1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
6 * Architecture independence:
7 * Copyright (c) 2005, Bull S.A.
8 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
12 * Extents support for EXT4
15 * - ext4*_error() should be used in some situations
16 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17 * - smart tree reduction
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/backing-dev.h>
31 #include "ext4_jbd2.h"
32 #include "ext4_extents.h"
35 #include <trace/events/ext4.h>
38 * used by extent splitting.
40 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
42 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
43 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
45 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
46 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
48 static __le32 ext4_extent_block_csum(struct inode *inode,
49 struct ext4_extent_header *eh)
51 struct ext4_inode_info *ei = EXT4_I(inode);
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
55 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
56 EXT4_EXTENT_TAIL_OFFSET(eh));
57 return cpu_to_le32(csum);
60 static int ext4_extent_block_csum_verify(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_extent_tail *et;
65 if (!ext4_has_metadata_csum(inode->i_sb))
68 et = find_ext4_extent_tail(eh);
69 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
74 static void ext4_extent_block_csum_set(struct inode *inode,
75 struct ext4_extent_header *eh)
77 struct ext4_extent_tail *et;
79 if (!ext4_has_metadata_csum(inode->i_sb))
82 et = find_ext4_extent_tail(eh);
83 et->et_checksum = ext4_extent_block_csum(inode, eh);
86 static int ext4_split_extent(handle_t *handle,
88 struct ext4_ext_path **ppath,
89 struct ext4_map_blocks *map,
93 static int ext4_split_extent_at(handle_t *handle,
95 struct ext4_ext_path **ppath,
100 static int ext4_find_delayed_extent(struct inode *inode,
101 struct extent_status *newes);
103 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
106 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
107 * moment, get_block can be called only for blocks inside i_size since
108 * page cache has been already dropped and writes are blocked by
109 * i_mutex. So we can safely drop the i_data_sem here.
111 BUG_ON(EXT4_JOURNAL(inode) == NULL);
112 ext4_discard_preallocations(inode);
113 up_write(&EXT4_I(inode)->i_data_sem);
119 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
120 * transaction with 'restart_cred' credits. The function drops i_data_sem
121 * when restarting transaction and gets it after transaction is restarted.
123 * The function returns 0 on success, 1 if transaction had to be restarted,
124 * and < 0 in case of fatal error.
126 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
127 int check_cred, int restart_cred,
133 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
134 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
136 down_write(&EXT4_I(inode)->i_data_sem);
145 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
146 struct ext4_ext_path *path)
149 /* path points to block */
150 BUFFER_TRACE(path->p_bh, "get_write_access");
151 return ext4_journal_get_write_access(handle, path->p_bh);
153 /* path points to leaf/index in inode body */
154 /* we use in-core data, no need to protect them */
164 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
165 struct inode *inode, struct ext4_ext_path *path)
169 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
171 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
172 /* path points to block */
173 err = __ext4_handle_dirty_metadata(where, line, handle,
176 /* path points to leaf/index in inode body */
177 err = ext4_mark_inode_dirty(handle, inode);
182 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
183 struct ext4_ext_path *path,
187 int depth = path->p_depth;
188 struct ext4_extent *ex;
191 * Try to predict block placement assuming that we are
192 * filling in a file which will eventually be
193 * non-sparse --- i.e., in the case of libbfd writing
194 * an ELF object sections out-of-order but in a way
195 * the eventually results in a contiguous object or
196 * executable file, or some database extending a table
197 * space file. However, this is actually somewhat
198 * non-ideal if we are writing a sparse file such as
199 * qemu or KVM writing a raw image file that is going
200 * to stay fairly sparse, since it will end up
201 * fragmenting the file system's free space. Maybe we
202 * should have some hueristics or some way to allow
203 * userspace to pass a hint to file system,
204 * especially if the latter case turns out to be
207 ex = path[depth].p_ext;
209 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
210 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
212 if (block > ext_block)
213 return ext_pblk + (block - ext_block);
215 return ext_pblk - (ext_block - block);
218 /* it looks like index is empty;
219 * try to find starting block from index itself */
220 if (path[depth].p_bh)
221 return path[depth].p_bh->b_blocknr;
224 /* OK. use inode's group */
225 return ext4_inode_to_goal_block(inode);
229 * Allocation for a meta data block
232 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
233 struct ext4_ext_path *path,
234 struct ext4_extent *ex, int *err, unsigned int flags)
236 ext4_fsblk_t goal, newblock;
238 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
239 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
244 static inline int ext4_ext_space_block(struct inode *inode, int check)
248 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
249 / sizeof(struct ext4_extent);
250 #ifdef AGGRESSIVE_TEST
251 if (!check && size > 6)
257 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
261 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
262 / sizeof(struct ext4_extent_idx);
263 #ifdef AGGRESSIVE_TEST
264 if (!check && size > 5)
270 static inline int ext4_ext_space_root(struct inode *inode, int check)
274 size = sizeof(EXT4_I(inode)->i_data);
275 size -= sizeof(struct ext4_extent_header);
276 size /= sizeof(struct ext4_extent);
277 #ifdef AGGRESSIVE_TEST
278 if (!check && size > 3)
284 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
288 size = sizeof(EXT4_I(inode)->i_data);
289 size -= sizeof(struct ext4_extent_header);
290 size /= sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292 if (!check && size > 4)
299 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
300 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
303 struct ext4_ext_path *path = *ppath;
304 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
306 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
307 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
308 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
309 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
313 ext4_ext_max_entries(struct inode *inode, int depth)
317 if (depth == ext_depth(inode)) {
319 max = ext4_ext_space_root(inode, 1);
321 max = ext4_ext_space_root_idx(inode, 1);
324 max = ext4_ext_space_block(inode, 1);
326 max = ext4_ext_space_block_idx(inode, 1);
332 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
334 ext4_fsblk_t block = ext4_ext_pblock(ext);
335 int len = ext4_ext_get_actual_len(ext);
336 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
341 * - overflow/wrap-around
343 if (lblock + len <= lblock)
345 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
348 static int ext4_valid_extent_idx(struct inode *inode,
349 struct ext4_extent_idx *ext_idx)
351 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
353 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
356 static int ext4_valid_extent_entries(struct inode *inode,
357 struct ext4_extent_header *eh,
360 unsigned short entries;
361 if (eh->eh_entries == 0)
364 entries = le16_to_cpu(eh->eh_entries);
368 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
369 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
370 ext4_fsblk_t pblock = 0;
371 ext4_lblk_t lblock = 0;
372 ext4_lblk_t prev = 0;
375 if (!ext4_valid_extent(inode, ext))
378 /* Check for overlapping extents */
379 lblock = le32_to_cpu(ext->ee_block);
380 len = ext4_ext_get_actual_len(ext);
381 if ((lblock <= prev) && prev) {
382 pblock = ext4_ext_pblock(ext);
383 es->s_last_error_block = cpu_to_le64(pblock);
388 prev = lblock + len - 1;
391 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
393 if (!ext4_valid_extent_idx(inode, ext_idx))
402 static int __ext4_ext_check(const char *function, unsigned int line,
403 struct inode *inode, struct ext4_extent_header *eh,
404 int depth, ext4_fsblk_t pblk)
406 const char *error_msg;
407 int max = 0, err = -EFSCORRUPTED;
409 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
410 error_msg = "invalid magic";
413 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
414 error_msg = "unexpected eh_depth";
417 if (unlikely(eh->eh_max == 0)) {
418 error_msg = "invalid eh_max";
421 max = ext4_ext_max_entries(inode, depth);
422 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
423 error_msg = "too large eh_max";
426 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
427 error_msg = "invalid eh_entries";
430 if (!ext4_valid_extent_entries(inode, eh, depth)) {
431 error_msg = "invalid extent entries";
434 if (unlikely(depth > 32)) {
435 error_msg = "too large eh_depth";
438 /* Verify checksum on non-root extent tree nodes */
439 if (ext_depth(inode) != depth &&
440 !ext4_extent_block_csum_verify(inode, eh)) {
441 error_msg = "extent tree corrupted";
448 ext4_set_errno(inode->i_sb, -err);
449 ext4_error_inode(inode, function, line, 0,
450 "pblk %llu bad header/extent: %s - magic %x, "
451 "entries %u, max %u(%u), depth %u(%u)",
452 (unsigned long long) pblk, error_msg,
453 le16_to_cpu(eh->eh_magic),
454 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
455 max, le16_to_cpu(eh->eh_depth), depth);
459 #define ext4_ext_check(inode, eh, depth, pblk) \
460 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
462 int ext4_ext_check_inode(struct inode *inode)
464 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
467 static struct buffer_head *
468 __read_extent_tree_block(const char *function, unsigned int line,
469 struct inode *inode, ext4_fsblk_t pblk, int depth,
472 struct buffer_head *bh;
475 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
477 return ERR_PTR(-ENOMEM);
479 if (!bh_uptodate_or_lock(bh)) {
480 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
481 err = bh_submit_read(bh);
485 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
487 if (!ext4_has_feature_journal(inode->i_sb) ||
489 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
490 err = __ext4_ext_check(function, line, inode,
491 ext_block_hdr(bh), depth, pblk);
495 set_buffer_verified(bh);
497 * If this is a leaf block, cache all of its entries
499 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
500 struct ext4_extent_header *eh = ext_block_hdr(bh);
501 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
502 ext4_lblk_t prev = 0;
505 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
506 unsigned int status = EXTENT_STATUS_WRITTEN;
507 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
508 int len = ext4_ext_get_actual_len(ex);
510 if (prev && (prev != lblk))
511 ext4_es_cache_extent(inode, prev,
515 if (ext4_ext_is_unwritten(ex))
516 status = EXTENT_STATUS_UNWRITTEN;
517 ext4_es_cache_extent(inode, lblk, len,
518 ext4_ext_pblock(ex), status);
529 #define read_extent_tree_block(inode, pblk, depth, flags) \
530 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
534 * This function is called to cache a file's extent information in the
537 int ext4_ext_precache(struct inode *inode)
539 struct ext4_inode_info *ei = EXT4_I(inode);
540 struct ext4_ext_path *path = NULL;
541 struct buffer_head *bh;
542 int i = 0, depth, ret = 0;
544 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
545 return 0; /* not an extent-mapped inode */
547 down_read(&ei->i_data_sem);
548 depth = ext_depth(inode);
550 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
553 up_read(&ei->i_data_sem);
557 /* Don't cache anything if there are no external extent blocks */
560 path[0].p_hdr = ext_inode_hdr(inode);
561 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
564 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
567 * If this is a leaf block or we've reached the end of
568 * the index block, go up
571 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
572 brelse(path[i].p_bh);
577 bh = read_extent_tree_block(inode,
578 ext4_idx_pblock(path[i].p_idx++),
580 EXT4_EX_FORCE_CACHE);
587 path[i].p_hdr = ext_block_hdr(bh);
588 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
590 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
592 up_read(&ei->i_data_sem);
593 ext4_ext_drop_refs(path);
599 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
601 int k, l = path->p_depth;
604 for (k = 0; k <= l; k++, path++) {
606 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
607 ext4_idx_pblock(path->p_idx));
608 } else if (path->p_ext) {
609 ext_debug(" %d:[%d]%d:%llu ",
610 le32_to_cpu(path->p_ext->ee_block),
611 ext4_ext_is_unwritten(path->p_ext),
612 ext4_ext_get_actual_len(path->p_ext),
613 ext4_ext_pblock(path->p_ext));
620 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
622 int depth = ext_depth(inode);
623 struct ext4_extent_header *eh;
624 struct ext4_extent *ex;
630 eh = path[depth].p_hdr;
631 ex = EXT_FIRST_EXTENT(eh);
633 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
635 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
636 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
637 ext4_ext_is_unwritten(ex),
638 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
643 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
644 ext4_fsblk_t newblock, int level)
646 int depth = ext_depth(inode);
647 struct ext4_extent *ex;
649 if (depth != level) {
650 struct ext4_extent_idx *idx;
651 idx = path[level].p_idx;
652 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
653 ext_debug("%d: move %d:%llu in new index %llu\n", level,
654 le32_to_cpu(idx->ei_block),
655 ext4_idx_pblock(idx),
663 ex = path[depth].p_ext;
664 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
665 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
666 le32_to_cpu(ex->ee_block),
668 ext4_ext_is_unwritten(ex),
669 ext4_ext_get_actual_len(ex),
676 #define ext4_ext_show_path(inode, path)
677 #define ext4_ext_show_leaf(inode, path)
678 #define ext4_ext_show_move(inode, path, newblock, level)
681 void ext4_ext_drop_refs(struct ext4_ext_path *path)
687 depth = path->p_depth;
688 for (i = 0; i <= depth; i++, path++)
696 * ext4_ext_binsearch_idx:
697 * binary search for the closest index of the given block
698 * the header must be checked before calling this
701 ext4_ext_binsearch_idx(struct inode *inode,
702 struct ext4_ext_path *path, ext4_lblk_t block)
704 struct ext4_extent_header *eh = path->p_hdr;
705 struct ext4_extent_idx *r, *l, *m;
708 ext_debug("binsearch for %u(idx): ", block);
710 l = EXT_FIRST_INDEX(eh) + 1;
711 r = EXT_LAST_INDEX(eh);
714 if (block < le32_to_cpu(m->ei_block))
718 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
719 m, le32_to_cpu(m->ei_block),
720 r, le32_to_cpu(r->ei_block));
724 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
725 ext4_idx_pblock(path->p_idx));
727 #ifdef CHECK_BINSEARCH
729 struct ext4_extent_idx *chix, *ix;
732 chix = ix = EXT_FIRST_INDEX(eh);
733 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
735 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
736 printk(KERN_DEBUG "k=%d, ix=0x%p, "
738 ix, EXT_FIRST_INDEX(eh));
739 printk(KERN_DEBUG "%u <= %u\n",
740 le32_to_cpu(ix->ei_block),
741 le32_to_cpu(ix[-1].ei_block));
743 BUG_ON(k && le32_to_cpu(ix->ei_block)
744 <= le32_to_cpu(ix[-1].ei_block));
745 if (block < le32_to_cpu(ix->ei_block))
749 BUG_ON(chix != path->p_idx);
756 * ext4_ext_binsearch:
757 * binary search for closest extent of the given block
758 * the header must be checked before calling this
761 ext4_ext_binsearch(struct inode *inode,
762 struct ext4_ext_path *path, ext4_lblk_t block)
764 struct ext4_extent_header *eh = path->p_hdr;
765 struct ext4_extent *r, *l, *m;
767 if (eh->eh_entries == 0) {
769 * this leaf is empty:
770 * we get such a leaf in split/add case
775 ext_debug("binsearch for %u: ", block);
777 l = EXT_FIRST_EXTENT(eh) + 1;
778 r = EXT_LAST_EXTENT(eh);
782 if (block < le32_to_cpu(m->ee_block))
786 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
787 m, le32_to_cpu(m->ee_block),
788 r, le32_to_cpu(r->ee_block));
792 ext_debug(" -> %d:%llu:[%d]%d ",
793 le32_to_cpu(path->p_ext->ee_block),
794 ext4_ext_pblock(path->p_ext),
795 ext4_ext_is_unwritten(path->p_ext),
796 ext4_ext_get_actual_len(path->p_ext));
798 #ifdef CHECK_BINSEARCH
800 struct ext4_extent *chex, *ex;
803 chex = ex = EXT_FIRST_EXTENT(eh);
804 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
805 BUG_ON(k && le32_to_cpu(ex->ee_block)
806 <= le32_to_cpu(ex[-1].ee_block));
807 if (block < le32_to_cpu(ex->ee_block))
811 BUG_ON(chex != path->p_ext);
817 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
819 struct ext4_extent_header *eh;
821 eh = ext_inode_hdr(inode);
824 eh->eh_magic = EXT4_EXT_MAGIC;
825 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
826 ext4_mark_inode_dirty(handle, inode);
830 struct ext4_ext_path *
831 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
832 struct ext4_ext_path **orig_path, int flags)
834 struct ext4_extent_header *eh;
835 struct buffer_head *bh;
836 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
837 short int depth, i, ppos = 0;
840 eh = ext_inode_hdr(inode);
841 depth = ext_depth(inode);
842 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
843 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
850 ext4_ext_drop_refs(path);
851 if (depth > path[0].p_maxdepth) {
853 *orig_path = path = NULL;
857 /* account possible depth increase */
858 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
861 return ERR_PTR(-ENOMEM);
862 path[0].p_maxdepth = depth + 1;
868 /* walk through the tree */
870 ext_debug("depth %d: num %d, max %d\n",
871 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
873 ext4_ext_binsearch_idx(inode, path + ppos, block);
874 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
875 path[ppos].p_depth = i;
876 path[ppos].p_ext = NULL;
878 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
885 eh = ext_block_hdr(bh);
887 path[ppos].p_bh = bh;
888 path[ppos].p_hdr = eh;
891 path[ppos].p_depth = i;
892 path[ppos].p_ext = NULL;
893 path[ppos].p_idx = NULL;
896 ext4_ext_binsearch(inode, path + ppos, block);
897 /* if not an empty leaf */
898 if (path[ppos].p_ext)
899 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
901 ext4_ext_show_path(inode, path);
906 ext4_ext_drop_refs(path);
914 * ext4_ext_insert_index:
915 * insert new index [@logical;@ptr] into the block at @curp;
916 * check where to insert: before @curp or after @curp
918 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
919 struct ext4_ext_path *curp,
920 int logical, ext4_fsblk_t ptr)
922 struct ext4_extent_idx *ix;
925 err = ext4_ext_get_access(handle, inode, curp);
929 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
930 EXT4_ERROR_INODE(inode,
931 "logical %d == ei_block %d!",
932 logical, le32_to_cpu(curp->p_idx->ei_block));
933 return -EFSCORRUPTED;
936 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
937 >= le16_to_cpu(curp->p_hdr->eh_max))) {
938 EXT4_ERROR_INODE(inode,
939 "eh_entries %d >= eh_max %d!",
940 le16_to_cpu(curp->p_hdr->eh_entries),
941 le16_to_cpu(curp->p_hdr->eh_max));
942 return -EFSCORRUPTED;
945 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
947 ext_debug("insert new index %d after: %llu\n", logical, ptr);
948 ix = curp->p_idx + 1;
951 ext_debug("insert new index %d before: %llu\n", logical, ptr);
955 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
958 ext_debug("insert new index %d: "
959 "move %d indices from 0x%p to 0x%p\n",
960 logical, len, ix, ix + 1);
961 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
964 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
965 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
966 return -EFSCORRUPTED;
969 ix->ei_block = cpu_to_le32(logical);
970 ext4_idx_store_pblock(ix, ptr);
971 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
973 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
974 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
975 return -EFSCORRUPTED;
978 err = ext4_ext_dirty(handle, inode, curp);
979 ext4_std_error(inode->i_sb, err);
986 * inserts new subtree into the path, using free index entry
988 * - allocates all needed blocks (new leaf and all intermediate index blocks)
989 * - makes decision where to split
990 * - moves remaining extents and index entries (right to the split point)
991 * into the newly allocated blocks
992 * - initializes subtree
994 static int ext4_ext_split(handle_t *handle, struct inode *inode,
996 struct ext4_ext_path *path,
997 struct ext4_extent *newext, int at)
999 struct buffer_head *bh = NULL;
1000 int depth = ext_depth(inode);
1001 struct ext4_extent_header *neh;
1002 struct ext4_extent_idx *fidx;
1003 int i = at, k, m, a;
1004 ext4_fsblk_t newblock, oldblock;
1006 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1008 size_t ext_size = 0;
1010 /* make decision: where to split? */
1011 /* FIXME: now decision is simplest: at current extent */
1013 /* if current leaf will be split, then we should use
1014 * border from split point */
1015 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1016 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1017 return -EFSCORRUPTED;
1019 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1020 border = path[depth].p_ext[1].ee_block;
1021 ext_debug("leaf will be split."
