2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
71 static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
74 struct ext4_extent_tail *et;
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
86 static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
89 struct ext4_extent_tail *et;
91 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
95 et = find_ext4_extent_tail(eh);
96 et->et_checksum = ext4_extent_block_csum(inode, eh);
99 static int ext4_split_extent(handle_t *handle,
101 struct ext4_ext_path *path,
102 struct ext4_map_blocks *map,
106 static int ext4_split_extent_at(handle_t *handle,
108 struct ext4_ext_path *path,
113 static int ext4_find_delayed_extent(struct inode *inode,
114 struct extent_status *newes);
116 static int ext4_ext_truncate_extend_restart(handle_t *handle,
122 if (!ext4_handle_valid(handle))
124 if (handle->h_buffer_credits > needed)
126 err = ext4_journal_extend(handle, needed);
129 err = ext4_truncate_restart_trans(handle, inode, needed);
141 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
142 struct ext4_ext_path *path)
145 /* path points to block */
146 BUFFER_TRACE(path->p_bh, "get_write_access");
147 return ext4_journal_get_write_access(handle, path->p_bh);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
165 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
167 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
168 /* path points to block */
169 err = __ext4_handle_dirty_metadata(where, line, handle,
172 /* path points to leaf/index in inode body */
173 err = ext4_mark_inode_dirty(handle, inode);
178 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
179 struct ext4_ext_path *path,
183 int depth = path->p_depth;
184 struct ext4_extent *ex;
187 * Try to predict block placement assuming that we are
188 * filling in a file which will eventually be
189 * non-sparse --- i.e., in the case of libbfd writing
190 * an ELF object sections out-of-order but in a way
191 * the eventually results in a contiguous object or
192 * executable file, or some database extending a table
193 * space file. However, this is actually somewhat
194 * non-ideal if we are writing a sparse file such as
195 * qemu or KVM writing a raw image file that is going
196 * to stay fairly sparse, since it will end up
197 * fragmenting the file system's free space. Maybe we
198 * should have some hueristics or some way to allow
199 * userspace to pass a hint to file system,
200 * especially if the latter case turns out to be
203 ex = path[depth].p_ext;
205 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
206 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
208 if (block > ext_block)
209 return ext_pblk + (block - ext_block);
211 return ext_pblk - (ext_block - block);
214 /* it looks like index is empty;
215 * try to find starting block from index itself */
216 if (path[depth].p_bh)
217 return path[depth].p_bh->b_blocknr;
220 /* OK. use inode's group */
221 return ext4_inode_to_goal_block(inode);
225 * Allocation for a meta data block
228 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
229 struct ext4_ext_path *path,
230 struct ext4_extent *ex, int *err, unsigned int flags)
232 ext4_fsblk_t goal, newblock;
234 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
235 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
240 static inline int ext4_ext_space_block(struct inode *inode, int check)
244 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
245 / sizeof(struct ext4_extent);
246 #ifdef AGGRESSIVE_TEST
247 if (!check && size > 6)
253 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
257 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
258 / sizeof(struct ext4_extent_idx);
259 #ifdef AGGRESSIVE_TEST
260 if (!check && size > 5)
266 static inline int ext4_ext_space_root(struct inode *inode, int check)
270 size = sizeof(EXT4_I(inode)->i_data);
271 size -= sizeof(struct ext4_extent_header);
272 size /= sizeof(struct ext4_extent);
273 #ifdef AGGRESSIVE_TEST
274 if (!check && size > 3)
280 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
284 size = sizeof(EXT4_I(inode)->i_data);
285 size -= sizeof(struct ext4_extent_header);
286 size /= sizeof(struct ext4_extent_idx);
287 #ifdef AGGRESSIVE_TEST
288 if (!check && size > 4)
295 * Calculate the number of metadata blocks needed
296 * to allocate @blocks
297 * Worse case is one block per extent
299 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
301 struct ext4_inode_info *ei = EXT4_I(inode);
304 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
305 / sizeof(struct ext4_extent_idx));
308 * If the new delayed allocation block is contiguous with the
309 * previous da block, it can share index blocks with the
310 * previous block, so we only need to allocate a new index
311 * block every idxs leaf blocks. At ldxs**2 blocks, we need
312 * an additional index block, and at ldxs**3 blocks, yet
313 * another index blocks.
315 if (ei->i_da_metadata_calc_len &&
316 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
319 if ((ei->i_da_metadata_calc_len % idxs) == 0)
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
323 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
325 ei->i_da_metadata_calc_len = 0;
327 ei->i_da_metadata_calc_len++;
328 ei->i_da_metadata_calc_last_lblock++;
333 * In the worst case we need a new set of index blocks at
334 * every level of the inode's extent tree.
336 ei->i_da_metadata_calc_len = 1;
337 ei->i_da_metadata_calc_last_lblock = lblock;
338 return ext_depth(inode) + 1;
342 ext4_ext_max_entries(struct inode *inode, int depth)
346 if (depth == ext_depth(inode)) {
348 max = ext4_ext_space_root(inode, 1);
350 max = ext4_ext_space_root_idx(inode, 1);
353 max = ext4_ext_space_block(inode, 1);
355 max = ext4_ext_space_block_idx(inode, 1);
361 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
363 ext4_fsblk_t block = ext4_ext_pblock(ext);
364 int len = ext4_ext_get_actual_len(ext);
365 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
366 ext4_lblk_t last = lblock + len - 1;
370 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
373 static int ext4_valid_extent_idx(struct inode *inode,
374 struct ext4_extent_idx *ext_idx)
376 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
378 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
381 static int ext4_valid_extent_entries(struct inode *inode,
382 struct ext4_extent_header *eh,
385 unsigned short entries;
386 if (eh->eh_entries == 0)
389 entries = le16_to_cpu(eh->eh_entries);
393 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
394 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
395 ext4_fsblk_t pblock = 0;
396 ext4_lblk_t lblock = 0;
397 ext4_lblk_t prev = 0;
400 if (!ext4_valid_extent(inode, ext))
403 /* Check for overlapping extents */
404 lblock = le32_to_cpu(ext->ee_block);
405 len = ext4_ext_get_actual_len(ext);
406 if ((lblock <= prev) && prev) {
407 pblock = ext4_ext_pblock(ext);
408 es->s_last_error_block = cpu_to_le64(pblock);
413 prev = lblock + len - 1;
416 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
418 if (!ext4_valid_extent_idx(inode, ext_idx))
427 static int __ext4_ext_check(const char *function, unsigned int line,
428 struct inode *inode, struct ext4_extent_header *eh,
429 int depth, ext4_fsblk_t pblk)
431 const char *error_msg;
434 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
435 error_msg = "invalid magic";
438 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
439 error_msg = "unexpected eh_depth";
442 if (unlikely(eh->eh_max == 0)) {
443 error_msg = "invalid eh_max";
446 max = ext4_ext_max_entries(inode, depth);
447 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
448 error_msg = "too large eh_max";
451 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
452 error_msg = "invalid eh_entries";
455 if (!ext4_valid_extent_entries(inode, eh, depth)) {
456 error_msg = "invalid extent entries";
459 /* Verify checksum on non-root extent tree nodes */
460 if (ext_depth(inode) != depth &&
461 !ext4_extent_block_csum_verify(inode, eh)) {
462 error_msg = "extent tree corrupted";
468 ext4_error_inode(inode, function, line, 0,
469 "pblk %llu bad header/extent: %s - magic %x, "
470 "entries %u, max %u(%u), depth %u(%u)",
471 (unsigned long long) pblk, error_msg,
472 le16_to_cpu(eh->eh_magic),
473 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
474 max, le16_to_cpu(eh->eh_depth), depth);
478 #define ext4_ext_check(inode, eh, depth, pblk) \
479 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
481 int ext4_ext_check_inode(struct inode *inode)
483 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
486 static struct buffer_head *
487 __read_extent_tree_block(const char *function, unsigned int line,
488 struct inode *inode, ext4_fsblk_t pblk, int depth,
491 struct buffer_head *bh;
494 bh = sb_getblk(inode->i_sb, pblk);
496 return ERR_PTR(-ENOMEM);
498 if (!bh_uptodate_or_lock(bh)) {
499 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
500 err = bh_submit_read(bh);
504 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
506 err = __ext4_ext_check(function, line, inode,
507 ext_block_hdr(bh), depth, pblk);
510 set_buffer_verified(bh);
512 * If this is a leaf block, cache all of its entries
514 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
515 struct ext4_extent_header *eh = ext_block_hdr(bh);
516 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
517 ext4_lblk_t prev = 0;
520 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
521 unsigned int status = EXTENT_STATUS_WRITTEN;
522 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
523 int len = ext4_ext_get_actual_len(ex);
525 if (prev && (prev != lblk))
526 ext4_es_cache_extent(inode, prev,
530 if (ext4_ext_is_unwritten(ex))
531 status = EXTENT_STATUS_UNWRITTEN;
532 ext4_es_cache_extent(inode, lblk, len,
533 ext4_ext_pblock(ex), status);
544 #define read_extent_tree_block(inode, pblk, depth, flags) \
545 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
549 * This function is called to cache a file's extent information in the
552 int ext4_ext_precache(struct inode *inode)
554 struct ext4_inode_info *ei = EXT4_I(inode);
555 struct ext4_ext_path *path = NULL;
556 struct buffer_head *bh;
557 int i = 0, depth, ret = 0;
559 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
560 return 0; /* not an extent-mapped inode */
562 down_read(&ei->i_data_sem);
563 depth = ext_depth(inode);
565 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
568 up_read(&ei->i_data_sem);
572 /* Don't cache anything if there are no external extent blocks */
575 path[0].p_hdr = ext_inode_hdr(inode);
576 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
579 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
582 * If this is a leaf block or we've reached the end of
583 * the index block, go up
586 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
587 brelse(path[i].p_bh);
592 bh = read_extent_tree_block(inode,
593 ext4_idx_pblock(path[i].p_idx++),
595 EXT4_EX_FORCE_CACHE);
602 path[i].p_hdr = ext_block_hdr(bh);
603 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
605 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
607 up_read(&ei->i_data_sem);
608 ext4_ext_drop_refs(path);
614 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
616 int k, l = path->p_depth;
619 for (k = 0; k <= l; k++, path++) {
621 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
622 ext4_idx_pblock(path->p_idx));
623 } else if (path->p_ext) {
624 ext_debug(" %d:[%d]%d:%llu ",
625 le32_to_cpu(path->p_ext->ee_block),
626 ext4_ext_is_unwritten(path->p_ext),
627 ext4_ext_get_actual_len(path->p_ext),
628 ext4_ext_pblock(path->p_ext));
635 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
637 int depth = ext_depth(inode);
638 struct ext4_extent_header *eh;
639 struct ext4_extent *ex;
645 eh = path[depth].p_hdr;
646 ex = EXT_FIRST_EXTENT(eh);
648 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
650 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
651 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
652 ext4_ext_is_unwritten(ex),
653 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
658 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
659 ext4_fsblk_t newblock, int level)
661 int depth = ext_depth(inode);
662 struct ext4_extent *ex;
664 if (depth != level) {
665 struct ext4_extent_idx *idx;
666 idx = path[level].p_idx;
667 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
668 ext_debug("%d: move %d:%llu in new index %llu\n", level,
669 le32_to_cpu(idx->ei_block),
670 ext4_idx_pblock(idx),
678 ex = path[depth].p_ext;
679 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
680 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
681 le32_to_cpu(ex->ee_block),
683 ext4_ext_is_unwritten(ex),
684 ext4_ext_get_actual_len(ex),
691 #define ext4_ext_show_path(inode, path)
692 #define ext4_ext_show_leaf(inode, path)
693 #define ext4_ext_show_move(inode, path, newblock, level)
696 void ext4_ext_drop_refs(struct ext4_ext_path *path)
698 int depth = path->p_depth;
701 for (i = 0; i <= depth; i++, path++)
709 * ext4_ext_binsearch_idx:
710 * binary search for the closest index of the given block
711 * the header must be checked before calling this
714 ext4_ext_binsearch_idx(struct inode *inode,
715 struct ext4_ext_path *path, ext4_lblk_t block)
717 struct ext4_extent_header *eh = path->p_hdr;
718 struct ext4_extent_idx *r, *l, *m;
721 ext_debug("binsearch for %u(idx): ", block);
723 l = EXT_FIRST_INDEX(eh) + 1;
724 r = EXT_LAST_INDEX(eh);
727 if (block < le32_to_cpu(m->ei_block))
731 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
732 m, le32_to_cpu(m->ei_block),
733 r, le32_to_cpu(r->ei_block));
737 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
738 ext4_idx_pblock(path->p_idx));
740 #ifdef CHECK_BINSEARCH
742 struct ext4_extent_idx *chix, *ix;
745 chix = ix = EXT_FIRST_INDEX(eh);
746 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
748 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
749 printk(KERN_DEBUG "k=%d, ix=0x%p, "
751 ix, EXT_FIRST_INDEX(eh));
752 printk(KERN_DEBUG "%u <= %u\n",
753 le32_to_cpu(ix->ei_block),
754 le32_to_cpu(ix[-1].ei_block));
756 BUG_ON(k && le32_to_cpu(ix->ei_block)
757 <= le32_to_cpu(ix[-1].ei_block));
758 if (block < le32_to_cpu(ix->ei_block))
762 BUG_ON(chix != path->p_idx);
769 * ext4_ext_binsearch:
770 * binary search for closest extent of the given block
771 * the header must be checked before calling this
774 ext4_ext_binsearch(struct inode *inode,
775 struct ext4_ext_path *path, ext4_lblk_t block)
777 struct ext4_extent_header *eh = path->p_hdr;
778 struct ext4_extent *r, *l, *m;
780 if (eh->eh_entries == 0) {
782 * this leaf is empty:
783 * we get such a leaf in split/add case
788 ext_debug("binsearch for %u: ", block);
790 l = EXT_FIRST_EXTENT(eh) + 1;
791 r = EXT_LAST_EXTENT(eh);
795 if (block < le32_to_cpu(m->ee_block))
799 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
800 m, le32_to_cpu(m->ee_block),
801 r, le32_to_cpu(r->ee_block));
805 ext_debug(" -> %d:%llu:[%d]%d ",
806 le32_to_cpu(path->p_ext->ee_block),
807 ext4_ext_pblock(path->p_ext),
808 ext4_ext_is_unwritten(path->p_ext),
809 ext4_ext_get_actual_len(path->p_ext));
811 #ifdef CHECK_BINSEARCH
813 struct ext4_extent *chex, *ex;
816 chex = ex = EXT_FIRST_EXTENT(eh);
817 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
818 BUG_ON(k && le32_to_cpu(ex->ee_block)
819 <= le32_to_cpu(ex[-1].ee_block));
820 if (block < le32_to_cpu(ex->ee_block))
824 BUG_ON(chex != path->p_ext);
830 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
832 struct ext4_extent_header *eh;
834 eh = ext_inode_hdr(inode);
837 eh->eh_magic = EXT4_EXT_MAGIC;
838 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
839 ext4_mark_inode_dirty(handle, inode);
843 struct ext4_ext_path *
844 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
845 struct ext4_ext_path *path, int flags)
847 struct ext4_extent_header *eh;
848 struct buffer_head *bh;
849 short int depth, i, ppos = 0, alloc = 0;
852 eh = ext_inode_hdr(inode);
853 depth = ext_depth(inode);
855 /* account possible depth increase */
857 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
860 return ERR_PTR(-ENOMEM);
867 /* walk through the tree */
869 ext_debug("depth %d: num %d, max %d\n",
870 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
872 ext4_ext_binsearch_idx(inode, path + ppos, block);
873 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
874 path[ppos].p_depth = i;
875 path[ppos].p_ext = NULL;
877 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
884 eh = ext_block_hdr(bh);
886 if (unlikely(ppos > depth)) {
888 EXT4_ERROR_INODE(inode,
889 "ppos %d > depth %d", ppos, depth);
893 path[ppos].p_bh = bh;
894 path[ppos].p_hdr = eh;
897 path[ppos].p_depth = i;
898 path[ppos].p_ext = NULL;
899 path[ppos].p_idx = NULL;
902 ext4_ext_binsearch(inode, path + ppos, block);
903 /* if not an empty leaf */
904 if (path[ppos].p_ext)
905 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
907 ext4_ext_show_path(inode, path);
912 ext4_ext_drop_refs(path);
919 * ext4_ext_insert_index:
920 * insert new index [@logical;@ptr] into the block at @curp;
921 * check where to insert: before @curp or after @curp
923 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
924 struct ext4_ext_path *curp,
925 int logical, ext4_fsblk_t ptr)
927 struct ext4_extent_idx *ix;
930 err = ext4_ext_get_access(handle, inode, curp);
934 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
935 EXT4_ERROR_INODE(inode,
936 "logical %d == ei_block %d!",
937 logical, le32_to_cpu(curp->p_idx->ei_block));
941 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
942 >= le16_to_cpu(curp->p_hdr->eh_max))) {
943 EXT4_ERROR_INODE(inode,
944 "eh_entries %d >= eh_max %d!",
945 le16_to_cpu(curp->p_hdr->eh_entries),
946 le16_to_cpu(curp->p_hdr->eh_max));
950 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
952 ext_debug("insert new index %d after: %llu\n", logical, ptr);
953 ix = curp->p_idx + 1;
956 ext_debug("insert new index %d before: %llu\n", logical, ptr);
960 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
963 ext_debug("insert new index %d: "
964 "move %d indices from 0x%p to 0x%p\n",
965 logical, len, ix, ix + 1);
966 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
969 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
970 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
974 ix->ei_block = cpu_to_le32(logical);
975 ext4_idx_store_pblock(ix, ptr);
976 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
978 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
979 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
983 err = ext4_ext_dirty(handle, inode, curp);
984 ext4_std_error(inode->i_sb, err);
991 * inserts new subtree into the path, using free index entry
993 * - allocates all needed blocks (new leaf and all intermediate index blocks)
994 * - makes decision where to split
995 * - moves remaining extents and index entries (right to the split point)
996 * into the newly allocated blocks
997 * - initializes subtree
999 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1001 struct ext4_ext_path *path,
1002 struct ext4_extent *newext, int at)
1004 struct buffer_head *bh = NULL;
1005 int depth = ext_depth(inode);
1006 struct ext4_extent_header *neh;
1007 struct ext4_extent_idx *fidx;
1008 int i = at, k, m, a;
1009 ext4_fsblk_t newblock, oldblock;
1011 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1014 /* make decision: where to split? */
1015 /* FIXME: now decision is simplest: at current extent */
1017 /* if current leaf will be split, then we should use
1018 * border from split point */
1019 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1020 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1023 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1024 border = path[depth].p_ext[1].ee_block;
1025 ext_debug("leaf will be split."
