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/iomap.h>
31 #include <linux/sched/mm.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_rwsem. 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);
109 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
115 depth = path->p_depth;
116 for (i = 0; i <= depth; i++, path++) {
122 void ext4_free_ext_path(struct ext4_ext_path *path)
124 ext4_ext_drop_refs(path);
129 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
130 * transaction with 'restart_cred' credits. The function drops i_data_sem
131 * when restarting transaction and gets it after transaction is restarted.
133 * The function returns 0 on success, 1 if transaction had to be restarted,
134 * and < 0 in case of fatal error.
136 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
137 int check_cred, int restart_cred,
143 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
144 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
146 down_write(&EXT4_I(inode)->i_data_sem);
155 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
156 struct ext4_ext_path *path)
161 /* path points to block */
162 BUFFER_TRACE(path->p_bh, "get_write_access");
163 err = ext4_journal_get_write_access(handle, inode->i_sb,
164 path->p_bh, EXT4_JTR_NONE);
166 * The extent buffer's verified bit will be set again in
167 * __ext4_ext_dirty(). We could leave an inconsistent
168 * buffer if the extents updating procudure break off du
169 * to some error happens, force to check it again.
172 clear_buffer_verified(path->p_bh);
174 /* path points to leaf/index in inode body */
175 /* we use in-core data, no need to protect them */
185 static int __ext4_ext_dirty(const char *where, unsigned int line,
186 handle_t *handle, struct inode *inode,
187 struct ext4_ext_path *path)
191 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
193 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
194 /* path points to block */
195 err = __ext4_handle_dirty_metadata(where, line, handle,
197 /* Extents updating done, re-set verified flag */
199 set_buffer_verified(path->p_bh);
201 /* path points to leaf/index in inode body */
202 err = ext4_mark_inode_dirty(handle, inode);
207 #define ext4_ext_dirty(handle, inode, path) \
208 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
210 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
211 struct ext4_ext_path *path,
215 int depth = path->p_depth;
216 struct ext4_extent *ex;
219 * Try to predict block placement assuming that we are
220 * filling in a file which will eventually be
221 * non-sparse --- i.e., in the case of libbfd writing
222 * an ELF object sections out-of-order but in a way
223 * the eventually results in a contiguous object or
224 * executable file, or some database extending a table
225 * space file. However, this is actually somewhat
226 * non-ideal if we are writing a sparse file such as
227 * qemu or KVM writing a raw image file that is going
228 * to stay fairly sparse, since it will end up
229 * fragmenting the file system's free space. Maybe we
230 * should have some hueristics or some way to allow
231 * userspace to pass a hint to file system,
232 * especially if the latter case turns out to be
235 ex = path[depth].p_ext;
237 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
238 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
240 if (block > ext_block)
241 return ext_pblk + (block - ext_block);
243 return ext_pblk - (ext_block - block);
246 /* it looks like index is empty;
247 * try to find starting block from index itself */
248 if (path[depth].p_bh)
249 return path[depth].p_bh->b_blocknr;
252 /* OK. use inode's group */
253 return ext4_inode_to_goal_block(inode);
257 * Allocation for a meta data block
260 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
261 struct ext4_ext_path *path,
262 struct ext4_extent *ex, int *err, unsigned int flags)
264 ext4_fsblk_t goal, newblock;
266 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
267 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
272 static inline int ext4_ext_space_block(struct inode *inode, int check)
276 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
277 / sizeof(struct ext4_extent);
278 #ifdef AGGRESSIVE_TEST
279 if (!check && size > 6)
285 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
289 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
290 / sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292 if (!check && size > 5)
298 static inline int ext4_ext_space_root(struct inode *inode, int check)
302 size = sizeof(EXT4_I(inode)->i_data);
303 size -= sizeof(struct ext4_extent_header);
304 size /= sizeof(struct ext4_extent);
305 #ifdef AGGRESSIVE_TEST
306 if (!check && size > 3)
312 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
316 size = sizeof(EXT4_I(inode)->i_data);
317 size -= sizeof(struct ext4_extent_header);
318 size /= sizeof(struct ext4_extent_idx);
319 #ifdef AGGRESSIVE_TEST
320 if (!check && size > 4)
327 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
328 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
331 struct ext4_ext_path *path = *ppath;
332 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
333 int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
336 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
338 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
339 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
344 ext4_ext_max_entries(struct inode *inode, int depth)
348 if (depth == ext_depth(inode)) {
350 max = ext4_ext_space_root(inode, 1);
352 max = ext4_ext_space_root_idx(inode, 1);
355 max = ext4_ext_space_block(inode, 1);
357 max = ext4_ext_space_block_idx(inode, 1);
363 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
365 ext4_fsblk_t block = ext4_ext_pblock(ext);
366 int len = ext4_ext_get_actual_len(ext);
367 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
372 * - overflow/wrap-around
374 if (lblock + len <= lblock)
376 return ext4_inode_block_valid(inode, block, len);
379 static int ext4_valid_extent_idx(struct inode *inode,
380 struct ext4_extent_idx *ext_idx)
382 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
384 return ext4_inode_block_valid(inode, block, 1);
387 static int ext4_valid_extent_entries(struct inode *inode,
388 struct ext4_extent_header *eh,
389 ext4_lblk_t lblk, ext4_fsblk_t *pblk,
392 unsigned short entries;
393 ext4_lblk_t lblock = 0;
396 if (eh->eh_entries == 0)
399 entries = le16_to_cpu(eh->eh_entries);
403 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
406 * The logical block in the first entry should equal to
407 * the number in the index block.
409 if (depth != ext_depth(inode) &&
410 lblk != le32_to_cpu(ext->ee_block))
413 if (!ext4_valid_extent(inode, ext))
416 /* Check for overlapping extents */
417 lblock = le32_to_cpu(ext->ee_block);
419 *pblk = ext4_ext_pblock(ext);
422 cur = lblock + ext4_ext_get_actual_len(ext);
427 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
430 * The logical block in the first entry should equal to
431 * the number in the parent index block.
433 if (depth != ext_depth(inode) &&
434 lblk != le32_to_cpu(ext_idx->ei_block))
437 if (!ext4_valid_extent_idx(inode, ext_idx))
440 /* Check for overlapping index extents */
441 lblock = le32_to_cpu(ext_idx->ei_block);
443 *pblk = ext4_idx_pblock(ext_idx);
454 static int __ext4_ext_check(const char *function, unsigned int line,
455 struct inode *inode, struct ext4_extent_header *eh,
456 int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
458 const char *error_msg;
459 int max = 0, err = -EFSCORRUPTED;
461 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
462 error_msg = "invalid magic";
465 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
466 error_msg = "unexpected eh_depth";
469 if (unlikely(eh->eh_max == 0)) {
470 error_msg = "invalid eh_max";
473 max = ext4_ext_max_entries(inode, depth);
474 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
475 error_msg = "too large eh_max";
478 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
479 error_msg = "invalid eh_entries";
482 if (unlikely((eh->eh_entries == 0) && (depth > 0))) {
483 error_msg = "eh_entries is 0 but eh_depth is > 0";
486 if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
487 error_msg = "invalid extent entries";
490 if (unlikely(depth > 32)) {
491 error_msg = "too large eh_depth";
494 /* Verify checksum on non-root extent tree nodes */
495 if (ext_depth(inode) != depth &&
496 !ext4_extent_block_csum_verify(inode, eh)) {
497 error_msg = "extent tree corrupted";
504 ext4_error_inode_err(inode, function, line, 0, -err,
505 "pblk %llu bad header/extent: %s - magic %x, "
506 "entries %u, max %u(%u), depth %u(%u)",
507 (unsigned long long) pblk, error_msg,
508 le16_to_cpu(eh->eh_magic),
509 le16_to_cpu(eh->eh_entries),
510 le16_to_cpu(eh->eh_max),
511 max, le16_to_cpu(eh->eh_depth), depth);
515 #define ext4_ext_check(inode, eh, depth, pblk) \
516 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
518 int ext4_ext_check_inode(struct inode *inode)
520 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
523 static void ext4_cache_extents(struct inode *inode,
524 struct ext4_extent_header *eh)
526 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
527 ext4_lblk_t prev = 0;
530 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
531 unsigned int status = EXTENT_STATUS_WRITTEN;
532 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
533 int len = ext4_ext_get_actual_len(ex);
535 if (prev && (prev != lblk))
536 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
539 if (ext4_ext_is_unwritten(ex))
540 status = EXTENT_STATUS_UNWRITTEN;
541 ext4_es_cache_extent(inode, lblk, len,
542 ext4_ext_pblock(ex), status);
547 static struct buffer_head *
548 __read_extent_tree_block(const char *function, unsigned int line,
549 struct inode *inode, struct ext4_extent_idx *idx,
550 int depth, int flags)
552 struct buffer_head *bh;
554 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
557 if (flags & EXT4_EX_NOFAIL)
558 gfp_flags |= __GFP_NOFAIL;
560 pblk = ext4_idx_pblock(idx);
561 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
563 return ERR_PTR(-ENOMEM);
565 if (!bh_uptodate_or_lock(bh)) {
566 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
567 err = ext4_read_bh(bh, 0, NULL);
571 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
573 err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
574 depth, pblk, le32_to_cpu(idx->ei_block));
577 set_buffer_verified(bh);
579 * If this is a leaf block, cache all of its entries
581 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
582 struct ext4_extent_header *eh = ext_block_hdr(bh);
583 ext4_cache_extents(inode, eh);
592 #define read_extent_tree_block(inode, idx, depth, flags) \
593 __read_extent_tree_block(__func__, __LINE__, (inode), (idx), \
597 * This function is called to cache a file's extent information in the
600 int ext4_ext_precache(struct inode *inode)
602 struct ext4_inode_info *ei = EXT4_I(inode);
603 struct ext4_ext_path *path = NULL;
604 struct buffer_head *bh;
605 int i = 0, depth, ret = 0;
607 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
608 return 0; /* not an extent-mapped inode */
610 down_read(&ei->i_data_sem);
611 depth = ext_depth(inode);
613 /* Don't cache anything if there are no external extent blocks */
615 up_read(&ei->i_data_sem);
619 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
622 up_read(&ei->i_data_sem);
626 path[0].p_hdr = ext_inode_hdr(inode);
627 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
630 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
633 * If this is a leaf block or we've reached the end of
634 * the index block, go up
637 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
638 brelse(path[i].p_bh);
643 bh = read_extent_tree_block(inode, path[i].p_idx++,
645 EXT4_EX_FORCE_CACHE);
652 path[i].p_hdr = ext_block_hdr(bh);
653 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
655 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
657 up_read(&ei->i_data_sem);
658 ext4_free_ext_path(path);
663 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
665 int k, l = path->p_depth;
667 ext_debug(inode, "path:");
668 for (k = 0; k <= l; k++, path++) {
670 ext_debug(inode, " %d->%llu",
671 le32_to_cpu(path->p_idx->ei_block),
672 ext4_idx_pblock(path->p_idx));
673 } else if (path->p_ext) {
674 ext_debug(inode, " %d:[%d]%d:%llu ",
675 le32_to_cpu(path->p_ext->ee_block),
676 ext4_ext_is_unwritten(path->p_ext),
677 ext4_ext_get_actual_len(path->p_ext),
678 ext4_ext_pblock(path->p_ext));
680 ext_debug(inode, " []");
682 ext_debug(inode, "\n");
685 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
687 int depth = ext_depth(inode);
688 struct ext4_extent_header *eh;
689 struct ext4_extent *ex;
695 eh = path[depth].p_hdr;
696 ex = EXT_FIRST_EXTENT(eh);
698 ext_debug(inode, "Displaying leaf extents\n");
700 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
701 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
702 ext4_ext_is_unwritten(ex),
703 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
705 ext_debug(inode, "\n");
708 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
709 ext4_fsblk_t newblock, int level)
711 int depth = ext_depth(inode);
712 struct ext4_extent *ex;
714 if (depth != level) {
715 struct ext4_extent_idx *idx;
716 idx = path[level].p_idx;
717 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
718 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
719 level, le32_to_cpu(idx->ei_block),
720 ext4_idx_pblock(idx), newblock);
727 ex = path[depth].p_ext;
728 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
729 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
730 le32_to_cpu(ex->ee_block),
732 ext4_ext_is_unwritten(ex),
733 ext4_ext_get_actual_len(ex),
740 #define ext4_ext_show_path(inode, path)
741 #define ext4_ext_show_leaf(inode, path)
742 #define ext4_ext_show_move(inode, path, newblock, level)
746 * ext4_ext_binsearch_idx:
747 * binary search for the closest index of the given block
748 * the header must be checked before calling this
751 ext4_ext_binsearch_idx(struct inode *inode,
752 struct ext4_ext_path *path, ext4_lblk_t block)
754 struct ext4_extent_header *eh = path->p_hdr;
755 struct ext4_extent_idx *r, *l, *m;
758 ext_debug(inode, "binsearch for %u(idx): ", block);
760 l = EXT_FIRST_INDEX(eh) + 1;
761 r = EXT_LAST_INDEX(eh);
764 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
765 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
766 r, le32_to_cpu(r->ei_block));
768 if (block < le32_to_cpu(m->ei_block))
775 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
776 ext4_idx_pblock(path->p_idx));
778 #ifdef CHECK_BINSEARCH
780 struct ext4_extent_idx *chix, *ix;
783 chix = ix = EXT_FIRST_INDEX(eh);
784 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
785 if (k != 0 && le32_to_cpu(ix->ei_block) <=
786 le32_to_cpu(ix[-1].ei_block)) {
787 printk(KERN_DEBUG "k=%d, ix=0x%p, "
789 ix, EXT_FIRST_INDEX(eh));
790 printk(KERN_DEBUG "%u <= %u\n",
791 le32_to_cpu(ix->ei_block),
792 le32_to_cpu(ix[-1].ei_block));
794 BUG_ON(k && le32_to_cpu(ix->ei_block)
795 <= le32_to_cpu(ix[-1].ei_block));
796 if (block < le32_to_cpu(ix->ei_block))
800 BUG_ON(chix != path->p_idx);
807 * ext4_ext_binsearch:
808 * binary search for closest extent of the given block
809 * the header must be checked before calling this
812 ext4_ext_binsearch(struct inode *inode,
813 struct ext4_ext_path *path, ext4_lblk_t block)
815 struct ext4_extent_header *eh = path->p_hdr;
816 struct ext4_extent *r, *l, *m;
818 if (eh->eh_entries == 0) {
820 * this leaf is empty:
821 * we get such a leaf in split/add case
826 ext_debug(inode, "binsearch for %u: ", block);
828 l = EXT_FIRST_EXTENT(eh) + 1;
829 r = EXT_LAST_EXTENT(eh);
833 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
834 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
835 r, le32_to_cpu(r->ee_block));
837 if (block < le32_to_cpu(m->ee_block))
844 ext_debug(inode, " -> %d:%llu:[%d]%d ",
845 le32_to_cpu(path->p_ext->ee_block),
846 ext4_ext_pblock(path->p_ext),
847 ext4_ext_is_unwritten(path->p_ext),
848 ext4_ext_get_actual_len(path->p_ext));
850 #ifdef CHECK_BINSEARCH
852 struct ext4_extent *chex, *ex;
855 chex = ex = EXT_FIRST_EXTENT(eh);
856 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
857 BUG_ON(k && le32_to_cpu(ex->ee_block)
858 <= le32_to_cpu(ex[-1].ee_block));
859 if (block < le32_to_cpu(ex->ee_block))
863 BUG_ON(chex != path->p_ext);
869 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
871 struct ext4_extent_header *eh;
873 eh = ext_inode_hdr(inode);
876 eh->eh_magic = EXT4_EXT_MAGIC;
877 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
878 eh->eh_generation = 0;
879 ext4_mark_inode_dirty(handle, inode);
882 struct ext4_ext_path *
883 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
884 struct ext4_ext_path **orig_path, int flags)
886 struct ext4_extent_header *eh;
887 struct buffer_head *bh;
888 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
889 short int depth, i, ppos = 0;
891 gfp_t gfp_flags = GFP_NOFS;
893 if (flags & EXT4_EX_NOFAIL)
894 gfp_flags |= __GFP_NOFAIL;
896 eh = ext_inode_hdr(inode);
897 depth = ext_depth(inode);
898 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
899 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
906 ext4_ext_drop_refs(path);
907 if (depth > path[0].p_maxdepth) {
909 *orig_path = path = NULL;
913 /* account possible depth increase */
914 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
917 return ERR_PTR(-ENOMEM);
918 path[0].p_maxdepth = depth + 1;
924 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
925 ext4_cache_extents(inode, eh);
926 /* walk through the tree */
928 ext_debug(inode, "depth %d: num %d, max %d\n",
929 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
931 ext4_ext_binsearch_idx(inode, path + ppos, block);
932 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
933 path[ppos].p_depth = i;
934 path[ppos].p_ext = NULL;
936 bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
942 eh = ext_block_hdr(bh);
944 path[ppos].p_bh = bh;
945 path[ppos].p_hdr = eh;
948 path[ppos].p_depth = i;
949 path[ppos].p_ext = NULL;
950 path[ppos].p_idx = NULL;
953 ext4_ext_binsearch(inode, path + ppos, block);
954 /* if not an empty leaf */
955 if (path[ppos].p_ext)
956 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
958 ext4_ext_show_path(inode, path);
963 ext4_free_ext_path(path);
970 * ext4_ext_insert_index:
971 * insert new index [@logical;@ptr] into the block at @curp;
972 * check where to insert: before @curp or after @curp
974 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
975 struct ext4_ext_path *curp,
976 int logical, ext4_fsblk_t ptr)
978 struct ext4_extent_idx *ix;
981 err = ext4_ext_get_access(handle, inode, curp);
985 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
986 EXT4_ERROR_INODE(inode,
987 "logical %d == ei_block %d!",
988 logical, le32_to_cpu(curp->p_idx->ei_block));
989 return -EFSCORRUPTED;
992 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
993 >= le16_to_cpu(curp->p_hdr->eh_max))) {
994 EXT4_ERROR_INODE(inode,
995 "eh_entries %d >= eh_max %d!",
996 le16_to_cpu(curp->p_hdr->eh_entries),
997 le16_to_cpu(curp->p_hdr->eh_max));
998 return -EFSCORRUPTED;
1001 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
1003 ext_debug(inode, "insert new index %d after: %llu\n",
1005 ix = curp->p_idx + 1;
1008 ext_debug(inode, "insert new index %d before: %llu\n",
1013 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1014 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1015 return -EFSCORRUPTED;
1018 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1021 ext_debug(inode, "insert new index %d: "
1022 "move %d indices from 0x%p to 0x%p\n",
1023 logical, len, ix, ix + 1);
1024 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1027 ix->ei_block = cpu_to_le32(logical);
1028 ext4_idx_store_pblock(ix, ptr);
1029 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1031 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1032 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1033 return -EFSCORRUPTED;
1036 err = ext4_ext_dirty(handle, inode, curp);
1037 ext4_std_error(inode->i_sb, err);
1044 * inserts new subtree into the path, using free index entry
1046 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1047 * - makes decision where to split
1048 * - moves remaining extents and index entries (right to the split point)
1049 * into the newly allocated blocks
1050 * - initializes subtree
1052 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1054 struct ext4_ext_path *path,
1055 struct ext4_extent *newext, int at)
1057 struct buffer_head *bh = NULL;
1058 int depth = ext_depth(inode);
1059 struct ext4_extent_header *neh;
1060 struct ext4_extent_idx *fidx;
1061 int i = at, k, m, a;
1062 ext4_fsblk_t newblock, oldblock;
1064 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1065 gfp_t gfp_flags = GFP_NOFS;
1067 size_t ext_size = 0;
1069 if (flags & EXT4_EX_NOFAIL)
1070 gfp_flags |= __GFP_NOFAIL;
1072 /* make decision: where to split? */
1073 /* FIXME: now decision is simplest: at current extent */
1075 /* if current leaf will be split, then we should use
1076 * border from split point */
1077 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1078 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1079 return -EFSCORRUPTED;
1081 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1082 border = path[depth].p_ext[1].ee_block;
1083 ext_debug(inode, "leaf will be split."
