1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
6 * Architecture independence:
7 * Copyright (c) 2005, Bull S.A.
8 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
12 * Extents support for EXT4
15 * - ext4*_error() should be used in some situations
16 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17 * - smart tree reduction
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"
36 #include <trace/events/ext4.h>
39 * used by extent splitting.
41 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
43 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
44 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
46 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
47 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
49 static __le32 ext4_extent_block_csum(struct inode *inode,
50 struct ext4_extent_header *eh)
52 struct ext4_inode_info *ei = EXT4_I(inode);
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
57 EXT4_EXTENT_TAIL_OFFSET(eh));
58 return cpu_to_le32(csum);
61 static int ext4_extent_block_csum_verify(struct inode *inode,
62 struct ext4_extent_header *eh)
64 struct ext4_extent_tail *et;
66 if (!ext4_has_metadata_csum(inode->i_sb))
69 et = find_ext4_extent_tail(eh);
70 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
75 static void ext4_extent_block_csum_set(struct inode *inode,
76 struct ext4_extent_header *eh)
78 struct ext4_extent_tail *et;
80 if (!ext4_has_metadata_csum(inode->i_sb))
83 et = find_ext4_extent_tail(eh);
84 et->et_checksum = ext4_extent_block_csum(inode, eh);
87 static int ext4_split_extent_at(handle_t *handle,
89 struct ext4_ext_path **ppath,
94 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
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.
102 BUG_ON(EXT4_JOURNAL(inode) == NULL);
103 ext4_discard_preallocations(inode, 0);
104 up_write(&EXT4_I(inode)->i_data_sem);
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.
114 * The function returns 0 on success, 1 if transaction had to be restarted,
115 * and < 0 in case of fatal error.
117 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
118 int check_cred, int restart_cred,
124 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
125 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
127 down_write(&EXT4_I(inode)->i_data_sem);
136 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
137 struct ext4_ext_path *path)
142 /* path points to block */
143 BUFFER_TRACE(path->p_bh, "get_write_access");
144 err = ext4_journal_get_write_access(handle, path->p_bh);
146 * The extent buffer's verified bit will be set again in
147 * __ext4_ext_dirty(). We could leave an inconsistent
148 * buffer if the extents updating procudure break off du
149 * to some error happens, force to check it again.
152 clear_buffer_verified(path->p_bh);
154 /* path points to leaf/index in inode body */
155 /* we use in-core data, no need to protect them */
165 static int __ext4_ext_dirty(const char *where, unsigned int line,
166 handle_t *handle, struct inode *inode,
167 struct ext4_ext_path *path)
171 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
173 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
174 /* path points to block */
175 err = __ext4_handle_dirty_metadata(where, line, handle,
177 /* Extents updating done, re-set verified flag */
179 set_buffer_verified(path->p_bh);
181 /* path points to leaf/index in inode body */
182 err = ext4_mark_inode_dirty(handle, inode);
187 #define ext4_ext_dirty(handle, inode, path) \
188 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
190 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
191 struct ext4_ext_path *path,
195 int depth = path->p_depth;
196 struct ext4_extent *ex;
199 * Try to predict block placement assuming that we are
200 * filling in a file which will eventually be
201 * non-sparse --- i.e., in the case of libbfd writing
202 * an ELF object sections out-of-order but in a way
203 * the eventually results in a contiguous object or
204 * executable file, or some database extending a table
205 * space file. However, this is actually somewhat
206 * non-ideal if we are writing a sparse file such as
207 * qemu or KVM writing a raw image file that is going
208 * to stay fairly sparse, since it will end up
209 * fragmenting the file system's free space. Maybe we
210 * should have some hueristics or some way to allow
211 * userspace to pass a hint to file system,
212 * especially if the latter case turns out to be
215 ex = path[depth].p_ext;
217 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
218 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
220 if (block > ext_block)
221 return ext_pblk + (block - ext_block);
223 return ext_pblk - (ext_block - block);
226 /* it looks like index is empty;
227 * try to find starting block from index itself */
228 if (path[depth].p_bh)
229 return path[depth].p_bh->b_blocknr;
232 /* OK. use inode's group */
233 return ext4_inode_to_goal_block(inode);
237 * Allocation for a meta data block
240 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
241 struct ext4_ext_path *path,
242 struct ext4_extent *ex, int *err, unsigned int flags)
244 ext4_fsblk_t goal, newblock;
246 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
247 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
252 static inline int ext4_ext_space_block(struct inode *inode, int check)
256 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
257 / sizeof(struct ext4_extent);
258 #ifdef AGGRESSIVE_TEST
259 if (!check && size > 6)
265 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
269 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
270 / sizeof(struct ext4_extent_idx);
271 #ifdef AGGRESSIVE_TEST
272 if (!check && size > 5)
278 static inline int ext4_ext_space_root(struct inode *inode, int check)
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent);
285 #ifdef AGGRESSIVE_TEST
286 if (!check && size > 3)
292 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
296 size = sizeof(EXT4_I(inode)->i_data);
297 size -= sizeof(struct ext4_extent_header);
298 size /= sizeof(struct ext4_extent_idx);
299 #ifdef AGGRESSIVE_TEST
300 if (!check && size > 4)
307 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
308 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
311 struct ext4_ext_path *path = *ppath;
312 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
313 int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
316 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
318 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
319 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
324 ext4_ext_max_entries(struct inode *inode, int depth)
328 if (depth == ext_depth(inode)) {
330 max = ext4_ext_space_root(inode, 1);
332 max = ext4_ext_space_root_idx(inode, 1);
335 max = ext4_ext_space_block(inode, 1);
337 max = ext4_ext_space_block_idx(inode, 1);
343 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
345 ext4_fsblk_t block = ext4_ext_pblock(ext);
346 int len = ext4_ext_get_actual_len(ext);
347 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
352 * - overflow/wrap-around
354 if (lblock + len <= lblock)
356 return ext4_inode_block_valid(inode, block, len);
359 static int ext4_valid_extent_idx(struct inode *inode,
360 struct ext4_extent_idx *ext_idx)
362 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
364 return ext4_inode_block_valid(inode, block, 1);
367 static int ext4_valid_extent_entries(struct inode *inode,
368 struct ext4_extent_header *eh,
369 ext4_lblk_t lblk, ext4_fsblk_t *pblk,
372 unsigned short entries;
373 ext4_lblk_t lblock = 0;
376 if (eh->eh_entries == 0)
379 entries = le16_to_cpu(eh->eh_entries);
383 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
386 * The logical block in the first entry should equal to
387 * the number in the index block.
389 if (depth != ext_depth(inode) &&
390 lblk != le32_to_cpu(ext->ee_block))
393 if (!ext4_valid_extent(inode, ext))
396 /* Check for overlapping extents */
397 lblock = le32_to_cpu(ext->ee_block);
399 *pblk = ext4_ext_pblock(ext);
402 cur = lblock + ext4_ext_get_actual_len(ext);
407 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
410 * The logical block in the first entry should equal to
411 * the number in the parent index block.
413 if (depth != ext_depth(inode) &&
414 lblk != le32_to_cpu(ext_idx->ei_block))
417 if (!ext4_valid_extent_idx(inode, ext_idx))
420 /* Check for overlapping index extents */
421 lblock = le32_to_cpu(ext_idx->ei_block);
423 *pblk = ext4_idx_pblock(ext_idx);
434 static int __ext4_ext_check(const char *function, unsigned int line,
435 struct inode *inode, struct ext4_extent_header *eh,
436 int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
438 const char *error_msg;
439 int max = 0, err = -EFSCORRUPTED;
441 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
442 error_msg = "invalid magic";
445 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
446 error_msg = "unexpected eh_depth";
449 if (unlikely(eh->eh_max == 0)) {
450 error_msg = "invalid eh_max";
453 max = ext4_ext_max_entries(inode, depth);
454 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
455 error_msg = "too large eh_max";
458 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
459 error_msg = "invalid eh_entries";
462 if (unlikely((eh->eh_entries == 0) && (depth > 0))) {
463 error_msg = "eh_entries is 0 but eh_depth is > 0";
466 if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
467 error_msg = "invalid extent entries";
470 if (unlikely(depth > 32)) {
471 error_msg = "too large eh_depth";
474 /* Verify checksum on non-root extent tree nodes */
475 if (ext_depth(inode) != depth &&
476 !ext4_extent_block_csum_verify(inode, eh)) {
477 error_msg = "extent tree corrupted";
484 ext4_error_inode_err(inode, function, line, 0, -err,
485 "pblk %llu bad header/extent: %s - magic %x, "
486 "entries %u, max %u(%u), depth %u(%u)",
487 (unsigned long long) pblk, error_msg,
488 le16_to_cpu(eh->eh_magic),
489 le16_to_cpu(eh->eh_entries),
490 le16_to_cpu(eh->eh_max),
491 max, le16_to_cpu(eh->eh_depth), depth);
495 #define ext4_ext_check(inode, eh, depth, pblk) \
496 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
498 int ext4_ext_check_inode(struct inode *inode)
500 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
503 static void ext4_cache_extents(struct inode *inode,
504 struct ext4_extent_header *eh)
506 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
507 ext4_lblk_t prev = 0;
510 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
511 unsigned int status = EXTENT_STATUS_WRITTEN;
512 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
513 int len = ext4_ext_get_actual_len(ex);
515 if (prev && (prev != lblk))
516 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
519 if (ext4_ext_is_unwritten(ex))
520 status = EXTENT_STATUS_UNWRITTEN;
521 ext4_es_cache_extent(inode, lblk, len,
522 ext4_ext_pblock(ex), status);
527 static struct buffer_head *
528 __read_extent_tree_block(const char *function, unsigned int line,
529 struct inode *inode, struct ext4_extent_idx *idx,
530 int depth, int flags)
532 struct buffer_head *bh;
534 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
537 if (flags & EXT4_EX_NOFAIL)
538 gfp_flags |= __GFP_NOFAIL;
540 pblk = ext4_idx_pblock(idx);
541 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
543 return ERR_PTR(-ENOMEM);
545 if (!bh_uptodate_or_lock(bh)) {
546 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
547 err = ext4_read_bh(bh, 0, NULL);
551 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
553 err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
554 depth, pblk, le32_to_cpu(idx->ei_block));
557 set_buffer_verified(bh);
559 * If this is a leaf block, cache all of its entries
561 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
562 struct ext4_extent_header *eh = ext_block_hdr(bh);
563 ext4_cache_extents(inode, eh);
572 #define read_extent_tree_block(inode, idx, depth, flags) \
573 __read_extent_tree_block(__func__, __LINE__, (inode), (idx), \
577 * This function is called to cache a file's extent information in the
580 int ext4_ext_precache(struct inode *inode)
582 struct ext4_inode_info *ei = EXT4_I(inode);
583 struct ext4_ext_path *path = NULL;
584 struct buffer_head *bh;
585 int i = 0, depth, ret = 0;
587 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
588 return 0; /* not an extent-mapped inode */
590 down_read(&ei->i_data_sem);
591 depth = ext_depth(inode);
593 /* Don't cache anything if there are no external extent blocks */
595 up_read(&ei->i_data_sem);
599 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
602 up_read(&ei->i_data_sem);
606 path[0].p_hdr = ext_inode_hdr(inode);
607 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
610 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
613 * If this is a leaf block or we've reached the end of
614 * the index block, go up
617 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
618 brelse(path[i].p_bh);
623 bh = read_extent_tree_block(inode, path[i].p_idx++,
625 EXT4_EX_FORCE_CACHE);
632 path[i].p_hdr = ext_block_hdr(bh);
633 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
635 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
637 up_read(&ei->i_data_sem);
638 ext4_ext_drop_refs(path);
644 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
646 int k, l = path->p_depth;
648 ext_debug(inode, "path:");
649 for (k = 0; k <= l; k++, path++) {
651 ext_debug(inode, " %d->%llu",
652 le32_to_cpu(path->p_idx->ei_block),
653 ext4_idx_pblock(path->p_idx));
654 } else if (path->p_ext) {
655 ext_debug(inode, " %d:[%d]%d:%llu ",
656 le32_to_cpu(path->p_ext->ee_block),
657 ext4_ext_is_unwritten(path->p_ext),
658 ext4_ext_get_actual_len(path->p_ext),
659 ext4_ext_pblock(path->p_ext));
661 ext_debug(inode, " []");
663 ext_debug(inode, "\n");
666 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
668 int depth = ext_depth(inode);
669 struct ext4_extent_header *eh;
670 struct ext4_extent *ex;
676 eh = path[depth].p_hdr;
677 ex = EXT_FIRST_EXTENT(eh);
679 ext_debug(inode, "Displaying leaf extents\n");
681 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
682 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
683 ext4_ext_is_unwritten(ex),
684 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
686 ext_debug(inode, "\n");
689 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
690 ext4_fsblk_t newblock, int level)
692 int depth = ext_depth(inode);
693 struct ext4_extent *ex;
695 if (depth != level) {
696 struct ext4_extent_idx *idx;
697 idx = path[level].p_idx;
698 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
699 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
700 level, le32_to_cpu(idx->ei_block),
701 ext4_idx_pblock(idx), newblock);
708 ex = path[depth].p_ext;
709 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
710 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
711 le32_to_cpu(ex->ee_block),
713 ext4_ext_is_unwritten(ex),
714 ext4_ext_get_actual_len(ex),
721 #define ext4_ext_show_path(inode, path)
722 #define ext4_ext_show_leaf(inode, path)
723 #define ext4_ext_show_move(inode, path, newblock, level)
726 void ext4_ext_drop_refs(struct ext4_ext_path *path)
732 depth = path->p_depth;
733 for (i = 0; i <= depth; i++, path++) {
740 * ext4_ext_binsearch_idx:
741 * binary search for the closest index of the given block
742 * the header must be checked before calling this
745 ext4_ext_binsearch_idx(struct inode *inode,
746 struct ext4_ext_path *path, ext4_lblk_t block)
748 struct ext4_extent_header *eh = path->p_hdr;
749 struct ext4_extent_idx *r, *l, *m;
752 ext_debug(inode, "binsearch for %u(idx): ", block);
754 l = EXT_FIRST_INDEX(eh) + 1;
755 r = EXT_LAST_INDEX(eh);
758 if (block < le32_to_cpu(m->ei_block))
762 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
763 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
764 r, le32_to_cpu(r->ei_block));
768 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
769 ext4_idx_pblock(path->p_idx));
771 #ifdef CHECK_BINSEARCH
773 struct ext4_extent_idx *chix, *ix;
776 chix = ix = EXT_FIRST_INDEX(eh);
777 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
778 if (k != 0 && le32_to_cpu(ix->ei_block) <=
779 le32_to_cpu(ix[-1].ei_block)) {
780 printk(KERN_DEBUG "k=%d, ix=0x%p, "
782 ix, EXT_FIRST_INDEX(eh));
783 printk(KERN_DEBUG "%u <= %u\n",
784 le32_to_cpu(ix->ei_block),
785 le32_to_cpu(ix[-1].ei_block));
787 BUG_ON(k && le32_to_cpu(ix->ei_block)
788 <= le32_to_cpu(ix[-1].ei_block));
789 if (block < le32_to_cpu(ix->ei_block))
793 BUG_ON(chix != path->p_idx);
800 * ext4_ext_binsearch:
801 * binary search for closest extent of the given block
802 * the header must be checked before calling this
805 ext4_ext_binsearch(struct inode *inode,
806 struct ext4_ext_path *path, ext4_lblk_t block)
808 struct ext4_extent_header *eh = path->p_hdr;
809 struct ext4_extent *r, *l, *m;
811 if (eh->eh_entries == 0) {
813 * this leaf is empty:
814 * we get such a leaf in split/add case
819 ext_debug(inode, "binsearch for %u: ", block);
821 l = EXT_FIRST_EXTENT(eh) + 1;
822 r = EXT_LAST_EXTENT(eh);
826 if (block < le32_to_cpu(m->ee_block))
830 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
831 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
832 r, le32_to_cpu(r->ee_block));
836 ext_debug(inode, " -> %d:%llu:[%d]%d ",
837 le32_to_cpu(path->p_ext->ee_block),
838 ext4_ext_pblock(path->p_ext),
839 ext4_ext_is_unwritten(path->p_ext),
840 ext4_ext_get_actual_len(path->p_ext));
842 #ifdef CHECK_BINSEARCH
844 struct ext4_extent *chex, *ex;
847 chex = ex = EXT_FIRST_EXTENT(eh);
848 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
849 BUG_ON(k && le32_to_cpu(ex->ee_block)
850 <= le32_to_cpu(ex[-1].ee_block));
851 if (block < le32_to_cpu(ex->ee_block))
855 BUG_ON(chex != path->p_ext);
861 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
863 struct ext4_extent_header *eh;
865 eh = ext_inode_hdr(inode);
868 eh->eh_magic = EXT4_EXT_MAGIC;
869 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
870 eh->eh_generation = 0;
871 ext4_mark_inode_dirty(handle, inode);
874 struct ext4_ext_path *
875 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
876 struct ext4_ext_path **orig_path, int flags)
878 struct ext4_extent_header *eh;
879 struct buffer_head *bh;
880 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
881 short int depth, i, ppos = 0;
883 gfp_t gfp_flags = GFP_NOFS;
885 if (flags & EXT4_EX_NOFAIL)
886 gfp_flags |= __GFP_NOFAIL;
888 eh = ext_inode_hdr(inode);
889 depth = ext_depth(inode);
890 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
891 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
898 ext4_ext_drop_refs(path);
899 if (depth > path[0].p_maxdepth) {
901 *orig_path = path = NULL;
905 /* account possible depth increase */
906 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
909 return ERR_PTR(-ENOMEM);
910 path[0].p_maxdepth = depth + 1;
916 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
917 ext4_cache_extents(inode, eh);
918 /* walk through the tree */
920 ext_debug(inode, "depth %d: num %d, max %d\n",
921 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
923 ext4_ext_binsearch_idx(inode, path + ppos, block);
924 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
925 path[ppos].p_depth = i;
926 path[ppos].p_ext = NULL;
928 bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
934 eh = ext_block_hdr(bh);
936 path[ppos].p_bh = bh;
937 path[ppos].p_hdr = eh;
940 path[ppos].p_depth = i;
941 path[ppos].p_ext = NULL;
942 path[ppos].p_idx = NULL;
945 ext4_ext_binsearch(inode, path + ppos, block);
946 /* if not an empty leaf */
947 if (path[ppos].p_ext)
948 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
950 ext4_ext_show_path(inode, path);
955 ext4_ext_drop_refs(path);
963 * ext4_ext_insert_index:
964 * insert new index [@logical;@ptr] into the block at @curp;
965 * check where to insert: before @curp or after @curp
967 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
968 struct ext4_ext_path *curp,
969 int logical, ext4_fsblk_t ptr)
971 struct ext4_extent_idx *ix;
974 err = ext4_ext_get_access(handle, inode, curp);
978 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
979 EXT4_ERROR_INODE(inode,
980 "logical %d == ei_block %d!",
981 logical, le32_to_cpu(curp->p_idx->ei_block));
982 return -EFSCORRUPTED;
985 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
986 >= le16_to_cpu(curp->p_hdr->eh_max))) {
987 EXT4_ERROR_INODE(inode,
988 "eh_entries %d >= eh_max %d!",
989 le16_to_cpu(curp->p_hdr->eh_entries),
990 le16_to_cpu(curp->p_hdr->eh_max));
991 return -EFSCORRUPTED;
994 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
996 ext_debug(inode, "insert new index %d after: %llu\n",
998 ix = curp->p_idx + 1;
1001 ext_debug(inode, "insert new index %d before: %llu\n",
1006 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1007 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1008 return -EFSCORRUPTED;
1011 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1014 ext_debug(inode, "insert new index %d: "
1015 "move %d indices from 0x%p to 0x%p\n",
1016 logical, len, ix, ix + 1);
1017 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1020 ix->ei_block = cpu_to_le32(logical);
1021 ext4_idx_store_pblock(ix, ptr);
1022 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1024 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1025 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1026 return -EFSCORRUPTED;
1029 err = ext4_ext_dirty(handle, inode, curp);
1030 ext4_std_error(inode->i_sb, err);
1037 * inserts new subtree into the path, using free index entry
1039 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1040 * - makes decision where to split
1041 * - moves remaining extents and index entries (right to the split point)
1042 * into the newly allocated blocks
1043 * - initializes subtree
1045 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1047 struct ext4_ext_path *path,
1048 struct ext4_extent *newext, int at)
1050 struct buffer_head *bh = NULL;
1051 int depth = ext_depth(inode);
1052 struct ext4_extent_header *neh;
1053 struct ext4_extent_idx *fidx;
1054 int i = at, k, m, a;
1055 ext4_fsblk_t newblock, oldblock;
1057 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1058 gfp_t gfp_flags = GFP_NOFS;
1060 size_t ext_size = 0;
1062 if (flags & EXT4_EX_NOFAIL)
1063 gfp_flags |= __GFP_NOFAIL;
1065 /* make decision: where to split? */
1066 /* FIXME: now decision is simplest: at current extent */
1068 /* if current leaf will be split, then we should use
1069 * border from split point */
1070 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1071 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1072 return -EFSCORRUPTED;
1074 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1075 border = path[depth].p_ext[1].ee_block;
1076 ext_debug(inode, "leaf will be split."