1022 " next leaf starts at %d\n",
1023 le32_to_cpu(border));
1025 border = newext->ee_block;
1026 ext_debug("leaf will be added."
1027 " next leaf starts at %d\n",
1028 le32_to_cpu(border));
1032 * If error occurs, then we break processing
1033 * and mark filesystem read-only. index won't
1034 * be inserted and tree will be in consistent
1035 * state. Next mount will repair buffers too.
1039 * Get array to track all allocated blocks.
1040 * We need this to handle errors and free blocks
1043 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1047 /* allocate all needed blocks */
1048 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1049 for (a = 0; a < depth - at; a++) {
1050 newblock = ext4_ext_new_meta_block(handle, inode, path,
1051 newext, &err, flags);
1054 ablocks[a] = newblock;
1057 /* initialize new leaf */
1058 newblock = ablocks[--a];
1059 if (unlikely(newblock == 0)) {
1060 EXT4_ERROR_INODE(inode, "newblock == 0!");
1061 err = -EFSCORRUPTED;
1064 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1065 if (unlikely(!bh)) {
1071 err = ext4_journal_get_create_access(handle, bh);
1075 neh = ext_block_hdr(bh);
1076 neh->eh_entries = 0;
1077 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1078 neh->eh_magic = EXT4_EXT_MAGIC;
1081 /* move remainder of path[depth] to the new leaf */
1082 if (unlikely(path[depth].p_hdr->eh_entries !=
1083 path[depth].p_hdr->eh_max)) {
1084 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1085 path[depth].p_hdr->eh_entries,
1086 path[depth].p_hdr->eh_max);
1087 err = -EFSCORRUPTED;
1090 /* start copy from next extent */
1091 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1092 ext4_ext_show_move(inode, path, newblock, depth);
1094 struct ext4_extent *ex;
1095 ex = EXT_FIRST_EXTENT(neh);
1096 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1097 le16_add_cpu(&neh->eh_entries, m);
1100 /* zero out unused area in the extent block */
1101 ext_size = sizeof(struct ext4_extent_header) +
1102 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1103 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1104 ext4_extent_block_csum_set(inode, neh);
1105 set_buffer_uptodate(bh);
1108 err = ext4_handle_dirty_metadata(handle, inode, bh);
1114 /* correct old leaf */
1116 err = ext4_ext_get_access(handle, inode, path + depth);
1119 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1120 err = ext4_ext_dirty(handle, inode, path + depth);
1126 /* create intermediate indexes */
1128 if (unlikely(k < 0)) {
1129 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1130 err = -EFSCORRUPTED;
1134 ext_debug("create %d intermediate indices\n", k);
1135 /* insert new index into current index block */
1136 /* current depth stored in i var */
1139 oldblock = newblock;
1140 newblock = ablocks[--a];
1141 bh = sb_getblk(inode->i_sb, newblock);
1142 if (unlikely(!bh)) {
1148 err = ext4_journal_get_create_access(handle, bh);
1152 neh = ext_block_hdr(bh);
1153 neh->eh_entries = cpu_to_le16(1);
1154 neh->eh_magic = EXT4_EXT_MAGIC;
1155 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1156 neh->eh_depth = cpu_to_le16(depth - i);
1157 fidx = EXT_FIRST_INDEX(neh);
1158 fidx->ei_block = border;
1159 ext4_idx_store_pblock(fidx, oldblock);
1161 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1162 i, newblock, le32_to_cpu(border), oldblock);
1164 /* move remainder of path[i] to the new index block */
1165 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1166 EXT_LAST_INDEX(path[i].p_hdr))) {
1167 EXT4_ERROR_INODE(inode,
1168 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1169 le32_to_cpu(path[i].p_ext->ee_block));
1170 err = -EFSCORRUPTED;
1173 /* start copy indexes */
1174 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1175 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1176 EXT_MAX_INDEX(path[i].p_hdr));
1177 ext4_ext_show_move(inode, path, newblock, i);
1179 memmove(++fidx, path[i].p_idx,
1180 sizeof(struct ext4_extent_idx) * m);
1181 le16_add_cpu(&neh->eh_entries, m);
1183 /* zero out unused area in the extent block */
1184 ext_size = sizeof(struct ext4_extent_header) +
1185 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1186 memset(bh->b_data + ext_size, 0,
1187 inode->i_sb->s_blocksize - ext_size);
1188 ext4_extent_block_csum_set(inode, neh);
1189 set_buffer_uptodate(bh);
1192 err = ext4_handle_dirty_metadata(handle, inode, bh);
1198 /* correct old index */
1200 err = ext4_ext_get_access(handle, inode, path + i);
1203 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1204 err = ext4_ext_dirty(handle, inode, path + i);
1212 /* insert new index */
1213 err = ext4_ext_insert_index(handle, inode, path + at,
1214 le32_to_cpu(border), newblock);
1218 if (buffer_locked(bh))
1224 /* free all allocated blocks in error case */
1225 for (i = 0; i < depth; i++) {
1228 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1229 EXT4_FREE_BLOCKS_METADATA);
1238 * ext4_ext_grow_indepth:
1239 * implements tree growing procedure:
1240 * - allocates new block
1241 * - moves top-level data (index block or leaf) into the new block
1242 * - initializes new top-level, creating index that points to the
1243 * just created block
1245 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1248 struct ext4_extent_header *neh;
1249 struct buffer_head *bh;
1250 ext4_fsblk_t newblock, goal = 0;
1251 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1253 size_t ext_size = 0;
1255 /* Try to prepend new index to old one */
1256 if (ext_depth(inode))
1257 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1258 if (goal > le32_to_cpu(es->s_first_data_block)) {
1259 flags |= EXT4_MB_HINT_TRY_GOAL;
1262 goal = ext4_inode_to_goal_block(inode);
1263 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1268 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1273 err = ext4_journal_get_create_access(handle, bh);
1279 ext_size = sizeof(EXT4_I(inode)->i_data);
1280 /* move top-level index/leaf into new block */
1281 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1282 /* zero out unused area in the extent block */
1283 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1285 /* set size of new block */
1286 neh = ext_block_hdr(bh);
1287 /* old root could have indexes or leaves
1288 * so calculate e_max right way */
1289 if (ext_depth(inode))
1290 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1292 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1293 neh->eh_magic = EXT4_EXT_MAGIC;
1294 ext4_extent_block_csum_set(inode, neh);
1295 set_buffer_uptodate(bh);
1298 err = ext4_handle_dirty_metadata(handle, inode, bh);
1302 /* Update top-level index: num,max,pointer */
1303 neh = ext_inode_hdr(inode);
1304 neh->eh_entries = cpu_to_le16(1);
1305 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1306 if (neh->eh_depth == 0) {
1307 /* Root extent block becomes index block */
1308 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1309 EXT_FIRST_INDEX(neh)->ei_block =
1310 EXT_FIRST_EXTENT(neh)->ee_block;
1312 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1313 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1314 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1315 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1317 le16_add_cpu(&neh->eh_depth, 1);
1318 ext4_mark_inode_dirty(handle, inode);
1326 * ext4_ext_create_new_leaf:
1327 * finds empty index and adds new leaf.
1328 * if no free index is found, then it requests in-depth growing.
1330 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1331 unsigned int mb_flags,
1332 unsigned int gb_flags,
1333 struct ext4_ext_path **ppath,
1334 struct ext4_extent *newext)
1336 struct ext4_ext_path *path = *ppath;
1337 struct ext4_ext_path *curp;
1338 int depth, i, err = 0;
1341 i = depth = ext_depth(inode);
1343 /* walk up to the tree and look for free index entry */
1344 curp = path + depth;
1345 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1350 /* we use already allocated block for index block,
1351 * so subsequent data blocks should be contiguous */
1352 if (EXT_HAS_FREE_INDEX(curp)) {
1353 /* if we found index with free entry, then use that
1354 * entry: create all needed subtree and add new leaf */
1355 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1360 path = ext4_find_extent(inode,
1361 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1364 err = PTR_ERR(path);
1366 /* tree is full, time to grow in depth */
1367 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1372 path = ext4_find_extent(inode,
1373 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1376 err = PTR_ERR(path);
1381 * only first (depth 0 -> 1) produces free space;
1382 * in all other cases we have to split the grown tree
1384 depth = ext_depth(inode);
1385 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1386 /* now we need to split */
1396 * search the closest allocated block to the left for *logical
1397 * and returns it at @logical + it's physical address at @phys
1398 * if *logical is the smallest allocated block, the function
1399 * returns 0 at @phys
1400 * return value contains 0 (success) or error code
1402 static int ext4_ext_search_left(struct inode *inode,
1403 struct ext4_ext_path *path,
1404 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1406 struct ext4_extent_idx *ix;
1407 struct ext4_extent *ex;
1410 if (unlikely(path == NULL)) {
1411 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1412 return -EFSCORRUPTED;
1414 depth = path->p_depth;
1417 if (depth == 0 && path->p_ext == NULL)
1420 /* usually extent in the path covers blocks smaller
1421 * then *logical, but it can be that extent is the
1422 * first one in the file */
1424 ex = path[depth].p_ext;
1425 ee_len = ext4_ext_get_actual_len(ex);
1426 if (*logical < le32_to_cpu(ex->ee_block)) {
1427 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1428 EXT4_ERROR_INODE(inode,
1429 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1430 *logical, le32_to_cpu(ex->ee_block));
1431 return -EFSCORRUPTED;
1433 while (--depth >= 0) {
1434 ix = path[depth].p_idx;
1435 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1436 EXT4_ERROR_INODE(inode,
1437 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1438 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1439 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1440 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1442 return -EFSCORRUPTED;
1448 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1449 EXT4_ERROR_INODE(inode,
1450 "logical %d < ee_block %d + ee_len %d!",
1451 *logical, le32_to_cpu(ex->ee_block), ee_len);
1452 return -EFSCORRUPTED;
1455 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1456 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1461 * search the closest allocated block to the right for *logical
1462 * and returns it at @logical + it's physical address at @phys
1463 * if *logical is the largest allocated block, the function
1464 * returns 0 at @phys
1465 * return value contains 0 (success) or error code
1467 static int ext4_ext_search_right(struct inode *inode,
1468 struct ext4_ext_path *path,
1469 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1470 struct ext4_extent **ret_ex)
1472 struct buffer_head *bh = NULL;
1473 struct ext4_extent_header *eh;
1474 struct ext4_extent_idx *ix;
1475 struct ext4_extent *ex;
1477 int depth; /* Note, NOT eh_depth; depth from top of tree */
1480 if (unlikely(path == NULL)) {
1481 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1482 return -EFSCORRUPTED;
1484 depth = path->p_depth;
1487 if (depth == 0 && path->p_ext == NULL)
1490 /* usually extent in the path covers blocks smaller
1491 * then *logical, but it can be that extent is the
1492 * first one in the file */
1494 ex = path[depth].p_ext;
1495 ee_len = ext4_ext_get_actual_len(ex);
1496 if (*logical < le32_to_cpu(ex->ee_block)) {
1497 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1498 EXT4_ERROR_INODE(inode,
1499 "first_extent(path[%d].p_hdr) != ex",
1501 return -EFSCORRUPTED;
1503 while (--depth >= 0) {
1504 ix = path[depth].p_idx;
1505 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1506 EXT4_ERROR_INODE(inode,
1507 "ix != EXT_FIRST_INDEX *logical %d!",
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 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1523 /* next allocated block in this leaf */
1528 /* go up and search for index to the right */
1529 while (--depth >= 0) {
1530 ix = path[depth].p_idx;
1531 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1535 /* we've gone up to the root and found no index to the right */
1539 /* we've found index to the right, let's
1540 * follow it and find the closest allocated
1541 * block to the right */
1543 block = ext4_idx_pblock(ix);
1544 while (++depth < path->p_depth) {
1545 /* subtract from p_depth to get proper eh_depth */
1546 bh = read_extent_tree_block(inode, block,
1547 path->p_depth - depth, 0);
1550 eh = ext_block_hdr(bh);
1551 ix = EXT_FIRST_INDEX(eh);
1552 block = ext4_idx_pblock(ix);
1556 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1559 eh = ext_block_hdr(bh);
1560 ex = EXT_FIRST_EXTENT(eh);
1562 *logical = le32_to_cpu(ex->ee_block);
1563 *phys = ext4_ext_pblock(ex);
1571 * ext4_ext_next_allocated_block:
1572 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1573 * NOTE: it considers block number from index entry as
1574 * allocated block. Thus, index entries have to be consistent
1578 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1582 BUG_ON(path == NULL);
1583 depth = path->p_depth;
1585 if (depth == 0 && path->p_ext == NULL)
1586 return EXT_MAX_BLOCKS;
1588 while (depth >= 0) {
1589 if (depth == path->p_depth) {
1591 if (path[depth].p_ext &&
1592 path[depth].p_ext !=
1593 EXT_LAST_EXTENT(path[depth].p_hdr))
1594 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1597 if (path[depth].p_idx !=
1598 EXT_LAST_INDEX(path[depth].p_hdr))
1599 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1604 return EXT_MAX_BLOCKS;
1608 * ext4_ext_next_leaf_block:
1609 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1611 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1615 BUG_ON(path == NULL);
1616 depth = path->p_depth;
1618 /* zero-tree has no leaf blocks at all */
1620 return EXT_MAX_BLOCKS;
1622 /* go to index block */
1625 while (depth >= 0) {
1626 if (path[depth].p_idx !=
1627 EXT_LAST_INDEX(path[depth].p_hdr))
1628 return (ext4_lblk_t)
1629 le32_to_cpu(path[depth].p_idx[1].ei_block);
1633 return EXT_MAX_BLOCKS;
1637 * ext4_ext_correct_indexes:
1638 * if leaf gets modified and modified extent is first in the leaf,
1639 * then we have to correct all indexes above.
1640 * TODO: do we need to correct tree in all cases?
1642 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1643 struct ext4_ext_path *path)
1645 struct ext4_extent_header *eh;
1646 int depth = ext_depth(inode);
1647 struct ext4_extent *ex;
1651 eh = path[depth].p_hdr;
1652 ex = path[depth].p_ext;
1654 if (unlikely(ex == NULL || eh == NULL)) {
1655 EXT4_ERROR_INODE(inode,
1656 "ex %p == NULL or eh %p == NULL", ex, eh);
1657 return -EFSCORRUPTED;
1661 /* there is no tree at all */
1665 if (ex != EXT_FIRST_EXTENT(eh)) {
1666 /* we correct tree if first leaf got modified only */
1671 * TODO: we need correction if border is smaller than current one
1674 border = path[depth].p_ext->ee_block;
1675 err = ext4_ext_get_access(handle, inode, path + k);
1678 path[k].p_idx->ei_block = border;
1679 err = ext4_ext_dirty(handle, inode, path + k);
1684 /* change all left-side indexes */
1685 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1687 err = ext4_ext_get_access(handle, inode, path + k);
1690 path[k].p_idx->ei_block = border;
1691 err = ext4_ext_dirty(handle, inode, path + k);
1700 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1701 struct ext4_extent *ex2)
1703 unsigned short ext1_ee_len, ext2_ee_len;
1705 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1708 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1709 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1711 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1712 le32_to_cpu(ex2->ee_block))
1716 * To allow future support for preallocated extents to be added
1717 * as an RO_COMPAT feature, refuse to merge to extents if
1718 * this can result in the top bit of ee_len being set.
1720 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1723 if (ext4_ext_is_unwritten(ex1) &&
1724 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1726 #ifdef AGGRESSIVE_TEST
1727 if (ext1_ee_len >= 4)
1731 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1737 * This function tries to merge the "ex" extent to the next extent in the tree.
1738 * It always tries to merge towards right. If you want to merge towards
1739 * left, pass "ex - 1" as argument instead of "ex".
