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 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
166 /* path points to block */
167 err = __ext4_handle_dirty_metadata(where, line, handle,
170 /* path points to leaf/index in inode body */
171 err = ext4_mark_inode_dirty(handle, inode);
176 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
177 struct ext4_ext_path *path,
181 int depth = path->p_depth;
182 struct ext4_extent *ex;
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
198 * especially if the latter case turns out to be
201 ex = path[depth].p_ext;
203 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
204 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
206 if (block > ext_block)
207 return ext_pblk + (block - ext_block);
209 return ext_pblk - (ext_block - block);
212 /* it looks like index is empty;
213 * try to find starting block from index itself */
214 if (path[depth].p_bh)
215 return path[depth].p_bh->b_blocknr;
218 /* OK. use inode's group */
219 return ext4_inode_to_goal_block(inode);
223 * Allocation for a meta data block
226 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
227 struct ext4_ext_path *path,
228 struct ext4_extent *ex, int *err, unsigned int flags)
230 ext4_fsblk_t goal, newblock;
232 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
233 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
238 static inline int ext4_ext_space_block(struct inode *inode, int check)
242 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent);
244 #ifdef AGGRESSIVE_TEST
245 if (!check && size > 6)
251 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
255 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
256 / sizeof(struct ext4_extent_idx);
257 #ifdef AGGRESSIVE_TEST
258 if (!check && size > 5)
264 static inline int ext4_ext_space_root(struct inode *inode, int check)
268 size = sizeof(EXT4_I(inode)->i_data);
269 size -= sizeof(struct ext4_extent_header);
270 size /= sizeof(struct ext4_extent);
271 #ifdef AGGRESSIVE_TEST
272 if (!check && size > 3)
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
285 #ifdef AGGRESSIVE_TEST
286 if (!check && size > 4)
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
297 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
299 struct ext4_inode_info *ei = EXT4_I(inode);
302 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
303 / sizeof(struct ext4_extent_idx));
306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
313 if (ei->i_da_metadata_calc_len &&
314 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
317 if ((ei->i_da_metadata_calc_len % idxs) == 0)
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
323 ei->i_da_metadata_calc_len = 0;
325 ei->i_da_metadata_calc_len++;
326 ei->i_da_metadata_calc_last_lblock++;
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
334 ei->i_da_metadata_calc_len = 1;
335 ei->i_da_metadata_calc_last_lblock = lblock;
336 return ext_depth(inode) + 1;
340 ext4_ext_max_entries(struct inode *inode, int depth)
344 if (depth == ext_depth(inode)) {
346 max = ext4_ext_space_root(inode, 1);
348 max = ext4_ext_space_root_idx(inode, 1);
351 max = ext4_ext_space_block(inode, 1);
353 max = ext4_ext_space_block_idx(inode, 1);
359 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
361 ext4_fsblk_t block = ext4_ext_pblock(ext);
362 int len = ext4_ext_get_actual_len(ext);
363 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
364 ext4_lblk_t last = lblock + len - 1;
368 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
371 static int ext4_valid_extent_idx(struct inode *inode,
372 struct ext4_extent_idx *ext_idx)
374 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
376 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
379 static int ext4_valid_extent_entries(struct inode *inode,
380 struct ext4_extent_header *eh,
383 unsigned short entries;
384 if (eh->eh_entries == 0)
387 entries = le16_to_cpu(eh->eh_entries);
391 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
392 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
393 ext4_fsblk_t pblock = 0;
394 ext4_lblk_t lblock = 0;
395 ext4_lblk_t prev = 0;
398 if (!ext4_valid_extent(inode, ext))
401 /* Check for overlapping extents */
402 lblock = le32_to_cpu(ext->ee_block);
403 len = ext4_ext_get_actual_len(ext);
404 if ((lblock <= prev) && prev) {
405 pblock = ext4_ext_pblock(ext);
406 es->s_last_error_block = cpu_to_le64(pblock);
411 prev = lblock + len - 1;
414 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
416 if (!ext4_valid_extent_idx(inode, ext_idx))
425 static int __ext4_ext_check(const char *function, unsigned int line,
426 struct inode *inode, struct ext4_extent_header *eh,
427 int depth, ext4_fsblk_t pblk)
429 const char *error_msg;
432 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
433 error_msg = "invalid magic";
436 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
437 error_msg = "unexpected eh_depth";
440 if (unlikely(eh->eh_max == 0)) {
441 error_msg = "invalid eh_max";
444 max = ext4_ext_max_entries(inode, depth);
445 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
446 error_msg = "too large eh_max";
449 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
450 error_msg = "invalid eh_entries";
453 if (!ext4_valid_extent_entries(inode, eh, depth)) {
454 error_msg = "invalid extent entries";
457 /* Verify checksum on non-root extent tree nodes */
458 if (ext_depth(inode) != depth &&
459 !ext4_extent_block_csum_verify(inode, eh)) {
460 error_msg = "extent tree corrupted";
466 ext4_error_inode(inode, function, line, 0,
467 "pblk %llu bad header/extent: %s - magic %x, "
468 "entries %u, max %u(%u), depth %u(%u)",
469 (unsigned long long) pblk, error_msg,
470 le16_to_cpu(eh->eh_magic),
471 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
472 max, le16_to_cpu(eh->eh_depth), depth);
476 #define ext4_ext_check(inode, eh, depth, pblk) \
477 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
479 int ext4_ext_check_inode(struct inode *inode)
481 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
484 static struct buffer_head *
485 __read_extent_tree_block(const char *function, unsigned int line,
486 struct inode *inode, ext4_fsblk_t pblk, int depth,
489 struct buffer_head *bh;
492 bh = sb_getblk(inode->i_sb, pblk);
494 return ERR_PTR(-ENOMEM);
496 if (!bh_uptodate_or_lock(bh)) {
497 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
498 err = bh_submit_read(bh);
502 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
504 err = __ext4_ext_check(function, line, inode,
505 ext_block_hdr(bh), depth, pblk);
508 set_buffer_verified(bh);
510 * If this is a leaf block, cache all of its entries
512 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
513 struct ext4_extent_header *eh = ext_block_hdr(bh);
514 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
515 ext4_lblk_t prev = 0;
518 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
519 unsigned int status = EXTENT_STATUS_WRITTEN;
520 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
521 int len = ext4_ext_get_actual_len(ex);
523 if (prev && (prev != lblk))
524 ext4_es_cache_extent(inode, prev,
528 if (ext4_ext_is_unwritten(ex))
529 status = EXTENT_STATUS_UNWRITTEN;
530 ext4_es_cache_extent(inode, lblk, len,
531 ext4_ext_pblock(ex), status);
542 #define read_extent_tree_block(inode, pblk, depth, flags) \
543 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
547 * This function is called to cache a file's extent information in the
550 int ext4_ext_precache(struct inode *inode)
552 struct ext4_inode_info *ei = EXT4_I(inode);
553 struct ext4_ext_path *path = NULL;
554 struct buffer_head *bh;
555 int i = 0, depth, ret = 0;
557 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
558 return 0; /* not an extent-mapped inode */
560 down_read(&ei->i_data_sem);
561 depth = ext_depth(inode);
563 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
566 up_read(&ei->i_data_sem);
570 /* Don't cache anything if there are no external extent blocks */
573 path[0].p_hdr = ext_inode_hdr(inode);
574 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
577 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
580 * If this is a leaf block or we've reached the end of
581 * the index block, go up
584 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
585 brelse(path[i].p_bh);
590 bh = read_extent_tree_block(inode,
591 ext4_idx_pblock(path[i].p_idx++),
593 EXT4_EX_FORCE_CACHE);
600 path[i].p_hdr = ext_block_hdr(bh);
601 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
603 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
605 up_read(&ei->i_data_sem);
606 ext4_ext_drop_refs(path);
612 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
614 int k, l = path->p_depth;
617 for (k = 0; k <= l; k++, path++) {
619 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
620 ext4_idx_pblock(path->p_idx));
621 } else if (path->p_ext) {
622 ext_debug(" %d:[%d]%d:%llu ",
623 le32_to_cpu(path->p_ext->ee_block),
624 ext4_ext_is_unwritten(path->p_ext),
625 ext4_ext_get_actual_len(path->p_ext),
626 ext4_ext_pblock(path->p_ext));
633 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
635 int depth = ext_depth(inode);
636 struct ext4_extent_header *eh;
637 struct ext4_extent *ex;
643 eh = path[depth].p_hdr;
644 ex = EXT_FIRST_EXTENT(eh);
646 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
648 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
649 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
650 ext4_ext_is_unwritten(ex),
651 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
656 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
657 ext4_fsblk_t newblock, int level)
659 int depth = ext_depth(inode);
660 struct ext4_extent *ex;
662 if (depth != level) {
663 struct ext4_extent_idx *idx;
664 idx = path[level].p_idx;
665 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
666 ext_debug("%d: move %d:%llu in new index %llu\n", level,
667 le32_to_cpu(idx->ei_block),
668 ext4_idx_pblock(idx),
676 ex = path[depth].p_ext;
677 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
678 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
679 le32_to_cpu(ex->ee_block),
681 ext4_ext_is_unwritten(ex),
682 ext4_ext_get_actual_len(ex),
689 #define ext4_ext_show_path(inode, path)
690 #define ext4_ext_show_leaf(inode, path)
691 #define ext4_ext_show_move(inode, path, newblock, level)
694 void ext4_ext_drop_refs(struct ext4_ext_path *path)
696 int depth = path->p_depth;
699 for (i = 0; i <= depth; i++, path++)
707 * ext4_ext_binsearch_idx:
708 * binary search for the closest index of the given block
709 * the header must be checked before calling this
712 ext4_ext_binsearch_idx(struct inode *inode,
713 struct ext4_ext_path *path, ext4_lblk_t block)
715 struct ext4_extent_header *eh = path->p_hdr;
716 struct ext4_extent_idx *r, *l, *m;
719 ext_debug("binsearch for %u(idx): ", block);
721 l = EXT_FIRST_INDEX(eh) + 1;
722 r = EXT_LAST_INDEX(eh);
725 if (block < le32_to_cpu(m->ei_block))
729 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
730 m, le32_to_cpu(m->ei_block),
731 r, le32_to_cpu(r->ei_block));
735 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
736 ext4_idx_pblock(path->p_idx));
738 #ifdef CHECK_BINSEARCH
740 struct ext4_extent_idx *chix, *ix;
743 chix = ix = EXT_FIRST_INDEX(eh);
744 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
746 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
747 printk(KERN_DEBUG "k=%d, ix=0x%p, "
749 ix, EXT_FIRST_INDEX(eh));
750 printk(KERN_DEBUG "%u <= %u\n",
751 le32_to_cpu(ix->ei_block),
752 le32_to_cpu(ix[-1].ei_block));
754 BUG_ON(k && le32_to_cpu(ix->ei_block)
755 <= le32_to_cpu(ix[-1].ei_block));
756 if (block < le32_to_cpu(ix->ei_block))
760 BUG_ON(chix != path->p_idx);
767 * ext4_ext_binsearch:
768 * binary search for closest extent of the given block
769 * the header must be checked before calling this
772 ext4_ext_binsearch(struct inode *inode,
773 struct ext4_ext_path *path, ext4_lblk_t block)
775 struct ext4_extent_header *eh = path->p_hdr;
776 struct ext4_extent *r, *l, *m;
778 if (eh->eh_entries == 0) {
780 * this leaf is empty:
781 * we get such a leaf in split/add case
786 ext_debug("binsearch for %u: ", block);
788 l = EXT_FIRST_EXTENT(eh) + 1;
789 r = EXT_LAST_EXTENT(eh);
793 if (block < le32_to_cpu(m->ee_block))
797 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
798 m, le32_to_cpu(m->ee_block),
799 r, le32_to_cpu(r->ee_block));
803 ext_debug(" -> %d:%llu:[%d]%d ",
804 le32_to_cpu(path->p_ext->ee_block),
805 ext4_ext_pblock(path->p_ext),
806 ext4_ext_is_unwritten(path->p_ext),
807 ext4_ext_get_actual_len(path->p_ext));
809 #ifdef CHECK_BINSEARCH
811 struct ext4_extent *chex, *ex;
814 chex = ex = EXT_FIRST_EXTENT(eh);
815 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
816 BUG_ON(k && le32_to_cpu(ex->ee_block)
817 <= le32_to_cpu(ex[-1].ee_block));
818 if (block < le32_to_cpu(ex->ee_block))
822 BUG_ON(chex != path->p_ext);
828 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
830 struct ext4_extent_header *eh;
832 eh = ext_inode_hdr(inode);
835 eh->eh_magic = EXT4_EXT_MAGIC;
836 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
837 ext4_mark_inode_dirty(handle, inode);
841 struct ext4_ext_path *
842 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
843 struct ext4_ext_path *path, int flags)
845 struct ext4_extent_header *eh;
846 struct buffer_head *bh;
847 short int depth, i, ppos = 0, alloc = 0;
850 eh = ext_inode_hdr(inode);
851 depth = ext_depth(inode);
853 /* account possible depth increase */
855 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
858 return ERR_PTR(-ENOMEM);
865 /* walk through the tree */
867 ext_debug("depth %d: num %d, max %d\n",
868 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
870 ext4_ext_binsearch_idx(inode, path + ppos, block);
871 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
872 path[ppos].p_depth = i;
873 path[ppos].p_ext = NULL;
875 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
882 eh = ext_block_hdr(bh);
884 if (unlikely(ppos > depth)) {
886 EXT4_ERROR_INODE(inode,
887 "ppos %d > depth %d", ppos, depth);
891 path[ppos].p_bh = bh;
892 path[ppos].p_hdr = eh;
895 path[ppos].p_depth = i;
896 path[ppos].p_ext = NULL;
897 path[ppos].p_idx = NULL;
900 ext4_ext_binsearch(inode, path + ppos, block);
901 /* if not an empty leaf */
902 if (path[ppos].p_ext)
903 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
905 ext4_ext_show_path(inode, path);
910 ext4_ext_drop_refs(path);
917 * ext4_ext_insert_index:
918 * insert new index [@logical;@ptr] into the block at @curp;
919 * check where to insert: before @curp or after @curp
921 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
922 struct ext4_ext_path *curp,
923 int logical, ext4_fsblk_t ptr)
925 struct ext4_extent_idx *ix;
928 err = ext4_ext_get_access(handle, inode, curp);
932 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
933 EXT4_ERROR_INODE(inode,
934 "logical %d == ei_block %d!",
935 logical, le32_to_cpu(curp->p_idx->ei_block));
939 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
940 >= le16_to_cpu(curp->p_hdr->eh_max))) {
941 EXT4_ERROR_INODE(inode,
942 "eh_entries %d >= eh_max %d!",
943 le16_to_cpu(curp->p_hdr->eh_entries),
944 le16_to_cpu(curp->p_hdr->eh_max));
948 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
950 ext_debug("insert new index %d after: %llu\n", logical, ptr);
951 ix = curp->p_idx + 1;
954 ext_debug("insert new index %d before: %llu\n", logical, ptr);
958 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
961 ext_debug("insert new index %d: "
962 "move %d indices from 0x%p to 0x%p\n",
963 logical, len, ix, ix + 1);
964 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
967 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
968 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
972 ix->ei_block = cpu_to_le32(logical);
973 ext4_idx_store_pblock(ix, ptr);
974 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
976 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
977 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
981 err = ext4_ext_dirty(handle, inode, curp);
982 ext4_std_error(inode->i_sb, err);
989 * inserts new subtree into the path, using free index entry
991 * - allocates all needed blocks (new leaf and all intermediate index blocks)
992 * - makes decision where to split
993 * - moves remaining extents and index entries (right to the split point)
994 * into the newly allocated blocks
995 * - initializes subtree
997 static int ext4_ext_split(handle_t *handle, struct inode *inode,
999 struct ext4_ext_path *path,
1000 struct ext4_extent *newext, int at)
1002 struct buffer_head *bh = NULL;
1003 int depth = ext_depth(inode);
1004 struct ext4_extent_header *neh;
1005 struct ext4_extent_idx *fidx;
1006 int i = at, k, m, a;
1007 ext4_fsblk_t newblock, oldblock;
1009 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1012 /* make decision: where to split? */
1013 /* FIXME: now decision is simplest: at current extent */
1015 /* if current leaf will be split, then we should use
1016 * border from split point */
1017 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1018 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1021 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1022 border = path[depth].p_ext[1].ee_block;
1023 ext_debug("leaf will be split."