1026 " next leaf starts at %d\n",
1027 le32_to_cpu(border));
1029 border = newext->ee_block;
1030 ext_debug("leaf will be added."
1031 " next leaf starts at %d\n",
1032 le32_to_cpu(border));
1036 * If error occurs, then we break processing
1037 * and mark filesystem read-only. index won't
1038 * be inserted and tree will be in consistent
1039 * state. Next mount will repair buffers too.
1043 * Get array to track all allocated blocks.
1044 * We need this to handle errors and free blocks
1047 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1051 /* allocate all needed blocks */
1052 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1053 for (a = 0; a < depth - at; a++) {
1054 newblock = ext4_ext_new_meta_block(handle, inode, path,
1055 newext, &err, flags);
1058 ablocks[a] = newblock;
1061 /* initialize new leaf */
1062 newblock = ablocks[--a];
1063 if (unlikely(newblock == 0)) {
1064 EXT4_ERROR_INODE(inode, "newblock == 0!");
1068 bh = sb_getblk(inode->i_sb, newblock);
1069 if (unlikely(!bh)) {
1075 err = ext4_journal_get_create_access(handle, bh);
1079 neh = ext_block_hdr(bh);
1080 neh->eh_entries = 0;
1081 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1082 neh->eh_magic = EXT4_EXT_MAGIC;
1085 /* move remainder of path[depth] to the new leaf */
1086 if (unlikely(path[depth].p_hdr->eh_entries !=
1087 path[depth].p_hdr->eh_max)) {
1088 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1089 path[depth].p_hdr->eh_entries,
1090 path[depth].p_hdr->eh_max);
1094 /* start copy from next extent */
1095 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1096 ext4_ext_show_move(inode, path, newblock, depth);
1098 struct ext4_extent *ex;
1099 ex = EXT_FIRST_EXTENT(neh);
1100 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1101 le16_add_cpu(&neh->eh_entries, m);
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);
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));
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 ext4_extent_block_csum_set(inode, neh);
1184 set_buffer_uptodate(bh);
1187 err = ext4_handle_dirty_metadata(handle, inode, bh);
1193 /* correct old index */
1195 err = ext4_ext_get_access(handle, inode, path + i);
1198 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1199 err = ext4_ext_dirty(handle, inode, path + i);
1207 /* insert new index */
1208 err = ext4_ext_insert_index(handle, inode, path + at,
1209 le32_to_cpu(border), newblock);
1213 if (buffer_locked(bh))
1219 /* free all allocated blocks in error case */
1220 for (i = 0; i < depth; i++) {
1223 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1224 EXT4_FREE_BLOCKS_METADATA);
1233 * ext4_ext_grow_indepth:
1234 * implements tree growing procedure:
1235 * - allocates new block
1236 * - moves top-level data (index block or leaf) into the new block
1237 * - initializes new top-level, creating index that points to the
1238 * just created block
1240 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1242 struct ext4_extent *newext)
1244 struct ext4_extent_header *neh;
1245 struct buffer_head *bh;
1246 ext4_fsblk_t newblock;
1249 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1250 newext, &err, flags);
1254 bh = sb_getblk(inode->i_sb, newblock);
1259 err = ext4_journal_get_create_access(handle, bh);
1265 /* move top-level index/leaf into new block */
1266 memmove(bh->b_data, EXT4_I(inode)->i_data,
1267 sizeof(EXT4_I(inode)->i_data));
1269 /* set size of new block */
1270 neh = ext_block_hdr(bh);
1271 /* old root could have indexes or leaves
1272 * so calculate e_max right way */
1273 if (ext_depth(inode))
1274 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1276 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1277 neh->eh_magic = EXT4_EXT_MAGIC;
1278 ext4_extent_block_csum_set(inode, neh);
1279 set_buffer_uptodate(bh);
1282 err = ext4_handle_dirty_metadata(handle, inode, bh);
1286 /* Update top-level index: num,max,pointer */
1287 neh = ext_inode_hdr(inode);
1288 neh->eh_entries = cpu_to_le16(1);
1289 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1290 if (neh->eh_depth == 0) {
1291 /* Root extent block becomes index block */
1292 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1293 EXT_FIRST_INDEX(neh)->ei_block =
1294 EXT_FIRST_EXTENT(neh)->ee_block;
1296 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1297 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1298 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1299 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1301 le16_add_cpu(&neh->eh_depth, 1);
1302 ext4_mark_inode_dirty(handle, inode);
1310 * ext4_ext_create_new_leaf:
1311 * finds empty index and adds new leaf.
1312 * if no free index is found, then it requests in-depth growing.
1314 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1315 unsigned int mb_flags,
1316 unsigned int gb_flags,
1317 struct ext4_ext_path *path,
1318 struct ext4_extent *newext)
1320 struct ext4_ext_path *curp;
1321 int depth, i, err = 0;
1324 i = depth = ext_depth(inode);
1326 /* walk up to the tree and look for free index entry */
1327 curp = path + depth;
1328 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1333 /* we use already allocated block for index block,
1334 * so subsequent data blocks should be contiguous */
1335 if (EXT_HAS_FREE_INDEX(curp)) {
1336 /* if we found index with free entry, then use that
1337 * entry: create all needed subtree and add new leaf */
1338 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1343 ext4_ext_drop_refs(path);
1344 path = ext4_ext_find_extent(inode,
1345 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1348 err = PTR_ERR(path);
1350 /* tree is full, time to grow in depth */
1351 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1356 ext4_ext_drop_refs(path);
1357 path = ext4_ext_find_extent(inode,
1358 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1361 err = PTR_ERR(path);
1366 * only first (depth 0 -> 1) produces free space;
1367 * in all other cases we have to split the grown tree
1369 depth = ext_depth(inode);
1370 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1371 /* now we need to split */
1381 * search the closest allocated block to the left for *logical
1382 * and returns it at @logical + it's physical address at @phys
1383 * if *logical is the smallest allocated block, the function
1384 * returns 0 at @phys
1385 * return value contains 0 (success) or error code
1387 static int ext4_ext_search_left(struct inode *inode,
1388 struct ext4_ext_path *path,
1389 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1391 struct ext4_extent_idx *ix;
1392 struct ext4_extent *ex;
1395 if (unlikely(path == NULL)) {
1396 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1399 depth = path->p_depth;
1402 if (depth == 0 && path->p_ext == NULL)
1405 /* usually extent in the path covers blocks smaller
1406 * then *logical, but it can be that extent is the
1407 * first one in the file */
1409 ex = path[depth].p_ext;
1410 ee_len = ext4_ext_get_actual_len(ex);
1411 if (*logical < le32_to_cpu(ex->ee_block)) {
1412 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1413 EXT4_ERROR_INODE(inode,
1414 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1415 *logical, le32_to_cpu(ex->ee_block));
1418 while (--depth >= 0) {
1419 ix = path[depth].p_idx;
1420 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1421 EXT4_ERROR_INODE(inode,
1422 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1423 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1424 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1425 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1433 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1434 EXT4_ERROR_INODE(inode,
1435 "logical %d < ee_block %d + ee_len %d!",
1436 *logical, le32_to_cpu(ex->ee_block), ee_len);
1440 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1441 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1446 * search the closest allocated block to the right for *logical
1447 * and returns it at @logical + it's physical address at @phys
1448 * if *logical is the largest allocated block, the function
1449 * returns 0 at @phys
1450 * return value contains 0 (success) or error code
1452 static int ext4_ext_search_right(struct inode *inode,
1453 struct ext4_ext_path *path,
1454 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1455 struct ext4_extent **ret_ex)
1457 struct buffer_head *bh = NULL;
1458 struct ext4_extent_header *eh;
1459 struct ext4_extent_idx *ix;
1460 struct ext4_extent *ex;
1462 int depth; /* Note, NOT eh_depth; depth from top of tree */
1465 if (unlikely(path == NULL)) {
1466 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1469 depth = path->p_depth;
1472 if (depth == 0 && path->p_ext == NULL)
1475 /* usually extent in the path covers blocks smaller
1476 * then *logical, but it can be that extent is the
1477 * first one in the file */
1479 ex = path[depth].p_ext;
1480 ee_len = ext4_ext_get_actual_len(ex);
1481 if (*logical < le32_to_cpu(ex->ee_block)) {
1482 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1483 EXT4_ERROR_INODE(inode,
1484 "first_extent(path[%d].p_hdr) != ex",
1488 while (--depth >= 0) {
1489 ix = path[depth].p_idx;
1490 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1491 EXT4_ERROR_INODE(inode,
1492 "ix != EXT_FIRST_INDEX *logical %d!",
1500 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1501 EXT4_ERROR_INODE(inode,
1502 "logical %d < ee_block %d + ee_len %d!",
1503 *logical, le32_to_cpu(ex->ee_block), ee_len);
1507 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1508 /* next allocated block in this leaf */
1513 /* go up and search for index to the right */
1514 while (--depth >= 0) {
1515 ix = path[depth].p_idx;
1516 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1520 /* we've gone up to the root and found no index to the right */
1524 /* we've found index to the right, let's
1525 * follow it and find the closest allocated
1526 * block to the right */
1528 block = ext4_idx_pblock(ix);
1529 while (++depth < path->p_depth) {
1530 /* subtract from p_depth to get proper eh_depth */
1531 bh = read_extent_tree_block(inode, block,
1532 path->p_depth - depth, 0);
1535 eh = ext_block_hdr(bh);
1536 ix = EXT_FIRST_INDEX(eh);
1537 block = ext4_idx_pblock(ix);
1541 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1544 eh = ext_block_hdr(bh);
1545 ex = EXT_FIRST_EXTENT(eh);
1547 *logical = le32_to_cpu(ex->ee_block);
1548 *phys = ext4_ext_pblock(ex);
1556 * ext4_ext_next_allocated_block:
1557 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1558 * NOTE: it considers block number from index entry as
1559 * allocated block. Thus, index entries have to be consistent
1563 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1567 BUG_ON(path == NULL);
1568 depth = path->p_depth;
1570 if (depth == 0 && path->p_ext == NULL)
1571 return EXT_MAX_BLOCKS;
1573 while (depth >= 0) {
1574 if (depth == path->p_depth) {
1576 if (path[depth].p_ext &&
1577 path[depth].p_ext !=
1578 EXT_LAST_EXTENT(path[depth].p_hdr))
1579 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1582 if (path[depth].p_idx !=
1583 EXT_LAST_INDEX(path[depth].p_hdr))
1584 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1589 return EXT_MAX_BLOCKS;
1593 * ext4_ext_next_leaf_block:
1594 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1596 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1600 BUG_ON(path == NULL);
1601 depth = path->p_depth;
1603 /* zero-tree has no leaf blocks at all */
1605 return EXT_MAX_BLOCKS;
1607 /* go to index block */
1610 while (depth >= 0) {
1611 if (path[depth].p_idx !=
1612 EXT_LAST_INDEX(path[depth].p_hdr))
1613 return (ext4_lblk_t)
1614 le32_to_cpu(path[depth].p_idx[1].ei_block);
1618 return EXT_MAX_BLOCKS;
1622 * ext4_ext_correct_indexes:
1623 * if leaf gets modified and modified extent is first in the leaf,
1624 * then we have to correct all indexes above.
1625 * TODO: do we need to correct tree in all cases?
1627 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1628 struct ext4_ext_path *path)
1630 struct ext4_extent_header *eh;
1631 int depth = ext_depth(inode);
1632 struct ext4_extent *ex;
1636 eh = path[depth].p_hdr;
1637 ex = path[depth].p_ext;
1639 if (unlikely(ex == NULL || eh == NULL)) {
1640 EXT4_ERROR_INODE(inode,
1641 "ex %p == NULL or eh %p == NULL", ex, eh);
1646 /* there is no tree at all */
1650 if (ex != EXT_FIRST_EXTENT(eh)) {
1651 /* we correct tree if first leaf got modified only */
1656 * TODO: we need correction if border is smaller than current one
1659 border = path[depth].p_ext->ee_block;
1660 err = ext4_ext_get_access(handle, inode, path + k);
1663 path[k].p_idx->ei_block = border;
1664 err = ext4_ext_dirty(handle, inode, path + k);
1669 /* change all left-side indexes */
1670 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1672 err = ext4_ext_get_access(handle, inode, path + k);
1675 path[k].p_idx->ei_block = border;
1676 err = ext4_ext_dirty(handle, inode, path + k);
1685 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1686 struct ext4_extent *ex2)
1688 unsigned short ext1_ee_len, ext2_ee_len;
1691 * Make sure that both extents are initialized. We don't merge
1692 * unwritten extents so that we can be sure that end_io code has
1693 * the extent that was written properly split out and conversion to
1694 * initialized is trivial.
1696 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1699 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1700 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1702 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1703 le32_to_cpu(ex2->ee_block))
1707 * To allow future support for preallocated extents to be added
1708 * as an RO_COMPAT feature, refuse to merge to extents if
1709 * this can result in the top bit of ee_len being set.
