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 License
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 <linux/uaccess.h>
41 #include <linux/fiemap.h>
42 #include <linux/backing-dev.h>
43 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
50 * used by extent splitting.
52 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
54 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
55 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
57 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
60 static __le32 ext4_extent_block_csum(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_inode_info *ei = EXT4_I(inode);
64 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
68 EXT4_EXTENT_TAIL_OFFSET(eh));
69 return cpu_to_le32(csum);
72 static int ext4_extent_block_csum_verify(struct inode *inode,
73 struct ext4_extent_header *eh)
75 struct ext4_extent_tail *et;
77 if (!ext4_has_metadata_csum(inode->i_sb))
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_metadata_csum(inode->i_sb))
94 et = find_ext4_extent_tail(eh);
95 et->et_checksum = ext4_extent_block_csum(inode, eh);
98 static int ext4_split_extent(handle_t *handle,
100 struct ext4_ext_path **ppath,
101 struct ext4_map_blocks *map,
105 static int ext4_split_extent_at(handle_t *handle,
107 struct ext4_ext_path **ppath,
112 static int ext4_find_delayed_extent(struct inode *inode,
113 struct extent_status *newes);
115 static int ext4_ext_truncate_extend_restart(handle_t *handle,
121 if (!ext4_handle_valid(handle))
123 if (handle->h_buffer_credits >= needed)
126 * If we need to extend the journal get a few extra blocks
127 * while we're at it for efficiency's sake.
130 err = ext4_journal_extend(handle, needed - handle->h_buffer_credits);
133 err = ext4_truncate_restart_trans(handle, inode, needed);
145 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
146 struct ext4_ext_path *path)
149 /* path points to block */
150 BUFFER_TRACE(path->p_bh, "get_write_access");
151 return ext4_journal_get_write_access(handle, path->p_bh);
153 /* path points to leaf/index in inode body */
154 /* we use in-core data, no need to protect them */
164 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
165 struct inode *inode, struct ext4_ext_path *path)
169 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
171 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
172 /* path points to block */
173 err = __ext4_handle_dirty_metadata(where, line, handle,
176 /* path points to leaf/index in inode body */
177 err = ext4_mark_inode_dirty(handle, inode);
182 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
183 struct ext4_ext_path *path,
187 int depth = path->p_depth;
188 struct ext4_extent *ex;
191 * Try to predict block placement assuming that we are
192 * filling in a file which will eventually be
193 * non-sparse --- i.e., in the case of libbfd writing
194 * an ELF object sections out-of-order but in a way
195 * the eventually results in a contiguous object or
196 * executable file, or some database extending a table
197 * space file. However, this is actually somewhat
198 * non-ideal if we are writing a sparse file such as
199 * qemu or KVM writing a raw image file that is going
200 * to stay fairly sparse, since it will end up
201 * fragmenting the file system's free space. Maybe we
202 * should have some hueristics or some way to allow
203 * userspace to pass a hint to file system,
204 * especially if the latter case turns out to be
207 ex = path[depth].p_ext;
209 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
210 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
212 if (block > ext_block)
213 return ext_pblk + (block - ext_block);
215 return ext_pblk - (ext_block - block);
218 /* it looks like index is empty;
219 * try to find starting block from index itself */
220 if (path[depth].p_bh)
221 return path[depth].p_bh->b_blocknr;
224 /* OK. use inode's group */
225 return ext4_inode_to_goal_block(inode);
229 * Allocation for a meta data block
232 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
233 struct ext4_ext_path *path,
234 struct ext4_extent *ex, int *err, unsigned int flags)
236 ext4_fsblk_t goal, newblock;
238 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
239 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
244 static inline int ext4_ext_space_block(struct inode *inode, int check)
248 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
249 / sizeof(struct ext4_extent);
250 #ifdef AGGRESSIVE_TEST
251 if (!check && size > 6)
257 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
261 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
262 / sizeof(struct ext4_extent_idx);
263 #ifdef AGGRESSIVE_TEST
264 if (!check && size > 5)
270 static inline int ext4_ext_space_root(struct inode *inode, int check)
274 size = sizeof(EXT4_I(inode)->i_data);
275 size -= sizeof(struct ext4_extent_header);
276 size /= sizeof(struct ext4_extent);
277 #ifdef AGGRESSIVE_TEST
278 if (!check && size > 3)
284 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
288 size = sizeof(EXT4_I(inode)->i_data);
289 size -= sizeof(struct ext4_extent_header);
290 size /= sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292 if (!check && size > 4)
299 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
300 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
303 struct ext4_ext_path *path = *ppath;
304 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
306 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
307 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
308 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
309 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
313 * Calculate the number of metadata blocks needed
314 * to allocate @blocks
315 * Worse case is one block per extent
317 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
319 struct ext4_inode_info *ei = EXT4_I(inode);
322 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
323 / sizeof(struct ext4_extent_idx));
326 * If the new delayed allocation block is contiguous with the
327 * previous da block, it can share index blocks with the
328 * previous block, so we only need to allocate a new index
329 * block every idxs leaf blocks. At ldxs**2 blocks, we need
330 * an additional index block, and at ldxs**3 blocks, yet
331 * another index blocks.
333 if (ei->i_da_metadata_calc_len &&
334 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
337 if ((ei->i_da_metadata_calc_len % idxs) == 0)
339 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
341 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
343 ei->i_da_metadata_calc_len = 0;
345 ei->i_da_metadata_calc_len++;
346 ei->i_da_metadata_calc_last_lblock++;
351 * In the worst case we need a new set of index blocks at
352 * every level of the inode's extent tree.
354 ei->i_da_metadata_calc_len = 1;
355 ei->i_da_metadata_calc_last_lblock = lblock;
356 return ext_depth(inode) + 1;
360 ext4_ext_max_entries(struct inode *inode, int depth)
364 if (depth == ext_depth(inode)) {
366 max = ext4_ext_space_root(inode, 1);
368 max = ext4_ext_space_root_idx(inode, 1);
371 max = ext4_ext_space_block(inode, 1);
373 max = ext4_ext_space_block_idx(inode, 1);
379 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
381 ext4_fsblk_t block = ext4_ext_pblock(ext);
382 int len = ext4_ext_get_actual_len(ext);
383 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
388 * - overflow/wrap-around
390 if (lblock + len <= lblock)
392 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
395 static int ext4_valid_extent_idx(struct inode *inode,
396 struct ext4_extent_idx *ext_idx)
398 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
400 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
403 static int ext4_valid_extent_entries(struct inode *inode,
404 struct ext4_extent_header *eh,
407 unsigned short entries;
408 if (eh->eh_entries == 0)
411 entries = le16_to_cpu(eh->eh_entries);
415 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
416 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
417 ext4_fsblk_t pblock = 0;
418 ext4_lblk_t lblock = 0;
419 ext4_lblk_t prev = 0;
422 if (!ext4_valid_extent(inode, ext))
425 /* Check for overlapping extents */
426 lblock = le32_to_cpu(ext->ee_block);
427 len = ext4_ext_get_actual_len(ext);
428 if ((lblock <= prev) && prev) {
429 pblock = ext4_ext_pblock(ext);
430 es->s_last_error_block = cpu_to_le64(pblock);
435 prev = lblock + len - 1;
438 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
440 if (!ext4_valid_extent_idx(inode, ext_idx))
449 static int __ext4_ext_check(const char *function, unsigned int line,
450 struct inode *inode, struct ext4_extent_header *eh,
451 int depth, ext4_fsblk_t pblk)
453 const char *error_msg;
454 int max = 0, err = -EFSCORRUPTED;
456 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
457 error_msg = "invalid magic";
460 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
461 error_msg = "unexpected eh_depth";
464 if (unlikely(eh->eh_max == 0)) {
465 error_msg = "invalid eh_max";
468 max = ext4_ext_max_entries(inode, depth);
469 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
470 error_msg = "too large eh_max";
473 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
474 error_msg = "invalid eh_entries";
477 if (!ext4_valid_extent_entries(inode, eh, depth)) {
478 error_msg = "invalid extent entries";
481 if (unlikely(depth > 32)) {
482 error_msg = "too large eh_depth";
485 /* Verify checksum on non-root extent tree nodes */
486 if (ext_depth(inode) != depth &&
487 !ext4_extent_block_csum_verify(inode, eh)) {
488 error_msg = "extent tree corrupted";
495 ext4_error_inode(inode, function, line, 0,
496 "pblk %llu bad header/extent: %s - magic %x, "
497 "entries %u, max %u(%u), depth %u(%u)",
498 (unsigned long long) pblk, error_msg,
499 le16_to_cpu(eh->eh_magic),
500 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
501 max, le16_to_cpu(eh->eh_depth), depth);
505 #define ext4_ext_check(inode, eh, depth, pblk) \
506 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
508 int ext4_ext_check_inode(struct inode *inode)
510 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
513 static struct buffer_head *
514 __read_extent_tree_block(const char *function, unsigned int line,
515 struct inode *inode, ext4_fsblk_t pblk, int depth,
518 struct buffer_head *bh;
521 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
523 return ERR_PTR(-ENOMEM);
525 if (!bh_uptodate_or_lock(bh)) {
526 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
527 err = bh_submit_read(bh);
531 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
533 err = __ext4_ext_check(function, line, inode,
534 ext_block_hdr(bh), depth, pblk);
537 set_buffer_verified(bh);
539 * If this is a leaf block, cache all of its entries
541 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
542 struct ext4_extent_header *eh = ext_block_hdr(bh);
543 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
544 ext4_lblk_t prev = 0;
547 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