1740 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1741 * 1 if they got merged.
1743 static int ext4_ext_try_to_merge_right(struct inode *inode,
1744 struct ext4_ext_path *path,
1745 struct ext4_extent *ex)
1747 struct ext4_extent_header *eh;
1748 unsigned int depth, len;
1749 int merge_done = 0, unwritten;
1751 depth = ext_depth(inode);
1752 BUG_ON(path[depth].p_hdr == NULL);
1753 eh = path[depth].p_hdr;
1755 while (ex < EXT_LAST_EXTENT(eh)) {
1756 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1758 /* merge with next extent! */
1759 unwritten = ext4_ext_is_unwritten(ex);
1760 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1761 + ext4_ext_get_actual_len(ex + 1));
1763 ext4_ext_mark_unwritten(ex);
1765 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1766 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1767 * sizeof(struct ext4_extent);
1768 memmove(ex + 1, ex + 2, len);
1770 le16_add_cpu(&eh->eh_entries, -1);
1772 WARN_ON(eh->eh_entries == 0);
1773 if (!eh->eh_entries)
1774 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1781 * This function does a very simple check to see if we can collapse
1782 * an extent tree with a single extent tree leaf block into the inode.
1784 static void ext4_ext_try_to_merge_up(handle_t *handle,
1785 struct inode *inode,
1786 struct ext4_ext_path *path)
1789 unsigned max_root = ext4_ext_space_root(inode, 0);
1792 if ((path[0].p_depth != 1) ||
1793 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1794 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1798 * We need to modify the block allocation bitmap and the block
1799 * group descriptor to release the extent tree block. If we
1800 * can't get the journal credits, give up.
1802 if (ext4_journal_extend(handle, 2,
1803 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1807 * Copy the extent data up to the inode
1809 blk = ext4_idx_pblock(path[0].p_idx);
1810 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1811 sizeof(struct ext4_extent_idx);
1812 s += sizeof(struct ext4_extent_header);
1814 path[1].p_maxdepth = path[0].p_maxdepth;
1815 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1816 path[0].p_depth = 0;
1817 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1818 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1819 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1821 brelse(path[1].p_bh);
1822 ext4_free_blocks(handle, inode, NULL, blk, 1,
1823 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1827 * This function tries to merge the @ex extent to neighbours in the tree.
1828 * return 1 if merge left else 0.
1830 static void ext4_ext_try_to_merge(handle_t *handle,
1831 struct inode *inode,
1832 struct ext4_ext_path *path,
1833 struct ext4_extent *ex) {
1834 struct ext4_extent_header *eh;
1838 depth = ext_depth(inode);
1839 BUG_ON(path[depth].p_hdr == NULL);
1840 eh = path[depth].p_hdr;
1842 if (ex > EXT_FIRST_EXTENT(eh))
1843 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1846 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1848 ext4_ext_try_to_merge_up(handle, inode, path);
1852 * check if a portion of the "newext" extent overlaps with an
1855 * If there is an overlap discovered, it updates the length of the newext
1856 * such that there will be no overlap, and then returns 1.
1857 * If there is no overlap found, it returns 0.
1859 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1860 struct inode *inode,
1861 struct ext4_extent *newext,
1862 struct ext4_ext_path *path)
1865 unsigned int depth, len1;
1866 unsigned int ret = 0;
1868 b1 = le32_to_cpu(newext->ee_block);
1869 len1 = ext4_ext_get_actual_len(newext);
1870 depth = ext_depth(inode);
1871 if (!path[depth].p_ext)
1873 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1876 * get the next allocated block if the extent in the path
1877 * is before the requested block(s)
1880 b2 = ext4_ext_next_allocated_block(path);
1881 if (b2 == EXT_MAX_BLOCKS)
1883 b2 = EXT4_LBLK_CMASK(sbi, b2);
1886 /* check for wrap through zero on extent logical start block*/
1887 if (b1 + len1 < b1) {
1888 len1 = EXT_MAX_BLOCKS - b1;
1889 newext->ee_len = cpu_to_le16(len1);
1893 /* check for overlap */
1894 if (b1 + len1 > b2) {
1895 newext->ee_len = cpu_to_le16(b2 - b1);
1903 * ext4_ext_insert_extent:
1904 * tries to merge requsted extent into the existing extent or
1905 * inserts requested extent as new one into the tree,
1906 * creating new leaf in the no-space case.
1908 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1909 struct ext4_ext_path **ppath,
1910 struct ext4_extent *newext, int gb_flags)
1912 struct ext4_ext_path *path = *ppath;
1913 struct ext4_extent_header *eh;
1914 struct ext4_extent *ex, *fex;
1915 struct ext4_extent *nearex; /* nearest extent */
1916 struct ext4_ext_path *npath = NULL;
1917 int depth, len, err;
1919 int mb_flags = 0, unwritten;
1921 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1922 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1923 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1924 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1925 return -EFSCORRUPTED;
1927 depth = ext_depth(inode);
1928 ex = path[depth].p_ext;
1929 eh = path[depth].p_hdr;
1930 if (unlikely(path[depth].p_hdr == NULL)) {
1931 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1932 return -EFSCORRUPTED;
1935 /* try to insert block into found extent and return */
1936 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1939 * Try to see whether we should rather test the extent on
1940 * right from ex, or from the left of ex. This is because
1941 * ext4_find_extent() can return either extent on the
1942 * left, or on the right from the searched position. This
1943 * will make merging more effective.
1945 if (ex < EXT_LAST_EXTENT(eh) &&
1946 (le32_to_cpu(ex->ee_block) +
1947 ext4_ext_get_actual_len(ex) <
1948 le32_to_cpu(newext->ee_block))) {
1951 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1952 (le32_to_cpu(newext->ee_block) +
1953 ext4_ext_get_actual_len(newext) <
1954 le32_to_cpu(ex->ee_block)))
1957 /* Try to append newex to the ex */
1958 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1959 ext_debug("append [%d]%d block to %u:[%d]%d"
1961 ext4_ext_is_unwritten(newext),
1962 ext4_ext_get_actual_len(newext),
1963 le32_to_cpu(ex->ee_block),
1964 ext4_ext_is_unwritten(ex),
1965 ext4_ext_get_actual_len(ex),
1966 ext4_ext_pblock(ex));
1967 err = ext4_ext_get_access(handle, inode,
1971 unwritten = ext4_ext_is_unwritten(ex);
1972 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1973 + ext4_ext_get_actual_len(newext));
1975 ext4_ext_mark_unwritten(ex);
1976 eh = path[depth].p_hdr;
1982 /* Try to prepend newex to the ex */
1983 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1984 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1986 le32_to_cpu(newext->ee_block),
1987 ext4_ext_is_unwritten(newext),
1988 ext4_ext_get_actual_len(newext),
1989 le32_to_cpu(ex->ee_block),
1990 ext4_ext_is_unwritten(ex),
1991 ext4_ext_get_actual_len(ex),
1992 ext4_ext_pblock(ex));
1993 err = ext4_ext_get_access(handle, inode,
1998 unwritten = ext4_ext_is_unwritten(ex);
1999 ex->ee_block = newext->ee_block;
2000 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2001 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2002 + ext4_ext_get_actual_len(newext));
2004 ext4_ext_mark_unwritten(ex);
2005 eh = path[depth].p_hdr;
2011 depth = ext_depth(inode);
2012 eh = path[depth].p_hdr;
2013 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2016 /* probably next leaf has space for us? */
2017 fex = EXT_LAST_EXTENT(eh);
2018 next = EXT_MAX_BLOCKS;
2019 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2020 next = ext4_ext_next_leaf_block(path);
2021 if (next != EXT_MAX_BLOCKS) {
2022 ext_debug("next leaf block - %u\n", next);
2023 BUG_ON(npath != NULL);
2024 npath = ext4_find_extent(inode, next, NULL, 0);
2026 return PTR_ERR(npath);
2027 BUG_ON(npath->p_depth != path->p_depth);
2028 eh = npath[depth].p_hdr;
2029 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2030 ext_debug("next leaf isn't full(%d)\n",
2031 le16_to_cpu(eh->eh_entries));
2035 ext_debug("next leaf has no free space(%d,%d)\n",
2036 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2040 * There is no free space in the found leaf.
2041 * We're gonna add a new leaf in the tree.
2043 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2044 mb_flags |= EXT4_MB_USE_RESERVED;
2045 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2049 depth = ext_depth(inode);
2050 eh = path[depth].p_hdr;
2053 nearex = path[depth].p_ext;
2055 err = ext4_ext_get_access(handle, inode, path + depth);
2060 /* there is no extent in this leaf, create first one */
2061 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2062 le32_to_cpu(newext->ee_block),
2063 ext4_ext_pblock(newext),
2064 ext4_ext_is_unwritten(newext),
2065 ext4_ext_get_actual_len(newext));
2066 nearex = EXT_FIRST_EXTENT(eh);
2068 if (le32_to_cpu(newext->ee_block)
2069 > le32_to_cpu(nearex->ee_block)) {
2071 ext_debug("insert %u:%llu:[%d]%d before: "
2073 le32_to_cpu(newext->ee_block),
2074 ext4_ext_pblock(newext),
2075 ext4_ext_is_unwritten(newext),
2076 ext4_ext_get_actual_len(newext),
2081 BUG_ON(newext->ee_block == nearex->ee_block);
2082 ext_debug("insert %u:%llu:[%d]%d after: "
2084 le32_to_cpu(newext->ee_block),
2085 ext4_ext_pblock(newext),
2086 ext4_ext_is_unwritten(newext),
2087 ext4_ext_get_actual_len(newext),
2090 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2092 ext_debug("insert %u:%llu:[%d]%d: "
2093 "move %d extents from 0x%p to 0x%p\n",
2094 le32_to_cpu(newext->ee_block),
2095 ext4_ext_pblock(newext),
2096 ext4_ext_is_unwritten(newext),
2097 ext4_ext_get_actual_len(newext),
2098 len, nearex, nearex + 1);
2099 memmove(nearex + 1, nearex,
2100 len * sizeof(struct ext4_extent));
2104 le16_add_cpu(&eh->eh_entries, 1);
2105 path[depth].p_ext = nearex;
2106 nearex->ee_block = newext->ee_block;
2107 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2108 nearex->ee_len = newext->ee_len;
2111 /* try to merge extents */
2112 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2113 ext4_ext_try_to_merge(handle, inode, path, nearex);
2116 /* time to correct all indexes above */
2117 err = ext4_ext_correct_indexes(handle, inode, path);
2121 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2124 ext4_ext_drop_refs(npath);
2129 static int ext4_fill_fiemap_extents(struct inode *inode,
2130 ext4_lblk_t block, ext4_lblk_t num,
2131 struct fiemap_extent_info *fieinfo)
2133 struct ext4_ext_path *path = NULL;
2134 struct ext4_extent *ex;
2135 struct extent_status es;
2136 ext4_lblk_t next, next_del, start = 0, end = 0;
2137 ext4_lblk_t last = block + num;
2138 int exists, depth = 0, err = 0;
2139 unsigned int flags = 0;
2140 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2142 while (block < last && block != EXT_MAX_BLOCKS) {
2144 /* find extent for this block */
2145 down_read(&EXT4_I(inode)->i_data_sem);
2147 path = ext4_find_extent(inode, block, &path, 0);
2149 up_read(&EXT4_I(inode)->i_data_sem);
2150 err = PTR_ERR(path);
2155 depth = ext_depth(inode);
2156 if (unlikely(path[depth].p_hdr == NULL)) {
2157 up_read(&EXT4_I(inode)->i_data_sem);
2158 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2159 err = -EFSCORRUPTED;
2162 ex = path[depth].p_ext;
2163 next = ext4_ext_next_allocated_block(path);
2168 /* there is no extent yet, so try to allocate
2169 * all requested space */
2172 } else if (le32_to_cpu(ex->ee_block) > block) {
2173 /* need to allocate space before found extent */
2175 end = le32_to_cpu(ex->ee_block);
2176 if (block + num < end)
2178 } else if (block >= le32_to_cpu(ex->ee_block)
2179 + ext4_ext_get_actual_len(ex)) {
2180 /* need to allocate space after found extent */
2185 } else if (block >= le32_to_cpu(ex->ee_block)) {
2187 * some part of requested space is covered
2191 end = le32_to_cpu(ex->ee_block)
2192 + ext4_ext_get_actual_len(ex);
2193 if (block + num < end)
2199 BUG_ON(end <= start);
2203 es.es_len = end - start;
2206 es.es_lblk = le32_to_cpu(ex->ee_block);
2207 es.es_len = ext4_ext_get_actual_len(ex);
2208 es.es_pblk = ext4_ext_pblock(ex);
2209 if (ext4_ext_is_unwritten(ex))
2210 flags |= FIEMAP_EXTENT_UNWRITTEN;
2214 * Find delayed extent and update es accordingly. We call
2215 * it even in !exists case to find out whether es is the
2216 * last existing extent or not.
2218 next_del = ext4_find_delayed_extent(inode, &es);
2219 if (!exists && next_del) {
2221 flags |= (FIEMAP_EXTENT_DELALLOC |
2222 FIEMAP_EXTENT_UNKNOWN);
2224 up_read(&EXT4_I(inode)->i_data_sem);
2226 if (unlikely(es.es_len == 0)) {
2227 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2228 err = -EFSCORRUPTED;
2233 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2234 * we need to check next == EXT_MAX_BLOCKS because it is
2235 * possible that an extent is with unwritten and delayed
2236 * status due to when an extent is delayed allocated and
2237 * is allocated by fallocate status tree will track both of
2240 * So we could return a unwritten and delayed extent, and
2241 * its block is equal to 'next'.
2243 if (next == next_del && next == EXT_MAX_BLOCKS) {
2244 flags |= FIEMAP_EXTENT_LAST;
2245 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2246 next != EXT_MAX_BLOCKS)) {
2247 EXT4_ERROR_INODE(inode,
2248 "next extent == %u, next "
2249 "delalloc extent = %u",
2251 err = -EFSCORRUPTED;
2257 err = fiemap_fill_next_extent(fieinfo,
2258 (__u64)es.es_lblk << blksize_bits,
2259 (__u64)es.es_pblk << blksize_bits,
2260 (__u64)es.es_len << blksize_bits,
2270 block = es.es_lblk + es.es_len;
2273 ext4_ext_drop_refs(path);
2278 static int ext4_fill_es_cache_info(struct inode *inode,
2279 ext4_lblk_t block, ext4_lblk_t num,
2280 struct fiemap_extent_info *fieinfo)
2282 ext4_lblk_t next, end = block + num - 1;
2283 struct extent_status es;
2284 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2288 while (block <= end) {
2291 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2293 if (ext4_es_is_unwritten(&es))
2294 flags |= FIEMAP_EXTENT_UNWRITTEN;
2295 if (ext4_es_is_delayed(&es))
2296 flags |= (FIEMAP_EXTENT_DELALLOC |
2297 FIEMAP_EXTENT_UNKNOWN);
2298 if (ext4_es_is_hole(&es))
2299 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2301 flags |= FIEMAP_EXTENT_LAST;
2302 if (flags & (FIEMAP_EXTENT_DELALLOC|
2303 EXT4_FIEMAP_EXTENT_HOLE))
2306 es.es_pblk = ext4_es_pblock(&es);
2307 err = fiemap_fill_next_extent(fieinfo,
2308 (__u64)es.es_lblk << blksize_bits,
2309 (__u64)es.es_pblk << blksize_bits,
2310 (__u64)es.es_len << blksize_bits,
2325 * ext4_ext_determine_hole - determine hole around given block
2326 * @inode: inode we lookup in
2327 * @path: path in extent tree to @lblk
2328 * @lblk: pointer to logical block around which we want to determine hole
2330 * Determine hole length (and start if easily possible) around given logical
2331 * block. We don't try too hard to find the beginning of the hole but @path
2332 * actually points to extent before @lblk, we provide it.
2334 * The function returns the length of a hole starting at @lblk. We update @lblk
2335 * to the beginning of the hole if we managed to find it.
2337 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2338 struct ext4_ext_path *path,
2341 int depth = ext_depth(inode);
2342 struct ext4_extent *ex;
2345 ex = path[depth].p_ext;
2347 /* there is no extent yet, so gap is [0;-] */
2349 len = EXT_MAX_BLOCKS;
2350 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2351 len = le32_to_cpu(ex->ee_block) - *lblk;
2352 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2353 + ext4_ext_get_actual_len(ex)) {
2356 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2357 next = ext4_ext_next_allocated_block(path);
2358 BUG_ON(next == *lblk);
2367 * ext4_ext_put_gap_in_cache:
2368 * calculate boundaries of the gap that the requested block fits into
2369 * and cache this gap
2372 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2373 ext4_lblk_t hole_len)
2375 struct extent_status es;
2377 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2378 hole_start + hole_len - 1, &es);
2380 /* There's delayed extent containing lblock? */
2381 if (es.es_lblk <= hole_start)
2383 hole_len = min(es.es_lblk - hole_start, hole_len);
2385 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2386 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2387 EXTENT_STATUS_HOLE);
2392 * removes index from the index block.
2394 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2395 struct ext4_ext_path *path, int depth)
2400 /* free index block */
2402 path = path + depth;
2403 leaf = ext4_idx_pblock(path->p_idx);
2404 if (unlikely(path->p_hdr->eh_entries == 0)) {
2405 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2406 return -EFSCORRUPTED;
2408 err = ext4_ext_get_access(handle, inode, path);
2412 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2413 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2414 len *= sizeof(struct ext4_extent_idx);
2415 memmove(path->p_idx, path->p_idx + 1, len);
2418 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2419 err = ext4_ext_dirty(handle, inode, path);
2422 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2423 trace_ext4_ext_rm_idx(inode, leaf);
2425 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2426 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2428 while (--depth >= 0) {
2429 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2432 err = ext4_ext_get_access(handle, inode, path);
2435 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2436 err = ext4_ext_dirty(handle, inode, path);
2444 * ext4_ext_calc_credits_for_single_extent:
2445 * This routine returns max. credits that needed to insert an extent
2446 * to the extent tree.
