4392 loff_t epos;
4393
4394 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4395 map.m_lblk = offset;
4396 map.m_len = len;
4397 /*
4398 * Don't normalize the request if it can fit in one extent so
4399 * that it doesn't get unnecessarily split into multiple
4400 * extents.
4401 */
4402 if (len <= EXT_UNWRITTEN_MAX_LEN)
4403 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4404
4405 /*
4406 * credits to insert 1 extent into extent tree
4407 */
4408 credits = ext4_chunk_trans_blocks(inode, len);
4409 depth = ext_depth(inode);
4410
4411retry:
4412 while (ret >= 0 && len) {
4413 /*
4414 * Recalculate credits when extent tree depth changes.
4415 */
4416 if (depth != ext_depth(inode)) {
4417 credits = ext4_chunk_trans_blocks(inode, len);
4418 depth = ext_depth(inode);
4419 }
4420
4421 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4422 credits);
4423 if (IS_ERR(handle)) {
4424 ret = PTR_ERR(handle);
4425 break;
4426 }
4427 ret = ext4_map_blocks(handle, inode, &map, flags);
4428 if (ret <= 0) {
4429 ext4_debug("inode #%lu: block %u: len %u: "
4430 "ext4_ext_map_blocks returned %d",
4431 inode->i_ino, map.m_lblk,
4432 map.m_len, ret);
4433 ext4_mark_inode_dirty(handle, inode);
4434 ret2 = ext4_journal_stop(handle);
4435 break;
4436 }
4437 map.m_lblk += ret;
4438 map.m_len = len = len - ret;
4439 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4440 inode->i_ctime = current_time(inode);
4441 if (new_size) {
4442 if (epos > new_size)
4443 epos = new_size;
4444 if (ext4_update_inode_size(inode, epos) & 0x1)
4445 inode->i_mtime = inode->i_ctime;
4446 }
4447 ret2 = ext4_mark_inode_dirty(handle, inode);
4448 ext4_update_inode_fsync_trans(handle, inode, 1);
4449 ret3 = ext4_journal_stop(handle);
4450 ret2 = ret3 ? ret3 : ret2;
4451 if (unlikely(ret2))
4452 break;
4453 }
4454 if (ret == -ENOSPC &&
4455 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4456 ret = 0;
4457 goto retry;
4458 }
4459
4460 return ret > 0 ? ret2 : ret;
4461}
4462
4463static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4464
4465static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4466
4467static long ext4_zero_range(struct file *file, loff_t offset,
4468 loff_t len, int mode)
4469{
4470 struct inode *inode = file_inode(file);
4471 handle_t *handle = NULL;
4472 unsigned int max_blocks;
4473 loff_t new_size = 0;
4474 int ret = 0;
4475 int flags;
4476 int credits;
4477 int partial_begin, partial_end;
4478 loff_t start, end;
4479 ext4_lblk_t lblk;
4480 unsigned int blkbits = inode->i_blkbits;
4481
4482 trace_ext4_zero_range(inode, offset, len, mode);
4483
4484 /* Call ext4_force_commit to flush all data in case of data=journal. */
4485 if (ext4_should_journal_data(inode)) {
4486 ret = ext4_force_commit(inode->i_sb);
4487 if (ret)
4488 return ret;
4489 }
4490
4491 /*
4492 * Round up offset. This is not fallocate, we need to zero out
4493 * blocks, so convert interior block aligned part of the range to
4494 * unwritten and possibly manually zero out unaligned parts of the
4495 * range.
4496 */
4497 start = round_up(offset, 1 << blkbits);
4498 end = round_down((offset + len), 1 << blkbits);
4499
4500 if (start < offset || end > offset + len)
4501 return -EINVAL;
4502 partial_begin = offset & ((1 << blkbits) - 1);
4503 partial_end = (offset + len) & ((1 << blkbits) - 1);
4504
4505 lblk = start >> blkbits;
4506 max_blocks = (end >> blkbits);
4507 if (max_blocks < lblk)
4508 max_blocks = 0;
4509 else
4510 max_blocks -= lblk;
4511
4512 inode_lock(inode);
4513
4514 /*
4515 * Indirect files do not support unwritten extents
4516 */
4517 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4518 ret = -EOPNOTSUPP;
4519 goto out_mutex;
4520 }
4521
4522 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4523 (offset + len > inode->i_size ||
4524 offset + len > EXT4_I(inode)->i_disksize)) {
4525 new_size = offset + len;
4526 ret = inode_newsize_ok(inode, new_size);
4527 if (ret)
4528 goto out_mutex;
4529 }
4530
4531 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4532
4533 /* Wait all existing dio workers, newcomers will block on i_mutex */
4534 inode_dio_wait(inode);
4535
4536 /* Preallocate the range including the unaligned edges */
4537 if (partial_begin || partial_end) {
4538 ret = ext4_alloc_file_blocks(file,
4539 round_down(offset, 1 << blkbits) >> blkbits,
4540 (round_up((offset + len), 1 << blkbits) -
4541 round_down(offset, 1 << blkbits)) >> blkbits,
4542 new_size, flags);
4543 if (ret)
4544 goto out_mutex;
4545
4546 }
4547
4548 /* Zero range excluding the unaligned edges */
4549 if (max_blocks > 0) {
4550 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4551 EXT4_EX_NOCACHE);
4552
4553 /*
4554 * Prevent page faults from reinstantiating pages we have
4555 * released from page cache.