1077 " next leaf starts at %d\n",
1078 le32_to_cpu(border));
1080 border = newext->ee_block;
1081 ext_debug(inode, "leaf will be added."
1082 " next leaf starts at %d\n",
1083 le32_to_cpu(border));
1087 * If error occurs, then we break processing
1088 * and mark filesystem read-only. index won't
1089 * be inserted and tree will be in consistent
1090 * state. Next mount will repair buffers too.
1094 * Get array to track all allocated blocks.
1095 * We need this to handle errors and free blocks
1098 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1102 /* allocate all needed blocks */
1103 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1104 for (a = 0; a < depth - at; a++) {
1105 newblock = ext4_ext_new_meta_block(handle, inode, path,
1106 newext, &err, flags);
1109 ablocks[a] = newblock;
1112 /* initialize new leaf */
1113 newblock = ablocks[--a];
1114 if (unlikely(newblock == 0)) {
1115 EXT4_ERROR_INODE(inode, "newblock == 0!");
1116 err = -EFSCORRUPTED;
1119 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1120 if (unlikely(!bh)) {
1126 err = ext4_journal_get_create_access(handle, bh);
1130 neh = ext_block_hdr(bh);
1131 neh->eh_entries = 0;
1132 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1133 neh->eh_magic = EXT4_EXT_MAGIC;
1135 neh->eh_generation = 0;
1137 /* move remainder of path[depth] to the new leaf */
1138 if (unlikely(path[depth].p_hdr->eh_entries !=
1139 path[depth].p_hdr->eh_max)) {
1140 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1141 path[depth].p_hdr->eh_entries,
1142 path[depth].p_hdr->eh_max);
1143 err = -EFSCORRUPTED;
1146 /* start copy from next extent */
1147 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1148 ext4_ext_show_move(inode, path, newblock, depth);
1150 struct ext4_extent *ex;
1151 ex = EXT_FIRST_EXTENT(neh);
1152 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1153 le16_add_cpu(&neh->eh_entries, m);
1156 /* zero out unused area in the extent block */
1157 ext_size = sizeof(struct ext4_extent_header) +
1158 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1159 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1160 ext4_extent_block_csum_set(inode, neh);
1161 set_buffer_uptodate(bh);
1164 err = ext4_handle_dirty_metadata(handle, inode, bh);
1170 /* correct old leaf */
1172 err = ext4_ext_get_access(handle, inode, path + depth);
1175 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1176 err = ext4_ext_dirty(handle, inode, path + depth);
1182 /* create intermediate indexes */
1184 if (unlikely(k < 0)) {
1185 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1186 err = -EFSCORRUPTED;
1190 ext_debug(inode, "create %d intermediate indices\n", k);
1191 /* insert new index into current index block */
1192 /* current depth stored in i var */
1195 oldblock = newblock;
1196 newblock = ablocks[--a];
1197 bh = sb_getblk(inode->i_sb, newblock);
1198 if (unlikely(!bh)) {
1204 err = ext4_journal_get_create_access(handle, bh);
1208 neh = ext_block_hdr(bh);
1209 neh->eh_entries = cpu_to_le16(1);
1210 neh->eh_magic = EXT4_EXT_MAGIC;
1211 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1212 neh->eh_depth = cpu_to_le16(depth - i);
1213 neh->eh_generation = 0;
1214 fidx = EXT_FIRST_INDEX(neh);
1215 fidx->ei_block = border;
1216 ext4_idx_store_pblock(fidx, oldblock);
1218 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1219 i, newblock, le32_to_cpu(border), oldblock);
1221 /* move remainder of path[i] to the new index block */
1222 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1223 EXT_LAST_INDEX(path[i].p_hdr))) {
1224 EXT4_ERROR_INODE(inode,
1225 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1226 le32_to_cpu(path[i].p_ext->ee_block));
1227 err = -EFSCORRUPTED;
1230 /* start copy indexes */
1231 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1232 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1233 EXT_MAX_INDEX(path[i].p_hdr));
1234 ext4_ext_show_move(inode, path, newblock, i);
1236 memmove(++fidx, path[i].p_idx,
1237 sizeof(struct ext4_extent_idx) * m);
1238 le16_add_cpu(&neh->eh_entries, m);
1240 /* zero out unused area in the extent block */
1241 ext_size = sizeof(struct ext4_extent_header) +
1242 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1243 memset(bh->b_data + ext_size, 0,
1244 inode->i_sb->s_blocksize - ext_size);
1245 ext4_extent_block_csum_set(inode, neh);
1246 set_buffer_uptodate(bh);
1249 err = ext4_handle_dirty_metadata(handle, inode, bh);
1255 /* correct old index */
1257 err = ext4_ext_get_access(handle, inode, path + i);
1260 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1261 err = ext4_ext_dirty(handle, inode, path + i);
1269 /* insert new index */
1270 err = ext4_ext_insert_index(handle, inode, path + at,
1271 le32_to_cpu(border), newblock);
1275 if (buffer_locked(bh))
1281 /* free all allocated blocks in error case */
1282 for (i = 0; i < depth; i++) {
1285 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1286 EXT4_FREE_BLOCKS_METADATA);
1295 * ext4_ext_grow_indepth:
1296 * implements tree growing procedure:
1297 * - allocates new block
1298 * - moves top-level data (index block or leaf) into the new block
1299 * - initializes new top-level, creating index that points to the
1300 * just created block
1302 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1305 struct ext4_extent_header *neh;
1306 struct buffer_head *bh;
1307 ext4_fsblk_t newblock, goal = 0;
1308 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1310 size_t ext_size = 0;
1312 /* Try to prepend new index to old one */
1313 if (ext_depth(inode))
1314 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1315 if (goal > le32_to_cpu(es->s_first_data_block)) {
1316 flags |= EXT4_MB_HINT_TRY_GOAL;
1319 goal = ext4_inode_to_goal_block(inode);
1320 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1325 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1330 err = ext4_journal_get_create_access(handle, bh);
1336 ext_size = sizeof(EXT4_I(inode)->i_data);
1337 /* move top-level index/leaf into new block */
1338 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1339 /* zero out unused area in the extent block */
1340 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1342 /* set size of new block */
1343 neh = ext_block_hdr(bh);
1344 /* old root could have indexes or leaves
1345 * so calculate e_max right way */
1346 if (ext_depth(inode))
1347 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1349 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1350 neh->eh_magic = EXT4_EXT_MAGIC;
1351 ext4_extent_block_csum_set(inode, neh);
1352 set_buffer_uptodate(bh);
1355 err = ext4_handle_dirty_metadata(handle, inode, bh);
1359 /* Update top-level index: num,max,pointer */
1360 neh = ext_inode_hdr(inode);
1361 neh->eh_entries = cpu_to_le16(1);
1362 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1363 if (neh->eh_depth == 0) {
1364 /* Root extent block becomes index block */
1365 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1366 EXT_FIRST_INDEX(neh)->ei_block =
1367 EXT_FIRST_EXTENT(neh)->ee_block;
1369 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1370 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1371 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1372 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1374 le16_add_cpu(&neh->eh_depth, 1);
1375 err = ext4_mark_inode_dirty(handle, inode);
1383 * ext4_ext_create_new_leaf:
1384 * finds empty index and adds new leaf.
1385 * if no free index is found, then it requests in-depth growing.
1387 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1388 unsigned int mb_flags,
1389 unsigned int gb_flags,
1390 struct ext4_ext_path **ppath,
1391 struct ext4_extent *newext)
1393 struct ext4_ext_path *path = *ppath;
1394 struct ext4_ext_path *curp;
1395 int depth, i, err = 0;
1398 i = depth = ext_depth(inode);
1400 /* walk up to the tree and look for free index entry */
1401 curp = path + depth;
1402 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1407 /* we use already allocated block for index block,
1408 * so subsequent data blocks should be contiguous */
1409 if (EXT_HAS_FREE_INDEX(curp)) {
1410 /* if we found index with free entry, then use that
1411 * entry: create all needed subtree and add new leaf */
1412 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1417 path = ext4_find_extent(inode,
1418 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1421 err = PTR_ERR(path);
1423 /* tree is full, time to grow in depth */
1424 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1429 path = ext4_find_extent(inode,
1430 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1433 err = PTR_ERR(path);
1438 * only first (depth 0 -> 1) produces free space;
1439 * in all other cases we have to split the grown tree
1441 depth = ext_depth(inode);
1442 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1443 /* now we need to split */
1453 * search the closest allocated block to the left for *logical
1454 * and returns it at @logical + it's physical address at @phys
1455 * if *logical is the smallest allocated block, the function
1456 * returns 0 at @phys
1457 * return value contains 0 (success) or error code
1459 static int ext4_ext_search_left(struct inode *inode,
1460 struct ext4_ext_path *path,
1461 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1463 struct ext4_extent_idx *ix;
1464 struct ext4_extent *ex;
1467 if (unlikely(path == NULL)) {
1468 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1469 return -EFSCORRUPTED;
1471 depth = path->p_depth;
1474 if (depth == 0 && path->p_ext == NULL)
1477 /* usually extent in the path covers blocks smaller
1478 * then *logical, but it can be that extent is the
1479 * first one in the file */
1481 ex = path[depth].p_ext;
1482 ee_len = ext4_ext_get_actual_len(ex);
1483 if (*logical < le32_to_cpu(ex->ee_block)) {
1484 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1485 EXT4_ERROR_INODE(inode,
1486 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1487 *logical, le32_to_cpu(ex->ee_block));
1488 return -EFSCORRUPTED;
1490 while (--depth >= 0) {
1491 ix = path[depth].p_idx;
1492 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1493 EXT4_ERROR_INODE(inode,
1494 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1495 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1496 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1497 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1499 return -EFSCORRUPTED;
1505 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1506 EXT4_ERROR_INODE(inode,
1507 "logical %d < ee_block %d + ee_len %d!",
1508 *logical, le32_to_cpu(ex->ee_block), ee_len);
1509 return -EFSCORRUPTED;
1512 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1513 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1518 * Search the closest allocated block to the right for *logical
1519 * and returns it at @logical + it's physical address at @phys.
1520 * If not exists, return 0 and @phys is set to 0. We will return
1521 * 1 which means we found an allocated block and ret_ex is valid.
1522 * Or return a (< 0) error code.
1524 static int ext4_ext_search_right(struct inode *inode,
1525 struct ext4_ext_path *path,
1526 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1527 struct ext4_extent *ret_ex)
1529 struct buffer_head *bh = NULL;
1530 struct ext4_extent_header *eh;
1531 struct ext4_extent_idx *ix;
1532 struct ext4_extent *ex;
1533 int depth; /* Note, NOT eh_depth; depth from top of tree */
1536 if (unlikely(path == NULL)) {
1537 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1538 return -EFSCORRUPTED;
1540 depth = path->p_depth;
1543 if (depth == 0 && path->p_ext == NULL)
1546 /* usually extent in the path covers blocks smaller
1547 * then *logical, but it can be that extent is the
1548 * first one in the file */
1550 ex = path[depth].p_ext;
1551 ee_len = ext4_ext_get_actual_len(ex);
1552 if (*logical < le32_to_cpu(ex->ee_block)) {
1553 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1554 EXT4_ERROR_INODE(inode,
1555 "first_extent(path[%d].p_hdr) != ex",
1557 return -EFSCORRUPTED;
1559 while (--depth >= 0) {
1560 ix = path[depth].p_idx;
1561 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1562 EXT4_ERROR_INODE(inode,
1563 "ix != EXT_FIRST_INDEX *logical %d!",
1565 return -EFSCORRUPTED;
1571 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1572 EXT4_ERROR_INODE(inode,
1573 "logical %d < ee_block %d + ee_len %d!",
1574 *logical, le32_to_cpu(ex->ee_block), ee_len);
1575 return -EFSCORRUPTED;
1578 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1579 /* next allocated block in this leaf */
1584 /* go up and search for index to the right */
1585 while (--depth >= 0) {
1586 ix = path[depth].p_idx;
1587 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1591 /* we've gone up to the root and found no index to the right */
1595 /* we've found index to the right, let's
1596 * follow it and find the closest allocated
1597 * block to the right */
1599 while (++depth < path->p_depth) {
1600 /* subtract from p_depth to get proper eh_depth */
1601 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1604 eh = ext_block_hdr(bh);
1605 ix = EXT_FIRST_INDEX(eh);
1609 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1612 eh = ext_block_hdr(bh);
1613 ex = EXT_FIRST_EXTENT(eh);
1615 *logical = le32_to_cpu(ex->ee_block);
1616 *phys = ext4_ext_pblock(ex);
1625 * ext4_ext_next_allocated_block:
1626 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1627 * NOTE: it considers block number from index entry as
1628 * allocated block. Thus, index entries have to be consistent
1632 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1636 BUG_ON(path == NULL);
1637 depth = path->p_depth;
1639 if (depth == 0 && path->p_ext == NULL)
1640 return EXT_MAX_BLOCKS;
1642 while (depth >= 0) {
1643 struct ext4_ext_path *p = &path[depth];
1645 if (depth == path->p_depth) {
1647 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1648 return le32_to_cpu(p->p_ext[1].ee_block);
1651 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1652 return le32_to_cpu(p->p_idx[1].ei_block);
1657 return EXT_MAX_BLOCKS;
1661 * ext4_ext_next_leaf_block:
1662 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1664 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1668 BUG_ON(path == NULL);
1669 depth = path->p_depth;
1671 /* zero-tree has no leaf blocks at all */
1673 return EXT_MAX_BLOCKS;
1675 /* go to index block */
1678 while (depth >= 0) {
1679 if (path[depth].p_idx !=
1680 EXT_LAST_INDEX(path[depth].p_hdr))
1681 return (ext4_lblk_t)
1682 le32_to_cpu(path[depth].p_idx[1].ei_block);
1686 return EXT_MAX_BLOCKS;
1690 * ext4_ext_correct_indexes:
1691 * if leaf gets modified and modified extent is first in the leaf,
1692 * then we have to correct all indexes above.
1693 * TODO: do we need to correct tree in all cases?
1695 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1696 struct ext4_ext_path *path)
1698 struct ext4_extent_header *eh;
1699 int depth = ext_depth(inode);
1700 struct ext4_extent *ex;
1704 eh = path[depth].p_hdr;
1705 ex = path[depth].p_ext;
1707 if (unlikely(ex == NULL || eh == NULL)) {
1708 EXT4_ERROR_INODE(inode,
1709 "ex %p == NULL or eh %p == NULL", ex, eh);
1710 return -EFSCORRUPTED;
1714 /* there is no tree at all */
1718 if (ex != EXT_FIRST_EXTENT(eh)) {
1719 /* we correct tree if first leaf got modified only */
1724 * TODO: we need correction if border is smaller than current one
1727 border = path[depth].p_ext->ee_block;
1728 err = ext4_ext_get_access(handle, inode, path + k);
1731 path[k].p_idx->ei_block = border;
1732 err = ext4_ext_dirty(handle, inode, path + k);
1737 /* change all left-side indexes */
1738 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1740 err = ext4_ext_get_access(handle, inode, path + k);
1743 path[k].p_idx->ei_block = border;
1744 err = ext4_ext_dirty(handle, inode, path + k);
1752 static int ext4_can_extents_be_merged(struct inode *inode,
1753 struct ext4_extent *ex1,
1754 struct ext4_extent *ex2)
1756 unsigned short ext1_ee_len, ext2_ee_len;
1758 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1761 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1762 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1764 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1765 le32_to_cpu(ex2->ee_block))
1768 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1771 if (ext4_ext_is_unwritten(ex1) &&
1772 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1774 #ifdef AGGRESSIVE_TEST
1775 if (ext1_ee_len >= 4)
1779 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1785 * This function tries to merge the "ex" extent to the next extent in the tree.