548 unsigned int status = EXTENT_STATUS_WRITTEN;
549 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
550 int len = ext4_ext_get_actual_len(ex);
552 if (prev && (prev != lblk))
553 ext4_es_cache_extent(inode, prev,
557 if (ext4_ext_is_unwritten(ex))
558 status = EXTENT_STATUS_UNWRITTEN;
559 ext4_es_cache_extent(inode, lblk, len,
560 ext4_ext_pblock(ex), status);
571 #define read_extent_tree_block(inode, pblk, depth, flags) \
572 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
576 * This function is called to cache a file's extent information in the
579 int ext4_ext_precache(struct inode *inode)
581 struct ext4_inode_info *ei = EXT4_I(inode);
582 struct ext4_ext_path *path = NULL;
583 struct buffer_head *bh;
584 int i = 0, depth, ret = 0;
586 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
587 return 0; /* not an extent-mapped inode */
589 down_read(&ei->i_data_sem);
590 depth = ext_depth(inode);
592 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
595 up_read(&ei->i_data_sem);
599 /* Don't cache anything if there are no external extent blocks */
602 path[0].p_hdr = ext_inode_hdr(inode);
603 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
606 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
609 * If this is a leaf block or we've reached the end of
610 * the index block, go up
613 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
614 brelse(path[i].p_bh);
619 bh = read_extent_tree_block(inode,
620 ext4_idx_pblock(path[i].p_idx++),
622 EXT4_EX_FORCE_CACHE);
629 path[i].p_hdr = ext_block_hdr(bh);
630 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
632 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
634 up_read(&ei->i_data_sem);
635 ext4_ext_drop_refs(path);
641 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
643 int k, l = path->p_depth;
646 for (k = 0; k <= l; k++, path++) {
648 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
649 ext4_idx_pblock(path->p_idx));
650 } else if (path->p_ext) {
651 ext_debug(" %d:[%d]%d:%llu ",
652 le32_to_cpu(path->p_ext->ee_block),
653 ext4_ext_is_unwritten(path->p_ext),
654 ext4_ext_get_actual_len(path->p_ext),
655 ext4_ext_pblock(path->p_ext));
662 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
664 int depth = ext_depth(inode);
665 struct ext4_extent_header *eh;
666 struct ext4_extent *ex;
672 eh = path[depth].p_hdr;
673 ex = EXT_FIRST_EXTENT(eh);
675 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
677 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
678 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
679 ext4_ext_is_unwritten(ex),
680 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
685 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
686 ext4_fsblk_t newblock, int level)
688 int depth = ext_depth(inode);
689 struct ext4_extent *ex;
691 if (depth != level) {
692 struct ext4_extent_idx *idx;
693 idx = path[level].p_idx;
694 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
695 ext_debug("%d: move %d:%llu in new index %llu\n", level,
696 le32_to_cpu(idx->ei_block),
697 ext4_idx_pblock(idx),
705 ex = path[depth].p_ext;
706 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
707 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
708 le32_to_cpu(ex->ee_block),
710 ext4_ext_is_unwritten(ex),
711 ext4_ext_get_actual_len(ex),
718 #define ext4_ext_show_path(inode, path)
719 #define ext4_ext_show_leaf(inode, path)
720 #define ext4_ext_show_move(inode, path, newblock, level)
723 void ext4_ext_drop_refs(struct ext4_ext_path *path)
729 depth = path->p_depth;
730 for (i = 0; i <= depth; i++, path++)
738 * ext4_ext_binsearch_idx:
739 * binary search for the closest index of the given block
740 * the header must be checked before calling this
743 ext4_ext_binsearch_idx(struct inode *inode,
744 struct ext4_ext_path *path, ext4_lblk_t block)
746 struct ext4_extent_header *eh = path->p_hdr;
747 struct ext4_extent_idx *r, *l, *m;
750 ext_debug("binsearch for %u(idx): ", block);
752 l = EXT_FIRST_INDEX(eh) + 1;
753 r = EXT_LAST_INDEX(eh);
756 if (block < le32_to_cpu(m->ei_block))
760 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
761 m, le32_to_cpu(m->ei_block),
762 r, le32_to_cpu(r->ei_block));
766 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
767 ext4_idx_pblock(path->p_idx));
769 #ifdef CHECK_BINSEARCH
771 struct ext4_extent_idx *chix, *ix;
774 chix = ix = EXT_FIRST_INDEX(eh);
775 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
777 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
778 printk(KERN_DEBUG "k=%d, ix=0x%p, "
780 ix, EXT_FIRST_INDEX(eh));
781 printk(KERN_DEBUG "%u <= %u\n",
782 le32_to_cpu(ix->ei_block),
783 le32_to_cpu(ix[-1].ei_block));
785 BUG_ON(k && le32_to_cpu(ix->ei_block)
786 <= le32_to_cpu(ix[-1].ei_block));
787 if (block < le32_to_cpu(ix->ei_block))
791 BUG_ON(chix != path->p_idx);
798 * ext4_ext_binsearch:
799 * binary search for closest extent of the given block
800 * the header must be checked before calling this
803 ext4_ext_binsearch(struct inode *inode,
804 struct ext4_ext_path *path, ext4_lblk_t block)
806 struct ext4_extent_header *eh = path->p_hdr;
807 struct ext4_extent *r, *l, *m;
809 if (eh->eh_entries == 0) {
811 * this leaf is empty:
812 * we get such a leaf in split/add case
817 ext_debug("binsearch for %u: ", block);
819 l = EXT_FIRST_EXTENT(eh) + 1;
820 r = EXT_LAST_EXTENT(eh);
824 if (block < le32_to_cpu(m->ee_block))
828 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
829 m, le32_to_cpu(m->ee_block),
830 r, le32_to_cpu(r->ee_block));
834 ext_debug(" -> %d:%llu:[%d]%d ",
835 le32_to_cpu(path->p_ext->ee_block),
836 ext4_ext_pblock(path->p_ext),
837 ext4_ext_is_unwritten(path->p_ext),
838 ext4_ext_get_actual_len(path->p_ext));
840 #ifdef CHECK_BINSEARCH
842 struct ext4_extent *chex, *ex;
845 chex = ex = EXT_FIRST_EXTENT(eh);
846 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
847 BUG_ON(k && le32_to_cpu(ex->ee_block)
848 <= le32_to_cpu(ex[-1].ee_block));
849 if (block < le32_to_cpu(ex->ee_block))
853 BUG_ON(chex != path->p_ext);
859 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
861 struct ext4_extent_header *eh;
863 eh = ext_inode_hdr(inode);
866 eh->eh_magic = EXT4_EXT_MAGIC;
867 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
868 ext4_mark_inode_dirty(handle, inode);
872 struct ext4_ext_path *
873 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
874 struct ext4_ext_path **orig_path, int flags)
876 struct ext4_extent_header *eh;
877 struct buffer_head *bh;
878 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
879 short int depth, i, ppos = 0;
882 eh = ext_inode_hdr(inode);
883 depth = ext_depth(inode);
886 ext4_ext_drop_refs(path);
887 if (depth > path[0].p_maxdepth) {
889 *orig_path = path = NULL;
893 /* account possible depth increase */
894 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
897 return ERR_PTR(-ENOMEM);
898 path[0].p_maxdepth = depth + 1;
904 /* walk through the tree */
906 ext_debug("depth %d: num %d, max %d\n",
907 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
909 ext4_ext_binsearch_idx(inode, path + ppos, block);
910 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
911 path[ppos].p_depth = i;
912 path[ppos].p_ext = NULL;
914 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
921 eh = ext_block_hdr(bh);
923 path[ppos].p_bh = bh;
924 path[ppos].p_hdr = eh;
927 path[ppos].p_depth = i;
928 path[ppos].p_ext = NULL;
929 path[ppos].p_idx = NULL;
932 ext4_ext_binsearch(inode, path + ppos, block);
933 /* if not an empty leaf */
934 if (path[ppos].p_ext)
935 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
937 ext4_ext_show_path(inode, path);
942 ext4_ext_drop_refs(path);
950 * ext4_ext_insert_index:
951 * insert new index [@logical;@ptr] into the block at @curp;
952 * check where to insert: before @curp or after @curp
954 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
955 struct ext4_ext_path *curp,
956 int logical, ext4_fsblk_t ptr)
958 struct ext4_extent_idx *ix;
961 err = ext4_ext_get_access(handle, inode, curp);
965 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
966 EXT4_ERROR_INODE(inode,
967 "logical %d == ei_block %d!",
968 logical, le32_to_cpu(curp->p_idx->ei_block));
969 return -EFSCORRUPTED;
972 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
973 >= le16_to_cpu(curp->p_hdr->eh_max))) {
974 EXT4_ERROR_INODE(inode,
975 "eh_entries %d >= eh_max %d!",
976 le16_to_cpu(curp->p_hdr->eh_entries),
977 le16_to_cpu(curp->p_hdr->eh_max));
978 return -EFSCORRUPTED;
981 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
983 ext_debug("insert new index %d after: %llu\n", logical, ptr);
984 ix = curp->p_idx + 1;
987 ext_debug("insert new index %d before: %llu\n", logical, ptr);
991 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
994 ext_debug("insert new index %d: "
995 "move %d indices from 0x%p to 0x%p\n",
996 logical, len, ix, ix + 1);
997 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1000 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1001 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1002 return -EFSCORRUPTED;
1005 ix->ei_block = cpu_to_le32(logical);
1006 ext4_idx_store_pblock(ix, ptr);
1007 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1009 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1010 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1011 return -EFSCORRUPTED;
1014 err = ext4_ext_dirty(handle, inode, curp);
1015 ext4_std_error(inode->i_sb, err);
1022 * inserts new subtree into the path, using free index entry
1024 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1025 * - makes decision where to split
1026 * - moves remaining extents and index entries (right to the split point)
1027 * into the newly allocated blocks
1028 * - initializes subtree
1030 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1032 struct ext4_ext_path *path,
1033 struct ext4_extent *newext, int at)
1035 struct buffer_head *bh = NULL;
1036 int depth = ext_depth(inode);
1037 struct ext4_extent_header *neh;
1038 struct ext4_extent_idx *fidx;
1039 int i = at, k, m, a;
1040 ext4_fsblk_t newblock, oldblock;
1042 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1045 /* make decision: where to split? */
1046 /* FIXME: now decision is simplest: at current extent */
1048 /* if current leaf will be split, then we should use
1049 * border from split point */
1050 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1051 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1052 return -EFSCORRUPTED;
1054 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1055 border = path[depth].p_ext[1].ee_block;
1056 ext_debug("leaf will be split."
1057 " next leaf starts at %d\n",
1058 le32_to_cpu(border));
1060 border = newext->ee_block;
1061 ext_debug("leaf will be added."
1062 " next leaf starts at %d\n",
1063 le32_to_cpu(border));
1067 * If error occurs, then we break processing
1068 * and mark filesystem read-only. index won't
1069 * be inserted and tree will be in consistent
1070 * state. Next mount will repair buffers too.
1074 * Get array to track all allocated blocks.