4556 */
4557 down_write(&EXT4_I(inode)->i_mmap_sem);
4558
4559 ret = ext4_break_layouts(inode);
4560 if (ret) {
4561 up_write(&EXT4_I(inode)->i_mmap_sem);
4562 goto out_mutex;
4563 }
4564
4565 ret = ext4_update_disksize_before_punch(inode, offset, len);
4566 if (ret) {
4567 up_write(&EXT4_I(inode)->i_mmap_sem);
4568 goto out_mutex;
4569 }
4570 /* Now release the pages and zero block aligned part of pages */
4571 truncate_pagecache_range(inode, start, end - 1);
4572 inode->i_mtime = inode->i_ctime = current_time(inode);
4573
4574 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4575 flags);
4576 up_write(&EXT4_I(inode)->i_mmap_sem);
4577 if (ret)
4578 goto out_mutex;
4579 }
4580 if (!partial_begin && !partial_end)
4581 goto out_mutex;
4582
4583 /*
4584 * In worst case we have to writeout two nonadjacent unwritten
4585 * blocks and update the inode
4586 */
4587 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4588 if (ext4_should_journal_data(inode))
4589 credits += 2;
4590 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4591 if (IS_ERR(handle)) {
4592 ret = PTR_ERR(handle);
4593 ext4_std_error(inode->i_sb, ret);
4594 goto out_mutex;
4595 }
4596
4597 inode->i_mtime = inode->i_ctime = current_time(inode);
4598 if (new_size)
4599 ext4_update_inode_size(inode, new_size);
4600 ret = ext4_mark_inode_dirty(handle, inode);
4601 if (unlikely(ret))
4602 goto out_handle;
4603
4604 /* Zero out partial block at the edges of the range */
4605 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4606 if (ret >= 0)
4607 ext4_update_inode_fsync_trans(handle, inode, 1);
4608
4609 if (file->f_flags & O_SYNC)
4610 ext4_handle_sync(handle);
4611
4612out_handle:
4613 ext4_journal_stop(handle);
4614out_mutex:
4615 inode_unlock(inode);
4616 return ret;
4617}
4618
4619/*
4620 * preallocate space for a file. This implements ext4's fallocate file
4621 * operation, which gets called from sys_fallocate system call.
4622 * For block-mapped files, posix_fallocate should fall back to the method
4623 * of writing zeroes to the required new blocks (the same behavior which is
4624 * expected for file systems which do not support fallocate() system call).
4625 */
4626long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4627{
4628 struct inode *inode = file_inode(file);
4629 loff_t new_size = 0;
4630 unsigned int max_blocks;
4631 int ret = 0;
4632 int flags;
4633 ext4_lblk_t lblk;
4634 unsigned int blkbits = inode->i_blkbits;
4635
4636 /*
4637 * Encrypted inodes can't handle collapse range or insert
4638 * range since we would need to re-encrypt blocks with a
4639 * different IV or XTS tweak (which are based on the logical
4640 * block number).
4641 */
4642 if (IS_ENCRYPTED(inode) &&
4643 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4644 return -EOPNOTSUPP;
4645
4646 /* Return error if mode is not supported */
4647 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4648 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4649 FALLOC_FL_INSERT_RANGE))
4650 return -EOPNOTSUPP;
4651
4652 if (mode & FALLOC_FL_PUNCH_HOLE)
4653 return ext4_punch_hole(inode, offset, len);
4654
4655 ret = ext4_convert_inline_data(inode);
4656 if (ret)
4657 return ret;
4658
4659 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4660 return ext4_collapse_range(inode, offset, len);
4661
4662 if (mode & FALLOC_FL_INSERT_RANGE)
4663 return ext4_insert_range(inode, offset, len);
4664
4665 if (mode & FALLOC_FL_ZERO_RANGE)
4666 return ext4_zero_range(file, offset, len, mode);
4667
4668 trace_ext4_fallocate_enter(inode, offset, len, mode);
4669 lblk = offset >> blkbits;
4670
4671 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4672 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4673
4674 inode_lock(inode);
4675
4676 /*
4677 * We only support preallocation for extent-based files only
4678 */
4679 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4680 ret = -EOPNOTSUPP;
4681 goto out;
4682 }
4683
4684 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4685 (offset + len > inode->i_size ||
4686 offset + len > EXT4_I(inode)->i_disksize)) {
4687 new_size = offset + len;
4688 ret = inode_newsize_ok(inode, new_size);
4689 if (ret)
4690 goto out;
4691 }
4692
4693 /* Wait all existing dio workers, newcomers will block on i_mutex */
4694 inode_dio_wait(inode);
4695
4696 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4697 if (ret)
4698 goto out;
4699
4700 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4701 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4702 EXT4_I(inode)->i_sync_tid);
4703 }
4704out:
4705 inode_unlock(inode);
4706 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4707 return ret;
4708}
4709
4710/*
4711 * This function convert a range of blocks to written extents
4712 * The caller of this function will pass the start offset and the size.
4713 * all unwritten extents within this range will be converted to
4714 * written extents.
4715 *
4716 * This function is called from the direct IO end io call back
4717 * function, to convert the fallocated extents after IO is completed.
4718 * Returns 0 on success.