1711 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1713 if (ext4_ext_is_unwritten(ex1) &&
1714 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1715 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1716 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1718 #ifdef AGGRESSIVE_TEST
1719 if (ext1_ee_len >= 4)
1723 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1729 * This function tries to merge the "ex" extent to the next extent in the tree.
1730 * It always tries to merge towards right. If you want to merge towards
1731 * left, pass "ex - 1" as argument instead of "ex".
1732 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1733 * 1 if they got merged.
1735 static int ext4_ext_try_to_merge_right(struct inode *inode,
1736 struct ext4_ext_path *path,
1737 struct ext4_extent *ex)
1739 struct ext4_extent_header *eh;
1740 unsigned int depth, len;
1741 int merge_done = 0, unwritten;
1743 depth = ext_depth(inode);
1744 BUG_ON(path[depth].p_hdr == NULL);
1745 eh = path[depth].p_hdr;
1747 while (ex < EXT_LAST_EXTENT(eh)) {
1748 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1750 /* merge with next extent! */
1751 unwritten = ext4_ext_is_unwritten(ex);
1752 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1753 + ext4_ext_get_actual_len(ex + 1));
1755 ext4_ext_mark_unwritten(ex);
1757 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1758 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1759 * sizeof(struct ext4_extent);
1760 memmove(ex + 1, ex + 2, len);
1762 le16_add_cpu(&eh->eh_entries, -1);
1764 WARN_ON(eh->eh_entries == 0);
1765 if (!eh->eh_entries)
1766 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1773 * This function does a very simple check to see if we can collapse
1774 * an extent tree with a single extent tree leaf block into the inode.
1776 static void ext4_ext_try_to_merge_up(handle_t *handle,
1777 struct inode *inode,
1778 struct ext4_ext_path *path)
1781 unsigned max_root = ext4_ext_space_root(inode, 0);
1784 if ((path[0].p_depth != 1) ||
1785 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1786 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1790 * We need to modify the block allocation bitmap and the block
1791 * group descriptor to release the extent tree block. If we
1792 * can't get the journal credits, give up.
1794 if (ext4_journal_extend(handle, 2))
1798 * Copy the extent data up to the inode
1800 blk = ext4_idx_pblock(path[0].p_idx);
1801 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1802 sizeof(struct ext4_extent_idx);
1803 s += sizeof(struct ext4_extent_header);
1805 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1806 path[0].p_depth = 0;
1807 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1808 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1809 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1811 brelse(path[1].p_bh);
1812 ext4_free_blocks(handle, inode, NULL, blk, 1,
1813 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1817 * This function tries to merge the @ex extent to neighbours in the tree.
1818 * return 1 if merge left else 0.
1820 static void ext4_ext_try_to_merge(handle_t *handle,
1821 struct inode *inode,
1822 struct ext4_ext_path *path,
1823 struct ext4_extent *ex) {
1824 struct ext4_extent_header *eh;
1828 depth = ext_depth(inode);
1829 BUG_ON(path[depth].p_hdr == NULL);
1830 eh = path[depth].p_hdr;
1832 if (ex > EXT_FIRST_EXTENT(eh))
1833 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1836 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1838 ext4_ext_try_to_merge_up(handle, inode, path);
1842 * check if a portion of the "newext" extent overlaps with an
1845 * If there is an overlap discovered, it updates the length of the newext
1846 * such that there will be no overlap, and then returns 1.
1847 * If there is no overlap found, it returns 0.
1849 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1850 struct inode *inode,
1851 struct ext4_extent *newext,
1852 struct ext4_ext_path *path)
1855 unsigned int depth, len1;
1856 unsigned int ret = 0;
1858 b1 = le32_to_cpu(newext->ee_block);
1859 len1 = ext4_ext_get_actual_len(newext);
1860 depth = ext_depth(inode);
1861 if (!path[depth].p_ext)
1863 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1866 * get the next allocated block if the extent in the path
1867 * is before the requested block(s)
1870 b2 = ext4_ext_next_allocated_block(path);
1871 if (b2 == EXT_MAX_BLOCKS)
1873 b2 = EXT4_LBLK_CMASK(sbi, b2);
1876 /* check for wrap through zero on extent logical start block*/
1877 if (b1 + len1 < b1) {
1878 len1 = EXT_MAX_BLOCKS - b1;
1879 newext->ee_len = cpu_to_le16(len1);
1883 /* check for overlap */
1884 if (b1 + len1 > b2) {
1885 newext->ee_len = cpu_to_le16(b2 - b1);
1893 * ext4_ext_insert_extent:
1894 * tries to merge requsted extent into the existing extent or
1895 * inserts requested extent as new one into the tree,
1896 * creating new leaf in the no-space case.
1898 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1899 struct ext4_ext_path *path,
1900 struct ext4_extent *newext, int gb_flags)
1902 struct ext4_extent_header *eh;
1903 struct ext4_extent *ex, *fex;
1904 struct ext4_extent *nearex; /* nearest extent */
1905 struct ext4_ext_path *npath = NULL;
1906 int depth, len, err;
1908 int mb_flags = 0, unwritten;
1910 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1911 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1914 depth = ext_depth(inode);
1915 ex = path[depth].p_ext;
1916 eh = path[depth].p_hdr;
1917 if (unlikely(path[depth].p_hdr == NULL)) {
1918 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1922 /* try to insert block into found extent and return */
1923 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1926 * Try to see whether we should rather test the extent on
1927 * right from ex, or from the left of ex. This is because
1928 * ext4_ext_find_extent() can return either extent on the
1929 * left, or on the right from the searched position. This
1930 * will make merging more effective.
1932 if (ex < EXT_LAST_EXTENT(eh) &&
1933 (le32_to_cpu(ex->ee_block) +
1934 ext4_ext_get_actual_len(ex) <
1935 le32_to_cpu(newext->ee_block))) {
1938 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1939 (le32_to_cpu(newext->ee_block) +
1940 ext4_ext_get_actual_len(newext) <
1941 le32_to_cpu(ex->ee_block)))
1944 /* Try to append newex to the ex */
1945 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1946 ext_debug("append [%d]%d block to %u:[%d]%d"
1948 ext4_ext_is_unwritten(newext),
1949 ext4_ext_get_actual_len(newext),
1950 le32_to_cpu(ex->ee_block),
1951 ext4_ext_is_unwritten(ex),
1952 ext4_ext_get_actual_len(ex),
1953 ext4_ext_pblock(ex));
1954 err = ext4_ext_get_access(handle, inode,
1958 unwritten = ext4_ext_is_unwritten(ex);
1959 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1960 + ext4_ext_get_actual_len(newext));
1962 ext4_ext_mark_unwritten(ex);
1963 eh = path[depth].p_hdr;
1969 /* Try to prepend newex to the ex */
1970 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1971 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1973 le32_to_cpu(newext->ee_block),
1974 ext4_ext_is_unwritten(newext),
1975 ext4_ext_get_actual_len(newext),
1976 le32_to_cpu(ex->ee_block),
1977 ext4_ext_is_unwritten(ex),
1978 ext4_ext_get_actual_len(ex),
1979 ext4_ext_pblock(ex));
1980 err = ext4_ext_get_access(handle, inode,
1985 unwritten = ext4_ext_is_unwritten(ex);
1986 ex->ee_block = newext->ee_block;
1987 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1988 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1989 + ext4_ext_get_actual_len(newext));
1991 ext4_ext_mark_unwritten(ex);
1992 eh = path[depth].p_hdr;
1998 depth = ext_depth(inode);
1999 eh = path[depth].p_hdr;
2000 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2003 /* probably next leaf has space for us? */
2004 fex = EXT_LAST_EXTENT(eh);
2005 next = EXT_MAX_BLOCKS;
2006 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2007 next = ext4_ext_next_leaf_block(path);
2008 if (next != EXT_MAX_BLOCKS) {
2009 ext_debug("next leaf block - %u\n", next);
2010 BUG_ON(npath != NULL);
2011 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2013 return PTR_ERR(npath);
2014 BUG_ON(npath->p_depth != path->p_depth);
2015 eh = npath[depth].p_hdr;
2016 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2017 ext_debug("next leaf isn't full(%d)\n",
2018 le16_to_cpu(eh->eh_entries));
2022 ext_debug("next leaf has no free space(%d,%d)\n",
2023 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2027 * There is no free space in the found leaf.
2028 * We're gonna add a new leaf in the tree.
2030 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2031 mb_flags = EXT4_MB_USE_RESERVED;
2032 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2036 depth = ext_depth(inode);
2037 eh = path[depth].p_hdr;
2040 nearex = path[depth].p_ext;
2042 err = ext4_ext_get_access(handle, inode, path + depth);
2047 /* there is no extent in this leaf, create first one */
2048 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2049 le32_to_cpu(newext->ee_block),
2050 ext4_ext_pblock(newext),
2051 ext4_ext_is_unwritten(newext),
2052 ext4_ext_get_actual_len(newext));
2053 nearex = EXT_FIRST_EXTENT(eh);
2055 if (le32_to_cpu(newext->ee_block)
2056 > le32_to_cpu(nearex->ee_block)) {
2058 ext_debug("insert %u:%llu:[%d]%d before: "
2060 le32_to_cpu(newext->ee_block),
2061 ext4_ext_pblock(newext),
2062 ext4_ext_is_unwritten(newext),
2063 ext4_ext_get_actual_len(newext),
2068 BUG_ON(newext->ee_block == nearex->ee_block);
2069 ext_debug("insert %u:%llu:[%d]%d after: "
2071 le32_to_cpu(newext->ee_block),
2072 ext4_ext_pblock(newext),
2073 ext4_ext_is_unwritten(newext),
2074 ext4_ext_get_actual_len(newext),
2077 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2079 ext_debug("insert %u:%llu:[%d]%d: "
2080 "move %d extents from 0x%p to 0x%p\n",
2081 le32_to_cpu(newext->ee_block),
2082 ext4_ext_pblock(newext),
2083 ext4_ext_is_unwritten(newext),
2084 ext4_ext_get_actual_len(newext),
2085 len, nearex, nearex + 1);
2086 memmove(nearex + 1, nearex,
2087 len * sizeof(struct ext4_extent));
2091 le16_add_cpu(&eh->eh_entries, 1);
2092 path[depth].p_ext = nearex;
2093 nearex->ee_block = newext->ee_block;
2094 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2095 nearex->ee_len = newext->ee_len;
2098 /* try to merge extents */
2099 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2100 ext4_ext_try_to_merge(handle, inode, path, nearex);
2103 /* time to correct all indexes above */
2104 err = ext4_ext_correct_indexes(handle, inode, path);
2108 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2112 ext4_ext_drop_refs(npath);
2118 static int ext4_fill_fiemap_extents(struct inode *inode,
2119 ext4_lblk_t block, ext4_lblk_t num,
2120 struct fiemap_extent_info *fieinfo)
2122 struct ext4_ext_path *path = NULL;
2123 struct ext4_extent *ex;
2124 struct extent_status es;
2125 ext4_lblk_t next, next_del, start = 0, end = 0;
2126 ext4_lblk_t last = block + num;
2127 int exists, depth = 0, err = 0;
2128 unsigned int flags = 0;
2129 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2131 while (block < last && block != EXT_MAX_BLOCKS) {
2133 /* find extent for this block */
2134 down_read(&EXT4_I(inode)->i_data_sem);
2136 if (path && ext_depth(inode) != depth) {
2137 /* depth was changed. we have to realloc path */
2142 path = ext4_ext_find_extent(inode, block, path, 0);
2144 up_read(&EXT4_I(inode)->i_data_sem);
2145 err = PTR_ERR(path);
2150 depth = ext_depth(inode);
2151 if (unlikely(path[depth].p_hdr == NULL)) {
2152 up_read(&EXT4_I(inode)->i_data_sem);
2153 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2157 ex = path[depth].p_ext;
2158 next = ext4_ext_next_allocated_block(path);
2159 ext4_ext_drop_refs(path);
2164 /* there is no extent yet, so try to allocate
2165 * all requested space */
2168 } else if (le32_to_cpu(ex->ee_block) > block) {
2169 /* need to allocate space before found extent */
2171 end = le32_to_cpu(ex->ee_block);
2172 if (block + num < end)
2174 } else if (block >= le32_to_cpu(ex->ee_block)
2175 + ext4_ext_get_actual_len(ex)) {
2176 /* need to allocate space after found extent */
2181 } else if (block >= le32_to_cpu(ex->ee_block)) {
2183 * some part of requested space is covered
2187 end = le32_to_cpu(ex->ee_block)
2188 + ext4_ext_get_actual_len(ex);
2189 if (block + num < end)
2195 BUG_ON(end <= start);
2199 es.es_len = end - start;
2202 es.es_lblk = le32_to_cpu(ex->ee_block);
2203 es.es_len = ext4_ext_get_actual_len(ex);
2204 es.es_pblk = ext4_ext_pblock(ex);
2205 if (ext4_ext_is_unwritten(ex))
2206 flags |= FIEMAP_EXTENT_UNWRITTEN;
2210 * Find delayed extent and update es accordingly. We call
2211 * it even in !exists case to find out whether es is the
2212 * last existing extent or not.
2214 next_del = ext4_find_delayed_extent(inode, &es);
2215 if (!exists && next_del) {
2217 flags |= (FIEMAP_EXTENT_DELALLOC |
2218 FIEMAP_EXTENT_UNKNOWN);
2220 up_read(&EXT4_I(inode)->i_data_sem);
2222 if (unlikely(es.es_len == 0)) {
2223 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2229 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2230 * we need to check next == EXT_MAX_BLOCKS because it is
2231 * possible that an extent is with unwritten and delayed
2232 * status due to when an extent is delayed allocated and
2233 * is allocated by fallocate status tree will track both of
2236 * So we could return a unwritten and delayed extent, and
2237 * its block is equal to 'next'.
2239 if (next == next_del && next == EXT_MAX_BLOCKS) {
2240 flags |= FIEMAP_EXTENT_LAST;
2241 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2242 next != EXT_MAX_BLOCKS)) {
2243 EXT4_ERROR_INODE(inode,
2244 "next extent == %u, next "
2245 "delalloc extent = %u",
2253 err = fiemap_fill_next_extent(fieinfo,
2254 (__u64)es.es_lblk << blksize_bits,
2255 (__u64)es.es_pblk << blksize_bits,
2256 (__u64)es.es_len << blksize_bits,
2266 block = es.es_lblk + es.es_len;
2270 ext4_ext_drop_refs(path);
2278 * ext4_ext_put_gap_in_cache:
2279 * calculate boundaries of the gap that the requested block fits into
2280 * and cache this gap
2283 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2286 int depth = ext_depth(inode);
2287 unsigned long len = 0;
2288 ext4_lblk_t lblock = 0;
2289 struct ext4_extent *ex;
2291 ex = path[depth].p_ext;
2294 * there is no extent yet, so gap is [0;-] and we
2297 ext_debug("cache gap(whole file):");
2298 } else if (block < le32_to_cpu(ex->ee_block)) {
2300 len = le32_to_cpu(ex->ee_block) - block;
2301 ext_debug("cache gap(before): %u [%u:%u]",
2303 le32_to_cpu(ex->ee_block),
2304 ext4_ext_get_actual_len(ex));
2305 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2306 ext4_es_insert_extent(inode, lblock, len, ~0,
2307 EXTENT_STATUS_HOLE);
2308 } else if (block >= le32_to_cpu(ex->ee_block)
2309 + ext4_ext_get_actual_len(ex)) {
2311 lblock = le32_to_cpu(ex->ee_block)
2312 + ext4_ext_get_actual_len(ex);
2314 next = ext4_ext_next_allocated_block(path);
2315 ext_debug("cache gap(after): [%u:%u] %u",
2316 le32_to_cpu(ex->ee_block),
2317 ext4_ext_get_actual_len(ex),
2319 BUG_ON(next == lblock);
2320 len = next - lblock;
2321 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2322 ext4_es_insert_extent(inode, lblock, len, ~0,
2323 EXTENT_STATUS_HOLE);
2328 ext_debug(" -> %u:%lu\n", lblock, len);
2333 * removes index from the index block.