1075 * We need this to handle errors and free blocks
1078 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1082 /* allocate all needed blocks */
1083 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1084 for (a = 0; a < depth - at; a++) {
1085 newblock = ext4_ext_new_meta_block(handle, inode, path,
1086 newext, &err, flags);
1089 ablocks[a] = newblock;
1092 /* initialize new leaf */
1093 newblock = ablocks[--a];
1094 if (unlikely(newblock == 0)) {
1095 EXT4_ERROR_INODE(inode, "newblock == 0!");
1096 err = -EFSCORRUPTED;
1099 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1100 if (unlikely(!bh)) {
1106 err = ext4_journal_get_create_access(handle, bh);
1110 neh = ext_block_hdr(bh);
1111 neh->eh_entries = 0;
1112 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1113 neh->eh_magic = EXT4_EXT_MAGIC;
1116 /* move remainder of path[depth] to the new leaf */
1117 if (unlikely(path[depth].p_hdr->eh_entries !=
1118 path[depth].p_hdr->eh_max)) {
1119 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1120 path[depth].p_hdr->eh_entries,
1121 path[depth].p_hdr->eh_max);
1122 err = -EFSCORRUPTED;
1125 /* start copy from next extent */
1126 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1127 ext4_ext_show_move(inode, path, newblock, depth);
1129 struct ext4_extent *ex;
1130 ex = EXT_FIRST_EXTENT(neh);
1131 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1132 le16_add_cpu(&neh->eh_entries, m);
1135 ext4_extent_block_csum_set(inode, neh);
1136 set_buffer_uptodate(bh);
1139 err = ext4_handle_dirty_metadata(handle, inode, bh);
1145 /* correct old leaf */
1147 err = ext4_ext_get_access(handle, inode, path + depth);
1150 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1151 err = ext4_ext_dirty(handle, inode, path + depth);
1157 /* create intermediate indexes */
1159 if (unlikely(k < 0)) {
1160 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1161 err = -EFSCORRUPTED;
1165 ext_debug("create %d intermediate indices\n", k);
1166 /* insert new index into current index block */
1167 /* current depth stored in i var */
1170 oldblock = newblock;
1171 newblock = ablocks[--a];
1172 bh = sb_getblk(inode->i_sb, newblock);
1173 if (unlikely(!bh)) {
1179 err = ext4_journal_get_create_access(handle, bh);
1183 neh = ext_block_hdr(bh);
1184 neh->eh_entries = cpu_to_le16(1);
1185 neh->eh_magic = EXT4_EXT_MAGIC;
1186 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1187 neh->eh_depth = cpu_to_le16(depth - i);
1188 fidx = EXT_FIRST_INDEX(neh);
1189 fidx->ei_block = border;
1190 ext4_idx_store_pblock(fidx, oldblock);
1192 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1193 i, newblock, le32_to_cpu(border), oldblock);
1195 /* move remainder of path[i] to the new index block */
1196 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1197 EXT_LAST_INDEX(path[i].p_hdr))) {
1198 EXT4_ERROR_INODE(inode,
1199 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1200 le32_to_cpu(path[i].p_ext->ee_block));
1201 err = -EFSCORRUPTED;
1204 /* start copy indexes */
1205 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1206 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1207 EXT_MAX_INDEX(path[i].p_hdr));
1208 ext4_ext_show_move(inode, path, newblock, i);
1210 memmove(++fidx, path[i].p_idx,
1211 sizeof(struct ext4_extent_idx) * m);
1212 le16_add_cpu(&neh->eh_entries, m);
1214 ext4_extent_block_csum_set(inode, neh);
1215 set_buffer_uptodate(bh);
1218 err = ext4_handle_dirty_metadata(handle, inode, bh);
1224 /* correct old index */
1226 err = ext4_ext_get_access(handle, inode, path + i);
1229 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1230 err = ext4_ext_dirty(handle, inode, path + i);
1238 /* insert new index */
1239 err = ext4_ext_insert_index(handle, inode, path + at,
1240 le32_to_cpu(border), newblock);
1244 if (buffer_locked(bh))
1250 /* free all allocated blocks in error case */
1251 for (i = 0; i < depth; i++) {
1254 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1255 EXT4_FREE_BLOCKS_METADATA);
1264 * ext4_ext_grow_indepth:
1265 * implements tree growing procedure:
1266 * - allocates new block
1267 * - moves top-level data (index block or leaf) into the new block
1268 * - initializes new top-level, creating index that points to the
1269 * just created block
1271 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1274 struct ext4_extent_header *neh;
1275 struct buffer_head *bh;
1276 ext4_fsblk_t newblock, goal = 0;
1277 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1280 /* Try to prepend new index to old one */
1281 if (ext_depth(inode))
1282 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1283 if (goal > le32_to_cpu(es->s_first_data_block)) {
1284 flags |= EXT4_MB_HINT_TRY_GOAL;
1287 goal = ext4_inode_to_goal_block(inode);
1288 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1293 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1298 err = ext4_journal_get_create_access(handle, bh);
1304 /* move top-level index/leaf into new block */
1305 memmove(bh->b_data, EXT4_I(inode)->i_data,
1306 sizeof(EXT4_I(inode)->i_data));
1308 /* set size of new block */
1309 neh = ext_block_hdr(bh);
1310 /* old root could have indexes or leaves
1311 * so calculate e_max right way */
1312 if (ext_depth(inode))
1313 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1315 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1316 neh->eh_magic = EXT4_EXT_MAGIC;
1317 ext4_extent_block_csum_set(inode, neh);
1318 set_buffer_uptodate(bh);
1321 err = ext4_handle_dirty_metadata(handle, inode, bh);
1325 /* Update top-level index: num,max,pointer */
1326 neh = ext_inode_hdr(inode);
1327 neh->eh_entries = cpu_to_le16(1);
1328 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1329 if (neh->eh_depth == 0) {
1330 /* Root extent block becomes index block */
1331 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1332 EXT_FIRST_INDEX(neh)->ei_block =
1333 EXT_FIRST_EXTENT(neh)->ee_block;
1335 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1336 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1337 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1338 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1340 le16_add_cpu(&neh->eh_depth, 1);
1341 ext4_mark_inode_dirty(handle, inode);
1349 * ext4_ext_create_new_leaf:
1350 * finds empty index and adds new leaf.
1351 * if no free index is found, then it requests in-depth growing.
1353 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1354 unsigned int mb_flags,
1355 unsigned int gb_flags,
1356 struct ext4_ext_path **ppath,
1357 struct ext4_extent *newext)
1359 struct ext4_ext_path *path = *ppath;
1360 struct ext4_ext_path *curp;
1361 int depth, i, err = 0;
1364 i = depth = ext_depth(inode);
1366 /* walk up to the tree and look for free index entry */
1367 curp = path + depth;
1368 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1373 /* we use already allocated block for index block,
1374 * so subsequent data blocks should be contiguous */
1375 if (EXT_HAS_FREE_INDEX(curp)) {
1376 /* if we found index with free entry, then use that
1377 * entry: create all needed subtree and add new leaf */
1378 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1383 path = ext4_find_extent(inode,
1384 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1387 err = PTR_ERR(path);
1389 /* tree is full, time to grow in depth */
1390 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1395 path = ext4_find_extent(inode,
1396 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1399 err = PTR_ERR(path);
1404 * only first (depth 0 -> 1) produces free space;
1405 * in all other cases we have to split the grown tree
1407 depth = ext_depth(inode);
1408 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1409 /* now we need to split */
1419 * search the closest allocated block to the left for *logical
1420 * and returns it at @logical + it's physical address at @phys
1421 * if *logical is the smallest allocated block, the function
1422 * returns 0 at @phys
1423 * return value contains 0 (success) or error code
1425 static int ext4_ext_search_left(struct inode *inode,
1426 struct ext4_ext_path *path,
1427 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1429 struct ext4_extent_idx *ix;
1430 struct ext4_extent *ex;
1433 if (unlikely(path == NULL)) {
1434 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1435 return -EFSCORRUPTED;
1437 depth = path->p_depth;
1440 if (depth == 0 && path->p_ext == NULL)
1443 /* usually extent in the path covers blocks smaller
1444 * then *logical, but it can be that extent is the
1445 * first one in the file */
1447 ex = path[depth].p_ext;
1448 ee_len = ext4_ext_get_actual_len(ex);
1449 if (*logical < le32_to_cpu(ex->ee_block)) {
1450 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1451 EXT4_ERROR_INODE(inode,
1452 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1453 *logical, le32_to_cpu(ex->ee_block));
1454 return -EFSCORRUPTED;
1456 while (--depth >= 0) {
1457 ix = path[depth].p_idx;
1458 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1459 EXT4_ERROR_INODE(inode,
1460 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1461 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1462 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1463 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1465 return -EFSCORRUPTED;
1471 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1472 EXT4_ERROR_INODE(inode,
1473 "logical %d < ee_block %d + ee_len %d!",
1474 *logical, le32_to_cpu(ex->ee_block), ee_len);
1475 return -EFSCORRUPTED;
1478 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1479 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1484 * search the closest allocated block to the right for *logical
1485 * and returns it at @logical + it's physical address at @phys
1486 * if *logical is the largest allocated block, the function
1487 * returns 0 at @phys
1488 * return value contains 0 (success) or error code
1490 static int ext4_ext_search_right(struct inode *inode,
1491 struct ext4_ext_path *path,
1492 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1493 struct ext4_extent **ret_ex)
1495 struct buffer_head *bh = NULL;
1496 struct ext4_extent_header *eh;
1497 struct ext4_extent_idx *ix;
1498 struct ext4_extent *ex;
1500 int depth; /* Note, NOT eh_depth; depth from top of tree */
1503 if (unlikely(path == NULL)) {
1504 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1505 return -EFSCORRUPTED;
1507 depth = path->p_depth;
1510 if (depth == 0 && path->p_ext == NULL)
1513 /* usually extent in the path covers blocks smaller
1514 * then *logical, but it can be that extent is the
1515 * first one in the file */
1517 ex = path[depth].p_ext;
1518 ee_len = ext4_ext_get_actual_len(ex);
1519 if (*logical < le32_to_cpu(ex->ee_block)) {
1520 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1521 EXT4_ERROR_INODE(inode,
1522 "first_extent(path[%d].p_hdr) != ex",
1524 return -EFSCORRUPTED;
1526 while (--depth >= 0) {
1527 ix = path[depth].p_idx;
1528 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1529 EXT4_ERROR_INODE(inode,
1530 "ix != EXT_FIRST_INDEX *logical %d!",
1532 return -EFSCORRUPTED;
1538 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1539 EXT4_ERROR_INODE(inode,
1540 "logical %d < ee_block %d + ee_len %d!",
1541 *logical, le32_to_cpu(ex->ee_block), ee_len);
1542 return -EFSCORRUPTED;
1545 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1546 /* next allocated block in this leaf */
1551 /* go up and search for index to the right */
1552 while (--depth >= 0) {
1553 ix = path[depth].p_idx;
1554 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1558 /* we've gone up to the root and found no index to the right */
1562 /* we've found index to the right, let's
1563 * follow it and find the closest allocated
1564 * block to the right */
1566 block = ext4_idx_pblock(ix);
1567 while (++depth < path->p_depth) {
1568 /* subtract from p_depth to get proper eh_depth */
1569 bh = read_extent_tree_block(inode, block,
1570 path->p_depth - depth, 0);
1573 eh = ext_block_hdr(bh);
1574 ix = EXT_FIRST_INDEX(eh);
1575 block = ext4_idx_pblock(ix);
1579 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1582 eh = ext_block_hdr(bh);
1583 ex = EXT_FIRST_EXTENT(eh);
1585 *logical = le32_to_cpu(ex->ee_block);
1586 *phys = ext4_ext_pblock(ex);
1594 * ext4_ext_next_allocated_block:
1595 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1596 * NOTE: it considers block number from index entry as
1597 * allocated block. Thus, index entries have to be consistent
1601 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1605 BUG_ON(path == NULL);
1606 depth = path->p_depth;
1608 if (depth == 0 && path->p_ext == NULL)
1609 return EXT_MAX_BLOCKS;
1611 while (depth >= 0) {
1612 if (depth == path->p_depth) {
1614 if (path[depth].p_ext &&
1615 path[depth].p_ext !=
1616 EXT_LAST_EXTENT(path[depth].p_hdr))
1617 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1620 if (path[depth].p_idx !=
1621 EXT_LAST_INDEX(path[depth].p_hdr))
1622 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1627 return EXT_MAX_BLOCKS;
1631 * ext4_ext_next_leaf_block:
1632 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1634 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1638 BUG_ON(path == NULL);
1639 depth = path->p_depth;
1641 /* zero-tree has no leaf blocks at all */
1643 return EXT_MAX_BLOCKS;
1645 /* go to index block */
1648 while (depth >= 0) {
1649 if (path[depth].p_idx !=
1650 EXT_LAST_INDEX(path[depth].p_hdr))
1651 return (ext4_lblk_t)
1652 le32_to_cpu(path[depth].p_idx[1].ei_block);
1656 return EXT_MAX_BLOCKS;
1660 * ext4_ext_correct_indexes:
1661 * if leaf gets modified and modified extent is first in the leaf,
1662 * then we have to correct all indexes above.