2447 * When pass the actual path, the caller should calculate credits
2450 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2451 struct ext4_ext_path *path)
2454 int depth = ext_depth(inode);
2457 /* probably there is space in leaf? */
2458 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2459 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2462 * There are some space in the leaf tree, no
2463 * need to account for leaf block credit
2465 * bitmaps and block group descriptor blocks
2466 * and other metadata blocks still need to be
2469 /* 1 bitmap, 1 block group descriptor */
2470 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2475 return ext4_chunk_trans_blocks(inode, nrblocks);
2479 * How many index/leaf blocks need to change/allocate to add @extents extents?
2481 * If we add a single extent, then in the worse case, each tree level
2482 * index/leaf need to be changed in case of the tree split.
2484 * If more extents are inserted, they could cause the whole tree split more
2485 * than once, but this is really rare.
2487 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2492 /* If we are converting the inline data, only one is needed here. */
2493 if (ext4_has_inline_data(inode))
2496 depth = ext_depth(inode);
2506 static inline int get_default_free_blocks_flags(struct inode *inode)
2508 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2509 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2510 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2511 else if (ext4_should_journal_data(inode))
2512 return EXT4_FREE_BLOCKS_FORGET;
2517 * ext4_rereserve_cluster - increment the reserved cluster count when
2518 * freeing a cluster with a pending reservation
2520 * @inode - file containing the cluster
2521 * @lblk - logical block in cluster to be reserved
2523 * Increments the reserved cluster count and adjusts quota in a bigalloc
2524 * file system when freeing a partial cluster containing at least one
2525 * delayed and unwritten block. A partial cluster meeting that
2526 * requirement will have a pending reservation. If so, the
2527 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2528 * defer reserved and allocated space accounting to a subsequent call
2531 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2533 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2534 struct ext4_inode_info *ei = EXT4_I(inode);
2536 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2538 spin_lock(&ei->i_block_reservation_lock);
2539 ei->i_reserved_data_blocks++;
2540 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2541 spin_unlock(&ei->i_block_reservation_lock);
2543 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2544 ext4_remove_pending(inode, lblk);
2547 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2548 struct ext4_extent *ex,
2549 struct partial_cluster *partial,
2550 ext4_lblk_t from, ext4_lblk_t to)
2552 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2553 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2554 ext4_fsblk_t last_pblk, pblk;
2558 /* only extent tail removal is allowed */
2559 if (from < le32_to_cpu(ex->ee_block) ||
2560 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2561 ext4_error(sbi->s_sb,
2562 "strange request: removal(2) %u-%u from %u:%u",
2563 from, to, le32_to_cpu(ex->ee_block), ee_len);
2567 #ifdef EXTENTS_STATS
2568 spin_lock(&sbi->s_ext_stats_lock);
2569 sbi->s_ext_blocks += ee_len;
2570 sbi->s_ext_extents++;
2571 if (ee_len < sbi->s_ext_min)
2572 sbi->s_ext_min = ee_len;
2573 if (ee_len > sbi->s_ext_max)
2574 sbi->s_ext_max = ee_len;
2575 if (ext_depth(inode) > sbi->s_depth_max)
2576 sbi->s_depth_max = ext_depth(inode);
2577 spin_unlock(&sbi->s_ext_stats_lock);
2580 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2583 * if we have a partial cluster, and it's different from the
2584 * cluster of the last block in the extent, we free it
2586 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2588 if (partial->state != initial &&
2589 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2590 if (partial->state == tofree) {
2591 flags = get_default_free_blocks_flags(inode);
2592 if (ext4_is_pending(inode, partial->lblk))
2593 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2594 ext4_free_blocks(handle, inode, NULL,
2595 EXT4_C2B(sbi, partial->pclu),
2596 sbi->s_cluster_ratio, flags);
2597 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2598 ext4_rereserve_cluster(inode, partial->lblk);
2600 partial->state = initial;
2603 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2604 pblk = ext4_ext_pblock(ex) + ee_len - num;
2607 * We free the partial cluster at the end of the extent (if any),
2608 * unless the cluster is used by another extent (partial_cluster
2609 * state is nofree). If a partial cluster exists here, it must be
2610 * shared with the last block in the extent.
2612 flags = get_default_free_blocks_flags(inode);
2614 /* partial, left end cluster aligned, right end unaligned */
2615 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2616 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2617 (partial->state != nofree)) {
2618 if (ext4_is_pending(inode, to))
2619 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2620 ext4_free_blocks(handle, inode, NULL,
2621 EXT4_PBLK_CMASK(sbi, last_pblk),
2622 sbi->s_cluster_ratio, flags);
2623 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2624 ext4_rereserve_cluster(inode, to);
2625 partial->state = initial;
2626 flags = get_default_free_blocks_flags(inode);
2629 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2632 * For bigalloc file systems, we never free a partial cluster
2633 * at the beginning of the extent. Instead, we check to see if we
2634 * need to free it on a subsequent call to ext4_remove_blocks,
2635 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2637 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2638 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2640 /* reset the partial cluster if we've freed past it */
2641 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2642 partial->state = initial;
2645 * If we've freed the entire extent but the beginning is not left
2646 * cluster aligned and is not marked as ineligible for freeing we
2647 * record the partial cluster at the beginning of the extent. It
2648 * wasn't freed by the preceding ext4_free_blocks() call, and we
2649 * need to look farther to the left to determine if it's to be freed
2650 * (not shared with another extent). Else, reset the partial
2651 * cluster - we're either done freeing or the beginning of the
2652 * extent is left cluster aligned.
2654 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2655 if (partial->state == initial) {
2656 partial->pclu = EXT4_B2C(sbi, pblk);
2657 partial->lblk = from;
2658 partial->state = tofree;
2661 partial->state = initial;
2668 * ext4_ext_rm_leaf() Removes the extents associated with the
2669 * blocks appearing between "start" and "end". Both "start"
2670 * and "end" must appear in the same extent or EIO is returned.
2672 * @handle: The journal handle
2673 * @inode: The files inode
2674 * @path: The path to the leaf
2675 * @partial_cluster: The cluster which we'll have to free if all extents
2676 * has been released from it. However, if this value is
2677 * negative, it's a cluster just to the right of the
2678 * punched region and it must not be freed.
2679 * @start: The first block to remove
2680 * @end: The last block to remove
2683 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2684 struct ext4_ext_path *path,
2685 struct partial_cluster *partial,
2686 ext4_lblk_t start, ext4_lblk_t end)
2688 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2689 int err = 0, correct_index = 0;
2690 int depth = ext_depth(inode), credits, revoke_credits;
2691 struct ext4_extent_header *eh;
2694 ext4_lblk_t ex_ee_block;
2695 unsigned short ex_ee_len;
2696 unsigned unwritten = 0;
2697 struct ext4_extent *ex;
2700 /* the header must be checked already in ext4_ext_remove_space() */
2701 ext_debug("truncate since %u in leaf to %u\n", start, end);
2702 if (!path[depth].p_hdr)
2703 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2704 eh = path[depth].p_hdr;
2705 if (unlikely(path[depth].p_hdr == NULL)) {
2706 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2707 return -EFSCORRUPTED;
2709 /* find where to start removing */
2710 ex = path[depth].p_ext;
2712 ex = EXT_LAST_EXTENT(eh);
2714 ex_ee_block = le32_to_cpu(ex->ee_block);
2715 ex_ee_len = ext4_ext_get_actual_len(ex);
2717 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2719 while (ex >= EXT_FIRST_EXTENT(eh) &&
2720 ex_ee_block + ex_ee_len > start) {
2722 if (ext4_ext_is_unwritten(ex))
2727 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2728 unwritten, ex_ee_len);
2729 path[depth].p_ext = ex;
2731 a = ex_ee_block > start ? ex_ee_block : start;
2732 b = ex_ee_block+ex_ee_len - 1 < end ?
2733 ex_ee_block+ex_ee_len - 1 : end;
2735 ext_debug(" border %u:%u\n", a, b);
2737 /* If this extent is beyond the end of the hole, skip it */
2738 if (end < ex_ee_block) {
2740 * We're going to skip this extent and move to another,
2741 * so note that its first cluster is in use to avoid
2742 * freeing it when removing blocks. Eventually, the
2743 * right edge of the truncated/punched region will
2744 * be just to the left.
2746 if (sbi->s_cluster_ratio > 1) {
2747 pblk = ext4_ext_pblock(ex);
2748 partial->pclu = EXT4_B2C(sbi, pblk);
2749 partial->state = nofree;
2752 ex_ee_block = le32_to_cpu(ex->ee_block);
2753 ex_ee_len = ext4_ext_get_actual_len(ex);
2755 } else if (b != ex_ee_block + ex_ee_len - 1) {
2756 EXT4_ERROR_INODE(inode,
2757 "can not handle truncate %u:%u "
2759 start, end, ex_ee_block,
2760 ex_ee_block + ex_ee_len - 1);
2761 err = -EFSCORRUPTED;
2763 } else if (a != ex_ee_block) {
2764 /* remove tail of the extent */
2765 num = a - ex_ee_block;
2767 /* remove whole extent: excellent! */
2771 * 3 for leaf, sb, and inode plus 2 (bmap and group
2772 * descriptor) for each block group; assume two block
2773 * groups plus ex_ee_len/blocks_per_block_group for
2776 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2777 if (ex == EXT_FIRST_EXTENT(eh)) {
2779 credits += (ext_depth(inode)) + 1;
2781 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2783 * We may end up freeing some index blocks and data from the
2784 * punched range. Note that partial clusters are accounted for
2785 * by ext4_free_data_revoke_credits().
2788 ext4_free_metadata_revoke_credits(inode->i_sb,
2790 ext4_free_data_revoke_credits(inode, b - a + 1);
2792 err = ext4_datasem_ensure_credits(handle, inode, credits,
2793 credits, revoke_credits);
2800 err = ext4_ext_get_access(handle, inode, path + depth);
2804 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2809 /* this extent is removed; mark slot entirely unused */
2810 ext4_ext_store_pblock(ex, 0);
2812 ex->ee_len = cpu_to_le16(num);
2814 * Do not mark unwritten if all the blocks in the
2815 * extent have been removed.
2817 if (unwritten && num)
2818 ext4_ext_mark_unwritten(ex);
2820 * If the extent was completely released,
2821 * we need to remove it from the leaf
2824 if (end != EXT_MAX_BLOCKS - 1) {
2826 * For hole punching, we need to scoot all the
2827 * extents up when an extent is removed so that
2828 * we dont have blank extents in the middle
2830 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2831 sizeof(struct ext4_extent));
2833 /* Now get rid of the one at the end */
2834 memset(EXT_LAST_EXTENT(eh), 0,
2835 sizeof(struct ext4_extent));
2837 le16_add_cpu(&eh->eh_entries, -1);
2840 err = ext4_ext_dirty(handle, inode, path + depth);
2844 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2845 ext4_ext_pblock(ex));
2847 ex_ee_block = le32_to_cpu(ex->ee_block);
2848 ex_ee_len = ext4_ext_get_actual_len(ex);
2851 if (correct_index && eh->eh_entries)
2852 err = ext4_ext_correct_indexes(handle, inode, path);
2855 * If there's a partial cluster and at least one extent remains in
2856 * the leaf, free the partial cluster if it isn't shared with the
2857 * current extent. If it is shared with the current extent
2858 * we reset the partial cluster because we've reached the start of the
2859 * truncated/punched region and we're done removing blocks.
2861 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2862 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2863 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2864 int flags = get_default_free_blocks_flags(inode);
2866 if (ext4_is_pending(inode, partial->lblk))
2867 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2868 ext4_free_blocks(handle, inode, NULL,
2869 EXT4_C2B(sbi, partial->pclu),
2870 sbi->s_cluster_ratio, flags);
2871 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2872 ext4_rereserve_cluster(inode, partial->lblk);
2874 partial->state = initial;
2877 /* if this leaf is free, then we should
2878 * remove it from index block above */
2879 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2880 err = ext4_ext_rm_idx(handle, inode, path, depth);
2887 * ext4_ext_more_to_rm:
2888 * returns 1 if current index has to be freed (even partial)
2891 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2893 BUG_ON(path->p_idx == NULL);
2895 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2899 * if truncate on deeper level happened, it wasn't partial,
2900 * so we have to consider current index for truncation
2902 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2907 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2910 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2911 int depth = ext_depth(inode);
2912 struct ext4_ext_path *path = NULL;
2913 struct partial_cluster partial;
2919 partial.state = initial;
2921 ext_debug("truncate since %u to %u\n", start, end);
2923 /* probably first extent we're gonna free will be last in block */
2924 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2926 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2928 return PTR_ERR(handle);
2931 trace_ext4_ext_remove_space(inode, start, end, depth);
2934 * Check if we are removing extents inside the extent tree. If that
2935 * is the case, we are going to punch a hole inside the extent tree
2936 * so we have to check whether we need to split the extent covering
2937 * the last block to remove so we can easily remove the part of it
2938 * in ext4_ext_rm_leaf().
2940 if (end < EXT_MAX_BLOCKS - 1) {
2941 struct ext4_extent *ex;
2942 ext4_lblk_t ee_block, ex_end, lblk;
2945 /* find extent for or closest extent to this block */
2946 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2948 ext4_journal_stop(handle);
2949 return PTR_ERR(path);
2951 depth = ext_depth(inode);
2952 /* Leaf not may not exist only if inode has no blocks at all */
2953 ex = path[depth].p_ext;
2956 EXT4_ERROR_INODE(inode,
2957 "path[%d].p_hdr == NULL",
2959 err = -EFSCORRUPTED;
2964 ee_block = le32_to_cpu(ex->ee_block);
2965 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2968 * See if the last block is inside the extent, if so split
2969 * the extent at 'end' block so we can easily remove the
2970 * tail of the first part of the split extent in
2971 * ext4_ext_rm_leaf().
2973 if (end >= ee_block && end < ex_end) {
2976 * If we're going to split the extent, note that
2977 * the cluster containing the block after 'end' is
2978 * in use to avoid freeing it when removing blocks.
2980 if (sbi->s_cluster_ratio > 1) {
2981 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2982 partial.pclu = EXT4_B2C(sbi, pblk);
2983 partial.state = nofree;
2987 * Split the extent in two so that 'end' is the last
2988 * block in the first new extent. Also we should not
2989 * fail removing space due to ENOSPC so try to use
2990 * reserved block if that happens.
2992 err = ext4_force_split_extent_at(handle, inode, &path,
2997 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2998 partial.state == initial) {
3000 * If we're punching, there's an extent to the right.
3001 * If the partial cluster hasn't been set, set it to
3002 * that extent's first cluster and its state to nofree
3003 * so it won't be freed should it contain blocks to be
3004 * removed. If it's already set (tofree/nofree), we're
3005 * retrying and keep the original partial cluster info
3006 * so a cluster marked tofree as a result of earlier
3007 * extent removal is not lost.
3010 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3015 partial.pclu = EXT4_B2C(sbi, pblk);
3016 partial.state = nofree;
3021 * We start scanning from right side, freeing all the blocks
3022 * after i_size and walking into the tree depth-wise.
3024 depth = ext_depth(inode);
3029 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3031 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3034 ext4_journal_stop(handle);
3037 path[0].p_maxdepth = path[0].p_depth = depth;
3038 path[0].p_hdr = ext_inode_hdr(inode);
3041 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3042 err = -EFSCORRUPTED;
3048 while (i >= 0 && err == 0) {
3050 /* this is leaf block */
3051 err = ext4_ext_rm_leaf(handle, inode, path,
3052 &partial, start, end);
3053 /* root level has p_bh == NULL, brelse() eats this */
3054 brelse(path[i].p_bh);
3055 path[i].p_bh = NULL;
3060 /* this is index block */
3061 if (!path[i].p_hdr) {
3062 ext_debug("initialize header\n");
3063 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3066 if (!path[i].p_idx) {
3067 /* this level hasn't been touched yet */
3068 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3069 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3070 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3072 le16_to_cpu(path[i].p_hdr->eh_entries));
3074 /* we were already here, see at next index */
3078 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3079 i, EXT_FIRST_INDEX(path[i].p_hdr),
3081 if (ext4_ext_more_to_rm(path + i)) {
3082 struct buffer_head *bh;
3083 /* go to the next level */
3084 ext_debug("move to level %d (block %llu)\n",
3085 i + 1, ext4_idx_pblock(path[i].p_idx));
3086 memset(path + i + 1, 0, sizeof(*path));
3087 bh = read_extent_tree_block(inode,
3088 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3091 /* should we reset i_size? */
3095 /* Yield here to deal with large extent trees.