1024 " next leaf starts at %d\n",
1025 le32_to_cpu(border));
1027 border = newext->ee_block;
1028 ext_debug("leaf will be added."
1029 " next leaf starts at %d\n",
1030 le32_to_cpu(border));
1034 * If error occurs, then we break processing
1035 * and mark filesystem read-only. index won't
1036 * be inserted and tree will be in consistent
1037 * state. Next mount will repair buffers too.
1041 * Get array to track all allocated blocks.
1042 * We need this to handle errors and free blocks
1045 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1049 /* allocate all needed blocks */
1050 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1051 for (a = 0; a < depth - at; a++) {
1052 newblock = ext4_ext_new_meta_block(handle, inode, path,
1053 newext, &err, flags);
1056 ablocks[a] = newblock;
1059 /* initialize new leaf */
1060 newblock = ablocks[--a];
1061 if (unlikely(newblock == 0)) {
1062 EXT4_ERROR_INODE(inode, "newblock == 0!");
1066 bh = sb_getblk(inode->i_sb, newblock);
1067 if (unlikely(!bh)) {
1073 err = ext4_journal_get_create_access(handle, bh);
1077 neh = ext_block_hdr(bh);
1078 neh->eh_entries = 0;
1079 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1080 neh->eh_magic = EXT4_EXT_MAGIC;
1083 /* move remainder of path[depth] to the new leaf */
1084 if (unlikely(path[depth].p_hdr->eh_entries !=
1085 path[depth].p_hdr->eh_max)) {
1086 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1087 path[depth].p_hdr->eh_entries,
1088 path[depth].p_hdr->eh_max);
1092 /* start copy from next extent */
1093 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1094 ext4_ext_show_move(inode, path, newblock, depth);
1096 struct ext4_extent *ex;
1097 ex = EXT_FIRST_EXTENT(neh);
1098 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1099 le16_add_cpu(&neh->eh_entries, m);
1102 ext4_extent_block_csum_set(inode, neh);
1103 set_buffer_uptodate(bh);
1106 err = ext4_handle_dirty_metadata(handle, inode, bh);
1112 /* correct old leaf */
1114 err = ext4_ext_get_access(handle, inode, path + depth);
1117 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1118 err = ext4_ext_dirty(handle, inode, path + depth);
1124 /* create intermediate indexes */
1126 if (unlikely(k < 0)) {
1127 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1132 ext_debug("create %d intermediate indices\n", k);
1133 /* insert new index into current index block */
1134 /* current depth stored in i var */
1137 oldblock = newblock;
1138 newblock = ablocks[--a];
1139 bh = sb_getblk(inode->i_sb, newblock);
1140 if (unlikely(!bh)) {
1146 err = ext4_journal_get_create_access(handle, bh);
1150 neh = ext_block_hdr(bh);
1151 neh->eh_entries = cpu_to_le16(1);
1152 neh->eh_magic = EXT4_EXT_MAGIC;
1153 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1154 neh->eh_depth = cpu_to_le16(depth - i);
1155 fidx = EXT_FIRST_INDEX(neh);
1156 fidx->ei_block = border;
1157 ext4_idx_store_pblock(fidx, oldblock);
1159 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1160 i, newblock, le32_to_cpu(border), oldblock);
1162 /* move remainder of path[i] to the new index block */
1163 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1164 EXT_LAST_INDEX(path[i].p_hdr))) {
1165 EXT4_ERROR_INODE(inode,
1166 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1167 le32_to_cpu(path[i].p_ext->ee_block));
1171 /* start copy indexes */
1172 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1173 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1174 EXT_MAX_INDEX(path[i].p_hdr));
1175 ext4_ext_show_move(inode, path, newblock, i);
1177 memmove(++fidx, path[i].p_idx,
1178 sizeof(struct ext4_extent_idx) * m);
1179 le16_add_cpu(&neh->eh_entries, m);
1181 ext4_extent_block_csum_set(inode, neh);
1182 set_buffer_uptodate(bh);
1185 err = ext4_handle_dirty_metadata(handle, inode, bh);
1191 /* correct old index */
1193 err = ext4_ext_get_access(handle, inode, path + i);
1196 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1197 err = ext4_ext_dirty(handle, inode, path + i);
1205 /* insert new index */
1206 err = ext4_ext_insert_index(handle, inode, path + at,
1207 le32_to_cpu(border), newblock);
1211 if (buffer_locked(bh))
1217 /* free all allocated blocks in error case */
1218 for (i = 0; i < depth; i++) {
1221 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1222 EXT4_FREE_BLOCKS_METADATA);
1231 * ext4_ext_grow_indepth:
1232 * implements tree growing procedure:
1233 * - allocates new block
1234 * - moves top-level data (index block or leaf) into the new block
1235 * - initializes new top-level, creating index that points to the
1236 * just created block
1238 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1240 struct ext4_extent *newext)
1242 struct ext4_extent_header *neh;
1243 struct buffer_head *bh;
1244 ext4_fsblk_t newblock;
1247 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1248 newext, &err, flags);
1252 bh = sb_getblk(inode->i_sb, newblock);
1257 err = ext4_journal_get_create_access(handle, bh);
1263 /* move top-level index/leaf into new block */
1264 memmove(bh->b_data, EXT4_I(inode)->i_data,
1265 sizeof(EXT4_I(inode)->i_data));
1267 /* set size of new block */
1268 neh = ext_block_hdr(bh);
1269 /* old root could have indexes or leaves
1270 * so calculate e_max right way */
1271 if (ext_depth(inode))
1272 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1274 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1275 neh->eh_magic = EXT4_EXT_MAGIC;
1276 ext4_extent_block_csum_set(inode, neh);
1277 set_buffer_uptodate(bh);
1280 err = ext4_handle_dirty_metadata(handle, inode, bh);
1284 /* Update top-level index: num,max,pointer */
1285 neh = ext_inode_hdr(inode);
1286 neh->eh_entries = cpu_to_le16(1);
1287 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1288 if (neh->eh_depth == 0) {
1289 /* Root extent block becomes index block */
1290 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1291 EXT_FIRST_INDEX(neh)->ei_block =
1292 EXT_FIRST_EXTENT(neh)->ee_block;
1294 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1295 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1296 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1297 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1299 le16_add_cpu(&neh->eh_depth, 1);
1300 ext4_mark_inode_dirty(handle, inode);
1308 * ext4_ext_create_new_leaf:
1309 * finds empty index and adds new leaf.
1310 * if no free index is found, then it requests in-depth growing.
1312 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1313 unsigned int mb_flags,
1314 unsigned int gb_flags,
1315 struct ext4_ext_path *path,
1316 struct ext4_extent *newext)
1318 struct ext4_ext_path *curp;
1319 int depth, i, err = 0;
1322 i = depth = ext_depth(inode);
1324 /* walk up to the tree and look for free index entry */
1325 curp = path + depth;
1326 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1331 /* we use already allocated block for index block,
1332 * so subsequent data blocks should be contiguous */
1333 if (EXT_HAS_FREE_INDEX(curp)) {
1334 /* if we found index with free entry, then use that
1335 * entry: create all needed subtree and add new leaf */
1336 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1341 ext4_ext_drop_refs(path);
1342 path = ext4_ext_find_extent(inode,
1343 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1346 err = PTR_ERR(path);
1348 /* tree is full, time to grow in depth */
1349 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1354 ext4_ext_drop_refs(path);
1355 path = ext4_ext_find_extent(inode,
1356 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1359 err = PTR_ERR(path);
1364 * only first (depth 0 -> 1) produces free space;
1365 * in all other cases we have to split the grown tree
1367 depth = ext_depth(inode);
1368 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1369 /* now we need to split */
1379 * search the closest allocated block to the left for *logical
1380 * and returns it at @logical + it's physical address at @phys
1381 * if *logical is the smallest allocated block, the function
1382 * returns 0 at @phys
1383 * return value contains 0 (success) or error code
1385 static int ext4_ext_search_left(struct inode *inode,
1386 struct ext4_ext_path *path,
1387 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1389 struct ext4_extent_idx *ix;
1390 struct ext4_extent *ex;
1393 if (unlikely(path == NULL)) {
1394 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1397 depth = path->p_depth;
1400 if (depth == 0 && path->p_ext == NULL)
1403 /* usually extent in the path covers blocks smaller
1404 * then *logical, but it can be that extent is the
1405 * first one in the file */
1407 ex = path[depth].p_ext;
1408 ee_len = ext4_ext_get_actual_len(ex);
1409 if (*logical < le32_to_cpu(ex->ee_block)) {
1410 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1411 EXT4_ERROR_INODE(inode,
1412 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1413 *logical, le32_to_cpu(ex->ee_block));
1416 while (--depth >= 0) {
1417 ix = path[depth].p_idx;
1418 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1419 EXT4_ERROR_INODE(inode,
1420 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1421 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1422 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1423 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1431 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1432 EXT4_ERROR_INODE(inode,
1433 "logical %d < ee_block %d + ee_len %d!",
1434 *logical, le32_to_cpu(ex->ee_block), ee_len);
1438 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1439 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1444 * search the closest allocated block to the right for *logical
1445 * and returns it at @logical + it's physical address at @phys
1446 * if *logical is the largest allocated block, the function
1447 * returns 0 at @phys
1448 * return value contains 0 (success) or error code
1450 static int ext4_ext_search_right(struct inode *inode,
1451 struct ext4_ext_path *path,
1452 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1453 struct ext4_extent **ret_ex)
1455 struct buffer_head *bh = NULL;
1456 struct ext4_extent_header *eh;
1457 struct ext4_extent_idx *ix;
1458 struct ext4_extent *ex;
1460 int depth; /* Note, NOT eh_depth; depth from top of tree */
1463 if (unlikely(path == NULL)) {
1464 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1467 depth = path->p_depth;
1470 if (depth == 0 && path->p_ext == NULL)
1473 /* usually extent in the path covers blocks smaller
1474 * then *logical, but it can be that extent is the
1475 * first one in the file */
1477 ex = path[depth].p_ext;
1478 ee_len = ext4_ext_get_actual_len(ex);
1479 if (*logical < le32_to_cpu(ex->ee_block)) {
1480 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1481 EXT4_ERROR_INODE(inode,
1482 "first_extent(path[%d].p_hdr) != ex",
1486 while (--depth >= 0) {
1487 ix = path[depth].p_idx;
1488 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1489 EXT4_ERROR_INODE(inode,
1490 "ix != EXT_FIRST_INDEX *logical %d!",
1498 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1499 EXT4_ERROR_INODE(inode,
1500 "logical %d < ee_block %d + ee_len %d!",
1501 *logical, le32_to_cpu(ex->ee_block), ee_len);
1505 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1506 /* next allocated block in this leaf */
1511 /* go up and search for index to the right */
1512 while (--depth >= 0) {
1513 ix = path[depth].p_idx;
1514 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1518 /* we've gone up to the root and found no index to the right */
1522 /* we've found index to the right, let's
1523 * follow it and find the closest allocated
1524 * block to the right */
1526 block = ext4_idx_pblock(ix);
1527 while (++depth < path->p_depth) {
1528 /* subtract from p_depth to get proper eh_depth */
1529 bh = read_extent_tree_block(inode, block,
1530 path->p_depth - depth, 0);
1533 eh = ext_block_hdr(bh);
1534 ix = EXT_FIRST_INDEX(eh);
1535 block = ext4_idx_pblock(ix);
1539 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1542 eh = ext_block_hdr(bh);
1543 ex = EXT_FIRST_EXTENT(eh);
1545 *logical = le32_to_cpu(ex->ee_block);
1546 *phys = ext4_ext_pblock(ex);
1554 * ext4_ext_next_allocated_block:
1555 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1556 * NOTE: it considers block number from index entry as
1557 * allocated block. Thus, index entries have to be consistent
1561 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1565 BUG_ON(path == NULL);
1566 depth = path->p_depth;
1568 if (depth == 0 && path->p_ext == NULL)
1569 return EXT_MAX_BLOCKS;
1571 while (depth >= 0) {
1572 if (depth == path->p_depth) {
1574 if (path[depth].p_ext &&
1575 path[depth].p_ext !=
1576 EXT_LAST_EXTENT(path[depth].p_hdr))
1577 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1580 if (path[depth].p_idx !=
1581 EXT_LAST_INDEX(path[depth].p_hdr))
1582 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1587 return EXT_MAX_BLOCKS;
1591 * ext4_ext_next_leaf_block:
1592 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1594 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1598 BUG_ON(path == NULL);
1599 depth = path->p_depth;
1601 /* zero-tree has no leaf blocks at all */
1603 return EXT_MAX_BLOCKS;
1605 /* go to index block */
1608 while (depth >= 0) {
1609 if (path[depth].p_idx !=
1610 EXT_LAST_INDEX(path[depth].p_hdr))
1611 return (ext4_lblk_t)
1612 le32_to_cpu(path[depth].p_idx[1].ei_block);
1616 return EXT_MAX_BLOCKS;
1620 * ext4_ext_correct_indexes:
1621 * if leaf gets modified and modified extent is first in the leaf,
1622 * then we have to correct all indexes above.