1663 * TODO: do we need to correct tree in all cases?
1665 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1666 struct ext4_ext_path *path)
1668 struct ext4_extent_header *eh;
1669 int depth = ext_depth(inode);
1670 struct ext4_extent *ex;
1674 eh = path[depth].p_hdr;
1675 ex = path[depth].p_ext;
1677 if (unlikely(ex == NULL || eh == NULL)) {
1678 EXT4_ERROR_INODE(inode,
1679 "ex %p == NULL or eh %p == NULL", ex, eh);
1680 return -EFSCORRUPTED;
1684 /* there is no tree at all */
1688 if (ex != EXT_FIRST_EXTENT(eh)) {
1689 /* we correct tree if first leaf got modified only */
1694 * TODO: we need correction if border is smaller than current one
1697 border = path[depth].p_ext->ee_block;
1698 err = ext4_ext_get_access(handle, inode, path + k);
1701 path[k].p_idx->ei_block = border;
1702 err = ext4_ext_dirty(handle, inode, path + k);
1707 /* change all left-side indexes */
1708 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1710 err = ext4_ext_get_access(handle, inode, path + k);
1713 path[k].p_idx->ei_block = border;
1714 err = ext4_ext_dirty(handle, inode, path + k);
1723 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1724 struct ext4_extent *ex2)
1726 unsigned short ext1_ee_len, ext2_ee_len;
1728 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1731 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1732 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1734 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1735 le32_to_cpu(ex2->ee_block))
1739 * To allow future support for preallocated extents to be added
1740 * as an RO_COMPAT feature, refuse to merge to extents if
1741 * this can result in the top bit of ee_len being set.
1743 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1746 * The check for IO to unwritten extent is somewhat racy as we
1747 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1748 * dropping i_data_sem. But reserved blocks should save us in that
1751 if (ext4_ext_is_unwritten(ex1) &&
1752 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1753 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1754 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1756 #ifdef AGGRESSIVE_TEST
1757 if (ext1_ee_len >= 4)
1761 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1767 * This function tries to merge the "ex" extent to the next extent in the tree.
1768 * It always tries to merge towards right. If you want to merge towards
1769 * left, pass "ex - 1" as argument instead of "ex".
1770 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1771 * 1 if they got merged.
1773 static int ext4_ext_try_to_merge_right(struct inode *inode,
1774 struct ext4_ext_path *path,
1775 struct ext4_extent *ex)
1777 struct ext4_extent_header *eh;
1778 unsigned int depth, len;
1779 int merge_done = 0, unwritten;
1781 depth = ext_depth(inode);
1782 BUG_ON(path[depth].p_hdr == NULL);
1783 eh = path[depth].p_hdr;
1785 while (ex < EXT_LAST_EXTENT(eh)) {
1786 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1788 /* merge with next extent! */
1789 unwritten = ext4_ext_is_unwritten(ex);
1790 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1791 + ext4_ext_get_actual_len(ex + 1));
1793 ext4_ext_mark_unwritten(ex);
1795 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1796 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1797 * sizeof(struct ext4_extent);
1798 memmove(ex + 1, ex + 2, len);
1800 le16_add_cpu(&eh->eh_entries, -1);
1802 WARN_ON(eh->eh_entries == 0);
1803 if (!eh->eh_entries)
1804 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1811 * This function does a very simple check to see if we can collapse
1812 * an extent tree with a single extent tree leaf block into the inode.
1814 static void ext4_ext_try_to_merge_up(handle_t *handle,
1815 struct inode *inode,
1816 struct ext4_ext_path *path)
1819 unsigned max_root = ext4_ext_space_root(inode, 0);
1822 if ((path[0].p_depth != 1) ||
1823 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1824 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1828 * We need to modify the block allocation bitmap and the block
1829 * group descriptor to release the extent tree block. If we
1830 * can't get the journal credits, give up.
1832 if (ext4_journal_extend(handle, 2))
1836 * Copy the extent data up to the inode
1838 blk = ext4_idx_pblock(path[0].p_idx);
1839 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1840 sizeof(struct ext4_extent_idx);
1841 s += sizeof(struct ext4_extent_header);
1843 path[1].p_maxdepth = path[0].p_maxdepth;
1844 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1845 path[0].p_depth = 0;
1846 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1847 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1848 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1850 brelse(path[1].p_bh);
1851 ext4_free_blocks(handle, inode, NULL, blk, 1,
1852 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1856 * This function tries to merge the @ex extent to neighbours in the tree.
1857 * return 1 if merge left else 0.
1859 static void ext4_ext_try_to_merge(handle_t *handle,
1860 struct inode *inode,
1861 struct ext4_ext_path *path,
1862 struct ext4_extent *ex) {
1863 struct ext4_extent_header *eh;
1867 depth = ext_depth(inode);
1868 BUG_ON(path[depth].p_hdr == NULL);
1869 eh = path[depth].p_hdr;
1871 if (ex > EXT_FIRST_EXTENT(eh))
1872 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1875 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1877 ext4_ext_try_to_merge_up(handle, inode, path);
1881 * check if a portion of the "newext" extent overlaps with an
1884 * If there is an overlap discovered, it updates the length of the newext
1885 * such that there will be no overlap, and then returns 1.
1886 * If there is no overlap found, it returns 0.
1888 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1889 struct inode *inode,
1890 struct ext4_extent *newext,
1891 struct ext4_ext_path *path)
1894 unsigned int depth, len1;
1895 unsigned int ret = 0;
1897 b1 = le32_to_cpu(newext->ee_block);
1898 len1 = ext4_ext_get_actual_len(newext);
1899 depth = ext_depth(inode);
1900 if (!path[depth].p_ext)
1902 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1905 * get the next allocated block if the extent in the path
1906 * is before the requested block(s)
1909 b2 = ext4_ext_next_allocated_block(path);
1910 if (b2 == EXT_MAX_BLOCKS)
1912 b2 = EXT4_LBLK_CMASK(sbi, b2);
1915 /* check for wrap through zero on extent logical start block*/
1916 if (b1 + len1 < b1) {
1917 len1 = EXT_MAX_BLOCKS - b1;
1918 newext->ee_len = cpu_to_le16(len1);
1922 /* check for overlap */
1923 if (b1 + len1 > b2) {
1924 newext->ee_len = cpu_to_le16(b2 - b1);
1932 * ext4_ext_insert_extent:
1933 * tries to merge requsted extent into the existing extent or
1934 * inserts requested extent as new one into the tree,
1935 * creating new leaf in the no-space case.
1937 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1938 struct ext4_ext_path **ppath,
1939 struct ext4_extent *newext, int gb_flags)
1941 struct ext4_ext_path *path = *ppath;
1942 struct ext4_extent_header *eh;
1943 struct ext4_extent *ex, *fex;
1944 struct ext4_extent *nearex; /* nearest extent */
1945 struct ext4_ext_path *npath = NULL;
1946 int depth, len, err;
1948 int mb_flags = 0, unwritten;
1950 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1951 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1952 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1953 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1954 return -EFSCORRUPTED;
1956 depth = ext_depth(inode);
1957 ex = path[depth].p_ext;
1958 eh = path[depth].p_hdr;
1959 if (unlikely(path[depth].p_hdr == NULL)) {
1960 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1961 return -EFSCORRUPTED;
1964 /* try to insert block into found extent and return */
1965 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1968 * Try to see whether we should rather test the extent on
1969 * right from ex, or from the left of ex. This is because
1970 * ext4_find_extent() can return either extent on the
1971 * left, or on the right from the searched position. This
1972 * will make merging more effective.
1974 if (ex < EXT_LAST_EXTENT(eh) &&
1975 (le32_to_cpu(ex->ee_block) +
1976 ext4_ext_get_actual_len(ex) <
1977 le32_to_cpu(newext->ee_block))) {
1980 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1981 (le32_to_cpu(newext->ee_block) +
1982 ext4_ext_get_actual_len(newext) <
1983 le32_to_cpu(ex->ee_block)))
1986 /* Try to append newex to the ex */
1987 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1988 ext_debug("append [%d]%d block to %u:[%d]%d"
1990 ext4_ext_is_unwritten(newext),
1991 ext4_ext_get_actual_len(newext),
1992 le32_to_cpu(ex->ee_block),
1993 ext4_ext_is_unwritten(ex),
1994 ext4_ext_get_actual_len(ex),
1995 ext4_ext_pblock(ex));
1996 err = ext4_ext_get_access(handle, inode,
2000 unwritten = ext4_ext_is_unwritten(ex);
2001 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2002 + ext4_ext_get_actual_len(newext));
2004 ext4_ext_mark_unwritten(ex);
2005 eh = path[depth].p_hdr;
2011 /* Try to prepend newex to the ex */
2012 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2013 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2015 le32_to_cpu(newext->ee_block),
2016 ext4_ext_is_unwritten(newext),
2017 ext4_ext_get_actual_len(newext),
2018 le32_to_cpu(ex->ee_block),
2019 ext4_ext_is_unwritten(ex),
2020 ext4_ext_get_actual_len(ex),
2021 ext4_ext_pblock(ex));
2022 err = ext4_ext_get_access(handle, inode,
2027 unwritten = ext4_ext_is_unwritten(ex);
2028 ex->ee_block = newext->ee_block;
2029 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2030 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2031 + ext4_ext_get_actual_len(newext));
2033 ext4_ext_mark_unwritten(ex);
2034 eh = path[depth].p_hdr;
2040 depth = ext_depth(inode);
2041 eh = path[depth].p_hdr;
2042 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2045 /* probably next leaf has space for us? */
2046 fex = EXT_LAST_EXTENT(eh);
2047 next = EXT_MAX_BLOCKS;
2048 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2049 next = ext4_ext_next_leaf_block(path);
2050 if (next != EXT_MAX_BLOCKS) {
2051 ext_debug("next leaf block - %u\n", next);
2052 BUG_ON(npath != NULL);
2053 npath = ext4_find_extent(inode, next, NULL, 0);
2055 return PTR_ERR(npath);
2056 BUG_ON(npath->p_depth != path->p_depth);
2057 eh = npath[depth].p_hdr;
2058 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2059 ext_debug("next leaf isn't full(%d)\n",
2060 le16_to_cpu(eh->eh_entries));
2064 ext_debug("next leaf has no free space(%d,%d)\n",
2065 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2069 * There is no free space in the found leaf.