2335 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2336 struct ext4_ext_path *path, int depth)
2341 /* free index block */
2343 path = path + depth;
2344 leaf = ext4_idx_pblock(path->p_idx);
2345 if (unlikely(path->p_hdr->eh_entries == 0)) {
2346 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2349 err = ext4_ext_get_access(handle, inode, path);
2353 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2354 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2355 len *= sizeof(struct ext4_extent_idx);
2356 memmove(path->p_idx, path->p_idx + 1, len);
2359 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2360 err = ext4_ext_dirty(handle, inode, path);
2363 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2364 trace_ext4_ext_rm_idx(inode, leaf);
2366 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2367 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2369 while (--depth >= 0) {
2370 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2373 err = ext4_ext_get_access(handle, inode, path);
2376 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2377 err = ext4_ext_dirty(handle, inode, path);
2385 * ext4_ext_calc_credits_for_single_extent:
2386 * This routine returns max. credits that needed to insert an extent
2387 * to the extent tree.
2388 * When pass the actual path, the caller should calculate credits
2391 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2392 struct ext4_ext_path *path)
2395 int depth = ext_depth(inode);
2398 /* probably there is space in leaf? */
2399 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2400 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2403 * There are some space in the leaf tree, no
2404 * need to account for leaf block credit
2406 * bitmaps and block group descriptor blocks
2407 * and other metadata blocks still need to be
2410 /* 1 bitmap, 1 block group descriptor */
2411 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2416 return ext4_chunk_trans_blocks(inode, nrblocks);
2420 * How many index/leaf blocks need to change/allocate to add @extents extents?
2422 * If we add a single extent, then in the worse case, each tree level
2423 * index/leaf need to be changed in case of the tree split.
2425 * If more extents are inserted, they could cause the whole tree split more
2426 * than once, but this is really rare.
2428 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2433 /* If we are converting the inline data, only one is needed here. */
2434 if (ext4_has_inline_data(inode))
2437 depth = ext_depth(inode);
2447 static inline int get_default_free_blocks_flags(struct inode *inode)
2449 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2450 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2451 else if (ext4_should_journal_data(inode))
2452 return EXT4_FREE_BLOCKS_FORGET;
2456 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2457 struct ext4_extent *ex,
2458 long long *partial_cluster,
2459 ext4_lblk_t from, ext4_lblk_t to)
2461 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2462 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2464 int flags = get_default_free_blocks_flags(inode);
2467 * For bigalloc file systems, we never free a partial cluster
2468 * at the beginning of the extent. Instead, we make a note
2469 * that we tried freeing the cluster, and check to see if we
2470 * need to free it on a subsequent call to ext4_remove_blocks,
2471 * or at the end of the ext4_truncate() operation.
2473 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2475 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2477 * If we have a partial cluster, and it's different from the
2478 * cluster of the last block, we need to explicitly free the
2479 * partial cluster here.
2481 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2482 if ((*partial_cluster > 0) &&
2483 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2484 ext4_free_blocks(handle, inode, NULL,
2485 EXT4_C2B(sbi, *partial_cluster),
2486 sbi->s_cluster_ratio, flags);
2487 *partial_cluster = 0;
2490 #ifdef EXTENTS_STATS
2492 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2493 spin_lock(&sbi->s_ext_stats_lock);
2494 sbi->s_ext_blocks += ee_len;
2495 sbi->s_ext_extents++;
2496 if (ee_len < sbi->s_ext_min)
2497 sbi->s_ext_min = ee_len;
2498 if (ee_len > sbi->s_ext_max)
2499 sbi->s_ext_max = ee_len;
2500 if (ext_depth(inode) > sbi->s_depth_max)
2501 sbi->s_depth_max = ext_depth(inode);
2502 spin_unlock(&sbi->s_ext_stats_lock);
2505 if (from >= le32_to_cpu(ex->ee_block)
2506 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2509 unsigned int unaligned;
2511 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2512 pblk = ext4_ext_pblock(ex) + ee_len - num;
2514 * Usually we want to free partial cluster at the end of the
2515 * extent, except for the situation when the cluster is still
2516 * used by any other extent (partial_cluster is negative).
2518 if (*partial_cluster < 0 &&
2519 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2520 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2522 ext_debug("free last %u blocks starting %llu partial %lld\n",
2523 num, pblk, *partial_cluster);
2524 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2526 * If the block range to be freed didn't start at the
2527 * beginning of a cluster, and we removed the entire
2528 * extent and the cluster is not used by any other extent,
2529 * save the partial cluster here, since we might need to
2530 * delete if we determine that the truncate operation has
2531 * removed all of the blocks in the cluster.
2533 * On the other hand, if we did not manage to free the whole
2534 * extent, we have to mark the cluster as used (store negative
2535 * cluster number in partial_cluster).
2537 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2538 if (unaligned && (ee_len == num) &&
2539 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2540 *partial_cluster = EXT4_B2C(sbi, pblk);
2542 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2543 else if (*partial_cluster > 0)
2544 *partial_cluster = 0;
2546 ext4_error(sbi->s_sb, "strange request: removal(2) "
2547 "%u-%u from %u:%u\n",
2548 from, to, le32_to_cpu(ex->ee_block), ee_len);
2554 * ext4_ext_rm_leaf() Removes the extents associated with the
2555 * blocks appearing between "start" and "end", and splits the extents
2556 * if "start" and "end" appear in the same extent
2558 * @handle: The journal handle
2559 * @inode: The files inode
2560 * @path: The path to the leaf
2561 * @partial_cluster: The cluster which we'll have to free if all extents
2562 * has been released from it. It gets negative in case
2563 * that the cluster is still used.
2564 * @start: The first block to remove
2565 * @end: The last block to remove
2568 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2569 struct ext4_ext_path *path,
2570 long long *partial_cluster,
2571 ext4_lblk_t start, ext4_lblk_t end)
2573 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2574 int err = 0, correct_index = 0;
2575 int depth = ext_depth(inode), credits;
2576 struct ext4_extent_header *eh;
2579 ext4_lblk_t ex_ee_block;
2580 unsigned short ex_ee_len;
2581 unsigned unwritten = 0;
2582 struct ext4_extent *ex;
2585 /* the header must be checked already in ext4_ext_remove_space() */
2586 ext_debug("truncate since %u in leaf to %u\n", start, end);
2587 if (!path[depth].p_hdr)
2588 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2589 eh = path[depth].p_hdr;
2590 if (unlikely(path[depth].p_hdr == NULL)) {
2591 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2594 /* find where to start removing */
2595 ex = path[depth].p_ext;
2597 ex = EXT_LAST_EXTENT(eh);
2599 ex_ee_block = le32_to_cpu(ex->ee_block);
2600 ex_ee_len = ext4_ext_get_actual_len(ex);
2603 * If we're starting with an extent other than the last one in the
2604 * node, we need to see if it shares a cluster with the extent to
2605 * the right (towards the end of the file). If its leftmost cluster
2606 * is this extent's rightmost cluster and it is not cluster aligned,
2607 * we'll mark it as a partial that is not to be deallocated.
2610 if (ex != EXT_LAST_EXTENT(eh)) {
2611 ext4_fsblk_t current_pblk, right_pblk;
2612 long long current_cluster, right_cluster;
2614 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2615 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2616 right_pblk = ext4_ext_pblock(ex + 1);
2617 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2618 if (current_cluster == right_cluster &&
2619 EXT4_PBLK_COFF(sbi, right_pblk))
2620 *partial_cluster = -right_cluster;
2623 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2625 while (ex >= EXT_FIRST_EXTENT(eh) &&
2626 ex_ee_block + ex_ee_len > start) {
2628 if (ext4_ext_is_unwritten(ex))
2633 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2634 unwritten, ex_ee_len);
2635 path[depth].p_ext = ex;
2637 a = ex_ee_block > start ? ex_ee_block : start;
2638 b = ex_ee_block+ex_ee_len - 1 < end ?
2639 ex_ee_block+ex_ee_len - 1 : end;
2641 ext_debug(" border %u:%u\n", a, b);
2643 /* If this extent is beyond the end of the hole, skip it */
2644 if (end < ex_ee_block) {
2646 * We're going to skip this extent and move to another,
2647 * so if this extent is not cluster aligned we have
2648 * to mark the current cluster as used to avoid
2649 * accidentally freeing it later on
2651 pblk = ext4_ext_pblock(ex);
2652 if (EXT4_PBLK_COFF(sbi, pblk))
2654 -((long long)EXT4_B2C(sbi, pblk));
2656 ex_ee_block = le32_to_cpu(ex->ee_block);
2657 ex_ee_len = ext4_ext_get_actual_len(ex);
2659 } else if (b != ex_ee_block + ex_ee_len - 1) {
2660 EXT4_ERROR_INODE(inode,
2661 "can not handle truncate %u:%u "
2663 start, end, ex_ee_block,
2664 ex_ee_block + ex_ee_len - 1);
2667 } else if (a != ex_ee_block) {
2668 /* remove tail of the extent */
2669 num = a - ex_ee_block;
2671 /* remove whole extent: excellent! */
2675 * 3 for leaf, sb, and inode plus 2 (bmap and group
2676 * descriptor) for each block group; assume two block
2677 * groups plus ex_ee_len/blocks_per_block_group for
2680 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2681 if (ex == EXT_FIRST_EXTENT(eh)) {
2683 credits += (ext_depth(inode)) + 1;
2685 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2687 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2691 err = ext4_ext_get_access(handle, inode, path + depth);
2695 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2701 /* this extent is removed; mark slot entirely unused */
2702 ext4_ext_store_pblock(ex, 0);
2704 ex->ee_len = cpu_to_le16(num);
2706 * Do not mark unwritten if all the blocks in the
2707 * extent have been removed.
2709 if (unwritten && num)
2710 ext4_ext_mark_unwritten(ex);
2712 * If the extent was completely released,
2713 * we need to remove it from the leaf
2716 if (end != EXT_MAX_BLOCKS - 1) {
2718 * For hole punching, we need to scoot all the
2719 * extents up when an extent is removed so that
2720 * we dont have blank extents in the middle
2722 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2723 sizeof(struct ext4_extent));
2725 /* Now get rid of the one at the end */
2726 memset(EXT_LAST_EXTENT(eh), 0,
2727 sizeof(struct ext4_extent));
2729 le16_add_cpu(&eh->eh_entries, -1);
2730 } else if (*partial_cluster > 0)
2731 *partial_cluster = 0;
2733 err = ext4_ext_dirty(handle, inode, path + depth);
2737 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2738 ext4_ext_pblock(ex));
2740 ex_ee_block = le32_to_cpu(ex->ee_block);
2741 ex_ee_len = ext4_ext_get_actual_len(ex);
2744 if (correct_index && eh->eh_entries)
2745 err = ext4_ext_correct_indexes(handle, inode, path);
2748 * If there's a partial cluster and at least one extent remains in
2749 * the leaf, free the partial cluster if it isn't shared with the
2750 * current extent. If there's a partial cluster and no extents
2751 * remain in the leaf, it can't be freed here. It can only be
2752 * freed when it's possible to determine if it's not shared with
2753 * any other extent - when the next leaf is processed or when space
2754 * removal is complete.
2756 if (*partial_cluster > 0 && eh->eh_entries &&
2757 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2758 *partial_cluster)) {
2759 int flags = get_default_free_blocks_flags(inode);
2761 ext4_free_blocks(handle, inode, NULL,
2762 EXT4_C2B(sbi, *partial_cluster),
2763 sbi->s_cluster_ratio, flags);
2764 *partial_cluster = 0;
2767 /* if this leaf is free, then we should
2768 * remove it from index block above */
2769 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2770 err = ext4_ext_rm_idx(handle, inode, path, depth);
2777 * ext4_ext_more_to_rm:
2778 * returns 1 if current index has to be freed (even partial)
2781 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2783 BUG_ON(path->p_idx == NULL);
2785 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2789 * if truncate on deeper level happened, it wasn't partial,
2790 * so we have to consider current index for truncation
2792 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2797 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2800 struct super_block *sb = inode->i_sb;
2801 int depth = ext_depth(inode);
2802 struct ext4_ext_path *path = NULL;
2803 long long partial_cluster = 0;
2807 ext_debug("truncate since %u to %u\n", start, end);
2809 /* probably first extent we're gonna free will be last in block */
2810 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2812 return PTR_ERR(handle);
2815 trace_ext4_ext_remove_space(inode, start, end, depth);
2818 * Check if we are removing extents inside the extent tree. If that
2819 * is the case, we are going to punch a hole inside the extent tree
2820 * so we have to check whether we need to split the extent covering
2821 * the last block to remove so we can easily remove the part of it
2822 * in ext4_ext_rm_leaf().
2824 if (end < EXT_MAX_BLOCKS - 1) {
2825 struct ext4_extent *ex;
2826 ext4_lblk_t ee_block;
2828 /* find extent for this block */
2829 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2831 ext4_journal_stop(handle);
2832 return PTR_ERR(path);
2834 depth = ext_depth(inode);
2835 /* Leaf not may not exist only if inode has no blocks at all */
2836 ex = path[depth].p_ext;
2839 EXT4_ERROR_INODE(inode,
2840 "path[%d].p_hdr == NULL",
2847 ee_block = le32_to_cpu(ex->ee_block);
2850 * See if the last block is inside the extent, if so split
2851 * the extent at 'end' block so we can easily remove the
2852 * tail of the first part of the split extent in
2853 * ext4_ext_rm_leaf().
2855 if (end >= ee_block &&
2856 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2859 if (ext4_ext_is_unwritten(ex))
2860 split_flag = EXT4_EXT_MARK_UNWRIT1 |
2861 EXT4_EXT_MARK_UNWRIT2;
2864 * Split the extent in two so that 'end' is the last
2865 * block in the first new extent. Also we should not
2866 * fail removing space due to ENOSPC so try to use
2867 * reserved block if that happens.
2869 err = ext4_split_extent_at(handle, inode, path,
2870 end + 1, split_flag,
2872 EXT4_GET_BLOCKS_PRE_IO |
2873 EXT4_GET_BLOCKS_METADATA_NOFAIL);
2880 * We start scanning from right side, freeing all the blocks
2881 * after i_size and walking into the tree depth-wise.
2883 depth = ext_depth(inode);
2888 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2890 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2893 ext4_journal_stop(handle);
2896 path[0].p_depth = depth;
2897 path[0].p_hdr = ext_inode_hdr(inode);
2900 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2907 while (i >= 0 && err == 0) {
2909 /* this is leaf block */
2910 err = ext4_ext_rm_leaf(handle, inode, path,
2911 &partial_cluster, start,
2913 /* root level has p_bh == NULL, brelse() eats this */
2914 brelse(path[i].p_bh);
2915 path[i].p_bh = NULL;
2920 /* this is index block */
2921 if (!path[i].p_hdr) {
2922 ext_debug("initialize header\n");
2923 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2926 if (!path[i].p_idx) {
2927 /* this level hasn't been touched yet */
2928 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2929 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2930 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2932 le16_to_cpu(path[i].p_hdr->eh_entries));
2934 /* we were already here, see at next index */
2938 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2939 i, EXT_FIRST_INDEX(path[i].p_hdr),
2941 if (ext4_ext_more_to_rm(path + i)) {
2942 struct buffer_head *bh;
2943 /* go to the next level */
2944 ext_debug("move to level %d (block %llu)\n",
2945 i + 1, ext4_idx_pblock(path[i].p_idx));
2946 memset(path + i + 1, 0, sizeof(*path));
2947 bh = read_extent_tree_block(inode,
2948 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2951 /* should we reset i_size? */
2955 /* Yield here to deal with large extent trees.