1084 " next leaf starts at %d\n",
1085 le32_to_cpu(border));
1087 border = newext->ee_block;
1088 ext_debug(inode, "leaf will be added."
1089 " next leaf starts at %d\n",
1090 le32_to_cpu(border));
1094 * If error occurs, then we break processing
1095 * and mark filesystem read-only. index won't
1096 * be inserted and tree will be in consistent
1097 * state. Next mount will repair buffers too.
1101 * Get array to track all allocated blocks.
1102 * We need this to handle errors and free blocks
1105 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1109 /* allocate all needed blocks */
1110 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1111 for (a = 0; a < depth - at; a++) {
1112 newblock = ext4_ext_new_meta_block(handle, inode, path,
1113 newext, &err, flags);
1116 ablocks[a] = newblock;
1119 /* initialize new leaf */
1120 newblock = ablocks[--a];
1121 if (unlikely(newblock == 0)) {
1122 EXT4_ERROR_INODE(inode, "newblock == 0!");
1123 err = -EFSCORRUPTED;
1126 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1127 if (unlikely(!bh)) {
1133 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1138 neh = ext_block_hdr(bh);
1139 neh->eh_entries = 0;
1140 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1141 neh->eh_magic = EXT4_EXT_MAGIC;
1143 neh->eh_generation = 0;
1145 /* move remainder of path[depth] to the new leaf */
1146 if (unlikely(path[depth].p_hdr->eh_entries !=
1147 path[depth].p_hdr->eh_max)) {
1148 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1149 path[depth].p_hdr->eh_entries,
1150 path[depth].p_hdr->eh_max);
1151 err = -EFSCORRUPTED;
1154 /* start copy from next extent */
1155 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1156 ext4_ext_show_move(inode, path, newblock, depth);
1158 struct ext4_extent *ex;
1159 ex = EXT_FIRST_EXTENT(neh);
1160 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1161 le16_add_cpu(&neh->eh_entries, m);
1164 /* zero out unused area in the extent block */
1165 ext_size = sizeof(struct ext4_extent_header) +
1166 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1167 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1168 ext4_extent_block_csum_set(inode, neh);
1169 set_buffer_uptodate(bh);
1172 err = ext4_handle_dirty_metadata(handle, inode, bh);
1178 /* correct old leaf */
1180 err = ext4_ext_get_access(handle, inode, path + depth);
1183 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1184 err = ext4_ext_dirty(handle, inode, path + depth);
1190 /* create intermediate indexes */
1192 if (unlikely(k < 0)) {
1193 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1194 err = -EFSCORRUPTED;
1198 ext_debug(inode, "create %d intermediate indices\n", k);
1199 /* insert new index into current index block */
1200 /* current depth stored in i var */
1203 oldblock = newblock;
1204 newblock = ablocks[--a];
1205 bh = sb_getblk(inode->i_sb, newblock);
1206 if (unlikely(!bh)) {
1212 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1217 neh = ext_block_hdr(bh);
1218 neh->eh_entries = cpu_to_le16(1);
1219 neh->eh_magic = EXT4_EXT_MAGIC;
1220 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1221 neh->eh_depth = cpu_to_le16(depth - i);
1222 neh->eh_generation = 0;
1223 fidx = EXT_FIRST_INDEX(neh);
1224 fidx->ei_block = border;
1225 ext4_idx_store_pblock(fidx, oldblock);
1227 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1228 i, newblock, le32_to_cpu(border), oldblock);
1230 /* move remainder of path[i] to the new index block */
1231 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1232 EXT_LAST_INDEX(path[i].p_hdr))) {
1233 EXT4_ERROR_INODE(inode,
1234 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1235 le32_to_cpu(path[i].p_ext->ee_block));
1236 err = -EFSCORRUPTED;
1239 /* start copy indexes */
1240 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1241 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1242 EXT_MAX_INDEX(path[i].p_hdr));
1243 ext4_ext_show_move(inode, path, newblock, i);
1245 memmove(++fidx, path[i].p_idx,
1246 sizeof(struct ext4_extent_idx) * m);
1247 le16_add_cpu(&neh->eh_entries, m);
1249 /* zero out unused area in the extent block */
1250 ext_size = sizeof(struct ext4_extent_header) +
1251 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1252 memset(bh->b_data + ext_size, 0,
1253 inode->i_sb->s_blocksize - ext_size);
1254 ext4_extent_block_csum_set(inode, neh);
1255 set_buffer_uptodate(bh);
1258 err = ext4_handle_dirty_metadata(handle, inode, bh);
1264 /* correct old index */
1266 err = ext4_ext_get_access(handle, inode, path + i);
1269 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1270 err = ext4_ext_dirty(handle, inode, path + i);
1278 /* insert new index */
1279 err = ext4_ext_insert_index(handle, inode, path + at,
1280 le32_to_cpu(border), newblock);
1284 if (buffer_locked(bh))
1290 /* free all allocated blocks in error case */
1291 for (i = 0; i < depth; i++) {
1294 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1295 EXT4_FREE_BLOCKS_METADATA);
1304 * ext4_ext_grow_indepth:
1305 * implements tree growing procedure:
1306 * - allocates new block
1307 * - moves top-level data (index block or leaf) into the new block
1308 * - initializes new top-level, creating index that points to the
1309 * just created block
1311 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1314 struct ext4_extent_header *neh;
1315 struct buffer_head *bh;
1316 ext4_fsblk_t newblock, goal = 0;
1317 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1319 size_t ext_size = 0;
1321 /* Try to prepend new index to old one */
1322 if (ext_depth(inode))
1323 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1324 if (goal > le32_to_cpu(es->s_first_data_block)) {
1325 flags |= EXT4_MB_HINT_TRY_GOAL;
1328 goal = ext4_inode_to_goal_block(inode);
1329 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1334 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1339 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1346 ext_size = sizeof(EXT4_I(inode)->i_data);
1347 /* move top-level index/leaf into new block */
1348 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1349 /* zero out unused area in the extent block */
1350 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1352 /* set size of new block */
1353 neh = ext_block_hdr(bh);
1354 /* old root could have indexes or leaves
1355 * so calculate e_max right way */
1356 if (ext_depth(inode))
1357 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1359 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1360 neh->eh_magic = EXT4_EXT_MAGIC;
1361 ext4_extent_block_csum_set(inode, neh);
1362 set_buffer_uptodate(bh);
1363 set_buffer_verified(bh);
1366 err = ext4_handle_dirty_metadata(handle, inode, bh);
1370 /* Update top-level index: num,max,pointer */
1371 neh = ext_inode_hdr(inode);
1372 neh->eh_entries = cpu_to_le16(1);
1373 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1374 if (neh->eh_depth == 0) {
1375 /* Root extent block becomes index block */
1376 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1377 EXT_FIRST_INDEX(neh)->ei_block =
1378 EXT_FIRST_EXTENT(neh)->ee_block;
1380 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1381 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1382 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1383 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1385 le16_add_cpu(&neh->eh_depth, 1);
1386 err = ext4_mark_inode_dirty(handle, inode);
1394 * ext4_ext_create_new_leaf:
1395 * finds empty index and adds new leaf.
1396 * if no free index is found, then it requests in-depth growing.
1398 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1399 unsigned int mb_flags,
1400 unsigned int gb_flags,
1401 struct ext4_ext_path **ppath,
1402 struct ext4_extent *newext)
1404 struct ext4_ext_path *path = *ppath;
1405 struct ext4_ext_path *curp;
1406 int depth, i, err = 0;
1409 i = depth = ext_depth(inode);
1411 /* walk up to the tree and look for free index entry */
1412 curp = path + depth;
1413 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1418 /* we use already allocated block for index block,
1419 * so subsequent data blocks should be contiguous */
1420 if (EXT_HAS_FREE_INDEX(curp)) {
1421 /* if we found index with free entry, then use that
1422 * entry: create all needed subtree and add new leaf */
1423 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1428 path = ext4_find_extent(inode,
1429 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1432 err = PTR_ERR(path);
1434 /* tree is full, time to grow in depth */
1435 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1440 path = ext4_find_extent(inode,
1441 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1444 err = PTR_ERR(path);
1449 * only first (depth 0 -> 1) produces free space;
1450 * in all other cases we have to split the grown tree
1452 depth = ext_depth(inode);
1453 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1454 /* now we need to split */
1464 * search the closest allocated block to the left for *logical
1465 * and returns it at @logical + it's physical address at @phys
1466 * if *logical is the smallest allocated block, the function
1467 * returns 0 at @phys
1468 * return value contains 0 (success) or error code
1470 static int ext4_ext_search_left(struct inode *inode,
1471 struct ext4_ext_path *path,
1472 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1474 struct ext4_extent_idx *ix;
1475 struct ext4_extent *ex;
1478 if (unlikely(path == NULL)) {
1479 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1480 return -EFSCORRUPTED;
1482 depth = path->p_depth;
1485 if (depth == 0 && path->p_ext == NULL)
1488 /* usually extent in the path covers blocks smaller
1489 * then *logical, but it can be that extent is the
1490 * first one in the file */
1492 ex = path[depth].p_ext;
1493 ee_len = ext4_ext_get_actual_len(ex);
1494 if (*logical < le32_to_cpu(ex->ee_block)) {
1495 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1496 EXT4_ERROR_INODE(inode,
1497 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1498 *logical, le32_to_cpu(ex->ee_block));
1499 return -EFSCORRUPTED;
1501 while (--depth >= 0) {
1502 ix = path[depth].p_idx;
1503 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1504 EXT4_ERROR_INODE(inode,
1505 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1506 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1507 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block),
1509 return -EFSCORRUPTED;
1515 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1516 EXT4_ERROR_INODE(inode,
1517 "logical %d < ee_block %d + ee_len %d!",
1518 *logical, le32_to_cpu(ex->ee_block), ee_len);
1519 return -EFSCORRUPTED;
1522 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1523 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1528 * Search the closest allocated block to the right for *logical
1529 * and returns it at @logical + it's physical address at @phys.
1530 * If not exists, return 0 and @phys is set to 0. We will return
1531 * 1 which means we found an allocated block and ret_ex is valid.
1532 * Or return a (< 0) error code.
1534 static int ext4_ext_search_right(struct inode *inode,
1535 struct ext4_ext_path *path,
1536 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1537 struct ext4_extent *ret_ex)
1539 struct buffer_head *bh = NULL;
1540 struct ext4_extent_header *eh;
1541 struct ext4_extent_idx *ix;
1542 struct ext4_extent *ex;
1543 int depth; /* Note, NOT eh_depth; depth from top of tree */
1546 if (unlikely(path == NULL)) {
1547 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1548 return -EFSCORRUPTED;
1550 depth = path->p_depth;
1553 if (depth == 0 && path->p_ext == NULL)
1556 /* usually extent in the path covers blocks smaller
1557 * then *logical, but it can be that extent is the
1558 * first one in the file */
1560 ex = path[depth].p_ext;
1561 ee_len = ext4_ext_get_actual_len(ex);
1562 if (*logical < le32_to_cpu(ex->ee_block)) {
1563 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1564 EXT4_ERROR_INODE(inode,
1565 "first_extent(path[%d].p_hdr) != ex",
1567 return -EFSCORRUPTED;
1569 while (--depth >= 0) {
1570 ix = path[depth].p_idx;
1571 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1572 EXT4_ERROR_INODE(inode,
1573 "ix != EXT_FIRST_INDEX *logical %d!",
1575 return -EFSCORRUPTED;
1581 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1582 EXT4_ERROR_INODE(inode,
1583 "logical %d < ee_block %d + ee_len %d!",
1584 *logical, le32_to_cpu(ex->ee_block), ee_len);
1585 return -EFSCORRUPTED;
1588 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1589 /* next allocated block in this leaf */
1594 /* go up and search for index to the right */
1595 while (--depth >= 0) {
1596 ix = path[depth].p_idx;
1597 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1601 /* we've gone up to the root and found no index to the right */
1605 /* we've found index to the right, let's
1606 * follow it and find the closest allocated
1607 * block to the right */
1609 while (++depth < path->p_depth) {
1610 /* subtract from p_depth to get proper eh_depth */
1611 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1614 eh = ext_block_hdr(bh);
1615 ix = EXT_FIRST_INDEX(eh);
1619 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1622 eh = ext_block_hdr(bh);
1623 ex = EXT_FIRST_EXTENT(eh);
1625 *logical = le32_to_cpu(ex->ee_block);
1626 *phys = ext4_ext_pblock(ex);
1635 * ext4_ext_next_allocated_block:
1636 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1637 * NOTE: it considers block number from index entry as
1638 * allocated block. Thus, index entries have to be consistent
1642 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1646 BUG_ON(path == NULL);
1647 depth = path->p_depth;
1649 if (depth == 0 && path->p_ext == NULL)
1650 return EXT_MAX_BLOCKS;
1652 while (depth >= 0) {
1653 struct ext4_ext_path *p = &path[depth];
1655 if (depth == path->p_depth) {
1657 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1658 return le32_to_cpu(p->p_ext[1].ee_block);
1661 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1662 return le32_to_cpu(p->p_idx[1].ei_block);
1667 return EXT_MAX_BLOCKS;
1671 * ext4_ext_next_leaf_block:
1672 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1674 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1678 BUG_ON(path == NULL);
1679 depth = path->p_depth;
1681 /* zero-tree has no leaf blocks at all */
1683 return EXT_MAX_BLOCKS;
1685 /* go to index block */
1688 while (depth >= 0) {
1689 if (path[depth].p_idx !=
1690 EXT_LAST_INDEX(path[depth].p_hdr))
1691 return (ext4_lblk_t)
1692 le32_to_cpu(path[depth].p_idx[1].ei_block);
1696 return EXT_MAX_BLOCKS;
1700 * ext4_ext_correct_indexes:
1701 * if leaf gets modified and modified extent is first in the leaf,
1702 * then we have to correct all indexes above.
1703 * TODO: do we need to correct tree in all cases?