2070 * We're gonna add a new leaf in the tree.
2072 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2073 mb_flags |= EXT4_MB_USE_RESERVED;
2074 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2078 depth = ext_depth(inode);
2079 eh = path[depth].p_hdr;
2082 nearex = path[depth].p_ext;
2084 err = ext4_ext_get_access(handle, inode, path + depth);
2089 /* there is no extent in this leaf, create first one */
2090 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2091 le32_to_cpu(newext->ee_block),
2092 ext4_ext_pblock(newext),
2093 ext4_ext_is_unwritten(newext),
2094 ext4_ext_get_actual_len(newext));
2095 nearex = EXT_FIRST_EXTENT(eh);
2097 if (le32_to_cpu(newext->ee_block)
2098 > le32_to_cpu(nearex->ee_block)) {
2100 ext_debug("insert %u:%llu:[%d]%d before: "
2102 le32_to_cpu(newext->ee_block),
2103 ext4_ext_pblock(newext),
2104 ext4_ext_is_unwritten(newext),
2105 ext4_ext_get_actual_len(newext),
2110 BUG_ON(newext->ee_block == nearex->ee_block);
2111 ext_debug("insert %u:%llu:[%d]%d after: "
2113 le32_to_cpu(newext->ee_block),
2114 ext4_ext_pblock(newext),
2115 ext4_ext_is_unwritten(newext),
2116 ext4_ext_get_actual_len(newext),
2119 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2121 ext_debug("insert %u:%llu:[%d]%d: "
2122 "move %d extents from 0x%p to 0x%p\n",
2123 le32_to_cpu(newext->ee_block),
2124 ext4_ext_pblock(newext),
2125 ext4_ext_is_unwritten(newext),
2126 ext4_ext_get_actual_len(newext),
2127 len, nearex, nearex + 1);
2128 memmove(nearex + 1, nearex,
2129 len * sizeof(struct ext4_extent));
2133 le16_add_cpu(&eh->eh_entries, 1);
2134 path[depth].p_ext = nearex;
2135 nearex->ee_block = newext->ee_block;
2136 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2137 nearex->ee_len = newext->ee_len;
2140 /* try to merge extents */
2141 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2142 ext4_ext_try_to_merge(handle, inode, path, nearex);
2145 /* time to correct all indexes above */
2146 err = ext4_ext_correct_indexes(handle, inode, path);
2150 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2153 ext4_ext_drop_refs(npath);
2158 static int ext4_fill_fiemap_extents(struct inode *inode,
2159 ext4_lblk_t block, ext4_lblk_t num,
2160 struct fiemap_extent_info *fieinfo)
2162 struct ext4_ext_path *path = NULL;
2163 struct ext4_extent *ex;
2164 struct extent_status es;
2165 ext4_lblk_t next, next_del, start = 0, end = 0;
2166 ext4_lblk_t last = block + num;
2167 int exists, depth = 0, err = 0;
2168 unsigned int flags = 0;
2169 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2171 while (block < last && block != EXT_MAX_BLOCKS) {
2173 /* find extent for this block */
2174 down_read(&EXT4_I(inode)->i_data_sem);
2176 path = ext4_find_extent(inode, block, &path, 0);
2178 up_read(&EXT4_I(inode)->i_data_sem);
2179 err = PTR_ERR(path);
2184 depth = ext_depth(inode);
2185 if (unlikely(path[depth].p_hdr == NULL)) {
2186 up_read(&EXT4_I(inode)->i_data_sem);
2187 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2188 err = -EFSCORRUPTED;
2191 ex = path[depth].p_ext;
2192 next = ext4_ext_next_allocated_block(path);
2197 /* there is no extent yet, so try to allocate
2198 * all requested space */
2201 } else if (le32_to_cpu(ex->ee_block) > block) {
2202 /* need to allocate space before found extent */
2204 end = le32_to_cpu(ex->ee_block);
2205 if (block + num < end)
2207 } else if (block >= le32_to_cpu(ex->ee_block)
2208 + ext4_ext_get_actual_len(ex)) {
2209 /* need to allocate space after found extent */
2214 } else if (block >= le32_to_cpu(ex->ee_block)) {
2216 * some part of requested space is covered
2220 end = le32_to_cpu(ex->ee_block)
2221 + ext4_ext_get_actual_len(ex);
2222 if (block + num < end)
2228 BUG_ON(end <= start);
2232 es.es_len = end - start;
2235 es.es_lblk = le32_to_cpu(ex->ee_block);
2236 es.es_len = ext4_ext_get_actual_len(ex);
2237 es.es_pblk = ext4_ext_pblock(ex);
2238 if (ext4_ext_is_unwritten(ex))
2239 flags |= FIEMAP_EXTENT_UNWRITTEN;
2243 * Find delayed extent and update es accordingly. We call
2244 * it even in !exists case to find out whether es is the
2245 * last existing extent or not.
2247 next_del = ext4_find_delayed_extent(inode, &es);
2248 if (!exists && next_del) {
2250 flags |= (FIEMAP_EXTENT_DELALLOC |
2251 FIEMAP_EXTENT_UNKNOWN);
2253 up_read(&EXT4_I(inode)->i_data_sem);
2255 if (unlikely(es.es_len == 0)) {
2256 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2257 err = -EFSCORRUPTED;
2262 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2263 * we need to check next == EXT_MAX_BLOCKS because it is
2264 * possible that an extent is with unwritten and delayed
2265 * status due to when an extent is delayed allocated and
2266 * is allocated by fallocate status tree will track both of
2269 * So we could return a unwritten and delayed extent, and
2270 * its block is equal to 'next'.
2272 if (next == next_del && next == EXT_MAX_BLOCKS) {
2273 flags |= FIEMAP_EXTENT_LAST;
2274 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2275 next != EXT_MAX_BLOCKS)) {
2276 EXT4_ERROR_INODE(inode,
2277 "next extent == %u, next "
2278 "delalloc extent = %u",
2280 err = -EFSCORRUPTED;
2286 err = fiemap_fill_next_extent(fieinfo,
2287 (__u64)es.es_lblk << blksize_bits,
2288 (__u64)es.es_pblk << blksize_bits,
2289 (__u64)es.es_len << blksize_bits,
2299 block = es.es_lblk + es.es_len;
2302 ext4_ext_drop_refs(path);
2308 * ext4_ext_determine_hole - determine hole around given block
2309 * @inode: inode we lookup in
2310 * @path: path in extent tree to @lblk
2311 * @lblk: pointer to logical block around which we want to determine hole
2313 * Determine hole length (and start if easily possible) around given logical
2314 * block. We don't try too hard to find the beginning of the hole but @path
2315 * actually points to extent before @lblk, we provide it.
2317 * The function returns the length of a hole starting at @lblk. We update @lblk
2318 * to the beginning of the hole if we managed to find it.
2320 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2321 struct ext4_ext_path *path,
2324 int depth = ext_depth(inode);
2325 struct ext4_extent *ex;
2328 ex = path[depth].p_ext;
2330 /* there is no extent yet, so gap is [0;-] */
2332 len = EXT_MAX_BLOCKS;
2333 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2334 len = le32_to_cpu(ex->ee_block) - *lblk;
2335 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2336 + ext4_ext_get_actual_len(ex)) {
2339 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2340 next = ext4_ext_next_allocated_block(path);
2341 BUG_ON(next == *lblk);
2350 * ext4_ext_put_gap_in_cache:
2351 * calculate boundaries of the gap that the requested block fits into
2352 * and cache this gap
2355 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2356 ext4_lblk_t hole_len)
2358 struct extent_status es;
2360 ext4_es_find_delayed_extent_range(inode, hole_start,
2361 hole_start + hole_len - 1, &es);
2363 /* There's delayed extent containing lblock? */
2364 if (es.es_lblk <= hole_start)
2366 hole_len = min(es.es_lblk - hole_start, hole_len);
2368 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2369 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2370 EXTENT_STATUS_HOLE);
2375 * removes index from the index block.
2377 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2378 struct ext4_ext_path *path, int depth)
2383 /* free index block */
2385 path = path + depth;
2386 leaf = ext4_idx_pblock(path->p_idx);
2387 if (unlikely(path->p_hdr->eh_entries == 0)) {
2388 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2389 return -EFSCORRUPTED;
2391 err = ext4_ext_get_access(handle, inode, path);
2395 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2396 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2397 len *= sizeof(struct ext4_extent_idx);
2398 memmove(path->p_idx, path->p_idx + 1, len);
2401 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2402 err = ext4_ext_dirty(handle, inode, path);
2405 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2406 trace_ext4_ext_rm_idx(inode, leaf);
2408 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2409 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2411 while (--depth >= 0) {
2412 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2415 err = ext4_ext_get_access(handle, inode, path);
2418 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2419 err = ext4_ext_dirty(handle, inode, path);
2427 * ext4_ext_calc_credits_for_single_extent:
2428 * This routine returns max. credits that needed to insert an extent
2429 * to the extent tree.
2430 * When pass the actual path, the caller should calculate credits
2433 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2434 struct ext4_ext_path *path)
2437 int depth = ext_depth(inode);
2440 /* probably there is space in leaf? */
2441 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2442 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2445 * There are some space in the leaf tree, no
2446 * need to account for leaf block credit
2448 * bitmaps and block group descriptor blocks
2449 * and other metadata blocks still need to be
2452 /* 1 bitmap, 1 block group descriptor */
2453 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2458 return ext4_chunk_trans_blocks(inode, nrblocks);
2462 * How many index/leaf blocks need to change/allocate to add @extents extents?
2464 * If we add a single extent, then in the worse case, each tree level
2465 * index/leaf need to be changed in case of the tree split.