1623 * TODO: do we need to correct tree in all cases?
1625 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1626 struct ext4_ext_path *path)
1628 struct ext4_extent_header *eh;
1629 int depth = ext_depth(inode);
1630 struct ext4_extent *ex;
1634 eh = path[depth].p_hdr;
1635 ex = path[depth].p_ext;
1637 if (unlikely(ex == NULL || eh == NULL)) {
1638 EXT4_ERROR_INODE(inode,
1639 "ex %p == NULL or eh %p == NULL", ex, eh);
1644 /* there is no tree at all */
1648 if (ex != EXT_FIRST_EXTENT(eh)) {
1649 /* we correct tree if first leaf got modified only */
1654 * TODO: we need correction if border is smaller than current one
1657 border = path[depth].p_ext->ee_block;
1658 err = ext4_ext_get_access(handle, inode, path + k);
1661 path[k].p_idx->ei_block = border;
1662 err = ext4_ext_dirty(handle, inode, path + k);
1667 /* change all left-side indexes */
1668 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1670 err = ext4_ext_get_access(handle, inode, path + k);
1673 path[k].p_idx->ei_block = border;
1674 err = ext4_ext_dirty(handle, inode, path + k);
1683 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1684 struct ext4_extent *ex2)
1686 unsigned short ext1_ee_len, ext2_ee_len;
1689 * Make sure that both extents are initialized. We don't merge
1690 * unwritten extents so that we can be sure that end_io code has
1691 * the extent that was written properly split out and conversion to
1692 * initialized is trivial.
1694 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1697 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1698 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1700 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1701 le32_to_cpu(ex2->ee_block))
1705 * To allow future support for preallocated extents to be added
1706 * as an RO_COMPAT feature, refuse to merge to extents if
1707 * this can result in the top bit of ee_len being set.
1709 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1711 if (ext4_ext_is_unwritten(ex1) &&
1712 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1713 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1714 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1716 #ifdef AGGRESSIVE_TEST
1717 if (ext1_ee_len >= 4)
1721 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1727 * This function tries to merge the "ex" extent to the next extent in the tree.
1728 * It always tries to merge towards right. If you want to merge towards
1729 * left, pass "ex - 1" as argument instead of "ex".
1730 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1731 * 1 if they got merged.
1733 static int ext4_ext_try_to_merge_right(struct inode *inode,
1734 struct ext4_ext_path *path,
1735 struct ext4_extent *ex)
1737 struct ext4_extent_header *eh;
1738 unsigned int depth, len;
1739 int merge_done = 0, unwritten;
1741 depth = ext_depth(inode);
1742 BUG_ON(path[depth].p_hdr == NULL);
1743 eh = path[depth].p_hdr;
1745 while (ex < EXT_LAST_EXTENT(eh)) {
1746 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1748 /* merge with next extent! */
1749 unwritten = ext4_ext_is_unwritten(ex);
1750 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1751 + ext4_ext_get_actual_len(ex + 1));
1753 ext4_ext_mark_unwritten(ex);
1755 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1756 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1757 * sizeof(struct ext4_extent);
1758 memmove(ex + 1, ex + 2, len);
1760 le16_add_cpu(&eh->eh_entries, -1);
1762 WARN_ON(eh->eh_entries == 0);
1763 if (!eh->eh_entries)
1764 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1771 * This function does a very simple check to see if we can collapse
1772 * an extent tree with a single extent tree leaf block into the inode.
1774 static void ext4_ext_try_to_merge_up(handle_t *handle,
1775 struct inode *inode,
1776 struct ext4_ext_path *path)
1779 unsigned max_root = ext4_ext_space_root(inode, 0);
1782 if ((path[0].p_depth != 1) ||
1783 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1784 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1788 * We need to modify the block allocation bitmap and the block
1789 * group descriptor to release the extent tree block. If we
1790 * can't get the journal credits, give up.
1792 if (ext4_journal_extend(handle, 2))
1796 * Copy the extent data up to the inode
1798 blk = ext4_idx_pblock(path[0].p_idx);
1799 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1800 sizeof(struct ext4_extent_idx);
1801 s += sizeof(struct ext4_extent_header);
1803 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1804 path[0].p_depth = 0;
1805 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1806 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1807 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1809 brelse(path[1].p_bh);
1810 ext4_free_blocks(handle, inode, NULL, blk, 1,
1811 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1815 * This function tries to merge the @ex extent to neighbours in the tree.
1816 * return 1 if merge left else 0.
1818 static void ext4_ext_try_to_merge(handle_t *handle,
1819 struct inode *inode,
1820 struct ext4_ext_path *path,
1821 struct ext4_extent *ex) {
1822 struct ext4_extent_header *eh;
1826 depth = ext_depth(inode);
1827 BUG_ON(path[depth].p_hdr == NULL);
1828 eh = path[depth].p_hdr;
1830 if (ex > EXT_FIRST_EXTENT(eh))
1831 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1834 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1836 ext4_ext_try_to_merge_up(handle, inode, path);
1840 * check if a portion of the "newext" extent overlaps with an
1843 * If there is an overlap discovered, it updates the length of the newext
1844 * such that there will be no overlap, and then returns 1.
1845 * If there is no overlap found, it returns 0.
1847 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1848 struct inode *inode,
1849 struct ext4_extent *newext,
1850 struct ext4_ext_path *path)
1853 unsigned int depth, len1;
1854 unsigned int ret = 0;
1856 b1 = le32_to_cpu(newext->ee_block);
1857 len1 = ext4_ext_get_actual_len(newext);
1858 depth = ext_depth(inode);
1859 if (!path[depth].p_ext)
1861 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1864 * get the next allocated block if the extent in the path
1865 * is before the requested block(s)
1868 b2 = ext4_ext_next_allocated_block(path);
1869 if (b2 == EXT_MAX_BLOCKS)
1871 b2 = EXT4_LBLK_CMASK(sbi, b2);
1874 /* check for wrap through zero on extent logical start block*/
1875 if (b1 + len1 < b1) {
1876 len1 = EXT_MAX_BLOCKS - b1;
1877 newext->ee_len = cpu_to_le16(len1);
1881 /* check for overlap */
1882 if (b1 + len1 > b2) {
1883 newext->ee_len = cpu_to_le16(b2 - b1);
1891 * ext4_ext_insert_extent:
1892 * tries to merge requsted extent into the existing extent or
1893 * inserts requested extent as new one into the tree,
1894 * creating new leaf in the no-space case.
1896 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1897 struct ext4_ext_path *path,
1898 struct ext4_extent *newext, int gb_flags)
1900 struct ext4_extent_header *eh;
1901 struct ext4_extent *ex, *fex;
1902 struct ext4_extent *nearex; /* nearest extent */
1903 struct ext4_ext_path *npath = NULL;
1904 int depth, len, err;
1906 int mb_flags = 0, unwritten;
1908 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1909 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1912 depth = ext_depth(inode);
1913 ex = path[depth].p_ext;
1914 eh = path[depth].p_hdr;
1915 if (unlikely(path[depth].p_hdr == NULL)) {
1916 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1920 /* try to insert block into found extent and return */
1921 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1924 * Try to see whether we should rather test the extent on
1925 * right from ex, or from the left of ex. This is because
1926 * ext4_ext_find_extent() can return either extent on the
1927 * left, or on the right from the searched position. This
1928 * will make merging more effective.
1930 if (ex < EXT_LAST_EXTENT(eh) &&
1931 (le32_to_cpu(ex->ee_block) +
1932 ext4_ext_get_actual_len(ex) <
1933 le32_to_cpu(newext->ee_block))) {
1936 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1937 (le32_to_cpu(newext->ee_block) +
1938 ext4_ext_get_actual_len(newext) <
1939 le32_to_cpu(ex->ee_block)))
1942 /* Try to append newex to the ex */
1943 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1944 ext_debug("append [%d]%d block to %u:[%d]%d"
1946 ext4_ext_is_unwritten(newext),
1947 ext4_ext_get_actual_len(newext),
1948 le32_to_cpu(ex->ee_block),
1949 ext4_ext_is_unwritten(ex),
1950 ext4_ext_get_actual_len(ex),
1951 ext4_ext_pblock(ex));
1952 err = ext4_ext_get_access(handle, inode,
1956 unwritten = ext4_ext_is_unwritten(ex);
1957 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1958 + ext4_ext_get_actual_len(newext));
1960 ext4_ext_mark_unwritten(ex);
1961 eh = path[depth].p_hdr;
1967 /* Try to prepend newex to the ex */
1968 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1969 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1971 le32_to_cpu(newext->ee_block),
1972 ext4_ext_is_unwritten(newext),
1973 ext4_ext_get_actual_len(newext),
1974 le32_to_cpu(ex->ee_block),
1975 ext4_ext_is_unwritten(ex),
1976 ext4_ext_get_actual_len(ex),
1977 ext4_ext_pblock(ex));
1978 err = ext4_ext_get_access(handle, inode,
1983 unwritten = ext4_ext_is_unwritten(ex);
1984 ex->ee_block = newext->ee_block;
1985 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1986 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1987 + ext4_ext_get_actual_len(newext));
1989 ext4_ext_mark_unwritten(ex);
1990 eh = path[depth].p_hdr;
1996 depth = ext_depth(inode);
1997 eh = path[depth].p_hdr;
1998 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2001 /* probably next leaf has space for us? */
2002 fex = EXT_LAST_EXTENT(eh);
2003 next = EXT_MAX_BLOCKS;
2004 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2005 next = ext4_ext_next_leaf_block(path);
2006 if (next != EXT_MAX_BLOCKS) {
2007 ext_debug("next leaf block - %u\n", next);
2008 BUG_ON(npath != NULL);
2009 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2011 return PTR_ERR(npath);
2012 BUG_ON(npath->p_depth != path->p_depth);
2013 eh = npath[depth].p_hdr;
2014 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2015 ext_debug("next leaf isn't full(%d)\n",
2016 le16_to_cpu(eh->eh_entries));
2020 ext_debug("next leaf has no free space(%d,%d)\n",
2021 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2025 * There is no free space in the found leaf.
2026 * We're gonna add a new leaf in the tree.
2028 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2029 mb_flags = EXT4_MB_USE_RESERVED;
2030 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2034 depth = ext_depth(inode);
2035 eh = path[depth].p_hdr;
2038 nearex = path[depth].p_ext;
2040 err = ext4_ext_get_access(handle, inode, path + depth);
2045 /* there is no extent in this leaf, create first one */
2046 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2047 le32_to_cpu(newext->ee_block),
2048 ext4_ext_pblock(newext),
2049 ext4_ext_is_unwritten(newext),
2050 ext4_ext_get_actual_len(newext));
2051 nearex = EXT_FIRST_EXTENT(eh);
2053 if (le32_to_cpu(newext->ee_block)
2054 > le32_to_cpu(nearex->ee_block)) {
2056 ext_debug("insert %u:%llu:[%d]%d before: "
2058 le32_to_cpu(newext->ee_block),
2059 ext4_ext_pblock(newext),
2060 ext4_ext_is_unwritten(newext),
2061 ext4_ext_get_actual_len(newext),
2066 BUG_ON(newext->ee_block == nearex->ee_block);
2067 ext_debug("insert %u:%llu:[%d]%d after: "
2069 le32_to_cpu(newext->ee_block),
2070 ext4_ext_pblock(newext),
2071 ext4_ext_is_unwritten(newext),
2072 ext4_ext_get_actual_len(newext),
2075 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2077 ext_debug("insert %u:%llu:[%d]%d: "
2078 "move %d extents from 0x%p to 0x%p\n",
2079 le32_to_cpu(newext->ee_block),
2080 ext4_ext_pblock(newext),
2081 ext4_ext_is_unwritten(newext),
2082 ext4_ext_get_actual_len(newext),
2083 len, nearex, nearex + 1);
2084 memmove(nearex + 1, nearex,
2085 len * sizeof(struct ext4_extent));
2089 le16_add_cpu(&eh->eh_entries, 1);
2090 path[depth].p_ext = nearex;
2091 nearex->ee_block = newext->ee_block;
2092 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2093 nearex->ee_len = newext->ee_len;
2096 /* try to merge extents */
2097 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2098 ext4_ext_try_to_merge(handle, inode, path, nearex);
2101 /* time to correct all indexes above */
2102 err = ext4_ext_correct_indexes(handle, inode, path);
2106 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2110 ext4_ext_drop_refs(npath);
2116 static int ext4_fill_fiemap_extents(struct inode *inode,
2117 ext4_lblk_t block, ext4_lblk_t num,
2118 struct fiemap_extent_info *fieinfo)
2120 struct ext4_ext_path *path = NULL;
2121 struct ext4_extent *ex;
2122 struct extent_status es;
2123 ext4_lblk_t next, next_del, start = 0, end = 0;
2124 ext4_lblk_t last = block + num;
2125 int exists, depth = 0, err = 0;
2126 unsigned int flags = 0;
2127 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2129 while (block < last && block != EXT_MAX_BLOCKS) {
2131 /* find extent for this block */
2132 down_read(&EXT4_I(inode)->i_data_sem);
2134 if (path && ext_depth(inode) != depth) {
2135 /* depth was changed. we have to realloc path */
2140 path = ext4_ext_find_extent(inode, block, path, 0);
2142 up_read(&EXT4_I(inode)->i_data_sem);
2143 err = PTR_ERR(path);
2148 depth = ext_depth(inode);
2149 if (unlikely(path[depth].p_hdr == NULL)) {
2150 up_read(&EXT4_I(inode)->i_data_sem);
2151 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2155 ex = path[depth].p_ext;
2156 next = ext4_ext_next_allocated_block(path);
2157 ext4_ext_drop_refs(path);
2162 /* there is no extent yet, so try to allocate
2163 * all requested space */
2166 } else if (le32_to_cpu(ex->ee_block) > block) {
2167 /* need to allocate space before found extent */
2169 end = le32_to_cpu(ex->ee_block);
2170 if (block + num < end)
2172 } else if (block >= le32_to_cpu(ex->ee_block)
2173 + ext4_ext_get_actual_len(ex)) {
2174 /* need to allocate space after found extent */
2179 } else if (block >= le32_to_cpu(ex->ee_block)) {
2181 * some part of requested space is covered
2185 end = le32_to_cpu(ex->ee_block)
2186 + ext4_ext_get_actual_len(ex);
2187 if (block + num < end)
2193 BUG_ON(end <= start);
2197 es.es_len = end - start;
2200 es.es_lblk = le32_to_cpu(ex->ee_block);
2201 es.es_len = ext4_ext_get_actual_len(ex);
2202 es.es_pblk = ext4_ext_pblock(ex);
2203 if (ext4_ext_is_unwritten(ex))
2204 flags |= FIEMAP_EXTENT_UNWRITTEN;
2208 * Find delayed extent and update es accordingly. We call
2209 * it even in !exists case to find out whether es is the
2210 * last existing extent or not.