1786 * It always tries to merge towards right. If you want to merge towards
1787 * left, pass "ex - 1" as argument instead of "ex".
1788 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1789 * 1 if they got merged.
1791 static int ext4_ext_try_to_merge_right(struct inode *inode,
1792 struct ext4_ext_path *path,
1793 struct ext4_extent *ex)
1795 struct ext4_extent_header *eh;
1796 unsigned int depth, len;
1797 int merge_done = 0, unwritten;
1799 depth = ext_depth(inode);
1800 BUG_ON(path[depth].p_hdr == NULL);
1801 eh = path[depth].p_hdr;
1803 while (ex < EXT_LAST_EXTENT(eh)) {
1804 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1806 /* merge with next extent! */
1807 unwritten = ext4_ext_is_unwritten(ex);
1808 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1809 + ext4_ext_get_actual_len(ex + 1));
1811 ext4_ext_mark_unwritten(ex);
1813 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1814 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1815 * sizeof(struct ext4_extent);
1816 memmove(ex + 1, ex + 2, len);
1818 le16_add_cpu(&eh->eh_entries, -1);
1820 WARN_ON(eh->eh_entries == 0);
1821 if (!eh->eh_entries)
1822 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1829 * This function does a very simple check to see if we can collapse
1830 * an extent tree with a single extent tree leaf block into the inode.
1832 static void ext4_ext_try_to_merge_up(handle_t *handle,
1833 struct inode *inode,
1834 struct ext4_ext_path *path)
1837 unsigned max_root = ext4_ext_space_root(inode, 0);
1840 if ((path[0].p_depth != 1) ||
1841 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1842 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1846 * We need to modify the block allocation bitmap and the block
1847 * group descriptor to release the extent tree block. If we
1848 * can't get the journal credits, give up.
1850 if (ext4_journal_extend(handle, 2,
1851 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1855 * Copy the extent data up to the inode
1857 blk = ext4_idx_pblock(path[0].p_idx);
1858 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1859 sizeof(struct ext4_extent_idx);
1860 s += sizeof(struct ext4_extent_header);
1862 path[1].p_maxdepth = path[0].p_maxdepth;
1863 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1864 path[0].p_depth = 0;
1865 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1866 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1867 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1869 brelse(path[1].p_bh);
1870 ext4_free_blocks(handle, inode, NULL, blk, 1,
1871 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1875 * This function tries to merge the @ex extent to neighbours in the tree, then
1876 * tries to collapse the extent tree into the inode.
1878 static void ext4_ext_try_to_merge(handle_t *handle,
1879 struct inode *inode,
1880 struct ext4_ext_path *path,
1881 struct ext4_extent *ex)
1883 struct ext4_extent_header *eh;
1887 depth = ext_depth(inode);
1888 BUG_ON(path[depth].p_hdr == NULL);
1889 eh = path[depth].p_hdr;
1891 if (ex > EXT_FIRST_EXTENT(eh))
1892 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1895 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1897 ext4_ext_try_to_merge_up(handle, inode, path);
1901 * check if a portion of the "newext" extent overlaps with an
1904 * If there is an overlap discovered, it updates the length of the newext
1905 * such that there will be no overlap, and then returns 1.
1906 * If there is no overlap found, it returns 0.
1908 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1909 struct inode *inode,
1910 struct ext4_extent *newext,
1911 struct ext4_ext_path *path)
1914 unsigned int depth, len1;
1915 unsigned int ret = 0;
1917 b1 = le32_to_cpu(newext->ee_block);
1918 len1 = ext4_ext_get_actual_len(newext);
1919 depth = ext_depth(inode);
1920 if (!path[depth].p_ext)
1922 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1925 * get the next allocated block if the extent in the path
1926 * is before the requested block(s)
1929 b2 = ext4_ext_next_allocated_block(path);
1930 if (b2 == EXT_MAX_BLOCKS)
1932 b2 = EXT4_LBLK_CMASK(sbi, b2);
1935 /* check for wrap through zero on extent logical start block*/
1936 if (b1 + len1 < b1) {
1937 len1 = EXT_MAX_BLOCKS - b1;
1938 newext->ee_len = cpu_to_le16(len1);
1942 /* check for overlap */
1943 if (b1 + len1 > b2) {
1944 newext->ee_len = cpu_to_le16(b2 - b1);
1952 * ext4_ext_insert_extent:
1953 * tries to merge requested extent into the existing extent or
1954 * inserts requested extent as new one into the tree,
1955 * creating new leaf in the no-space case.
1957 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1958 struct ext4_ext_path **ppath,
1959 struct ext4_extent *newext, int gb_flags)
1961 struct ext4_ext_path *path = *ppath;
1962 struct ext4_extent_header *eh;
1963 struct ext4_extent *ex, *fex;
1964 struct ext4_extent *nearex; /* nearest extent */
1965 struct ext4_ext_path *npath = NULL;
1966 int depth, len, err;
1968 int mb_flags = 0, unwritten;
1970 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1971 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1972 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1973 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1974 return -EFSCORRUPTED;
1976 depth = ext_depth(inode);
1977 ex = path[depth].p_ext;
1978 eh = path[depth].p_hdr;
1979 if (unlikely(path[depth].p_hdr == NULL)) {
1980 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1981 return -EFSCORRUPTED;
1984 /* try to insert block into found extent and return */
1985 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1988 * Try to see whether we should rather test the extent on
1989 * right from ex, or from the left of ex. This is because
1990 * ext4_find_extent() can return either extent on the
1991 * left, or on the right from the searched position. This
1992 * will make merging more effective.
1994 if (ex < EXT_LAST_EXTENT(eh) &&
1995 (le32_to_cpu(ex->ee_block) +
1996 ext4_ext_get_actual_len(ex) <
1997 le32_to_cpu(newext->ee_block))) {
2000 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2001 (le32_to_cpu(newext->ee_block) +
2002 ext4_ext_get_actual_len(newext) <
2003 le32_to_cpu(ex->ee_block)))
2006 /* Try to append newex to the ex */
2007 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2008 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
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,
2020 unwritten = ext4_ext_is_unwritten(ex);
2021 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2022 + ext4_ext_get_actual_len(newext));
2024 ext4_ext_mark_unwritten(ex);
2025 eh = path[depth].p_hdr;
2031 /* Try to prepend newex to the ex */
2032 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2033 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2035 le32_to_cpu(newext->ee_block),
2036 ext4_ext_is_unwritten(newext),
2037 ext4_ext_get_actual_len(newext),
2038 le32_to_cpu(ex->ee_block),
2039 ext4_ext_is_unwritten(ex),
2040 ext4_ext_get_actual_len(ex),
2041 ext4_ext_pblock(ex));
2042 err = ext4_ext_get_access(handle, inode,
2047 unwritten = ext4_ext_is_unwritten(ex);
2048 ex->ee_block = newext->ee_block;
2049 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2050 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2051 + ext4_ext_get_actual_len(newext));
2053 ext4_ext_mark_unwritten(ex);
2054 eh = path[depth].p_hdr;
2060 depth = ext_depth(inode);
2061 eh = path[depth].p_hdr;
2062 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2065 /* probably next leaf has space for us? */
2066 fex = EXT_LAST_EXTENT(eh);
2067 next = EXT_MAX_BLOCKS;
2068 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2069 next = ext4_ext_next_leaf_block(path);
2070 if (next != EXT_MAX_BLOCKS) {
2071 ext_debug(inode, "next leaf block - %u\n", next);
2072 BUG_ON(npath != NULL);
2073 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2075 return PTR_ERR(npath);
2076 BUG_ON(npath->p_depth != path->p_depth);
2077 eh = npath[depth].p_hdr;
2078 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2079 ext_debug(inode, "next leaf isn't full(%d)\n",
2080 le16_to_cpu(eh->eh_entries));
2084 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2085 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2089 * There is no free space in the found leaf.
2090 * We're gonna add a new leaf in the tree.
2092 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2093 mb_flags |= EXT4_MB_USE_RESERVED;
2094 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2098 depth = ext_depth(inode);
2099 eh = path[depth].p_hdr;
2102 nearex = path[depth].p_ext;
2104 err = ext4_ext_get_access(handle, inode, path + depth);
2109 /* there is no extent in this leaf, create first one */
2110 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2111 le32_to_cpu(newext->ee_block),
2112 ext4_ext_pblock(newext),
2113 ext4_ext_is_unwritten(newext),
2114 ext4_ext_get_actual_len(newext));
2115 nearex = EXT_FIRST_EXTENT(eh);
2117 if (le32_to_cpu(newext->ee_block)
2118 > le32_to_cpu(nearex->ee_block)) {
2120 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2122 le32_to_cpu(newext->ee_block),
2123 ext4_ext_pblock(newext),
2124 ext4_ext_is_unwritten(newext),
2125 ext4_ext_get_actual_len(newext),
2130 BUG_ON(newext->ee_block == nearex->ee_block);
2131 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2133 le32_to_cpu(newext->ee_block),
2134 ext4_ext_pblock(newext),
2135 ext4_ext_is_unwritten(newext),
2136 ext4_ext_get_actual_len(newext),
2139 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2141 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2142 "move %d extents from 0x%p to 0x%p\n",
2143 le32_to_cpu(newext->ee_block),
2144 ext4_ext_pblock(newext),
2145 ext4_ext_is_unwritten(newext),
2146 ext4_ext_get_actual_len(newext),
2147 len, nearex, nearex + 1);
2148 memmove(nearex + 1, nearex,
2149 len * sizeof(struct ext4_extent));
2153 le16_add_cpu(&eh->eh_entries, 1);
2154 path[depth].p_ext = nearex;
2155 nearex->ee_block = newext->ee_block;
2156 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2157 nearex->ee_len = newext->ee_len;
2160 /* try to merge extents */
2161 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2162 ext4_ext_try_to_merge(handle, inode, path, nearex);
2165 /* time to correct all indexes above */
2166 err = ext4_ext_correct_indexes(handle, inode, path);
2170 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2173 ext4_ext_drop_refs(npath);
2178 static int ext4_fill_es_cache_info(struct inode *inode,
2179 ext4_lblk_t block, ext4_lblk_t num,
2180 struct fiemap_extent_info *fieinfo)
2182 ext4_lblk_t next, end = block + num - 1;
2183 struct extent_status es;
2184 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2188 while (block <= end) {
2191 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2193 if (ext4_es_is_unwritten(&es))
2194 flags |= FIEMAP_EXTENT_UNWRITTEN;
2195 if (ext4_es_is_delayed(&es))
2196 flags |= (FIEMAP_EXTENT_DELALLOC |
2197 FIEMAP_EXTENT_UNKNOWN);
2198 if (ext4_es_is_hole(&es))
2199 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2201 flags |= FIEMAP_EXTENT_LAST;
2202 if (flags & (FIEMAP_EXTENT_DELALLOC|
2203 EXT4_FIEMAP_EXTENT_HOLE))
2206 es.es_pblk = ext4_es_pblock(&es);
2207 err = fiemap_fill_next_extent(fieinfo,
2208 (__u64)es.es_lblk << blksize_bits,
2209 (__u64)es.es_pblk << blksize_bits,
2210 (__u64)es.es_len << blksize_bits,
2225 * ext4_ext_find_hole - find hole around given block according to the given path
2226 * @inode: inode we lookup in
2227 * @path: path in extent tree to @lblk
2228 * @lblk: pointer to logical block around which we want to determine hole
2230 * Determine hole length (and start if easily possible) around given logical
2231 * block. We don't try too hard to find the beginning of the hole but @path
2232 * actually points to extent before @lblk, we provide it.
2234 * The function returns the length of a hole starting at @lblk. We update @lblk
2235 * to the beginning of the hole if we managed to find it.
2237 static ext4_lblk_t ext4_ext_find_hole(struct inode *inode,
2238 struct ext4_ext_path *path,
2241 int depth = ext_depth(inode);
2242 struct ext4_extent *ex;
2245 ex = path[depth].p_ext;
2247 /* there is no extent yet, so gap is [0;-] */
2249 len = EXT_MAX_BLOCKS;
2250 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2251 len = le32_to_cpu(ex->ee_block) - *lblk;
2252 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2253 + ext4_ext_get_actual_len(ex)) {
2256 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2257 next = ext4_ext_next_allocated_block(path);
2258 BUG_ON(next == *lblk);
2268 * removes index from the index block.
2270 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2271 struct ext4_ext_path *path, int depth)
2276 /* free index block */
2278 path = path + depth;
2279 leaf = ext4_idx_pblock(path->p_idx);
2280 if (unlikely(path->p_hdr->eh_entries == 0)) {
2281 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2282 return -EFSCORRUPTED;
2284 err = ext4_ext_get_access(handle, inode, path);
2288 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2289 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2290 len *= sizeof(struct ext4_extent_idx);
2291 memmove(path->p_idx, path->p_idx + 1, len);
2294 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2295 err = ext4_ext_dirty(handle, inode, path);
2298 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2299 trace_ext4_ext_rm_idx(inode, leaf);
2301 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2302 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2304 while (--depth >= 0) {
2305 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2308 err = ext4_ext_get_access(handle, inode, path);
2311 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2312 err = ext4_ext_dirty(handle, inode, path);
2320 * ext4_ext_calc_credits_for_single_extent:
2321 * This routine returns max. credits that needed to insert an extent
2322 * to the extent tree.
2323 * When pass the actual path, the caller should calculate credits
2326 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2327 struct ext4_ext_path *path)
2330 int depth = ext_depth(inode);
2333 /* probably there is space in leaf? */
2334 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2335 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2338 * There are some space in the leaf tree, no
2339 * need to account for leaf block credit
2341 * bitmaps and block group descriptor blocks
2342 * and other metadata blocks still need to be
2345 /* 1 bitmap, 1 block group descriptor */
2346 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2351 return ext4_chunk_trans_blocks(inode, nrblocks);
2355 * How many index/leaf blocks need to change/allocate to add @extents extents?
2357 * If we add a single extent, then in the worse case, each tree level
2358 * index/leaf need to be changed in case of the tree split.
2360 * If more extents are inserted, they could cause the whole tree split more
2361 * than once, but this is really rare.
2363 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2368 /* If we are converting the inline data, only one is needed here. */
2369 if (ext4_has_inline_data(inode))
2372 depth = ext_depth(inode);
2382 static inline int get_default_free_blocks_flags(struct inode *inode)
2384 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2385 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2386 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2387 else if (ext4_should_journal_data(inode))
2388 return EXT4_FREE_BLOCKS_FORGET;
2393 * ext4_rereserve_cluster - increment the reserved cluster count when
2394 * freeing a cluster with a pending reservation
2396 * @inode - file containing the cluster
2397 * @lblk - logical block in cluster to be reserved
2399 * Increments the reserved cluster count and adjusts quota in a bigalloc
2400 * file system when freeing a partial cluster containing at least one
2401 * delayed and unwritten block. A partial cluster meeting that
2402 * requirement will have a pending reservation. If so, the
2403 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2404 * defer reserved and allocated space accounting to a subsequent call
2407 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2409 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2410 struct ext4_inode_info *ei = EXT4_I(inode);
2412 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2414 spin_lock(&ei->i_block_reservation_lock);
2415 ei->i_reserved_data_blocks++;
2416 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2417 spin_unlock(&ei->i_block_reservation_lock);
2419 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2420 ext4_remove_pending(inode, lblk);
2423 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2424 struct ext4_extent *ex,
2425 struct partial_cluster *partial,
2426 ext4_lblk_t from, ext4_lblk_t to)
2428 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2429 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2430 ext4_fsblk_t last_pblk, pblk;
2434 /* only extent tail removal is allowed */
2435 if (from < le32_to_cpu(ex->ee_block) ||
2436 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2437 ext4_error(sbi->s_sb,
2438 "strange request: removal(2) %u-%u from %u:%u",
2439 from, to, le32_to_cpu(ex->ee_block), ee_len);
2443 #ifdef EXTENTS_STATS
2444 spin_lock(&sbi->s_ext_stats_lock);
2445 sbi->s_ext_blocks += ee_len;
2446 sbi->s_ext_extents++;
2447 if (ee_len < sbi->s_ext_min)
2448 sbi->s_ext_min = ee_len;
2449 if (ee_len > sbi->s_ext_max)
2450 sbi->s_ext_max = ee_len;
2451 if (ext_depth(inode) > sbi->s_depth_max)
2452 sbi->s_depth_max = ext_depth(inode);
2453 spin_unlock(&sbi->s_ext_stats_lock);
2456 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2459 * if we have a partial cluster, and it's different from the
2460 * cluster of the last block in the extent, we free it
2462 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2464 if (partial->state != initial &&
2465 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2466 if (partial->state == tofree) {
2467 flags = get_default_free_blocks_flags(inode);
2468 if (ext4_is_pending(inode, partial->lblk))
2469 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2470 ext4_free_blocks(handle, inode, NULL,
2471 EXT4_C2B(sbi, partial->pclu),
2472 sbi->s_cluster_ratio, flags);
2473 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2474 ext4_rereserve_cluster(inode, partial->lblk);
2476 partial->state = initial;
2479 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2480 pblk = ext4_ext_pblock(ex) + ee_len - num;
2483 * We free the partial cluster at the end of the extent (if any),
2484 * unless the cluster is used by another extent (partial_cluster
2485 * state is nofree). If a partial cluster exists here, it must be
2486 * shared with the last block in the extent.