4719 */
4720int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4721 loff_t offset, ssize_t len)
4722{
4723 unsigned int max_blocks;
4724 int ret = 0, ret2 = 0, ret3 = 0;
4725 struct ext4_map_blocks map;
4726 unsigned int blkbits = inode->i_blkbits;
4727 unsigned int credits = 0;
4728
4729 map.m_lblk = offset >> blkbits;
4730 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4731
4732 if (!handle) {
4733 /*
4734 * credits to insert 1 extent into extent tree
4735 */
4736 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4737 }
4738 while (ret >= 0 && ret < max_blocks) {
4739 map.m_lblk += ret;
4740 map.m_len = (max_blocks -= ret);
4741 if (credits) {
4742 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4743 credits);
4744 if (IS_ERR(handle)) {
4745 ret = PTR_ERR(handle);
4746 break;
4747 }
4748 }
4749 ret = ext4_map_blocks(handle, inode, &map,
4750 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4751 if (ret <= 0)
4752 ext4_warning(inode->i_sb,
4753 "inode #%lu: block %u: len %u: "
4754 "ext4_ext_map_blocks returned %d",
4755 inode->i_ino, map.m_lblk,
4756 map.m_len, ret);
4757 ret2 = ext4_mark_inode_dirty(handle, inode);
4758 if (credits) {
4759 ret3 = ext4_journal_stop(handle);
4760 if (unlikely(ret3))
4761 ret2 = ret3;
4762 }
4763
4764 if (ret <= 0 || ret2)
4765 break;
4766 }
4767 return ret > 0 ? ret2 : ret;
4768}
4769
4770int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4771{
4772 int ret, err = 0;
4773 struct ext4_io_end_vec *io_end_vec;
4774
4775 /*
4776 * This is somewhat ugly but the idea is clear: When transaction is
4777 * reserved, everything goes into it. Otherwise we rather start several
4778 * smaller transactions for conversion of each extent separately.
4779 */
4780 if (handle) {
4781 handle = ext4_journal_start_reserved(handle,
4782 EXT4_HT_EXT_CONVERT);
4783 if (IS_ERR(handle))
4784 return PTR_ERR(handle);
4785 }
4786
4787 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4788 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4789 io_end_vec->offset,
4790 io_end_vec->size);
4791 if (ret)
4792 break;
4793 }
4794
4795 if (handle)
4796 err = ext4_journal_stop(handle);
4797
4798 return ret < 0 ? ret : err;
4799}
4800
4801static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4802{
4803 __u64 physical = 0;
4804 __u64 length = 0;
4805 int blockbits = inode->i_sb->s_blocksize_bits;
4806 int error = 0;
4807 u16 iomap_type;
4808
4809 /* in-inode? */
4810 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4811 struct ext4_iloc iloc;
4812 int offset; /* offset of xattr in inode */
4813
4814 error = ext4_get_inode_loc(inode, &iloc);
4815 if (error)
4816 return error;
4817 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4818 offset = EXT4_GOOD_OLD_INODE_SIZE +
4819 EXT4_I(inode)->i_extra_isize;
4820 physical += offset;
4821 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4822 brelse(iloc.bh);
4823 iomap_type = IOMAP_INLINE;
4824 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4825 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4826 length = inode->i_sb->s_blocksize;
4827 iomap_type = IOMAP_MAPPED;
4828 } else {
4829 /* no in-inode or external block for xattr, so return -ENOENT */
4830 error = -ENOENT;
4831 goto out;
4832 }
4833
4834 iomap->addr = physical;
4835 iomap->offset = 0;
4836 iomap->length = length;
4837 iomap->type = iomap_type;
4838 iomap->flags = 0;
4839out:
4840 return error;
4841}
4842
4843static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4844 loff_t length, unsigned flags,
4845 struct iomap *iomap, struct iomap *srcmap)
4846{
4847 int error;
4848
4849 error = ext4_iomap_xattr_fiemap(inode, iomap);
4850 if (error == 0 && (offset >= iomap->length))
4851 error = -ENOENT;
4852 return error;
4853}
4854
4855static const struct iomap_ops ext4_iomap_xattr_ops = {
4856 .iomap_begin = ext4_iomap_xattr_begin,
4857};
4858
4859static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4860{
4861 u64 maxbytes;
4862
4863 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4864 maxbytes = inode->i_sb->s_maxbytes;
4865 else
4866 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4867
4868 if (*len == 0)
4869 return -EINVAL;
4870 if (start > maxbytes)
4871 return -EFBIG;
4872
4873 /*
4874 * Shrink request scope to what the fs can actually handle.
4875 */
4876 if (*len > maxbytes || (maxbytes - *len) < start)
4877 *len = maxbytes - start;
4878 return 0;
4879}
4880
4881int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4882 u64 start, u64 len)
4883{
4884 int error = 0;
4885
4886 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4887 error = ext4_ext_precache(inode);
4888 if (error)
4889 return error;
4890 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4891 }
4892
4893 /*
4894 * For bitmap files the maximum size limit could be smaller than
4895 * s_maxbytes, so check len here manually instead of just relying on the
4896 * generic check.
4897 */
4898 error = ext4_fiemap_check_ranges(inode, start, &len);
4899 if (error)
4900 return error;
4901
4902 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4903 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4904 return iomap_fiemap(inode, fieinfo, start, len,
4905 &ext4_iomap_xattr_ops);
4906 }
4907
4908 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4909}
4910
4911int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4912 __u64 start, __u64 len)
4913{
4914 ext4_lblk_t start_blk, len_blks;
4915 __u64 last_blk;
4916 int error = 0;
4917
4918 if (ext4_has_inline_data(inode)) {
4919 int has_inline;
4920
4921 down_read(&EXT4_I(inode)->xattr_sem);
4922 has_inline = ext4_has_inline_data(inode);
4923 up_read(&EXT4_I(inode)->xattr_sem);
4924 if (has_inline)
4925 return 0;
4926 }
4927
4928 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4929 error = ext4_ext_precache(inode);
4930 if (error)
4931 return error;
4932 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4933 }
4934
4935 error = fiemap_prep(inode, fieinfo, start, &len, 0);
4936 if (error)
4937 return error;
4938
4939 error = ext4_fiemap_check_ranges(inode, start, &len);
4940 if (error)
4941 return error;
4942
4943 start_blk = start >> inode->i_sb->s_blocksize_bits;
4944 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4945 if (last_blk >= EXT_MAX_BLOCKS)
4946 last_blk = EXT_MAX_BLOCKS-1;
4947 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4948
4949 /*
4950 * Walk the extent tree gathering extent information
4951 * and pushing extents back to the user.