2212 next_del = ext4_find_delayed_extent(inode, &es);
2213 if (!exists && next_del) {
2215 flags |= (FIEMAP_EXTENT_DELALLOC |
2216 FIEMAP_EXTENT_UNKNOWN);
2218 up_read(&EXT4_I(inode)->i_data_sem);
2220 if (unlikely(es.es_len == 0)) {
2221 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2227 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2228 * we need to check next == EXT_MAX_BLOCKS because it is
2229 * possible that an extent is with unwritten and delayed
2230 * status due to when an extent is delayed allocated and
2231 * is allocated by fallocate status tree will track both of
2234 * So we could return a unwritten and delayed extent, and
2235 * its block is equal to 'next'.
2237 if (next == next_del && next == EXT_MAX_BLOCKS) {
2238 flags |= FIEMAP_EXTENT_LAST;
2239 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2240 next != EXT_MAX_BLOCKS)) {
2241 EXT4_ERROR_INODE(inode,
2242 "next extent == %u, next "
2243 "delalloc extent = %u",
2251 err = fiemap_fill_next_extent(fieinfo,
2252 (__u64)es.es_lblk << blksize_bits,
2253 (__u64)es.es_pblk << blksize_bits,
2254 (__u64)es.es_len << blksize_bits,
2264 block = es.es_lblk + es.es_len;
2268 ext4_ext_drop_refs(path);
2276 * ext4_ext_put_gap_in_cache:
2277 * calculate boundaries of the gap that the requested block fits into
2278 * and cache this gap
2281 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2284 int depth = ext_depth(inode);
2285 unsigned long len = 0;
2286 ext4_lblk_t lblock = 0;
2287 struct ext4_extent *ex;
2289 ex = path[depth].p_ext;
2292 * there is no extent yet, so gap is [0;-] and we
2295 ext_debug("cache gap(whole file):");
2296 } else if (block < le32_to_cpu(ex->ee_block)) {
2298 len = le32_to_cpu(ex->ee_block) - block;
2299 ext_debug("cache gap(before): %u [%u:%u]",
2301 le32_to_cpu(ex->ee_block),
2302 ext4_ext_get_actual_len(ex));
2303 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2304 ext4_es_insert_extent(inode, lblock, len, ~0,
2305 EXTENT_STATUS_HOLE);
2306 } else if (block >= le32_to_cpu(ex->ee_block)
2307 + ext4_ext_get_actual_len(ex)) {
2309 lblock = le32_to_cpu(ex->ee_block)
2310 + ext4_ext_get_actual_len(ex);
2312 next = ext4_ext_next_allocated_block(path);
2313 ext_debug("cache gap(after): [%u:%u] %u",
2314 le32_to_cpu(ex->ee_block),
2315 ext4_ext_get_actual_len(ex),
2317 BUG_ON(next == lblock);
2318 len = next - lblock;
2319 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2320 ext4_es_insert_extent(inode, lblock, len, ~0,
2321 EXTENT_STATUS_HOLE);
2326 ext_debug(" -> %u:%lu\n", lblock, len);
2331 * removes index from the index block.
2333 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2334 struct ext4_ext_path *path, int depth)
2339 /* free index block */
2341 path = path + depth;
2342 leaf = ext4_idx_pblock(path->p_idx);
2343 if (unlikely(path->p_hdr->eh_entries == 0)) {
2344 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2347 err = ext4_ext_get_access(handle, inode, path);
2351 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2352 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2353 len *= sizeof(struct ext4_extent_idx);
2354 memmove(path->p_idx, path->p_idx + 1, len);
2357 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2358 err = ext4_ext_dirty(handle, inode, path);
2361 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2362 trace_ext4_ext_rm_idx(inode, leaf);
2364 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2365 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2367 while (--depth >= 0) {
2368 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2371 err = ext4_ext_get_access(handle, inode, path);
2374 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2375 err = ext4_ext_dirty(handle, inode, path);
2383 * ext4_ext_calc_credits_for_single_extent:
2384 * This routine returns max. credits that needed to insert an extent
2385 * to the extent tree.
2386 * When pass the actual path, the caller should calculate credits
2389 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2390 struct ext4_ext_path *path)
2393 int depth = ext_depth(inode);
2396 /* probably there is space in leaf? */
2397 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2398 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2401 * There are some space in the leaf tree, no
2402 * need to account for leaf block credit
2404 * bitmaps and block group descriptor blocks
2405 * and other metadata blocks still need to be
2408 /* 1 bitmap, 1 block group descriptor */
2409 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2414 return ext4_chunk_trans_blocks(inode, nrblocks);
2418 * How many index/leaf blocks need to change/allocate to add @extents extents?
2420 * If we add a single extent, then in the worse case, each tree level
2421 * index/leaf need to be changed in case of the tree split.
2423 * If more extents are inserted, they could cause the whole tree split more
2424 * than once, but this is really rare.
2426 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2431 /* If we are converting the inline data, only one is needed here. */
2432 if (ext4_has_inline_data(inode))
2435 depth = ext_depth(inode);
2445 static inline int get_default_free_blocks_flags(struct inode *inode)
2447 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2448 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2449 else if (ext4_should_journal_data(inode))
2450 return EXT4_FREE_BLOCKS_FORGET;
2454 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2455 struct ext4_extent *ex,
2456 long long *partial_cluster,
2457 ext4_lblk_t from, ext4_lblk_t to)
2459 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2460 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2462 int flags = get_default_free_blocks_flags(inode);
2465 * For bigalloc file systems, we never free a partial cluster
2466 * at the beginning of the extent. Instead, we make a note
2467 * that we tried freeing the cluster, and check to see if we
2468 * need to free it on a subsequent call to ext4_remove_blocks,
2469 * or at the end of the ext4_truncate() operation.
2471 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2473 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2475 * If we have a partial cluster, and it's different from the
2476 * cluster of the last block, we need to explicitly free the
2477 * partial cluster here.
2479 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2480 if ((*partial_cluster > 0) &&
2481 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2482 ext4_free_blocks(handle, inode, NULL,
2483 EXT4_C2B(sbi, *partial_cluster),
2484 sbi->s_cluster_ratio, flags);
2485 *partial_cluster = 0;
2488 #ifdef EXTENTS_STATS
2490 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2491 spin_lock(&sbi->s_ext_stats_lock);
2492 sbi->s_ext_blocks += ee_len;
2493 sbi->s_ext_extents++;
2494 if (ee_len < sbi->s_ext_min)
2495 sbi->s_ext_min = ee_len;
2496 if (ee_len > sbi->s_ext_max)
2497 sbi->s_ext_max = ee_len;
2498 if (ext_depth(inode) > sbi->s_depth_max)
2499 sbi->s_depth_max = ext_depth(inode);
2500 spin_unlock(&sbi->s_ext_stats_lock);
2503 if (from >= le32_to_cpu(ex->ee_block)
2504 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2507 unsigned int unaligned;
2509 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2510 pblk = ext4_ext_pblock(ex) + ee_len - num;
2512 * Usually we want to free partial cluster at the end of the
2513 * extent, except for the situation when the cluster is still
2514 * used by any other extent (partial_cluster is negative).
2516 if (*partial_cluster < 0 &&
2517 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2518 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2520 ext_debug("free last %u blocks starting %llu partial %lld\n",
2521 num, pblk, *partial_cluster);
2522 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2524 * If the block range to be freed didn't start at the
2525 * beginning of a cluster, and we removed the entire
2526 * extent and the cluster is not used by any other extent,
2527 * save the partial cluster here, since we might need to
2528 * delete if we determine that the truncate operation has
2529 * removed all of the blocks in the cluster.
2531 * On the other hand, if we did not manage to free the whole
2532 * extent, we have to mark the cluster as used (store negative
2533 * cluster number in partial_cluster).
2535 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2536 if (unaligned && (ee_len == num) &&
2537 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2538 *partial_cluster = EXT4_B2C(sbi, pblk);
2540 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2541 else if (*partial_cluster > 0)
2542 *partial_cluster = 0;
2544 ext4_error(sbi->s_sb, "strange request: removal(2) "
2545 "%u-%u from %u:%u\n",
2546 from, to, le32_to_cpu(ex->ee_block), ee_len);
2552 * ext4_ext_rm_leaf() Removes the extents associated with the
2553 * blocks appearing between "start" and "end", and splits the extents
2554 * if "start" and "end" appear in the same extent
2556 * @handle: The journal handle
2557 * @inode: The files inode
2558 * @path: The path to the leaf
2559 * @partial_cluster: The cluster which we'll have to free if all extents
2560 * has been released from it. It gets negative in case
2561 * that the cluster is still used.
2562 * @start: The first block to remove
2563 * @end: The last block to remove
2566 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2567 struct ext4_ext_path *path,
2568 long long *partial_cluster,
2569 ext4_lblk_t start, ext4_lblk_t end)
2571 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2572 int err = 0, correct_index = 0;
2573 int depth = ext_depth(inode), credits;
2574 struct ext4_extent_header *eh;
2577 ext4_lblk_t ex_ee_block;
2578 unsigned short ex_ee_len;
2579 unsigned unwritten = 0;
2580 struct ext4_extent *ex;
2583 /* the header must be checked already in ext4_ext_remove_space() */
2584 ext_debug("truncate since %u in leaf to %u\n", start, end);
2585 if (!path[depth].p_hdr)
2586 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2587 eh = path[depth].p_hdr;
2588 if (unlikely(path[depth].p_hdr == NULL)) {
2589 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2592 /* find where to start removing */
2593 ex = path[depth].p_ext;
2595 ex = EXT_LAST_EXTENT(eh);
2597 ex_ee_block = le32_to_cpu(ex->ee_block);
2598 ex_ee_len = ext4_ext_get_actual_len(ex);
2601 * If we're starting with an extent other than the last one in the
2602 * node, we need to see if it shares a cluster with the extent to
2603 * the right (towards the end of the file). If its leftmost cluster
2604 * is this extent's rightmost cluster and it is not cluster aligned,
2605 * we'll mark it as a partial that is not to be deallocated.
2608 if (ex != EXT_LAST_EXTENT(eh)) {
2609 ext4_fsblk_t current_pblk, right_pblk;
2610 long long current_cluster, right_cluster;
2612 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2613 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2614 right_pblk = ext4_ext_pblock(ex + 1);
2615 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2616 if (current_cluster == right_cluster &&
2617 EXT4_PBLK_COFF(sbi, right_pblk))
2618 *partial_cluster = -right_cluster;
2621 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2623 while (ex >= EXT_FIRST_EXTENT(eh) &&
2624 ex_ee_block + ex_ee_len > start) {
2626 if (ext4_ext_is_unwritten(ex))
2631 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2632 unwritten, ex_ee_len);
2633 path[depth].p_ext = ex;
2635 a = ex_ee_block > start ? ex_ee_block : start;
2636 b = ex_ee_block+ex_ee_len - 1 < end ?
2637 ex_ee_block+ex_ee_len - 1 : end;
2639 ext_debug(" border %u:%u\n", a, b);
2641 /* If this extent is beyond the end of the hole, skip it */
2642 if (end < ex_ee_block) {
2644 * We're going to skip this extent and move to another,
2645 * so if this extent is not cluster aligned we have
2646 * to mark the current cluster as used to avoid
2647 * accidentally freeing it later on
2649 pblk = ext4_ext_pblock(ex);
2650 if (EXT4_PBLK_COFF(sbi, pblk))
2652 -((long long)EXT4_B2C(sbi, pblk));
2654 ex_ee_block = le32_to_cpu(ex->ee_block);
2655 ex_ee_len = ext4_ext_get_actual_len(ex);
2657 } else if (b != ex_ee_block + ex_ee_len - 1) {
2658 EXT4_ERROR_INODE(inode,
2659 "can not handle truncate %u:%u "
2661 start, end, ex_ee_block,
2662 ex_ee_block + ex_ee_len - 1);
2665 } else if (a != ex_ee_block) {
2666 /* remove tail of the extent */
2667 num = a - ex_ee_block;
2669 /* remove whole extent: excellent! */
2673 * 3 for leaf, sb, and inode plus 2 (bmap and group
2674 * descriptor) for each block group; assume two block
2675 * groups plus ex_ee_len/blocks_per_block_group for
2678 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2679 if (ex == EXT_FIRST_EXTENT(eh)) {
2681 credits += (ext_depth(inode)) + 1;
2683 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2685 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2689 err = ext4_ext_get_access(handle, inode, path + depth);
2693 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2699 /* this extent is removed; mark slot entirely unused */
2700 ext4_ext_store_pblock(ex, 0);
2702 ex->ee_len = cpu_to_le16(num);
2704 * Do not mark unwritten if all the blocks in the
2705 * extent have been removed.
2707 if (unwritten && num)
2708 ext4_ext_mark_unwritten(ex);
2710 * If the extent was completely released,
2711 * we need to remove it from the leaf
2714 if (end != EXT_MAX_BLOCKS - 1) {
2716 * For hole punching, we need to scoot all the
2717 * extents up when an extent is removed so that
2718 * we dont have blank extents in the middle
2720 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2721 sizeof(struct ext4_extent));
2723 /* Now get rid of the one at the end */
2724 memset(EXT_LAST_EXTENT(eh), 0,
2725 sizeof(struct ext4_extent));
2727 le16_add_cpu(&eh->eh_entries, -1);
2728 } else if (*partial_cluster > 0)
2729 *partial_cluster = 0;
2731 err = ext4_ext_dirty(handle, inode, path + depth);
2735 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2736 ext4_ext_pblock(ex));
2738 ex_ee_block = le32_to_cpu(ex->ee_block);
2739 ex_ee_len = ext4_ext_get_actual_len(ex);
2742 if (correct_index && eh->eh_entries)
2743 err = ext4_ext_correct_indexes(handle, inode, path);
2746 * If there's a partial cluster and at least one extent remains in
2747 * the leaf, free the partial cluster if it isn't shared with the
2748 * current extent. If there's a partial cluster and no extents
2749 * remain in the leaf, it can't be freed here. It can only be
2750 * freed when it's possible to determine if it's not shared with
2751 * any other extent - when the next leaf is processed or when space
2752 * removal is complete.