2488 flags = get_default_free_blocks_flags(inode);
2490 /* partial, left end cluster aligned, right end unaligned */
2491 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2492 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2493 (partial->state != nofree)) {
2494 if (ext4_is_pending(inode, to))
2495 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2496 ext4_free_blocks(handle, inode, NULL,
2497 EXT4_PBLK_CMASK(sbi, last_pblk),
2498 sbi->s_cluster_ratio, flags);
2499 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2500 ext4_rereserve_cluster(inode, to);
2501 partial->state = initial;
2502 flags = get_default_free_blocks_flags(inode);
2505 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2508 * For bigalloc file systems, we never free a partial cluster
2509 * at the beginning of the extent. Instead, we check to see if we
2510 * need to free it on a subsequent call to ext4_remove_blocks,
2511 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2513 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2514 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2516 /* reset the partial cluster if we've freed past it */
2517 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2518 partial->state = initial;
2521 * If we've freed the entire extent but the beginning is not left
2522 * cluster aligned and is not marked as ineligible for freeing we
2523 * record the partial cluster at the beginning of the extent. It
2524 * wasn't freed by the preceding ext4_free_blocks() call, and we
2525 * need to look farther to the left to determine if it's to be freed
2526 * (not shared with another extent). Else, reset the partial
2527 * cluster - we're either done freeing or the beginning of the
2528 * extent is left cluster aligned.
2530 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2531 if (partial->state == initial) {
2532 partial->pclu = EXT4_B2C(sbi, pblk);
2533 partial->lblk = from;
2534 partial->state = tofree;
2537 partial->state = initial;
2544 * ext4_ext_rm_leaf() Removes the extents associated with the
2545 * blocks appearing between "start" and "end". Both "start"
2546 * and "end" must appear in the same extent or EIO is returned.
2548 * @handle: The journal handle
2549 * @inode: The files inode
2550 * @path: The path to the leaf
2551 * @partial_cluster: The cluster which we'll have to free if all extents
2552 * has been released from it. However, if this value is
2553 * negative, it's a cluster just to the right of the
2554 * punched region and it must not be freed.
2555 * @start: The first block to remove
2556 * @end: The last block to remove
2559 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2560 struct ext4_ext_path *path,
2561 struct partial_cluster *partial,
2562 ext4_lblk_t start, ext4_lblk_t end)
2564 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2565 int err = 0, correct_index = 0;
2566 int depth = ext_depth(inode), credits, revoke_credits;
2567 struct ext4_extent_header *eh;
2570 ext4_lblk_t ex_ee_block;
2571 unsigned short ex_ee_len;
2572 unsigned unwritten = 0;
2573 struct ext4_extent *ex;
2576 /* the header must be checked already in ext4_ext_remove_space() */
2577 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2578 if (!path[depth].p_hdr)
2579 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2580 eh = path[depth].p_hdr;
2581 if (unlikely(path[depth].p_hdr == NULL)) {
2582 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2583 return -EFSCORRUPTED;
2585 /* find where to start removing */
2586 ex = path[depth].p_ext;
2588 ex = EXT_LAST_EXTENT(eh);
2590 ex_ee_block = le32_to_cpu(ex->ee_block);
2591 ex_ee_len = ext4_ext_get_actual_len(ex);
2593 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2595 while (ex >= EXT_FIRST_EXTENT(eh) &&
2596 ex_ee_block + ex_ee_len > start) {
2598 if (ext4_ext_is_unwritten(ex))
2603 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2604 unwritten, ex_ee_len);
2605 path[depth].p_ext = ex;
2607 a = ex_ee_block > start ? ex_ee_block : start;
2608 b = ex_ee_block+ex_ee_len - 1 < end ?
2609 ex_ee_block+ex_ee_len - 1 : end;
2611 ext_debug(inode, " border %u:%u\n", a, b);
2613 /* If this extent is beyond the end of the hole, skip it */
2614 if (end < ex_ee_block) {
2616 * We're going to skip this extent and move to another,
2617 * so note that its first cluster is in use to avoid
2618 * freeing it when removing blocks. Eventually, the
2619 * right edge of the truncated/punched region will
2620 * be just to the left.
2622 if (sbi->s_cluster_ratio > 1) {
2623 pblk = ext4_ext_pblock(ex);
2624 partial->pclu = EXT4_B2C(sbi, pblk);
2625 partial->state = nofree;
2628 ex_ee_block = le32_to_cpu(ex->ee_block);
2629 ex_ee_len = ext4_ext_get_actual_len(ex);
2631 } else if (b != ex_ee_block + ex_ee_len - 1) {
2632 EXT4_ERROR_INODE(inode,
2633 "can not handle truncate %u:%u "
2635 start, end, ex_ee_block,
2636 ex_ee_block + ex_ee_len - 1);
2637 err = -EFSCORRUPTED;
2639 } else if (a != ex_ee_block) {
2640 /* remove tail of the extent */
2641 num = a - ex_ee_block;
2643 /* remove whole extent: excellent! */
2647 * 3 for leaf, sb, and inode plus 2 (bmap and group
2648 * descriptor) for each block group; assume two block
2649 * groups plus ex_ee_len/blocks_per_block_group for
2652 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2653 if (ex == EXT_FIRST_EXTENT(eh)) {
2655 credits += (ext_depth(inode)) + 1;
2657 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2659 * We may end up freeing some index blocks and data from the
2660 * punched range. Note that partial clusters are accounted for
2661 * by ext4_free_data_revoke_credits().
2664 ext4_free_metadata_revoke_credits(inode->i_sb,
2666 ext4_free_data_revoke_credits(inode, b - a + 1);
2668 err = ext4_datasem_ensure_credits(handle, inode, credits,
2669 credits, revoke_credits);
2676 err = ext4_ext_get_access(handle, inode, path + depth);
2680 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2685 /* this extent is removed; mark slot entirely unused */
2686 ext4_ext_store_pblock(ex, 0);
2688 ex->ee_len = cpu_to_le16(num);
2690 * Do not mark unwritten if all the blocks in the
2691 * extent have been removed.
2693 if (unwritten && num)
2694 ext4_ext_mark_unwritten(ex);
2696 * If the extent was completely released,
2697 * we need to remove it from the leaf
2700 if (end != EXT_MAX_BLOCKS - 1) {
2702 * For hole punching, we need to scoot all the
2703 * extents up when an extent is removed so that
2704 * we dont have blank extents in the middle
2706 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2707 sizeof(struct ext4_extent));
2709 /* Now get rid of the one at the end */
2710 memset(EXT_LAST_EXTENT(eh), 0,
2711 sizeof(struct ext4_extent));
2713 le16_add_cpu(&eh->eh_entries, -1);
2716 err = ext4_ext_dirty(handle, inode, path + depth);
2720 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2721 ext4_ext_pblock(ex));
2723 ex_ee_block = le32_to_cpu(ex->ee_block);
2724 ex_ee_len = ext4_ext_get_actual_len(ex);
2727 if (correct_index && eh->eh_entries)
2728 err = ext4_ext_correct_indexes(handle, inode, path);
2731 * If there's a partial cluster and at least one extent remains in
2732 * the leaf, free the partial cluster if it isn't shared with the
2733 * current extent. If it is shared with the current extent
2734 * we reset the partial cluster because we've reached the start of the
2735 * truncated/punched region and we're done removing blocks.
2737 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2738 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2739 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2740 int flags = get_default_free_blocks_flags(inode);
2742 if (ext4_is_pending(inode, partial->lblk))
2743 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2744 ext4_free_blocks(handle, inode, NULL,
2745 EXT4_C2B(sbi, partial->pclu),
2746 sbi->s_cluster_ratio, flags);
2747 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2748 ext4_rereserve_cluster(inode, partial->lblk);
2750 partial->state = initial;
2753 /* if this leaf is free, then we should
2754 * remove it from index block above */
2755 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2756 err = ext4_ext_rm_idx(handle, inode, path, depth);
2763 * ext4_ext_more_to_rm:
2764 * returns 1 if current index has to be freed (even partial)
2767 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2769 BUG_ON(path->p_idx == NULL);
2771 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2775 * if truncate on deeper level happened, it wasn't partial,
2776 * so we have to consider current index for truncation
2778 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2783 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2786 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2787 int depth = ext_depth(inode);
2788 struct ext4_ext_path *path = NULL;
2789 struct partial_cluster partial;
2795 partial.state = initial;
2797 ext_debug(inode, "truncate since %u to %u\n", start, end);
2799 /* probably first extent we're gonna free will be last in block */
2800 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2802 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2804 return PTR_ERR(handle);
2807 trace_ext4_ext_remove_space(inode, start, end, depth);
2810 * Check if we are removing extents inside the extent tree. If that
2811 * is the case, we are going to punch a hole inside the extent tree
2812 * so we have to check whether we need to split the extent covering
2813 * the last block to remove so we can easily remove the part of it
2814 * in ext4_ext_rm_leaf().
2816 if (end < EXT_MAX_BLOCKS - 1) {
2817 struct ext4_extent *ex;
2818 ext4_lblk_t ee_block, ex_end, lblk;
2821 /* find extent for or closest extent to this block */
2822 path = ext4_find_extent(inode, end, NULL,
2823 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2825 ext4_journal_stop(handle);
2826 return PTR_ERR(path);
2828 depth = ext_depth(inode);
2829 /* Leaf not may not exist only if inode has no blocks at all */
2830 ex = path[depth].p_ext;
2833 EXT4_ERROR_INODE(inode,
2834 "path[%d].p_hdr == NULL",
2836 err = -EFSCORRUPTED;
2841 ee_block = le32_to_cpu(ex->ee_block);
2842 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2845 * See if the last block is inside the extent, if so split
2846 * the extent at 'end' block so we can easily remove the
2847 * tail of the first part of the split extent in
2848 * ext4_ext_rm_leaf().
2850 if (end >= ee_block && end < ex_end) {
2853 * If we're going to split the extent, note that
2854 * the cluster containing the block after 'end' is
2855 * in use to avoid freeing it when removing blocks.
2857 if (sbi->s_cluster_ratio > 1) {
2858 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2859 partial.pclu = EXT4_B2C(sbi, pblk);
2860 partial.state = nofree;
2864 * Split the extent in two so that 'end' is the last
2865 * block in the first new extent. Also we should not
2866 * fail removing space due to ENOSPC so try to use
2867 * reserved block if that happens.
2869 err = ext4_force_split_extent_at(handle, inode, &path,
2874 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2875 partial.state == initial) {
2877 * If we're punching, there's an extent to the right.
2878 * If the partial cluster hasn't been set, set it to
2879 * that extent's first cluster and its state to nofree
2880 * so it won't be freed should it contain blocks to be
2881 * removed. If it's already set (tofree/nofree), we're
2882 * retrying and keep the original partial cluster info
2883 * so a cluster marked tofree as a result of earlier
2884 * extent removal is not lost.
2887 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2892 partial.pclu = EXT4_B2C(sbi, pblk);
2893 partial.state = nofree;
2898 * We start scanning from right side, freeing all the blocks
2899 * after i_size and walking into the tree depth-wise.
2901 depth = ext_depth(inode);
2906 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2908 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2909 GFP_NOFS | __GFP_NOFAIL);
2911 ext4_journal_stop(handle);
2914 path[0].p_maxdepth = path[0].p_depth = depth;
2915 path[0].p_hdr = ext_inode_hdr(inode);
2918 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2919 err = -EFSCORRUPTED;
2925 while (i >= 0 && err == 0) {
2927 /* this is leaf block */
2928 err = ext4_ext_rm_leaf(handle, inode, path,
2929 &partial, start, end);
2930 /* root level has p_bh == NULL, brelse() eats this */
2931 brelse(path[i].p_bh);
2932 path[i].p_bh = NULL;
2937 /* this is index block */
2938 if (!path[i].p_hdr) {
2939 ext_debug(inode, "initialize header\n");
2940 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2943 if (!path[i].p_idx) {
2944 /* this level hasn't been touched yet */
2945 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2946 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2947 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2949 le16_to_cpu(path[i].p_hdr->eh_entries));
2951 /* we were already here, see at next index */
2955 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2956 i, EXT_FIRST_INDEX(path[i].p_hdr),
2958 if (ext4_ext_more_to_rm(path + i)) {
2959 struct buffer_head *bh;
2960 /* go to the next level */
2961 ext_debug(inode, "move to level %d (block %llu)\n",
2962 i + 1, ext4_idx_pblock(path[i].p_idx));
2963 memset(path + i + 1, 0, sizeof(*path));
2964 bh = read_extent_tree_block(inode, path[i].p_idx,
2968 /* should we reset i_size? */
2972 /* Yield here to deal with large extent trees.
2973 * Should be a no-op if we did IO above. */
2975 if (WARN_ON(i + 1 > depth)) {
2976 err = -EFSCORRUPTED;
2979 path[i + 1].p_bh = bh;
2981 /* save actual number of indexes since this
2982 * number is changed at the next iteration */
2983 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2986 /* we finished processing this index, go up */
2987 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2988 /* index is empty, remove it;
2989 * handle must be already prepared by the
2990 * truncatei_leaf() */
2991 err = ext4_ext_rm_idx(handle, inode, path, i);
2993 /* root level has p_bh == NULL, brelse() eats this */
2994 brelse(path[i].p_bh);
2995 path[i].p_bh = NULL;
2997 ext_debug(inode, "return to level %d\n", i);
3001 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3002 path->p_hdr->eh_entries);
3005 * if there's a partial cluster and we have removed the first extent
3006 * in the file, then we also free the partial cluster, if any
3008 if (partial.state == tofree && err == 0) {
3009 int flags = get_default_free_blocks_flags(inode);
3011 if (ext4_is_pending(inode, partial.lblk))
3012 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3013 ext4_free_blocks(handle, inode, NULL,
3014 EXT4_C2B(sbi, partial.pclu),
3015 sbi->s_cluster_ratio, flags);
3016 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3017 ext4_rereserve_cluster(inode, partial.lblk);
3018 partial.state = initial;
3021 /* TODO: flexible tree reduction should be here */
3022 if (path->p_hdr->eh_entries == 0) {
3024 * truncate to zero freed all the tree,
3025 * so we need to correct eh_depth
3027 err = ext4_ext_get_access(handle, inode, path);
3029 ext_inode_hdr(inode)->eh_depth = 0;
3030 ext_inode_hdr(inode)->eh_max =
3031 cpu_to_le16(ext4_ext_space_root(inode, 0));
3032 err = ext4_ext_dirty(handle, inode, path);
3036 ext4_ext_drop_refs(path);
3041 ext4_journal_stop(handle);
3047 * called at mount time
3049 void ext4_ext_init(struct super_block *sb)
3052 * possible initialization would be here
3055 if (ext4_has_feature_extents(sb)) {
3056 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3057 printk(KERN_INFO "EXT4-fs: file extents enabled"
3058 #ifdef AGGRESSIVE_TEST
3059 ", aggressive tests"
3061 #ifdef CHECK_BINSEARCH
3064 #ifdef EXTENTS_STATS
3069 #ifdef EXTENTS_STATS
3070 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3071 EXT4_SB(sb)->s_ext_min = 1 << 30;
3072 EXT4_SB(sb)->s_ext_max = 0;
3078 * called at umount time
3080 void ext4_ext_release(struct super_block *sb)
3082 if (!ext4_has_feature_extents(sb))
3085 #ifdef EXTENTS_STATS
3086 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3087 struct ext4_sb_info *sbi = EXT4_SB(sb);
3088 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3089 sbi->s_ext_blocks, sbi->s_ext_extents,
3090 sbi->s_ext_blocks / sbi->s_ext_extents);
3091 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3092 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3097 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3099 ext4_lblk_t ee_block;
3100 ext4_fsblk_t ee_pblock;
3101 unsigned int ee_len;
3103 ee_block = le32_to_cpu(ex->ee_block);
3104 ee_len = ext4_ext_get_actual_len(ex);
3105 ee_pblock = ext4_ext_pblock(ex);
3110 ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3111 EXTENT_STATUS_WRITTEN);
3115 /* FIXME!! we need to try to merge to left or right after zero-out */
3116 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3118 ext4_fsblk_t ee_pblock;
3119 unsigned int ee_len;
3121 ee_len = ext4_ext_get_actual_len(ex);
3122 ee_pblock = ext4_ext_pblock(ex);
3123 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3128 * ext4_split_extent_at() splits an extent at given block.
3130 * @handle: the journal handle
3131 * @inode: the file inode
3132 * @path: the path to the extent
3133 * @split: the logical block where the extent is splitted.
3134 * @split_flags: indicates if the extent could be zeroout if split fails, and
3135 * the states(init or unwritten) of new extents.
3136 * @flags: flags used to insert new extent to extent tree.
3139 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3140 * of which are determined by split_flag.
3142 * There are two cases:
3143 * a> the extent are splitted into two extent.
3144 * b> split is not needed, and just mark the extent.
3146 * return 0 on success.