1705 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1706 struct ext4_ext_path *path)
1708 struct ext4_extent_header *eh;
1709 int depth = ext_depth(inode);
1710 struct ext4_extent *ex;
1714 eh = path[depth].p_hdr;
1715 ex = path[depth].p_ext;
1717 if (unlikely(ex == NULL || eh == NULL)) {
1718 EXT4_ERROR_INODE(inode,
1719 "ex %p == NULL or eh %p == NULL", ex, eh);
1720 return -EFSCORRUPTED;
1724 /* there is no tree at all */
1728 if (ex != EXT_FIRST_EXTENT(eh)) {
1729 /* we correct tree if first leaf got modified only */
1734 * TODO: we need correction if border is smaller than current one
1737 border = path[depth].p_ext->ee_block;
1738 err = ext4_ext_get_access(handle, inode, path + k);
1741 path[k].p_idx->ei_block = border;
1742 err = ext4_ext_dirty(handle, inode, path + k);
1747 /* change all left-side indexes */
1748 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1750 err = ext4_ext_get_access(handle, inode, path + k);
1753 path[k].p_idx->ei_block = border;
1754 err = ext4_ext_dirty(handle, inode, path + k);
1762 static int ext4_can_extents_be_merged(struct inode *inode,
1763 struct ext4_extent *ex1,
1764 struct ext4_extent *ex2)
1766 unsigned short ext1_ee_len, ext2_ee_len;
1768 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1771 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1772 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1774 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1775 le32_to_cpu(ex2->ee_block))
1778 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1781 if (ext4_ext_is_unwritten(ex1) &&
1782 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1784 #ifdef AGGRESSIVE_TEST
1785 if (ext1_ee_len >= 4)
1789 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1795 * This function tries to merge the "ex" extent to the next extent in the tree.
1796 * It always tries to merge towards right. If you want to merge towards
1797 * left, pass "ex - 1" as argument instead of "ex".
1798 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1799 * 1 if they got merged.
1801 static int ext4_ext_try_to_merge_right(struct inode *inode,
1802 struct ext4_ext_path *path,
1803 struct ext4_extent *ex)
1805 struct ext4_extent_header *eh;
1806 unsigned int depth, len;
1807 int merge_done = 0, unwritten;
1809 depth = ext_depth(inode);
1810 BUG_ON(path[depth].p_hdr == NULL);
1811 eh = path[depth].p_hdr;
1813 while (ex < EXT_LAST_EXTENT(eh)) {
1814 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1816 /* merge with next extent! */
1817 unwritten = ext4_ext_is_unwritten(ex);
1818 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1819 + ext4_ext_get_actual_len(ex + 1));
1821 ext4_ext_mark_unwritten(ex);
1823 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1824 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1825 * sizeof(struct ext4_extent);
1826 memmove(ex + 1, ex + 2, len);
1828 le16_add_cpu(&eh->eh_entries, -1);
1830 WARN_ON(eh->eh_entries == 0);
1831 if (!eh->eh_entries)
1832 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1839 * This function does a very simple check to see if we can collapse
1840 * an extent tree with a single extent tree leaf block into the inode.
1842 static void ext4_ext_try_to_merge_up(handle_t *handle,
1843 struct inode *inode,
1844 struct ext4_ext_path *path)
1847 unsigned max_root = ext4_ext_space_root(inode, 0);
1850 if ((path[0].p_depth != 1) ||
1851 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1852 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1856 * We need to modify the block allocation bitmap and the block
1857 * group descriptor to release the extent tree block. If we
1858 * can't get the journal credits, give up.
1860 if (ext4_journal_extend(handle, 2,
1861 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1865 * Copy the extent data up to the inode
1867 blk = ext4_idx_pblock(path[0].p_idx);
1868 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1869 sizeof(struct ext4_extent_idx);
1870 s += sizeof(struct ext4_extent_header);
1872 path[1].p_maxdepth = path[0].p_maxdepth;
1873 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1874 path[0].p_depth = 0;
1875 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1876 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1877 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1879 brelse(path[1].p_bh);
1880 ext4_free_blocks(handle, inode, NULL, blk, 1,
1881 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1885 * This function tries to merge the @ex extent to neighbours in the tree, then
1886 * tries to collapse the extent tree into the inode.
1888 static void ext4_ext_try_to_merge(handle_t *handle,
1889 struct inode *inode,
1890 struct ext4_ext_path *path,
1891 struct ext4_extent *ex)
1893 struct ext4_extent_header *eh;
1897 depth = ext_depth(inode);
1898 BUG_ON(path[depth].p_hdr == NULL);
1899 eh = path[depth].p_hdr;
1901 if (ex > EXT_FIRST_EXTENT(eh))
1902 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1905 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1907 ext4_ext_try_to_merge_up(handle, inode, path);
1911 * check if a portion of the "newext" extent overlaps with an
1914 * If there is an overlap discovered, it updates the length of the newext
1915 * such that there will be no overlap, and then returns 1.
1916 * If there is no overlap found, it returns 0.
1918 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1919 struct inode *inode,
1920 struct ext4_extent *newext,
1921 struct ext4_ext_path *path)
1924 unsigned int depth, len1;
1925 unsigned int ret = 0;
1927 b1 = le32_to_cpu(newext->ee_block);
1928 len1 = ext4_ext_get_actual_len(newext);
1929 depth = ext_depth(inode);
1930 if (!path[depth].p_ext)
1932 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1935 * get the next allocated block if the extent in the path
1936 * is before the requested block(s)
1939 b2 = ext4_ext_next_allocated_block(path);
1940 if (b2 == EXT_MAX_BLOCKS)
1942 b2 = EXT4_LBLK_CMASK(sbi, b2);
1945 /* check for wrap through zero on extent logical start block*/
1946 if (b1 + len1 < b1) {
1947 len1 = EXT_MAX_BLOCKS - b1;
1948 newext->ee_len = cpu_to_le16(len1);
1952 /* check for overlap */
1953 if (b1 + len1 > b2) {
1954 newext->ee_len = cpu_to_le16(b2 - b1);
1962 * ext4_ext_insert_extent:
1963 * tries to merge requested extent into the existing extent or
1964 * inserts requested extent as new one into the tree,
1965 * creating new leaf in the no-space case.
1967 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1968 struct ext4_ext_path **ppath,
1969 struct ext4_extent *newext, int gb_flags)
1971 struct ext4_ext_path *path = *ppath;
1972 struct ext4_extent_header *eh;
1973 struct ext4_extent *ex, *fex;
1974 struct ext4_extent *nearex; /* nearest extent */
1975 struct ext4_ext_path *npath = NULL;
1976 int depth, len, err;
1978 int mb_flags = 0, unwritten;
1980 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1981 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1982 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1983 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1984 return -EFSCORRUPTED;
1986 depth = ext_depth(inode);
1987 ex = path[depth].p_ext;
1988 eh = path[depth].p_hdr;
1989 if (unlikely(path[depth].p_hdr == NULL)) {
1990 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1991 return -EFSCORRUPTED;
1994 /* try to insert block into found extent and return */
1995 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1998 * Try to see whether we should rather test the extent on
1999 * right from ex, or from the left of ex. This is because
2000 * ext4_find_extent() can return either extent on the
2001 * left, or on the right from the searched position. This
2002 * will make merging more effective.
2004 if (ex < EXT_LAST_EXTENT(eh) &&
2005 (le32_to_cpu(ex->ee_block) +
2006 ext4_ext_get_actual_len(ex) <
2007 le32_to_cpu(newext->ee_block))) {
2010 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2011 (le32_to_cpu(newext->ee_block) +
2012 ext4_ext_get_actual_len(newext) <
2013 le32_to_cpu(ex->ee_block)))
2016 /* Try to append newex to the ex */
2017 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2018 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2020 ext4_ext_is_unwritten(newext),
2021 ext4_ext_get_actual_len(newext),
2022 le32_to_cpu(ex->ee_block),
2023 ext4_ext_is_unwritten(ex),
2024 ext4_ext_get_actual_len(ex),
2025 ext4_ext_pblock(ex));
2026 err = ext4_ext_get_access(handle, inode,
2030 unwritten = ext4_ext_is_unwritten(ex);
2031 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2032 + ext4_ext_get_actual_len(newext));
2034 ext4_ext_mark_unwritten(ex);
2040 /* Try to prepend newex to the ex */
2041 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2042 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2044 le32_to_cpu(newext->ee_block),
2045 ext4_ext_is_unwritten(newext),
2046 ext4_ext_get_actual_len(newext),
2047 le32_to_cpu(ex->ee_block),
2048 ext4_ext_is_unwritten(ex),
2049 ext4_ext_get_actual_len(ex),
2050 ext4_ext_pblock(ex));
2051 err = ext4_ext_get_access(handle, inode,
2056 unwritten = ext4_ext_is_unwritten(ex);
2057 ex->ee_block = newext->ee_block;
2058 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2059 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2060 + ext4_ext_get_actual_len(newext));
2062 ext4_ext_mark_unwritten(ex);
2068 depth = ext_depth(inode);
2069 eh = path[depth].p_hdr;
2070 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2073 /* probably next leaf has space for us? */
2074 fex = EXT_LAST_EXTENT(eh);
2075 next = EXT_MAX_BLOCKS;
2076 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2077 next = ext4_ext_next_leaf_block(path);
2078 if (next != EXT_MAX_BLOCKS) {
2079 ext_debug(inode, "next leaf block - %u\n", next);
2080 BUG_ON(npath != NULL);
2081 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2083 return PTR_ERR(npath);
2084 BUG_ON(npath->p_depth != path->p_depth);
2085 eh = npath[depth].p_hdr;
2086 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2087 ext_debug(inode, "next leaf isn't full(%d)\n",
2088 le16_to_cpu(eh->eh_entries));
2092 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2093 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2097 * There is no free space in the found leaf.
2098 * We're gonna add a new leaf in the tree.
2100 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2101 mb_flags |= EXT4_MB_USE_RESERVED;
2102 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2106 depth = ext_depth(inode);
2107 eh = path[depth].p_hdr;
2110 nearex = path[depth].p_ext;
2112 err = ext4_ext_get_access(handle, inode, path + depth);
2117 /* there is no extent in this leaf, create first one */
2118 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2119 le32_to_cpu(newext->ee_block),
2120 ext4_ext_pblock(newext),
2121 ext4_ext_is_unwritten(newext),
2122 ext4_ext_get_actual_len(newext));
2123 nearex = EXT_FIRST_EXTENT(eh);
2125 if (le32_to_cpu(newext->ee_block)
2126 > le32_to_cpu(nearex->ee_block)) {
2128 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2130 le32_to_cpu(newext->ee_block),
2131 ext4_ext_pblock(newext),
2132 ext4_ext_is_unwritten(newext),
2133 ext4_ext_get_actual_len(newext),
2138 BUG_ON(newext->ee_block == nearex->ee_block);
2139 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2141 le32_to_cpu(newext->ee_block),
2142 ext4_ext_pblock(newext),
2143 ext4_ext_is_unwritten(newext),
2144 ext4_ext_get_actual_len(newext),
2147 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2149 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2150 "move %d extents from 0x%p to 0x%p\n",
2151 le32_to_cpu(newext->ee_block),
2152 ext4_ext_pblock(newext),
2153 ext4_ext_is_unwritten(newext),
2154 ext4_ext_get_actual_len(newext),
2155 len, nearex, nearex + 1);
2156 memmove(nearex + 1, nearex,
2157 len * sizeof(struct ext4_extent));
2161 le16_add_cpu(&eh->eh_entries, 1);
2162 path[depth].p_ext = nearex;
2163 nearex->ee_block = newext->ee_block;
2164 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2165 nearex->ee_len = newext->ee_len;
2168 /* try to merge extents */
2169 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2170 ext4_ext_try_to_merge(handle, inode, path, nearex);
2173 /* time to correct all indexes above */
2174 err = ext4_ext_correct_indexes(handle, inode, path);
2178 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2181 ext4_free_ext_path(npath);
2185 static int ext4_fill_es_cache_info(struct inode *inode,
2186 ext4_lblk_t block, ext4_lblk_t num,
2187 struct fiemap_extent_info *fieinfo)
2189 ext4_lblk_t next, end = block + num - 1;
2190 struct extent_status es;
2191 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2195 while (block <= end) {
2198 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2200 if (ext4_es_is_unwritten(&es))
2201 flags |= FIEMAP_EXTENT_UNWRITTEN;
2202 if (ext4_es_is_delayed(&es))
2203 flags |= (FIEMAP_EXTENT_DELALLOC |
2204 FIEMAP_EXTENT_UNKNOWN);
2205 if (ext4_es_is_hole(&es))
2206 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2208 flags |= FIEMAP_EXTENT_LAST;
2209 if (flags & (FIEMAP_EXTENT_DELALLOC|
2210 EXT4_FIEMAP_EXTENT_HOLE))
2213 es.es_pblk = ext4_es_pblock(&es);
2214 err = fiemap_fill_next_extent(fieinfo,
2215 (__u64)es.es_lblk << blksize_bits,
2216 (__u64)es.es_pblk << blksize_bits,
2217 (__u64)es.es_len << blksize_bits,
2232 * ext4_ext_find_hole - find hole around given block according to the given path
2233 * @inode: inode we lookup in
2234 * @path: path in extent tree to @lblk
2235 * @lblk: pointer to logical block around which we want to determine hole
2237 * Determine hole length (and start if easily possible) around given logical
2238 * block. We don't try too hard to find the beginning of the hole but @path
2239 * actually points to extent before @lblk, we provide it.
2241 * The function returns the length of a hole starting at @lblk. We update @lblk
2242 * to the beginning of the hole if we managed to find it.
2244 static ext4_lblk_t ext4_ext_find_hole(struct inode *inode,
2245 struct ext4_ext_path *path,
2248 int depth = ext_depth(inode);
2249 struct ext4_extent *ex;
2252 ex = path[depth].p_ext;
2254 /* there is no extent yet, so gap is [0;-] */
2256 len = EXT_MAX_BLOCKS;
2257 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2258 len = le32_to_cpu(ex->ee_block) - *lblk;
2259 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2260 + ext4_ext_get_actual_len(ex)) {
2263 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2264 next = ext4_ext_next_allocated_block(path);
2265 BUG_ON(next == *lblk);
2275 * removes index from the index block.
2277 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2278 struct ext4_ext_path *path, int depth)
2283 /* free index block */
2285 path = path + depth;
2286 leaf = ext4_idx_pblock(path->p_idx);
2287 if (unlikely(path->p_hdr->eh_entries == 0)) {
2288 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2289 return -EFSCORRUPTED;
2291 err = ext4_ext_get_access(handle, inode, path);
2295 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2296 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2297 len *= sizeof(struct ext4_extent_idx);
2298 memmove(path->p_idx, path->p_idx + 1, len);
2301 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2302 err = ext4_ext_dirty(handle, inode, path);
2305 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2306 trace_ext4_ext_rm_idx(inode, leaf);
2308 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2309 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2311 while (--depth >= 0) {
2312 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2315 err = ext4_ext_get_access(handle, inode, path);
2318 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2319 err = ext4_ext_dirty(handle, inode, path);
2327 * ext4_ext_calc_credits_for_single_extent:
2328 * This routine returns max. credits that needed to insert an extent
2329 * to the extent tree.
2330 * When pass the actual path, the caller should calculate credits
2333 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2334 struct ext4_ext_path *path)
2337 int depth = ext_depth(inode);
2340 /* probably there is space in leaf? */
2341 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2342 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2345 * There are some space in the leaf tree, no
2346 * need to account for leaf block credit
2348 * bitmaps and block group descriptor blocks
2349 * and other metadata blocks still need to be
2352 /* 1 bitmap, 1 block group descriptor */
2353 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2358 return ext4_chunk_trans_blocks(inode, nrblocks);
2362 * How many index/leaf blocks need to change/allocate to add @extents extents?
2364 * If we add a single extent, then in the worse case, each tree level
2365 * index/leaf need to be changed in case of the tree split.
2367 * If more extents are inserted, they could cause the whole tree split more
2368 * than once, but this is really rare.
2370 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2375 /* If we are converting the inline data, only one is needed here. */
2376 if (ext4_has_inline_data(inode))
2379 depth = ext_depth(inode);
2389 static inline int get_default_free_blocks_flags(struct inode *inode)
2391 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2392 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2393 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2394 else if (ext4_should_journal_data(inode))
2395 return EXT4_FREE_BLOCKS_FORGET;
2400 * ext4_rereserve_cluster - increment the reserved cluster count when
2401 * freeing a cluster with a pending reservation
2403 * @inode - file containing the cluster
2404 * @lblk - logical block in cluster to be reserved
2406 * Increments the reserved cluster count and adjusts quota in a bigalloc
2407 * file system when freeing a partial cluster containing at least one
2408 * delayed and unwritten block. A partial cluster meeting that
2409 * requirement will have a pending reservation. If so, the
2410 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2411 * defer reserved and allocated space accounting to a subsequent call
2414 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2416 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2417 struct ext4_inode_info *ei = EXT4_I(inode);
2419 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2421 spin_lock(&ei->i_block_reservation_lock);
2422 ei->i_reserved_data_blocks++;
2423 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2424 spin_unlock(&ei->i_block_reservation_lock);
2426 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2427 ext4_remove_pending(inode, lblk);
2430 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2431 struct ext4_extent *ex,
2432 struct partial_cluster *partial,
2433 ext4_lblk_t from, ext4_lblk_t to)
2435 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2436 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2437 ext4_fsblk_t last_pblk, pblk;
2441 /* only extent tail removal is allowed */
2442 if (from < le32_to_cpu(ex->ee_block) ||
2443 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2444 ext4_error(sbi->s_sb,
2445 "strange request: removal(2) %u-%u from %u:%u",
2446 from, to, le32_to_cpu(ex->ee_block), ee_len);
2450 #ifdef EXTENTS_STATS
2451 spin_lock(&sbi->s_ext_stats_lock);
2452 sbi->s_ext_blocks += ee_len;
2453 sbi->s_ext_extents++;
2454 if (ee_len < sbi->s_ext_min)
2455 sbi->s_ext_min = ee_len;
2456 if (ee_len > sbi->s_ext_max)
2457 sbi->s_ext_max = ee_len;
2458 if (ext_depth(inode) > sbi->s_depth_max)
2459 sbi->s_depth_max = ext_depth(inode);
2460 spin_unlock(&sbi->s_ext_stats_lock);
2463 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2466 * if we have a partial cluster, and it's different from the
2467 * cluster of the last block in the extent, we free it
2469 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2471 if (partial->state != initial &&
2472 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2473 if (partial->state == tofree) {
2474 flags = get_default_free_blocks_flags(inode);
2475 if (ext4_is_pending(inode, partial->lblk))
2476 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2477 ext4_free_blocks(handle, inode, NULL,
2478 EXT4_C2B(sbi, partial->pclu),
2479 sbi->s_cluster_ratio, flags);
2480 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2481 ext4_rereserve_cluster(inode, partial->lblk);
2483 partial->state = initial;
2486 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2487 pblk = ext4_ext_pblock(ex) + ee_len - num;
2490 * We free the partial cluster at the end of the extent (if any),
2491 * unless the cluster is used by another extent (partial_cluster
2492 * state is nofree). If a partial cluster exists here, it must be
2493 * shared with the last block in the extent.