2754 if (*partial_cluster > 0 && eh->eh_entries &&
2755 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2756 *partial_cluster)) {
2757 int flags = get_default_free_blocks_flags(inode);
2759 ext4_free_blocks(handle, inode, NULL,
2760 EXT4_C2B(sbi, *partial_cluster),
2761 sbi->s_cluster_ratio, flags);
2762 *partial_cluster = 0;
2765 /* if this leaf is free, then we should
2766 * remove it from index block above */
2767 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2768 err = ext4_ext_rm_idx(handle, inode, path, depth);
2775 * ext4_ext_more_to_rm:
2776 * returns 1 if current index has to be freed (even partial)
2779 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2781 BUG_ON(path->p_idx == NULL);
2783 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2787 * if truncate on deeper level happened, it wasn't partial,
2788 * so we have to consider current index for truncation
2790 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2795 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2798 struct super_block *sb = inode->i_sb;
2799 int depth = ext_depth(inode);
2800 struct ext4_ext_path *path = NULL;
2801 long long partial_cluster = 0;
2805 ext_debug("truncate since %u to %u\n", start, end);
2807 /* probably first extent we're gonna free will be last in block */
2808 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2810 return PTR_ERR(handle);
2813 trace_ext4_ext_remove_space(inode, start, end, depth);
2816 * Check if we are removing extents inside the extent tree. If that
2817 * is the case, we are going to punch a hole inside the extent tree
2818 * so we have to check whether we need to split the extent covering
2819 * the last block to remove so we can easily remove the part of it
2820 * in ext4_ext_rm_leaf().
2822 if (end < EXT_MAX_BLOCKS - 1) {
2823 struct ext4_extent *ex;
2824 ext4_lblk_t ee_block;
2826 /* find extent for this block */
2827 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2829 ext4_journal_stop(handle);
2830 return PTR_ERR(path);
2832 depth = ext_depth(inode);
2833 /* Leaf not may not exist only if inode has no blocks at all */
2834 ex = path[depth].p_ext;
2837 EXT4_ERROR_INODE(inode,
2838 "path[%d].p_hdr == NULL",
2845 ee_block = le32_to_cpu(ex->ee_block);
2848 * See if the last block is inside the extent, if so split
2849 * the extent at 'end' block so we can easily remove the
2850 * tail of the first part of the split extent in
2851 * ext4_ext_rm_leaf().
2853 if (end >= ee_block &&
2854 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2857 if (ext4_ext_is_unwritten(ex))
2858 split_flag = EXT4_EXT_MARK_UNWRIT1 |
2859 EXT4_EXT_MARK_UNWRIT2;
2862 * Split the extent in two so that 'end' is the last
2863 * block in the first new extent. Also we should not
2864 * fail removing space due to ENOSPC so try to use
2865 * reserved block if that happens.
2867 err = ext4_split_extent_at(handle, inode, path,
2868 end + 1, split_flag,
2870 EXT4_GET_BLOCKS_PRE_IO |
2871 EXT4_GET_BLOCKS_METADATA_NOFAIL);
2878 * We start scanning from right side, freeing all the blocks
2879 * after i_size and walking into the tree depth-wise.
2881 depth = ext_depth(inode);
2886 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2888 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2891 ext4_journal_stop(handle);
2894 path[0].p_depth = depth;
2895 path[0].p_hdr = ext_inode_hdr(inode);
2898 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2905 while (i >= 0 && err == 0) {
2907 /* this is leaf block */
2908 err = ext4_ext_rm_leaf(handle, inode, path,
2909 &partial_cluster, start,
2911 /* root level has p_bh == NULL, brelse() eats this */
2912 brelse(path[i].p_bh);
2913 path[i].p_bh = NULL;
2918 /* this is index block */
2919 if (!path[i].p_hdr) {
2920 ext_debug("initialize header\n");
2921 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2924 if (!path[i].p_idx) {
2925 /* this level hasn't been touched yet */
2926 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2927 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2928 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2930 le16_to_cpu(path[i].p_hdr->eh_entries));
2932 /* we were already here, see at next index */
2936 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2937 i, EXT_FIRST_INDEX(path[i].p_hdr),
2939 if (ext4_ext_more_to_rm(path + i)) {
2940 struct buffer_head *bh;
2941 /* go to the next level */
2942 ext_debug("move to level %d (block %llu)\n",
2943 i + 1, ext4_idx_pblock(path[i].p_idx));
2944 memset(path + i + 1, 0, sizeof(*path));
2945 bh = read_extent_tree_block(inode,
2946 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2949 /* should we reset i_size? */
2953 /* Yield here to deal with large extent trees.
2954 * Should be a no-op if we did IO above. */
2956 if (WARN_ON(i + 1 > depth)) {
2960 path[i + 1].p_bh = bh;
2962 /* save actual number of indexes since this
2963 * number is changed at the next iteration */
2964 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2967 /* we finished processing this index, go up */
2968 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2969 /* index is empty, remove it;
2970 * handle must be already prepared by the
2971 * truncatei_leaf() */
2972 err = ext4_ext_rm_idx(handle, inode, path, i);
2974 /* root level has p_bh == NULL, brelse() eats this */
2975 brelse(path[i].p_bh);
2976 path[i].p_bh = NULL;
2978 ext_debug("return to level %d\n", i);
2982 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2983 partial_cluster, path->p_hdr->eh_entries);
2985 /* If we still have something in the partial cluster and we have removed
2986 * even the first extent, then we should free the blocks in the partial
2987 * cluster as well. */
2988 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2989 int flags = get_default_free_blocks_flags(inode);
2991 ext4_free_blocks(handle, inode, NULL,
2992 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2993 EXT4_SB(sb)->s_cluster_ratio, flags);
2994 partial_cluster = 0;
2997 /* TODO: flexible tree reduction should be here */
2998 if (path->p_hdr->eh_entries == 0) {
3000 * truncate to zero freed all the tree,
3001 * so we need to correct eh_depth
3003 err = ext4_ext_get_access(handle, inode, path);
3005 ext_inode_hdr(inode)->eh_depth = 0;
3006 ext_inode_hdr(inode)->eh_max =
3007 cpu_to_le16(ext4_ext_space_root(inode, 0));
3008 err = ext4_ext_dirty(handle, inode, path);
3012 ext4_ext_drop_refs(path);
3014 if (err == -EAGAIN) {
3018 ext4_journal_stop(handle);
3024 * called at mount time
3026 void ext4_ext_init(struct super_block *sb)
3029 * possible initialization would be here
3032 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3033 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3034 printk(KERN_INFO "EXT4-fs: file extents enabled"
3035 #ifdef AGGRESSIVE_TEST
3036 ", aggressive tests"
3038 #ifdef CHECK_BINSEARCH
3041 #ifdef EXTENTS_STATS
3046 #ifdef EXTENTS_STATS
3047 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3048 EXT4_SB(sb)->s_ext_min = 1 << 30;
3049 EXT4_SB(sb)->s_ext_max = 0;
3055 * called at umount time
3057 void ext4_ext_release(struct super_block *sb)
3059 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3062 #ifdef EXTENTS_STATS
3063 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3064 struct ext4_sb_info *sbi = EXT4_SB(sb);
3065 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3066 sbi->s_ext_blocks, sbi->s_ext_extents,
3067 sbi->s_ext_blocks / sbi->s_ext_extents);
3068 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3069 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3074 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3076 ext4_lblk_t ee_block;
3077 ext4_fsblk_t ee_pblock;
3078 unsigned int ee_len;
3080 ee_block = le32_to_cpu(ex->ee_block);
3081 ee_len = ext4_ext_get_actual_len(ex);
3082 ee_pblock = ext4_ext_pblock(ex);
3087 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3088 EXTENT_STATUS_WRITTEN);
3091 /* FIXME!! we need to try to merge to left or right after zero-out */
3092 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3094 ext4_fsblk_t ee_pblock;
3095 unsigned int ee_len;
3098 ee_len = ext4_ext_get_actual_len(ex);
3099 ee_pblock = ext4_ext_pblock(ex);
3101 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3109 * ext4_split_extent_at() splits an extent at given block.
3111 * @handle: the journal handle
3112 * @inode: the file inode
3113 * @path: the path to the extent
3114 * @split: the logical block where the extent is splitted.
3115 * @split_flags: indicates if the extent could be zeroout if split fails, and
3116 * the states(init or unwritten) of new extents.
3117 * @flags: flags used to insert new extent to extent tree.
3120 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3121 * of which are deterimined by split_flag.
3123 * There are two cases:
3124 * a> the extent are splitted into two extent.
3125 * b> split is not needed, and just mark the extent.
3127 * return 0 on success.
3129 static int ext4_split_extent_at(handle_t *handle,
3130 struct inode *inode,
3131 struct ext4_ext_path *path,
3136 ext4_fsblk_t newblock;
3137 ext4_lblk_t ee_block;
3138 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3139 struct ext4_extent *ex2 = NULL;
3140 unsigned int ee_len, depth;
3143 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3144 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3146 ext_debug("ext4_split_extents_at: inode %lu, logical"
3147 "block %llu\n", inode->i_ino, (unsigned long long)split);
3149 ext4_ext_show_leaf(inode, path);
3151 depth = ext_depth(inode);
3152 ex = path[depth].p_ext;
3153 ee_block = le32_to_cpu(ex->ee_block);
3154 ee_len = ext4_ext_get_actual_len(ex);
3155 newblock = split - ee_block + ext4_ext_pblock(ex);
3157 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3158 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3159 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3160 EXT4_EXT_MARK_UNWRIT1 |
3161 EXT4_EXT_MARK_UNWRIT2));
3163 err = ext4_ext_get_access(handle, inode, path + depth);
3167 if (split == ee_block) {
3169 * case b: block @split is the block that the extent begins with
3170 * then we just change the state of the extent, and splitting
3173 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3174 ext4_ext_mark_unwritten(ex);
3176 ext4_ext_mark_initialized(ex);
3178 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3179 ext4_ext_try_to_merge(handle, inode, path, ex);
3181 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3186 memcpy(&orig_ex, ex, sizeof(orig_ex));
3187 ex->ee_len = cpu_to_le16(split - ee_block);
3188 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3189 ext4_ext_mark_unwritten(ex);
3192 * path may lead to new leaf, not to original leaf any more
3193 * after ext4_ext_insert_extent() returns,
3195 err = ext4_ext_dirty(handle, inode, path + depth);
3197 goto fix_extent_len;
3200 ex2->ee_block = cpu_to_le32(split);
3201 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3202 ext4_ext_store_pblock(ex2, newblock);
3203 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3204 ext4_ext_mark_unwritten(ex2);
3206 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3207 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3208 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3209 if (split_flag & EXT4_EXT_DATA_VALID1) {
3210 err = ext4_ext_zeroout(inode, ex2);
3211 zero_ex.ee_block = ex2->ee_block;
3212 zero_ex.ee_len = cpu_to_le16(
3213 ext4_ext_get_actual_len(ex2));
3214 ext4_ext_store_pblock(&zero_ex,
3215 ext4_ext_pblock(ex2));
3217 err = ext4_ext_zeroout(inode, ex);
3218 zero_ex.ee_block = ex->ee_block;
3219 zero_ex.ee_len = cpu_to_le16(
3220 ext4_ext_get_actual_len(ex));
3221 ext4_ext_store_pblock(&zero_ex,
3222 ext4_ext_pblock(ex));
3225 err = ext4_ext_zeroout(inode, &orig_ex);
3226 zero_ex.ee_block = orig_ex.ee_block;
3227 zero_ex.ee_len = cpu_to_le16(
3228 ext4_ext_get_actual_len(&orig_ex));
3229 ext4_ext_store_pblock(&zero_ex,
3230 ext4_ext_pblock(&orig_ex));
3234 goto fix_extent_len;
3235 /* update the extent length and mark as initialized */
3236 ex->ee_len = cpu_to_le16(ee_len);
3237 ext4_ext_try_to_merge(handle, inode, path, ex);
3238 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3240 goto fix_extent_len;
3242 /* update extent status tree */
3243 err = ext4_zeroout_es(inode, &zero_ex);
3247 goto fix_extent_len;
3250 ext4_ext_show_leaf(inode, path);
3254 ex->ee_len = orig_ex.ee_len;
3255 ext4_ext_dirty(handle, inode, path + depth);
3260 * ext4_split_extents() splits an extent and mark extent which is covered
3261 * by @map as split_flags indicates
3263 * It may result in splitting the extent into multiple extents (up to three)
3264 * There are three possibilities:
3265 * a> There is no split required
3266 * b> Splits in two extents: Split is happening at either end of the extent
3267 * c> Splits in three extents: Somone is splitting in middle of the extent
3270 static int ext4_split_extent(handle_t *handle,
3271 struct inode *inode,
3272 struct ext4_ext_path *path,
3273 struct ext4_map_blocks *map,
3277 ext4_lblk_t ee_block;
3278 struct ext4_extent *ex;
3279 unsigned int ee_len, depth;
3282 int split_flag1, flags1;
3283 int allocated = map->m_len;
3285 depth = ext_depth(inode);
3286 ex = path[depth].p_ext;
3287 ee_block = le32_to_cpu(ex->ee_block);
3288 ee_len = ext4_ext_get_actual_len(ex);
3289 unwritten = ext4_ext_is_unwritten(ex);
3291 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3292 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3293 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3295 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3296 EXT4_EXT_MARK_UNWRIT2;
3297 if (split_flag & EXT4_EXT_DATA_VALID2)
3298 split_flag1 |= EXT4_EXT_DATA_VALID1;
3299 err = ext4_split_extent_at(handle, inode, path,
3300 map->m_lblk + map->m_len, split_flag1, flags1);
3304 allocated = ee_len - (map->m_lblk - ee_block);
3307 * Update path is required because previous ext4_split_extent_at() may
3308 * result in split of original leaf or extent zeroout.
3310 ext4_ext_drop_refs(path);
3311 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3313 return PTR_ERR(path);
3314 depth = ext_depth(inode);
3315 ex = path[depth].p_ext;
3317 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3318 (unsigned long) map->m_lblk);
3321 unwritten = ext4_ext_is_unwritten(ex);
3324 if (map->m_lblk >= ee_block) {
3325 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3327 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3328 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3329 EXT4_EXT_MARK_UNWRIT2);
3331 err = ext4_split_extent_at(handle, inode, path,
3332 map->m_lblk, split_flag1, flags);
3337 ext4_ext_show_leaf(inode, path);
3339 return err ? err : allocated;
3343 * This function is called by ext4_ext_map_blocks() if someone tries to write
3344 * to an unwritten extent. It may result in splitting the unwritten
3345 * extent into multiple extents (up to three - one initialized and two
3347 * There are three possibilities:
3348 * a> There is no split required: Entire extent should be initialized
3349 * b> Splits in two extents: Write is happening at either end of the extent
3350 * c> Splits in three extents: Somone is writing in middle of the extent
3353 * - The extent pointed to by 'path' is unwritten.