4952 */
4953 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
4954}
4955
4956/*
4957 * ext4_access_path:
4958 * Function to access the path buffer for marking it dirty.
4959 * It also checks if there are sufficient credits left in the journal handle
4960 * to update path.
4961 */
4962static int
4963ext4_access_path(handle_t *handle, struct inode *inode,
4964 struct ext4_ext_path *path)
4965{
4966 int credits, err;
4967
4968 if (!ext4_handle_valid(handle))
4969 return 0;
4970
4971 /*
4972 * Check if need to extend journal credits
4973 * 3 for leaf, sb, and inode plus 2 (bmap and group
4974 * descriptor) for each block group; assume two block
4975 * groups
4976 */
4977 credits = ext4_writepage_trans_blocks(inode);
4978 err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
4979 if (err < 0)
4980 return err;
4981
4982 err = ext4_ext_get_access(handle, inode, path);
4983 return err;
4984}
4985
4986/*
4987 * ext4_ext_shift_path_extents:
4988 * Shift the extents of a path structure lying between path[depth].p_ext
4989 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
4990 * if it is right shift or left shift operation.
4991 */
4992static int
4993ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
4994 struct inode *inode, handle_t *handle,
4995 enum SHIFT_DIRECTION SHIFT)
4996{
4997 int depth, err = 0;
4998 struct ext4_extent *ex_start, *ex_last;
4999 bool update = false;
5000 depth = path->p_depth;
5001
5002 while (depth >= 0) {
5003 if (depth == path->p_depth) {
5004 ex_start = path[depth].p_ext;
5005 if (!ex_start)
5006 return -EFSCORRUPTED;
5007
5008 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5009
5010 err = ext4_access_path(handle, inode, path + depth);
5011 if (err)
5012 goto out;
5013
5014 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5015 update = true;
5016
5017 while (ex_start <= ex_last) {
5018 if (SHIFT == SHIFT_LEFT) {
5019 le32_add_cpu(&ex_start->ee_block,
5020 -shift);
5021 /* Try to merge to the left. */
5022 if ((ex_start >
5023 EXT_FIRST_EXTENT(path[depth].p_hdr))
5024 &&
5025 ext4_ext_try_to_merge_right(inode,
5026 path, ex_start - 1))
5027 ex_last--;
5028 else
5029 ex_start++;
5030 } else {
5031 le32_add_cpu(&ex_last->ee_block, shift);
5032 ext4_ext_try_to_merge_right(inode, path,
5033 ex_last);
5034 ex_last--;
5035 }
5036 }
5037 err = ext4_ext_dirty(handle, inode, path + depth);
5038 if (err)
5039 goto out;
5040
5041 if (--depth < 0 || !update)
5042 break;
5043 }
5044
5045 /* Update index too */
5046 err = ext4_access_path(handle, inode, path + depth);
5047 if (err)
5048 goto out;
5049
5050 if (SHIFT == SHIFT_LEFT)
5051 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5052 else
5053 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5054 err = ext4_ext_dirty(handle, inode, path + depth);
5055 if (err)
5056 goto out;
5057
5058 /* we are done if current index is not a starting index */
5059 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5060 break;
5061
5062 depth--;
5063 }
5064
5065out:
5066 return err;
5067}
5068
5069/*
5070 * ext4_ext_shift_extents:
5071 * All the extents which lies in the range from @start to the last allocated
5072 * block for the @inode are shifted either towards left or right (depending
5073 * upon @SHIFT) by @shift blocks.
5074 * On success, 0 is returned, error otherwise.
5075 */
5076static int
5077ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5078 ext4_lblk_t start, ext4_lblk_t shift,
5079 enum SHIFT_DIRECTION SHIFT)
5080{
5081 struct ext4_ext_path *path;
5082 int ret = 0, depth;
5083 struct ext4_extent *extent;
5084 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5085
5086 /* Let path point to the last extent */
5087 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5088 EXT4_EX_NOCACHE);
5089 if (IS_ERR(path))
5090 return PTR_ERR(path);
5091
5092 depth = path->p_depth;
5093 extent = path[depth].p_ext;
5094 if (!extent)
5095 goto out;
5096
5097 stop = le32_to_cpu(extent->ee_block);
5098
5099 /*
5100 * For left shifts, make sure the hole on the left is big enough to
5101 * accommodate the shift. For right shifts, make sure the last extent
5102 * won't be shifted beyond EXT_MAX_BLOCKS.
5103 */
5104 if (SHIFT == SHIFT_LEFT) {
5105 path = ext4_find_extent(inode, start - 1, &path,
5106 EXT4_EX_NOCACHE);
5107 if (IS_ERR(path))
5108 return PTR_ERR(path);
5109 depth = path->p_depth;
5110 extent = path[depth].p_ext;
5111 if (extent) {
5112 ex_start = le32_to_cpu(extent->ee_block);
5113 ex_end = le32_to_cpu(extent->ee_block) +
5114 ext4_ext_get_actual_len(extent);
5115 } else {
5116 ex_start = 0;
5117 ex_end = 0;
5118 }
5119
5120 if ((start == ex_start && shift > ex_start) ||
5121 (shift > start - ex_end)) {
5122 ret = -EINVAL;
5123 goto out;
5124 }
5125 } else {
5126 if (shift > EXT_MAX_BLOCKS -
5127 (stop + ext4_ext_get_actual_len(extent))) {
5128 ret = -EINVAL;
5129 goto out;
5130 }
5131 }
5132
5133 /*
5134 * In case of left shift, iterator points to start and it is increased
5135 * till we reach stop. In case of right shift, iterator points to stop
5136 * and it is decreased till we reach start.