3148 static int ext4_split_extent_at(handle_t *handle,
3149 struct inode *inode,
3150 struct ext4_ext_path **ppath,
3155 struct ext4_ext_path *path = *ppath;
3156 ext4_fsblk_t newblock;
3157 ext4_lblk_t ee_block;
3158 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3159 struct ext4_extent *ex2 = NULL;
3160 unsigned int ee_len, depth;
3163 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3164 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3166 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3168 ext4_ext_show_leaf(inode, path);
3170 depth = ext_depth(inode);
3171 ex = path[depth].p_ext;
3172 ee_block = le32_to_cpu(ex->ee_block);
3173 ee_len = ext4_ext_get_actual_len(ex);
3174 newblock = split - ee_block + ext4_ext_pblock(ex);
3176 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3177 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3178 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3179 EXT4_EXT_MARK_UNWRIT1 |
3180 EXT4_EXT_MARK_UNWRIT2));
3182 err = ext4_ext_get_access(handle, inode, path + depth);
3186 if (split == ee_block) {
3188 * case b: block @split is the block that the extent begins with
3189 * then we just change the state of the extent, and splitting
3192 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3193 ext4_ext_mark_unwritten(ex);
3195 ext4_ext_mark_initialized(ex);
3197 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3198 ext4_ext_try_to_merge(handle, inode, path, ex);
3200 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3205 memcpy(&orig_ex, ex, sizeof(orig_ex));
3206 ex->ee_len = cpu_to_le16(split - ee_block);
3207 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3208 ext4_ext_mark_unwritten(ex);
3211 * path may lead to new leaf, not to original leaf any more
3212 * after ext4_ext_insert_extent() returns,
3214 err = ext4_ext_dirty(handle, inode, path + depth);
3216 goto fix_extent_len;
3219 ex2->ee_block = cpu_to_le32(split);
3220 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3221 ext4_ext_store_pblock(ex2, newblock);
3222 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3223 ext4_ext_mark_unwritten(ex2);
3225 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3226 if (err != -ENOSPC && err != -EDQUOT)
3229 if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3230 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3231 if (split_flag & EXT4_EXT_DATA_VALID1) {
3232 err = ext4_ext_zeroout(inode, ex2);
3233 zero_ex.ee_block = ex2->ee_block;
3234 zero_ex.ee_len = cpu_to_le16(
3235 ext4_ext_get_actual_len(ex2));
3236 ext4_ext_store_pblock(&zero_ex,
3237 ext4_ext_pblock(ex2));
3239 err = ext4_ext_zeroout(inode, ex);
3240 zero_ex.ee_block = ex->ee_block;
3241 zero_ex.ee_len = cpu_to_le16(
3242 ext4_ext_get_actual_len(ex));
3243 ext4_ext_store_pblock(&zero_ex,
3244 ext4_ext_pblock(ex));
3247 err = ext4_ext_zeroout(inode, &orig_ex);
3248 zero_ex.ee_block = orig_ex.ee_block;
3249 zero_ex.ee_len = cpu_to_le16(
3250 ext4_ext_get_actual_len(&orig_ex));
3251 ext4_ext_store_pblock(&zero_ex,
3252 ext4_ext_pblock(&orig_ex));
3256 /* update the extent length and mark as initialized */
3257 ex->ee_len = cpu_to_le16(ee_len);
3258 ext4_ext_try_to_merge(handle, inode, path, ex);
3259 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3261 /* update extent status tree */
3262 err = ext4_zeroout_es(inode, &zero_ex);
3263 /* If we failed at this point, we don't know in which
3264 * state the extent tree exactly is so don't try to fix
3265 * length of the original extent as it may do even more
3273 ex->ee_len = orig_ex.ee_len;
3275 * Ignore ext4_ext_dirty return value since we are already in error path
3276 * and err is a non-zero error code.
3278 ext4_ext_dirty(handle, inode, path + path->p_depth);
3281 ext4_ext_show_leaf(inode, path);
3286 * ext4_split_extents() splits an extent and mark extent which is covered
3287 * by @map as split_flags indicates
3289 * It may result in splitting the extent into multiple extents (up to three)
3290 * There are three possibilities:
3291 * a> There is no split required
3292 * b> Splits in two extents: Split is happening at either end of the extent
3293 * c> Splits in three extents: Somone is splitting in middle of the extent
3296 static int ext4_split_extent(handle_t *handle,
3297 struct inode *inode,
3298 struct ext4_ext_path **ppath,
3299 struct ext4_map_blocks *map,
3303 struct ext4_ext_path *path = *ppath;
3304 ext4_lblk_t ee_block;
3305 struct ext4_extent *ex;
3306 unsigned int ee_len, depth;
3309 int split_flag1, flags1;
3310 int allocated = map->m_len;
3312 depth = ext_depth(inode);
3313 ex = path[depth].p_ext;
3314 ee_block = le32_to_cpu(ex->ee_block);
3315 ee_len = ext4_ext_get_actual_len(ex);
3316 unwritten = ext4_ext_is_unwritten(ex);
3318 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3319 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3320 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3322 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3323 EXT4_EXT_MARK_UNWRIT2;
3324 if (split_flag & EXT4_EXT_DATA_VALID2)
3325 split_flag1 |= EXT4_EXT_DATA_VALID1;
3326 err = ext4_split_extent_at(handle, inode, ppath,
3327 map->m_lblk + map->m_len, split_flag1, flags1);
3331 allocated = ee_len - (map->m_lblk - ee_block);
3334 * Update path is required because previous ext4_split_extent_at() may
3335 * result in split of original leaf or extent zeroout.
3337 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3339 return PTR_ERR(path);
3340 depth = ext_depth(inode);
3341 ex = path[depth].p_ext;
3343 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3344 (unsigned long) map->m_lblk);
3345 return -EFSCORRUPTED;
3347 unwritten = ext4_ext_is_unwritten(ex);
3350 if (map->m_lblk >= ee_block) {
3351 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3353 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3354 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3355 EXT4_EXT_MARK_UNWRIT2);
3357 err = ext4_split_extent_at(handle, inode, ppath,
3358 map->m_lblk, split_flag1, flags);
3363 ext4_ext_show_leaf(inode, path);
3365 return err ? err : allocated;
3369 * This function is called by ext4_ext_map_blocks() if someone tries to write
3370 * to an unwritten extent. It may result in splitting the unwritten
3371 * extent into multiple extents (up to three - one initialized and two
3373 * There are three possibilities:
3374 * a> There is no split required: Entire extent should be initialized
3375 * b> Splits in two extents: Write is happening at either end of the extent
3376 * c> Splits in three extents: Somone is writing in middle of the extent
3379 * - The extent pointed to by 'path' is unwritten.
3380 * - The extent pointed to by 'path' contains a superset
3381 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3383 * Post-conditions on success:
3384 * - the returned value is the number of blocks beyond map->l_lblk
3385 * that are allocated and initialized.
3386 * It is guaranteed to be >= map->m_len.
3388 static int ext4_ext_convert_to_initialized(handle_t *handle,
3389 struct inode *inode,
3390 struct ext4_map_blocks *map,
3391 struct ext4_ext_path **ppath,
3394 struct ext4_ext_path *path = *ppath;
3395 struct ext4_sb_info *sbi;
3396 struct ext4_extent_header *eh;
3397 struct ext4_map_blocks split_map;
3398 struct ext4_extent zero_ex1, zero_ex2;
3399 struct ext4_extent *ex, *abut_ex;
3400 ext4_lblk_t ee_block, eof_block;
3401 unsigned int ee_len, depth, map_len = map->m_len;
3402 int allocated = 0, max_zeroout = 0;
3404 int split_flag = EXT4_EXT_DATA_VALID2;
3406 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3407 (unsigned long long)map->m_lblk, map_len);
3409 sbi = EXT4_SB(inode->i_sb);
3410 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3411 >> inode->i_sb->s_blocksize_bits;
3412 if (eof_block < map->m_lblk + map_len)
3413 eof_block = map->m_lblk + map_len;
3415 depth = ext_depth(inode);
3416 eh = path[depth].p_hdr;
3417 ex = path[depth].p_ext;
3418 ee_block = le32_to_cpu(ex->ee_block);
3419 ee_len = ext4_ext_get_actual_len(ex);
3420 zero_ex1.ee_len = 0;
3421 zero_ex2.ee_len = 0;
3423 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3425 /* Pre-conditions */
3426 BUG_ON(!ext4_ext_is_unwritten(ex));
3427 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3430 * Attempt to transfer newly initialized blocks from the currently
3431 * unwritten extent to its neighbor. This is much cheaper
3432 * than an insertion followed by a merge as those involve costly
3433 * memmove() calls. Transferring to the left is the common case in
3434 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3435 * followed by append writes.
3437 * Limitations of the current logic:
3438 * - L1: we do not deal with writes covering the whole extent.
3439 * This would require removing the extent if the transfer
3441 * - L2: we only attempt to merge with an extent stored in the
3442 * same extent tree node.
3444 if ((map->m_lblk == ee_block) &&
3445 /* See if we can merge left */
3446 (map_len < ee_len) && /*L1*/
3447 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3448 ext4_lblk_t prev_lblk;
3449 ext4_fsblk_t prev_pblk, ee_pblk;
3450 unsigned int prev_len;
3453 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3454 prev_len = ext4_ext_get_actual_len(abut_ex);
3455 prev_pblk = ext4_ext_pblock(abut_ex);
3456 ee_pblk = ext4_ext_pblock(ex);
3459 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3460 * upon those conditions:
3461 * - C1: abut_ex is initialized,
3462 * - C2: abut_ex is logically abutting ex,
3463 * - C3: abut_ex is physically abutting ex,
3464 * - C4: abut_ex can receive the additional blocks without
3465 * overflowing the (initialized) length limit.
3467 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3468 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3469 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3470 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3471 err = ext4_ext_get_access(handle, inode, path + depth);
3475 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3478 /* Shift the start of ex by 'map_len' blocks */
3479 ex->ee_block = cpu_to_le32(ee_block + map_len);
3480 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3481 ex->ee_len = cpu_to_le16(ee_len - map_len);
3482 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3484 /* Extend abut_ex by 'map_len' blocks */
3485 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3487 /* Result: number of initialized blocks past m_lblk */
3488 allocated = map_len;
3490 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3491 (map_len < ee_len) && /*L1*/
3492 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3493 /* See if we can merge right */
3494 ext4_lblk_t next_lblk;
3495 ext4_fsblk_t next_pblk, ee_pblk;
3496 unsigned int next_len;
3499 next_lblk = le32_to_cpu(abut_ex->ee_block);
3500 next_len = ext4_ext_get_actual_len(abut_ex);
3501 next_pblk = ext4_ext_pblock(abut_ex);
3502 ee_pblk = ext4_ext_pblock(ex);
3505 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3506 * upon those conditions:
3507 * - C1: abut_ex is initialized,
3508 * - C2: abut_ex is logically abutting ex,
3509 * - C3: abut_ex is physically abutting ex,
3510 * - C4: abut_ex can receive the additional blocks without
3511 * overflowing the (initialized) length limit.
3513 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3514 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3515 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3516 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3517 err = ext4_ext_get_access(handle, inode, path + depth);
3521 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3524 /* Shift the start of abut_ex by 'map_len' blocks */
3525 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3526 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3527 ex->ee_len = cpu_to_le16(ee_len - map_len);
3528 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3530 /* Extend abut_ex by 'map_len' blocks */
3531 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3533 /* Result: number of initialized blocks past m_lblk */
3534 allocated = map_len;
3538 /* Mark the block containing both extents as dirty */
3539 err = ext4_ext_dirty(handle, inode, path + depth);
3541 /* Update path to point to the right extent */
3542 path[depth].p_ext = abut_ex;
3545 allocated = ee_len - (map->m_lblk - ee_block);
3547 WARN_ON(map->m_lblk < ee_block);
3549 * It is safe to convert extent to initialized via explicit
3550 * zeroout only if extent is fully inside i_size or new_size.
3552 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3554 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3555 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3556 (inode->i_sb->s_blocksize_bits - 10);
3560 * 1. split the extent into three extents.
3561 * 2. split the extent into two extents, zeroout the head of the first
3563 * 3. split the extent into two extents, zeroout the tail of the second
3565 * 4. split the extent into two extents with out zeroout.
3566 * 5. no splitting needed, just possibly zeroout the head and / or the
3567 * tail of the extent.
3569 split_map.m_lblk = map->m_lblk;
3570 split_map.m_len = map->m_len;
3572 if (max_zeroout && (allocated > split_map.m_len)) {
3573 if (allocated <= max_zeroout) {
3576 cpu_to_le32(split_map.m_lblk +
3579 cpu_to_le16(allocated - split_map.m_len);
3580 ext4_ext_store_pblock(&zero_ex1,
3581 ext4_ext_pblock(ex) + split_map.m_lblk +
3582 split_map.m_len - ee_block);
3583 err = ext4_ext_zeroout(inode, &zero_ex1);
3586 split_map.m_len = allocated;
3588 if (split_map.m_lblk - ee_block + split_map.m_len <
3591 if (split_map.m_lblk != ee_block) {
3592 zero_ex2.ee_block = ex->ee_block;
3593 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3595 ext4_ext_store_pblock(&zero_ex2,
3596 ext4_ext_pblock(ex));
3597 err = ext4_ext_zeroout(inode, &zero_ex2);
3602 split_map.m_len += split_map.m_lblk - ee_block;
3603 split_map.m_lblk = ee_block;
3604 allocated = map->m_len;
3608 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3613 /* If we have gotten a failure, don't zero out status tree */
3615 err = ext4_zeroout_es(inode, &zero_ex1);
3617 err = ext4_zeroout_es(inode, &zero_ex2);
3619 return err ? err : allocated;
3623 * This function is called by ext4_ext_map_blocks() from
3624 * ext4_get_blocks_dio_write() when DIO to write
3625 * to an unwritten extent.
3627 * Writing to an unwritten extent may result in splitting the unwritten
3628 * extent into multiple initialized/unwritten extents (up to three)
3629 * There are three possibilities:
3630 * a> There is no split required: Entire extent should be unwritten
3631 * b> Splits in two extents: Write is happening at either end of the extent
3632 * c> Splits in three extents: Somone is writing in middle of the extent
3634 * This works the same way in the case of initialized -> unwritten conversion.
3636 * One of more index blocks maybe needed if the extent tree grow after
3637 * the unwritten extent split. To prevent ENOSPC occur at the IO
3638 * complete, we need to split the unwritten extent before DIO submit
3639 * the IO. The unwritten extent called at this time will be split
3640 * into three unwritten extent(at most). After IO complete, the part
3641 * being filled will be convert to initialized by the end_io callback function
3642 * via ext4_convert_unwritten_extents().
3644 * Returns the size of unwritten extent to be written on success.
3646 static int ext4_split_convert_extents(handle_t *handle,
3647 struct inode *inode,
3648 struct ext4_map_blocks *map,
3649 struct ext4_ext_path **ppath,
3652 struct ext4_ext_path *path = *ppath;
3653 ext4_lblk_t eof_block;
3654 ext4_lblk_t ee_block;
3655 struct ext4_extent *ex;
3656 unsigned int ee_len;
3657 int split_flag = 0, depth;
3659 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3660 (unsigned long long)map->m_lblk, map->m_len);
3662 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3663 >> inode->i_sb->s_blocksize_bits;
3664 if (eof_block < map->m_lblk + map->m_len)
3665 eof_block = map->m_lblk + map->m_len;
3667 * It is safe to convert extent to initialized via explicit
3668 * zeroout only if extent is fully inside i_size or new_size.
3670 depth = ext_depth(inode);
3671 ex = path[depth].p_ext;
3672 ee_block = le32_to_cpu(ex->ee_block);
3673 ee_len = ext4_ext_get_actual_len(ex);
3675 /* Convert to unwritten */
3676 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3677 split_flag |= EXT4_EXT_DATA_VALID1;
3678 /* Convert to initialized */
3679 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3680 split_flag |= ee_block + ee_len <= eof_block ?
3681 EXT4_EXT_MAY_ZEROOUT : 0;
3682 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3684 flags |= EXT4_GET_BLOCKS_PRE_IO;
3685 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3688 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3689 struct inode *inode,
3690 struct ext4_map_blocks *map,
3691 struct ext4_ext_path **ppath)
3693 struct ext4_ext_path *path = *ppath;
3694 struct ext4_extent *ex;
3695 ext4_lblk_t ee_block;
3696 unsigned int ee_len;
3700 depth = ext_depth(inode);
3701 ex = path[depth].p_ext;
3702 ee_block = le32_to_cpu(ex->ee_block);
3703 ee_len = ext4_ext_get_actual_len(ex);
3705 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3706 (unsigned long long)ee_block, ee_len);
3708 /* If extent is larger than requested it is a clear sign that we still
3709 * have some extent state machine issues left. So extent_split is still
3711 * TODO: Once all related issues will be fixed this situation should be
3714 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3715 #ifdef CONFIG_EXT4_DEBUG
3716 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3717 " len %u; IO logical block %llu, len %u",
3718 inode->i_ino, (unsigned long long)ee_block, ee_len,
3719 (unsigned long long)map->m_lblk, map->m_len);
3721 err = ext4_split_convert_extents(handle, inode, map, ppath,
3722 EXT4_GET_BLOCKS_CONVERT);
3725 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3727 return PTR_ERR(path);
3728 depth = ext_depth(inode);
3729 ex = path[depth].p_ext;
3732 err = ext4_ext_get_access(handle, inode, path + depth);
3735 /* first mark the extent as initialized */
3736 ext4_ext_mark_initialized(ex);
3738 /* note: ext4_ext_correct_indexes() isn't needed here because
3739 * borders are not changed
3741 ext4_ext_try_to_merge(handle, inode, path, ex);
3743 /* Mark modified extent as dirty */
3744 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3746 ext4_ext_show_leaf(inode, path);
3751 convert_initialized_extent(handle_t *handle, struct inode *inode,
3752 struct ext4_map_blocks *map,
3753 struct ext4_ext_path **ppath,
3754 unsigned int *allocated)
3756 struct ext4_ext_path *path = *ppath;
3757 struct ext4_extent *ex;
3758 ext4_lblk_t ee_block;
3759 unsigned int ee_len;
3764 * Make sure that the extent is no bigger than we support with
3767 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3768 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3770 depth = ext_depth(inode);
3771 ex = path[depth].p_ext;
3772 ee_block = le32_to_cpu(ex->ee_block);
3773 ee_len = ext4_ext_get_actual_len(ex);
3775 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3776 (unsigned long long)ee_block, ee_len);
3778 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3779 err = ext4_split_convert_extents(handle, inode, map, ppath,
3780 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3783 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3785 return PTR_ERR(path);
3786 depth = ext_depth(inode);
3787 ex = path[depth].p_ext;
3789 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3790 (unsigned long) map->m_lblk);
3791 return -EFSCORRUPTED;
3795 err = ext4_ext_get_access(handle, inode, path + depth);
3798 /* first mark the extent as unwritten */
3799 ext4_ext_mark_unwritten(ex);
3801 /* note: ext4_ext_correct_indexes() isn't needed here because
3802 * borders are not changed
3804 ext4_ext_try_to_merge(handle, inode, path, ex);
3806 /* Mark modified extent as dirty */
3807 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3810 ext4_ext_show_leaf(inode, path);
3812 ext4_update_inode_fsync_trans(handle, inode, 1);
3814 map->m_flags |= EXT4_MAP_UNWRITTEN;
3815 if (*allocated > map->m_len)
3816 *allocated = map->m_len;
3817 map->m_len = *allocated;
3822 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3823 struct ext4_map_blocks *map,
3824 struct ext4_ext_path **ppath, int flags,
3825 unsigned int allocated, ext4_fsblk_t newblock)
3827 struct ext4_ext_path __maybe_unused *path = *ppath;
3831 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3832 (unsigned long long)map->m_lblk, map->m_len, flags,
3834 ext4_ext_show_leaf(inode, path);
3837 * When writing into unwritten space, we should not fail to
3838 * allocate metadata blocks for the new extent block if needed.