2956 * Should be a no-op if we did IO above. */
2958 if (WARN_ON(i + 1 > depth)) {
2962 path[i + 1].p_bh = bh;
2964 /* save actual number of indexes since this
2965 * number is changed at the next iteration */
2966 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2969 /* we finished processing this index, go up */
2970 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2971 /* index is empty, remove it;
2972 * handle must be already prepared by the
2973 * truncatei_leaf() */
2974 err = ext4_ext_rm_idx(handle, inode, path, i);
2976 /* root level has p_bh == NULL, brelse() eats this */
2977 brelse(path[i].p_bh);
2978 path[i].p_bh = NULL;
2980 ext_debug("return to level %d\n", i);
2984 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2985 partial_cluster, path->p_hdr->eh_entries);
2987 /* If we still have something in the partial cluster and we have removed
2988 * even the first extent, then we should free the blocks in the partial
2989 * cluster as well. */
2990 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2991 int flags = get_default_free_blocks_flags(inode);
2993 ext4_free_blocks(handle, inode, NULL,
2994 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2995 EXT4_SB(sb)->s_cluster_ratio, flags);
2996 partial_cluster = 0;
2999 /* TODO: flexible tree reduction should be here */
3000 if (path->p_hdr->eh_entries == 0) {
3002 * truncate to zero freed all the tree,
3003 * so we need to correct eh_depth
3005 err = ext4_ext_get_access(handle, inode, path);
3007 ext_inode_hdr(inode)->eh_depth = 0;
3008 ext_inode_hdr(inode)->eh_max =
3009 cpu_to_le16(ext4_ext_space_root(inode, 0));
3010 err = ext4_ext_dirty(handle, inode, path);
3014 ext4_ext_drop_refs(path);
3016 if (err == -EAGAIN) {
3020 ext4_journal_stop(handle);
3026 * called at mount time
3028 void ext4_ext_init(struct super_block *sb)
3031 * possible initialization would be here
3034 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3035 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3036 printk(KERN_INFO "EXT4-fs: file extents enabled"
3037 #ifdef AGGRESSIVE_TEST
3038 ", aggressive tests"
3040 #ifdef CHECK_BINSEARCH
3043 #ifdef EXTENTS_STATS
3048 #ifdef EXTENTS_STATS
3049 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3050 EXT4_SB(sb)->s_ext_min = 1 << 30;
3051 EXT4_SB(sb)->s_ext_max = 0;
3057 * called at umount time
3059 void ext4_ext_release(struct super_block *sb)
3061 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3064 #ifdef EXTENTS_STATS
3065 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3066 struct ext4_sb_info *sbi = EXT4_SB(sb);
3067 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3068 sbi->s_ext_blocks, sbi->s_ext_extents,
3069 sbi->s_ext_blocks / sbi->s_ext_extents);
3070 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3071 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3076 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3078 ext4_lblk_t ee_block;
3079 ext4_fsblk_t ee_pblock;
3080 unsigned int ee_len;
3082 ee_block = le32_to_cpu(ex->ee_block);
3083 ee_len = ext4_ext_get_actual_len(ex);
3084 ee_pblock = ext4_ext_pblock(ex);
3089 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3090 EXTENT_STATUS_WRITTEN);
3093 /* FIXME!! we need to try to merge to left or right after zero-out */
3094 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3096 ext4_fsblk_t ee_pblock;
3097 unsigned int ee_len;
3100 ee_len = ext4_ext_get_actual_len(ex);
3101 ee_pblock = ext4_ext_pblock(ex);
3103 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3111 * ext4_split_extent_at() splits an extent at given block.
3113 * @handle: the journal handle
3114 * @inode: the file inode
3115 * @path: the path to the extent
3116 * @split: the logical block where the extent is splitted.
3117 * @split_flags: indicates if the extent could be zeroout if split fails, and
3118 * the states(init or unwritten) of new extents.
3119 * @flags: flags used to insert new extent to extent tree.
3122 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3123 * of which are deterimined by split_flag.
3125 * There are two cases:
3126 * a> the extent are splitted into two extent.
3127 * b> split is not needed, and just mark the extent.
3129 * return 0 on success.
3131 static int ext4_split_extent_at(handle_t *handle,
3132 struct inode *inode,
3133 struct ext4_ext_path *path,
3138 ext4_fsblk_t newblock;
3139 ext4_lblk_t ee_block;
3140 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3141 struct ext4_extent *ex2 = NULL;
3142 unsigned int ee_len, depth;
3145 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3146 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3148 ext_debug("ext4_split_extents_at: inode %lu, logical"
3149 "block %llu\n", inode->i_ino, (unsigned long long)split);
3151 ext4_ext_show_leaf(inode, path);
3153 depth = ext_depth(inode);
3154 ex = path[depth].p_ext;
3155 ee_block = le32_to_cpu(ex->ee_block);
3156 ee_len = ext4_ext_get_actual_len(ex);
3157 newblock = split - ee_block + ext4_ext_pblock(ex);
3159 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3160 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3161 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3162 EXT4_EXT_MARK_UNWRIT1 |
3163 EXT4_EXT_MARK_UNWRIT2));
3165 err = ext4_ext_get_access(handle, inode, path + depth);
3169 if (split == ee_block) {
3171 * case b: block @split is the block that the extent begins with
3172 * then we just change the state of the extent, and splitting
3175 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3176 ext4_ext_mark_unwritten(ex);
3178 ext4_ext_mark_initialized(ex);
3180 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3181 ext4_ext_try_to_merge(handle, inode, path, ex);
3183 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3188 memcpy(&orig_ex, ex, sizeof(orig_ex));
3189 ex->ee_len = cpu_to_le16(split - ee_block);
3190 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3191 ext4_ext_mark_unwritten(ex);
3194 * path may lead to new leaf, not to original leaf any more
3195 * after ext4_ext_insert_extent() returns,
3197 err = ext4_ext_dirty(handle, inode, path + depth);
3199 goto fix_extent_len;
3202 ex2->ee_block = cpu_to_le32(split);
3203 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3204 ext4_ext_store_pblock(ex2, newblock);
3205 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3206 ext4_ext_mark_unwritten(ex2);
3208 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3209 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3210 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3211 if (split_flag & EXT4_EXT_DATA_VALID1) {
3212 err = ext4_ext_zeroout(inode, ex2);
3213 zero_ex.ee_block = ex2->ee_block;
3214 zero_ex.ee_len = cpu_to_le16(
3215 ext4_ext_get_actual_len(ex2));
3216 ext4_ext_store_pblock(&zero_ex,
3217 ext4_ext_pblock(ex2));
3219 err = ext4_ext_zeroout(inode, ex);
3220 zero_ex.ee_block = ex->ee_block;
3221 zero_ex.ee_len = cpu_to_le16(
3222 ext4_ext_get_actual_len(ex));
3223 ext4_ext_store_pblock(&zero_ex,
3224 ext4_ext_pblock(ex));
3227 err = ext4_ext_zeroout(inode, &orig_ex);
3228 zero_ex.ee_block = orig_ex.ee_block;
3229 zero_ex.ee_len = cpu_to_le16(
3230 ext4_ext_get_actual_len(&orig_ex));
3231 ext4_ext_store_pblock(&zero_ex,
3232 ext4_ext_pblock(&orig_ex));
3236 goto fix_extent_len;
3237 /* update the extent length and mark as initialized */
3238 ex->ee_len = cpu_to_le16(ee_len);
3239 ext4_ext_try_to_merge(handle, inode, path, ex);
3240 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3242 goto fix_extent_len;
3244 /* update extent status tree */
3245 err = ext4_zeroout_es(inode, &zero_ex);
3249 goto fix_extent_len;
3252 ext4_ext_show_leaf(inode, path);
3256 ex->ee_len = orig_ex.ee_len;
3257 ext4_ext_dirty(handle, inode, path + path->p_depth);
3262 * ext4_split_extents() splits an extent and mark extent which is covered
3263 * by @map as split_flags indicates
3265 * It may result in splitting the extent into multiple extents (up to three)
3266 * There are three possibilities:
3267 * a> There is no split required
3268 * b> Splits in two extents: Split is happening at either end of the extent
3269 * c> Splits in three extents: Somone is splitting in middle of the extent
3272 static int ext4_split_extent(handle_t *handle,
3273 struct inode *inode,
3274 struct ext4_ext_path *path,
3275 struct ext4_map_blocks *map,
3279 ext4_lblk_t ee_block;
3280 struct ext4_extent *ex;
3281 unsigned int ee_len, depth;
3284 int split_flag1, flags1;
3285 int allocated = map->m_len;
3287 depth = ext_depth(inode);
3288 ex = path[depth].p_ext;
3289 ee_block = le32_to_cpu(ex->ee_block);
3290 ee_len = ext4_ext_get_actual_len(ex);
3291 unwritten = ext4_ext_is_unwritten(ex);
3293 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3294 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3295 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3297 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3298 EXT4_EXT_MARK_UNWRIT2;
3299 if (split_flag & EXT4_EXT_DATA_VALID2)
3300 split_flag1 |= EXT4_EXT_DATA_VALID1;
3301 err = ext4_split_extent_at(handle, inode, path,
3302 map->m_lblk + map->m_len, split_flag1, flags1);
3306 allocated = ee_len - (map->m_lblk - ee_block);
3309 * Update path is required because previous ext4_split_extent_at() may
3310 * result in split of original leaf or extent zeroout.
3312 ext4_ext_drop_refs(path);
3313 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3315 return PTR_ERR(path);
3316 depth = ext_depth(inode);
3317 ex = path[depth].p_ext;
3319 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3320 (unsigned long) map->m_lblk);
3323 unwritten = ext4_ext_is_unwritten(ex);
3326 if (map->m_lblk >= ee_block) {
3327 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3329 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3330 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3331 EXT4_EXT_MARK_UNWRIT2);
3333 err = ext4_split_extent_at(handle, inode, path,
3334 map->m_lblk, split_flag1, flags);
3339 ext4_ext_show_leaf(inode, path);
3341 return err ? err : allocated;
3345 * This function is called by ext4_ext_map_blocks() if someone tries to write
3346 * to an unwritten extent. It may result in splitting the unwritten
3347 * extent into multiple extents (up to three - one initialized and two
3349 * There are three possibilities:
3350 * a> There is no split required: Entire extent should be initialized
3351 * b> Splits in two extents: Write is happening at either end of the extent
3352 * c> Splits in three extents: Somone is writing in middle of the extent
3355 * - The extent pointed to by 'path' is unwritten.
3356 * - The extent pointed to by 'path' contains a superset
3357 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3359 * Post-conditions on success:
3360 * - the returned value is the number of blocks beyond map->l_lblk
3361 * that are allocated and initialized.
3362 * It is guaranteed to be >= map->m_len.
3364 static int ext4_ext_convert_to_initialized(handle_t *handle,
3365 struct inode *inode,
3366 struct ext4_map_blocks *map,
3367 struct ext4_ext_path *path,
3370 struct ext4_sb_info *sbi;
3371 struct ext4_extent_header *eh;
3372 struct ext4_map_blocks split_map;
3373 struct ext4_extent zero_ex;
3374 struct ext4_extent *ex, *abut_ex;
3375 ext4_lblk_t ee_block, eof_block;
3376 unsigned int ee_len, depth, map_len = map->m_len;
3377 int allocated = 0, max_zeroout = 0;
3381 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3382 "block %llu, max_blocks %u\n", inode->i_ino,
3383 (unsigned long long)map->m_lblk, map_len);
3385 sbi = EXT4_SB(inode->i_sb);
3386 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3387 inode->i_sb->s_blocksize_bits;
3388 if (eof_block < map->m_lblk + map_len)
3389 eof_block = map->m_lblk + map_len;
3391 depth = ext_depth(inode);
3392 eh = path[depth].p_hdr;
3393 ex = path[depth].p_ext;
3394 ee_block = le32_to_cpu(ex->ee_block);
3395 ee_len = ext4_ext_get_actual_len(ex);
3398 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3400 /* Pre-conditions */
3401 BUG_ON(!ext4_ext_is_unwritten(ex));
3402 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3405 * Attempt to transfer newly initialized blocks from the currently
3406 * unwritten extent to its neighbor. This is much cheaper
3407 * than an insertion followed by a merge as those involve costly
3408 * memmove() calls. Transferring to the left is the common case in
3409 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3410 * followed by append writes.
3412 * Limitations of the current logic:
3413 * - L1: we do not deal with writes covering the whole extent.
3414 * This would require removing the extent if the transfer
3416 * - L2: we only attempt to merge with an extent stored in the
3417 * same extent tree node.
3419 if ((map->m_lblk == ee_block) &&
3420 /* See if we can merge left */
3421 (map_len < ee_len) && /*L1*/
3422 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3423 ext4_lblk_t prev_lblk;
3424 ext4_fsblk_t prev_pblk, ee_pblk;
3425 unsigned int prev_len;
3428 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3429 prev_len = ext4_ext_get_actual_len(abut_ex);
3430 prev_pblk = ext4_ext_pblock(abut_ex);
3431 ee_pblk = ext4_ext_pblock(ex);
3434 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3435 * upon those conditions:
3436 * - C1: abut_ex is initialized,
3437 * - C2: abut_ex is logically abutting ex,
3438 * - C3: abut_ex is physically abutting ex,
3439 * - C4: abut_ex can receive the additional blocks without
3440 * overflowing the (initialized) length limit.
3442 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3443 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3444 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3445 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3446 err = ext4_ext_get_access(handle, inode, path + depth);
3450 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3453 /* Shift the start of ex by 'map_len' blocks */
3454 ex->ee_block = cpu_to_le32(ee_block + map_len);
3455 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3456 ex->ee_len = cpu_to_le16(ee_len - map_len);
3457 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3459 /* Extend abut_ex by 'map_len' blocks */
3460 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3462 /* Result: number of initialized blocks past m_lblk */
3463 allocated = map_len;
3465 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3466 (map_len < ee_len) && /*L1*/
3467 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3468 /* See if we can merge right */
3469 ext4_lblk_t next_lblk;
3470 ext4_fsblk_t next_pblk, ee_pblk;
3471 unsigned int next_len;
3474 next_lblk = le32_to_cpu(abut_ex->ee_block);
3475 next_len = ext4_ext_get_actual_len(abut_ex);
3476 next_pblk = ext4_ext_pblock(abut_ex);
3477 ee_pblk = ext4_ext_pblock(ex);
3480 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3481 * upon those conditions:
3482 * - C1: abut_ex is initialized,
3483 * - C2: abut_ex is logically abutting ex,
3484 * - C3: abut_ex is physically abutting ex,
3485 * - C4: abut_ex can receive the additional blocks without
3486 * overflowing the (initialized) length limit.
3488 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3489 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3490 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3491 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3492 err = ext4_ext_get_access(handle, inode, path + depth);
3496 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3499 /* Shift the start of abut_ex by 'map_len' blocks */
3500 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3501 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3502 ex->ee_len = cpu_to_le16(ee_len - map_len);
3503 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3505 /* Extend abut_ex by 'map_len' blocks */
3506 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3508 /* Result: number of initialized blocks past m_lblk */
3509 allocated = map_len;
3513 /* Mark the block containing both extents as dirty */
3514 ext4_ext_dirty(handle, inode, path + depth);
3516 /* Update path to point to the right extent */
3517 path[depth].p_ext = abut_ex;
3520 allocated = ee_len - (map->m_lblk - ee_block);
3522 WARN_ON(map->m_lblk < ee_block);
3524 * It is safe to convert extent to initialized via explicit
3525 * zeroout only if extent is fully inside i_size or new_size.