2467 * If more extents are inserted, they could cause the whole tree split more
2468 * than once, but this is really rare.
2470 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2475 /* If we are converting the inline data, only one is needed here. */
2476 if (ext4_has_inline_data(inode))
2479 depth = ext_depth(inode);
2489 static inline int get_default_free_blocks_flags(struct inode *inode)
2491 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2492 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2493 else if (ext4_should_journal_data(inode))
2494 return EXT4_FREE_BLOCKS_FORGET;
2498 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2499 struct ext4_extent *ex,
2500 long long *partial_cluster,
2501 ext4_lblk_t from, ext4_lblk_t to)
2503 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2504 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2506 int flags = get_default_free_blocks_flags(inode);
2509 * For bigalloc file systems, we never free a partial cluster
2510 * at the beginning of the extent. Instead, we make a note
2511 * that we tried freeing the cluster, and check to see if we
2512 * need to free it on a subsequent call to ext4_remove_blocks,
2513 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2515 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2517 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2519 * If we have a partial cluster, and it's different from the
2520 * cluster of the last block, we need to explicitly free the
2521 * partial cluster here.
2523 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2524 if (*partial_cluster > 0 &&
2525 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2526 ext4_free_blocks(handle, inode, NULL,
2527 EXT4_C2B(sbi, *partial_cluster),
2528 sbi->s_cluster_ratio, flags);
2529 *partial_cluster = 0;
2532 #ifdef EXTENTS_STATS
2534 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2535 spin_lock(&sbi->s_ext_stats_lock);
2536 sbi->s_ext_blocks += ee_len;
2537 sbi->s_ext_extents++;
2538 if (ee_len < sbi->s_ext_min)
2539 sbi->s_ext_min = ee_len;
2540 if (ee_len > sbi->s_ext_max)
2541 sbi->s_ext_max = ee_len;
2542 if (ext_depth(inode) > sbi->s_depth_max)
2543 sbi->s_depth_max = ext_depth(inode);
2544 spin_unlock(&sbi->s_ext_stats_lock);
2547 if (from >= le32_to_cpu(ex->ee_block)
2548 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2551 long long first_cluster;
2553 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2554 pblk = ext4_ext_pblock(ex) + ee_len - num;
2556 * Usually we want to free partial cluster at the end of the
2557 * extent, except for the situation when the cluster is still
2558 * used by any other extent (partial_cluster is negative).
2560 if (*partial_cluster < 0 &&
2561 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2562 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2564 ext_debug("free last %u blocks starting %llu partial %lld\n",
2565 num, pblk, *partial_cluster);
2566 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2568 * If the block range to be freed didn't start at the
2569 * beginning of a cluster, and we removed the entire
2570 * extent and the cluster is not used by any other extent,
2571 * save the partial cluster here, since we might need to
2572 * delete if we determine that the truncate or punch hole
2573 * operation has removed all of the blocks in the cluster.
2574 * If that cluster is used by another extent, preserve its
2575 * negative value so it isn't freed later on.
2577 * If the whole extent wasn't freed, we've reached the
2578 * start of the truncated/punched region and have finished
2579 * removing blocks. If there's a partial cluster here it's
2580 * shared with the remainder of the extent and is no longer
2581 * a candidate for removal.
2583 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2584 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2585 if (first_cluster != -*partial_cluster)
2586 *partial_cluster = first_cluster;
2588 *partial_cluster = 0;
2591 ext4_error(sbi->s_sb, "strange request: removal(2) "
2593 from, to, le32_to_cpu(ex->ee_block), ee_len);
2599 * ext4_ext_rm_leaf() Removes the extents associated with the
2600 * blocks appearing between "start" and "end". Both "start"
2601 * and "end" must appear in the same extent or EIO is returned.
2603 * @handle: The journal handle
2604 * @inode: The files inode
2605 * @path: The path to the leaf
2606 * @partial_cluster: The cluster which we'll have to free if all extents
2607 * has been released from it. However, if this value is
2608 * negative, it's a cluster just to the right of the
2609 * punched region and it must not be freed.
2610 * @start: The first block to remove
2611 * @end: The last block to remove
2614 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2615 struct ext4_ext_path *path,
2616 long long *partial_cluster,
2617 ext4_lblk_t start, ext4_lblk_t end)
2619 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2620 int err = 0, correct_index = 0;
2621 int depth = ext_depth(inode), credits;
2622 struct ext4_extent_header *eh;
2625 ext4_lblk_t ex_ee_block;
2626 unsigned short ex_ee_len;
2627 unsigned unwritten = 0;
2628 struct ext4_extent *ex;
2631 /* the header must be checked already in ext4_ext_remove_space() */
2632 ext_debug("truncate since %u in leaf to %u\n", start, end);
2633 if (!path[depth].p_hdr)
2634 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2635 eh = path[depth].p_hdr;
2636 if (unlikely(path[depth].p_hdr == NULL)) {
2637 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2638 return -EFSCORRUPTED;
2640 /* find where to start removing */
2641 ex = path[depth].p_ext;
2643 ex = EXT_LAST_EXTENT(eh);
2645 ex_ee_block = le32_to_cpu(ex->ee_block);
2646 ex_ee_len = ext4_ext_get_actual_len(ex);
2648 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2650 while (ex >= EXT_FIRST_EXTENT(eh) &&
2651 ex_ee_block + ex_ee_len > start) {
2653 if (ext4_ext_is_unwritten(ex))
2658 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2659 unwritten, ex_ee_len);
2660 path[depth].p_ext = ex;
2662 a = ex_ee_block > start ? ex_ee_block : start;
2663 b = ex_ee_block+ex_ee_len - 1 < end ?
2664 ex_ee_block+ex_ee_len - 1 : end;
2666 ext_debug(" border %u:%u\n", a, b);
2668 /* If this extent is beyond the end of the hole, skip it */
2669 if (end < ex_ee_block) {
2671 * We're going to skip this extent and move to another,
2672 * so note that its first cluster is in use to avoid
2673 * freeing it when removing blocks. Eventually, the
2674 * right edge of the truncated/punched region will
2675 * be just to the left.
2677 if (sbi->s_cluster_ratio > 1) {
2678 pblk = ext4_ext_pblock(ex);
2680 -(long long) EXT4_B2C(sbi, pblk);
2683 ex_ee_block = le32_to_cpu(ex->ee_block);
2684 ex_ee_len = ext4_ext_get_actual_len(ex);
2686 } else if (b != ex_ee_block + ex_ee_len - 1) {
2687 EXT4_ERROR_INODE(inode,
2688 "can not handle truncate %u:%u "
2690 start, end, ex_ee_block,
2691 ex_ee_block + ex_ee_len - 1);
2692 err = -EFSCORRUPTED;
2694 } else if (a != ex_ee_block) {
2695 /* remove tail of the extent */
2696 num = a - ex_ee_block;
2698 /* remove whole extent: excellent! */
2702 * 3 for leaf, sb, and inode plus 2 (bmap and group
2703 * descriptor) for each block group; assume two block
2704 * groups plus ex_ee_len/blocks_per_block_group for
2707 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2708 if (ex == EXT_FIRST_EXTENT(eh)) {
2710 credits += (ext_depth(inode)) + 1;
2712 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2714 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2718 err = ext4_ext_get_access(handle, inode, path + depth);
2722 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2728 /* this extent is removed; mark slot entirely unused */
2729 ext4_ext_store_pblock(ex, 0);
2731 ex->ee_len = cpu_to_le16(num);
2733 * Do not mark unwritten if all the blocks in the
2734 * extent have been removed.
2736 if (unwritten && num)
2737 ext4_ext_mark_unwritten(ex);
2739 * If the extent was completely released,
2740 * we need to remove it from the leaf
2743 if (end != EXT_MAX_BLOCKS - 1) {
2745 * For hole punching, we need to scoot all the
2746 * extents up when an extent is removed so that
2747 * we dont have blank extents in the middle
2749 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2750 sizeof(struct ext4_extent));
2752 /* Now get rid of the one at the end */
2753 memset(EXT_LAST_EXTENT(eh), 0,
2754 sizeof(struct ext4_extent));
2756 le16_add_cpu(&eh->eh_entries, -1);
2759 err = ext4_ext_dirty(handle, inode, path + depth);
2763 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2764 ext4_ext_pblock(ex));
2766 ex_ee_block = le32_to_cpu(ex->ee_block);
2767 ex_ee_len = ext4_ext_get_actual_len(ex);
2770 if (correct_index && eh->eh_entries)
2771 err = ext4_ext_correct_indexes(handle, inode, path);
2774 * If there's a partial cluster and at least one extent remains in
2775 * the leaf, free the partial cluster if it isn't shared with the
2776 * current extent. If it is shared with the current extent
2777 * we zero partial_cluster because we've reached the start of the
2778 * truncated/punched region and we're done removing blocks.
2780 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2781 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2782 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2783 ext4_free_blocks(handle, inode, NULL,
2784 EXT4_C2B(sbi, *partial_cluster),
2785 sbi->s_cluster_ratio,
2786 get_default_free_blocks_flags(inode));
2788 *partial_cluster = 0;
2791 /* if this leaf is free, then we should
2792 * remove it from index block above */
2793 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2794 err = ext4_ext_rm_idx(handle, inode, path, depth);
2801 * ext4_ext_more_to_rm:
2802 * returns 1 if current index has to be freed (even partial)
2805 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2807 BUG_ON(path->p_idx == NULL);
2809 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2813 * if truncate on deeper level happened, it wasn't partial,
2814 * so we have to consider current index for truncation
2816 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2821 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2824 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2825 int depth = ext_depth(inode);
2826 struct ext4_ext_path *path = NULL;
2827 long long partial_cluster = 0;
2831 ext_debug("truncate since %u to %u\n", start, end);
2833 /* probably first extent we're gonna free will be last in block */
2834 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2836 return PTR_ERR(handle);
2839 trace_ext4_ext_remove_space(inode, start, end, depth);
2842 * Check if we are removing extents inside the extent tree. If that
2843 * is the case, we are going to punch a hole inside the extent tree
2844 * so we have to check whether we need to split the extent covering
2845 * the last block to remove so we can easily remove the part of it
2846 * in ext4_ext_rm_leaf().
2848 if (end < EXT_MAX_BLOCKS - 1) {
2849 struct ext4_extent *ex;
2850 ext4_lblk_t ee_block, ex_end, lblk;
2853 /* find extent for or closest extent to this block */
2854 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2856 ext4_journal_stop(handle);
2857 return PTR_ERR(path);
2859 depth = ext_depth(inode);
2860 /* Leaf not may not exist only if inode has no blocks at all */
2861 ex = path[depth].p_ext;
2864 EXT4_ERROR_INODE(inode,
2865 "path[%d].p_hdr == NULL",
2867 err = -EFSCORRUPTED;
2872 ee_block = le32_to_cpu(ex->ee_block);
2873 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2876 * See if the last block is inside the extent, if so split
2877 * the extent at 'end' block so we can easily remove the
2878 * tail of the first part of the split extent in
2879 * ext4_ext_rm_leaf().