3840 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3842 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3843 allocated, newblock);
3845 /* get_block() before submitting IO, split the extent */
3846 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3847 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3848 flags | EXT4_GET_BLOCKS_CONVERT);
3854 * shouldn't get a 0 return when splitting an extent unless
3855 * m_len is 0 (bug) or extent has been corrupted
3857 if (unlikely(ret == 0)) {
3858 EXT4_ERROR_INODE(inode,
3859 "unexpected ret == 0, m_len = %u",
3861 err = -EFSCORRUPTED;
3864 map->m_flags |= EXT4_MAP_UNWRITTEN;
3867 /* IO end_io complete, convert the filled extent to written */
3868 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3869 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3873 ext4_update_inode_fsync_trans(handle, inode, 1);
3876 /* buffered IO cases */
3878 * repeat fallocate creation request
3879 * we already have an unwritten extent
3881 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3882 map->m_flags |= EXT4_MAP_UNWRITTEN;
3886 /* buffered READ or buffered write_begin() lookup */
3887 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3889 * We have blocks reserved already. We
3890 * return allocated blocks so that delalloc
3891 * won't do block reservation for us. But
3892 * the buffer head will be unmapped so that
3893 * a read from the block returns 0s.
3895 map->m_flags |= EXT4_MAP_UNWRITTEN;
3900 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3901 * For buffered writes, at writepage time, etc. Convert a
3902 * discovered unwritten extent to written.
3904 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3909 ext4_update_inode_fsync_trans(handle, inode, 1);
3911 * shouldn't get a 0 return when converting an unwritten extent
3912 * unless m_len is 0 (bug) or extent has been corrupted
3914 if (unlikely(ret == 0)) {
3915 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3917 err = -EFSCORRUPTED;
3923 map->m_flags |= EXT4_MAP_NEW;
3925 map->m_flags |= EXT4_MAP_MAPPED;
3927 map->m_pblk = newblock;
3928 if (allocated > map->m_len)
3929 allocated = map->m_len;
3930 map->m_len = allocated;
3931 ext4_ext_show_leaf(inode, path);
3933 return err ? err : allocated;
3937 * get_implied_cluster_alloc - check to see if the requested
3938 * allocation (in the map structure) overlaps with a cluster already
3939 * allocated in an extent.
3940 * @sb The filesystem superblock structure
3941 * @map The requested lblk->pblk mapping
3942 * @ex The extent structure which might contain an implied
3943 * cluster allocation
3945 * This function is called by ext4_ext_map_blocks() after we failed to
3946 * find blocks that were already in the inode's extent tree. Hence,
3947 * we know that the beginning of the requested region cannot overlap
3948 * the extent from the inode's extent tree. There are three cases we
3949 * want to catch. The first is this case:
3951 * |--- cluster # N--|
3952 * |--- extent ---| |---- requested region ---|
3955 * The second case that we need to test for is this one:
3957 * |--------- cluster # N ----------------|
3958 * |--- requested region --| |------- extent ----|
3959 * |=======================|
3961 * The third case is when the requested region lies between two extents
3962 * within the same cluster:
3963 * |------------- cluster # N-------------|
3964 * |----- ex -----| |---- ex_right ----|
3965 * |------ requested region ------|
3966 * |================|
3968 * In each of the above cases, we need to set the map->m_pblk and
3969 * map->m_len so it corresponds to the return the extent labelled as
3970 * "|====|" from cluster #N, since it is already in use for data in
3971 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3972 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3973 * as a new "allocated" block region. Otherwise, we will return 0 and
3974 * ext4_ext_map_blocks() will then allocate one or more new clusters
3975 * by calling ext4_mb_new_blocks().
3977 static int get_implied_cluster_alloc(struct super_block *sb,
3978 struct ext4_map_blocks *map,
3979 struct ext4_extent *ex,
3980 struct ext4_ext_path *path)
3982 struct ext4_sb_info *sbi = EXT4_SB(sb);
3983 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3984 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3985 ext4_lblk_t rr_cluster_start;
3986 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3987 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3988 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3990 /* The extent passed in that we are trying to match */
3991 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3992 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3994 /* The requested region passed into ext4_map_blocks() */
3995 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3997 if ((rr_cluster_start == ex_cluster_end) ||
3998 (rr_cluster_start == ex_cluster_start)) {
3999 if (rr_cluster_start == ex_cluster_end)
4000 ee_start += ee_len - 1;
4001 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4002 map->m_len = min(map->m_len,
4003 (unsigned) sbi->s_cluster_ratio - c_offset);
4005 * Check for and handle this case:
4007 * |--------- cluster # N-------------|
4008 * |------- extent ----|
4009 * |--- requested region ---|
4013 if (map->m_lblk < ee_block)
4014 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4017 * Check for the case where there is already another allocated
4018 * block to the right of 'ex' but before the end of the cluster.
4020 * |------------- cluster # N-------------|
4021 * |----- ex -----| |---- ex_right ----|
4022 * |------ requested region ------|
4023 * |================|
4025 if (map->m_lblk > ee_block) {
4026 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4027 map->m_len = min(map->m_len, next - map->m_lblk);
4030 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4034 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4039 * Determine hole length around the given logical block, first try to
4040 * locate and expand the hole from the given @path, and then adjust it
4041 * if it's partially or completely converted to delayed extents, insert
4042 * it into the extent cache tree if it's indeed a hole, finally return
4043 * the length of the determined extent.
4045 static ext4_lblk_t ext4_ext_determine_insert_hole(struct inode *inode,
4046 struct ext4_ext_path *path,
4049 ext4_lblk_t hole_start, len;
4050 struct extent_status es;
4053 len = ext4_ext_find_hole(inode, path, &hole_start);
4055 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
4056 hole_start + len - 1, &es);
4061 * There's a delalloc extent in the hole, handle it if the delalloc
4062 * extent is in front of, behind and straddle the queried range.
4064 if (lblk >= es.es_lblk + es.es_len) {
4066 * The delalloc extent is in front of the queried range,
4067 * find again from the queried start block.
4069 len -= lblk - hole_start;
4072 } else if (in_range(lblk, es.es_lblk, es.es_len)) {
4074 * The delalloc extent containing lblk, it must have been
4075 * added after ext4_map_blocks() checked the extent status
4076 * tree, adjust the length to the delalloc extent's after
4079 len = es.es_lblk + es.es_len - lblk;
4083 * The delalloc extent is partially or completely behind
4084 * the queried range, update hole length until the
4085 * beginning of the delalloc extent.
4087 len = min(es.es_lblk - hole_start, len);
4091 /* Put just found gap into cache to speed up subsequent requests */
4092 ext_debug(inode, " -> %u:%u\n", hole_start, len);
4093 ext4_es_insert_extent(inode, hole_start, len, ~0, EXTENT_STATUS_HOLE);
4095 /* Update hole_len to reflect hole size after lblk */
4096 if (hole_start != lblk)
4097 len -= lblk - hole_start;
4103 * Block allocation/map/preallocation routine for extents based files
4106 * Need to be called with
4107 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4108 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4110 * return > 0, number of blocks already mapped/allocated
4111 * if create == 0 and these are pre-allocated blocks
4112 * buffer head is unmapped
4113 * otherwise blocks are mapped
4115 * return = 0, if plain look up failed (blocks have not been allocated)
4116 * buffer head is unmapped
4118 * return < 0, error case.
4120 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4121 struct ext4_map_blocks *map, int flags)
4123 struct ext4_ext_path *path = NULL;
4124 struct ext4_extent newex, *ex, ex2;
4125 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4126 ext4_fsblk_t newblock = 0, pblk;
4127 int err = 0, depth, ret;
4128 unsigned int allocated = 0, offset = 0;
4129 unsigned int allocated_clusters = 0;
4130 struct ext4_allocation_request ar;
4131 ext4_lblk_t cluster_offset;
4133 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4134 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4136 /* find extent for this block */
4137 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4139 err = PTR_ERR(path);
4144 depth = ext_depth(inode);
4147 * consistent leaf must not be empty;
4148 * this situation is possible, though, _during_ tree modification;
4149 * this is why assert can't be put in ext4_find_extent()
4151 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4152 EXT4_ERROR_INODE(inode, "bad extent address "
4153 "lblock: %lu, depth: %d pblock %lld",
4154 (unsigned long) map->m_lblk, depth,
4155 path[depth].p_block);
4156 err = -EFSCORRUPTED;
4160 ex = path[depth].p_ext;
4162 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4163 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4164 unsigned short ee_len;
4168 * unwritten extents are treated as holes, except that
4169 * we split out initialized portions during a write.
4171 ee_len = ext4_ext_get_actual_len(ex);
4173 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4175 /* if found extent covers block, simply return it */
4176 if (in_range(map->m_lblk, ee_block, ee_len)) {
4177 newblock = map->m_lblk - ee_block + ee_start;
4178 /* number of remaining blocks in the extent */
4179 allocated = ee_len - (map->m_lblk - ee_block);
4180 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4181 map->m_lblk, ee_block, ee_len, newblock);
4184 * If the extent is initialized check whether the
4185 * caller wants to convert it to unwritten.
4187 if ((!ext4_ext_is_unwritten(ex)) &&
4188 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4189 err = convert_initialized_extent(handle,
4190 inode, map, &path, &allocated);
4192 } else if (!ext4_ext_is_unwritten(ex)) {
4193 map->m_flags |= EXT4_MAP_MAPPED;
4194 map->m_pblk = newblock;
4195 if (allocated > map->m_len)
4196 allocated = map->m_len;
4197 map->m_len = allocated;
4198 ext4_ext_show_leaf(inode, path);
4202 ret = ext4_ext_handle_unwritten_extents(
4203 handle, inode, map, &path, flags,
4204 allocated, newblock);
4214 * requested block isn't allocated yet;
4215 * we couldn't try to create block if create flag is zero
4217 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4220 len = ext4_ext_determine_insert_hole(inode, path, map->m_lblk);
4223 map->m_len = min_t(unsigned int, map->m_len, len);
4228 * Okay, we need to do block allocation.
4230 newex.ee_block = cpu_to_le32(map->m_lblk);
4231 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4234 * If we are doing bigalloc, check to see if the extent returned
4235 * by ext4_find_extent() implies a cluster we can use.
4237 if (cluster_offset && ex &&
4238 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4239 ar.len = allocated = map->m_len;
4240 newblock = map->m_pblk;
4241 goto got_allocated_blocks;
4244 /* find neighbour allocated blocks */
4245 ar.lleft = map->m_lblk;
4246 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4249 ar.lright = map->m_lblk;
4250 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4254 /* Check if the extent after searching to the right implies a
4255 * cluster we can use. */
4256 if ((sbi->s_cluster_ratio > 1) && err &&
4257 get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4258 ar.len = allocated = map->m_len;
4259 newblock = map->m_pblk;
4260 goto got_allocated_blocks;
4264 * See if request is beyond maximum number of blocks we can have in
4265 * a single extent. For an initialized extent this limit is
4266 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4267 * EXT_UNWRITTEN_MAX_LEN.
4269 if (map->m_len > EXT_INIT_MAX_LEN &&
4270 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4271 map->m_len = EXT_INIT_MAX_LEN;
4272 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4273 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4274 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4276 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4277 newex.ee_len = cpu_to_le16(map->m_len);
4278 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4280 allocated = ext4_ext_get_actual_len(&newex);
4282 allocated = map->m_len;
4284 /* allocate new block */
4286 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4287 ar.logical = map->m_lblk;
4289 * We calculate the offset from the beginning of the cluster
4290 * for the logical block number, since when we allocate a
4291 * physical cluster, the physical block should start at the
4292 * same offset from the beginning of the cluster. This is
4293 * needed so that future calls to get_implied_cluster_alloc()
4296 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4297 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4299 ar.logical -= offset;
4300 if (S_ISREG(inode->i_mode))
4301 ar.flags = EXT4_MB_HINT_DATA;
4303 /* disable in-core preallocation for non-regular files */
4305 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4306 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4307 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4308 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4309 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4310 ar.flags |= EXT4_MB_USE_RESERVED;
4311 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4314 allocated_clusters = ar.len;
4315 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4316 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4317 ar.goal, newblock, ar.len, allocated);
4318 if (ar.len > allocated)
4321 got_allocated_blocks:
4322 /* try to insert new extent into found leaf and return */
4323 pblk = newblock + offset;
4324 ext4_ext_store_pblock(&newex, pblk);
4325 newex.ee_len = cpu_to_le16(ar.len);
4326 /* Mark unwritten */
4327 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4328 ext4_ext_mark_unwritten(&newex);
4329 map->m_flags |= EXT4_MAP_UNWRITTEN;
4332 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4334 if (allocated_clusters) {
4338 * free data blocks we just allocated.
4339 * not a good idea to call discard here directly,
4340 * but otherwise we'd need to call it every free().
4342 ext4_discard_preallocations(inode, 0);
4343 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4344 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4345 ext4_free_blocks(handle, inode, NULL, newblock,
4346 EXT4_C2B(sbi, allocated_clusters),
4353 * Reduce the reserved cluster count to reflect successful deferred
4354 * allocation of delayed allocated clusters or direct allocation of
4355 * clusters discovered to be delayed allocated. Once allocated, a
4356 * cluster is not included in the reserved count.
4358 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4359 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4361 * When allocating delayed allocated clusters, simply
4362 * reduce the reserved cluster count and claim quota
4364 ext4_da_update_reserve_space(inode, allocated_clusters,
4367 ext4_lblk_t lblk, len;
4371 * When allocating non-delayed allocated clusters
4372 * (from fallocate, filemap, DIO, or clusters
4373 * allocated when delalloc has been disabled by
4374 * ext4_nonda_switch), reduce the reserved cluster
4375 * count by the number of allocated clusters that
4376 * have previously been delayed allocated. Quota
4377 * has been claimed by ext4_mb_new_blocks() above,
4378 * so release the quota reservations made for any
4379 * previously delayed allocated clusters.
4381 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4382 len = allocated_clusters << sbi->s_cluster_bits;
4383 n = ext4_es_delayed_clu(inode, lblk, len);
4385 ext4_da_update_reserve_space(inode, (int) n, 0);
4390 * Cache the extent and update transaction to commit on fdatasync only
4391 * when it is _not_ an unwritten extent.
4393 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4394 ext4_update_inode_fsync_trans(handle, inode, 1);
4396 ext4_update_inode_fsync_trans(handle, inode, 0);
4398 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4400 map->m_len = ar.len;
4401 allocated = map->m_len;
4402 ext4_ext_show_leaf(inode, path);
4404 ext4_ext_drop_refs(path);
4407 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4408 err ? err : allocated);
4409 return err ? err : allocated;
4412 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4414 struct super_block *sb = inode->i_sb;
4415 ext4_lblk_t last_block;
4419 * TODO: optimization is possible here.
4420 * Probably we need not scan at all,
4421 * because page truncation is enough.
4424 /* we have to know where to truncate from in crash case */
4425 EXT4_I(inode)->i_disksize = inode->i_size;
4426 err = ext4_mark_inode_dirty(handle, inode);
4430 last_block = (inode->i_size + sb->s_blocksize - 1)
4431 >> EXT4_BLOCK_SIZE_BITS(sb);
4433 err = ext4_es_remove_extent(inode, last_block,
4434 EXT_MAX_BLOCKS - last_block);
4435 if (err == -ENOMEM) {
4437 congestion_wait(BLK_RW_ASYNC, HZ/50);
4443 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4444 if (err == -ENOMEM) {
4446 congestion_wait(BLK_RW_ASYNC, HZ/50);
4447 goto retry_remove_space;
4452 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4453 ext4_lblk_t len, loff_t new_size,
4456 struct inode *inode = file_inode(file);
4459 int ret2 = 0, ret3 = 0;
4462 struct ext4_map_blocks map;
4463 unsigned int credits;
4466 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4467 map.m_lblk = offset;
4470 * Don't normalize the request if it can fit in one extent so
4471 * that it doesn't get unnecessarily split into multiple
4474 if (len <= EXT_UNWRITTEN_MAX_LEN)
4475 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4478 * credits to insert 1 extent into extent tree
4480 credits = ext4_chunk_trans_blocks(inode, len);
4481 depth = ext_depth(inode);
4484 while (ret >= 0 && len) {
4486 * Recalculate credits when extent tree depth changes.