3096 * Should be a no-op if we did IO above. */
3098 if (WARN_ON(i + 1 > depth)) {
3099 err = -EFSCORRUPTED;
3102 path[i + 1].p_bh = bh;
3104 /* save actual number of indexes since this
3105 * number is changed at the next iteration */
3106 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3109 /* we finished processing this index, go up */
3110 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3111 /* index is empty, remove it;
3112 * handle must be already prepared by the
3113 * truncatei_leaf() */
3114 err = ext4_ext_rm_idx(handle, inode, path, i);
3116 /* root level has p_bh == NULL, brelse() eats this */
3117 brelse(path[i].p_bh);
3118 path[i].p_bh = NULL;
3120 ext_debug("return to level %d\n", i);
3124 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3125 path->p_hdr->eh_entries);
3128 * if there's a partial cluster and we have removed the first extent
3129 * in the file, then we also free the partial cluster, if any
3131 if (partial.state == tofree && err == 0) {
3132 int flags = get_default_free_blocks_flags(inode);
3134 if (ext4_is_pending(inode, partial.lblk))
3135 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3136 ext4_free_blocks(handle, inode, NULL,
3137 EXT4_C2B(sbi, partial.pclu),
3138 sbi->s_cluster_ratio, flags);
3139 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3140 ext4_rereserve_cluster(inode, partial.lblk);
3141 partial.state = initial;
3144 /* TODO: flexible tree reduction should be here */
3145 if (path->p_hdr->eh_entries == 0) {
3147 * truncate to zero freed all the tree,
3148 * so we need to correct eh_depth
3150 err = ext4_ext_get_access(handle, inode, path);
3152 ext_inode_hdr(inode)->eh_depth = 0;
3153 ext_inode_hdr(inode)->eh_max =
3154 cpu_to_le16(ext4_ext_space_root(inode, 0));
3155 err = ext4_ext_dirty(handle, inode, path);
3159 ext4_ext_drop_refs(path);
3164 ext4_journal_stop(handle);
3170 * called at mount time
3172 void ext4_ext_init(struct super_block *sb)
3175 * possible initialization would be here
3178 if (ext4_has_feature_extents(sb)) {
3179 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3180 printk(KERN_INFO "EXT4-fs: file extents enabled"
3181 #ifdef AGGRESSIVE_TEST
3182 ", aggressive tests"
3184 #ifdef CHECK_BINSEARCH
3187 #ifdef EXTENTS_STATS
3192 #ifdef EXTENTS_STATS
3193 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3194 EXT4_SB(sb)->s_ext_min = 1 << 30;
3195 EXT4_SB(sb)->s_ext_max = 0;
3201 * called at umount time
3203 void ext4_ext_release(struct super_block *sb)
3205 if (!ext4_has_feature_extents(sb))
3208 #ifdef EXTENTS_STATS
3209 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3210 struct ext4_sb_info *sbi = EXT4_SB(sb);
3211 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3212 sbi->s_ext_blocks, sbi->s_ext_extents,
3213 sbi->s_ext_blocks / sbi->s_ext_extents);
3214 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3215 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3220 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3222 ext4_lblk_t ee_block;
3223 ext4_fsblk_t ee_pblock;
3224 unsigned int ee_len;
3226 ee_block = le32_to_cpu(ex->ee_block);
3227 ee_len = ext4_ext_get_actual_len(ex);
3228 ee_pblock = ext4_ext_pblock(ex);
3233 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3234 EXTENT_STATUS_WRITTEN);
3237 /* FIXME!! we need to try to merge to left or right after zero-out */
3238 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3240 ext4_fsblk_t ee_pblock;
3241 unsigned int ee_len;
3243 ee_len = ext4_ext_get_actual_len(ex);
3244 ee_pblock = ext4_ext_pblock(ex);
3245 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3250 * ext4_split_extent_at() splits an extent at given block.
3252 * @handle: the journal handle
3253 * @inode: the file inode
3254 * @path: the path to the extent
3255 * @split: the logical block where the extent is splitted.
3256 * @split_flags: indicates if the extent could be zeroout if split fails, and
3257 * the states(init or unwritten) of new extents.
3258 * @flags: flags used to insert new extent to extent tree.
3261 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3262 * of which are deterimined by split_flag.
3264 * There are two cases:
3265 * a> the extent are splitted into two extent.
3266 * b> split is not needed, and just mark the extent.
3268 * return 0 on success.
3270 static int ext4_split_extent_at(handle_t *handle,
3271 struct inode *inode,
3272 struct ext4_ext_path **ppath,
3277 struct ext4_ext_path *path = *ppath;
3278 ext4_fsblk_t newblock;
3279 ext4_lblk_t ee_block;
3280 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3281 struct ext4_extent *ex2 = NULL;
3282 unsigned int ee_len, depth;
3285 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3286 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3288 ext_debug("ext4_split_extents_at: inode %lu, logical"
3289 "block %llu\n", inode->i_ino, (unsigned long long)split);
3291 ext4_ext_show_leaf(inode, path);
3293 depth = ext_depth(inode);
3294 ex = path[depth].p_ext;
3295 ee_block = le32_to_cpu(ex->ee_block);
3296 ee_len = ext4_ext_get_actual_len(ex);
3297 newblock = split - ee_block + ext4_ext_pblock(ex);
3299 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3300 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3301 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3302 EXT4_EXT_MARK_UNWRIT1 |
3303 EXT4_EXT_MARK_UNWRIT2));
3305 err = ext4_ext_get_access(handle, inode, path + depth);
3309 if (split == ee_block) {
3311 * case b: block @split is the block that the extent begins with
3312 * then we just change the state of the extent, and splitting
3315 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3316 ext4_ext_mark_unwritten(ex);
3318 ext4_ext_mark_initialized(ex);
3320 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3321 ext4_ext_try_to_merge(handle, inode, path, ex);
3323 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3328 memcpy(&orig_ex, ex, sizeof(orig_ex));
3329 ex->ee_len = cpu_to_le16(split - ee_block);
3330 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3331 ext4_ext_mark_unwritten(ex);
3334 * path may lead to new leaf, not to original leaf any more
3335 * after ext4_ext_insert_extent() returns,
3337 err = ext4_ext_dirty(handle, inode, path + depth);
3339 goto fix_extent_len;
3342 ex2->ee_block = cpu_to_le32(split);
3343 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3344 ext4_ext_store_pblock(ex2, newblock);
3345 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3346 ext4_ext_mark_unwritten(ex2);
3348 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3349 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3350 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3351 if (split_flag & EXT4_EXT_DATA_VALID1) {
3352 err = ext4_ext_zeroout(inode, ex2);
3353 zero_ex.ee_block = ex2->ee_block;
3354 zero_ex.ee_len = cpu_to_le16(
3355 ext4_ext_get_actual_len(ex2));
3356 ext4_ext_store_pblock(&zero_ex,
3357 ext4_ext_pblock(ex2));
3359 err = ext4_ext_zeroout(inode, ex);
3360 zero_ex.ee_block = ex->ee_block;
3361 zero_ex.ee_len = cpu_to_le16(
3362 ext4_ext_get_actual_len(ex));
3363 ext4_ext_store_pblock(&zero_ex,
3364 ext4_ext_pblock(ex));
3367 err = ext4_ext_zeroout(inode, &orig_ex);
3368 zero_ex.ee_block = orig_ex.ee_block;
3369 zero_ex.ee_len = cpu_to_le16(
3370 ext4_ext_get_actual_len(&orig_ex));
3371 ext4_ext_store_pblock(&zero_ex,
3372 ext4_ext_pblock(&orig_ex));
3376 goto fix_extent_len;
3377 /* update the extent length and mark as initialized */
3378 ex->ee_len = cpu_to_le16(ee_len);
3379 ext4_ext_try_to_merge(handle, inode, path, ex);
3380 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3382 goto fix_extent_len;
3384 /* update extent status tree */
3385 err = ext4_zeroout_es(inode, &zero_ex);
3389 goto fix_extent_len;
3392 ext4_ext_show_leaf(inode, path);
3396 ex->ee_len = orig_ex.ee_len;
3397 ext4_ext_dirty(handle, inode, path + path->p_depth);
3402 * ext4_split_extents() splits an extent and mark extent which is covered
3403 * by @map as split_flags indicates
3405 * It may result in splitting the extent into multiple extents (up to three)
3406 * There are three possibilities:
3407 * a> There is no split required
3408 * b> Splits in two extents: Split is happening at either end of the extent
3409 * c> Splits in three extents: Somone is splitting in middle of the extent
3412 static int ext4_split_extent(handle_t *handle,
3413 struct inode *inode,
3414 struct ext4_ext_path **ppath,
3415 struct ext4_map_blocks *map,
3419 struct ext4_ext_path *path = *ppath;
3420 ext4_lblk_t ee_block;
3421 struct ext4_extent *ex;
3422 unsigned int ee_len, depth;
3425 int split_flag1, flags1;
3426 int allocated = map->m_len;
3428 depth = ext_depth(inode);
3429 ex = path[depth].p_ext;
3430 ee_block = le32_to_cpu(ex->ee_block);
3431 ee_len = ext4_ext_get_actual_len(ex);
3432 unwritten = ext4_ext_is_unwritten(ex);
3434 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3435 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3436 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3438 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3439 EXT4_EXT_MARK_UNWRIT2;
3440 if (split_flag & EXT4_EXT_DATA_VALID2)
3441 split_flag1 |= EXT4_EXT_DATA_VALID1;
3442 err = ext4_split_extent_at(handle, inode, ppath,
3443 map->m_lblk + map->m_len, split_flag1, flags1);
3447 allocated = ee_len - (map->m_lblk - ee_block);
3450 * Update path is required because previous ext4_split_extent_at() may
3451 * result in split of original leaf or extent zeroout.
3453 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3455 return PTR_ERR(path);
3456 depth = ext_depth(inode);
3457 ex = path[depth].p_ext;
3459 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3460 (unsigned long) map->m_lblk);
3461 return -EFSCORRUPTED;
3463 unwritten = ext4_ext_is_unwritten(ex);
3466 if (map->m_lblk >= ee_block) {
3467 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3469 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3470 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3471 EXT4_EXT_MARK_UNWRIT2);
3473 err = ext4_split_extent_at(handle, inode, ppath,
3474 map->m_lblk, split_flag1, flags);
3479 ext4_ext_show_leaf(inode, path);
3481 return err ? err : allocated;
3485 * This function is called by ext4_ext_map_blocks() if someone tries to write
3486 * to an unwritten extent. It may result in splitting the unwritten
3487 * extent into multiple extents (up to three - one initialized and two
3489 * There are three possibilities:
3490 * a> There is no split required: Entire extent should be initialized
3491 * b> Splits in two extents: Write is happening at either end of the extent
3492 * c> Splits in three extents: Somone is writing in middle of the extent
3495 * - The extent pointed to by 'path' is unwritten.
3496 * - The extent pointed to by 'path' contains a superset
3497 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3499 * Post-conditions on success:
3500 * - the returned value is the number of blocks beyond map->l_lblk
3501 * that are allocated and initialized.
3502 * It is guaranteed to be >= map->m_len.
3504 static int ext4_ext_convert_to_initialized(handle_t *handle,
3505 struct inode *inode,
3506 struct ext4_map_blocks *map,
3507 struct ext4_ext_path **ppath,
3510 struct ext4_ext_path *path = *ppath;
3511 struct ext4_sb_info *sbi;
3512 struct ext4_extent_header *eh;
3513 struct ext4_map_blocks split_map;
3514 struct ext4_extent zero_ex1, zero_ex2;
3515 struct ext4_extent *ex, *abut_ex;
3516 ext4_lblk_t ee_block, eof_block;
3517 unsigned int ee_len, depth, map_len = map->m_len;
3518 int allocated = 0, max_zeroout = 0;
3520 int split_flag = EXT4_EXT_DATA_VALID2;
3522 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3523 "block %llu, max_blocks %u\n", inode->i_ino,
3524 (unsigned long long)map->m_lblk, map_len);
3526 sbi = EXT4_SB(inode->i_sb);
3527 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3528 inode->i_sb->s_blocksize_bits;
3529 if (eof_block < map->m_lblk + map_len)
3530 eof_block = map->m_lblk + map_len;
3532 depth = ext_depth(inode);
3533 eh = path[depth].p_hdr;
3534 ex = path[depth].p_ext;
3535 ee_block = le32_to_cpu(ex->ee_block);
3536 ee_len = ext4_ext_get_actual_len(ex);
3537 zero_ex1.ee_len = 0;
3538 zero_ex2.ee_len = 0;
3540 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3542 /* Pre-conditions */
3543 BUG_ON(!ext4_ext_is_unwritten(ex));
3544 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3547 * Attempt to transfer newly initialized blocks from the currently
3548 * unwritten extent to its neighbor. This is much cheaper
3549 * than an insertion followed by a merge as those involve costly
3550 * memmove() calls. Transferring to the left is the common case in
3551 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3552 * followed by append writes.
3554 * Limitations of the current logic:
3555 * - L1: we do not deal with writes covering the whole extent.
3556 * This would require removing the extent if the transfer
3558 * - L2: we only attempt to merge with an extent stored in the
3559 * same extent tree node.
3561 if ((map->m_lblk == ee_block) &&
3562 /* See if we can merge left */
3563 (map_len < ee_len) && /*L1*/
3564 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3565 ext4_lblk_t prev_lblk;
3566 ext4_fsblk_t prev_pblk, ee_pblk;
3567 unsigned int prev_len;
3570 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3571 prev_len = ext4_ext_get_actual_len(abut_ex);
3572 prev_pblk = ext4_ext_pblock(abut_ex);
3573 ee_pblk = ext4_ext_pblock(ex);
3576 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3577 * upon those conditions:
3578 * - C1: abut_ex is initialized,
3579 * - C2: abut_ex is logically abutting ex,
3580 * - C3: abut_ex is physically abutting ex,
3581 * - C4: abut_ex can receive the additional blocks without
3582 * overflowing the (initialized) length limit.
3584 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3585 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3586 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3587 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3588 err = ext4_ext_get_access(handle, inode, path + depth);
3592 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3595 /* Shift the start of ex by 'map_len' blocks */
3596 ex->ee_block = cpu_to_le32(ee_block + map_len);
3597 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3598 ex->ee_len = cpu_to_le16(ee_len - map_len);
3599 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3601 /* Extend abut_ex by 'map_len' blocks */
3602 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3604 /* Result: number of initialized blocks past m_lblk */
3605 allocated = map_len;
3607 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3608 (map_len < ee_len) && /*L1*/
3609 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3610 /* See if we can merge right */
3611 ext4_lblk_t next_lblk;
3612 ext4_fsblk_t next_pblk, ee_pblk;
3613 unsigned int next_len;
3616 next_lblk = le32_to_cpu(abut_ex->ee_block);
3617 next_len = ext4_ext_get_actual_len(abut_ex);
3618 next_pblk = ext4_ext_pblock(abut_ex);
3619 ee_pblk = ext4_ext_pblock(ex);
3622 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3623 * upon those conditions:
3624 * - C1: abut_ex is initialized,
3625 * - C2: abut_ex is logically abutting ex,
3626 * - C3: abut_ex is physically abutting ex,
3627 * - C4: abut_ex can receive the additional blocks without
3628 * overflowing the (initialized) length limit.
3630 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3631 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3632 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3633 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3634 err = ext4_ext_get_access(handle, inode, path + depth);
3638 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3641 /* Shift the start of abut_ex by 'map_len' blocks */
3642 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3643 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3644 ex->ee_len = cpu_to_le16(ee_len - map_len);
3645 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3647 /* Extend abut_ex by 'map_len' blocks */
3648 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3650 /* Result: number of initialized blocks past m_lblk */
3651 allocated = map_len;
3655 /* Mark the block containing both extents as dirty */
3656 ext4_ext_dirty(handle, inode, path + depth);
3658 /* Update path to point to the right extent */
3659 path[depth].p_ext = abut_ex;
3662 allocated = ee_len - (map->m_lblk - ee_block);
3664 WARN_ON(map->m_lblk < ee_block);
3666 * It is safe to convert extent to initialized via explicit
3667 * zeroout only if extent is fully inside i_size or new_size.
3669 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3671 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3672 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3673 (inode->i_sb->s_blocksize_bits - 10);
3677 * 1. split the extent into three extents.
3678 * 2. split the extent into two extents, zeroout the head of the first
3680 * 3. split the extent into two extents, zeroout the tail of the second
3682 * 4. split the extent into two extents with out zeroout.
3683 * 5. no splitting needed, just possibly zeroout the head and / or the
3684 * tail of the extent.
3686 split_map.m_lblk = map->m_lblk;
3687 split_map.m_len = map->m_len;
3689 if (max_zeroout && (allocated > split_map.m_len)) {
3690 if (allocated <= max_zeroout) {
3693 cpu_to_le32(split_map.m_lblk +
3696 cpu_to_le16(allocated - split_map.m_len);
3697 ext4_ext_store_pblock(&zero_ex1,
3698 ext4_ext_pblock(ex) + split_map.m_lblk +
3699 split_map.m_len - ee_block);
3700 err = ext4_ext_zeroout(inode, &zero_ex1);
3703 split_map.m_len = allocated;
3705 if (split_map.m_lblk - ee_block + split_map.m_len <
3708 if (split_map.m_lblk != ee_block) {
3709 zero_ex2.ee_block = ex->ee_block;
3710 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3712 ext4_ext_store_pblock(&zero_ex2,
3713 ext4_ext_pblock(ex));
3714 err = ext4_ext_zeroout(inode, &zero_ex2);
3719 split_map.m_len += split_map.m_lblk - ee_block;
3720 split_map.m_lblk = ee_block;
3721 allocated = map->m_len;
3725 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3730 /* If we have gotten a failure, don't zero out status tree */
3732 err = ext4_zeroout_es(inode, &zero_ex1);
3734 err = ext4_zeroout_es(inode, &zero_ex2);
3736 return err ? err : allocated;
3740 * This function is called by ext4_ext_map_blocks() from
3741 * ext4_get_blocks_dio_write() when DIO to write
3742 * to an unwritten extent.
3744 * Writing to an unwritten extent may result in splitting the unwritten
3745 * extent into multiple initialized/unwritten extents (up to three)
3746 * There are three possibilities:
3747 * a> There is no split required: Entire extent should be unwritten
3748 * b> Splits in two extents: Write is happening at either end of the extent
3749 * c> Splits in three extents: Somone is writing in middle of the extent
3751 * This works the same way in the case of initialized -> unwritten conversion.
3753 * One of more index blocks maybe needed if the extent tree grow after
3754 * the unwritten extent split. To prevent ENOSPC occur at the IO
3755 * complete, we need to split the unwritten extent before DIO submit
3756 * the IO. The unwritten extent called at this time will be split
3757 * into three unwritten extent(at most). After IO complete, the part
3758 * being filled will be convert to initialized by the end_io callback function
3759 * via ext4_convert_unwritten_extents().