5137 */
5138 if (SHIFT == SHIFT_LEFT)
5139 iterator = &start;
5140 else
5141 iterator = &stop;
5142
5143 /*
5144 * Its safe to start updating extents. Start and stop are unsigned, so
5145 * in case of right shift if extent with 0 block is reached, iterator
5146 * becomes NULL to indicate the end of the loop.
5147 */
5148 while (iterator && start <= stop) {
5149 path = ext4_find_extent(inode, *iterator, &path,
5150 EXT4_EX_NOCACHE);
5151 if (IS_ERR(path))
5152 return PTR_ERR(path);
5153 depth = path->p_depth;
5154 extent = path[depth].p_ext;
5155 if (!extent) {
5156 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5157 (unsigned long) *iterator);
5158 return -EFSCORRUPTED;
5159 }
5160 if (SHIFT == SHIFT_LEFT && *iterator >
5161 le32_to_cpu(extent->ee_block)) {
5162 /* Hole, move to the next extent */
5163 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5164 path[depth].p_ext++;
5165 } else {
5166 *iterator = ext4_ext_next_allocated_block(path);
5167 continue;
5168 }
5169 }
5170
5171 if (SHIFT == SHIFT_LEFT) {
5172 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5173 *iterator = le32_to_cpu(extent->ee_block) +
5174 ext4_ext_get_actual_len(extent);
5175 } else {
5176 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5177 if (le32_to_cpu(extent->ee_block) > 0)
5178 *iterator = le32_to_cpu(extent->ee_block) - 1;
5179 else
5180 /* Beginning is reached, end of the loop */
5181 iterator = NULL;
5182 /* Update path extent in case we need to stop */
5183 while (le32_to_cpu(extent->ee_block) < start)
5184 extent++;
5185 path[depth].p_ext = extent;
5186 }
5187 ret = ext4_ext_shift_path_extents(path, shift, inode,
5188 handle, SHIFT);
5189 if (ret)
5190 break;
5191 }
5192out:
5193 ext4_ext_drop_refs(path);
5194 kfree(path);
5195 return ret;
5196}
5197
5198/*
5199 * ext4_collapse_range:
5200 * This implements the fallocate's collapse range functionality for ext4
5201 * Returns: 0 and non-zero on error.
5202 */
5203static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5204{
5205 struct super_block *sb = inode->i_sb;
5206 ext4_lblk_t punch_start, punch_stop;
5207 handle_t *handle;
5208 unsigned int credits;
5209 loff_t new_size, ioffset;
5210 int ret;
5211
5212 /*
5213 * We need to test this early because xfstests assumes that a
5214 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5215 * system does not support collapse range.
5216 */
5217 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5218 return -EOPNOTSUPP;
5219
5220 /* Collapse range works only on fs cluster size aligned regions. */
5221 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5222 return -EINVAL;
5223
5224 trace_ext4_collapse_range(inode, offset, len);
5225
5226 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5227 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5228
5229 /* Call ext4_force_commit to flush all data in case of data=journal. */
5230 if (ext4_should_journal_data(inode)) {
5231 ret = ext4_force_commit(inode->i_sb);
5232 if (ret)
5233 return ret;
5234 }
5235
5236 inode_lock(inode);
5237 /*
5238 * There is no need to overlap collapse range with EOF, in which case
5239 * it is effectively a truncate operation
5240 */
5241 if (offset + len >= inode->i_size) {
5242 ret = -EINVAL;
5243 goto out_mutex;
5244 }
5245
5246 /* Currently just for extent based files */
5247 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5248 ret = -EOPNOTSUPP;
5249 goto out_mutex;
5250 }
5251
5252 /* Wait for existing dio to complete */
5253 inode_dio_wait(inode);
5254
5255 /*
5256 * Prevent page faults from reinstantiating pages we have released from
5257 * page cache.
5258 */
5259 down_write(&EXT4_I(inode)->i_mmap_sem);
5260
5261 ret = ext4_break_layouts(inode);
5262 if (ret)
5263 goto out_mmap;
5264
5265 /*
5266 * Need to round down offset to be aligned with page size boundary
5267 * for page size > block size.
5268 */
5269 ioffset = round_down(offset, PAGE_SIZE);
5270 /*
5271 * Write tail of the last page before removed range since it will get
5272 * removed from the page cache below.
5273 */
5274 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5275 if (ret)
5276 goto out_mmap;
5277 /*
5278 * Write data that will be shifted to preserve them when discarding
5279 * page cache below. We are also protected from pages becoming dirty
5280 * by i_mmap_sem.