4488 if (depth != ext_depth(inode)) {
4489 credits = ext4_chunk_trans_blocks(inode, len);
4490 depth = ext_depth(inode);
4493 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4495 if (IS_ERR(handle)) {
4496 ret = PTR_ERR(handle);
4499 ret = ext4_map_blocks(handle, inode, &map, flags);
4501 ext4_debug("inode #%lu: block %u: len %u: "
4502 "ext4_ext_map_blocks returned %d",
4503 inode->i_ino, map.m_lblk,
4505 ext4_mark_inode_dirty(handle, inode);
4506 ret2 = ext4_journal_stop(handle);
4510 map.m_len = len = len - ret;
4511 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4512 inode->i_ctime = current_time(inode);
4514 if (epos > new_size)
4516 if (ext4_update_inode_size(inode, epos) & 0x1)
4517 inode->i_mtime = inode->i_ctime;
4519 ret2 = ext4_mark_inode_dirty(handle, inode);
4520 ext4_update_inode_fsync_trans(handle, inode, 1);
4521 ret3 = ext4_journal_stop(handle);
4522 ret2 = ret3 ? ret3 : ret2;
4526 if (ret == -ENOSPC &&
4527 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4532 return ret > 0 ? ret2 : ret;
4535 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4537 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4539 static long ext4_zero_range(struct file *file, loff_t offset,
4540 loff_t len, int mode)
4542 struct inode *inode = file_inode(file);
4543 handle_t *handle = NULL;
4544 unsigned int max_blocks;
4545 loff_t new_size = 0;
4549 int partial_begin, partial_end;
4552 unsigned int blkbits = inode->i_blkbits;
4554 trace_ext4_zero_range(inode, offset, len, mode);
4556 /* Call ext4_force_commit to flush all data in case of data=journal. */
4557 if (ext4_should_journal_data(inode)) {
4558 ret = ext4_force_commit(inode->i_sb);
4564 * Round up offset. This is not fallocate, we need to zero out
4565 * blocks, so convert interior block aligned part of the range to
4566 * unwritten and possibly manually zero out unaligned parts of the
4569 start = round_up(offset, 1 << blkbits);
4570 end = round_down((offset + len), 1 << blkbits);
4572 if (start < offset || end > offset + len)
4574 partial_begin = offset & ((1 << blkbits) - 1);
4575 partial_end = (offset + len) & ((1 << blkbits) - 1);
4577 lblk = start >> blkbits;
4578 max_blocks = (end >> blkbits);
4579 if (max_blocks < lblk)
4587 * Indirect files do not support unwritten extents
4589 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4594 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4595 (offset + len > inode->i_size ||
4596 offset + len > EXT4_I(inode)->i_disksize)) {
4597 new_size = offset + len;
4598 ret = inode_newsize_ok(inode, new_size);
4603 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4605 /* Wait all existing dio workers, newcomers will block on i_mutex */
4606 inode_dio_wait(inode);
4608 ret = file_modified(file);
4612 /* Preallocate the range including the unaligned edges */
4613 if (partial_begin || partial_end) {
4614 ret = ext4_alloc_file_blocks(file,
4615 round_down(offset, 1 << blkbits) >> blkbits,
4616 (round_up((offset + len), 1 << blkbits) -
4617 round_down(offset, 1 << blkbits)) >> blkbits,
4624 /* Zero range excluding the unaligned edges */
4625 if (max_blocks > 0) {
4626 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4630 * Prevent page faults from reinstantiating pages we have
4631 * released from page cache.
4633 down_write(&EXT4_I(inode)->i_mmap_sem);
4635 ret = ext4_break_layouts(inode);
4637 up_write(&EXT4_I(inode)->i_mmap_sem);
4641 ret = ext4_update_disksize_before_punch(inode, offset, len);
4643 up_write(&EXT4_I(inode)->i_mmap_sem);
4646 /* Now release the pages and zero block aligned part of pages */
4647 truncate_pagecache_range(inode, start, end - 1);
4648 inode->i_mtime = inode->i_ctime = current_time(inode);
4650 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4652 up_write(&EXT4_I(inode)->i_mmap_sem);
4656 if (!partial_begin && !partial_end)
4660 * In worst case we have to writeout two nonadjacent unwritten
4661 * blocks and update the inode
4663 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4664 if (ext4_should_journal_data(inode))
4666 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4667 if (IS_ERR(handle)) {
4668 ret = PTR_ERR(handle);
4669 ext4_std_error(inode->i_sb, ret);
4673 inode->i_mtime = inode->i_ctime = current_time(inode);
4675 ext4_update_inode_size(inode, new_size);
4676 ret = ext4_mark_inode_dirty(handle, inode);
4679 /* Zero out partial block at the edges of the range */
4680 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4682 ext4_update_inode_fsync_trans(handle, inode, 1);
4684 if (file->f_flags & O_SYNC)
4685 ext4_handle_sync(handle);
4688 ext4_journal_stop(handle);
4690 inode_unlock(inode);
4695 * preallocate space for a file. This implements ext4's fallocate file
4696 * operation, which gets called from sys_fallocate system call.
4697 * For block-mapped files, posix_fallocate should fall back to the method
4698 * of writing zeroes to the required new blocks (the same behavior which is
4699 * expected for file systems which do not support fallocate() system call).
4701 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4703 struct inode *inode = file_inode(file);
4704 loff_t new_size = 0;
4705 unsigned int max_blocks;
4709 unsigned int blkbits = inode->i_blkbits;
4712 * Encrypted inodes can't handle collapse range or insert
4713 * range since we would need to re-encrypt blocks with a
4714 * different IV or XTS tweak (which are based on the logical
4717 if (IS_ENCRYPTED(inode) &&
4718 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4721 /* Return error if mode is not supported */
4722 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4723 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4724 FALLOC_FL_INSERT_RANGE))
4728 ret = ext4_convert_inline_data(inode);
4729 inode_unlock(inode);
4733 if (mode & FALLOC_FL_PUNCH_HOLE) {
4734 ret = ext4_punch_hole(file, offset, len);
4738 if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4739 ret = ext4_collapse_range(file, offset, len);
4743 if (mode & FALLOC_FL_INSERT_RANGE) {
4744 ret = ext4_insert_range(file, offset, len);
4748 if (mode & FALLOC_FL_ZERO_RANGE) {
4749 ret = ext4_zero_range(file, offset, len, mode);
4752 trace_ext4_fallocate_enter(inode, offset, len, mode);
4753 lblk = offset >> blkbits;
4755 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4756 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4761 * We only support preallocation for extent-based files only
4763 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4768 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4769 (offset + len > inode->i_size ||
4770 offset + len > EXT4_I(inode)->i_disksize)) {
4771 new_size = offset + len;
4772 ret = inode_newsize_ok(inode, new_size);
4777 /* Wait all existing dio workers, newcomers will block on i_mutex */
4778 inode_dio_wait(inode);
4780 ret = file_modified(file);
4784 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4788 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4789 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4790 EXT4_I(inode)->i_sync_tid);
4793 inode_unlock(inode);
4794 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4800 * This function convert a range of blocks to written extents
4801 * The caller of this function will pass the start offset and the size.
4802 * all unwritten extents within this range will be converted to
4805 * This function is called from the direct IO end io call back
4806 * function, to convert the fallocated extents after IO is completed.
4807 * Returns 0 on success.
4809 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4810 loff_t offset, ssize_t len)
4812 unsigned int max_blocks;
4813 int ret = 0, ret2 = 0, ret3 = 0;
4814 struct ext4_map_blocks map;
4815 unsigned int blkbits = inode->i_blkbits;
4816 unsigned int credits = 0;
4818 map.m_lblk = offset >> blkbits;
4819 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4823 * credits to insert 1 extent into extent tree
4825 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4827 while (ret >= 0 && ret < max_blocks) {
4829 map.m_len = (max_blocks -= ret);
4831 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4833 if (IS_ERR(handle)) {
4834 ret = PTR_ERR(handle);
4838 ret = ext4_map_blocks(handle, inode, &map,
4839 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4841 ext4_warning(inode->i_sb,
4842 "inode #%lu: block %u: len %u: "
4843 "ext4_ext_map_blocks returned %d",
4844 inode->i_ino, map.m_lblk,
4846 ret2 = ext4_mark_inode_dirty(handle, inode);
4848 ret3 = ext4_journal_stop(handle);
4853 if (ret <= 0 || ret2)
4856 return ret > 0 ? ret2 : ret;
4859 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4861 int ret = 0, err = 0;
4862 struct ext4_io_end_vec *io_end_vec;
4865 * This is somewhat ugly but the idea is clear: When transaction is
4866 * reserved, everything goes into it. Otherwise we rather start several
4867 * smaller transactions for conversion of each extent separately.
4870 handle = ext4_journal_start_reserved(handle,
4871 EXT4_HT_EXT_CONVERT);
4873 return PTR_ERR(handle);
4876 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4877 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4885 err = ext4_journal_stop(handle);
4887 return ret < 0 ? ret : err;
4890 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4894 int blockbits = inode->i_sb->s_blocksize_bits;
4899 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4900 struct ext4_iloc iloc;
4901 int offset; /* offset of xattr in inode */
4903 error = ext4_get_inode_loc(inode, &iloc);
4906 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4907 offset = EXT4_GOOD_OLD_INODE_SIZE +
4908 EXT4_I(inode)->i_extra_isize;
4910 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4912 iomap_type = IOMAP_INLINE;
4913 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4914 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4915 length = inode->i_sb->s_blocksize;
4916 iomap_type = IOMAP_MAPPED;
4918 /* no in-inode or external block for xattr, so return -ENOENT */
4923 iomap->addr = physical;
4925 iomap->length = length;
4926 iomap->type = iomap_type;
4932 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4933 loff_t length, unsigned flags,
4934 struct iomap *iomap, struct iomap *srcmap)
4938 error = ext4_iomap_xattr_fiemap(inode, iomap);
4939 if (error == 0 && (offset >= iomap->length))
4944 static const struct iomap_ops ext4_iomap_xattr_ops = {
4945 .iomap_begin = ext4_iomap_xattr_begin,
4948 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4952 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4953 maxbytes = inode->i_sb->s_maxbytes;
4955 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4959 if (start > maxbytes)
4963 * Shrink request scope to what the fs can actually handle.
4965 if (*len > maxbytes || (maxbytes - *len) < start)
4966 *len = maxbytes - start;
4970 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4975 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4976 error = ext4_ext_precache(inode);
4979 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4983 * For bitmap files the maximum size limit could be smaller than
4984 * s_maxbytes, so check len here manually instead of just relying on the
4987 error = ext4_fiemap_check_ranges(inode, start, &len);
4991 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4992 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4993 return iomap_fiemap(inode, fieinfo, start, len,
4994 &ext4_iomap_xattr_ops);
4997 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
5000 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5001 __u64 start, __u64 len)
5003 ext4_lblk_t start_blk, len_blks;
5007 if (ext4_has_inline_data(inode)) {
5010 down_read(&EXT4_I(inode)->xattr_sem);
5011 has_inline = ext4_has_inline_data(inode);
5012 up_read(&EXT4_I(inode)->xattr_sem);
5017 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5018 error = ext4_ext_precache(inode);
5021 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5024 error = fiemap_prep(inode, fieinfo, start, &len, 0);
5028 error = ext4_fiemap_check_ranges(inode, start, &len);
5032 start_blk = start >> inode->i_sb->s_blocksize_bits;
5033 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5034 if (last_blk >= EXT_MAX_BLOCKS)
5035 last_blk = EXT_MAX_BLOCKS-1;
5036 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5039 * Walk the extent tree gathering extent information
5040 * and pushing extents back to the user.
5042 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5046 * ext4_ext_shift_path_extents:
5047 * Shift the extents of a path structure lying between path[depth].p_ext
5048 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5049 * if it is right shift or left shift operation.
5052 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5053 struct inode *inode, handle_t *handle,
5054 enum SHIFT_DIRECTION SHIFT)
5057 struct ext4_extent *ex_start, *ex_last;
5058 bool update = false;
5059 int credits, restart_credits;
5060 depth = path->p_depth;
5062 while (depth >= 0) {
5063 if (depth == path->p_depth) {
5064 ex_start = path[depth].p_ext;
5066 return -EFSCORRUPTED;
5068 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5069 /* leaf + sb + inode */
5071 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5073 /* extent tree + sb + inode */
5074 credits = depth + 2;
5077 restart_credits = ext4_writepage_trans_blocks(inode);
5078 err = ext4_datasem_ensure_credits(handle, inode, credits,
5079 restart_credits, 0);
5086 err = ext4_ext_get_access(handle, inode, path + depth);
5090 while (ex_start <= ex_last) {
5091 if (SHIFT == SHIFT_LEFT) {
5092 le32_add_cpu(&ex_start->ee_block,
5094 /* Try to merge to the left. */
5096 EXT_FIRST_EXTENT(path[depth].p_hdr))
5098 ext4_ext_try_to_merge_right(inode,
5099 path, ex_start - 1))
5104 le32_add_cpu(&ex_last->ee_block, shift);
5105 ext4_ext_try_to_merge_right(inode, path,
5110 err = ext4_ext_dirty(handle, inode, path + depth);
5114 if (--depth < 0 || !update)
5118 /* Update index too */
5119 err = ext4_ext_get_access(handle, inode, path + depth);
5123 if (SHIFT == SHIFT_LEFT)
5124 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5126 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5127 err = ext4_ext_dirty(handle, inode, path + depth);
5131 /* we are done if current index is not a starting index */
5132 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5143 * ext4_ext_shift_extents:
5144 * All the extents which lies in the range from @start to the last allocated
5145 * block for the @inode are shifted either towards left or right (depending
5146 * upon @SHIFT) by @shift blocks.
5147 * On success, 0 is returned, error otherwise.
5150 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5151 ext4_lblk_t start, ext4_lblk_t shift,
5152 enum SHIFT_DIRECTION SHIFT)
5154 struct ext4_ext_path *path;
5156 struct ext4_extent *extent;
5157 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5158 ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5160 /* Let path point to the last extent */
5161 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5164 return PTR_ERR(path);
5166 depth = path->p_depth;
5167 extent = path[depth].p_ext;
5171 stop = le32_to_cpu(extent->ee_block);
5174 * For left shifts, make sure the hole on the left is big enough to
5175 * accommodate the shift. For right shifts, make sure the last extent
5176 * won't be shifted beyond EXT_MAX_BLOCKS.
5178 if (SHIFT == SHIFT_LEFT) {
5179 path = ext4_find_extent(inode, start - 1, &path,
5182 return PTR_ERR(path);
5183 depth = path->p_depth;
5184 extent = path[depth].p_ext;
5186 ex_start = le32_to_cpu(extent->ee_block);
5187 ex_end = le32_to_cpu(extent->ee_block) +
5188 ext4_ext_get_actual_len(extent);
5194 if ((start == ex_start && shift > ex_start) ||
5195 (shift > start - ex_end)) {
5200 if (shift > EXT_MAX_BLOCKS -
5201 (stop + ext4_ext_get_actual_len(extent))) {
5208 * In case of left shift, iterator points to start and it is increased
5209 * till we reach stop. In case of right shift, iterator points to stop
5210 * and it is decreased till we reach start.
5214 if (SHIFT == SHIFT_LEFT)
5219 if (tmp != EXT_MAX_BLOCKS)
5223 * Its safe to start updating extents. Start and stop are unsigned, so
5224 * in case of right shift if extent with 0 block is reached, iterator
5225 * becomes NULL to indicate the end of the loop.
5227 while (iterator && start <= stop) {
5228 path = ext4_find_extent(inode, *iterator, &path,
5231 return PTR_ERR(path);
5232 depth = path->p_depth;
5233 extent = path[depth].p_ext;
5235 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5236 (unsigned long) *iterator);
5237 return -EFSCORRUPTED;
5239 if (SHIFT == SHIFT_LEFT && *iterator >
5240 le32_to_cpu(extent->ee_block)) {
5241 /* Hole, move to the next extent */
5242 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5243 path[depth].p_ext++;
5245 *iterator = ext4_ext_next_allocated_block(path);
5251 if (SHIFT == SHIFT_LEFT) {
5252 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5253 *iterator = le32_to_cpu(extent->ee_block) +
5254 ext4_ext_get_actual_len(extent);
5256 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5257 if (le32_to_cpu(extent->ee_block) > start)
5258 *iterator = le32_to_cpu(extent->ee_block) - 1;
5259 else if (le32_to_cpu(extent->ee_block) == start)
5262 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5263 while (le32_to_cpu(extent->ee_block) >= start)
5266 if (extent == EXT_LAST_EXTENT(path[depth].p_hdr))
5272 path[depth].p_ext = extent;
5274 ret = ext4_ext_shift_path_extents(path, shift, inode,
5276 /* iterator can be NULL which means we should break */
5283 ext4_ext_drop_refs(path);
5289 * ext4_collapse_range:
5290 * This implements the fallocate's collapse range functionality for ext4
5291 * Returns: 0 and non-zero on error.
5293 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5295 struct inode *inode = file_inode(file);
5296 struct super_block *sb = inode->i_sb;
5297 ext4_lblk_t punch_start, punch_stop;
5299 unsigned int credits;
5300 loff_t new_size, ioffset;
5304 * We need to test this early because xfstests assumes that a
5305 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5306 * system does not support collapse range.
5308 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5311 /* Collapse range works only on fs cluster size aligned regions. */
5312 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5315 trace_ext4_collapse_range(inode, offset, len);
5317 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5318 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5320 /* Call ext4_force_commit to flush all data in case of data=journal. */
5321 if (ext4_should_journal_data(inode)) {
5322 ret = ext4_force_commit(inode->i_sb);
5329 * There is no need to overlap collapse range with EOF, in which case
5330 * it is effectively a truncate operation
5332 if (offset + len >= inode->i_size) {
5337 /* Currently just for extent based files */
5338 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5343 /* Wait for existing dio to complete */
5344 inode_dio_wait(inode);
5346 ret = file_modified(file);
5351 * Prevent page faults from reinstantiating pages we have released from
5354 down_write(&EXT4_I(inode)->i_mmap_sem);
5356 ret = ext4_break_layouts(inode);
5361 * Need to round down offset to be aligned with page size boundary
5362 * for page size > block size.
5364 ioffset = round_down(offset, PAGE_SIZE);
5366 * Write tail of the last page before removed range since it will get
5367 * removed from the page cache below.