2495 flags = get_default_free_blocks_flags(inode);
2497 /* partial, left end cluster aligned, right end unaligned */
2498 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2499 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2500 (partial->state != nofree)) {
2501 if (ext4_is_pending(inode, to))
2502 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2503 ext4_free_blocks(handle, inode, NULL,
2504 EXT4_PBLK_CMASK(sbi, last_pblk),
2505 sbi->s_cluster_ratio, flags);
2506 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2507 ext4_rereserve_cluster(inode, to);
2508 partial->state = initial;
2509 flags = get_default_free_blocks_flags(inode);
2512 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2515 * For bigalloc file systems, we never free a partial cluster
2516 * at the beginning of the extent. Instead, we check to see if we
2517 * need to free it on a subsequent call to ext4_remove_blocks,
2518 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2520 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2521 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2523 /* reset the partial cluster if we've freed past it */
2524 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2525 partial->state = initial;
2528 * If we've freed the entire extent but the beginning is not left
2529 * cluster aligned and is not marked as ineligible for freeing we
2530 * record the partial cluster at the beginning of the extent. It
2531 * wasn't freed by the preceding ext4_free_blocks() call, and we
2532 * need to look farther to the left to determine if it's to be freed
2533 * (not shared with another extent). Else, reset the partial
2534 * cluster - we're either done freeing or the beginning of the
2535 * extent is left cluster aligned.
2537 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2538 if (partial->state == initial) {
2539 partial->pclu = EXT4_B2C(sbi, pblk);
2540 partial->lblk = from;
2541 partial->state = tofree;
2544 partial->state = initial;
2551 * ext4_ext_rm_leaf() Removes the extents associated with the
2552 * blocks appearing between "start" and "end". Both "start"
2553 * and "end" must appear in the same extent or EIO is returned.
2555 * @handle: The journal handle
2556 * @inode: The files inode
2557 * @path: The path to the leaf
2558 * @partial_cluster: The cluster which we'll have to free if all extents
2559 * has been released from it. However, if this value is
2560 * negative, it's a cluster just to the right of the
2561 * punched region and it must not be freed.
2562 * @start: The first block to remove
2563 * @end: The last block to remove
2566 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2567 struct ext4_ext_path *path,
2568 struct partial_cluster *partial,
2569 ext4_lblk_t start, ext4_lblk_t end)
2571 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2572 int err = 0, correct_index = 0;
2573 int depth = ext_depth(inode), credits, revoke_credits;
2574 struct ext4_extent_header *eh;
2577 ext4_lblk_t ex_ee_block;
2578 unsigned short ex_ee_len;
2579 unsigned unwritten = 0;
2580 struct ext4_extent *ex;
2583 /* the header must be checked already in ext4_ext_remove_space() */
2584 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2585 if (!path[depth].p_hdr)
2586 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2587 eh = path[depth].p_hdr;
2588 if (unlikely(path[depth].p_hdr == NULL)) {
2589 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2590 return -EFSCORRUPTED;
2592 /* find where to start removing */
2593 ex = path[depth].p_ext;
2595 ex = EXT_LAST_EXTENT(eh);
2597 ex_ee_block = le32_to_cpu(ex->ee_block);
2598 ex_ee_len = ext4_ext_get_actual_len(ex);
2600 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2602 while (ex >= EXT_FIRST_EXTENT(eh) &&
2603 ex_ee_block + ex_ee_len > start) {
2605 if (ext4_ext_is_unwritten(ex))
2610 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2611 unwritten, ex_ee_len);
2612 path[depth].p_ext = ex;
2614 a = ex_ee_block > start ? ex_ee_block : start;
2615 b = ex_ee_block+ex_ee_len - 1 < end ?
2616 ex_ee_block+ex_ee_len - 1 : end;
2618 ext_debug(inode, " border %u:%u\n", a, b);
2620 /* If this extent is beyond the end of the hole, skip it */
2621 if (end < ex_ee_block) {
2623 * We're going to skip this extent and move to another,
2624 * so note that its first cluster is in use to avoid
2625 * freeing it when removing blocks. Eventually, the
2626 * right edge of the truncated/punched region will
2627 * be just to the left.
2629 if (sbi->s_cluster_ratio > 1) {
2630 pblk = ext4_ext_pblock(ex);
2631 partial->pclu = EXT4_B2C(sbi, pblk);
2632 partial->state = nofree;
2635 ex_ee_block = le32_to_cpu(ex->ee_block);
2636 ex_ee_len = ext4_ext_get_actual_len(ex);
2638 } else if (b != ex_ee_block + ex_ee_len - 1) {
2639 EXT4_ERROR_INODE(inode,
2640 "can not handle truncate %u:%u "
2642 start, end, ex_ee_block,
2643 ex_ee_block + ex_ee_len - 1);
2644 err = -EFSCORRUPTED;
2646 } else if (a != ex_ee_block) {
2647 /* remove tail of the extent */
2648 num = a - ex_ee_block;
2650 /* remove whole extent: excellent! */
2654 * 3 for leaf, sb, and inode plus 2 (bmap and group
2655 * descriptor) for each block group; assume two block
2656 * groups plus ex_ee_len/blocks_per_block_group for
2659 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2660 if (ex == EXT_FIRST_EXTENT(eh)) {
2662 credits += (ext_depth(inode)) + 1;
2664 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2666 * We may end up freeing some index blocks and data from the
2667 * punched range. Note that partial clusters are accounted for
2668 * by ext4_free_data_revoke_credits().
2671 ext4_free_metadata_revoke_credits(inode->i_sb,
2673 ext4_free_data_revoke_credits(inode, b - a + 1);
2675 err = ext4_datasem_ensure_credits(handle, inode, credits,
2676 credits, revoke_credits);
2683 err = ext4_ext_get_access(handle, inode, path + depth);
2687 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2692 /* this extent is removed; mark slot entirely unused */
2693 ext4_ext_store_pblock(ex, 0);
2695 ex->ee_len = cpu_to_le16(num);
2697 * Do not mark unwritten if all the blocks in the
2698 * extent have been removed.
2700 if (unwritten && num)
2701 ext4_ext_mark_unwritten(ex);
2703 * If the extent was completely released,
2704 * we need to remove it from the leaf
2707 if (end != EXT_MAX_BLOCKS - 1) {
2709 * For hole punching, we need to scoot all the
2710 * extents up when an extent is removed so that
2711 * we dont have blank extents in the middle
2713 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2714 sizeof(struct ext4_extent));
2716 /* Now get rid of the one at the end */
2717 memset(EXT_LAST_EXTENT(eh), 0,
2718 sizeof(struct ext4_extent));
2720 le16_add_cpu(&eh->eh_entries, -1);
2723 err = ext4_ext_dirty(handle, inode, path + depth);
2727 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2728 ext4_ext_pblock(ex));
2730 ex_ee_block = le32_to_cpu(ex->ee_block);
2731 ex_ee_len = ext4_ext_get_actual_len(ex);
2734 if (correct_index && eh->eh_entries)
2735 err = ext4_ext_correct_indexes(handle, inode, path);
2738 * If there's a partial cluster and at least one extent remains in
2739 * the leaf, free the partial cluster if it isn't shared with the
2740 * current extent. If it is shared with the current extent
2741 * we reset the partial cluster because we've reached the start of the
2742 * truncated/punched region and we're done removing blocks.
2744 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2745 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2746 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2747 int flags = get_default_free_blocks_flags(inode);
2749 if (ext4_is_pending(inode, partial->lblk))
2750 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2751 ext4_free_blocks(handle, inode, NULL,
2752 EXT4_C2B(sbi, partial->pclu),
2753 sbi->s_cluster_ratio, flags);
2754 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2755 ext4_rereserve_cluster(inode, partial->lblk);
2757 partial->state = initial;
2760 /* if this leaf is free, then we should
2761 * remove it from index block above */
2762 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2763 err = ext4_ext_rm_idx(handle, inode, path, depth);
2770 * ext4_ext_more_to_rm:
2771 * returns 1 if current index has to be freed (even partial)
2774 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2776 BUG_ON(path->p_idx == NULL);
2778 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2782 * if truncate on deeper level happened, it wasn't partial,
2783 * so we have to consider current index for truncation
2785 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2790 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2793 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2794 int depth = ext_depth(inode);
2795 struct ext4_ext_path *path = NULL;
2796 struct partial_cluster partial;
2802 partial.state = initial;
2804 ext_debug(inode, "truncate since %u to %u\n", start, end);
2806 /* probably first extent we're gonna free will be last in block */
2807 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2809 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2811 return PTR_ERR(handle);
2814 trace_ext4_ext_remove_space(inode, start, end, depth);
2817 * Check if we are removing extents inside the extent tree. If that
2818 * is the case, we are going to punch a hole inside the extent tree
2819 * so we have to check whether we need to split the extent covering
2820 * the last block to remove so we can easily remove the part of it
2821 * in ext4_ext_rm_leaf().
2823 if (end < EXT_MAX_BLOCKS - 1) {
2824 struct ext4_extent *ex;
2825 ext4_lblk_t ee_block, ex_end, lblk;
2828 /* find extent for or closest extent to this block */
2829 path = ext4_find_extent(inode, end, NULL,
2830 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2832 ext4_journal_stop(handle);
2833 return PTR_ERR(path);
2835 depth = ext_depth(inode);
2836 /* Leaf not may not exist only if inode has no blocks at all */
2837 ex = path[depth].p_ext;
2840 EXT4_ERROR_INODE(inode,
2841 "path[%d].p_hdr == NULL",
2843 err = -EFSCORRUPTED;
2848 ee_block = le32_to_cpu(ex->ee_block);
2849 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2852 * See if the last block is inside the extent, if so split
2853 * the extent at 'end' block so we can easily remove the
2854 * tail of the first part of the split extent in
2855 * ext4_ext_rm_leaf().
2857 if (end >= ee_block && end < ex_end) {
2860 * If we're going to split the extent, note that
2861 * the cluster containing the block after 'end' is
2862 * in use to avoid freeing it when removing blocks.
2864 if (sbi->s_cluster_ratio > 1) {
2865 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2866 partial.pclu = EXT4_B2C(sbi, pblk);
2867 partial.state = nofree;
2871 * Split the extent in two so that 'end' is the last
2872 * block in the first new extent. Also we should not
2873 * fail removing space due to ENOSPC so try to use
2874 * reserved block if that happens.
2876 err = ext4_force_split_extent_at(handle, inode, &path,
2881 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2882 partial.state == initial) {
2884 * If we're punching, there's an extent to the right.
2885 * If the partial cluster hasn't been set, set it to
2886 * that extent's first cluster and its state to nofree
2887 * so it won't be freed should it contain blocks to be
2888 * removed. If it's already set (tofree/nofree), we're
2889 * retrying and keep the original partial cluster info
2890 * so a cluster marked tofree as a result of earlier
2891 * extent removal is not lost.
2894 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2899 partial.pclu = EXT4_B2C(sbi, pblk);
2900 partial.state = nofree;
2905 * We start scanning from right side, freeing all the blocks
2906 * after i_size and walking into the tree depth-wise.
2908 depth = ext_depth(inode);
2913 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2915 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2916 GFP_NOFS | __GFP_NOFAIL);
2918 ext4_journal_stop(handle);
2921 path[0].p_maxdepth = path[0].p_depth = depth;
2922 path[0].p_hdr = ext_inode_hdr(inode);
2925 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2926 err = -EFSCORRUPTED;
2932 while (i >= 0 && err == 0) {
2934 /* this is leaf block */
2935 err = ext4_ext_rm_leaf(handle, inode, path,
2936 &partial, start, end);
2937 /* root level has p_bh == NULL, brelse() eats this */
2938 brelse(path[i].p_bh);
2939 path[i].p_bh = NULL;
2944 /* this is index block */
2945 if (!path[i].p_hdr) {
2946 ext_debug(inode, "initialize header\n");
2947 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2950 if (!path[i].p_idx) {
2951 /* this level hasn't been touched yet */
2952 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2953 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2954 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2956 le16_to_cpu(path[i].p_hdr->eh_entries));
2958 /* we were already here, see at next index */
2962 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2963 i, EXT_FIRST_INDEX(path[i].p_hdr),
2965 if (ext4_ext_more_to_rm(path + i)) {
2966 struct buffer_head *bh;
2967 /* go to the next level */
2968 ext_debug(inode, "move to level %d (block %llu)\n",
2969 i + 1, ext4_idx_pblock(path[i].p_idx));
2970 memset(path + i + 1, 0, sizeof(*path));
2971 bh = read_extent_tree_block(inode, path[i].p_idx,
2975 /* should we reset i_size? */
2979 /* Yield here to deal with large extent trees.
2980 * Should be a no-op if we did IO above. */
2982 if (WARN_ON(i + 1 > depth)) {
2983 err = -EFSCORRUPTED;
2986 path[i + 1].p_bh = bh;
2988 /* save actual number of indexes since this
2989 * number is changed at the next iteration */
2990 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2993 /* we finished processing this index, go up */
2994 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2995 /* index is empty, remove it;
2996 * handle must be already prepared by the
2997 * truncatei_leaf() */
2998 err = ext4_ext_rm_idx(handle, inode, path, i);
3000 /* root level has p_bh == NULL, brelse() eats this */
3001 brelse(path[i].p_bh);
3002 path[i].p_bh = NULL;
3004 ext_debug(inode, "return to level %d\n", i);
3008 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3009 path->p_hdr->eh_entries);
3012 * if there's a partial cluster and we have removed the first extent
3013 * in the file, then we also free the partial cluster, if any
3015 if (partial.state == tofree && err == 0) {
3016 int flags = get_default_free_blocks_flags(inode);
3018 if (ext4_is_pending(inode, partial.lblk))
3019 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3020 ext4_free_blocks(handle, inode, NULL,
3021 EXT4_C2B(sbi, partial.pclu),
3022 sbi->s_cluster_ratio, flags);
3023 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3024 ext4_rereserve_cluster(inode, partial.lblk);
3025 partial.state = initial;
3028 /* TODO: flexible tree reduction should be here */
3029 if (path->p_hdr->eh_entries == 0) {
3031 * truncate to zero freed all the tree,
3032 * so we need to correct eh_depth
3034 err = ext4_ext_get_access(handle, inode, path);
3036 ext_inode_hdr(inode)->eh_depth = 0;
3037 ext_inode_hdr(inode)->eh_max =
3038 cpu_to_le16(ext4_ext_space_root(inode, 0));
3039 err = ext4_ext_dirty(handle, inode, path);
3043 ext4_free_ext_path(path);
3047 ext4_journal_stop(handle);
3053 * called at mount time
3055 void ext4_ext_init(struct super_block *sb)
3058 * possible initialization would be here
3061 if (ext4_has_feature_extents(sb)) {
3062 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3063 printk(KERN_INFO "EXT4-fs: file extents enabled"
3064 #ifdef AGGRESSIVE_TEST
3065 ", aggressive tests"
3067 #ifdef CHECK_BINSEARCH
3070 #ifdef EXTENTS_STATS
3075 #ifdef EXTENTS_STATS
3076 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3077 EXT4_SB(sb)->s_ext_min = 1 << 30;
3078 EXT4_SB(sb)->s_ext_max = 0;
3084 * called at umount time
3086 void ext4_ext_release(struct super_block *sb)
3088 if (!ext4_has_feature_extents(sb))
3091 #ifdef EXTENTS_STATS
3092 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3093 struct ext4_sb_info *sbi = EXT4_SB(sb);
3094 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3095 sbi->s_ext_blocks, sbi->s_ext_extents,
3096 sbi->s_ext_blocks / sbi->s_ext_extents);
3097 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3098 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3103 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3105 ext4_lblk_t ee_block;
3106 ext4_fsblk_t ee_pblock;
3107 unsigned int ee_len;
3109 ee_block = le32_to_cpu(ex->ee_block);
3110 ee_len = ext4_ext_get_actual_len(ex);
3111 ee_pblock = ext4_ext_pblock(ex);
3116 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3117 EXTENT_STATUS_WRITTEN);
3120 /* FIXME!! we need to try to merge to left or right after zero-out */
3121 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3123 ext4_fsblk_t ee_pblock;
3124 unsigned int ee_len;
3126 ee_len = ext4_ext_get_actual_len(ex);
3127 ee_pblock = ext4_ext_pblock(ex);
3128 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3133 * ext4_split_extent_at() splits an extent at given block.