3761 * Returns the size of unwritten extent to be written on success.
3763 static int ext4_split_convert_extents(handle_t *handle,
3764 struct inode *inode,
3765 struct ext4_map_blocks *map,
3766 struct ext4_ext_path **ppath,
3769 struct ext4_ext_path *path = *ppath;
3770 ext4_lblk_t eof_block;
3771 ext4_lblk_t ee_block;
3772 struct ext4_extent *ex;
3773 unsigned int ee_len;
3774 int split_flag = 0, depth;
3776 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3777 __func__, inode->i_ino,
3778 (unsigned long long)map->m_lblk, map->m_len);
3780 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3781 inode->i_sb->s_blocksize_bits;
3782 if (eof_block < map->m_lblk + map->m_len)
3783 eof_block = map->m_lblk + map->m_len;
3785 * It is safe to convert extent to initialized via explicit
3786 * zeroout only if extent is fully insde i_size or new_size.
3788 depth = ext_depth(inode);
3789 ex = path[depth].p_ext;
3790 ee_block = le32_to_cpu(ex->ee_block);
3791 ee_len = ext4_ext_get_actual_len(ex);
3793 /* Convert to unwritten */
3794 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3795 split_flag |= EXT4_EXT_DATA_VALID1;
3796 /* Convert to initialized */
3797 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3798 split_flag |= ee_block + ee_len <= eof_block ?
3799 EXT4_EXT_MAY_ZEROOUT : 0;
3800 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3802 flags |= EXT4_GET_BLOCKS_PRE_IO;
3803 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3806 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3807 struct inode *inode,
3808 struct ext4_map_blocks *map,
3809 struct ext4_ext_path **ppath)
3811 struct ext4_ext_path *path = *ppath;
3812 struct ext4_extent *ex;
3813 ext4_lblk_t ee_block;
3814 unsigned int ee_len;
3818 depth = ext_depth(inode);
3819 ex = path[depth].p_ext;
3820 ee_block = le32_to_cpu(ex->ee_block);
3821 ee_len = ext4_ext_get_actual_len(ex);
3823 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3824 "block %llu, max_blocks %u\n", inode->i_ino,
3825 (unsigned long long)ee_block, ee_len);
3827 /* If extent is larger than requested it is a clear sign that we still
3828 * have some extent state machine issues left. So extent_split is still
3830 * TODO: Once all related issues will be fixed this situation should be
3833 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3834 #ifdef CONFIG_EXT4_DEBUG
3835 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3836 " len %u; IO logical block %llu, len %u",
3837 inode->i_ino, (unsigned long long)ee_block, ee_len,
3838 (unsigned long long)map->m_lblk, map->m_len);
3840 err = ext4_split_convert_extents(handle, inode, map, ppath,
3841 EXT4_GET_BLOCKS_CONVERT);
3844 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3846 return PTR_ERR(path);
3847 depth = ext_depth(inode);
3848 ex = path[depth].p_ext;
3851 err = ext4_ext_get_access(handle, inode, path + depth);
3854 /* first mark the extent as initialized */
3855 ext4_ext_mark_initialized(ex);
3857 /* note: ext4_ext_correct_indexes() isn't needed here because
3858 * borders are not changed
3860 ext4_ext_try_to_merge(handle, inode, path, ex);
3862 /* Mark modified extent as dirty */
3863 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3865 ext4_ext_show_leaf(inode, path);
3870 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3872 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3874 struct ext4_ext_path *path,
3878 struct ext4_extent_header *eh;
3879 struct ext4_extent *last_ex;
3881 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3884 depth = ext_depth(inode);
3885 eh = path[depth].p_hdr;
3888 * We're going to remove EOFBLOCKS_FL entirely in future so we
3889 * do not care for this case anymore. Simply remove the flag
3890 * if there are no extents.
3892 if (unlikely(!eh->eh_entries))
3894 last_ex = EXT_LAST_EXTENT(eh);
3896 * We should clear the EOFBLOCKS_FL flag if we are writing the
3897 * last block in the last extent in the file. We test this by
3898 * first checking to see if the caller to
3899 * ext4_ext_get_blocks() was interested in the last block (or
3900 * a block beyond the last block) in the current extent. If
3901 * this turns out to be false, we can bail out from this
3902 * function immediately.
3904 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3905 ext4_ext_get_actual_len(last_ex))
3908 * If the caller does appear to be planning to write at or
3909 * beyond the end of the current extent, we then test to see
3910 * if the current extent is the last extent in the file, by
3911 * checking to make sure it was reached via the rightmost node
3912 * at each level of the tree.
3914 for (i = depth-1; i >= 0; i--)
3915 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3918 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3919 return ext4_mark_inode_dirty(handle, inode);
3923 convert_initialized_extent(handle_t *handle, struct inode *inode,
3924 struct ext4_map_blocks *map,
3925 struct ext4_ext_path **ppath,
3926 unsigned int allocated)
3928 struct ext4_ext_path *path = *ppath;
3929 struct ext4_extent *ex;
3930 ext4_lblk_t ee_block;
3931 unsigned int ee_len;
3936 * Make sure that the extent is no bigger than we support with
3939 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3940 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3942 depth = ext_depth(inode);
3943 ex = path[depth].p_ext;
3944 ee_block = le32_to_cpu(ex->ee_block);
3945 ee_len = ext4_ext_get_actual_len(ex);
3947 ext_debug("%s: inode %lu, logical"
3948 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3949 (unsigned long long)ee_block, ee_len);
3951 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3952 err = ext4_split_convert_extents(handle, inode, map, ppath,
3953 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3956 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3958 return PTR_ERR(path);
3959 depth = ext_depth(inode);
3960 ex = path[depth].p_ext;
3962 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3963 (unsigned long) map->m_lblk);
3964 return -EFSCORRUPTED;
3968 err = ext4_ext_get_access(handle, inode, path + depth);
3971 /* first mark the extent as unwritten */
3972 ext4_ext_mark_unwritten(ex);
3974 /* note: ext4_ext_correct_indexes() isn't needed here because
3975 * borders are not changed
3977 ext4_ext_try_to_merge(handle, inode, path, ex);
3979 /* Mark modified extent as dirty */
3980 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3983 ext4_ext_show_leaf(inode, path);
3985 ext4_update_inode_fsync_trans(handle, inode, 1);
3986 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3989 map->m_flags |= EXT4_MAP_UNWRITTEN;
3990 if (allocated > map->m_len)
3991 allocated = map->m_len;
3992 map->m_len = allocated;
3997 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3998 struct ext4_map_blocks *map,
3999 struct ext4_ext_path **ppath, int flags,
4000 unsigned int allocated, ext4_fsblk_t newblock)
4002 struct ext4_ext_path *path = *ppath;
4006 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4007 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4008 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4010 ext4_ext_show_leaf(inode, path);
4013 * When writing into unwritten space, we should not fail to
4014 * allocate metadata blocks for the new extent block if needed.
4016 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4018 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4019 allocated, newblock);
4021 /* get_block() before submit the IO, split the extent */
4022 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4023 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4024 flags | EXT4_GET_BLOCKS_CONVERT);
4027 map->m_flags |= EXT4_MAP_UNWRITTEN;
4030 /* IO end_io complete, convert the filled extent to written */
4031 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4032 if (flags & EXT4_GET_BLOCKS_ZERO) {
4033 if (allocated > map->m_len)
4034 allocated = map->m_len;
4035 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4040 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4043 ext4_update_inode_fsync_trans(handle, inode, 1);
4044 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4048 map->m_flags |= EXT4_MAP_MAPPED;
4049 map->m_pblk = newblock;
4050 if (allocated > map->m_len)
4051 allocated = map->m_len;
4052 map->m_len = allocated;
4055 /* buffered IO case */
4057 * repeat fallocate creation request
4058 * we already have an unwritten extent
4060 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4061 map->m_flags |= EXT4_MAP_UNWRITTEN;
4065 /* buffered READ or buffered write_begin() lookup */
4066 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4068 * We have blocks reserved already. We
4069 * return allocated blocks so that delalloc
4070 * won't do block reservation for us. But
4071 * the buffer head will be unmapped so that
4072 * a read from the block returns 0s.
4074 map->m_flags |= EXT4_MAP_UNWRITTEN;
4078 /* buffered write, writepage time, convert*/
4079 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4081 ext4_update_inode_fsync_trans(handle, inode, 1);
4088 map->m_flags |= EXT4_MAP_NEW;
4089 if (allocated > map->m_len)
4090 allocated = map->m_len;
4091 map->m_len = allocated;
4094 map->m_flags |= EXT4_MAP_MAPPED;
4095 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4096 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4102 if (allocated > map->m_len)
4103 allocated = map->m_len;
4104 ext4_ext_show_leaf(inode, path);
4105 map->m_pblk = newblock;
4106 map->m_len = allocated;
4108 return err ? err : allocated;
4112 * get_implied_cluster_alloc - check to see if the requested
4113 * allocation (in the map structure) overlaps with a cluster already
4114 * allocated in an extent.
4115 * @sb The filesystem superblock structure
4116 * @map The requested lblk->pblk mapping
4117 * @ex The extent structure which might contain an implied
4118 * cluster allocation
4120 * This function is called by ext4_ext_map_blocks() after we failed to
4121 * find blocks that were already in the inode's extent tree. Hence,
4122 * we know that the beginning of the requested region cannot overlap
4123 * the extent from the inode's extent tree. There are three cases we
4124 * want to catch. The first is this case:
4126 * |--- cluster # N--|
4127 * |--- extent ---| |---- requested region ---|
4130 * The second case that we need to test for is this one:
4132 * |--------- cluster # N ----------------|
4133 * |--- requested region --| |------- extent ----|
4134 * |=======================|
4136 * The third case is when the requested region lies between two extents
4137 * within the same cluster:
4138 * |------------- cluster # N-------------|
4139 * |----- ex -----| |---- ex_right ----|
4140 * |------ requested region ------|
4141 * |================|
4143 * In each of the above cases, we need to set the map->m_pblk and
4144 * map->m_len so it corresponds to the return the extent labelled as
4145 * "|====|" from cluster #N, since it is already in use for data in
4146 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4147 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4148 * as a new "allocated" block region. Otherwise, we will return 0 and
4149 * ext4_ext_map_blocks() will then allocate one or more new clusters
4150 * by calling ext4_mb_new_blocks().
4152 static int get_implied_cluster_alloc(struct super_block *sb,
4153 struct ext4_map_blocks *map,
4154 struct ext4_extent *ex,
4155 struct ext4_ext_path *path)
4157 struct ext4_sb_info *sbi = EXT4_SB(sb);
4158 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4159 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4160 ext4_lblk_t rr_cluster_start;
4161 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4162 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4163 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4165 /* The extent passed in that we are trying to match */
4166 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4167 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4169 /* The requested region passed into ext4_map_blocks() */
4170 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4172 if ((rr_cluster_start == ex_cluster_end) ||
4173 (rr_cluster_start == ex_cluster_start)) {
4174 if (rr_cluster_start == ex_cluster_end)
4175 ee_start += ee_len - 1;
4176 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4177 map->m_len = min(map->m_len,
4178 (unsigned) sbi->s_cluster_ratio - c_offset);
4180 * Check for and handle this case:
4182 * |--------- cluster # N-------------|
4183 * |------- extent ----|
4184 * |--- requested region ---|
4188 if (map->m_lblk < ee_block)
4189 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4192 * Check for the case where there is already another allocated
4193 * block to the right of 'ex' but before the end of the cluster.
4195 * |------------- cluster # N-------------|
4196 * |----- ex -----| |---- ex_right ----|
4197 * |------ requested region ------|
4198 * |================|
4200 if (map->m_lblk > ee_block) {
4201 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4202 map->m_len = min(map->m_len, next - map->m_lblk);
4205 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4209 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4215 * Block allocation/map/preallocation routine for extents based files
4218 * Need to be called with
4219 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4220 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4222 * return > 0, number of of blocks already mapped/allocated
4223 * if create == 0 and these are pre-allocated blocks
4224 * buffer head is unmapped
4225 * otherwise blocks are mapped
4227 * return = 0, if plain look up failed (blocks have not been allocated)
4228 * buffer head is unmapped
4230 * return < 0, error case.
4232 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4233 struct ext4_map_blocks *map, int flags)
4235 struct ext4_ext_path *path = NULL;
4236 struct ext4_extent newex, *ex, *ex2;
4237 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4238 ext4_fsblk_t newblock = 0;
4239 int free_on_err = 0, err = 0, depth, ret;
4240 unsigned int allocated = 0, offset = 0;
4241 unsigned int allocated_clusters = 0;
4242 struct ext4_allocation_request ar;
4243 ext4_lblk_t cluster_offset;
4244 bool map_from_cluster = false;
4246 ext_debug("blocks %u/%u requested for inode %lu\n",
4247 map->m_lblk, map->m_len, inode->i_ino);
4248 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4250 /* find extent for this block */
4251 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4253 err = PTR_ERR(path);
4258 depth = ext_depth(inode);
4261 * consistent leaf must not be empty;
4262 * this situation is possible, though, _during_ tree modification;
4263 * this is why assert can't be put in ext4_find_extent()
4265 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4266 EXT4_ERROR_INODE(inode, "bad extent address "
4267 "lblock: %lu, depth: %d pblock %lld",
4268 (unsigned long) map->m_lblk, depth,
4269 path[depth].p_block);
4270 err = -EFSCORRUPTED;
4274 ex = path[depth].p_ext;
4276 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4277 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4278 unsigned short ee_len;
4282 * unwritten extents are treated as holes, except that
4283 * we split out initialized portions during a write.
4285 ee_len = ext4_ext_get_actual_len(ex);
4287 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4289 /* if found extent covers block, simply return it */
4290 if (in_range(map->m_lblk, ee_block, ee_len)) {
4291 newblock = map->m_lblk - ee_block + ee_start;
4292 /* number of remaining blocks in the extent */
4293 allocated = ee_len - (map->m_lblk - ee_block);
4294 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4295 ee_block, ee_len, newblock);
4298 * If the extent is initialized check whether the
4299 * caller wants to convert it to unwritten.
4301 if ((!ext4_ext_is_unwritten(ex)) &&
4302 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4303 allocated = convert_initialized_extent(
4304 handle, inode, map, &path,
4307 } else if (!ext4_ext_is_unwritten(ex))
4310 ret = ext4_ext_handle_unwritten_extents(
4311 handle, inode, map, &path, flags,
4312 allocated, newblock);
4322 * requested block isn't allocated yet;
4323 * we couldn't try to create block if create flag is zero
4325 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4326 ext4_lblk_t hole_start, hole_len;
4328 hole_start = map->m_lblk;
4329 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4331 * put just found gap into cache to speed up
4332 * subsequent requests
4334 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4336 /* Update hole_len to reflect hole size after map->m_lblk */
4337 if (hole_start != map->m_lblk)
4338 hole_len -= map->m_lblk - hole_start;
4340 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4346 * Okay, we need to do block allocation.
4348 newex.ee_block = cpu_to_le32(map->m_lblk);
4349 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4352 * If we are doing bigalloc, check to see if the extent returned
4353 * by ext4_find_extent() implies a cluster we can use.
4355 if (cluster_offset && ex &&
4356 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4357 ar.len = allocated = map->m_len;
4358 newblock = map->m_pblk;
4359 map_from_cluster = true;
4360 goto got_allocated_blocks;
4363 /* find neighbour allocated blocks */
4364 ar.lleft = map->m_lblk;
4365 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4368 ar.lright = map->m_lblk;
4370 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4374 /* Check if the extent after searching to the right implies a
4375 * cluster we can use. */
4376 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4377 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4378 ar.len = allocated = map->m_len;
4379 newblock = map->m_pblk;
4380 map_from_cluster = true;
4381 goto got_allocated_blocks;
4385 * See if request is beyond maximum number of blocks we can have in
4386 * a single extent. For an initialized extent this limit is
4387 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4388 * EXT_UNWRITTEN_MAX_LEN.
4390 if (map->m_len > EXT_INIT_MAX_LEN &&
4391 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4392 map->m_len = EXT_INIT_MAX_LEN;
4393 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4394 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4395 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4397 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4398 newex.ee_len = cpu_to_le16(map->m_len);
4399 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4401 allocated = ext4_ext_get_actual_len(&newex);
4403 allocated = map->m_len;
4405 /* allocate new block */
4407 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4408 ar.logical = map->m_lblk;
4410 * We calculate the offset from the beginning of the cluster
4411 * for the logical block number, since when we allocate a
4412 * physical cluster, the physical block should start at the
4413 * same offset from the beginning of the cluster. This is
4414 * needed so that future calls to get_implied_cluster_alloc()
4417 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4418 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4420 ar.logical -= offset;
4421 if (S_ISREG(inode->i_mode))
4422 ar.flags = EXT4_MB_HINT_DATA;
4424 /* disable in-core preallocation for non-regular files */
4426 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4427 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4428 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4429 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4430 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4431 ar.flags |= EXT4_MB_USE_RESERVED;
4432 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4435 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4436 ar.goal, newblock, allocated);
4438 allocated_clusters = ar.len;
4439 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4440 if (ar.len > allocated)
4443 got_allocated_blocks:
4444 /* try to insert new extent into found leaf and return */
4445 ext4_ext_store_pblock(&newex, newblock + offset);
4446 newex.ee_len = cpu_to_le16(ar.len);
4447 /* Mark unwritten */
4448 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4449 ext4_ext_mark_unwritten(&newex);
4450 map->m_flags |= EXT4_MAP_UNWRITTEN;
4454 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4455 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4458 err = ext4_ext_insert_extent(handle, inode, &path,
4461 if (err && free_on_err) {
4462 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4463 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4464 /* free data blocks we just allocated */
4465 /* not a good idea to call discard here directly,
4466 * but otherwise we'd need to call it every free() */
4467 ext4_discard_preallocations(inode);
4468 ext4_free_blocks(handle, inode, NULL, newblock,
4469 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4473 /* previous routine could use block we allocated */
4474 newblock = ext4_ext_pblock(&newex);
4475 allocated = ext4_ext_get_actual_len(&newex);
4476 if (allocated > map->m_len)
4477 allocated = map->m_len;
4478 map->m_flags |= EXT4_MAP_NEW;
4481 * Reduce the reserved cluster count to reflect successful deferred
4482 * allocation of delayed allocated clusters or direct allocation of
4483 * clusters discovered to be delayed allocated. Once allocated, a
4484 * cluster is not included in the reserved count.