2881 if (end >= ee_block && end < ex_end) {
2884 * If we're going to split the extent, note that
2885 * the cluster containing the block after 'end' is
2886 * in use to avoid freeing it when removing blocks.
2888 if (sbi->s_cluster_ratio > 1) {
2889 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2891 -(long long) EXT4_B2C(sbi, pblk);
2895 * Split the extent in two so that 'end' is the last
2896 * block in the first new extent. Also we should not
2897 * fail removing space due to ENOSPC so try to use
2898 * reserved block if that happens.
2900 err = ext4_force_split_extent_at(handle, inode, &path,
2905 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2907 * If there's an extent to the right its first cluster
2908 * contains the immediate right boundary of the
2909 * truncated/punched region. Set partial_cluster to
2910 * its negative value so it won't be freed if shared
2911 * with the current extent. The end < ee_block case
2912 * is handled in ext4_ext_rm_leaf().
2915 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2921 -(long long) EXT4_B2C(sbi, pblk);
2925 * We start scanning from right side, freeing all the blocks
2926 * after i_size and walking into the tree depth-wise.
2928 depth = ext_depth(inode);
2933 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2935 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2938 ext4_journal_stop(handle);
2941 path[0].p_maxdepth = path[0].p_depth = depth;
2942 path[0].p_hdr = ext_inode_hdr(inode);
2945 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2946 err = -EFSCORRUPTED;
2952 while (i >= 0 && err == 0) {
2954 /* this is leaf block */
2955 err = ext4_ext_rm_leaf(handle, inode, path,
2956 &partial_cluster, start,
2958 /* root level has p_bh == NULL, brelse() eats this */
2959 brelse(path[i].p_bh);
2960 path[i].p_bh = NULL;
2965 /* this is index block */
2966 if (!path[i].p_hdr) {
2967 ext_debug("initialize header\n");
2968 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2971 if (!path[i].p_idx) {
2972 /* this level hasn't been touched yet */
2973 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2974 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2975 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2977 le16_to_cpu(path[i].p_hdr->eh_entries));
2979 /* we were already here, see at next index */
2983 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2984 i, EXT_FIRST_INDEX(path[i].p_hdr),
2986 if (ext4_ext_more_to_rm(path + i)) {
2987 struct buffer_head *bh;
2988 /* go to the next level */
2989 ext_debug("move to level %d (block %llu)\n",
2990 i + 1, ext4_idx_pblock(path[i].p_idx));
2991 memset(path + i + 1, 0, sizeof(*path));
2992 bh = read_extent_tree_block(inode,
2993 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2996 /* should we reset i_size? */
3000 /* Yield here to deal with large extent trees.
3001 * Should be a no-op if we did IO above. */
3003 if (WARN_ON(i + 1 > depth)) {
3004 err = -EFSCORRUPTED;
3007 path[i + 1].p_bh = bh;
3009 /* save actual number of indexes since this
3010 * number is changed at the next iteration */
3011 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3014 /* we finished processing this index, go up */
3015 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3016 /* index is empty, remove it;
3017 * handle must be already prepared by the
3018 * truncatei_leaf() */
3019 err = ext4_ext_rm_idx(handle, inode, path, i);
3021 /* root level has p_bh == NULL, brelse() eats this */
3022 brelse(path[i].p_bh);
3023 path[i].p_bh = NULL;
3025 ext_debug("return to level %d\n", i);
3029 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3030 partial_cluster, path->p_hdr->eh_entries);
3033 * If we still have something in the partial cluster and we have removed
3034 * even the first extent, then we should free the blocks in the partial
3035 * cluster as well. (This code will only run when there are no leaves
3036 * to the immediate left of the truncated/punched region.)
3038 if (partial_cluster > 0 && err == 0) {
3039 /* don't zero partial_cluster since it's not used afterwards */
3040 ext4_free_blocks(handle, inode, NULL,
3041 EXT4_C2B(sbi, partial_cluster),
3042 sbi->s_cluster_ratio,
3043 get_default_free_blocks_flags(inode));
3046 /* TODO: flexible tree reduction should be here */
3047 if (path->p_hdr->eh_entries == 0) {
3049 * truncate to zero freed all the tree,
3050 * so we need to correct eh_depth
3052 err = ext4_ext_get_access(handle, inode, path);
3054 ext_inode_hdr(inode)->eh_depth = 0;
3055 ext_inode_hdr(inode)->eh_max =
3056 cpu_to_le16(ext4_ext_space_root(inode, 0));
3057 err = ext4_ext_dirty(handle, inode, path);
3061 ext4_ext_drop_refs(path);
3066 ext4_journal_stop(handle);
3072 * called at mount time
3074 void ext4_ext_init(struct super_block *sb)
3077 * possible initialization would be here
3080 if (ext4_has_feature_extents(sb)) {
3081 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3082 printk(KERN_INFO "EXT4-fs: file extents enabled"
3083 #ifdef AGGRESSIVE_TEST
3084 ", aggressive tests"
3086 #ifdef CHECK_BINSEARCH
3089 #ifdef EXTENTS_STATS
3094 #ifdef EXTENTS_STATS
3095 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3096 EXT4_SB(sb)->s_ext_min = 1 << 30;
3097 EXT4_SB(sb)->s_ext_max = 0;
3103 * called at umount time
3105 void ext4_ext_release(struct super_block *sb)
3107 if (!ext4_has_feature_extents(sb))
3110 #ifdef EXTENTS_STATS
3111 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3112 struct ext4_sb_info *sbi = EXT4_SB(sb);
3113 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3114 sbi->s_ext_blocks, sbi->s_ext_extents,
3115 sbi->s_ext_blocks / sbi->s_ext_extents);
3116 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3117 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3122 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3124 ext4_lblk_t ee_block;
3125 ext4_fsblk_t ee_pblock;
3126 unsigned int ee_len;
3128 ee_block = le32_to_cpu(ex->ee_block);
3129 ee_len = ext4_ext_get_actual_len(ex);
3130 ee_pblock = ext4_ext_pblock(ex);
3135 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3136 EXTENT_STATUS_WRITTEN);
3139 /* FIXME!! we need to try to merge to left or right after zero-out */
3140 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3142 ext4_fsblk_t ee_pblock;
3143 unsigned int ee_len;
3145 ee_len = ext4_ext_get_actual_len(ex);
3146 ee_pblock = ext4_ext_pblock(ex);
3147 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3152 * ext4_split_extent_at() splits an extent at given block.
3154 * @handle: the journal handle
3155 * @inode: the file inode
3156 * @path: the path to the extent
3157 * @split: the logical block where the extent is splitted.
3158 * @split_flags: indicates if the extent could be zeroout if split fails, and
3159 * the states(init or unwritten) of new extents.
3160 * @flags: flags used to insert new extent to extent tree.
3163 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3164 * of which are deterimined by split_flag.
3166 * There are two cases:
3167 * a> the extent are splitted into two extent.
3168 * b> split is not needed, and just mark the extent.
3170 * return 0 on success.
3172 static int ext4_split_extent_at(handle_t *handle,
3173 struct inode *inode,
3174 struct ext4_ext_path **ppath,
3179 struct ext4_ext_path *path = *ppath;
3180 ext4_fsblk_t newblock;
3181 ext4_lblk_t ee_block;
3182 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3183 struct ext4_extent *ex2 = NULL;
3184 unsigned int ee_len, depth;
3187 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3188 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3190 ext_debug("ext4_split_extents_at: inode %lu, logical"
3191 "block %llu\n", inode->i_ino, (unsigned long long)split);
3193 ext4_ext_show_leaf(inode, path);
3195 depth = ext_depth(inode);
3196 ex = path[depth].p_ext;
3197 ee_block = le32_to_cpu(ex->ee_block);
3198 ee_len = ext4_ext_get_actual_len(ex);
3199 newblock = split - ee_block + ext4_ext_pblock(ex);
3201 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3202 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3203 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3204 EXT4_EXT_MARK_UNWRIT1 |
3205 EXT4_EXT_MARK_UNWRIT2));
3207 err = ext4_ext_get_access(handle, inode, path + depth);
3211 if (split == ee_block) {
3213 * case b: block @split is the block that the extent begins with
3214 * then we just change the state of the extent, and splitting
3217 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3218 ext4_ext_mark_unwritten(ex);
3220 ext4_ext_mark_initialized(ex);
3222 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3223 ext4_ext_try_to_merge(handle, inode, path, ex);
3225 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3230 memcpy(&orig_ex, ex, sizeof(orig_ex));
3231 ex->ee_len = cpu_to_le16(split - ee_block);
3232 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3233 ext4_ext_mark_unwritten(ex);
3236 * path may lead to new leaf, not to original leaf any more
3237 * after ext4_ext_insert_extent() returns,
3239 err = ext4_ext_dirty(handle, inode, path + depth);
3241 goto fix_extent_len;
3244 ex2->ee_block = cpu_to_le32(split);
3245 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3246 ext4_ext_store_pblock(ex2, newblock);
3247 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3248 ext4_ext_mark_unwritten(ex2);
3250 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3251 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3252 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3253 if (split_flag & EXT4_EXT_DATA_VALID1) {
3254 err = ext4_ext_zeroout(inode, ex2);
3255 zero_ex.ee_block = ex2->ee_block;
3256 zero_ex.ee_len = cpu_to_le16(
3257 ext4_ext_get_actual_len(ex2));
3258 ext4_ext_store_pblock(&zero_ex,
3259 ext4_ext_pblock(ex2));
3261 err = ext4_ext_zeroout(inode, ex);
3262 zero_ex.ee_block = ex->ee_block;
3263 zero_ex.ee_len = cpu_to_le16(
3264 ext4_ext_get_actual_len(ex));
3265 ext4_ext_store_pblock(&zero_ex,
3266 ext4_ext_pblock(ex));
3269 err = ext4_ext_zeroout(inode, &orig_ex);
3270 zero_ex.ee_block = orig_ex.ee_block;
3271 zero_ex.ee_len = cpu_to_le16(
3272 ext4_ext_get_actual_len(&orig_ex));
3273 ext4_ext_store_pblock(&zero_ex,
3274 ext4_ext_pblock(&orig_ex));
3278 goto fix_extent_len;
3279 /* update the extent length and mark as initialized */
3280 ex->ee_len = cpu_to_le16(ee_len);
3281 ext4_ext_try_to_merge(handle, inode, path, ex);
3282 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3284 goto fix_extent_len;
3286 /* update extent status tree */
3287 err = ext4_zeroout_es(inode, &zero_ex);
3291 goto fix_extent_len;
3294 ext4_ext_show_leaf(inode, path);
3298 ex->ee_len = orig_ex.ee_len;
3299 ext4_ext_dirty(handle, inode, path + path->p_depth);
3304 * ext4_split_extents() splits an extent and mark extent which is covered
3305 * by @map as split_flags indicates
3307 * It may result in splitting the extent into multiple extents (up to three)
3308 * There are three possibilities:
3309 * a> There is no split required
3310 * b> Splits in two extents: Split is happening at either end of the extent
3311 * c> Splits in three extents: Somone is splitting in middle of the extent
3314 static int ext4_split_extent(handle_t *handle,
3315 struct inode *inode,
3316 struct ext4_ext_path **ppath,
3317 struct ext4_map_blocks *map,
3321 struct ext4_ext_path *path = *ppath;
3322 ext4_lblk_t ee_block;
3323 struct ext4_extent *ex;
3324 unsigned int ee_len, depth;
3327 int split_flag1, flags1;
3328 int allocated = map->m_len;
3330 depth = ext_depth(inode);
3331 ex = path[depth].p_ext;
3332 ee_block = le32_to_cpu(ex->ee_block);
3333 ee_len = ext4_ext_get_actual_len(ex);
3334 unwritten = ext4_ext_is_unwritten(ex);
3336 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3337 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3338 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3340 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3341 EXT4_EXT_MARK_UNWRIT2;
3342 if (split_flag & EXT4_EXT_DATA_VALID2)
3343 split_flag1 |= EXT4_EXT_DATA_VALID1;
3344 err = ext4_split_extent_at(handle, inode, ppath,
3345 map->m_lblk + map->m_len, split_flag1, flags1);
3349 allocated = ee_len - (map->m_lblk - ee_block);
3352 * Update path is required because previous ext4_split_extent_at() may
3353 * result in split of original leaf or extent zeroout.