3527 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3529 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3530 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3531 (inode->i_sb->s_blocksize_bits - 10);
3533 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3534 if (max_zeroout && (ee_len <= max_zeroout)) {
3535 err = ext4_ext_zeroout(inode, ex);
3538 zero_ex.ee_block = ex->ee_block;
3539 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3540 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3542 err = ext4_ext_get_access(handle, inode, path + depth);
3545 ext4_ext_mark_initialized(ex);
3546 ext4_ext_try_to_merge(handle, inode, path, ex);
3547 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3553 * 1. split the extent into three extents.
3554 * 2. split the extent into two extents, zeroout the first half.
3555 * 3. split the extent into two extents, zeroout the second half.
3556 * 4. split the extent into two extents with out zeroout.
3558 split_map.m_lblk = map->m_lblk;
3559 split_map.m_len = map->m_len;
3561 if (max_zeroout && (allocated > map->m_len)) {
3562 if (allocated <= max_zeroout) {
3565 cpu_to_le32(map->m_lblk);
3566 zero_ex.ee_len = cpu_to_le16(allocated);
3567 ext4_ext_store_pblock(&zero_ex,
3568 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3569 err = ext4_ext_zeroout(inode, &zero_ex);
3572 split_map.m_lblk = map->m_lblk;
3573 split_map.m_len = allocated;
3574 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3576 if (map->m_lblk != ee_block) {
3577 zero_ex.ee_block = ex->ee_block;
3578 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3580 ext4_ext_store_pblock(&zero_ex,
3581 ext4_ext_pblock(ex));
3582 err = ext4_ext_zeroout(inode, &zero_ex);
3587 split_map.m_lblk = ee_block;
3588 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3589 allocated = map->m_len;
3593 allocated = ext4_split_extent(handle, inode, path,
3594 &split_map, split_flag, flags);
3599 /* If we have gotten a failure, don't zero out status tree */
3601 err = ext4_zeroout_es(inode, &zero_ex);
3602 return err ? err : allocated;
3606 * This function is called by ext4_ext_map_blocks() from
3607 * ext4_get_blocks_dio_write() when DIO to write
3608 * to an unwritten extent.
3610 * Writing to an unwritten extent may result in splitting the unwritten
3611 * extent into multiple initialized/unwritten extents (up to three)
3612 * There are three possibilities:
3613 * a> There is no split required: Entire extent should be unwritten
3614 * b> Splits in two extents: Write is happening at either end of the extent
3615 * c> Splits in three extents: Somone is writing in middle of the extent
3617 * This works the same way in the case of initialized -> unwritten conversion.
3619 * One of more index blocks maybe needed if the extent tree grow after
3620 * the unwritten extent split. To prevent ENOSPC occur at the IO
3621 * complete, we need to split the unwritten extent before DIO submit
3622 * the IO. The unwritten extent called at this time will be split
3623 * into three unwritten extent(at most). After IO complete, the part
3624 * being filled will be convert to initialized by the end_io callback function
3625 * via ext4_convert_unwritten_extents().
3627 * Returns the size of unwritten extent to be written on success.
3629 static int ext4_split_convert_extents(handle_t *handle,
3630 struct inode *inode,
3631 struct ext4_map_blocks *map,
3632 struct ext4_ext_path *path,
3635 ext4_lblk_t eof_block;
3636 ext4_lblk_t ee_block;
3637 struct ext4_extent *ex;
3638 unsigned int ee_len;
3639 int split_flag = 0, depth;
3641 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3642 __func__, inode->i_ino,
3643 (unsigned long long)map->m_lblk, map->m_len);
3645 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3646 inode->i_sb->s_blocksize_bits;
3647 if (eof_block < map->m_lblk + map->m_len)
3648 eof_block = map->m_lblk + map->m_len;
3650 * It is safe to convert extent to initialized via explicit
3651 * zeroout only if extent is fully insde i_size or new_size.
3653 depth = ext_depth(inode);
3654 ex = path[depth].p_ext;
3655 ee_block = le32_to_cpu(ex->ee_block);
3656 ee_len = ext4_ext_get_actual_len(ex);
3658 /* Convert to unwritten */
3659 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3660 split_flag |= EXT4_EXT_DATA_VALID1;
3661 /* Convert to initialized */
3662 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3663 split_flag |= ee_block + ee_len <= eof_block ?
3664 EXT4_EXT_MAY_ZEROOUT : 0;
3665 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3667 flags |= EXT4_GET_BLOCKS_PRE_IO;
3668 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3671 static int ext4_convert_initialized_extents(handle_t *handle,
3672 struct inode *inode,
3673 struct ext4_map_blocks *map,
3674 struct ext4_ext_path *path)
3676 struct ext4_extent *ex;
3677 ext4_lblk_t ee_block;
3678 unsigned int ee_len;
3682 depth = ext_depth(inode);
3683 ex = path[depth].p_ext;
3684 ee_block = le32_to_cpu(ex->ee_block);
3685 ee_len = ext4_ext_get_actual_len(ex);
3687 ext_debug("%s: inode %lu, logical"
3688 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3689 (unsigned long long)ee_block, ee_len);
3691 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3692 err = ext4_split_convert_extents(handle, inode, map, path,
3693 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3696 ext4_ext_drop_refs(path);
3697 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3699 err = PTR_ERR(path);
3702 depth = ext_depth(inode);
3703 ex = path[depth].p_ext;
3705 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3706 (unsigned long) map->m_lblk);
3712 err = ext4_ext_get_access(handle, inode, path + depth);
3715 /* first mark the extent as unwritten */
3716 ext4_ext_mark_unwritten(ex);
3718 /* note: ext4_ext_correct_indexes() isn't needed here because
3719 * borders are not changed
3721 ext4_ext_try_to_merge(handle, inode, path, ex);
3723 /* Mark modified extent as dirty */
3724 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3726 ext4_ext_show_leaf(inode, path);
3731 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3732 struct inode *inode,
3733 struct ext4_map_blocks *map,
3734 struct ext4_ext_path *path)
3736 struct ext4_extent *ex;
3737 ext4_lblk_t ee_block;
3738 unsigned int ee_len;
3742 depth = ext_depth(inode);
3743 ex = path[depth].p_ext;
3744 ee_block = le32_to_cpu(ex->ee_block);
3745 ee_len = ext4_ext_get_actual_len(ex);
3747 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3748 "block %llu, max_blocks %u\n", inode->i_ino,
3749 (unsigned long long)ee_block, ee_len);
3751 /* If extent is larger than requested it is a clear sign that we still
3752 * have some extent state machine issues left. So extent_split is still
3754 * TODO: Once all related issues will be fixed this situation should be
3757 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3759 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3760 " len %u; IO logical block %llu, len %u\n",
3761 inode->i_ino, (unsigned long long)ee_block, ee_len,
3762 (unsigned long long)map->m_lblk, map->m_len);
3764 err = ext4_split_convert_extents(handle, inode, map, path,
3765 EXT4_GET_BLOCKS_CONVERT);
3768 ext4_ext_drop_refs(path);
3769 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3771 err = PTR_ERR(path);
3774 depth = ext_depth(inode);
3775 ex = path[depth].p_ext;
3778 err = ext4_ext_get_access(handle, inode, path + depth);
3781 /* first mark the extent as initialized */
3782 ext4_ext_mark_initialized(ex);
3784 /* note: ext4_ext_correct_indexes() isn't needed here because
3785 * borders are not changed
3787 ext4_ext_try_to_merge(handle, inode, path, ex);
3789 /* Mark modified extent as dirty */
3790 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3792 ext4_ext_show_leaf(inode, path);
3796 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3797 sector_t block, int count)
3800 for (i = 0; i < count; i++)
3801 unmap_underlying_metadata(bdev, block + i);
3805 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3807 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3809 struct ext4_ext_path *path,
3813 struct ext4_extent_header *eh;
3814 struct ext4_extent *last_ex;
3816 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3819 depth = ext_depth(inode);
3820 eh = path[depth].p_hdr;
3823 * We're going to remove EOFBLOCKS_FL entirely in future so we
3824 * do not care for this case anymore. Simply remove the flag
3825 * if there are no extents.
3827 if (unlikely(!eh->eh_entries))
3829 last_ex = EXT_LAST_EXTENT(eh);
3831 * We should clear the EOFBLOCKS_FL flag if we are writing the
3832 * last block in the last extent in the file. We test this by
3833 * first checking to see if the caller to
3834 * ext4_ext_get_blocks() was interested in the last block (or
3835 * a block beyond the last block) in the current extent. If
3836 * this turns out to be false, we can bail out from this
3837 * function immediately.
3839 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3840 ext4_ext_get_actual_len(last_ex))
3843 * If the caller does appear to be planning to write at or
3844 * beyond the end of the current extent, we then test to see
3845 * if the current extent is the last extent in the file, by
3846 * checking to make sure it was reached via the rightmost node
3847 * at each level of the tree.
3849 for (i = depth-1; i >= 0; i--)
3850 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3853 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3854 return ext4_mark_inode_dirty(handle, inode);
3858 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3860 * Return 1 if there is a delalloc block in the range, otherwise 0.
3862 int ext4_find_delalloc_range(struct inode *inode,
3863 ext4_lblk_t lblk_start,
3864 ext4_lblk_t lblk_end)
3866 struct extent_status es;
3868 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3870 return 0; /* there is no delay extent in this tree */
3871 else if (es.es_lblk <= lblk_start &&
3872 lblk_start < es.es_lblk + es.es_len)
3874 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3880 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3882 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3883 ext4_lblk_t lblk_start, lblk_end;
3884 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3885 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3887 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3891 * Determines how many complete clusters (out of those specified by the 'map')
3892 * are under delalloc and were reserved quota for.
3893 * This function is called when we are writing out the blocks that were
3894 * originally written with their allocation delayed, but then the space was
3895 * allocated using fallocate() before the delayed allocation could be resolved.
3896 * The cases to look for are:
3897 * ('=' indicated delayed allocated blocks
3898 * '-' indicates non-delayed allocated blocks)
3899 * (a) partial clusters towards beginning and/or end outside of allocated range
3900 * are not delalloc'ed.
3902 * |----c---=|====c====|====c====|===-c----|
3903 * |++++++ allocated ++++++|
3904 * ==> 4 complete clusters in above example
3906 * (b) partial cluster (outside of allocated range) towards either end is
3907 * marked for delayed allocation. In this case, we will exclude that
3910 * |----====c========|========c========|
3911 * |++++++ allocated ++++++|
3912 * ==> 1 complete clusters in above example
3915 * |================c================|
3916 * |++++++ allocated ++++++|
3917 * ==> 0 complete clusters in above example
3919 * The ext4_da_update_reserve_space will be called only if we
3920 * determine here that there were some "entire" clusters that span
3921 * this 'allocated' range.
3922 * In the non-bigalloc case, this function will just end up returning num_blks
3923 * without ever calling ext4_find_delalloc_range.
3926 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3927 unsigned int num_blks)
3929 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3930 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3931 ext4_lblk_t lblk_from, lblk_to, c_offset;
3932 unsigned int allocated_clusters = 0;
3934 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3935 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3937 /* max possible clusters for this allocation */
3938 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3940 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3942 /* Check towards left side */
3943 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3945 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3946 lblk_to = lblk_from + c_offset - 1;
3948 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3949 allocated_clusters--;
3952 /* Now check towards right. */
3953 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3954 if (allocated_clusters && c_offset) {
3955 lblk_from = lblk_start + num_blks;
3956 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3958 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3959 allocated_clusters--;
3962 return allocated_clusters;
3966 ext4_ext_convert_initialized_extent(handle_t *handle, struct inode *inode,
3967 struct ext4_map_blocks *map,
3968 struct ext4_ext_path *path, int flags,
3969 unsigned int allocated, ext4_fsblk_t newblock)
3975 * Make sure that the extent is no bigger than we support with
3978 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3979 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3981 ret = ext4_convert_initialized_extents(handle, inode, map,
3984 ext4_update_inode_fsync_trans(handle, inode, 1);
3985 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3989 map->m_flags |= EXT4_MAP_UNWRITTEN;
3990 if (allocated > map->m_len)
3991 allocated = map->m_len;
3992 map->m_len = allocated;
3994 return err ? err : allocated;
3998 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3999 struct ext4_map_blocks *map,
4000 struct ext4_ext_path *path, int flags,
4001 unsigned int allocated, ext4_fsblk_t newblock)
4005 ext4_io_end_t *io = ext4_inode_aio(inode);
4007 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4008 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4009 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4011 ext4_ext_show_leaf(inode, path);
4014 * When writing into unwritten space, we should not fail to
4015 * allocate metadata blocks for the new extent block if needed.
4017 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4019 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4020 allocated, newblock);
4022 /* get_block() before submit the IO, split the extent */
4023 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4024 ret = ext4_split_convert_extents(handle, inode, map,
4025 path, flags | EXT4_GET_BLOCKS_CONVERT);
4029 * Flag the inode(non aio case) or end_io struct (aio case)
4030 * that this IO needs to conversion to written when IO is
4034 ext4_set_io_unwritten_flag(inode, io);
4036 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4037 map->m_flags |= EXT4_MAP_UNWRITTEN;
4040 /* IO end_io complete, convert the filled extent to written */
4041 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4042 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4045 ext4_update_inode_fsync_trans(handle, inode, 1);
4046 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4050 map->m_flags |= EXT4_MAP_MAPPED;
4051 map->m_pblk = newblock;
4052 if (allocated > map->m_len)
4053 allocated = map->m_len;
4054 map->m_len = allocated;
4057 /* buffered IO case */
4059 * repeat fallocate creation request
4060 * we already have an unwritten extent
4062 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4063 map->m_flags |= EXT4_MAP_UNWRITTEN;
4067 /* buffered READ or buffered write_begin() lookup */
4068 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4070 * We have blocks reserved already. We
4071 * return allocated blocks so that delalloc
4072 * won't do block reservation for us. But
4073 * the buffer head will be unmapped so that
4074 * a read from the block returns 0s.
4076 map->m_flags |= EXT4_MAP_UNWRITTEN;
4080 /* buffered write, writepage time, convert*/
4081 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
4083 ext4_update_inode_fsync_trans(handle, inode, 1);
4090 map->m_flags |= EXT4_MAP_NEW;
4092 * if we allocated more blocks than requested
4093 * we need to make sure we unmap the extra block
4094 * allocated. The actual needed block will get
4095 * unmapped later when we find the buffer_head marked
4098 if (allocated > map->m_len) {
4099 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4100 newblock + map->m_len,
4101 allocated - map->m_len);
4102 allocated = map->m_len;
4104 map->m_len = allocated;
4107 * If we have done fallocate with the offset that is already
4108 * delayed allocated, we would have block reservation
4109 * and quota reservation done in the delayed write path.
4110 * But fallocate would have already updated quota and block
4111 * count for this offset. So cancel these reservation
4113 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4114 unsigned int reserved_clusters;
4115 reserved_clusters = get_reserved_cluster_alloc(inode,
4116 map->m_lblk, map->m_len);
4117 if (reserved_clusters)
4118 ext4_da_update_reserve_space(inode,
4124 map->m_flags |= EXT4_MAP_MAPPED;
4125 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4126 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4132 if (allocated > map->m_len)
4133 allocated = map->m_len;
4134 ext4_ext_show_leaf(inode, path);
4135 map->m_pblk = newblock;
4136 map->m_len = allocated;
4138 return err ? err : allocated;
4142 * get_implied_cluster_alloc - check to see if the requested
4143 * allocation (in the map structure) overlaps with a cluster already
4144 * allocated in an extent.