4486 if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4487 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4489 * When allocating delayed allocated clusters, simply
4490 * reduce the reserved cluster count and claim quota
4492 ext4_da_update_reserve_space(inode, allocated_clusters,
4495 ext4_lblk_t lblk, len;
4499 * When allocating non-delayed allocated clusters
4500 * (from fallocate, filemap, DIO, or clusters
4501 * allocated when delalloc has been disabled by
4502 * ext4_nonda_switch), reduce the reserved cluster
4503 * count by the number of allocated clusters that
4504 * have previously been delayed allocated. Quota
4505 * has been claimed by ext4_mb_new_blocks() above,
4506 * so release the quota reservations made for any
4507 * previously delayed allocated clusters.
4509 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4510 len = allocated_clusters << sbi->s_cluster_bits;
4511 n = ext4_es_delayed_clu(inode, lblk, len);
4513 ext4_da_update_reserve_space(inode, (int) n, 0);
4518 * Cache the extent and update transaction to commit on fdatasync only
4519 * when it is _not_ an unwritten extent.
4521 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4522 ext4_update_inode_fsync_trans(handle, inode, 1);
4524 ext4_update_inode_fsync_trans(handle, inode, 0);
4526 if (allocated > map->m_len)
4527 allocated = map->m_len;
4528 ext4_ext_show_leaf(inode, path);
4529 map->m_flags |= EXT4_MAP_MAPPED;
4530 map->m_pblk = newblock;
4531 map->m_len = allocated;
4533 ext4_ext_drop_refs(path);
4536 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4537 err ? err : allocated);
4538 return err ? err : allocated;
4541 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4543 struct super_block *sb = inode->i_sb;
4544 ext4_lblk_t last_block;
4548 * TODO: optimization is possible here.
4549 * Probably we need not scan at all,
4550 * because page truncation is enough.
4553 /* we have to know where to truncate from in crash case */
4554 EXT4_I(inode)->i_disksize = inode->i_size;
4555 err = ext4_mark_inode_dirty(handle, inode);
4559 last_block = (inode->i_size + sb->s_blocksize - 1)
4560 >> EXT4_BLOCK_SIZE_BITS(sb);
4562 err = ext4_es_remove_extent(inode, last_block,
4563 EXT_MAX_BLOCKS - last_block);
4564 if (err == -ENOMEM) {
4566 congestion_wait(BLK_RW_ASYNC, HZ/50);
4571 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4574 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4575 ext4_lblk_t len, loff_t new_size,
4578 struct inode *inode = file_inode(file);
4584 struct ext4_map_blocks map;
4585 unsigned int credits;
4588 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4589 map.m_lblk = offset;
4592 * Don't normalize the request if it can fit in one extent so
4593 * that it doesn't get unnecessarily split into multiple
4596 if (len <= EXT_UNWRITTEN_MAX_LEN)
4597 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4600 * credits to insert 1 extent into extent tree
4602 credits = ext4_chunk_trans_blocks(inode, len);
4603 depth = ext_depth(inode);
4606 while (ret >= 0 && len) {
4608 * Recalculate credits when extent tree depth changes.
4610 if (depth != ext_depth(inode)) {
4611 credits = ext4_chunk_trans_blocks(inode, len);
4612 depth = ext_depth(inode);
4615 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4617 if (IS_ERR(handle)) {
4618 ret = PTR_ERR(handle);
4621 ret = ext4_map_blocks(handle, inode, &map, flags);
4623 ext4_debug("inode #%lu: block %u: len %u: "
4624 "ext4_ext_map_blocks returned %d",
4625 inode->i_ino, map.m_lblk,
4627 ext4_mark_inode_dirty(handle, inode);
4628 ret2 = ext4_journal_stop(handle);
4632 map.m_len = len = len - ret;
4633 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4634 inode->i_ctime = current_time(inode);
4636 if (epos > new_size)
4638 if (ext4_update_inode_size(inode, epos) & 0x1)
4639 inode->i_mtime = inode->i_ctime;
4641 if (epos > inode->i_size)
4642 ext4_set_inode_flag(inode,
4643 EXT4_INODE_EOFBLOCKS);
4645 ext4_mark_inode_dirty(handle, inode);
4646 ext4_update_inode_fsync_trans(handle, inode, 1);
4647 ret2 = ext4_journal_stop(handle);
4651 if (ret == -ENOSPC &&
4652 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4657 return ret > 0 ? ret2 : ret;
4660 static long ext4_zero_range(struct file *file, loff_t offset,
4661 loff_t len, int mode)
4663 struct inode *inode = file_inode(file);
4664 handle_t *handle = NULL;
4665 unsigned int max_blocks;
4666 loff_t new_size = 0;
4670 int partial_begin, partial_end;
4673 unsigned int blkbits = inode->i_blkbits;
4675 trace_ext4_zero_range(inode, offset, len, mode);
4677 if (!S_ISREG(inode->i_mode))
4680 /* Call ext4_force_commit to flush all data in case of data=journal. */
4681 if (ext4_should_journal_data(inode)) {
4682 ret = ext4_force_commit(inode->i_sb);
4688 * Round up offset. This is not fallocate, we neet to zero out
4689 * blocks, so convert interior block aligned part of the range to
4690 * unwritten and possibly manually zero out unaligned parts of the
4693 start = round_up(offset, 1 << blkbits);
4694 end = round_down((offset + len), 1 << blkbits);
4696 if (start < offset || end > offset + len)
4698 partial_begin = offset & ((1 << blkbits) - 1);
4699 partial_end = (offset + len) & ((1 << blkbits) - 1);
4701 lblk = start >> blkbits;
4702 max_blocks = (end >> blkbits);
4703 if (max_blocks < lblk)
4711 * Indirect files do not support unwritten extnets
4713 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4718 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4719 (offset + len > inode->i_size ||
4720 offset + len > EXT4_I(inode)->i_disksize)) {
4721 new_size = offset + len;
4722 ret = inode_newsize_ok(inode, new_size);
4727 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4728 if (mode & FALLOC_FL_KEEP_SIZE)
4729 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4731 /* Wait all existing dio workers, newcomers will block on i_mutex */
4732 inode_dio_wait(inode);
4734 /* Preallocate the range including the unaligned edges */
4735 if (partial_begin || partial_end) {
4736 ret = ext4_alloc_file_blocks(file,
4737 round_down(offset, 1 << blkbits) >> blkbits,
4738 (round_up((offset + len), 1 << blkbits) -
4739 round_down(offset, 1 << blkbits)) >> blkbits,
4746 /* Zero range excluding the unaligned edges */
4747 if (max_blocks > 0) {
4748 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4752 * Prevent page faults from reinstantiating pages we have
4753 * released from page cache.
4755 down_write(&EXT4_I(inode)->i_mmap_sem);
4757 ret = ext4_break_layouts(inode);
4759 up_write(&EXT4_I(inode)->i_mmap_sem);
4763 ret = ext4_update_disksize_before_punch(inode, offset, len);
4765 up_write(&EXT4_I(inode)->i_mmap_sem);
4768 /* Now release the pages and zero block aligned part of pages */
4769 truncate_pagecache_range(inode, start, end - 1);
4770 inode->i_mtime = inode->i_ctime = current_time(inode);
4772 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4774 up_write(&EXT4_I(inode)->i_mmap_sem);
4778 if (!partial_begin && !partial_end)
4782 * In worst case we have to writeout two nonadjacent unwritten
4783 * blocks and update the inode
4785 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4786 if (ext4_should_journal_data(inode))
4788 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4789 if (IS_ERR(handle)) {
4790 ret = PTR_ERR(handle);
4791 ext4_std_error(inode->i_sb, ret);
4795 inode->i_mtime = inode->i_ctime = current_time(inode);
4797 ext4_update_inode_size(inode, new_size);
4800 * Mark that we allocate beyond EOF so the subsequent truncate
4801 * can proceed even if the new size is the same as i_size.
4803 if (offset + len > inode->i_size)
4804 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4806 ext4_mark_inode_dirty(handle, inode);
4808 /* Zero out partial block at the edges of the range */
4809 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4811 ext4_update_inode_fsync_trans(handle, inode, 1);
4813 if (file->f_flags & O_SYNC)
4814 ext4_handle_sync(handle);
4816 ext4_journal_stop(handle);
4818 inode_unlock(inode);
4823 * preallocate space for a file. This implements ext4's fallocate file
4824 * operation, which gets called from sys_fallocate system call.
4825 * For block-mapped files, posix_fallocate should fall back to the method
4826 * of writing zeroes to the required new blocks (the same behavior which is
4827 * expected for file systems which do not support fallocate() system call).
4829 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4831 struct inode *inode = file_inode(file);
4832 loff_t new_size = 0;
4833 unsigned int max_blocks;
4837 unsigned int blkbits = inode->i_blkbits;
4840 * Encrypted inodes can't handle collapse range or insert
4841 * range since we would need to re-encrypt blocks with a
4842 * different IV or XTS tweak (which are based on the logical
4845 if (IS_ENCRYPTED(inode) &&
4846 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4849 /* Return error if mode is not supported */
4850 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4851 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4852 FALLOC_FL_INSERT_RANGE))
4855 if (mode & FALLOC_FL_PUNCH_HOLE)
4856 return ext4_punch_hole(inode, offset, len);
4858 ret = ext4_convert_inline_data(inode);
4862 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4863 return ext4_collapse_range(inode, offset, len);
4865 if (mode & FALLOC_FL_INSERT_RANGE)
4866 return ext4_insert_range(inode, offset, len);
4868 if (mode & FALLOC_FL_ZERO_RANGE)
4869 return ext4_zero_range(file, offset, len, mode);
4871 trace_ext4_fallocate_enter(inode, offset, len, mode);
4872 lblk = offset >> blkbits;
4874 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4875 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4876 if (mode & FALLOC_FL_KEEP_SIZE)
4877 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4882 * We only support preallocation for extent-based files only
4884 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4889 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4890 (offset + len > inode->i_size ||
4891 offset + len > EXT4_I(inode)->i_disksize)) {
4892 new_size = offset + len;
4893 ret = inode_newsize_ok(inode, new_size);
4898 /* Wait all existing dio workers, newcomers will block on i_mutex */
4899 inode_dio_wait(inode);
4901 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4905 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4906 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4907 EXT4_I(inode)->i_sync_tid);
4910 inode_unlock(inode);
4911 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4916 * This function convert a range of blocks to written extents
4917 * The caller of this function will pass the start offset and the size.
4918 * all unwritten extents within this range will be converted to
4921 * This function is called from the direct IO end io call back
4922 * function, to convert the fallocated extents after IO is completed.
4923 * Returns 0 on success.
4925 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4926 loff_t offset, ssize_t len)
4928 unsigned int max_blocks;
4931 struct ext4_map_blocks map;
4932 unsigned int blkbits = inode->i_blkbits;
4933 unsigned int credits = 0;
4935 map.m_lblk = offset >> blkbits;
4936 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4940 * credits to insert 1 extent into extent tree
4942 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4944 while (ret >= 0 && ret < max_blocks) {
4946 map.m_len = (max_blocks -= ret);
4948 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4950 if (IS_ERR(handle)) {
4951 ret = PTR_ERR(handle);
4955 ret = ext4_map_blocks(handle, inode, &map,
4956 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4958 ext4_warning(inode->i_sb,
4959 "inode #%lu: block %u: len %u: "
4960 "ext4_ext_map_blocks returned %d",
4961 inode->i_ino, map.m_lblk,
4963 ext4_mark_inode_dirty(handle, inode);
4965 ret2 = ext4_journal_stop(handle);
4966 if (ret <= 0 || ret2)
4969 return ret > 0 ? ret2 : ret;
4972 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4975 struct ext4_io_end_vec *io_end_vec;
4978 * This is somewhat ugly but the idea is clear: When transaction is
4979 * reserved, everything goes into it. Otherwise we rather start several
4980 * smaller transactions for conversion of each extent separately.
4983 handle = ext4_journal_start_reserved(handle,
4984 EXT4_HT_EXT_CONVERT);
4986 return PTR_ERR(handle);
4989 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4990 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4998 err = ext4_journal_stop(handle);
5000 return ret < 0 ? ret : err;
5004 * If newes is not existing extent (newes->ec_pblk equals zero) find
5005 * delayed extent at start of newes and update newes accordingly and
5006 * return start of the next delayed extent.
5008 * If newes is existing extent (newes->ec_pblk is not equal zero)
5009 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5010 * extent found. Leave newes unmodified.
5012 static int ext4_find_delayed_extent(struct inode *inode,
5013 struct extent_status *newes)
5015 struct extent_status es;
5016 ext4_lblk_t block, next_del;
5018 if (newes->es_pblk == 0) {
5019 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5021 newes->es_lblk + newes->es_len - 1,
5025 * No extent in extent-tree contains block @newes->es_pblk,
5026 * then the block may stay in 1)a hole or 2)delayed-extent.
5032 if (es.es_lblk > newes->es_lblk) {
5034 newes->es_len = min(es.es_lblk - newes->es_lblk,
5039 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5042 block = newes->es_lblk + newes->es_len;
5043 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5044 EXT_MAX_BLOCKS, &es);
5046 next_del = EXT_MAX_BLOCKS;
5048 next_del = es.es_lblk;
5053 static int ext4_xattr_fiemap(struct inode *inode,
5054 struct fiemap_extent_info *fieinfo)
5058 __u32 flags = FIEMAP_EXTENT_LAST;
5059 int blockbits = inode->i_sb->s_blocksize_bits;
5063 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5064 struct ext4_iloc iloc;
5065 int offset; /* offset of xattr in inode */
5067 error = ext4_get_inode_loc(inode, &iloc);
5070 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5071 offset = EXT4_GOOD_OLD_INODE_SIZE +
5072 EXT4_I(inode)->i_extra_isize;
5074 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5075 flags |= FIEMAP_EXTENT_DATA_INLINE;
5077 } else { /* external block */
5078 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5079 length = inode->i_sb->s_blocksize;
5083 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5085 return (error < 0 ? error : 0);
5088 static int _ext4_fiemap(struct inode *inode,
5089 struct fiemap_extent_info *fieinfo,
5090 __u64 start, __u64 len,
5091 int (*fill)(struct inode *, ext4_lblk_t,
5093 struct fiemap_extent_info *))
5095 ext4_lblk_t start_blk;
5096 u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5100 if (ext4_has_inline_data(inode)) {
5103 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5110 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5111 error = ext4_ext_precache(inode);
5114 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5117 /* fallback to generic here if not in extents fmt */
5118 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5119 fill == ext4_fill_fiemap_extents)
5120 return generic_block_fiemap(inode, fieinfo, start, len,
5123 if (fill == ext4_fill_es_cache_info)
5124 ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5125 if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5128 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5129 error = ext4_xattr_fiemap(inode, fieinfo);
5131 ext4_lblk_t len_blks;
5134 start_blk = start >> inode->i_sb->s_blocksize_bits;
5135 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5136 if (last_blk >= EXT_MAX_BLOCKS)
5137 last_blk = EXT_MAX_BLOCKS-1;
5138 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5141 * Walk the extent tree gathering extent information
5142 * and pushing extents back to the user.
5144 error = fill(inode, start_blk, len_blks, fieinfo);
5149 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5150 __u64 start, __u64 len)
5152 return _ext4_fiemap(inode, fieinfo, start, len,
5153 ext4_fill_fiemap_extents);
5156 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5157 __u64 start, __u64 len)
5159 if (ext4_has_inline_data(inode)) {
5162 down_read(&EXT4_I(inode)->xattr_sem);
5163 has_inline = ext4_has_inline_data(inode);
5164 up_read(&EXT4_I(inode)->xattr_sem);
5169 return _ext4_fiemap(inode, fieinfo, start, len,
5170 ext4_fill_es_cache_info);
5176 * Function to access the path buffer for marking it dirty.
5177 * It also checks if there are sufficient credits left in the journal handle
5181 ext4_access_path(handle_t *handle, struct inode *inode,
5182 struct ext4_ext_path *path)
5186 if (!ext4_handle_valid(handle))
5190 * Check if need to extend journal credits
5191 * 3 for leaf, sb, and inode plus 2 (bmap and group
5192 * descriptor) for each block group; assume two block
5195 credits = ext4_writepage_trans_blocks(inode);
5196 err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
5200 err = ext4_ext_get_access(handle, inode, path);
5205 * ext4_ext_shift_path_extents:
5206 * Shift the extents of a path structure lying between path[depth].p_ext
5207 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5208 * if it is right shift or left shift operation.