5281 */
5282 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5283 LLONG_MAX);
5284 if (ret)
5285 goto out_mmap;
5286 truncate_pagecache(inode, ioffset);
5287
5288 credits = ext4_writepage_trans_blocks(inode);
5289 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5290 if (IS_ERR(handle)) {
5291 ret = PTR_ERR(handle);
5292 goto out_mmap;
5293 }
5294
5295 down_write(&EXT4_I(inode)->i_data_sem);
5296 ext4_discard_preallocations(inode, 0);
5297
5298 ret = ext4_es_remove_extent(inode, punch_start,
5299 EXT_MAX_BLOCKS - punch_start);
5300 if (ret) {
5301 up_write(&EXT4_I(inode)->i_data_sem);
5302 goto out_stop;
5303 }
5304
5305 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5306 if (ret) {
5307 up_write(&EXT4_I(inode)->i_data_sem);
5308 goto out_stop;
5309 }
5310 ext4_discard_preallocations(inode, 0);
5311
5312 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5313 punch_stop - punch_start, SHIFT_LEFT);
5314 if (ret) {
5315 up_write(&EXT4_I(inode)->i_data_sem);
5316 goto out_stop;
5317 }
5318
5319 new_size = inode->i_size - len;
5320 i_size_write(inode, new_size);
5321 EXT4_I(inode)->i_disksize = new_size;
5322
5323 up_write(&EXT4_I(inode)->i_data_sem);
5324 if (IS_SYNC(inode))
5325 ext4_handle_sync(handle);
5326 inode->i_mtime = inode->i_ctime = current_time(inode);
5327 ret = ext4_mark_inode_dirty(handle, inode);
5328 ext4_update_inode_fsync_trans(handle, inode, 1);
5329
5330out_stop:
5331 ext4_journal_stop(handle);
5332out_mmap:
5333 up_write(&EXT4_I(inode)->i_mmap_sem);
5334out_mutex:
5335 inode_unlock(inode);
5336 return ret;
5337}
5338
5339/*
5340 * ext4_insert_range:
5341 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5342 * The data blocks starting from @offset to the EOF are shifted by @len
5343 * towards right to create a hole in the @inode. Inode size is increased
5344 * by len bytes.
5345 * Returns 0 on success, error otherwise.
5346 */
5347static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5348{
5349 struct super_block *sb = inode->i_sb;
5350 handle_t *handle;
5351 struct ext4_ext_path *path;
5352 struct ext4_extent *extent;
5353 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5354 unsigned int credits, ee_len;
5355 int ret = 0, depth, split_flag = 0;
5356 loff_t ioffset;
5357
5358 /*
5359 * We need to test this early because xfstests assumes that an
5360 * insert range of (0, 1) will return EOPNOTSUPP if the file
5361 * system does not support insert range.
5362 */
5363 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5364 return -EOPNOTSUPP;
5365
5366 /* Insert range works only on fs cluster size aligned regions. */
5367 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5368 return -EINVAL;
5369
5370 trace_ext4_insert_range(inode, offset, len);
5371
5372 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5373 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5374
5375 /* Call ext4_force_commit to flush all data in case of data=journal */
5376 if (ext4_should_journal_data(inode)) {
5377 ret = ext4_force_commit(inode->i_sb);
5378 if (ret)
5379 return ret;
5380 }
5381
5382 inode_lock(inode);
5383 /* Currently just for extent based files */
5384 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5385 ret = -EOPNOTSUPP;
5386 goto out_mutex;
5387 }
5388
5389 /* Check whether the maximum file size would be exceeded */
5390 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5391 ret = -EFBIG;
5392 goto out_mutex;
5393 }
5394
5395 /* Offset must be less than i_size */
5396 if (offset >= inode->i_size) {
5397 ret = -EINVAL;
5398 goto out_mutex;
5399 }
5400
5401 /* Wait for existing dio to complete */
5402 inode_dio_wait(inode);
5403
5404 /*
5405 * Prevent page faults from reinstantiating pages we have released from
5406 * page cache.
5407 */
5408 down_write(&EXT4_I(inode)->i_mmap_sem);
5409
5410 ret = ext4_break_layouts(inode);
5411 if (ret)
5412 goto out_mmap;
5413
5414 /*
5415 * Need to round down to align start offset to page size boundary
5416 * for page size > block size.
5417 */
5418 ioffset = round_down(offset, PAGE_SIZE);
5419 /* Write out all dirty pages */
5420 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5421 LLONG_MAX);
5422 if (ret)
5423 goto out_mmap;
5424 truncate_pagecache(inode, ioffset);
5425
5426 credits = ext4_writepage_trans_blocks(inode);
5427 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5428 if (IS_ERR(handle)) {
5429 ret = PTR_ERR(handle);
5430 goto out_mmap;
5431 }
5432
5433 /* Expand file to avoid data loss if there is error while shifting */
5434 inode->i_size += len;
5435 EXT4_I(inode)->i_disksize += len;
5436 inode->i_mtime = inode->i_ctime = current_time(inode);
5437 ret = ext4_mark_inode_dirty(handle, inode);
5438 if (ret)
5439 goto out_stop;
5440
5441 down_write(&EXT4_I(inode)->i_data_sem);
5442 ext4_discard_preallocations(inode, 0);
5443
5444 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5445 if (IS_ERR(path)) {
5446 up_write(&EXT4_I(inode)->i_data_sem);
5447 goto out_stop;
5448 }
5449
5450 depth = ext_depth(inode);
5451 extent = path[depth].p_ext;