4145 * @sb The filesystem superblock structure
4146 * @map The requested lblk->pblk mapping
4147 * @ex The extent structure which might contain an implied
4148 * cluster allocation
4150 * This function is called by ext4_ext_map_blocks() after we failed to
4151 * find blocks that were already in the inode's extent tree. Hence,
4152 * we know that the beginning of the requested region cannot overlap
4153 * the extent from the inode's extent tree. There are three cases we
4154 * want to catch. The first is this case:
4156 * |--- cluster # N--|
4157 * |--- extent ---| |---- requested region ---|
4160 * The second case that we need to test for is this one:
4162 * |--------- cluster # N ----------------|
4163 * |--- requested region --| |------- extent ----|
4164 * |=======================|
4166 * The third case is when the requested region lies between two extents
4167 * within the same cluster:
4168 * |------------- cluster # N-------------|
4169 * |----- ex -----| |---- ex_right ----|
4170 * |------ requested region ------|
4171 * |================|
4173 * In each of the above cases, we need to set the map->m_pblk and
4174 * map->m_len so it corresponds to the return the extent labelled as
4175 * "|====|" from cluster #N, since it is already in use for data in
4176 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4177 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4178 * as a new "allocated" block region. Otherwise, we will return 0 and
4179 * ext4_ext_map_blocks() will then allocate one or more new clusters
4180 * by calling ext4_mb_new_blocks().
4182 static int get_implied_cluster_alloc(struct super_block *sb,
4183 struct ext4_map_blocks *map,
4184 struct ext4_extent *ex,
4185 struct ext4_ext_path *path)
4187 struct ext4_sb_info *sbi = EXT4_SB(sb);
4188 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4189 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4190 ext4_lblk_t rr_cluster_start;
4191 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4192 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4193 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4195 /* The extent passed in that we are trying to match */
4196 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4197 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4199 /* The requested region passed into ext4_map_blocks() */
4200 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4202 if ((rr_cluster_start == ex_cluster_end) ||
4203 (rr_cluster_start == ex_cluster_start)) {
4204 if (rr_cluster_start == ex_cluster_end)
4205 ee_start += ee_len - 1;
4206 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4207 map->m_len = min(map->m_len,
4208 (unsigned) sbi->s_cluster_ratio - c_offset);
4210 * Check for and handle this case:
4212 * |--------- cluster # N-------------|
4213 * |------- extent ----|
4214 * |--- requested region ---|
4218 if (map->m_lblk < ee_block)
4219 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4222 * Check for the case where there is already another allocated
4223 * block to the right of 'ex' but before the end of the cluster.
4225 * |------------- cluster # N-------------|
4226 * |----- ex -----| |---- ex_right ----|
4227 * |------ requested region ------|
4228 * |================|
4230 if (map->m_lblk > ee_block) {
4231 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4232 map->m_len = min(map->m_len, next - map->m_lblk);
4235 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4239 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4245 * Block allocation/map/preallocation routine for extents based files
4248 * Need to be called with
4249 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4250 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4252 * return > 0, number of of blocks already mapped/allocated
4253 * if create == 0 and these are pre-allocated blocks
4254 * buffer head is unmapped
4255 * otherwise blocks are mapped
4257 * return = 0, if plain look up failed (blocks have not been allocated)
4258 * buffer head is unmapped
4260 * return < 0, error case.
4262 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4263 struct ext4_map_blocks *map, int flags)
4265 struct ext4_ext_path *path = NULL;
4266 struct ext4_extent newex, *ex, *ex2;
4267 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4268 ext4_fsblk_t newblock = 0;
4269 int free_on_err = 0, err = 0, depth, ret;
4270 unsigned int allocated = 0, offset = 0;
4271 unsigned int allocated_clusters = 0;
4272 struct ext4_allocation_request ar;
4273 ext4_io_end_t *io = ext4_inode_aio(inode);
4274 ext4_lblk_t cluster_offset;
4275 int set_unwritten = 0;
4277 ext_debug("blocks %u/%u requested for inode %lu\n",
4278 map->m_lblk, map->m_len, inode->i_ino);
4279 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4281 /* find extent for this block */
4282 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4284 err = PTR_ERR(path);
4289 depth = ext_depth(inode);
4292 * consistent leaf must not be empty;
4293 * this situation is possible, though, _during_ tree modification;
4294 * this is why assert can't be put in ext4_ext_find_extent()
4296 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4297 EXT4_ERROR_INODE(inode, "bad extent address "
4298 "lblock: %lu, depth: %d pblock %lld",
4299 (unsigned long) map->m_lblk, depth,
4300 path[depth].p_block);
4305 ex = path[depth].p_ext;
4307 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4308 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4309 unsigned short ee_len;
4313 * unwritten extents are treated as holes, except that
4314 * we split out initialized portions during a write.
4316 ee_len = ext4_ext_get_actual_len(ex);
4318 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4320 /* if found extent covers block, simply return it */
4321 if (in_range(map->m_lblk, ee_block, ee_len)) {
4322 newblock = map->m_lblk - ee_block + ee_start;
4323 /* number of remaining blocks in the extent */
4324 allocated = ee_len - (map->m_lblk - ee_block);
4325 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4326 ee_block, ee_len, newblock);
4329 * If the extent is initialized check whether the
4330 * caller wants to convert it to unwritten.
4332 if ((!ext4_ext_is_unwritten(ex)) &&
4333 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4334 allocated = ext4_ext_convert_initialized_extent(
4335 handle, inode, map, path, flags,
4336 allocated, newblock);
4338 } else if (!ext4_ext_is_unwritten(ex))
4341 ret = ext4_ext_handle_unwritten_extents(
4342 handle, inode, map, path, flags,
4343 allocated, newblock);
4352 if ((sbi->s_cluster_ratio > 1) &&
4353 ext4_find_delalloc_cluster(inode, map->m_lblk))
4354 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4357 * requested block isn't allocated yet;
4358 * we couldn't try to create block if create flag is zero
4360 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4362 * put just found gap into cache to speed up
4363 * subsequent requests
4365 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4366 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4371 * Okay, we need to do block allocation.
4373 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4374 newex.ee_block = cpu_to_le32(map->m_lblk);
4375 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4378 * If we are doing bigalloc, check to see if the extent returned
4379 * by ext4_ext_find_extent() implies a cluster we can use.
4381 if (cluster_offset && ex &&
4382 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4383 ar.len = allocated = map->m_len;
4384 newblock = map->m_pblk;
4385 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4386 goto got_allocated_blocks;
4389 /* find neighbour allocated blocks */
4390 ar.lleft = map->m_lblk;
4391 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4394 ar.lright = map->m_lblk;
4396 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4400 /* Check if the extent after searching to the right implies a
4401 * cluster we can use. */
4402 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4403 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4404 ar.len = allocated = map->m_len;
4405 newblock = map->m_pblk;
4406 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4407 goto got_allocated_blocks;
4411 * See if request is beyond maximum number of blocks we can have in
4412 * a single extent. For an initialized extent this limit is
4413 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4414 * EXT_UNWRITTEN_MAX_LEN.
4416 if (map->m_len > EXT_INIT_MAX_LEN &&
4417 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4418 map->m_len = EXT_INIT_MAX_LEN;
4419 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4420 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4421 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4423 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4424 newex.ee_len = cpu_to_le16(map->m_len);
4425 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4427 allocated = ext4_ext_get_actual_len(&newex);
4429 allocated = map->m_len;
4431 /* allocate new block */
4433 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4434 ar.logical = map->m_lblk;
4436 * We calculate the offset from the beginning of the cluster
4437 * for the logical block number, since when we allocate a
4438 * physical cluster, the physical block should start at the
4439 * same offset from the beginning of the cluster. This is
4440 * needed so that future calls to get_implied_cluster_alloc()
4443 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4444 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4446 ar.logical -= offset;
4447 if (S_ISREG(inode->i_mode))
4448 ar.flags = EXT4_MB_HINT_DATA;
4450 /* disable in-core preallocation for non-regular files */
4452 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4453 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4454 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4457 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4458 ar.goal, newblock, allocated);
4460 allocated_clusters = ar.len;
4461 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4462 if (ar.len > allocated)
4465 got_allocated_blocks:
4466 /* try to insert new extent into found leaf and return */
4467 ext4_ext_store_pblock(&newex, newblock + offset);
4468 newex.ee_len = cpu_to_le16(ar.len);
4469 /* Mark unwritten */
4470 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4471 ext4_ext_mark_unwritten(&newex);
4472 map->m_flags |= EXT4_MAP_UNWRITTEN;
4474 * io_end structure was created for every IO write to an
4475 * unwritten extent. To avoid unnecessary conversion,
4476 * here we flag the IO that really needs the conversion.
4477 * For non asycn direct IO case, flag the inode state
4478 * that we need to perform conversion when IO is done.
4480 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4485 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4486 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4489 err = ext4_ext_insert_extent(handle, inode, path,
4492 if (!err && set_unwritten) {
4494 ext4_set_io_unwritten_flag(inode, io);
4496 ext4_set_inode_state(inode,
4497 EXT4_STATE_DIO_UNWRITTEN);
4500 if (err && free_on_err) {
4501 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4502 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4503 /* free data blocks we just allocated */
4504 /* not a good idea to call discard here directly,
4505 * but otherwise we'd need to call it every free() */
4506 ext4_discard_preallocations(inode);
4507 ext4_free_blocks(handle, inode, NULL, newblock,
4508 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4512 /* previous routine could use block we allocated */
4513 newblock = ext4_ext_pblock(&newex);
4514 allocated = ext4_ext_get_actual_len(&newex);
4515 if (allocated > map->m_len)
4516 allocated = map->m_len;
4517 map->m_flags |= EXT4_MAP_NEW;
4520 * Update reserved blocks/metadata blocks after successful
4521 * block allocation which had been deferred till now.
4523 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4524 unsigned int reserved_clusters;
4526 * Check how many clusters we had reserved this allocated range
4528 reserved_clusters = get_reserved_cluster_alloc(inode,
4529 map->m_lblk, allocated);
4530 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4531 if (reserved_clusters) {
4533 * We have clusters reserved for this range.
4534 * But since we are not doing actual allocation
4535 * and are simply using blocks from previously
4536 * allocated cluster, we should release the
4537 * reservation and not claim quota.
4539 ext4_da_update_reserve_space(inode,
4540 reserved_clusters, 0);
4543 BUG_ON(allocated_clusters < reserved_clusters);
4544 if (reserved_clusters < allocated_clusters) {
4545 struct ext4_inode_info *ei = EXT4_I(inode);
4546 int reservation = allocated_clusters -
4549 * It seems we claimed few clusters outside of
4550 * the range of this allocation. We should give
4551 * it back to the reservation pool. This can
4552 * happen in the following case:
4554 * * Suppose s_cluster_ratio is 4 (i.e., each
4555 * cluster has 4 blocks. Thus, the clusters
4556 * are [0-3],[4-7],[8-11]...
4557 * * First comes delayed allocation write for
4558 * logical blocks 10 & 11. Since there were no
4559 * previous delayed allocated blocks in the
4560 * range [8-11], we would reserve 1 cluster
4562 * * Next comes write for logical blocks 3 to 8.
4563 * In this case, we will reserve 2 clusters
4564 * (for [0-3] and [4-7]; and not for [8-11] as
4565 * that range has a delayed allocated blocks.
4566 * Thus total reserved clusters now becomes 3.
4567 * * Now, during the delayed allocation writeout
4568 * time, we will first write blocks [3-8] and
4569 * allocate 3 clusters for writing these
4570 * blocks. Also, we would claim all these
4571 * three clusters above.
4572 * * Now when we come here to writeout the
4573 * blocks [10-11], we would expect to claim
4574 * the reservation of 1 cluster we had made
4575 * (and we would claim it since there are no
4576 * more delayed allocated blocks in the range
4577 * [8-11]. But our reserved cluster count had
4578 * already gone to 0.
4580 * Thus, at the step 4 above when we determine
4581 * that there are still some unwritten delayed
4582 * allocated blocks outside of our current
4583 * block range, we should increment the
4584 * reserved clusters count so that when the
4585 * remaining blocks finally gets written, we
4588 dquot_reserve_block(inode,
4589 EXT4_C2B(sbi, reservation));
4590 spin_lock(&ei->i_block_reservation_lock);
4591 ei->i_reserved_data_blocks += reservation;
4592 spin_unlock(&ei->i_block_reservation_lock);
4595 * We will claim quota for all newly allocated blocks.
4596 * We're updating the reserved space *after* the
4597 * correction above so we do not accidentally free
4598 * all the metadata reservation because we might
4599 * actually need it later on.
4601 ext4_da_update_reserve_space(inode, allocated_clusters,
4607 * Cache the extent and update transaction to commit on fdatasync only
4608 * when it is _not_ an unwritten extent.
4610 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4611 ext4_update_inode_fsync_trans(handle, inode, 1);
4613 ext4_update_inode_fsync_trans(handle, inode, 0);
4615 if (allocated > map->m_len)
4616 allocated = map->m_len;
4617 ext4_ext_show_leaf(inode, path);
4618 map->m_flags |= EXT4_MAP_MAPPED;
4619 map->m_pblk = newblock;
4620 map->m_len = allocated;
4623 ext4_ext_drop_refs(path);
4627 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4628 err ? err : allocated);
4629 ext4_es_lru_add(inode);
4630 return err ? err : allocated;
4633 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4635 struct super_block *sb = inode->i_sb;
4636 ext4_lblk_t last_block;
4640 * TODO: optimization is possible here.
4641 * Probably we need not scan at all,
4642 * because page truncation is enough.
4645 /* we have to know where to truncate from in crash case */
4646 EXT4_I(inode)->i_disksize = inode->i_size;
4647 ext4_mark_inode_dirty(handle, inode);
4649 last_block = (inode->i_size + sb->s_blocksize - 1)
4650 >> EXT4_BLOCK_SIZE_BITS(sb);
4652 err = ext4_es_remove_extent(inode, last_block,
4653 EXT_MAX_BLOCKS - last_block);
4654 if (err == -ENOMEM) {
4656 congestion_wait(BLK_RW_ASYNC, HZ/50);
4660 ext4_std_error(inode->i_sb, err);
4663 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4664 ext4_std_error(inode->i_sb, err);
4667 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4668 ext4_lblk_t len, int flags, int mode)
4670 struct inode *inode = file_inode(file);
4675 struct ext4_map_blocks map;
4676 unsigned int credits;
4678 map.m_lblk = offset;
4680 * Don't normalize the request if it can fit in one extent so
4681 * that it doesn't get unnecessarily split into multiple
4684 if (len <= EXT_UNWRITTEN_MAX_LEN)
4685 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4688 * credits to insert 1 extent into extent tree
4690 credits = ext4_chunk_trans_blocks(inode, len);
4693 while (ret >= 0 && ret < len) {
4694 map.m_lblk = map.m_lblk + ret;
4695 map.m_len = len = len - ret;
4696 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4698 if (IS_ERR(handle)) {
4699 ret = PTR_ERR(handle);
4702 ret = ext4_map_blocks(handle, inode, &map, flags);
4704 ext4_debug("inode #%lu: block %u: len %u: "
4705 "ext4_ext_map_blocks returned %d",
4706 inode->i_ino, map.m_lblk,
4708 ext4_mark_inode_dirty(handle, inode);
4709 ret2 = ext4_journal_stop(handle);
4712 ret2 = ext4_journal_stop(handle);
4716 if (ret == -ENOSPC &&
4717 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4722 return ret > 0 ? ret2 : ret;
4725 static long ext4_zero_range(struct file *file, loff_t offset,
4726 loff_t len, int mode)
4728 struct inode *inode = file_inode(file);
4729 handle_t *handle = NULL;
4730 unsigned int max_blocks;
4731 loff_t new_size = 0;
4738 struct address_space *mapping = inode->i_mapping;
4739 unsigned int blkbits = inode->i_blkbits;
4741 trace_ext4_zero_range(inode, offset, len, mode);
4743 if (!S_ISREG(inode->i_mode))
4746 /* Call ext4_force_commit to flush all data in case of data=journal. */
4747 if (ext4_should_journal_data(inode)) {
4748 ret = ext4_force_commit(inode->i_sb);
4754 * Write out all dirty pages to avoid race conditions
4755 * Then release them.