3135 * @handle: the journal handle
3136 * @inode: the file inode
3137 * @path: the path to the extent
3138 * @split: the logical block where the extent is splitted.
3139 * @split_flags: indicates if the extent could be zeroout if split fails, and
3140 * the states(init or unwritten) of new extents.
3141 * @flags: flags used to insert new extent to extent tree.
3144 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3145 * of which are determined by split_flag.
3147 * There are two cases:
3148 * a> the extent are splitted into two extent.
3149 * b> split is not needed, and just mark the extent.
3151 * return 0 on success.
3153 static int ext4_split_extent_at(handle_t *handle,
3154 struct inode *inode,
3155 struct ext4_ext_path **ppath,
3160 struct ext4_ext_path *path = *ppath;
3161 ext4_fsblk_t newblock;
3162 ext4_lblk_t ee_block;
3163 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3164 struct ext4_extent *ex2 = NULL;
3165 unsigned int ee_len, depth;
3168 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3169 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3171 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3173 ext4_ext_show_leaf(inode, path);
3175 depth = ext_depth(inode);
3176 ex = path[depth].p_ext;
3177 ee_block = le32_to_cpu(ex->ee_block);
3178 ee_len = ext4_ext_get_actual_len(ex);
3179 newblock = split - ee_block + ext4_ext_pblock(ex);
3181 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3182 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3183 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3184 EXT4_EXT_MARK_UNWRIT1 |
3185 EXT4_EXT_MARK_UNWRIT2));
3187 err = ext4_ext_get_access(handle, inode, path + depth);
3191 if (split == ee_block) {
3193 * case b: block @split is the block that the extent begins with
3194 * then we just change the state of the extent, and splitting
3197 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3198 ext4_ext_mark_unwritten(ex);
3200 ext4_ext_mark_initialized(ex);
3202 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3203 ext4_ext_try_to_merge(handle, inode, path, ex);
3205 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3210 memcpy(&orig_ex, ex, sizeof(orig_ex));
3211 ex->ee_len = cpu_to_le16(split - ee_block);
3212 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3213 ext4_ext_mark_unwritten(ex);
3216 * path may lead to new leaf, not to original leaf any more
3217 * after ext4_ext_insert_extent() returns,
3219 err = ext4_ext_dirty(handle, inode, path + depth);
3221 goto fix_extent_len;
3224 ex2->ee_block = cpu_to_le32(split);
3225 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3226 ext4_ext_store_pblock(ex2, newblock);
3227 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3228 ext4_ext_mark_unwritten(ex2);
3230 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3231 if (err != -ENOSPC && err != -EDQUOT)
3234 if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3235 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3236 if (split_flag & EXT4_EXT_DATA_VALID1) {
3237 err = ext4_ext_zeroout(inode, ex2);
3238 zero_ex.ee_block = ex2->ee_block;
3239 zero_ex.ee_len = cpu_to_le16(
3240 ext4_ext_get_actual_len(ex2));
3241 ext4_ext_store_pblock(&zero_ex,
3242 ext4_ext_pblock(ex2));
3244 err = ext4_ext_zeroout(inode, ex);
3245 zero_ex.ee_block = ex->ee_block;
3246 zero_ex.ee_len = cpu_to_le16(
3247 ext4_ext_get_actual_len(ex));
3248 ext4_ext_store_pblock(&zero_ex,
3249 ext4_ext_pblock(ex));
3252 err = ext4_ext_zeroout(inode, &orig_ex);
3253 zero_ex.ee_block = orig_ex.ee_block;
3254 zero_ex.ee_len = cpu_to_le16(
3255 ext4_ext_get_actual_len(&orig_ex));
3256 ext4_ext_store_pblock(&zero_ex,
3257 ext4_ext_pblock(&orig_ex));
3261 /* update the extent length and mark as initialized */
3262 ex->ee_len = cpu_to_le16(ee_len);
3263 ext4_ext_try_to_merge(handle, inode, path, ex);
3264 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3266 /* update extent status tree */
3267 err = ext4_zeroout_es(inode, &zero_ex);
3268 /* If we failed at this point, we don't know in which
3269 * state the extent tree exactly is so don't try to fix
3270 * length of the original extent as it may do even more
3278 ex->ee_len = orig_ex.ee_len;
3280 * Ignore ext4_ext_dirty return value since we are already in error path
3281 * and err is a non-zero error code.
3283 ext4_ext_dirty(handle, inode, path + path->p_depth);
3286 ext4_ext_show_leaf(inode, path);
3291 * ext4_split_extents() splits an extent and mark extent which is covered
3292 * by @map as split_flags indicates
3294 * It may result in splitting the extent into multiple extents (up to three)
3295 * There are three possibilities:
3296 * a> There is no split required
3297 * b> Splits in two extents: Split is happening at either end of the extent
3298 * c> Splits in three extents: Somone is splitting in middle of the extent
3301 static int ext4_split_extent(handle_t *handle,
3302 struct inode *inode,
3303 struct ext4_ext_path **ppath,
3304 struct ext4_map_blocks *map,
3308 struct ext4_ext_path *path = *ppath;
3309 ext4_lblk_t ee_block;
3310 struct ext4_extent *ex;
3311 unsigned int ee_len, depth;
3314 int split_flag1, flags1;
3315 int allocated = map->m_len;
3317 depth = ext_depth(inode);
3318 ex = path[depth].p_ext;
3319 ee_block = le32_to_cpu(ex->ee_block);
3320 ee_len = ext4_ext_get_actual_len(ex);
3321 unwritten = ext4_ext_is_unwritten(ex);
3323 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3324 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3325 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3327 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3328 EXT4_EXT_MARK_UNWRIT2;
3329 if (split_flag & EXT4_EXT_DATA_VALID2)
3330 split_flag1 |= EXT4_EXT_DATA_VALID1;
3331 err = ext4_split_extent_at(handle, inode, ppath,
3332 map->m_lblk + map->m_len, split_flag1, flags1);
3336 allocated = ee_len - (map->m_lblk - ee_block);
3339 * Update path is required because previous ext4_split_extent_at() may
3340 * result in split of original leaf or extent zeroout.
3342 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3344 return PTR_ERR(path);
3345 depth = ext_depth(inode);
3346 ex = path[depth].p_ext;
3348 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3349 (unsigned long) map->m_lblk);
3350 return -EFSCORRUPTED;
3352 unwritten = ext4_ext_is_unwritten(ex);
3354 if (map->m_lblk >= ee_block) {
3355 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3357 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3358 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3359 EXT4_EXT_MARK_UNWRIT2);
3361 err = ext4_split_extent_at(handle, inode, ppath,
3362 map->m_lblk, split_flag1, flags);
3367 ext4_ext_show_leaf(inode, path);
3369 return err ? err : allocated;
3373 * This function is called by ext4_ext_map_blocks() if someone tries to write
3374 * to an unwritten extent. It may result in splitting the unwritten
3375 * extent into multiple extents (up to three - one initialized and two
3377 * There are three possibilities:
3378 * a> There is no split required: Entire extent should be initialized
3379 * b> Splits in two extents: Write is happening at either end of the extent
3380 * c> Splits in three extents: Somone is writing in middle of the extent
3383 * - The extent pointed to by 'path' is unwritten.
3384 * - The extent pointed to by 'path' contains a superset
3385 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3387 * Post-conditions on success:
3388 * - the returned value is the number of blocks beyond map->l_lblk
3389 * that are allocated and initialized.
3390 * It is guaranteed to be >= map->m_len.
3392 static int ext4_ext_convert_to_initialized(handle_t *handle,
3393 struct inode *inode,
3394 struct ext4_map_blocks *map,
3395 struct ext4_ext_path **ppath,
3398 struct ext4_ext_path *path = *ppath;
3399 struct ext4_sb_info *sbi;
3400 struct ext4_extent_header *eh;
3401 struct ext4_map_blocks split_map;
3402 struct ext4_extent zero_ex1, zero_ex2;
3403 struct ext4_extent *ex, *abut_ex;
3404 ext4_lblk_t ee_block, eof_block;
3405 unsigned int ee_len, depth, map_len = map->m_len;
3406 int allocated = 0, max_zeroout = 0;
3408 int split_flag = EXT4_EXT_DATA_VALID2;
3410 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3411 (unsigned long long)map->m_lblk, map_len);
3413 sbi = EXT4_SB(inode->i_sb);
3414 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3415 >> inode->i_sb->s_blocksize_bits;
3416 if (eof_block < map->m_lblk + map_len)
3417 eof_block = map->m_lblk + map_len;
3419 depth = ext_depth(inode);
3420 eh = path[depth].p_hdr;
3421 ex = path[depth].p_ext;
3422 ee_block = le32_to_cpu(ex->ee_block);
3423 ee_len = ext4_ext_get_actual_len(ex);
3424 zero_ex1.ee_len = 0;
3425 zero_ex2.ee_len = 0;
3427 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3429 /* Pre-conditions */
3430 BUG_ON(!ext4_ext_is_unwritten(ex));
3431 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3434 * Attempt to transfer newly initialized blocks from the currently
3435 * unwritten extent to its neighbor. This is much cheaper
3436 * than an insertion followed by a merge as those involve costly
3437 * memmove() calls. Transferring to the left is the common case in
3438 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3439 * followed by append writes.
3441 * Limitations of the current logic:
3442 * - L1: we do not deal with writes covering the whole extent.
3443 * This would require removing the extent if the transfer
3445 * - L2: we only attempt to merge with an extent stored in the
3446 * same extent tree node.
3448 if ((map->m_lblk == ee_block) &&
3449 /* See if we can merge left */
3450 (map_len < ee_len) && /*L1*/
3451 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3452 ext4_lblk_t prev_lblk;
3453 ext4_fsblk_t prev_pblk, ee_pblk;
3454 unsigned int prev_len;
3457 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3458 prev_len = ext4_ext_get_actual_len(abut_ex);
3459 prev_pblk = ext4_ext_pblock(abut_ex);
3460 ee_pblk = ext4_ext_pblock(ex);
3463 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3464 * upon those conditions:
3465 * - C1: abut_ex is initialized,
3466 * - C2: abut_ex is logically abutting ex,
3467 * - C3: abut_ex is physically abutting ex,
3468 * - C4: abut_ex can receive the additional blocks without
3469 * overflowing the (initialized) length limit.
3471 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3472 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3473 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3474 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3475 err = ext4_ext_get_access(handle, inode, path + depth);
3479 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3482 /* Shift the start of ex by 'map_len' blocks */
3483 ex->ee_block = cpu_to_le32(ee_block + map_len);
3484 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3485 ex->ee_len = cpu_to_le16(ee_len - map_len);
3486 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3488 /* Extend abut_ex by 'map_len' blocks */
3489 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3491 /* Result: number of initialized blocks past m_lblk */
3492 allocated = map_len;
3494 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3495 (map_len < ee_len) && /*L1*/
3496 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3497 /* See if we can merge right */
3498 ext4_lblk_t next_lblk;
3499 ext4_fsblk_t next_pblk, ee_pblk;
3500 unsigned int next_len;
3503 next_lblk = le32_to_cpu(abut_ex->ee_block);
3504 next_len = ext4_ext_get_actual_len(abut_ex);
3505 next_pblk = ext4_ext_pblock(abut_ex);
3506 ee_pblk = ext4_ext_pblock(ex);
3509 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3510 * upon those conditions:
3511 * - C1: abut_ex is initialized,
3512 * - C2: abut_ex is logically abutting ex,
3513 * - C3: abut_ex is physically abutting ex,
3514 * - C4: abut_ex can receive the additional blocks without
3515 * overflowing the (initialized) length limit.
3517 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3518 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3519 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3520 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3521 err = ext4_ext_get_access(handle, inode, path + depth);
3525 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3528 /* Shift the start of abut_ex by 'map_len' blocks */
3529 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3530 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3531 ex->ee_len = cpu_to_le16(ee_len - map_len);
3532 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3534 /* Extend abut_ex by 'map_len' blocks */
3535 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3537 /* Result: number of initialized blocks past m_lblk */
3538 allocated = map_len;
3542 /* Mark the block containing both extents as dirty */
3543 err = ext4_ext_dirty(handle, inode, path + depth);
3545 /* Update path to point to the right extent */
3546 path[depth].p_ext = abut_ex;
3549 allocated = ee_len - (map->m_lblk - ee_block);
3551 WARN_ON(map->m_lblk < ee_block);
3553 * It is safe to convert extent to initialized via explicit
3554 * zeroout only if extent is fully inside i_size or new_size.
3556 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3558 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3559 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3560 (inode->i_sb->s_blocksize_bits - 10);
3564 * 1. split the extent into three extents.
3565 * 2. split the extent into two extents, zeroout the head of the first
3567 * 3. split the extent into two extents, zeroout the tail of the second
3569 * 4. split the extent into two extents with out zeroout.
3570 * 5. no splitting needed, just possibly zeroout the head and / or the
3571 * tail of the extent.
3573 split_map.m_lblk = map->m_lblk;
3574 split_map.m_len = map->m_len;
3576 if (max_zeroout && (allocated > split_map.m_len)) {
3577 if (allocated <= max_zeroout) {
3580 cpu_to_le32(split_map.m_lblk +
3583 cpu_to_le16(allocated - split_map.m_len);
3584 ext4_ext_store_pblock(&zero_ex1,
3585 ext4_ext_pblock(ex) + split_map.m_lblk +
3586 split_map.m_len - ee_block);
3587 err = ext4_ext_zeroout(inode, &zero_ex1);
3590 split_map.m_len = allocated;
3592 if (split_map.m_lblk - ee_block + split_map.m_len <
3595 if (split_map.m_lblk != ee_block) {
3596 zero_ex2.ee_block = ex->ee_block;
3597 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3599 ext4_ext_store_pblock(&zero_ex2,
3600 ext4_ext_pblock(ex));
3601 err = ext4_ext_zeroout(inode, &zero_ex2);
3606 split_map.m_len += split_map.m_lblk - ee_block;
3607 split_map.m_lblk = ee_block;
3608 allocated = map->m_len;
3613 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3618 /* If we have gotten a failure, don't zero out status tree */
3620 err = ext4_zeroout_es(inode, &zero_ex1);
3622 err = ext4_zeroout_es(inode, &zero_ex2);
3624 return err ? err : allocated;
3628 * This function is called by ext4_ext_map_blocks() from
3629 * ext4_get_blocks_dio_write() when DIO to write
3630 * to an unwritten extent.
3632 * Writing to an unwritten extent may result in splitting the unwritten
3633 * extent into multiple initialized/unwritten extents (up to three)
3634 * There are three possibilities:
3635 * a> There is no split required: Entire extent should be unwritten
3636 * b> Splits in two extents: Write is happening at either end of the extent
3637 * c> Splits in three extents: Somone is writing in middle of the extent
3639 * This works the same way in the case of initialized -> unwritten conversion.
3641 * One of more index blocks maybe needed if the extent tree grow after
3642 * the unwritten extent split. To prevent ENOSPC occur at the IO
3643 * complete, we need to split the unwritten extent before DIO submit
3644 * the IO. The unwritten extent called at this time will be split
3645 * into three unwritten extent(at most). After IO complete, the part
3646 * being filled will be convert to initialized by the end_io callback function
3647 * via ext4_convert_unwritten_extents().
3649 * Returns the size of unwritten extent to be written on success.
3651 static int ext4_split_convert_extents(handle_t *handle,
3652 struct inode *inode,
3653 struct ext4_map_blocks *map,
3654 struct ext4_ext_path **ppath,
3657 struct ext4_ext_path *path = *ppath;
3658 ext4_lblk_t eof_block;
3659 ext4_lblk_t ee_block;
3660 struct ext4_extent *ex;
3661 unsigned int ee_len;
3662 int split_flag = 0, depth;
3664 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3665 (unsigned long long)map->m_lblk, map->m_len);
3667 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3668 >> inode->i_sb->s_blocksize_bits;
3669 if (eof_block < map->m_lblk + map->m_len)
3670 eof_block = map->m_lblk + map->m_len;
3672 * It is safe to convert extent to initialized via explicit
3673 * zeroout only if extent is fully inside i_size or new_size.
3675 depth = ext_depth(inode);
3676 ex = path[depth].p_ext;
3677 ee_block = le32_to_cpu(ex->ee_block);
3678 ee_len = ext4_ext_get_actual_len(ex);
3680 /* Convert to unwritten */
3681 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3682 split_flag |= EXT4_EXT_DATA_VALID1;
3683 /* Convert to initialized */
3684 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3685 split_flag |= ee_block + ee_len <= eof_block ?