5211 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5212 struct inode *inode, handle_t *handle,
5213 enum SHIFT_DIRECTION SHIFT)
5216 struct ext4_extent *ex_start, *ex_last;
5217 bool update = false;
5218 depth = path->p_depth;
5220 while (depth >= 0) {
5221 if (depth == path->p_depth) {
5222 ex_start = path[depth].p_ext;
5224 return -EFSCORRUPTED;
5226 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5228 err = ext4_access_path(handle, inode, path + depth);
5232 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5235 while (ex_start <= ex_last) {
5236 if (SHIFT == SHIFT_LEFT) {
5237 le32_add_cpu(&ex_start->ee_block,
5239 /* Try to merge to the left. */
5241 EXT_FIRST_EXTENT(path[depth].p_hdr))
5243 ext4_ext_try_to_merge_right(inode,
5244 path, ex_start - 1))
5249 le32_add_cpu(&ex_last->ee_block, shift);
5250 ext4_ext_try_to_merge_right(inode, path,
5255 err = ext4_ext_dirty(handle, inode, path + depth);
5259 if (--depth < 0 || !update)
5263 /* Update index too */
5264 err = ext4_access_path(handle, inode, path + depth);
5268 if (SHIFT == SHIFT_LEFT)
5269 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5271 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5272 err = ext4_ext_dirty(handle, inode, path + depth);
5276 /* we are done if current index is not a starting index */
5277 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5288 * ext4_ext_shift_extents:
5289 * All the extents which lies in the range from @start to the last allocated
5290 * block for the @inode are shifted either towards left or right (depending
5291 * upon @SHIFT) by @shift blocks.
5292 * On success, 0 is returned, error otherwise.
5295 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5296 ext4_lblk_t start, ext4_lblk_t shift,
5297 enum SHIFT_DIRECTION SHIFT)
5299 struct ext4_ext_path *path;
5301 struct ext4_extent *extent;
5302 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5304 /* Let path point to the last extent */
5305 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5308 return PTR_ERR(path);
5310 depth = path->p_depth;
5311 extent = path[depth].p_ext;
5315 stop = le32_to_cpu(extent->ee_block);
5318 * For left shifts, make sure the hole on the left is big enough to
5319 * accommodate the shift. For right shifts, make sure the last extent
5320 * won't be shifted beyond EXT_MAX_BLOCKS.
5322 if (SHIFT == SHIFT_LEFT) {
5323 path = ext4_find_extent(inode, start - 1, &path,
5326 return PTR_ERR(path);
5327 depth = path->p_depth;
5328 extent = path[depth].p_ext;
5330 ex_start = le32_to_cpu(extent->ee_block);
5331 ex_end = le32_to_cpu(extent->ee_block) +
5332 ext4_ext_get_actual_len(extent);
5338 if ((start == ex_start && shift > ex_start) ||
5339 (shift > start - ex_end)) {
5344 if (shift > EXT_MAX_BLOCKS -
5345 (stop + ext4_ext_get_actual_len(extent))) {
5352 * In case of left shift, iterator points to start and it is increased
5353 * till we reach stop. In case of right shift, iterator points to stop
5354 * and it is decreased till we reach start.
5356 if (SHIFT == SHIFT_LEFT)
5362 * Its safe to start updating extents. Start and stop are unsigned, so
5363 * in case of right shift if extent with 0 block is reached, iterator
5364 * becomes NULL to indicate the end of the loop.
5366 while (iterator && start <= stop) {
5367 path = ext4_find_extent(inode, *iterator, &path,
5370 return PTR_ERR(path);
5371 depth = path->p_depth;
5372 extent = path[depth].p_ext;
5374 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5375 (unsigned long) *iterator);
5376 return -EFSCORRUPTED;
5378 if (SHIFT == SHIFT_LEFT && *iterator >
5379 le32_to_cpu(extent->ee_block)) {
5380 /* Hole, move to the next extent */
5381 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5382 path[depth].p_ext++;
5384 *iterator = ext4_ext_next_allocated_block(path);
5389 if (SHIFT == SHIFT_LEFT) {
5390 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5391 *iterator = le32_to_cpu(extent->ee_block) +
5392 ext4_ext_get_actual_len(extent);
5394 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5395 if (le32_to_cpu(extent->ee_block) > 0)
5396 *iterator = le32_to_cpu(extent->ee_block) - 1;
5398 /* Beginning is reached, end of the loop */
5400 /* Update path extent in case we need to stop */
5401 while (le32_to_cpu(extent->ee_block) < start)
5403 path[depth].p_ext = extent;
5405 ret = ext4_ext_shift_path_extents(path, shift, inode,
5411 ext4_ext_drop_refs(path);
5417 * ext4_collapse_range:
5418 * This implements the fallocate's collapse range functionality for ext4
5419 * Returns: 0 and non-zero on error.
5421 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5423 struct super_block *sb = inode->i_sb;
5424 ext4_lblk_t punch_start, punch_stop;
5426 unsigned int credits;
5427 loff_t new_size, ioffset;
5431 * We need to test this early because xfstests assumes that a
5432 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5433 * system does not support collapse range.
5435 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5438 /* Collapse range works only on fs cluster size aligned regions. */
5439 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5442 if (!S_ISREG(inode->i_mode))
5445 trace_ext4_collapse_range(inode, offset, len);
5447 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5448 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5450 /* Call ext4_force_commit to flush all data in case of data=journal. */
5451 if (ext4_should_journal_data(inode)) {
5452 ret = ext4_force_commit(inode->i_sb);
5459 * There is no need to overlap collapse range with EOF, in which case
5460 * it is effectively a truncate operation
5462 if (offset + len >= inode->i_size) {
5467 /* Currently just for extent based files */
5468 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5473 /* Wait for existing dio to complete */
5474 inode_dio_wait(inode);
5477 * Prevent page faults from reinstantiating pages we have released from
5480 down_write(&EXT4_I(inode)->i_mmap_sem);
5482 ret = ext4_break_layouts(inode);
5487 * Need to round down offset to be aligned with page size boundary
5488 * for page size > block size.
5490 ioffset = round_down(offset, PAGE_SIZE);
5492 * Write tail of the last page before removed range since it will get
5493 * removed from the page cache below.
5495 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5499 * Write data that will be shifted to preserve them when discarding
5500 * page cache below. We are also protected from pages becoming dirty
5503 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5507 truncate_pagecache(inode, ioffset);
5509 credits = ext4_writepage_trans_blocks(inode);
5510 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5511 if (IS_ERR(handle)) {
5512 ret = PTR_ERR(handle);
5516 down_write(&EXT4_I(inode)->i_data_sem);
5517 ext4_discard_preallocations(inode);
5519 ret = ext4_es_remove_extent(inode, punch_start,
5520 EXT_MAX_BLOCKS - punch_start);
5522 up_write(&EXT4_I(inode)->i_data_sem);
5526 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5528 up_write(&EXT4_I(inode)->i_data_sem);
5531 ext4_discard_preallocations(inode);
5533 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5534 punch_stop - punch_start, SHIFT_LEFT);
5536 up_write(&EXT4_I(inode)->i_data_sem);
5540 new_size = inode->i_size - len;
5541 i_size_write(inode, new_size);
5542 EXT4_I(inode)->i_disksize = new_size;
5544 up_write(&EXT4_I(inode)->i_data_sem);
5546 ext4_handle_sync(handle);
5547 inode->i_mtime = inode->i_ctime = current_time(inode);
5548 ext4_mark_inode_dirty(handle, inode);
5549 ext4_update_inode_fsync_trans(handle, inode, 1);
5552 ext4_journal_stop(handle);
5554 up_write(&EXT4_I(inode)->i_mmap_sem);
5556 inode_unlock(inode);
5561 * ext4_insert_range:
5562 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5563 * The data blocks starting from @offset to the EOF are shifted by @len
5564 * towards right to create a hole in the @inode. Inode size is increased
5566 * Returns 0 on success, error otherwise.
5568 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5570 struct super_block *sb = inode->i_sb;
5572 struct ext4_ext_path *path;
5573 struct ext4_extent *extent;
5574 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5575 unsigned int credits, ee_len;
5576 int ret = 0, depth, split_flag = 0;
5580 * We need to test this early because xfstests assumes that an
5581 * insert range of (0, 1) will return EOPNOTSUPP if the file
5582 * system does not support insert range.
5584 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5587 /* Insert range works only on fs cluster size aligned regions. */
5588 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5591 if (!S_ISREG(inode->i_mode))
5594 trace_ext4_insert_range(inode, offset, len);
5596 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5597 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5599 /* Call ext4_force_commit to flush all data in case of data=journal */
5600 if (ext4_should_journal_data(inode)) {
5601 ret = ext4_force_commit(inode->i_sb);
5607 /* Currently just for extent based files */
5608 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5613 /* Check whether the maximum file size would be exceeded */
5614 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5619 /* Offset must be less than i_size */
5620 if (offset >= inode->i_size) {
5625 /* Wait for existing dio to complete */
5626 inode_dio_wait(inode);
5629 * Prevent page faults from reinstantiating pages we have released from
5632 down_write(&EXT4_I(inode)->i_mmap_sem);
5634 ret = ext4_break_layouts(inode);
5639 * Need to round down to align start offset to page size boundary
5640 * for page size > block size.
5642 ioffset = round_down(offset, PAGE_SIZE);
5643 /* Write out all dirty pages */
5644 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5648 truncate_pagecache(inode, ioffset);
5650 credits = ext4_writepage_trans_blocks(inode);
5651 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5652 if (IS_ERR(handle)) {
5653 ret = PTR_ERR(handle);
5657 /* Expand file to avoid data loss if there is error while shifting */
5658 inode->i_size += len;
5659 EXT4_I(inode)->i_disksize += len;
5660 inode->i_mtime = inode->i_ctime = current_time(inode);
5661 ret = ext4_mark_inode_dirty(handle, inode);
5665 down_write(&EXT4_I(inode)->i_data_sem);
5666 ext4_discard_preallocations(inode);
5668 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5670 up_write(&EXT4_I(inode)->i_data_sem);
5674 depth = ext_depth(inode);
5675 extent = path[depth].p_ext;
5677 ee_start_lblk = le32_to_cpu(extent->ee_block);
5678 ee_len = ext4_ext_get_actual_len(extent);
5681 * If offset_lblk is not the starting block of extent, split
5682 * the extent @offset_lblk
5684 if ((offset_lblk > ee_start_lblk) &&
5685 (offset_lblk < (ee_start_lblk + ee_len))) {
5686 if (ext4_ext_is_unwritten(extent))
5687 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5688 EXT4_EXT_MARK_UNWRIT2;
5689 ret = ext4_split_extent_at(handle, inode, &path,
5690 offset_lblk, split_flag,
5692 EXT4_GET_BLOCKS_PRE_IO |
5693 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5696 ext4_ext_drop_refs(path);
5699 up_write(&EXT4_I(inode)->i_data_sem);
5703 ext4_ext_drop_refs(path);
5707 ret = ext4_es_remove_extent(inode, offset_lblk,
5708 EXT_MAX_BLOCKS - offset_lblk);
5710 up_write(&EXT4_I(inode)->i_data_sem);
5715 * if offset_lblk lies in a hole which is at start of file, use
5716 * ee_start_lblk to shift extents
5718 ret = ext4_ext_shift_extents(inode, handle,
5719 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5720 len_lblk, SHIFT_RIGHT);
5722 up_write(&EXT4_I(inode)->i_data_sem);
5724 ext4_handle_sync(handle);
5726 ext4_update_inode_fsync_trans(handle, inode, 1);
5729 ext4_journal_stop(handle);
5731 up_write(&EXT4_I(inode)->i_mmap_sem);
5733 inode_unlock(inode);
5738 * ext4_swap_extents() - Swap extents between two inodes
5739 * @handle: handle for this transaction
5740 * @inode1: First inode
5741 * @inode2: Second inode
5742 * @lblk1: Start block for first inode
5743 * @lblk2: Start block for second inode
5744 * @count: Number of blocks to swap
5745 * @unwritten: Mark second inode's extents as unwritten after swap
5746 * @erp: Pointer to save error value
5748 * This helper routine does exactly what is promise "swap extents". All other
5749 * stuff such as page-cache locking consistency, bh mapping consistency or
5750 * extent's data copying must be performed by caller.
5752 * i_mutex is held for both inodes
5753 * i_data_sem is locked for write for both inodes
5755 * All pages from requested range are locked for both inodes
5758 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5759 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5760 ext4_lblk_t count, int unwritten, int *erp)
5762 struct ext4_ext_path *path1 = NULL;
5763 struct ext4_ext_path *path2 = NULL;
5764 int replaced_count = 0;
5766 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5767 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5768 BUG_ON(!inode_is_locked(inode1));
5769 BUG_ON(!inode_is_locked(inode2));
5771 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5774 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5779 struct ext4_extent *ex1, *ex2, tmp_ex;
5780 ext4_lblk_t e1_blk, e2_blk;
5781 int e1_len, e2_len, len;
5784 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5785 if (IS_ERR(path1)) {
5786 *erp = PTR_ERR(path1);
5792 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5793 if (IS_ERR(path2)) {
5794 *erp = PTR_ERR(path2);
5798 ex1 = path1[path1->p_depth].p_ext;
5799 ex2 = path2[path2->p_depth].p_ext;
5800 /* Do we have somthing to swap ? */
5801 if (unlikely(!ex2 || !ex1))
5804 e1_blk = le32_to_cpu(ex1->ee_block);
5805 e2_blk = le32_to_cpu(ex2->ee_block);
5806 e1_len = ext4_ext_get_actual_len(ex1);
5807 e2_len = ext4_ext_get_actual_len(ex2);
5810 if (!in_range(lblk1, e1_blk, e1_len) ||
5811 !in_range(lblk2, e2_blk, e2_len)) {
5812 ext4_lblk_t next1, next2;
5814 /* if hole after extent, then go to next extent */
5815 next1 = ext4_ext_next_allocated_block(path1);
5816 next2 = ext4_ext_next_allocated_block(path2);
5817 /* If hole before extent, then shift to that extent */
5822 /* Do we have something to swap */
5823 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5825 /* Move to the rightest boundary */
5826 len = next1 - lblk1;
5827 if (len < next2 - lblk2)
5828 len = next2 - lblk2;
5837 /* Prepare left boundary */
5838 if (e1_blk < lblk1) {
5840 *erp = ext4_force_split_extent_at(handle, inode1,
5845 if (e2_blk < lblk2) {
5847 *erp = ext4_force_split_extent_at(handle, inode2,
5852 /* ext4_split_extent_at() may result in leaf extent split,
5853 * path must to be revalidated. */
5857 /* Prepare right boundary */
5859 if (len > e1_blk + e1_len - lblk1)
5860 len = e1_blk + e1_len - lblk1;
5861 if (len > e2_blk + e2_len - lblk2)
5862 len = e2_blk + e2_len - lblk2;
5864 if (len != e1_len) {
5866 *erp = ext4_force_split_extent_at(handle, inode1,
5867 &path1, lblk1 + len, 0);
5871 if (len != e2_len) {
5873 *erp = ext4_force_split_extent_at(handle, inode2,
5874 &path2, lblk2 + len, 0);
5878 /* ext4_split_extent_at() may result in leaf extent split,
5879 * path must to be revalidated. */
5883 BUG_ON(e2_len != e1_len);
5884 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5887 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5891 /* Both extents are fully inside boundaries. Swap it now */
5893 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5894 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5895 ex1->ee_len = cpu_to_le16(e2_len);
5896 ex2->ee_len = cpu_to_le16(e1_len);
5898 ext4_ext_mark_unwritten(ex2);
5899 if (ext4_ext_is_unwritten(&tmp_ex))
5900 ext4_ext_mark_unwritten(ex1);
5902 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5903 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5904 *erp = ext4_ext_dirty(handle, inode2, path2 +
5908 *erp = ext4_ext_dirty(handle, inode1, path1 +
5911 * Looks scarry ah..? second inode already points to new blocks,
5912 * and it was successfully dirtied. But luckily error may happen
5913 * only due to journal error, so full transaction will be
5920 replaced_count += len;
5924 ext4_ext_drop_refs(path1);
5926 ext4_ext_drop_refs(path2);
5928 path1 = path2 = NULL;
5930 return replaced_count;
5934 * ext4_clu_mapped - determine whether any block in a logical cluster has
5935 * been mapped to a physical cluster
5937 * @inode - file containing the logical cluster
5938 * @lclu - logical cluster of interest
5940 * Returns 1 if any block in the logical cluster is mapped, signifying
5941 * that a physical cluster has been allocated for it. Otherwise,
5942 * returns 0. Can also return negative error codes. Derived from
5943 * ext4_ext_map_blocks().
5945 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5947 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5948 struct ext4_ext_path *path;
5949 int depth, mapped = 0, err = 0;
5950 struct ext4_extent *extent;
5951 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5953 /* search for the extent closest to the first block in the cluster */
5954 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5956 err = PTR_ERR(path);
5961 depth = ext_depth(inode);
5964 * A consistent leaf must not be empty. This situation is possible,
5965 * though, _during_ tree modification, and it's why an assert can't
5966 * be put in ext4_find_extent().
5968 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5969 EXT4_ERROR_INODE(inode,
5970 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5971 (unsigned long) EXT4_C2B(sbi, lclu),
5972 depth, path[depth].p_block);
5973 err = -EFSCORRUPTED;
5977 extent = path[depth].p_ext;
5979 /* can't be mapped if the extent tree is empty */
5983 first_lblk = le32_to_cpu(extent->ee_block);
5984 first_lclu = EXT4_B2C(sbi, first_lblk);
5987 * Three possible outcomes at this point - found extent spanning
5988 * the target cluster, to the left of the target cluster, or to the
5989 * right of the target cluster. The first two cases are handled here.
5990 * The last case indicates the target cluster is not mapped.
5992 if (lclu >= first_lclu) {
5993 last_lclu = EXT4_B2C(sbi, first_lblk +
5994 ext4_ext_get_actual_len(extent) - 1);
5995 if (lclu <= last_lclu) {
5998 first_lblk = ext4_ext_next_allocated_block(path);
5999 first_lclu = EXT4_B2C(sbi, first_lblk);
6000 if (lclu == first_lclu)
6006 ext4_ext_drop_refs(path);
6009 return err ? err : mapped;