5452 if (extent) {
5453 ee_start_lblk = le32_to_cpu(extent->ee_block);
5454 ee_len = ext4_ext_get_actual_len(extent);
5455
5456 /*
5457 * If offset_lblk is not the starting block of extent, split
5458 * the extent @offset_lblk
5459 */
5460 if ((offset_lblk > ee_start_lblk) &&
5461 (offset_lblk < (ee_start_lblk + ee_len))) {
5462 if (ext4_ext_is_unwritten(extent))
5463 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5464 EXT4_EXT_MARK_UNWRIT2;
5465 ret = ext4_split_extent_at(handle, inode, &path,
5466 offset_lblk, split_flag,
5467 EXT4_EX_NOCACHE |
5468 EXT4_GET_BLOCKS_PRE_IO |
5469 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5470 }
5471
5472 ext4_ext_drop_refs(path);
5473 kfree(path);
5474 if (ret < 0) {
5475 up_write(&EXT4_I(inode)->i_data_sem);
5476 goto out_stop;
5477 }
5478 } else {
5479 ext4_ext_drop_refs(path);
5480 kfree(path);
5481 }
5482
5483 ret = ext4_es_remove_extent(inode, offset_lblk,
5484 EXT_MAX_BLOCKS - offset_lblk);
5485 if (ret) {
5486 up_write(&EXT4_I(inode)->i_data_sem);
5487 goto out_stop;
5488 }
5489
5490 /*
5491 * if offset_lblk lies in a hole which is at start of file, use
5492 * ee_start_lblk to shift extents
5493 */
5494 ret = ext4_ext_shift_extents(inode, handle,
5495 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5496 len_lblk, SHIFT_RIGHT);
5497
5498 up_write(&EXT4_I(inode)->i_data_sem);
5499 if (IS_SYNC(inode))
5500 ext4_handle_sync(handle);
5501 if (ret >= 0)
5502 ext4_update_inode_fsync_trans(handle, inode, 1);
5503
5504out_stop:
5505 ext4_journal_stop(handle);
5506out_mmap:
5507 up_write(&EXT4_I(inode)->i_mmap_sem);
5508out_mutex:
5509 inode_unlock(inode);
5510 return ret;
5511}
5512
5513/**
5514 * ext4_swap_extents() - Swap extents between two inodes
5515 * @handle: handle for this transaction
5516 * @inode1: First inode
5517 * @inode2: Second inode
5518 * @lblk1: Start block for first inode
5519 * @lblk2: Start block for second inode
5520 * @count: Number of blocks to swap
5521 * @unwritten: Mark second inode's extents as unwritten after swap
5522 * @erp: Pointer to save error value
5523 *
5524 * This helper routine does exactly what is promise "swap extents". All other
5525 * stuff such as page-cache locking consistency, bh mapping consistency or
5526 * extent's data copying must be performed by caller.
5527 * Locking:
5528 * i_mutex is held for both inodes
5529 * i_data_sem is locked for write for both inodes
5530 * Assumptions:
5531 * All pages from requested range are locked for both inodes
5532 */
5533int
5534ext4_swap_extents(handle_t *handle, struct inode *inode1,
5535 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5536 ext4_lblk_t count, int unwritten, int *erp)
5537{
5538 struct ext4_ext_path *path1 = NULL;
5539 struct ext4_ext_path *path2 = NULL;
5540 int replaced_count = 0;
5541
5542 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5543 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5544 BUG_ON(!inode_is_locked(inode1));
5545 BUG_ON(!inode_is_locked(inode2));
5546
5547 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5548 if (unlikely(*erp))
5549 return 0;
5550 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5551 if (unlikely(*erp))
5552 return 0;
5553
5554 while (count) {
5555 struct ext4_extent *ex1, *ex2, tmp_ex;
5556 ext4_lblk_t e1_blk, e2_blk;
5557 int e1_len, e2_len, len;
5558 int split = 0;
5559
5560 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5561 if (IS_ERR(path1)) {
5562 *erp = PTR_ERR(path1);
5563 path1 = NULL;
5564 finish:
5565 count = 0;
5566 goto repeat;
5567 }
5568 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5569 if (IS_ERR(path2)) {
5570 *erp = PTR_ERR(path2);
5571 path2 = NULL;
5572 goto finish;
5573 }
5574 ex1 = path1[path1->p_depth].p_ext;
5575 ex2 = path2[path2->p_depth].p_ext;
5576 /* Do we have something to swap ? */
5577 if (unlikely(!ex2 || !ex1))
5578 goto finish;
5579
5580 e1_blk = le32_to_cpu(ex1->ee_block);
5581 e2_blk = le32_to_cpu(ex2->ee_block);
5582 e1_len = ext4_ext_get_actual_len(ex1);
5583 e2_len = ext4_ext_get_actual_len(ex2);
5584
5585 /* Hole handling */
5586 if (!in_range(lblk1, e1_blk, e1_len) ||
5587 !in_range(lblk2, e2_blk, e2_len)) {
5588 ext4_lblk_t next1, next2;
5589
5590 /* if hole after extent, then go to next extent */
5591 next1 = ext4_ext_next_allocated_block(path1);
5592 next2 = ext4_ext_next_allocated_block(path2);
5593 /* If hole before extent, then shift to that extent */
5594 if (e1_blk > lblk1)
5595 next1 = e1_blk;
5596 if (e2_blk > lblk2)
5597 next2 = e2_blk;
5598 /* Do we have something to swap */
5599 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5600 goto finish;
5601 /* Move to the rightest boundary */
5602 len = next1 - lblk1;
5603 if (len < next2 - lblk2)
5604 len = next2 - lblk2;
5605 if (len > count)
5606 len = count;
5607 lblk1 += len;
5608 lblk2 += len;
5609 count -= len;
5610 goto repeat;
5611 }
5612
5613 /* Prepare left boundary */
5614 if (e1_blk < lblk1) {
5615 split = 1;
5616 *erp = ext4_force_split_extent_at(handle, inode1,
5617 &path1, lblk1, 0);
5618 if (unlikely(*erp))