4757 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4758 ret = filemap_write_and_wait_range(mapping, offset,
4765 * Round up offset. This is not fallocate, we neet to zero out
4766 * blocks, so convert interior block aligned part of the range to
4767 * unwritten and possibly manually zero out unaligned parts of the
4770 start = round_up(offset, 1 << blkbits);
4771 end = round_down((offset + len), 1 << blkbits);
4773 if (start < offset || end > offset + len)
4775 partial = (offset + len) & ((1 << blkbits) - 1);
4777 lblk = start >> blkbits;
4778 max_blocks = (end >> blkbits);
4779 if (max_blocks < lblk)
4784 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4785 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN;
4786 if (mode & FALLOC_FL_KEEP_SIZE)
4787 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4789 mutex_lock(&inode->i_mutex);
4792 * Indirect files do not support unwritten extnets
4794 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4799 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4800 offset + len > i_size_read(inode)) {
4801 new_size = offset + len;
4802 ret = inode_newsize_ok(inode, new_size);
4806 * If we have a partial block after EOF we have to allocate
4813 if (max_blocks > 0) {
4815 /* Now release the pages and zero block aligned part of pages*/
4816 truncate_pagecache_range(inode, start, end - 1);
4818 /* Wait all existing dio workers, newcomers will block on i_mutex */
4819 ext4_inode_block_unlocked_dio(inode);
4820 inode_dio_wait(inode);
4823 * Remove entire range from the extent status tree.
4825 ret = ext4_es_remove_extent(inode, lblk, max_blocks);
4829 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, flags,
4835 * In worst case we have to writeout two nonadjacent unwritten
4836 * blocks and update the inode
4838 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4839 if (ext4_should_journal_data(inode))
4841 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4842 if (IS_ERR(handle)) {
4843 ret = PTR_ERR(handle);
4844 ext4_std_error(inode->i_sb, ret);
4848 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4850 ext4_update_inode_size(inode, new_size);
4853 * Mark that we allocate beyond EOF so the subsequent truncate
4854 * can proceed even if the new size is the same as i_size.
4856 if ((offset + len) > i_size_read(inode))
4857 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4860 ext4_mark_inode_dirty(handle, inode);
4862 /* Zero out partial block at the edges of the range */
4863 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4865 if (file->f_flags & O_SYNC)
4866 ext4_handle_sync(handle);
4868 ext4_journal_stop(handle);
4870 ext4_inode_resume_unlocked_dio(inode);
4872 mutex_unlock(&inode->i_mutex);
4877 * preallocate space for a file. This implements ext4's fallocate file
4878 * operation, which gets called from sys_fallocate system call.
4879 * For block-mapped files, posix_fallocate should fall back to the method
4880 * of writing zeroes to the required new blocks (the same behavior which is
4881 * expected for file systems which do not support fallocate() system call).
4883 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4885 struct inode *inode = file_inode(file);
4887 loff_t new_size = 0;
4888 unsigned int max_blocks;
4892 unsigned int blkbits = inode->i_blkbits;
4894 /* Return error if mode is not supported */
4895 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4896 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4899 if (mode & FALLOC_FL_PUNCH_HOLE)
4900 return ext4_punch_hole(inode, offset, len);
4902 ret = ext4_convert_inline_data(inode);
4907 * currently supporting (pre)allocate mode for extent-based
4910 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4913 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4914 return ext4_collapse_range(inode, offset, len);
4916 if (mode & FALLOC_FL_ZERO_RANGE)
4917 return ext4_zero_range(file, offset, len, mode);
4919 trace_ext4_fallocate_enter(inode, offset, len, mode);
4920 lblk = offset >> blkbits;
4922 * We can't just convert len to max_blocks because
4923 * If blocksize = 4096 offset = 3072 and len = 2048
4925 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4928 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4929 if (mode & FALLOC_FL_KEEP_SIZE)
4930 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4932 mutex_lock(&inode->i_mutex);
4934 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4935 offset + len > i_size_read(inode)) {
4936 new_size = offset + len;
4937 ret = inode_newsize_ok(inode, new_size);
4942 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, flags, mode);
4946 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4950 inode->i_ctime = ext4_current_time(inode);
4953 if (ext4_update_inode_size(inode, new_size) & 0x1)
4954 inode->i_mtime = inode->i_ctime;
4957 * Mark that we allocate beyond EOF so the subsequent truncate
4958 * can proceed even if the new size is the same as i_size.
4960 if ((offset + len) > i_size_read(inode))
4961 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4963 ext4_mark_inode_dirty(handle, inode);
4964 if (file->f_flags & O_SYNC)
4965 ext4_handle_sync(handle);
4967 ext4_journal_stop(handle);
4969 mutex_unlock(&inode->i_mutex);
4970 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4975 * This function convert a range of blocks to written extents
4976 * The caller of this function will pass the start offset and the size.
4977 * all unwritten extents within this range will be converted to
4980 * This function is called from the direct IO end io call back
4981 * function, to convert the fallocated extents after IO is completed.
4982 * Returns 0 on success.
4984 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4985 loff_t offset, ssize_t len)
4987 unsigned int max_blocks;
4990 struct ext4_map_blocks map;
4991 unsigned int credits, blkbits = inode->i_blkbits;
4993 map.m_lblk = offset >> blkbits;
4995 * We can't just convert len to max_blocks because
4996 * If blocksize = 4096 offset = 3072 and len = 2048
4998 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5001 * This is somewhat ugly but the idea is clear: When transaction is
5002 * reserved, everything goes into it. Otherwise we rather start several
5003 * smaller transactions for conversion of each extent separately.
5006 handle = ext4_journal_start_reserved(handle,
5007 EXT4_HT_EXT_CONVERT);
5009 return PTR_ERR(handle);
5013 * credits to insert 1 extent into extent tree
5015 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5017 while (ret >= 0 && ret < max_blocks) {
5019 map.m_len = (max_blocks -= ret);
5021 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5023 if (IS_ERR(handle)) {
5024 ret = PTR_ERR(handle);
5028 ret = ext4_map_blocks(handle, inode, &map,
5029 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5031 ext4_warning(inode->i_sb,
5032 "inode #%lu: block %u: len %u: "
5033 "ext4_ext_map_blocks returned %d",
5034 inode->i_ino, map.m_lblk,
5036 ext4_mark_inode_dirty(handle, inode);
5038 ret2 = ext4_journal_stop(handle);
5039 if (ret <= 0 || ret2)
5043 ret2 = ext4_journal_stop(handle);
5044 return ret > 0 ? ret2 : ret;
5048 * If newes is not existing extent (newes->ec_pblk equals zero) find
5049 * delayed extent at start of newes and update newes accordingly and
5050 * return start of the next delayed extent.
5052 * If newes is existing extent (newes->ec_pblk is not equal zero)
5053 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5054 * extent found. Leave newes unmodified.
5056 static int ext4_find_delayed_extent(struct inode *inode,
5057 struct extent_status *newes)
5059 struct extent_status es;
5060 ext4_lblk_t block, next_del;
5062 if (newes->es_pblk == 0) {
5063 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5064 newes->es_lblk + newes->es_len - 1, &es);
5067 * No extent in extent-tree contains block @newes->es_pblk,
5068 * then the block may stay in 1)a hole or 2)delayed-extent.
5074 if (es.es_lblk > newes->es_lblk) {
5076 newes->es_len = min(es.es_lblk - newes->es_lblk,
5081 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5084 block = newes->es_lblk + newes->es_len;
5085 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5087 next_del = EXT_MAX_BLOCKS;
5089 next_del = es.es_lblk;
5093 /* fiemap flags we can handle specified here */
5094 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5096 static int ext4_xattr_fiemap(struct inode *inode,
5097 struct fiemap_extent_info *fieinfo)
5101 __u32 flags = FIEMAP_EXTENT_LAST;
5102 int blockbits = inode->i_sb->s_blocksize_bits;
5106 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5107 struct ext4_iloc iloc;
5108 int offset; /* offset of xattr in inode */
5110 error = ext4_get_inode_loc(inode, &iloc);
5113 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5114 offset = EXT4_GOOD_OLD_INODE_SIZE +
5115 EXT4_I(inode)->i_extra_isize;
5117 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5118 flags |= FIEMAP_EXTENT_DATA_INLINE;
5120 } else { /* external block */
5121 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5122 length = inode->i_sb->s_blocksize;
5126 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5128 return (error < 0 ? error : 0);
5131 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5132 __u64 start, __u64 len)
5134 ext4_lblk_t start_blk;
5137 if (ext4_has_inline_data(inode)) {
5140 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5146 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5147 error = ext4_ext_precache(inode);
5152 /* fallback to generic here if not in extents fmt */
5153 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5154 return generic_block_fiemap(inode, fieinfo, start, len,
5157 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5160 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5161 error = ext4_xattr_fiemap(inode, fieinfo);
5163 ext4_lblk_t len_blks;
5166 start_blk = start >> inode->i_sb->s_blocksize_bits;
5167 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5168 if (last_blk >= EXT_MAX_BLOCKS)
5169 last_blk = EXT_MAX_BLOCKS-1;
5170 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5173 * Walk the extent tree gathering extent information
5174 * and pushing extents back to the user.
5176 error = ext4_fill_fiemap_extents(inode, start_blk,
5179 ext4_es_lru_add(inode);
5185 * Function to access the path buffer for marking it dirty.
5186 * It also checks if there are sufficient credits left in the journal handle
5190 ext4_access_path(handle_t *handle, struct inode *inode,
5191 struct ext4_ext_path *path)
5195 if (!ext4_handle_valid(handle))
5199 * Check if need to extend journal credits
5200 * 3 for leaf, sb, and inode plus 2 (bmap and group
5201 * descriptor) for each block group; assume two block
5204 if (handle->h_buffer_credits < 7) {
5205 credits = ext4_writepage_trans_blocks(inode);
5206 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5207 /* EAGAIN is success */
5208 if (err && err != -EAGAIN)
5212 err = ext4_ext_get_access(handle, inode, path);
5217 * ext4_ext_shift_path_extents:
5218 * Shift the extents of a path structure lying between path[depth].p_ext
5219 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5220 * from starting block for each extent.
5223 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5224 struct inode *inode, handle_t *handle,
5228 struct ext4_extent *ex_start, *ex_last;
5230 depth = path->p_depth;
5232 while (depth >= 0) {
5233 if (depth == path->p_depth) {
5234 ex_start = path[depth].p_ext;
5238 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5242 err = ext4_access_path(handle, inode, path + depth);
5246 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5249 *start = le32_to_cpu(ex_last->ee_block) +
5250 ext4_ext_get_actual_len(ex_last);
5252 while (ex_start <= ex_last) {
5253 le32_add_cpu(&ex_start->ee_block, -shift);
5254 /* Try to merge to the left. */
5256 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5257 ext4_ext_try_to_merge_right(inode,
5258 path, ex_start - 1))
5263 err = ext4_ext_dirty(handle, inode, path + depth);
5267 if (--depth < 0 || !update)
5271 /* Update index too */
5272 err = ext4_access_path(handle, inode, path + depth);
5276 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5277 err = ext4_ext_dirty(handle, inode, path + depth);
5281 /* we are done if current index is not a starting index */
5282 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5293 * ext4_ext_shift_extents:
5294 * All the extents which lies in the range from start to the last allocated
5295 * block for the file are shifted downwards by shift blocks.
5296 * On success, 0 is returned, error otherwise.
5299 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5300 ext4_lblk_t start, ext4_lblk_t shift)
5302 struct ext4_ext_path *path;
5304 struct ext4_extent *extent;
5305 ext4_lblk_t stop_block, current_block;
5306 ext4_lblk_t ex_start, ex_end;
5308 /* Let path point to the last extent */
5309 path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5311 return PTR_ERR(path);
5313 depth = path->p_depth;
5314 extent = path[depth].p_ext;
5316 ext4_ext_drop_refs(path);
5321 stop_block = le32_to_cpu(extent->ee_block) +
5322 ext4_ext_get_actual_len(extent);
5323 ext4_ext_drop_refs(path);
5326 /* Nothing to shift, if hole is at the end of file */
5327 if (start >= stop_block)
5331 * Don't start shifting extents until we make sure the hole is big
5332 * enough to accomodate the shift.
5334 path = ext4_ext_find_extent(inode, start - 1, NULL, 0);
5336 return PTR_ERR(path);
5337 depth = path->p_depth;
5338 extent = path[depth].p_ext;
5340 ex_start = le32_to_cpu(extent->ee_block);
5341 ex_end = le32_to_cpu(extent->ee_block) +
5342 ext4_ext_get_actual_len(extent);
5347 ext4_ext_drop_refs(path);
5350 if ((start == ex_start && shift > ex_start) ||
5351 (shift > start - ex_end))
5354 /* Its safe to start updating extents */
5355 while (start < stop_block) {
5356 path = ext4_ext_find_extent(inode, start, NULL, 0);
5358 return PTR_ERR(path);
5359 depth = path->p_depth;
5360 extent = path[depth].p_ext;
5362 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5363 (unsigned long) start);
5367 current_block = le32_to_cpu(extent->ee_block);
5368 if (start > current_block) {
5369 /* Hole, move to the next extent */
5370 ret = mext_next_extent(inode, path, &extent);
5372 ext4_ext_drop_refs(path);
5379 ret = ext4_ext_shift_path_extents(path, shift, inode,
5381 ext4_ext_drop_refs(path);
5391 * ext4_collapse_range:
5392 * This implements the fallocate's collapse range functionality for ext4
5393 * Returns: 0 and non-zero on error.
5395 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5397 struct super_block *sb = inode->i_sb;
5398 ext4_lblk_t punch_start, punch_stop;
5400 unsigned int credits;
5401 loff_t new_size, ioffset;
5404 /* Collapse range works only on fs block size aligned offsets. */
5405 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5406 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5409 if (!S_ISREG(inode->i_mode))
5412 trace_ext4_collapse_range(inode, offset, len);
5414 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5415 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5417 /* Call ext4_force_commit to flush all data in case of data=journal. */
5418 if (ext4_should_journal_data(inode)) {
5419 ret = ext4_force_commit(inode->i_sb);
5425 * Need to round down offset to be aligned with page size boundary
5426 * for page size > block size.
5428 ioffset = round_down(offset, PAGE_SIZE);
5430 /* Write out all dirty pages */
5431 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5436 /* Take mutex lock */
5437 mutex_lock(&inode->i_mutex);
5440 * There is no need to overlap collapse range with EOF, in which case
5441 * it is effectively a truncate operation
5443 if (offset + len >= i_size_read(inode)) {
5448 /* Currently just for extent based files */
5449 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5454 truncate_pagecache(inode, ioffset);
5456 /* Wait for existing dio to complete */
5457 ext4_inode_block_unlocked_dio(inode);
5458 inode_dio_wait(inode);
5460 credits = ext4_writepage_trans_blocks(inode);
5461 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5462 if (IS_ERR(handle)) {
5463 ret = PTR_ERR(handle);
5467 down_write(&EXT4_I(inode)->i_data_sem);
5468 ext4_discard_preallocations(inode);
5470 ret = ext4_es_remove_extent(inode, punch_start,
5471 EXT_MAX_BLOCKS - punch_start);
5473 up_write(&EXT4_I(inode)->i_data_sem);
5477 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5479 up_write(&EXT4_I(inode)->i_data_sem);
5482 ext4_discard_preallocations(inode);
5484 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5485 punch_stop - punch_start);
5487 up_write(&EXT4_I(inode)->i_data_sem);
5491 new_size = i_size_read(inode) - len;
5492 i_size_write(inode, new_size);
5493 EXT4_I(inode)->i_disksize = new_size;
5495 up_write(&EXT4_I(inode)->i_data_sem);
5497 ext4_handle_sync(handle);
5498 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5499 ext4_mark_inode_dirty(handle, inode);
5502 ext4_journal_stop(handle);
5504 ext4_inode_resume_unlocked_dio(inode);
5506 mutex_unlock(&inode->i_mutex);