3686 EXT4_EXT_MAY_ZEROOUT : 0;
3687 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3689 flags |= EXT4_GET_BLOCKS_PRE_IO;
3690 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3693 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3694 struct inode *inode,
3695 struct ext4_map_blocks *map,
3696 struct ext4_ext_path **ppath)
3698 struct ext4_ext_path *path = *ppath;
3699 struct ext4_extent *ex;
3700 ext4_lblk_t ee_block;
3701 unsigned int ee_len;
3705 depth = ext_depth(inode);
3706 ex = path[depth].p_ext;
3707 ee_block = le32_to_cpu(ex->ee_block);
3708 ee_len = ext4_ext_get_actual_len(ex);
3710 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3711 (unsigned long long)ee_block, ee_len);
3713 /* If extent is larger than requested it is a clear sign that we still
3714 * have some extent state machine issues left. So extent_split is still
3716 * TODO: Once all related issues will be fixed this situation should be
3719 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3720 #ifdef CONFIG_EXT4_DEBUG
3721 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3722 " len %u; IO logical block %llu, len %u",
3723 inode->i_ino, (unsigned long long)ee_block, ee_len,
3724 (unsigned long long)map->m_lblk, map->m_len);
3726 err = ext4_split_convert_extents(handle, inode, map, ppath,
3727 EXT4_GET_BLOCKS_CONVERT);
3730 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3732 return PTR_ERR(path);
3733 depth = ext_depth(inode);
3734 ex = path[depth].p_ext;
3737 err = ext4_ext_get_access(handle, inode, path + depth);
3740 /* first mark the extent as initialized */
3741 ext4_ext_mark_initialized(ex);
3743 /* note: ext4_ext_correct_indexes() isn't needed here because
3744 * borders are not changed
3746 ext4_ext_try_to_merge(handle, inode, path, ex);
3748 /* Mark modified extent as dirty */
3749 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3751 ext4_ext_show_leaf(inode, path);
3756 convert_initialized_extent(handle_t *handle, struct inode *inode,
3757 struct ext4_map_blocks *map,
3758 struct ext4_ext_path **ppath,
3759 unsigned int *allocated)
3761 struct ext4_ext_path *path = *ppath;
3762 struct ext4_extent *ex;
3763 ext4_lblk_t ee_block;
3764 unsigned int ee_len;
3769 * Make sure that the extent is no bigger than we support with
3772 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3773 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3775 depth = ext_depth(inode);
3776 ex = path[depth].p_ext;
3777 ee_block = le32_to_cpu(ex->ee_block);
3778 ee_len = ext4_ext_get_actual_len(ex);
3780 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3781 (unsigned long long)ee_block, ee_len);
3783 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3784 err = ext4_split_convert_extents(handle, inode, map, ppath,
3785 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3788 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3790 return PTR_ERR(path);
3791 depth = ext_depth(inode);
3792 ex = path[depth].p_ext;
3794 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3795 (unsigned long) map->m_lblk);
3796 return -EFSCORRUPTED;
3800 err = ext4_ext_get_access(handle, inode, path + depth);
3803 /* first mark the extent as unwritten */
3804 ext4_ext_mark_unwritten(ex);
3806 /* note: ext4_ext_correct_indexes() isn't needed here because
3807 * borders are not changed
3809 ext4_ext_try_to_merge(handle, inode, path, ex);
3811 /* Mark modified extent as dirty */
3812 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3815 ext4_ext_show_leaf(inode, path);
3817 ext4_update_inode_fsync_trans(handle, inode, 1);
3819 map->m_flags |= EXT4_MAP_UNWRITTEN;
3820 if (*allocated > map->m_len)
3821 *allocated = map->m_len;
3822 map->m_len = *allocated;
3827 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3828 struct ext4_map_blocks *map,
3829 struct ext4_ext_path **ppath, int flags,
3830 unsigned int allocated, ext4_fsblk_t newblock)
3832 struct ext4_ext_path __maybe_unused *path = *ppath;
3836 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3837 (unsigned long long)map->m_lblk, map->m_len, flags,
3839 ext4_ext_show_leaf(inode, path);
3842 * When writing into unwritten space, we should not fail to
3843 * allocate metadata blocks for the new extent block if needed.
3845 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3847 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3848 allocated, newblock);
3850 /* get_block() before submitting IO, split the extent */
3851 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3852 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3853 flags | EXT4_GET_BLOCKS_CONVERT);
3859 * shouldn't get a 0 return when splitting an extent unless
3860 * m_len is 0 (bug) or extent has been corrupted
3862 if (unlikely(ret == 0)) {
3863 EXT4_ERROR_INODE(inode,
3864 "unexpected ret == 0, m_len = %u",
3866 err = -EFSCORRUPTED;
3869 map->m_flags |= EXT4_MAP_UNWRITTEN;
3872 /* IO end_io complete, convert the filled extent to written */
3873 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3874 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3878 ext4_update_inode_fsync_trans(handle, inode, 1);
3881 /* buffered IO cases */
3883 * repeat fallocate creation request
3884 * we already have an unwritten extent
3886 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3887 map->m_flags |= EXT4_MAP_UNWRITTEN;
3891 /* buffered READ or buffered write_begin() lookup */
3892 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3894 * We have blocks reserved already. We
3895 * return allocated blocks so that delalloc
3896 * won't do block reservation for us. But
3897 * the buffer head will be unmapped so that
3898 * a read from the block returns 0s.
3900 map->m_flags |= EXT4_MAP_UNWRITTEN;
3905 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3906 * For buffered writes, at writepage time, etc. Convert a
3907 * discovered unwritten extent to written.
3909 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3914 ext4_update_inode_fsync_trans(handle, inode, 1);
3916 * shouldn't get a 0 return when converting an unwritten extent
3917 * unless m_len is 0 (bug) or extent has been corrupted
3919 if (unlikely(ret == 0)) {
3920 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3922 err = -EFSCORRUPTED;
3928 map->m_flags |= EXT4_MAP_NEW;
3930 map->m_flags |= EXT4_MAP_MAPPED;
3932 map->m_pblk = newblock;
3933 if (allocated > map->m_len)
3934 allocated = map->m_len;
3935 map->m_len = allocated;
3936 ext4_ext_show_leaf(inode, path);
3938 return err ? err : allocated;
3942 * get_implied_cluster_alloc - check to see if the requested
3943 * allocation (in the map structure) overlaps with a cluster already
3944 * allocated in an extent.
3945 * @sb The filesystem superblock structure
3946 * @map The requested lblk->pblk mapping
3947 * @ex The extent structure which might contain an implied
3948 * cluster allocation
3950 * This function is called by ext4_ext_map_blocks() after we failed to
3951 * find blocks that were already in the inode's extent tree. Hence,
3952 * we know that the beginning of the requested region cannot overlap
3953 * the extent from the inode's extent tree. There are three cases we
3954 * want to catch. The first is this case:
3956 * |--- cluster # N--|
3957 * |--- extent ---| |---- requested region ---|
3960 * The second case that we need to test for is this one:
3962 * |--------- cluster # N ----------------|
3963 * |--- requested region --| |------- extent ----|
3964 * |=======================|
3966 * The third case is when the requested region lies between two extents
3967 * within the same cluster:
3968 * |------------- cluster # N-------------|
3969 * |----- ex -----| |---- ex_right ----|
3970 * |------ requested region ------|
3971 * |================|
3973 * In each of the above cases, we need to set the map->m_pblk and
3974 * map->m_len so it corresponds to the return the extent labelled as
3975 * "|====|" from cluster #N, since it is already in use for data in
3976 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3977 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3978 * as a new "allocated" block region. Otherwise, we will return 0 and
3979 * ext4_ext_map_blocks() will then allocate one or more new clusters
3980 * by calling ext4_mb_new_blocks().
3982 static int get_implied_cluster_alloc(struct super_block *sb,
3983 struct ext4_map_blocks *map,
3984 struct ext4_extent *ex,
3985 struct ext4_ext_path *path)
3987 struct ext4_sb_info *sbi = EXT4_SB(sb);
3988 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3989 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3990 ext4_lblk_t rr_cluster_start;
3991 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3992 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3993 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3995 /* The extent passed in that we are trying to match */
3996 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3997 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3999 /* The requested region passed into ext4_map_blocks() */
4000 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4002 if ((rr_cluster_start == ex_cluster_end) ||
4003 (rr_cluster_start == ex_cluster_start)) {
4004 if (rr_cluster_start == ex_cluster_end)
4005 ee_start += ee_len - 1;
4006 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4007 map->m_len = min(map->m_len,
4008 (unsigned) sbi->s_cluster_ratio - c_offset);
4010 * Check for and handle this case:
4012 * |--------- cluster # N-------------|
4013 * |------- extent ----|
4014 * |--- requested region ---|
4018 if (map->m_lblk < ee_block)
4019 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4022 * Check for the case where there is already another allocated
4023 * block to the right of 'ex' but before the end of the cluster.
4025 * |------------- cluster # N-------------|
4026 * |----- ex -----| |---- ex_right ----|
4027 * |------ requested region ------|
4028 * |================|
4030 if (map->m_lblk > ee_block) {
4031 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4032 map->m_len = min(map->m_len, next - map->m_lblk);
4035 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4039 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4044 * Determine hole length around the given logical block, first try to
4045 * locate and expand the hole from the given @path, and then adjust it
4046 * if it's partially or completely converted to delayed extents, insert
4047 * it into the extent cache tree if it's indeed a hole, finally return
4048 * the length of the determined extent.
4050 static ext4_lblk_t ext4_ext_determine_insert_hole(struct inode *inode,
4051 struct ext4_ext_path *path,
4054 ext4_lblk_t hole_start, len;
4055 struct extent_status es;
4058 len = ext4_ext_find_hole(inode, path, &hole_start);
4060 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
4061 hole_start + len - 1, &es);
4066 * There's a delalloc extent in the hole, handle it if the delalloc
4067 * extent is in front of, behind and straddle the queried range.
4069 if (lblk >= es.es_lblk + es.es_len) {
4071 * The delalloc extent is in front of the queried range,
4072 * find again from the queried start block.
4074 len -= lblk - hole_start;
4077 } else if (in_range(lblk, es.es_lblk, es.es_len)) {
4079 * The delalloc extent containing lblk, it must have been
4080 * added after ext4_map_blocks() checked the extent status
4081 * tree, adjust the length to the delalloc extent's after
4084 len = es.es_lblk + es.es_len - lblk;
4088 * The delalloc extent is partially or completely behind
4089 * the queried range, update hole length until the
4090 * beginning of the delalloc extent.
4092 len = min(es.es_lblk - hole_start, len);
4096 /* Put just found gap into cache to speed up subsequent requests */
4097 ext_debug(inode, " -> %u:%u\n", hole_start, len);
4098 ext4_es_insert_extent(inode, hole_start, len, ~0, EXTENT_STATUS_HOLE);
4100 /* Update hole_len to reflect hole size after lblk */
4101 if (hole_start != lblk)
4102 len -= lblk - hole_start;
4108 * Block allocation/map/preallocation routine for extents based files
4111 * Need to be called with
4112 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4113 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4115 * return > 0, number of blocks already mapped/allocated
4116 * if create == 0 and these are pre-allocated blocks
4117 * buffer head is unmapped
4118 * otherwise blocks are mapped
4120 * return = 0, if plain look up failed (blocks have not been allocated)
4121 * buffer head is unmapped
4123 * return < 0, error case.
4125 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4126 struct ext4_map_blocks *map, int flags)
4128 struct ext4_ext_path *path = NULL;
4129 struct ext4_extent newex, *ex, ex2;
4130 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4131 ext4_fsblk_t newblock = 0, pblk;
4132 int err = 0, depth, ret;
4133 unsigned int allocated = 0, offset = 0;
4134 unsigned int allocated_clusters = 0;
4135 struct ext4_allocation_request ar;
4136 ext4_lblk_t cluster_offset;
4138 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4139 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4141 /* find extent for this block */
4142 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4144 err = PTR_ERR(path);
4149 depth = ext_depth(inode);
4152 * consistent leaf must not be empty;
4153 * this situation is possible, though, _during_ tree modification;
4154 * this is why assert can't be put in ext4_find_extent()
4156 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4157 EXT4_ERROR_INODE(inode, "bad extent address "
4158 "lblock: %lu, depth: %d pblock %lld",
4159 (unsigned long) map->m_lblk, depth,
4160 path[depth].p_block);
4161 err = -EFSCORRUPTED;
4165 ex = path[depth].p_ext;
4167 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4168 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4169 unsigned short ee_len;
4173 * unwritten extents are treated as holes, except that
4174 * we split out initialized portions during a write.
4176 ee_len = ext4_ext_get_actual_len(ex);
4178 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4180 /* if found extent covers block, simply return it */
4181 if (in_range(map->m_lblk, ee_block, ee_len)) {
4182 newblock = map->m_lblk - ee_block + ee_start;
4183 /* number of remaining blocks in the extent */
4184 allocated = ee_len - (map->m_lblk - ee_block);
4185 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4186 map->m_lblk, ee_block, ee_len, newblock);
4189 * If the extent is initialized check whether the
4190 * caller wants to convert it to unwritten.
4192 if ((!ext4_ext_is_unwritten(ex)) &&
4193 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4194 err = convert_initialized_extent(handle,
4195 inode, map, &path, &allocated);
4197 } else if (!ext4_ext_is_unwritten(ex)) {
4198 map->m_flags |= EXT4_MAP_MAPPED;
4199 map->m_pblk = newblock;
4200 if (allocated > map->m_len)
4201 allocated = map->m_len;
4202 map->m_len = allocated;
4203 ext4_ext_show_leaf(inode, path);
4207 ret = ext4_ext_handle_unwritten_extents(
4208 handle, inode, map, &path, flags,
4209 allocated, newblock);
4219 * requested block isn't allocated yet;
4220 * we couldn't try to create block if create flag is zero
4222 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4225 len = ext4_ext_determine_insert_hole(inode, path, map->m_lblk);
4228 map->m_len = min_t(unsigned int, map->m_len, len);
4233 * Okay, we need to do block allocation.
4235 newex.ee_block = cpu_to_le32(map->m_lblk);
4236 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4239 * If we are doing bigalloc, check to see if the extent returned
4240 * by ext4_find_extent() implies a cluster we can use.
4242 if (cluster_offset && ex &&
4243 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4244 ar.len = allocated = map->m_len;
4245 newblock = map->m_pblk;
4246 goto got_allocated_blocks;
4249 /* find neighbour allocated blocks */
4250 ar.lleft = map->m_lblk;
4251 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4254 ar.lright = map->m_lblk;
4255 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4259 /* Check if the extent after searching to the right implies a
4260 * cluster we can use. */
4261 if ((sbi->s_cluster_ratio > 1) && err &&
4262 get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4263 ar.len = allocated = map->m_len;
4264 newblock = map->m_pblk;
4265 goto got_allocated_blocks;
4269 * See if request is beyond maximum number of blocks we can have in
4270 * a single extent. For an initialized extent this limit is
4271 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4272 * EXT_UNWRITTEN_MAX_LEN.
4274 if (map->m_len > EXT_INIT_MAX_LEN &&
4275 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4276 map->m_len = EXT_INIT_MAX_LEN;
4277 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4278 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4279 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4281 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4282 newex.ee_len = cpu_to_le16(map->m_len);
4283 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4285 allocated = ext4_ext_get_actual_len(&newex);
4287 allocated = map->m_len;
4289 /* allocate new block */
4291 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4292 ar.logical = map->m_lblk;
4294 * We calculate the offset from the beginning of the cluster
4295 * for the logical block number, since when we allocate a
4296 * physical cluster, the physical block should start at the
4297 * same offset from the beginning of the cluster. This is
4298 * needed so that future calls to get_implied_cluster_alloc()
4301 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4302 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4304 ar.logical -= offset;
4305 if (S_ISREG(inode->i_mode))
4306 ar.flags = EXT4_MB_HINT_DATA;
4308 /* disable in-core preallocation for non-regular files */
4310 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4311 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4312 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4313 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4314 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4315 ar.flags |= EXT4_MB_USE_RESERVED;
4316 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4319 allocated_clusters = ar.len;
4320 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4321 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4322 ar.goal, newblock, ar.len, allocated);
4323 if (ar.len > allocated)
4326 got_allocated_blocks:
4327 /* try to insert new extent into found leaf and return */
4328 pblk = newblock + offset;
4329 ext4_ext_store_pblock(&newex, pblk);
4330 newex.ee_len = cpu_to_le16(ar.len);
4331 /* Mark unwritten */
4332 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4333 ext4_ext_mark_unwritten(&newex);
4334 map->m_flags |= EXT4_MAP_UNWRITTEN;
4337 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4339 if (allocated_clusters) {
4343 * free data blocks we just allocated.
4344 * not a good idea to call discard here directly,
4345 * but otherwise we'd need to call it every free().
4347 ext4_discard_preallocations(inode, 0);
4348 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4349 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4350 ext4_free_blocks(handle, inode, NULL, newblock,
4351 EXT4_C2B(sbi, allocated_clusters),
4358 * Reduce the reserved cluster count to reflect successful deferred
4359 * allocation of delayed allocated clusters or direct allocation of
4360 * clusters discovered to be delayed allocated. Once allocated, a
4361 * cluster is not included in the reserved count.
4363 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4364 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4366 * When allocating delayed allocated clusters, simply
4367 * reduce the reserved cluster count and claim quota
4369 ext4_da_update_reserve_space(inode, allocated_clusters,
4372 ext4_lblk_t lblk, len;
4376 * When allocating non-delayed allocated clusters
4377 * (from fallocate, filemap, DIO, or clusters
4378 * allocated when delalloc has been disabled by
4379 * ext4_nonda_switch), reduce the reserved cluster
4380 * count by the number of allocated clusters that
4381 * have previously been delayed allocated. Quota
4382 * has been claimed by ext4_mb_new_blocks() above,
4383 * so release the quota reservations made for any
4384 * previously delayed allocated clusters.