5619 goto finish;
5620 }
5621 if (e2_blk < lblk2) {
5622 split = 1;
5623 *erp = ext4_force_split_extent_at(handle, inode2,
5624 &path2, lblk2, 0);
5625 if (unlikely(*erp))
5626 goto finish;
5627 }
5628 /* ext4_split_extent_at() may result in leaf extent split,
5629 * path must to be revalidated. */
5630 if (split)
5631 goto repeat;
5632
5633 /* Prepare right boundary */
5634 len = count;
5635 if (len > e1_blk + e1_len - lblk1)
5636 len = e1_blk + e1_len - lblk1;
5637 if (len > e2_blk + e2_len - lblk2)
5638 len = e2_blk + e2_len - lblk2;
5639
5640 if (len != e1_len) {
5641 split = 1;
5642 *erp = ext4_force_split_extent_at(handle, inode1,
5643 &path1, lblk1 + len, 0);
5644 if (unlikely(*erp))
5645 goto finish;
5646 }
5647 if (len != e2_len) {
5648 split = 1;
5649 *erp = ext4_force_split_extent_at(handle, inode2,
5650 &path2, lblk2 + len, 0);
5651 if (*erp)
5652 goto finish;
5653 }
5654 /* ext4_split_extent_at() may result in leaf extent split,
5655 * path must to be revalidated. */
5656 if (split)
5657 goto repeat;
5658
5659 BUG_ON(e2_len != e1_len);
5660 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5661 if (unlikely(*erp))
5662 goto finish;
5663 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5664 if (unlikely(*erp))
5665 goto finish;
5666
5667 /* Both extents are fully inside boundaries. Swap it now */
5668 tmp_ex = *ex1;
5669 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5670 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5671 ex1->ee_len = cpu_to_le16(e2_len);
5672 ex2->ee_len = cpu_to_le16(e1_len);
5673 if (unwritten)
5674 ext4_ext_mark_unwritten(ex2);
5675 if (ext4_ext_is_unwritten(&tmp_ex))
5676 ext4_ext_mark_unwritten(ex1);
5677
5678 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5679 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5680 *erp = ext4_ext_dirty(handle, inode2, path2 +
5681 path2->p_depth);
5682 if (unlikely(*erp))
5683 goto finish;
5684 *erp = ext4_ext_dirty(handle, inode1, path1 +
5685 path1->p_depth);
5686 /*
5687 * Looks scarry ah..? second inode already points to new blocks,
5688 * and it was successfully dirtied. But luckily error may happen
5689 * only due to journal error, so full transaction will be
5690 * aborted anyway.
5691 */
5692 if (unlikely(*erp))
5693 goto finish;
5694 lblk1 += len;
5695 lblk2 += len;
5696 replaced_count += len;
5697 count -= len;
5698
5699 repeat:
5700 ext4_ext_drop_refs(path1);
5701 kfree(path1);
5702 ext4_ext_drop_refs(path2);
5703 kfree(path2);
5704 path1 = path2 = NULL;
5705 }
5706 return replaced_count;
5707}
5708
5709/*
5710 * ext4_clu_mapped - determine whether any block in a logical cluster has
5711 * been mapped to a physical cluster
5712 *
5713 * @inode - file containing the logical cluster
5714 * @lclu - logical cluster of interest
5715 *
5716 * Returns 1 if any block in the logical cluster is mapped, signifying
5717 * that a physical cluster has been allocated for it. Otherwise,
5718 * returns 0. Can also return negative error codes. Derived from
5719 * ext4_ext_map_blocks().
5720 */
5721int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5722{
5723 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5724 struct ext4_ext_path *path;
5725 int depth, mapped = 0, err = 0;
5726 struct ext4_extent *extent;
5727 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5728
5729 /* search for the extent closest to the first block in the cluster */
5730 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5731 if (IS_ERR(path)) {
5732 err = PTR_ERR(path);
5733 path = NULL;
5734 goto out;
5735 }
5736
5737 depth = ext_depth(inode);
5738
5739 /*
5740 * A consistent leaf must not be empty. This situation is possible,
5741 * though, _during_ tree modification, and it's why an assert can't
5742 * be put in ext4_find_extent().
5743 */
5744 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5745 EXT4_ERROR_INODE(inode,
5746 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5747 (unsigned long) EXT4_C2B(sbi, lclu),
5748 depth, path[depth].p_block);
5749 err = -EFSCORRUPTED;
5750 goto out;
5751 }
5752
5753 extent = path[depth].p_ext;
5754
5755 /* can't be mapped if the extent tree is empty */
5756 if (extent == NULL)
5757 goto out;
5758
5759 first_lblk = le32_to_cpu(extent->ee_block);
5760 first_lclu = EXT4_B2C(sbi, first_lblk);
5761
5762 /*
5763 * Three possible outcomes at this point - found extent spanning
5764 * the target cluster, to the left of the target cluster, or to the
5765 * right of the target cluster. The first two cases are handled here.
5766 * The last case indicates the target cluster is not mapped.
5767 */
5768 if (lclu >= first_lclu) {
5769 last_lclu = EXT4_B2C(sbi, first_lblk +
5770 ext4_ext_get_actual_len(extent) - 1);
5771 if (lclu <= last_lclu) {
5772 mapped = 1;
5773 } else {
5774 first_lblk = ext4_ext_next_allocated_block(path);
5775 first_lclu = EXT4_B2C(sbi, first_lblk);
5776 if (lclu == first_lclu)
5777 mapped = 1;
5778 }
5779 }
5780
5781out:
5782 ext4_ext_drop_refs(path);
5783 kfree(path);
5784
5785 return err ? err : mapped;
5786}