3354 * - The extent pointed to by 'path' contains a superset
3355 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3357 * Post-conditions on success:
3358 * - the returned value is the number of blocks beyond map->l_lblk
3359 * that are allocated and initialized.
3360 * It is guaranteed to be >= map->m_len.
3362 static int ext4_ext_convert_to_initialized(handle_t *handle,
3363 struct inode *inode,
3364 struct ext4_map_blocks *map,
3365 struct ext4_ext_path *path,
3368 struct ext4_sb_info *sbi;
3369 struct ext4_extent_header *eh;
3370 struct ext4_map_blocks split_map;
3371 struct ext4_extent zero_ex;
3372 struct ext4_extent *ex, *abut_ex;
3373 ext4_lblk_t ee_block, eof_block;
3374 unsigned int ee_len, depth, map_len = map->m_len;
3375 int allocated = 0, max_zeroout = 0;
3379 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3380 "block %llu, max_blocks %u\n", inode->i_ino,
3381 (unsigned long long)map->m_lblk, map_len);
3383 sbi = EXT4_SB(inode->i_sb);
3384 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3385 inode->i_sb->s_blocksize_bits;
3386 if (eof_block < map->m_lblk + map_len)
3387 eof_block = map->m_lblk + map_len;
3389 depth = ext_depth(inode);
3390 eh = path[depth].p_hdr;
3391 ex = path[depth].p_ext;
3392 ee_block = le32_to_cpu(ex->ee_block);
3393 ee_len = ext4_ext_get_actual_len(ex);
3396 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3398 /* Pre-conditions */
3399 BUG_ON(!ext4_ext_is_unwritten(ex));
3400 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3403 * Attempt to transfer newly initialized blocks from the currently
3404 * unwritten extent to its neighbor. This is much cheaper
3405 * than an insertion followed by a merge as those involve costly
3406 * memmove() calls. Transferring to the left is the common case in
3407 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3408 * followed by append writes.
3410 * Limitations of the current logic:
3411 * - L1: we do not deal with writes covering the whole extent.
3412 * This would require removing the extent if the transfer
3414 * - L2: we only attempt to merge with an extent stored in the
3415 * same extent tree node.
3417 if ((map->m_lblk == ee_block) &&
3418 /* See if we can merge left */
3419 (map_len < ee_len) && /*L1*/
3420 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3421 ext4_lblk_t prev_lblk;
3422 ext4_fsblk_t prev_pblk, ee_pblk;
3423 unsigned int prev_len;
3426 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3427 prev_len = ext4_ext_get_actual_len(abut_ex);
3428 prev_pblk = ext4_ext_pblock(abut_ex);
3429 ee_pblk = ext4_ext_pblock(ex);
3432 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3433 * upon those conditions:
3434 * - C1: abut_ex is initialized,
3435 * - C2: abut_ex is logically abutting ex,
3436 * - C3: abut_ex is physically abutting ex,
3437 * - C4: abut_ex can receive the additional blocks without
3438 * overflowing the (initialized) length limit.
3440 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3441 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3442 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3443 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3444 err = ext4_ext_get_access(handle, inode, path + depth);
3448 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3451 /* Shift the start of ex by 'map_len' blocks */
3452 ex->ee_block = cpu_to_le32(ee_block + map_len);
3453 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3454 ex->ee_len = cpu_to_le16(ee_len - map_len);
3455 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3457 /* Extend abut_ex by 'map_len' blocks */
3458 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3460 /* Result: number of initialized blocks past m_lblk */
3461 allocated = map_len;
3463 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3464 (map_len < ee_len) && /*L1*/
3465 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3466 /* See if we can merge right */
3467 ext4_lblk_t next_lblk;
3468 ext4_fsblk_t next_pblk, ee_pblk;
3469 unsigned int next_len;
3472 next_lblk = le32_to_cpu(abut_ex->ee_block);
3473 next_len = ext4_ext_get_actual_len(abut_ex);
3474 next_pblk = ext4_ext_pblock(abut_ex);
3475 ee_pblk = ext4_ext_pblock(ex);
3478 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3479 * upon those conditions:
3480 * - C1: abut_ex is initialized,
3481 * - C2: abut_ex is logically abutting ex,
3482 * - C3: abut_ex is physically abutting ex,
3483 * - C4: abut_ex can receive the additional blocks without
3484 * overflowing the (initialized) length limit.
3486 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3487 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3488 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3489 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3490 err = ext4_ext_get_access(handle, inode, path + depth);
3494 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3497 /* Shift the start of abut_ex by 'map_len' blocks */
3498 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3499 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3500 ex->ee_len = cpu_to_le16(ee_len - map_len);
3501 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3503 /* Extend abut_ex by 'map_len' blocks */
3504 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3506 /* Result: number of initialized blocks past m_lblk */
3507 allocated = map_len;
3511 /* Mark the block containing both extents as dirty */
3512 ext4_ext_dirty(handle, inode, path + depth);
3514 /* Update path to point to the right extent */
3515 path[depth].p_ext = abut_ex;
3518 allocated = ee_len - (map->m_lblk - ee_block);
3520 WARN_ON(map->m_lblk < ee_block);
3522 * It is safe to convert extent to initialized via explicit
3523 * zeroout only if extent is fully inside i_size or new_size.
3525 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3527 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3528 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3529 (inode->i_sb->s_blocksize_bits - 10);
3531 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3532 if (max_zeroout && (ee_len <= max_zeroout)) {
3533 err = ext4_ext_zeroout(inode, ex);
3536 zero_ex.ee_block = ex->ee_block;
3537 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3538 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3540 err = ext4_ext_get_access(handle, inode, path + depth);
3543 ext4_ext_mark_initialized(ex);
3544 ext4_ext_try_to_merge(handle, inode, path, ex);
3545 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3551 * 1. split the extent into three extents.
3552 * 2. split the extent into two extents, zeroout the first half.
3553 * 3. split the extent into two extents, zeroout the second half.
3554 * 4. split the extent into two extents with out zeroout.
3556 split_map.m_lblk = map->m_lblk;
3557 split_map.m_len = map->m_len;
3559 if (max_zeroout && (allocated > map->m_len)) {
3560 if (allocated <= max_zeroout) {
3563 cpu_to_le32(map->m_lblk);
3564 zero_ex.ee_len = cpu_to_le16(allocated);
3565 ext4_ext_store_pblock(&zero_ex,
3566 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3567 err = ext4_ext_zeroout(inode, &zero_ex);
3570 split_map.m_lblk = map->m_lblk;
3571 split_map.m_len = allocated;
3572 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3574 if (map->m_lblk != ee_block) {
3575 zero_ex.ee_block = ex->ee_block;
3576 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3578 ext4_ext_store_pblock(&zero_ex,
3579 ext4_ext_pblock(ex));
3580 err = ext4_ext_zeroout(inode, &zero_ex);
3585 split_map.m_lblk = ee_block;
3586 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3587 allocated = map->m_len;
3591 allocated = ext4_split_extent(handle, inode, path,
3592 &split_map, split_flag, flags);
3597 /* If we have gotten a failure, don't zero out status tree */
3599 err = ext4_zeroout_es(inode, &zero_ex);
3600 return err ? err : allocated;
3604 * This function is called by ext4_ext_map_blocks() from
3605 * ext4_get_blocks_dio_write() when DIO to write
3606 * to an unwritten extent.
3608 * Writing to an unwritten extent may result in splitting the unwritten
3609 * extent into multiple initialized/unwritten extents (up to three)
3610 * There are three possibilities:
3611 * a> There is no split required: Entire extent should be unwritten
3612 * b> Splits in two extents: Write is happening at either end of the extent
3613 * c> Splits in three extents: Somone is writing in middle of the extent
3615 * This works the same way in the case of initialized -> unwritten conversion.
3617 * One of more index blocks maybe needed if the extent tree grow after
3618 * the unwritten extent split. To prevent ENOSPC occur at the IO
3619 * complete, we need to split the unwritten extent before DIO submit
3620 * the IO. The unwritten extent called at this time will be split
3621 * into three unwritten extent(at most). After IO complete, the part
3622 * being filled will be convert to initialized by the end_io callback function
3623 * via ext4_convert_unwritten_extents().
3625 * Returns the size of unwritten extent to be written on success.
3627 static int ext4_split_convert_extents(handle_t *handle,
3628 struct inode *inode,
3629 struct ext4_map_blocks *map,
3630 struct ext4_ext_path *path,
3633 ext4_lblk_t eof_block;
3634 ext4_lblk_t ee_block;
3635 struct ext4_extent *ex;
3636 unsigned int ee_len;
3637 int split_flag = 0, depth;
3639 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3640 __func__, inode->i_ino,
3641 (unsigned long long)map->m_lblk, map->m_len);
3643 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3644 inode->i_sb->s_blocksize_bits;
3645 if (eof_block < map->m_lblk + map->m_len)
3646 eof_block = map->m_lblk + map->m_len;
3648 * It is safe to convert extent to initialized via explicit
3649 * zeroout only if extent is fully insde i_size or new_size.
3651 depth = ext_depth(inode);
3652 ex = path[depth].p_ext;
3653 ee_block = le32_to_cpu(ex->ee_block);
3654 ee_len = ext4_ext_get_actual_len(ex);
3656 /* Convert to unwritten */
3657 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3658 split_flag |= EXT4_EXT_DATA_VALID1;
3659 /* Convert to initialized */
3660 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3661 split_flag |= ee_block + ee_len <= eof_block ?
3662 EXT4_EXT_MAY_ZEROOUT : 0;
3663 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3665 flags |= EXT4_GET_BLOCKS_PRE_IO;
3666 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3669 static int ext4_convert_initialized_extents(handle_t *handle,
3670 struct inode *inode,
3671 struct ext4_map_blocks *map,
3672 struct ext4_ext_path *path)
3674 struct ext4_extent *ex;
3675 ext4_lblk_t ee_block;
3676 unsigned int ee_len;
3680 depth = ext_depth(inode);
3681 ex = path[depth].p_ext;
3682 ee_block = le32_to_cpu(ex->ee_block);
3683 ee_len = ext4_ext_get_actual_len(ex);
3685 ext_debug("%s: inode %lu, logical"
3686 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3687 (unsigned long long)ee_block, ee_len);
3689 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3690 err = ext4_split_convert_extents(handle, inode, map, path,
3691 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3694 ext4_ext_drop_refs(path);
3695 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3697 err = PTR_ERR(path);
3700 depth = ext_depth(inode);
3701 ex = path[depth].p_ext;
3703 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3704 (unsigned long) map->m_lblk);
3710 err = ext4_ext_get_access(handle, inode, path + depth);
3713 /* first mark the extent as unwritten */
3714 ext4_ext_mark_unwritten(ex);
3716 /* note: ext4_ext_correct_indexes() isn't needed here because
3717 * borders are not changed
3719 ext4_ext_try_to_merge(handle, inode, path, ex);
3721 /* Mark modified extent as dirty */
3722 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3724 ext4_ext_show_leaf(inode, path);
3729 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3730 struct inode *inode,
3731 struct ext4_map_blocks *map,
3732 struct ext4_ext_path *path)
3734 struct ext4_extent *ex;
3735 ext4_lblk_t ee_block;
3736 unsigned int ee_len;
3740 depth = ext_depth(inode);
3741 ex = path[depth].p_ext;
3742 ee_block = le32_to_cpu(ex->ee_block);
3743 ee_len = ext4_ext_get_actual_len(ex);
3745 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3746 "block %llu, max_blocks %u\n", inode->i_ino,
3747 (unsigned long long)ee_block, ee_len);
3749 /* If extent is larger than requested it is a clear sign that we still
3750 * have some extent state machine issues left. So extent_split is still
3752 * TODO: Once all related issues will be fixed this situation should be
3755 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3757 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3758 " len %u; IO logical block %llu, len %u\n",
3759 inode->i_ino, (unsigned long long)ee_block, ee_len,
3760 (unsigned long long)map->m_lblk, map->m_len);
3762 err = ext4_split_convert_extents(handle, inode, map, path,
3763 EXT4_GET_BLOCKS_CONVERT);
3766 ext4_ext_drop_refs(path);
3767 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3769 err = PTR_ERR(path);
3772 depth = ext_depth(inode);
3773 ex = path[depth].p_ext;
3776 err = ext4_ext_get_access(handle, inode, path + depth);
3779 /* first mark the extent as initialized */
3780 ext4_ext_mark_initialized(ex);
3782 /* note: ext4_ext_correct_indexes() isn't needed here because
3783 * borders are not changed
3785 ext4_ext_try_to_merge(handle, inode, path, ex);
3787 /* Mark modified extent as dirty */
3788 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3790 ext4_ext_show_leaf(inode, path);
3794 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3795 sector_t block, int count)
3798 for (i = 0; i < count; i++)
3799 unmap_underlying_metadata(bdev, block + i);
3803 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3805 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3807 struct ext4_ext_path *path,
3811 struct ext4_extent_header *eh;
3812 struct ext4_extent *last_ex;
3814 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3817 depth = ext_depth(inode);
3818 eh = path[depth].p_hdr;
3821 * We're going to remove EOFBLOCKS_FL entirely in future so we
3822 * do not care for this case anymore. Simply remove the flag
3823 * if there are no extents.
3825 if (unlikely(!eh->eh_entries))
3827 last_ex = EXT_LAST_EXTENT(eh);
3829 * We should clear the EOFBLOCKS_FL flag if we are writing the
3830 * last block in the last extent in the file. We test this by
3831 * first checking to see if the caller to
3832 * ext4_ext_get_blocks() was interested in the last block (or
3833 * a block beyond the last block) in the current extent. If
3834 * this turns out to be false, we can bail out from this
3835 * function immediately.
3837 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3838 ext4_ext_get_actual_len(last_ex))
3841 * If the caller does appear to be planning to write at or
3842 * beyond the end of the current extent, we then test to see
3843 * if the current extent is the last extent in the file, by
3844 * checking to make sure it was reached via the rightmost node
3845 * at each level of the tree.
3847 for (i = depth-1; i >= 0; i--)
3848 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3851 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3852 return ext4_mark_inode_dirty(handle, inode);
3856 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3858 * Return 1 if there is a delalloc block in the range, otherwise 0.