5369 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5373 * Write data that will be shifted to preserve them when discarding
5374 * page cache below. We are also protected from pages becoming dirty
5377 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5381 truncate_pagecache(inode, ioffset);
5383 credits = ext4_writepage_trans_blocks(inode);
5384 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5385 if (IS_ERR(handle)) {
5386 ret = PTR_ERR(handle);
5389 ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5391 down_write(&EXT4_I(inode)->i_data_sem);
5392 ext4_discard_preallocations(inode, 0);
5394 ret = ext4_es_remove_extent(inode, punch_start,
5395 EXT_MAX_BLOCKS - punch_start);
5397 up_write(&EXT4_I(inode)->i_data_sem);
5401 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5403 up_write(&EXT4_I(inode)->i_data_sem);
5406 ext4_discard_preallocations(inode, 0);
5408 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5409 punch_stop - punch_start, SHIFT_LEFT);
5411 up_write(&EXT4_I(inode)->i_data_sem);
5415 new_size = inode->i_size - len;
5416 i_size_write(inode, new_size);
5417 EXT4_I(inode)->i_disksize = new_size;
5419 up_write(&EXT4_I(inode)->i_data_sem);
5421 ext4_handle_sync(handle);
5422 inode->i_mtime = inode->i_ctime = current_time(inode);
5423 ret = ext4_mark_inode_dirty(handle, inode);
5424 ext4_update_inode_fsync_trans(handle, inode, 1);
5427 ext4_journal_stop(handle);
5428 ext4_fc_stop_ineligible(sb);
5430 up_write(&EXT4_I(inode)->i_mmap_sem);
5432 inode_unlock(inode);
5437 * ext4_insert_range:
5438 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5439 * The data blocks starting from @offset to the EOF are shifted by @len
5440 * towards right to create a hole in the @inode. Inode size is increased
5442 * Returns 0 on success, error otherwise.
5444 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5446 struct inode *inode = file_inode(file);
5447 struct super_block *sb = inode->i_sb;
5449 struct ext4_ext_path *path;
5450 struct ext4_extent *extent;
5451 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5452 unsigned int credits, ee_len;
5453 int ret = 0, depth, split_flag = 0;
5457 * We need to test this early because xfstests assumes that an
5458 * insert range of (0, 1) will return EOPNOTSUPP if the file
5459 * system does not support insert range.
5461 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5464 /* Insert range works only on fs cluster size aligned regions. */
5465 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5468 trace_ext4_insert_range(inode, offset, len);
5470 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5471 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5473 /* Call ext4_force_commit to flush all data in case of data=journal */
5474 if (ext4_should_journal_data(inode)) {
5475 ret = ext4_force_commit(inode->i_sb);
5481 /* Currently just for extent based files */
5482 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5487 /* Check whether the maximum file size would be exceeded */
5488 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5493 /* Offset must be less than i_size */
5494 if (offset >= inode->i_size) {
5499 /* Wait for existing dio to complete */
5500 inode_dio_wait(inode);
5502 ret = file_modified(file);
5507 * Prevent page faults from reinstantiating pages we have released from
5510 down_write(&EXT4_I(inode)->i_mmap_sem);
5512 ret = ext4_break_layouts(inode);
5517 * Need to round down to align start offset to page size boundary
5518 * for page size > block size.
5520 ioffset = round_down(offset, PAGE_SIZE);
5521 /* Write out all dirty pages */
5522 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5526 truncate_pagecache(inode, ioffset);
5528 credits = ext4_writepage_trans_blocks(inode);
5529 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5530 if (IS_ERR(handle)) {
5531 ret = PTR_ERR(handle);
5534 ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5536 /* Expand file to avoid data loss if there is error while shifting */
5537 inode->i_size += len;
5538 EXT4_I(inode)->i_disksize += len;
5539 inode->i_mtime = inode->i_ctime = current_time(inode);
5540 ret = ext4_mark_inode_dirty(handle, inode);
5544 down_write(&EXT4_I(inode)->i_data_sem);
5545 ext4_discard_preallocations(inode, 0);
5547 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5549 up_write(&EXT4_I(inode)->i_data_sem);
5553 depth = ext_depth(inode);
5554 extent = path[depth].p_ext;
5556 ee_start_lblk = le32_to_cpu(extent->ee_block);
5557 ee_len = ext4_ext_get_actual_len(extent);
5560 * If offset_lblk is not the starting block of extent, split
5561 * the extent @offset_lblk
5563 if ((offset_lblk > ee_start_lblk) &&
5564 (offset_lblk < (ee_start_lblk + ee_len))) {
5565 if (ext4_ext_is_unwritten(extent))
5566 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5567 EXT4_EXT_MARK_UNWRIT2;
5568 ret = ext4_split_extent_at(handle, inode, &path,
5569 offset_lblk, split_flag,
5571 EXT4_GET_BLOCKS_PRE_IO |
5572 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5575 ext4_ext_drop_refs(path);
5578 up_write(&EXT4_I(inode)->i_data_sem);
5582 ext4_ext_drop_refs(path);
5586 ret = ext4_es_remove_extent(inode, offset_lblk,
5587 EXT_MAX_BLOCKS - offset_lblk);
5589 up_write(&EXT4_I(inode)->i_data_sem);
5594 * if offset_lblk lies in a hole which is at start of file, use
5595 * ee_start_lblk to shift extents
5597 ret = ext4_ext_shift_extents(inode, handle,
5598 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5599 len_lblk, SHIFT_RIGHT);
5601 up_write(&EXT4_I(inode)->i_data_sem);
5603 ext4_handle_sync(handle);
5605 ext4_update_inode_fsync_trans(handle, inode, 1);
5608 ext4_journal_stop(handle);
5609 ext4_fc_stop_ineligible(sb);
5611 up_write(&EXT4_I(inode)->i_mmap_sem);
5613 inode_unlock(inode);
5618 * ext4_swap_extents() - Swap extents between two inodes
5619 * @handle: handle for this transaction
5620 * @inode1: First inode
5621 * @inode2: Second inode
5622 * @lblk1: Start block for first inode
5623 * @lblk2: Start block for second inode
5624 * @count: Number of blocks to swap
5625 * @unwritten: Mark second inode's extents as unwritten after swap
5626 * @erp: Pointer to save error value
5628 * This helper routine does exactly what is promise "swap extents". All other
5629 * stuff such as page-cache locking consistency, bh mapping consistency or
5630 * extent's data copying must be performed by caller.
5632 * i_mutex is held for both inodes
5633 * i_data_sem is locked for write for both inodes
5635 * All pages from requested range are locked for both inodes
5638 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5639 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5640 ext4_lblk_t count, int unwritten, int *erp)
5642 struct ext4_ext_path *path1 = NULL;
5643 struct ext4_ext_path *path2 = NULL;
5644 int replaced_count = 0;
5646 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5647 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5648 BUG_ON(!inode_is_locked(inode1));
5649 BUG_ON(!inode_is_locked(inode2));
5651 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5654 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5659 struct ext4_extent *ex1, *ex2, tmp_ex;
5660 ext4_lblk_t e1_blk, e2_blk;
5661 int e1_len, e2_len, len;
5664 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5665 if (IS_ERR(path1)) {
5666 *erp = PTR_ERR(path1);
5672 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5673 if (IS_ERR(path2)) {
5674 *erp = PTR_ERR(path2);
5678 ex1 = path1[path1->p_depth].p_ext;
5679 ex2 = path2[path2->p_depth].p_ext;
5680 /* Do we have something to swap ? */
5681 if (unlikely(!ex2 || !ex1))
5684 e1_blk = le32_to_cpu(ex1->ee_block);
5685 e2_blk = le32_to_cpu(ex2->ee_block);
5686 e1_len = ext4_ext_get_actual_len(ex1);
5687 e2_len = ext4_ext_get_actual_len(ex2);
5690 if (!in_range(lblk1, e1_blk, e1_len) ||
5691 !in_range(lblk2, e2_blk, e2_len)) {
5692 ext4_lblk_t next1, next2;
5694 /* if hole after extent, then go to next extent */
5695 next1 = ext4_ext_next_allocated_block(path1);
5696 next2 = ext4_ext_next_allocated_block(path2);
5697 /* If hole before extent, then shift to that extent */
5702 /* Do we have something to swap */
5703 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5705 /* Move to the rightest boundary */
5706 len = next1 - lblk1;
5707 if (len < next2 - lblk2)
5708 len = next2 - lblk2;
5717 /* Prepare left boundary */
5718 if (e1_blk < lblk1) {
5720 *erp = ext4_force_split_extent_at(handle, inode1,
5725 if (e2_blk < lblk2) {
5727 *erp = ext4_force_split_extent_at(handle, inode2,
5732 /* ext4_split_extent_at() may result in leaf extent split,
5733 * path must to be revalidated. */
5737 /* Prepare right boundary */
5739 if (len > e1_blk + e1_len - lblk1)
5740 len = e1_blk + e1_len - lblk1;
5741 if (len > e2_blk + e2_len - lblk2)
5742 len = e2_blk + e2_len - lblk2;
5744 if (len != e1_len) {
5746 *erp = ext4_force_split_extent_at(handle, inode1,
5747 &path1, lblk1 + len, 0);
5751 if (len != e2_len) {
5753 *erp = ext4_force_split_extent_at(handle, inode2,
5754 &path2, lblk2 + len, 0);
5758 /* ext4_split_extent_at() may result in leaf extent split,
5759 * path must to be revalidated. */
5763 BUG_ON(e2_len != e1_len);
5764 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5767 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5771 /* Both extents are fully inside boundaries. Swap it now */
5773 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5774 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5775 ex1->ee_len = cpu_to_le16(e2_len);
5776 ex2->ee_len = cpu_to_le16(e1_len);
5778 ext4_ext_mark_unwritten(ex2);
5779 if (ext4_ext_is_unwritten(&tmp_ex))
5780 ext4_ext_mark_unwritten(ex1);
5782 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5783 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5784 *erp = ext4_ext_dirty(handle, inode2, path2 +
5788 *erp = ext4_ext_dirty(handle, inode1, path1 +
5791 * Looks scarry ah..? second inode already points to new blocks,
5792 * and it was successfully dirtied. But luckily error may happen
5793 * only due to journal error, so full transaction will be
5800 replaced_count += len;
5804 ext4_ext_drop_refs(path1);
5806 ext4_ext_drop_refs(path2);
5808 path1 = path2 = NULL;
5810 return replaced_count;
5814 * ext4_clu_mapped - determine whether any block in a logical cluster has
5815 * been mapped to a physical cluster
5817 * @inode - file containing the logical cluster
5818 * @lclu - logical cluster of interest
5820 * Returns 1 if any block in the logical cluster is mapped, signifying
5821 * that a physical cluster has been allocated for it. Otherwise,
5822 * returns 0. Can also return negative error codes. Derived from
5823 * ext4_ext_map_blocks().
5825 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5827 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5828 struct ext4_ext_path *path;
5829 int depth, mapped = 0, err = 0;
5830 struct ext4_extent *extent;
5831 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5834 * if data can be stored inline, the logical cluster isn't
5835 * mapped - no physical clusters have been allocated, and the
5836 * file has no extents
5838 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) ||
5839 ext4_has_inline_data(inode))
5842 /* search for the extent closest to the first block in the cluster */
5843 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5845 err = PTR_ERR(path);
5850 depth = ext_depth(inode);
5853 * A consistent leaf must not be empty. This situation is possible,
5854 * though, _during_ tree modification, and it's why an assert can't
5855 * be put in ext4_find_extent().
5857 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5858 EXT4_ERROR_INODE(inode,
5859 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5860 (unsigned long) EXT4_C2B(sbi, lclu),
5861 depth, path[depth].p_block);
5862 err = -EFSCORRUPTED;
5866 extent = path[depth].p_ext;
5868 /* can't be mapped if the extent tree is empty */
5872 first_lblk = le32_to_cpu(extent->ee_block);
5873 first_lclu = EXT4_B2C(sbi, first_lblk);
5876 * Three possible outcomes at this point - found extent spanning
5877 * the target cluster, to the left of the target cluster, or to the
5878 * right of the target cluster. The first two cases are handled here.
5879 * The last case indicates the target cluster is not mapped.
5881 if (lclu >= first_lclu) {
5882 last_lclu = EXT4_B2C(sbi, first_lblk +
5883 ext4_ext_get_actual_len(extent) - 1);
5884 if (lclu <= last_lclu) {
5887 first_lblk = ext4_ext_next_allocated_block(path);
5888 first_lclu = EXT4_B2C(sbi, first_lblk);
5889 if (lclu == first_lclu)
5895 ext4_ext_drop_refs(path);
5898 return err ? err : mapped;
5902 * Updates physical block address and unwritten status of extent
5903 * starting at lblk start and of len. If such an extent doesn't exist,
5904 * this function splits the extent tree appropriately to create an
5905 * extent like this. This function is called in the fast commit
5906 * replay path. Returns 0 on success and error on failure.
5908 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5909 int len, int unwritten, ext4_fsblk_t pblk)
5911 struct ext4_ext_path *path = NULL, *ppath;
5912 struct ext4_extent *ex;
5915 path = ext4_find_extent(inode, start, NULL, 0);
5917 return PTR_ERR(path);
5918 ex = path[path->p_depth].p_ext;
5920 ret = -EFSCORRUPTED;
5924 if (le32_to_cpu(ex->ee_block) != start ||
5925 ext4_ext_get_actual_len(ex) != len) {
5926 /* We need to split this extent to match our extent first */
5928 down_write(&EXT4_I(inode)->i_data_sem);
5929 ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5930 up_write(&EXT4_I(inode)->i_data_sem);
5934 path = ext4_find_extent(inode, start, NULL, 0);
5938 ex = path[path->p_depth].p_ext;
5939 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5940 if (ext4_ext_get_actual_len(ex) != len) {
5941 down_write(&EXT4_I(inode)->i_data_sem);
5942 ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5944 up_write(&EXT4_I(inode)->i_data_sem);
5948 path = ext4_find_extent(inode, start, NULL, 0);
5951 ex = path[path->p_depth].p_ext;
5955 ext4_ext_mark_unwritten(ex);
5957 ext4_ext_mark_initialized(ex);
5958 ext4_ext_store_pblock(ex, pblk);
5959 down_write(&EXT4_I(inode)->i_data_sem);
5960 ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5961 up_write(&EXT4_I(inode)->i_data_sem);
5963 ext4_ext_drop_refs(path);
5965 ext4_mark_inode_dirty(NULL, inode);
5969 /* Try to shrink the extent tree */
5970 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5972 struct ext4_ext_path *path = NULL;
5973 struct ext4_extent *ex;
5974 ext4_lblk_t old_cur, cur = 0;
5977 path = ext4_find_extent(inode, cur, NULL, 0);
5980 ex = path[path->p_depth].p_ext;
5982 ext4_ext_drop_refs(path);
5984 ext4_mark_inode_dirty(NULL, inode);
5988 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5991 ext4_ext_try_to_merge(NULL, inode, path, ex);
5992 down_write(&EXT4_I(inode)->i_data_sem);
5993 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5994 up_write(&EXT4_I(inode)->i_data_sem);
5995 ext4_mark_inode_dirty(NULL, inode);
5996 ext4_ext_drop_refs(path);
6001 /* Check if *cur is a hole and if it is, skip it */
6002 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
6005 struct ext4_map_blocks map;
6008 map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
6010 ret = ext4_map_blocks(NULL, inode, &map, 0);
6015 *cur = *cur + map.m_len;
6019 /* Count number of blocks used by this inode and update i_blocks */
6020 int ext4_ext_replay_set_iblocks(struct inode *inode)
6022 struct ext4_ext_path *path = NULL, *path2 = NULL;
6023 struct ext4_extent *ex;
6024 ext4_lblk_t cur = 0, end;
6025 int numblks = 0, i, ret = 0;
6026 ext4_fsblk_t cmp1, cmp2;
6027 struct ext4_map_blocks map;
6029 /* Determin the size of the file first */
6030 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6033 return PTR_ERR(path);
6034 ex = path[path->p_depth].p_ext;
6036 ext4_ext_drop_refs(path);
6040 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6041 ext4_ext_drop_refs(path);
6044 /* Count the number of data blocks */
6048 map.m_len = end - cur;
6049 ret = ext4_map_blocks(NULL, inode, &map, 0);
6054 cur = cur + map.m_len;
6058 * Count the number of extent tree blocks. We do it by looking up
6059 * two successive extents and determining the difference between
6060 * their paths. When path is different for 2 successive extents
6061 * we compare the blocks in the path at each level and increment
6062 * iblocks by total number of differences found.
6065 ret = skip_hole(inode, &cur);
6068 path = ext4_find_extent(inode, cur, NULL, 0);
6071 numblks += path->p_depth;
6072 ext4_ext_drop_refs(path);
6075 path = ext4_find_extent(inode, cur, NULL, 0);
6078 ex = path[path->p_depth].p_ext;
6080 ext4_ext_drop_refs(path);
6084 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6085 ext4_ext_get_actual_len(ex));
6086 ret = skip_hole(inode, &cur);
6088 ext4_ext_drop_refs(path);
6092 path2 = ext4_find_extent(inode, cur, NULL, 0);
6093 if (IS_ERR(path2)) {
6094 ext4_ext_drop_refs(path);
6098 ex = path2[path2->p_depth].p_ext;
6099 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6101 if (i <= path->p_depth)
6102 cmp1 = path[i].p_bh ?
6103 path[i].p_bh->b_blocknr : 0;
6104 if (i <= path2->p_depth)
6105 cmp2 = path2[i].p_bh ?
6106 path2[i].p_bh->b_blocknr : 0;
6107 if (cmp1 != cmp2 && cmp2 != 0)
6110 ext4_ext_drop_refs(path);
6111 ext4_ext_drop_refs(path2);
6117 inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6118 ext4_mark_inode_dirty(NULL, inode);
6122 int ext4_ext_clear_bb(struct inode *inode)
6124 struct ext4_ext_path *path = NULL;
6125 struct ext4_extent *ex;
6126 ext4_lblk_t cur = 0, end;
6128 struct ext4_map_blocks map;
6130 /* Determin the size of the file first */
6131 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6134 return PTR_ERR(path);
6135 ex = path[path->p_depth].p_ext;
6137 ext4_ext_drop_refs(path);
6141 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6142 ext4_ext_drop_refs(path);
6148 map.m_len = end - cur;
6149 ret = ext4_map_blocks(NULL, inode, &map, 0);
6153 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6154 if (!IS_ERR_OR_NULL(path)) {
6155 for (j = 0; j < path->p_depth; j++) {
6157 ext4_mb_mark_bb(inode->i_sb,
6158 path[j].p_block, 1, 0);
6159 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6160 0, path[j].p_block, 1, 1);
6162 ext4_ext_drop_refs(path);
6165 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6166 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6167 map.m_lblk, map.m_pblk, map.m_len, 1);
6169 cur = cur + map.m_len;