4386 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4387 len = allocated_clusters << sbi->s_cluster_bits;
4388 n = ext4_es_delayed_clu(inode, lblk, len);
4390 ext4_da_update_reserve_space(inode, (int) n, 0);
4395 * Cache the extent and update transaction to commit on fdatasync only
4396 * when it is _not_ an unwritten extent.
4398 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4399 ext4_update_inode_fsync_trans(handle, inode, 1);
4401 ext4_update_inode_fsync_trans(handle, inode, 0);
4403 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4405 map->m_len = ar.len;
4406 allocated = map->m_len;
4407 ext4_ext_show_leaf(inode, path);
4409 ext4_free_ext_path(path);
4411 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4412 err ? err : allocated);
4413 return err ? err : allocated;
4416 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4418 struct super_block *sb = inode->i_sb;
4419 ext4_lblk_t last_block;
4423 * TODO: optimization is possible here.
4424 * Probably we need not scan at all,
4425 * because page truncation is enough.
4428 /* we have to know where to truncate from in crash case */
4429 EXT4_I(inode)->i_disksize = inode->i_size;
4430 err = ext4_mark_inode_dirty(handle, inode);
4434 last_block = (inode->i_size + sb->s_blocksize - 1)
4435 >> EXT4_BLOCK_SIZE_BITS(sb);
4437 err = ext4_es_remove_extent(inode, last_block,
4438 EXT_MAX_BLOCKS - last_block);
4439 if (err == -ENOMEM) {
4440 memalloc_retry_wait(GFP_ATOMIC);
4446 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4447 if (err == -ENOMEM) {
4448 memalloc_retry_wait(GFP_ATOMIC);
4449 goto retry_remove_space;
4454 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4455 ext4_lblk_t len, loff_t new_size,
4458 struct inode *inode = file_inode(file);
4460 int ret = 0, ret2 = 0, ret3 = 0;
4463 struct ext4_map_blocks map;
4464 unsigned int credits;
4467 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4468 map.m_lblk = offset;
4471 * Don't normalize the request if it can fit in one extent so
4472 * that it doesn't get unnecessarily split into multiple
4475 if (len <= EXT_UNWRITTEN_MAX_LEN)
4476 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4479 * credits to insert 1 extent into extent tree
4481 credits = ext4_chunk_trans_blocks(inode, len);
4482 depth = ext_depth(inode);
4487 * Recalculate credits when extent tree depth changes.
4489 if (depth != ext_depth(inode)) {
4490 credits = ext4_chunk_trans_blocks(inode, len);
4491 depth = ext_depth(inode);
4494 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4496 if (IS_ERR(handle)) {
4497 ret = PTR_ERR(handle);
4500 ret = ext4_map_blocks(handle, inode, &map, flags);
4502 ext4_debug("inode #%lu: block %u: len %u: "
4503 "ext4_ext_map_blocks returned %d",
4504 inode->i_ino, map.m_lblk,
4506 ext4_mark_inode_dirty(handle, inode);
4507 ext4_journal_stop(handle);
4511 * allow a full retry cycle for any remaining allocations
4515 map.m_len = len = len - ret;
4516 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4517 inode->i_ctime = current_time(inode);
4519 if (epos > new_size)
4521 if (ext4_update_inode_size(inode, epos) & 0x1)
4522 inode->i_mtime = inode->i_ctime;
4524 ret2 = ext4_mark_inode_dirty(handle, inode);
4525 ext4_update_inode_fsync_trans(handle, inode, 1);
4526 ret3 = ext4_journal_stop(handle);
4527 ret2 = ret3 ? ret3 : ret2;
4531 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4534 return ret > 0 ? ret2 : ret;
4537 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4539 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4541 static long ext4_zero_range(struct file *file, loff_t offset,
4542 loff_t len, int mode)
4544 struct inode *inode = file_inode(file);
4545 struct address_space *mapping = file->f_mapping;
4546 handle_t *handle = NULL;
4547 unsigned int max_blocks;
4548 loff_t new_size = 0;
4552 int partial_begin, partial_end;
4555 unsigned int blkbits = inode->i_blkbits;
4557 trace_ext4_zero_range(inode, offset, len, mode);
4559 /* Call ext4_force_commit to flush all data in case of data=journal. */
4560 if (ext4_should_journal_data(inode)) {
4561 ret = ext4_force_commit(inode->i_sb);
4567 * Round up offset. This is not fallocate, we need to zero out
4568 * blocks, so convert interior block aligned part of the range to
4569 * unwritten and possibly manually zero out unaligned parts of the
4572 start = round_up(offset, 1 << blkbits);
4573 end = round_down((offset + len), 1 << blkbits);
4575 if (start < offset || end > offset + len)
4577 partial_begin = offset & ((1 << blkbits) - 1);
4578 partial_end = (offset + len) & ((1 << blkbits) - 1);
4580 lblk = start >> blkbits;
4581 max_blocks = (end >> blkbits);
4582 if (max_blocks < lblk)
4590 * Indirect files do not support unwritten extents
4592 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4597 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4598 (offset + len > inode->i_size ||
4599 offset + len > EXT4_I(inode)->i_disksize)) {
4600 new_size = offset + len;
4601 ret = inode_newsize_ok(inode, new_size);
4606 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4608 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4609 inode_dio_wait(inode);
4611 ret = file_modified(file);
4615 /* Preallocate the range including the unaligned edges */
4616 if (partial_begin || partial_end) {
4617 ret = ext4_alloc_file_blocks(file,
4618 round_down(offset, 1 << blkbits) >> blkbits,
4619 (round_up((offset + len), 1 << blkbits) -
4620 round_down(offset, 1 << blkbits)) >> blkbits,
4627 /* Zero range excluding the unaligned edges */
4628 if (max_blocks > 0) {
4629 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4633 * Prevent page faults from reinstantiating pages we have
4634 * released from page cache.
4636 filemap_invalidate_lock(mapping);
4638 ret = ext4_break_layouts(inode);
4640 filemap_invalidate_unlock(mapping);
4644 ret = ext4_update_disksize_before_punch(inode, offset, len);
4646 filemap_invalidate_unlock(mapping);
4649 /* Now release the pages and zero block aligned part of pages */
4650 truncate_pagecache_range(inode, start, end - 1);
4651 inode->i_mtime = inode->i_ctime = current_time(inode);
4653 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4655 filemap_invalidate_unlock(mapping);
4659 if (!partial_begin && !partial_end)
4663 * In worst case we have to writeout two nonadjacent unwritten
4664 * blocks and update the inode
4666 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4667 if (ext4_should_journal_data(inode))
4669 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4670 if (IS_ERR(handle)) {
4671 ret = PTR_ERR(handle);
4672 ext4_std_error(inode->i_sb, ret);
4676 inode->i_mtime = inode->i_ctime = current_time(inode);
4678 ext4_update_inode_size(inode, new_size);
4679 ret = ext4_mark_inode_dirty(handle, inode);
4682 /* Zero out partial block at the edges of the range */
4683 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4685 ext4_update_inode_fsync_trans(handle, inode, 1);
4687 if (file->f_flags & O_SYNC)
4688 ext4_handle_sync(handle);
4691 ext4_journal_stop(handle);
4693 inode_unlock(inode);
4698 * preallocate space for a file. This implements ext4's fallocate file
4699 * operation, which gets called from sys_fallocate system call.
4700 * For block-mapped files, posix_fallocate should fall back to the method
4701 * of writing zeroes to the required new blocks (the same behavior which is
4702 * expected for file systems which do not support fallocate() system call).
4704 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4706 struct inode *inode = file_inode(file);
4707 loff_t new_size = 0;
4708 unsigned int max_blocks;
4712 unsigned int blkbits = inode->i_blkbits;
4715 * Encrypted inodes can't handle collapse range or insert
4716 * range since we would need to re-encrypt blocks with a
4717 * different IV or XTS tweak (which are based on the logical
4720 if (IS_ENCRYPTED(inode) &&
4721 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4724 /* Return error if mode is not supported */
4725 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4726 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4727 FALLOC_FL_INSERT_RANGE))
4731 ret = ext4_convert_inline_data(inode);
4732 inode_unlock(inode);
4736 if (mode & FALLOC_FL_PUNCH_HOLE) {
4737 ret = ext4_punch_hole(file, offset, len);
4741 if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4742 ret = ext4_collapse_range(file, offset, len);
4746 if (mode & FALLOC_FL_INSERT_RANGE) {
4747 ret = ext4_insert_range(file, offset, len);
4751 if (mode & FALLOC_FL_ZERO_RANGE) {
4752 ret = ext4_zero_range(file, offset, len, mode);
4755 trace_ext4_fallocate_enter(inode, offset, len, mode);
4756 lblk = offset >> blkbits;
4758 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4759 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4764 * We only support preallocation for extent-based files only
4766 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4771 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4772 (offset + len > inode->i_size ||
4773 offset + len > EXT4_I(inode)->i_disksize)) {
4774 new_size = offset + len;
4775 ret = inode_newsize_ok(inode, new_size);
4780 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4781 inode_dio_wait(inode);
4783 ret = file_modified(file);
4787 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4791 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4792 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4793 EXT4_I(inode)->i_sync_tid);
4796 inode_unlock(inode);
4797 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4803 * This function convert a range of blocks to written extents
4804 * The caller of this function will pass the start offset and the size.
4805 * all unwritten extents within this range will be converted to
4808 * This function is called from the direct IO end io call back
4809 * function, to convert the fallocated extents after IO is completed.
4810 * Returns 0 on success.
4812 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4813 loff_t offset, ssize_t len)
4815 unsigned int max_blocks;
4816 int ret = 0, ret2 = 0, ret3 = 0;
4817 struct ext4_map_blocks map;
4818 unsigned int blkbits = inode->i_blkbits;
4819 unsigned int credits = 0;
4821 map.m_lblk = offset >> blkbits;
4822 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4826 * credits to insert 1 extent into extent tree
4828 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4830 while (ret >= 0 && ret < max_blocks) {
4832 map.m_len = (max_blocks -= ret);
4834 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4836 if (IS_ERR(handle)) {
4837 ret = PTR_ERR(handle);
4841 ret = ext4_map_blocks(handle, inode, &map,
4842 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4844 ext4_warning(inode->i_sb,
4845 "inode #%lu: block %u: len %u: "
4846 "ext4_ext_map_blocks returned %d",
4847 inode->i_ino, map.m_lblk,
4849 ret2 = ext4_mark_inode_dirty(handle, inode);
4851 ret3 = ext4_journal_stop(handle);
4856 if (ret <= 0 || ret2)
4859 return ret > 0 ? ret2 : ret;
4862 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4864 int ret = 0, err = 0;
4865 struct ext4_io_end_vec *io_end_vec;
4868 * This is somewhat ugly but the idea is clear: When transaction is
4869 * reserved, everything goes into it. Otherwise we rather start several
4870 * smaller transactions for conversion of each extent separately.
4873 handle = ext4_journal_start_reserved(handle,
4874 EXT4_HT_EXT_CONVERT);
4876 return PTR_ERR(handle);
4879 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4880 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4888 err = ext4_journal_stop(handle);
4890 return ret < 0 ? ret : err;
4893 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4897 int blockbits = inode->i_sb->s_blocksize_bits;
4902 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4903 struct ext4_iloc iloc;
4904 int offset; /* offset of xattr in inode */
4906 error = ext4_get_inode_loc(inode, &iloc);
4909 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4910 offset = EXT4_GOOD_OLD_INODE_SIZE +
4911 EXT4_I(inode)->i_extra_isize;
4913 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4915 iomap_type = IOMAP_INLINE;
4916 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4917 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4918 length = inode->i_sb->s_blocksize;
4919 iomap_type = IOMAP_MAPPED;
4921 /* no in-inode or external block for xattr, so return -ENOENT */
4926 iomap->addr = physical;
4928 iomap->length = length;
4929 iomap->type = iomap_type;
4935 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4936 loff_t length, unsigned flags,
4937 struct iomap *iomap, struct iomap *srcmap)
4941 error = ext4_iomap_xattr_fiemap(inode, iomap);
4942 if (error == 0 && (offset >= iomap->length))
4947 static const struct iomap_ops ext4_iomap_xattr_ops = {
4948 .iomap_begin = ext4_iomap_xattr_begin,
4951 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4955 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4956 maxbytes = inode->i_sb->s_maxbytes;
4958 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4962 if (start > maxbytes)
4966 * Shrink request scope to what the fs can actually handle.
4968 if (*len > maxbytes || (maxbytes - *len) < start)
4969 *len = maxbytes - start;
4973 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4978 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4979 error = ext4_ext_precache(inode);
4982 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4986 * For bitmap files the maximum size limit could be smaller than
4987 * s_maxbytes, so check len here manually instead of just relying on the
4990 error = ext4_fiemap_check_ranges(inode, start, &len);
4994 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4995 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4996 return iomap_fiemap(inode, fieinfo, start, len,
4997 &ext4_iomap_xattr_ops);
5000 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
5003 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5004 __u64 start, __u64 len)
5006 ext4_lblk_t start_blk, len_blks;
5010 if (ext4_has_inline_data(inode)) {
5013 down_read(&EXT4_I(inode)->xattr_sem);
5014 has_inline = ext4_has_inline_data(inode);
5015 up_read(&EXT4_I(inode)->xattr_sem);
5020 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5021 error = ext4_ext_precache(inode);
5024 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5027 error = fiemap_prep(inode, fieinfo, start, &len, 0);
5031 error = ext4_fiemap_check_ranges(inode, start, &len);
5035 start_blk = start >> inode->i_sb->s_blocksize_bits;
5036 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5037 if (last_blk >= EXT_MAX_BLOCKS)
5038 last_blk = EXT_MAX_BLOCKS-1;
5039 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5042 * Walk the extent tree gathering extent information
5043 * and pushing extents back to the user.
5045 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5049 * ext4_ext_shift_path_extents:
5050 * Shift the extents of a path structure lying between path[depth].p_ext
5051 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5052 * if it is right shift or left shift operation.
5055 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5056 struct inode *inode, handle_t *handle,
5057 enum SHIFT_DIRECTION SHIFT)
5060 struct ext4_extent *ex_start, *ex_last;
5061 bool update = false;
5062 int credits, restart_credits;
5063 depth = path->p_depth;
5065 while (depth >= 0) {
5066 if (depth == path->p_depth) {
5067 ex_start = path[depth].p_ext;
5069 return -EFSCORRUPTED;
5071 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5072 /* leaf + sb + inode */
5074 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5076 /* extent tree + sb + inode */
5077 credits = depth + 2;
5080 restart_credits = ext4_writepage_trans_blocks(inode);
5081 err = ext4_datasem_ensure_credits(handle, inode, credits,
5082 restart_credits, 0);
5089 err = ext4_ext_get_access(handle, inode, path + depth);
5093 while (ex_start <= ex_last) {
5094 if (SHIFT == SHIFT_LEFT) {
5095 le32_add_cpu(&ex_start->ee_block,
5097 /* Try to merge to the left. */
5099 EXT_FIRST_EXTENT(path[depth].p_hdr))
5101 ext4_ext_try_to_merge_right(inode,
5102 path, ex_start - 1))
5107 le32_add_cpu(&ex_last->ee_block, shift);
5108 ext4_ext_try_to_merge_right(inode, path,
5113 err = ext4_ext_dirty(handle, inode, path + depth);
5117 if (--depth < 0 || !update)
5121 /* Update index too */
5122 err = ext4_ext_get_access(handle, inode, path + depth);
5126 if (SHIFT == SHIFT_LEFT)
5127 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5129 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5130 err = ext4_ext_dirty(handle, inode, path + depth);
5134 /* we are done if current index is not a starting index */
5135 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5146 * ext4_ext_shift_extents:
5147 * All the extents which lies in the range from @start to the last allocated
5148 * block for the @inode are shifted either towards left or right (depending
5149 * upon @SHIFT) by @shift blocks.
5150 * On success, 0 is returned, error otherwise.
5153 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5154 ext4_lblk_t start, ext4_lblk_t shift,
5155 enum SHIFT_DIRECTION SHIFT)
5157 struct ext4_ext_path *path;
5159 struct ext4_extent *extent;
5160 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5161 ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5163 /* Let path point to the last extent */
5164 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5167 return PTR_ERR(path);
5169 depth = path->p_depth;
5170 extent = path[depth].p_ext;
5174 stop = le32_to_cpu(extent->ee_block);
5177 * For left shifts, make sure the hole on the left is big enough to
5178 * accommodate the shift. For right shifts, make sure the last extent
5179 * won't be shifted beyond EXT_MAX_BLOCKS.
5181 if (SHIFT == SHIFT_LEFT) {
5182 path = ext4_find_extent(inode, start - 1, &path,
5185 return PTR_ERR(path);
5186 depth = path->p_depth;
5187 extent = path[depth].p_ext;
5189 ex_start = le32_to_cpu(extent->ee_block);
5190 ex_end = le32_to_cpu(extent->ee_block) +
5191 ext4_ext_get_actual_len(extent);
5197 if ((start == ex_start && shift > ex_start) ||
5198 (shift > start - ex_end)) {
5203 if (shift > EXT_MAX_BLOCKS -
5204 (stop + ext4_ext_get_actual_len(extent))) {
5211 * In case of left shift, iterator points to start and it is increased
5212 * till we reach stop. In case of right shift, iterator points to stop
5213 * and it is decreased till we reach start.