3860 int ext4_find_delalloc_range(struct inode *inode,
3861 ext4_lblk_t lblk_start,
3862 ext4_lblk_t lblk_end)
3864 struct extent_status es;
3866 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3868 return 0; /* there is no delay extent in this tree */
3869 else if (es.es_lblk <= lblk_start &&
3870 lblk_start < es.es_lblk + es.es_len)
3872 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3878 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3881 ext4_lblk_t lblk_start, lblk_end;
3882 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3883 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3885 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3889 * Determines how many complete clusters (out of those specified by the 'map')
3890 * are under delalloc and were reserved quota for.
3891 * This function is called when we are writing out the blocks that were
3892 * originally written with their allocation delayed, but then the space was
3893 * allocated using fallocate() before the delayed allocation could be resolved.
3894 * The cases to look for are:
3895 * ('=' indicated delayed allocated blocks
3896 * '-' indicates non-delayed allocated blocks)
3897 * (a) partial clusters towards beginning and/or end outside of allocated range
3898 * are not delalloc'ed.
3900 * |----c---=|====c====|====c====|===-c----|
3901 * |++++++ allocated ++++++|
3902 * ==> 4 complete clusters in above example
3904 * (b) partial cluster (outside of allocated range) towards either end is
3905 * marked for delayed allocation. In this case, we will exclude that
3908 * |----====c========|========c========|
3909 * |++++++ allocated ++++++|
3910 * ==> 1 complete clusters in above example
3913 * |================c================|
3914 * |++++++ allocated ++++++|
3915 * ==> 0 complete clusters in above example
3917 * The ext4_da_update_reserve_space will be called only if we
3918 * determine here that there were some "entire" clusters that span
3919 * this 'allocated' range.
3920 * In the non-bigalloc case, this function will just end up returning num_blks
3921 * without ever calling ext4_find_delalloc_range.
3924 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3925 unsigned int num_blks)
3927 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3928 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3929 ext4_lblk_t lblk_from, lblk_to, c_offset;
3930 unsigned int allocated_clusters = 0;
3932 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3933 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3935 /* max possible clusters for this allocation */
3936 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3938 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3940 /* Check towards left side */
3941 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3943 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3944 lblk_to = lblk_from + c_offset - 1;
3946 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3947 allocated_clusters--;
3950 /* Now check towards right. */
3951 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3952 if (allocated_clusters && c_offset) {
3953 lblk_from = lblk_start + num_blks;
3954 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3956 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3957 allocated_clusters--;
3960 return allocated_clusters;
3964 ext4_ext_convert_initialized_extent(handle_t *handle, struct inode *inode,
3965 struct ext4_map_blocks *map,
3966 struct ext4_ext_path *path, int flags,
3967 unsigned int allocated, ext4_fsblk_t newblock)
3973 * Make sure that the extent is no bigger than we support with
3976 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3977 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3979 ret = ext4_convert_initialized_extents(handle, inode, map,
3982 ext4_update_inode_fsync_trans(handle, inode, 1);
3983 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3987 map->m_flags |= EXT4_MAP_UNWRITTEN;
3988 if (allocated > map->m_len)
3989 allocated = map->m_len;
3990 map->m_len = allocated;
3992 return err ? err : allocated;
3996 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3997 struct ext4_map_blocks *map,
3998 struct ext4_ext_path *path, int flags,
3999 unsigned int allocated, ext4_fsblk_t newblock)
4003 ext4_io_end_t *io = ext4_inode_aio(inode);
4005 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4006 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4007 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4009 ext4_ext_show_leaf(inode, path);
4012 * When writing into unwritten space, we should not fail to
4013 * allocate metadata blocks for the new extent block if needed.
4015 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4017 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4018 allocated, newblock);
4020 /* get_block() before submit the IO, split the extent */
4021 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4022 ret = ext4_split_convert_extents(handle, inode, map,
4023 path, flags | EXT4_GET_BLOCKS_CONVERT);
4027 * Flag the inode(non aio case) or end_io struct (aio case)
4028 * that this IO needs to conversion to written when IO is
4032 ext4_set_io_unwritten_flag(inode, io);
4034 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4035 map->m_flags |= EXT4_MAP_UNWRITTEN;
4038 /* IO end_io complete, convert the filled extent to written */
4039 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4040 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4043 ext4_update_inode_fsync_trans(handle, inode, 1);
4044 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4048 map->m_flags |= EXT4_MAP_MAPPED;
4049 map->m_pblk = newblock;
4050 if (allocated > map->m_len)
4051 allocated = map->m_len;
4052 map->m_len = allocated;
4055 /* buffered IO case */
4057 * repeat fallocate creation request
4058 * we already have an unwritten extent
4060 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4061 map->m_flags |= EXT4_MAP_UNWRITTEN;
4065 /* buffered READ or buffered write_begin() lookup */
4066 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4068 * We have blocks reserved already. We
4069 * return allocated blocks so that delalloc
4070 * won't do block reservation for us. But
4071 * the buffer head will be unmapped so that
4072 * a read from the block returns 0s.
4074 map->m_flags |= EXT4_MAP_UNWRITTEN;
4078 /* buffered write, writepage time, convert*/
4079 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
4081 ext4_update_inode_fsync_trans(handle, inode, 1);
4088 map->m_flags |= EXT4_MAP_NEW;
4090 * if we allocated more blocks than requested
4091 * we need to make sure we unmap the extra block
4092 * allocated. The actual needed block will get
4093 * unmapped later when we find the buffer_head marked
4096 if (allocated > map->m_len) {
4097 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4098 newblock + map->m_len,
4099 allocated - map->m_len);
4100 allocated = map->m_len;
4102 map->m_len = allocated;
4105 * If we have done fallocate with the offset that is already
4106 * delayed allocated, we would have block reservation
4107 * and quota reservation done in the delayed write path.
4108 * But fallocate would have already updated quota and block
4109 * count for this offset. So cancel these reservation
4111 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4112 unsigned int reserved_clusters;
4113 reserved_clusters = get_reserved_cluster_alloc(inode,
4114 map->m_lblk, map->m_len);
4115 if (reserved_clusters)
4116 ext4_da_update_reserve_space(inode,
4122 map->m_flags |= EXT4_MAP_MAPPED;
4123 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4124 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4130 if (allocated > map->m_len)
4131 allocated = map->m_len;
4132 ext4_ext_show_leaf(inode, path);
4133 map->m_pblk = newblock;
4134 map->m_len = allocated;
4136 return err ? err : allocated;
4140 * get_implied_cluster_alloc - check to see if the requested
4141 * allocation (in the map structure) overlaps with a cluster already
4142 * allocated in an extent.
4143 * @sb The filesystem superblock structure
4144 * @map The requested lblk->pblk mapping
4145 * @ex The extent structure which might contain an implied
4146 * cluster allocation
4148 * This function is called by ext4_ext_map_blocks() after we failed to
4149 * find blocks that were already in the inode's extent tree. Hence,
4150 * we know that the beginning of the requested region cannot overlap
4151 * the extent from the inode's extent tree. There are three cases we
4152 * want to catch. The first is this case:
4154 * |--- cluster # N--|
4155 * |--- extent ---| |---- requested region ---|
4158 * The second case that we need to test for is this one:
4160 * |--------- cluster # N ----------------|
4161 * |--- requested region --| |------- extent ----|
4162 * |=======================|
4164 * The third case is when the requested region lies between two extents
4165 * within the same cluster:
4166 * |------------- cluster # N-------------|
4167 * |----- ex -----| |---- ex_right ----|
4168 * |------ requested region ------|
4169 * |================|
4171 * In each of the above cases, we need to set the map->m_pblk and
4172 * map->m_len so it corresponds to the return the extent labelled as
4173 * "|====|" from cluster #N, since it is already in use for data in
4174 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4175 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4176 * as a new "allocated" block region. Otherwise, we will return 0 and
4177 * ext4_ext_map_blocks() will then allocate one or more new clusters
4178 * by calling ext4_mb_new_blocks().
4180 static int get_implied_cluster_alloc(struct super_block *sb,
4181 struct ext4_map_blocks *map,
4182 struct ext4_extent *ex,
4183 struct ext4_ext_path *path)
4185 struct ext4_sb_info *sbi = EXT4_SB(sb);
4186 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4187 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4188 ext4_lblk_t rr_cluster_start;
4189 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4190 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4191 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4193 /* The extent passed in that we are trying to match */
4194 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4195 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4197 /* The requested region passed into ext4_map_blocks() */
4198 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4200 if ((rr_cluster_start == ex_cluster_end) ||
4201 (rr_cluster_start == ex_cluster_start)) {
4202 if (rr_cluster_start == ex_cluster_end)
4203 ee_start += ee_len - 1;
4204 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4205 map->m_len = min(map->m_len,
4206 (unsigned) sbi->s_cluster_ratio - c_offset);
4208 * Check for and handle this case:
4210 * |--------- cluster # N-------------|
4211 * |------- extent ----|
4212 * |--- requested region ---|
4216 if (map->m_lblk < ee_block)
4217 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4220 * Check for the case where there is already another allocated
4221 * block to the right of 'ex' but before the end of the cluster.
4223 * |------------- cluster # N-------------|
4224 * |----- ex -----| |---- ex_right ----|
4225 * |------ requested region ------|
4226 * |================|
4228 if (map->m_lblk > ee_block) {
4229 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4230 map->m_len = min(map->m_len, next - map->m_lblk);
4233 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4237 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4243 * Block allocation/map/preallocation routine for extents based files
4246 * Need to be called with
4247 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4248 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4250 * return > 0, number of of blocks already mapped/allocated
4251 * if create == 0 and these are pre-allocated blocks
4252 * buffer head is unmapped
4253 * otherwise blocks are mapped
4255 * return = 0, if plain look up failed (blocks have not been allocated)
4256 * buffer head is unmapped
4258 * return < 0, error case.
4260 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4261 struct ext4_map_blocks *map, int flags)
4263 struct ext4_ext_path *path = NULL;
4264 struct ext4_extent newex, *ex, *ex2;
4265 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4266 ext4_fsblk_t newblock = 0;
4267 int free_on_err = 0, err = 0, depth, ret;
4268 unsigned int allocated = 0, offset = 0;
4269 unsigned int allocated_clusters = 0;
4270 struct ext4_allocation_request ar;
4271 ext4_io_end_t *io = ext4_inode_aio(inode);
4272 ext4_lblk_t cluster_offset;
4273 int set_unwritten = 0;
4275 ext_debug("blocks %u/%u requested for inode %lu\n",
4276 map->m_lblk, map->m_len, inode->i_ino);
4277 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4279 /* find extent for this block */
4280 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4282 err = PTR_ERR(path);
4287 depth = ext_depth(inode);
4290 * consistent leaf must not be empty;
4291 * this situation is possible, though, _during_ tree modification;
4292 * this is why assert can't be put in ext4_ext_find_extent()
4294 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4295 EXT4_ERROR_INODE(inode, "bad extent address "
4296 "lblock: %lu, depth: %d pblock %lld",
4297 (unsigned long) map->m_lblk, depth,
4298 path[depth].p_block);
4303 ex = path[depth].p_ext;
4305 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4306 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4307 unsigned short ee_len;
4311 * unwritten extents are treated as holes, except that
4312 * we split out initialized portions during a write.
4314 ee_len = ext4_ext_get_actual_len(ex);
4316 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4318 /* if found extent covers block, simply return it */
4319 if (in_range(map->m_lblk, ee_block, ee_len)) {
4320 newblock = map->m_lblk - ee_block + ee_start;
4321 /* number of remaining blocks in the extent */
4322 allocated = ee_len - (map->m_lblk - ee_block);
4323 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4324 ee_block, ee_len, newblock);
4327 * If the extent is initialized check whether the
4328 * caller wants to convert it to unwritten.
4330 if ((!ext4_ext_is_unwritten(ex)) &&
4331 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4332 allocated = ext4_ext_convert_initialized_extent(
4333 handle, inode, map, path, flags,
4334 allocated, newblock);
4336 } else if (!ext4_ext_is_unwritten(ex))
4339 ret = ext4_ext_handle_unwritten_extents(
4340 handle, inode, map, path, flags,
4341 allocated, newblock);
4350 if ((sbi->s_cluster_ratio > 1) &&
4351 ext4_find_delalloc_cluster(inode, map->m_lblk))
4352 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4355 * requested block isn't allocated yet;
4356 * we couldn't try to create block if create flag is zero
4358 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4360 * put just found gap into cache to speed up
4361 * subsequent requests
4363 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4364 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4369 * Okay, we need to do block allocation.
4371 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4372 newex.ee_block = cpu_to_le32(map->m_lblk);
4373 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4376 * If we are doing bigalloc, check to see if the extent returned
4377 * by ext4_ext_find_extent() implies a cluster we can use.
4379 if (cluster_offset && ex &&
4380 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4381 ar.len = allocated = map->m_len;
4382 newblock = map->m_pblk;
4383 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4384 goto got_allocated_blocks;
4387 /* find neighbour allocated blocks */
4388 ar.lleft = map->m_lblk;
4389 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4392 ar.lright = map->m_lblk;
4394 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4398 /* Check if the extent after searching to the right implies a
4399 * cluster we can use. */
4400 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4401 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4402 ar.len = allocated = map->m_len;
4403 newblock = map->m_pblk;
4404 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4405 goto got_allocated_blocks;
4409 * See if request is beyond maximum number of blocks we can have in
4410 * a single extent. For an initialized extent this limit is
4411 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4412 * EXT_UNWRITTEN_MAX_LEN.
4414 if (map->m_len > EXT_INIT_MAX_LEN &&
4415 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4416 map->m_len = EXT_INIT_MAX_LEN;
4417 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4418 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4419 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4421 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4422 newex.ee_len = cpu_to_le16(map->m_len);
4423 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4425 allocated = ext4_ext_get_actual_len(&newex);
4427 allocated = map->m_len;
4429 /* allocate new block */
4431 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4432 ar.logical = map->m_lblk;
4434 * We calculate the offset from the beginning of the cluster
4435 * for the logical block number, since when we allocate a
4436 * physical cluster, the physical block should start at the
4437 * same offset from the beginning of the cluster. This is
4438 * needed so that future calls to get_implied_cluster_alloc()
4441 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4442 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4444 ar.logical -= offset;
4445 if (S_ISREG(inode->i_mode))
4446 ar.flags = EXT4_MB_HINT_DATA;
4448 /* disable in-core preallocation for non-regular files */
4450 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4451 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4452 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4455 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4456 ar.goal, newblock, allocated);
4458 allocated_clusters = ar.len;
4459 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4460 if (ar.len > allocated)
4463 got_allocated_blocks:
4464 /* try to insert new extent into found leaf and return */
4465 ext4_ext_store_pblock(&newex, newblock + offset);
4466 newex.ee_len = cpu_to_le16(ar.len);
4467 /* Mark unwritten */
4468 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4469 ext4_ext_mark_unwritten(&newex);
4470 map->m_flags |= EXT4_MAP_UNWRITTEN;
4472 * io_end structure was created for every IO write to an
4473 * unwritten extent. To avoid unnecessary conversion,
4474 * here we flag the IO that really needs the conversion.