5217 if (SHIFT == SHIFT_LEFT)
5222 if (tmp != EXT_MAX_BLOCKS)
5226 * Its safe to start updating extents. Start and stop are unsigned, so
5227 * in case of right shift if extent with 0 block is reached, iterator
5228 * becomes NULL to indicate the end of the loop.
5230 while (iterator && start <= stop) {
5231 path = ext4_find_extent(inode, *iterator, &path,
5234 return PTR_ERR(path);
5235 depth = path->p_depth;
5236 extent = path[depth].p_ext;
5238 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5239 (unsigned long) *iterator);
5240 return -EFSCORRUPTED;
5242 if (SHIFT == SHIFT_LEFT && *iterator >
5243 le32_to_cpu(extent->ee_block)) {
5244 /* Hole, move to the next extent */
5245 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5246 path[depth].p_ext++;
5248 *iterator = ext4_ext_next_allocated_block(path);
5254 if (SHIFT == SHIFT_LEFT) {
5255 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5256 *iterator = le32_to_cpu(extent->ee_block) +
5257 ext4_ext_get_actual_len(extent);
5259 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5260 if (le32_to_cpu(extent->ee_block) > start)
5261 *iterator = le32_to_cpu(extent->ee_block) - 1;
5262 else if (le32_to_cpu(extent->ee_block) == start)
5265 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5266 while (le32_to_cpu(extent->ee_block) >= start)
5269 if (extent == EXT_LAST_EXTENT(path[depth].p_hdr))
5275 path[depth].p_ext = extent;
5277 ret = ext4_ext_shift_path_extents(path, shift, inode,
5279 /* iterator can be NULL which means we should break */
5286 ext4_free_ext_path(path);
5291 * ext4_collapse_range:
5292 * This implements the fallocate's collapse range functionality for ext4
5293 * Returns: 0 and non-zero on error.
5295 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5297 struct inode *inode = file_inode(file);
5298 struct super_block *sb = inode->i_sb;
5299 struct address_space *mapping = inode->i_mapping;
5300 ext4_lblk_t punch_start, punch_stop;
5302 unsigned int credits;
5303 loff_t new_size, ioffset;
5307 * We need to test this early because xfstests assumes that a
5308 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5309 * system does not support collapse range.
5311 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5314 /* Collapse range works only on fs cluster size aligned regions. */
5315 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5318 trace_ext4_collapse_range(inode, offset, len);
5320 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5321 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5323 /* Call ext4_force_commit to flush all data in case of data=journal. */
5324 if (ext4_should_journal_data(inode)) {
5325 ret = ext4_force_commit(inode->i_sb);
5332 * There is no need to overlap collapse range with EOF, in which case
5333 * it is effectively a truncate operation
5335 if (offset + len >= inode->i_size) {
5340 /* Currently just for extent based files */
5341 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5346 /* Wait for existing dio to complete */
5347 inode_dio_wait(inode);
5349 ret = file_modified(file);
5354 * Prevent page faults from reinstantiating pages we have released from
5357 filemap_invalidate_lock(mapping);
5359 ret = ext4_break_layouts(inode);
5364 * Need to round down offset to be aligned with page size boundary
5365 * for page size > block size.
5367 ioffset = round_down(offset, PAGE_SIZE);
5369 * Write tail of the last page before removed range since it will get
5370 * removed from the page cache below.
5372 ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5376 * Write data that will be shifted to preserve them when discarding
5377 * page cache below. We are also protected from pages becoming dirty
5378 * by i_rwsem and invalidate_lock.
5380 ret = filemap_write_and_wait_range(mapping, offset + len,
5384 truncate_pagecache(inode, ioffset);
5386 credits = ext4_writepage_trans_blocks(inode);
5387 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5388 if (IS_ERR(handle)) {
5389 ret = PTR_ERR(handle);
5392 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5394 down_write(&EXT4_I(inode)->i_data_sem);
5395 ext4_discard_preallocations(inode, 0);
5397 ret = ext4_es_remove_extent(inode, punch_start,
5398 EXT_MAX_BLOCKS - punch_start);
5400 up_write(&EXT4_I(inode)->i_data_sem);
5404 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5406 up_write(&EXT4_I(inode)->i_data_sem);
5409 ext4_discard_preallocations(inode, 0);
5411 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5412 punch_stop - punch_start, SHIFT_LEFT);
5414 up_write(&EXT4_I(inode)->i_data_sem);
5418 new_size = inode->i_size - len;
5419 i_size_write(inode, new_size);
5420 EXT4_I(inode)->i_disksize = new_size;
5422 up_write(&EXT4_I(inode)->i_data_sem);
5424 ext4_handle_sync(handle);
5425 inode->i_mtime = inode->i_ctime = current_time(inode);
5426 ret = ext4_mark_inode_dirty(handle, inode);
5427 ext4_update_inode_fsync_trans(handle, inode, 1);
5430 ext4_journal_stop(handle);
5432 filemap_invalidate_unlock(mapping);
5434 inode_unlock(inode);
5439 * ext4_insert_range:
5440 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5441 * The data blocks starting from @offset to the EOF are shifted by @len
5442 * towards right to create a hole in the @inode. Inode size is increased
5444 * Returns 0 on success, error otherwise.
5446 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5448 struct inode *inode = file_inode(file);
5449 struct super_block *sb = inode->i_sb;
5450 struct address_space *mapping = inode->i_mapping;
5452 struct ext4_ext_path *path;
5453 struct ext4_extent *extent;
5454 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5455 unsigned int credits, ee_len;
5456 int ret = 0, depth, split_flag = 0;
5460 * We need to test this early because xfstests assumes that an
5461 * insert range of (0, 1) will return EOPNOTSUPP if the file
5462 * system does not support insert range.
5464 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5467 /* Insert range works only on fs cluster size aligned regions. */
5468 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5471 trace_ext4_insert_range(inode, offset, len);
5473 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5474 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5476 /* Call ext4_force_commit to flush all data in case of data=journal */
5477 if (ext4_should_journal_data(inode)) {
5478 ret = ext4_force_commit(inode->i_sb);
5484 /* Currently just for extent based files */
5485 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5490 /* Check whether the maximum file size would be exceeded */
5491 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5496 /* Offset must be less than i_size */
5497 if (offset >= inode->i_size) {
5502 /* Wait for existing dio to complete */
5503 inode_dio_wait(inode);
5505 ret = file_modified(file);
5510 * Prevent page faults from reinstantiating pages we have released from
5513 filemap_invalidate_lock(mapping);
5515 ret = ext4_break_layouts(inode);
5520 * Need to round down to align start offset to page size boundary
5521 * for page size > block size.
5523 ioffset = round_down(offset, PAGE_SIZE);
5524 /* Write out all dirty pages */
5525 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5529 truncate_pagecache(inode, ioffset);
5531 credits = ext4_writepage_trans_blocks(inode);
5532 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5533 if (IS_ERR(handle)) {
5534 ret = PTR_ERR(handle);
5537 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5539 /* Expand file to avoid data loss if there is error while shifting */
5540 inode->i_size += len;
5541 EXT4_I(inode)->i_disksize += len;
5542 inode->i_mtime = inode->i_ctime = current_time(inode);
5543 ret = ext4_mark_inode_dirty(handle, inode);
5547 down_write(&EXT4_I(inode)->i_data_sem);
5548 ext4_discard_preallocations(inode, 0);
5550 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5552 up_write(&EXT4_I(inode)->i_data_sem);
5556 depth = ext_depth(inode);
5557 extent = path[depth].p_ext;
5559 ee_start_lblk = le32_to_cpu(extent->ee_block);
5560 ee_len = ext4_ext_get_actual_len(extent);
5563 * If offset_lblk is not the starting block of extent, split
5564 * the extent @offset_lblk
5566 if ((offset_lblk > ee_start_lblk) &&
5567 (offset_lblk < (ee_start_lblk + ee_len))) {
5568 if (ext4_ext_is_unwritten(extent))
5569 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5570 EXT4_EXT_MARK_UNWRIT2;
5571 ret = ext4_split_extent_at(handle, inode, &path,
5572 offset_lblk, split_flag,
5574 EXT4_GET_BLOCKS_PRE_IO |
5575 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5578 ext4_free_ext_path(path);
5580 up_write(&EXT4_I(inode)->i_data_sem);
5584 ext4_free_ext_path(path);
5587 ret = ext4_es_remove_extent(inode, offset_lblk,
5588 EXT_MAX_BLOCKS - offset_lblk);
5590 up_write(&EXT4_I(inode)->i_data_sem);
5595 * if offset_lblk lies in a hole which is at start of file, use
5596 * ee_start_lblk to shift extents
5598 ret = ext4_ext_shift_extents(inode, handle,
5599 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5600 len_lblk, SHIFT_RIGHT);
5602 up_write(&EXT4_I(inode)->i_data_sem);
5604 ext4_handle_sync(handle);
5606 ext4_update_inode_fsync_trans(handle, inode, 1);
5609 ext4_journal_stop(handle);
5611 filemap_invalidate_unlock(mapping);
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_rwsem 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_free_ext_path(path1);
5805 ext4_free_ext_path(path2);
5806 path1 = path2 = NULL;
5808 return replaced_count;
5812 * ext4_clu_mapped - determine whether any block in a logical cluster has
5813 * been mapped to a physical cluster
5815 * @inode - file containing the logical cluster
5816 * @lclu - logical cluster of interest
5818 * Returns 1 if any block in the logical cluster is mapped, signifying
5819 * that a physical cluster has been allocated for it. Otherwise,
5820 * returns 0. Can also return negative error codes. Derived from
5821 * ext4_ext_map_blocks().
5823 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5825 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5826 struct ext4_ext_path *path;
5827 int depth, mapped = 0, err = 0;
5828 struct ext4_extent *extent;
5829 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5832 * if data can be stored inline, the logical cluster isn't
5833 * mapped - no physical clusters have been allocated, and the
5834 * file has no extents
5836 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) ||
5837 ext4_has_inline_data(inode))
5840 /* search for the extent closest to the first block in the cluster */
5841 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5843 err = PTR_ERR(path);
5848 depth = ext_depth(inode);
5851 * A consistent leaf must not be empty. This situation is possible,
5852 * though, _during_ tree modification, and it's why an assert can't
5853 * be put in ext4_find_extent().
5855 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5856 EXT4_ERROR_INODE(inode,
5857 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5858 (unsigned long) EXT4_C2B(sbi, lclu),
5859 depth, path[depth].p_block);
5860 err = -EFSCORRUPTED;
5864 extent = path[depth].p_ext;
5866 /* can't be mapped if the extent tree is empty */
5870 first_lblk = le32_to_cpu(extent->ee_block);
5871 first_lclu = EXT4_B2C(sbi, first_lblk);
5874 * Three possible outcomes at this point - found extent spanning
5875 * the target cluster, to the left of the target cluster, or to the
5876 * right of the target cluster. The first two cases are handled here.
5877 * The last case indicates the target cluster is not mapped.
5879 if (lclu >= first_lclu) {
5880 last_lclu = EXT4_B2C(sbi, first_lblk +
5881 ext4_ext_get_actual_len(extent) - 1);
5882 if (lclu <= last_lclu) {
5885 first_lblk = ext4_ext_next_allocated_block(path);
5886 first_lclu = EXT4_B2C(sbi, first_lblk);
5887 if (lclu == first_lclu)
5893 ext4_free_ext_path(path);
5895 return err ? err : mapped;
5899 * Updates physical block address and unwritten status of extent
5900 * starting at lblk start and of len. If such an extent doesn't exist,
5901 * this function splits the extent tree appropriately to create an
5902 * extent like this. This function is called in the fast commit
5903 * replay path. Returns 0 on success and error on failure.
5905 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5906 int len, int unwritten, ext4_fsblk_t pblk)
5908 struct ext4_ext_path *path = NULL, *ppath;
5909 struct ext4_extent *ex;
5912 path = ext4_find_extent(inode, start, NULL, 0);
5914 return PTR_ERR(path);
5915 ex = path[path->p_depth].p_ext;
5917 ret = -EFSCORRUPTED;
5921 if (le32_to_cpu(ex->ee_block) != start ||
5922 ext4_ext_get_actual_len(ex) != len) {
5923 /* We need to split this extent to match our extent first */
5925 down_write(&EXT4_I(inode)->i_data_sem);
5926 ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5927 up_write(&EXT4_I(inode)->i_data_sem);
5931 path = ext4_find_extent(inode, start, NULL, 0);
5935 ex = path[path->p_depth].p_ext;
5936 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5937 if (ext4_ext_get_actual_len(ex) != len) {
5938 down_write(&EXT4_I(inode)->i_data_sem);
5939 ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5941 up_write(&EXT4_I(inode)->i_data_sem);
5945 path = ext4_find_extent(inode, start, NULL, 0);
5948 ex = path[path->p_depth].p_ext;
5952 ext4_ext_mark_unwritten(ex);
5954 ext4_ext_mark_initialized(ex);
5955 ext4_ext_store_pblock(ex, pblk);
5956 down_write(&EXT4_I(inode)->i_data_sem);
5957 ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5958 up_write(&EXT4_I(inode)->i_data_sem);
5960 ext4_free_ext_path(path);
5961 ext4_mark_inode_dirty(NULL, inode);
5965 /* Try to shrink the extent tree */
5966 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5968 struct ext4_ext_path *path = NULL;
5969 struct ext4_extent *ex;
5970 ext4_lblk_t old_cur, cur = 0;
5973 path = ext4_find_extent(inode, cur, NULL, 0);
5976 ex = path[path->p_depth].p_ext;
5978 ext4_free_ext_path(path);
5979 ext4_mark_inode_dirty(NULL, inode);
5983 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5986 ext4_ext_try_to_merge(NULL, inode, path, ex);
5987 down_write(&EXT4_I(inode)->i_data_sem);
5988 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5989 up_write(&EXT4_I(inode)->i_data_sem);
5990 ext4_mark_inode_dirty(NULL, inode);
5991 ext4_free_ext_path(path);
5995 /* Check if *cur is a hole and if it is, skip it */
5996 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5999 struct ext4_map_blocks map;
6002 map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
6004 ret = ext4_map_blocks(NULL, inode, &map, 0);
6009 *cur = *cur + map.m_len;
6013 /* Count number of blocks used by this inode and update i_blocks */
6014 int ext4_ext_replay_set_iblocks(struct inode *inode)
6016 struct ext4_ext_path *path = NULL, *path2 = NULL;
6017 struct ext4_extent *ex;
6018 ext4_lblk_t cur = 0, end;
6019 int numblks = 0, i, ret = 0;
6020 ext4_fsblk_t cmp1, cmp2;
6021 struct ext4_map_blocks map;
6023 /* Determin the size of the file first */
6024 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6027 return PTR_ERR(path);
6028 ex = path[path->p_depth].p_ext;
6030 ext4_free_ext_path(path);
6033 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6034 ext4_free_ext_path(path);
6036 /* Count the number of data blocks */
6040 map.m_len = end - cur;
6041 ret = ext4_map_blocks(NULL, inode, &map, 0);
6046 cur = cur + map.m_len;
6050 * Count the number of extent tree blocks. We do it by looking up
6051 * two successive extents and determining the difference between
6052 * their paths. When path is different for 2 successive extents
6053 * we compare the blocks in the path at each level and increment
6054 * iblocks by total number of differences found.
6057 ret = skip_hole(inode, &cur);
6060 path = ext4_find_extent(inode, cur, NULL, 0);
6063 numblks += path->p_depth;
6064 ext4_free_ext_path(path);
6066 path = ext4_find_extent(inode, cur, NULL, 0);
6069 ex = path[path->p_depth].p_ext;
6071 ext4_free_ext_path(path);
6074 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6075 ext4_ext_get_actual_len(ex));
6076 ret = skip_hole(inode, &cur);
6078 ext4_free_ext_path(path);
6081 path2 = ext4_find_extent(inode, cur, NULL, 0);
6082 if (IS_ERR(path2)) {
6083 ext4_free_ext_path(path);
6086 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6088 if (i <= path->p_depth)
6089 cmp1 = path[i].p_bh ?
6090 path[i].p_bh->b_blocknr : 0;
6091 if (i <= path2->p_depth)
6092 cmp2 = path2[i].p_bh ?
6093 path2[i].p_bh->b_blocknr : 0;
6094 if (cmp1 != cmp2 && cmp2 != 0)
6097 ext4_free_ext_path(path);
6098 ext4_free_ext_path(path2);
6102 inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6103 ext4_mark_inode_dirty(NULL, inode);
6107 int ext4_ext_clear_bb(struct inode *inode)
6109 struct ext4_ext_path *path = NULL;
6110 struct ext4_extent *ex;
6111 ext4_lblk_t cur = 0, end;
6113 struct ext4_map_blocks map;
6115 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6118 /* Determin the size of the file first */
6119 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6122 return PTR_ERR(path);
6123 ex = path[path->p_depth].p_ext;
6125 ext4_free_ext_path(path);
6128 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6129 ext4_free_ext_path(path);
6134 map.m_len = end - cur;
6135 ret = ext4_map_blocks(NULL, inode, &map, 0);
6139 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6140 if (!IS_ERR_OR_NULL(path)) {
6141 for (j = 0; j < path->p_depth; j++) {
6143 ext4_mb_mark_bb(inode->i_sb,
6144 path[j].p_block, 1, 0);
6145 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6146 0, path[j].p_block, 1, 1);
6148 ext4_free_ext_path(path);
6150 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6151 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6152 map.m_lblk, map.m_pblk, map.m_len, 1);
6154 cur = cur + map.m_len;