4475 * For non asycn direct IO case, flag the inode state
4476 * that we need to perform conversion when IO is done.
4478 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4483 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4484 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4487 err = ext4_ext_insert_extent(handle, inode, path,
4490 if (!err && set_unwritten) {
4492 ext4_set_io_unwritten_flag(inode, io);
4494 ext4_set_inode_state(inode,
4495 EXT4_STATE_DIO_UNWRITTEN);
4498 if (err && free_on_err) {
4499 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4500 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4501 /* free data blocks we just allocated */
4502 /* not a good idea to call discard here directly,
4503 * but otherwise we'd need to call it every free() */
4504 ext4_discard_preallocations(inode);
4505 ext4_free_blocks(handle, inode, NULL, newblock,
4506 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4510 /* previous routine could use block we allocated */
4511 newblock = ext4_ext_pblock(&newex);
4512 allocated = ext4_ext_get_actual_len(&newex);
4513 if (allocated > map->m_len)
4514 allocated = map->m_len;
4515 map->m_flags |= EXT4_MAP_NEW;
4518 * Update reserved blocks/metadata blocks after successful
4519 * block allocation which had been deferred till now.
4521 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4522 unsigned int reserved_clusters;
4524 * Check how many clusters we had reserved this allocated range
4526 reserved_clusters = get_reserved_cluster_alloc(inode,
4527 map->m_lblk, allocated);
4528 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4529 if (reserved_clusters) {
4531 * We have clusters reserved for this range.
4532 * But since we are not doing actual allocation
4533 * and are simply using blocks from previously
4534 * allocated cluster, we should release the
4535 * reservation and not claim quota.
4537 ext4_da_update_reserve_space(inode,
4538 reserved_clusters, 0);
4541 BUG_ON(allocated_clusters < reserved_clusters);
4542 if (reserved_clusters < allocated_clusters) {
4543 struct ext4_inode_info *ei = EXT4_I(inode);
4544 int reservation = allocated_clusters -
4547 * It seems we claimed few clusters outside of
4548 * the range of this allocation. We should give
4549 * it back to the reservation pool. This can
4550 * happen in the following case:
4552 * * Suppose s_cluster_ratio is 4 (i.e., each
4553 * cluster has 4 blocks. Thus, the clusters
4554 * are [0-3],[4-7],[8-11]...
4555 * * First comes delayed allocation write for
4556 * logical blocks 10 & 11. Since there were no
4557 * previous delayed allocated blocks in the
4558 * range [8-11], we would reserve 1 cluster
4560 * * Next comes write for logical blocks 3 to 8.
4561 * In this case, we will reserve 2 clusters
4562 * (for [0-3] and [4-7]; and not for [8-11] as
4563 * that range has a delayed allocated blocks.
4564 * Thus total reserved clusters now becomes 3.
4565 * * Now, during the delayed allocation writeout
4566 * time, we will first write blocks [3-8] and
4567 * allocate 3 clusters for writing these
4568 * blocks. Also, we would claim all these
4569 * three clusters above.
4570 * * Now when we come here to writeout the
4571 * blocks [10-11], we would expect to claim
4572 * the reservation of 1 cluster we had made
4573 * (and we would claim it since there are no
4574 * more delayed allocated blocks in the range
4575 * [8-11]. But our reserved cluster count had
4576 * already gone to 0.
4578 * Thus, at the step 4 above when we determine
4579 * that there are still some unwritten delayed
4580 * allocated blocks outside of our current
4581 * block range, we should increment the
4582 * reserved clusters count so that when the
4583 * remaining blocks finally gets written, we
4586 dquot_reserve_block(inode,
4587 EXT4_C2B(sbi, reservation));
4588 spin_lock(&ei->i_block_reservation_lock);
4589 ei->i_reserved_data_blocks += reservation;
4590 spin_unlock(&ei->i_block_reservation_lock);
4593 * We will claim quota for all newly allocated blocks.
4594 * We're updating the reserved space *after* the
4595 * correction above so we do not accidentally free
4596 * all the metadata reservation because we might
4597 * actually need it later on.
4599 ext4_da_update_reserve_space(inode, allocated_clusters,
4605 * Cache the extent and update transaction to commit on fdatasync only
4606 * when it is _not_ an unwritten extent.
4608 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4609 ext4_update_inode_fsync_trans(handle, inode, 1);
4611 ext4_update_inode_fsync_trans(handle, inode, 0);
4613 if (allocated > map->m_len)
4614 allocated = map->m_len;
4615 ext4_ext_show_leaf(inode, path);
4616 map->m_flags |= EXT4_MAP_MAPPED;
4617 map->m_pblk = newblock;
4618 map->m_len = allocated;
4621 ext4_ext_drop_refs(path);
4625 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4626 err ? err : allocated);
4627 ext4_es_lru_add(inode);
4628 return err ? err : allocated;
4631 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4633 struct super_block *sb = inode->i_sb;
4634 ext4_lblk_t last_block;
4638 * TODO: optimization is possible here.
4639 * Probably we need not scan at all,
4640 * because page truncation is enough.
4643 /* we have to know where to truncate from in crash case */
4644 EXT4_I(inode)->i_disksize = inode->i_size;
4645 ext4_mark_inode_dirty(handle, inode);
4647 last_block = (inode->i_size + sb->s_blocksize - 1)
4648 >> EXT4_BLOCK_SIZE_BITS(sb);
4650 err = ext4_es_remove_extent(inode, last_block,
4651 EXT_MAX_BLOCKS - last_block);
4652 if (err == -ENOMEM) {
4654 congestion_wait(BLK_RW_ASYNC, HZ/50);
4658 ext4_std_error(inode->i_sb, err);
4661 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4662 ext4_std_error(inode->i_sb, err);
4665 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4666 ext4_lblk_t len, int flags, int mode)
4668 struct inode *inode = file_inode(file);
4673 struct ext4_map_blocks map;
4674 unsigned int credits;
4676 map.m_lblk = offset;
4678 * Don't normalize the request if it can fit in one extent so
4679 * that it doesn't get unnecessarily split into multiple
4682 if (len <= EXT_UNWRITTEN_MAX_LEN)
4683 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4686 * credits to insert 1 extent into extent tree
4688 credits = ext4_chunk_trans_blocks(inode, len);
4691 while (ret >= 0 && ret < len) {
4692 map.m_lblk = map.m_lblk + ret;
4693 map.m_len = len = len - ret;
4694 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4696 if (IS_ERR(handle)) {
4697 ret = PTR_ERR(handle);
4700 ret = ext4_map_blocks(handle, inode, &map, flags);
4702 ext4_debug("inode #%lu: block %u: len %u: "
4703 "ext4_ext_map_blocks returned %d",
4704 inode->i_ino, map.m_lblk,
4706 ext4_mark_inode_dirty(handle, inode);
4707 ret2 = ext4_journal_stop(handle);
4710 ret2 = ext4_journal_stop(handle);
4714 if (ret == -ENOSPC &&
4715 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4720 return ret > 0 ? ret2 : ret;
4723 static long ext4_zero_range(struct file *file, loff_t offset,
4724 loff_t len, int mode)
4726 struct inode *inode = file_inode(file);
4727 handle_t *handle = NULL;
4728 unsigned int max_blocks;
4729 loff_t new_size = 0;
4735 struct address_space *mapping = inode->i_mapping;
4736 unsigned int blkbits = inode->i_blkbits;
4738 trace_ext4_zero_range(inode, offset, len, mode);
4740 if (!S_ISREG(inode->i_mode))
4743 /* Call ext4_force_commit to flush all data in case of data=journal. */
4744 if (ext4_should_journal_data(inode)) {
4745 ret = ext4_force_commit(inode->i_sb);
4751 * Write out all dirty pages to avoid race conditions
4752 * Then release them.
4754 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4755 ret = filemap_write_and_wait_range(mapping, offset,
4762 * Round up offset. This is not fallocate, we neet to zero out
4763 * blocks, so convert interior block aligned part of the range to
4764 * unwritten and possibly manually zero out unaligned parts of the
4767 start = round_up(offset, 1 << blkbits);
4768 end = round_down((offset + len), 1 << blkbits);
4770 if (start < offset || end > offset + len)
4772 partial = (offset + len) & ((1 << blkbits) - 1);
4774 lblk = start >> blkbits;
4775 max_blocks = (end >> blkbits);
4776 if (max_blocks < lblk)
4781 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4782 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN;
4783 if (mode & FALLOC_FL_KEEP_SIZE)
4784 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4786 mutex_lock(&inode->i_mutex);
4789 * Indirect files do not support unwritten extnets
4791 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4796 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4797 offset + len > i_size_read(inode)) {
4798 new_size = offset + len;
4799 ret = inode_newsize_ok(inode, new_size);
4803 * If we have a partial block after EOF we have to allocate
4810 if (max_blocks > 0) {
4812 /* Now release the pages and zero block aligned part of pages*/
4813 truncate_pagecache_range(inode, start, end - 1);
4815 /* Wait all existing dio workers, newcomers will block on i_mutex */
4816 ext4_inode_block_unlocked_dio(inode);
4817 inode_dio_wait(inode);
4820 * Remove entire range from the extent status tree.
4822 ret = ext4_es_remove_extent(inode, lblk, max_blocks);
4826 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, flags,
4832 handle = ext4_journal_start(inode, EXT4_HT_MISC, 4);
4833 if (IS_ERR(handle)) {
4834 ret = PTR_ERR(handle);
4835 ext4_std_error(inode->i_sb, ret);
4839 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4842 if (new_size > i_size_read(inode))
4843 i_size_write(inode, new_size);
4844 if (new_size > EXT4_I(inode)->i_disksize)
4845 ext4_update_i_disksize(inode, new_size);
4848 * Mark that we allocate beyond EOF so the subsequent truncate
4849 * can proceed even if the new size is the same as i_size.
4851 if ((offset + len) > i_size_read(inode))
4852 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4855 ext4_mark_inode_dirty(handle, inode);
4857 /* Zero out partial block at the edges of the range */
4858 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4860 if (file->f_flags & O_SYNC)
4861 ext4_handle_sync(handle);
4863 ext4_journal_stop(handle);
4865 ext4_inode_resume_unlocked_dio(inode);
4867 mutex_unlock(&inode->i_mutex);
4872 * preallocate space for a file. This implements ext4's fallocate file
4873 * operation, which gets called from sys_fallocate system call.
4874 * For block-mapped files, posix_fallocate should fall back to the method
4875 * of writing zeroes to the required new blocks (the same behavior which is
4876 * expected for file systems which do not support fallocate() system call).
4878 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4880 struct inode *inode = file_inode(file);
4882 loff_t new_size = 0;
4883 unsigned int max_blocks;
4888 unsigned int blkbits = inode->i_blkbits;
4890 /* Return error if mode is not supported */
4891 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4892 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4895 if (mode & FALLOC_FL_PUNCH_HOLE)
4896 return ext4_punch_hole(inode, offset, len);
4898 ret = ext4_convert_inline_data(inode);
4903 * currently supporting (pre)allocate mode for extent-based
4906 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4909 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4910 return ext4_collapse_range(inode, offset, len);
4912 if (mode & FALLOC_FL_ZERO_RANGE)
4913 return ext4_zero_range(file, offset, len, mode);
4915 trace_ext4_fallocate_enter(inode, offset, len, mode);
4916 lblk = offset >> blkbits;
4918 * We can't just convert len to max_blocks because
4919 * If blocksize = 4096 offset = 3072 and len = 2048
4921 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4924 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4925 if (mode & FALLOC_FL_KEEP_SIZE)
4926 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4928 mutex_lock(&inode->i_mutex);
4930 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4931 offset + len > i_size_read(inode)) {
4932 new_size = offset + len;
4933 ret = inode_newsize_ok(inode, new_size);
4938 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, flags, mode);
4942 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
4946 tv = inode->i_ctime = ext4_current_time(inode);
4949 if (new_size > i_size_read(inode)) {
4950 i_size_write(inode, new_size);
4951 inode->i_mtime = tv;
4953 if (new_size > EXT4_I(inode)->i_disksize)
4954 ext4_update_i_disksize(inode, new_size);
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_BLOCK_SIZE(sb) - 1) ||
5406 len & (EXT4_BLOCK_SIZE(sb) - 1))
5409 if (!S_ISREG(inode->i_mode))
5412 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1)
5415 trace_ext4_collapse_range(inode, offset, len);
5417 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5418 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5420 /* Call ext4_force_commit to flush all data in case of data=journal. */
5421 if (ext4_should_journal_data(inode)) {
5422 ret = ext4_force_commit(inode->i_sb);
5428 * Need to round down offset to be aligned with page size boundary
5429 * for page size > block size.
5431 ioffset = round_down(offset, PAGE_SIZE);
5433 /* Write out all dirty pages */
5434 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5439 /* Take mutex lock */
5440 mutex_lock(&inode->i_mutex);
5443 * There is no need to overlap collapse range with EOF, in which case
5444 * it is effectively a truncate operation
5446 if (offset + len >= i_size_read(inode)) {
5451 /* Currently just for extent based files */
5452 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5457 truncate_pagecache(inode, ioffset);
5459 /* Wait for existing dio to complete */
5460 ext4_inode_block_unlocked_dio(inode);
5461 inode_dio_wait(inode);
5463 credits = ext4_writepage_trans_blocks(inode);
5464 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5465 if (IS_ERR(handle)) {
5466 ret = PTR_ERR(handle);
5470 down_write(&EXT4_I(inode)->i_data_sem);
5471 ext4_discard_preallocations(inode);
5473 ret = ext4_es_remove_extent(inode, punch_start,
5474 EXT_MAX_BLOCKS - punch_start);
5476 up_write(&EXT4_I(inode)->i_data_sem);
5480 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5482 up_write(&EXT4_I(inode)->i_data_sem);
5485 ext4_discard_preallocations(inode);
5487 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5488 punch_stop - punch_start);
5490 up_write(&EXT4_I(inode)->i_data_sem);
5494 new_size = i_size_read(inode) - len;
5495 i_size_write(inode, new_size);
5496 EXT4_I(inode)->i_disksize = new_size;
5498 up_write(&EXT4_I(inode)->i_data_sem);
5500 ext4_handle_sync(handle);
5501 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5502 ext4_mark_inode_dirty(handle, inode);
5505 ext4_journal_stop(handle);
5507 ext4_inode_resume_unlocked_dio(inode);
5509 mutex_unlock(&inode->i_mutex);