3355 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3357 return PTR_ERR(path);
3358 depth = ext_depth(inode);
3359 ex = path[depth].p_ext;
3361 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3362 (unsigned long) map->m_lblk);
3363 return -EFSCORRUPTED;
3365 unwritten = ext4_ext_is_unwritten(ex);
3368 if (map->m_lblk >= ee_block) {
3369 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3371 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3372 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3373 EXT4_EXT_MARK_UNWRIT2);
3375 err = ext4_split_extent_at(handle, inode, ppath,
3376 map->m_lblk, split_flag1, flags);
3381 ext4_ext_show_leaf(inode, path);
3383 return err ? err : allocated;
3387 * This function is called by ext4_ext_map_blocks() if someone tries to write
3388 * to an unwritten extent. It may result in splitting the unwritten
3389 * extent into multiple extents (up to three - one initialized and two
3391 * There are three possibilities:
3392 * a> There is no split required: Entire extent should be initialized
3393 * b> Splits in two extents: Write is happening at either end of the extent
3394 * c> Splits in three extents: Somone is writing in middle of the extent
3397 * - The extent pointed to by 'path' is unwritten.
3398 * - The extent pointed to by 'path' contains a superset
3399 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3401 * Post-conditions on success:
3402 * - the returned value is the number of blocks beyond map->l_lblk
3403 * that are allocated and initialized.
3404 * It is guaranteed to be >= map->m_len.
3406 static int ext4_ext_convert_to_initialized(handle_t *handle,
3407 struct inode *inode,
3408 struct ext4_map_blocks *map,
3409 struct ext4_ext_path **ppath,
3412 struct ext4_ext_path *path = *ppath;
3413 struct ext4_sb_info *sbi;
3414 struct ext4_extent_header *eh;
3415 struct ext4_map_blocks split_map;
3416 struct ext4_extent zero_ex1, zero_ex2;
3417 struct ext4_extent *ex, *abut_ex;
3418 ext4_lblk_t ee_block, eof_block;
3419 unsigned int ee_len, depth, map_len = map->m_len;
3420 int allocated = 0, max_zeroout = 0;
3422 int split_flag = EXT4_EXT_DATA_VALID2;
3424 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3425 "block %llu, max_blocks %u\n", inode->i_ino,
3426 (unsigned long long)map->m_lblk, map_len);
3428 sbi = EXT4_SB(inode->i_sb);
3429 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3430 inode->i_sb->s_blocksize_bits;
3431 if (eof_block < map->m_lblk + map_len)
3432 eof_block = map->m_lblk + map_len;
3434 depth = ext_depth(inode);
3435 eh = path[depth].p_hdr;
3436 ex = path[depth].p_ext;
3437 ee_block = le32_to_cpu(ex->ee_block);
3438 ee_len = ext4_ext_get_actual_len(ex);
3439 zero_ex1.ee_len = 0;
3440 zero_ex2.ee_len = 0;
3442 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3444 /* Pre-conditions */
3445 BUG_ON(!ext4_ext_is_unwritten(ex));
3446 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3449 * Attempt to transfer newly initialized blocks from the currently
3450 * unwritten extent to its neighbor. This is much cheaper
3451 * than an insertion followed by a merge as those involve costly
3452 * memmove() calls. Transferring to the left is the common case in
3453 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3454 * followed by append writes.
3456 * Limitations of the current logic:
3457 * - L1: we do not deal with writes covering the whole extent.
3458 * This would require removing the extent if the transfer
3460 * - L2: we only attempt to merge with an extent stored in the
3461 * same extent tree node.
3463 if ((map->m_lblk == ee_block) &&
3464 /* See if we can merge left */
3465 (map_len < ee_len) && /*L1*/
3466 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3467 ext4_lblk_t prev_lblk;
3468 ext4_fsblk_t prev_pblk, ee_pblk;
3469 unsigned int prev_len;
3472 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3473 prev_len = ext4_ext_get_actual_len(abut_ex);
3474 prev_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 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3488 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3489 (prev_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 ex by 'map_len' blocks */
3498 ex->ee_block = cpu_to_le32(ee_block + map_len);
3499 ext4_ext_store_pblock(ex, ee_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(prev_len + map_len);
3506 /* Result: number of initialized blocks past m_lblk */
3507 allocated = map_len;
3509 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3510 (map_len < ee_len) && /*L1*/
3511 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3512 /* See if we can merge right */
3513 ext4_lblk_t next_lblk;
3514 ext4_fsblk_t next_pblk, ee_pblk;
3515 unsigned int next_len;
3518 next_lblk = le32_to_cpu(abut_ex->ee_block);
3519 next_len = ext4_ext_get_actual_len(abut_ex);
3520 next_pblk = ext4_ext_pblock(abut_ex);
3521 ee_pblk = ext4_ext_pblock(ex);
3524 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3525 * upon those conditions:
3526 * - C1: abut_ex is initialized,
3527 * - C2: abut_ex is logically abutting ex,
3528 * - C3: abut_ex is physically abutting ex,
3529 * - C4: abut_ex can receive the additional blocks without
3530 * overflowing the (initialized) length limit.
3532 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3533 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3534 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3535 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3536 err = ext4_ext_get_access(handle, inode, path + depth);
3540 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3543 /* Shift the start of abut_ex by 'map_len' blocks */
3544 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3545 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3546 ex->ee_len = cpu_to_le16(ee_len - map_len);
3547 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3549 /* Extend abut_ex by 'map_len' blocks */
3550 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3552 /* Result: number of initialized blocks past m_lblk */
3553 allocated = map_len;
3557 /* Mark the block containing both extents as dirty */
3558 ext4_ext_dirty(handle, inode, path + depth);
3560 /* Update path to point to the right extent */
3561 path[depth].p_ext = abut_ex;
3564 allocated = ee_len - (map->m_lblk - ee_block);
3566 WARN_ON(map->m_lblk < ee_block);
3568 * It is safe to convert extent to initialized via explicit
3569 * zeroout only if extent is fully inside i_size or new_size.
3571 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3573 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3574 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3575 (inode->i_sb->s_blocksize_bits - 10);
3577 if (ext4_encrypted_inode(inode))
3582 * 1. split the extent into three extents.
3583 * 2. split the extent into two extents, zeroout the head of the first
3585 * 3. split the extent into two extents, zeroout the tail of the second
3587 * 4. split the extent into two extents with out zeroout.
3588 * 5. no splitting needed, just possibly zeroout the head and / or the
3589 * tail of the extent.
3591 split_map.m_lblk = map->m_lblk;
3592 split_map.m_len = map->m_len;
3594 if (max_zeroout && (allocated > split_map.m_len)) {
3595 if (allocated <= max_zeroout) {
3598 cpu_to_le32(split_map.m_lblk +
3601 cpu_to_le16(allocated - split_map.m_len);
3602 ext4_ext_store_pblock(&zero_ex1,
3603 ext4_ext_pblock(ex) + split_map.m_lblk +
3604 split_map.m_len - ee_block);
3605 err = ext4_ext_zeroout(inode, &zero_ex1);
3608 split_map.m_len = allocated;
3610 if (split_map.m_lblk - ee_block + split_map.m_len <
3613 if (split_map.m_lblk != ee_block) {
3614 zero_ex2.ee_block = ex->ee_block;
3615 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3617 ext4_ext_store_pblock(&zero_ex2,
3618 ext4_ext_pblock(ex));
3619 err = ext4_ext_zeroout(inode, &zero_ex2);
3624 split_map.m_len += split_map.m_lblk - ee_block;
3625 split_map.m_lblk = ee_block;
3626 allocated = map->m_len;
3630 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3635 /* If we have gotten a failure, don't zero out status tree */
3637 err = ext4_zeroout_es(inode, &zero_ex1);
3639 err = ext4_zeroout_es(inode, &zero_ex2);
3641 return err ? err : allocated;
3645 * This function is called by ext4_ext_map_blocks() from
3646 * ext4_get_blocks_dio_write() when DIO to write
3647 * to an unwritten extent.
3649 * Writing to an unwritten extent may result in splitting the unwritten
3650 * extent into multiple initialized/unwritten extents (up to three)
3651 * There are three possibilities:
3652 * a> There is no split required: Entire extent should be unwritten
3653 * b> Splits in two extents: Write is happening at either end of the extent
3654 * c> Splits in three extents: Somone is writing in middle of the extent
3656 * This works the same way in the case of initialized -> unwritten conversion.
3658 * One of more index blocks maybe needed if the extent tree grow after
3659 * the unwritten extent split. To prevent ENOSPC occur at the IO
3660 * complete, we need to split the unwritten extent before DIO submit
3661 * the IO. The unwritten extent called at this time will be split
3662 * into three unwritten extent(at most). After IO complete, the part
3663 * being filled will be convert to initialized by the end_io callback function
3664 * via ext4_convert_unwritten_extents().
3666 * Returns the size of unwritten extent to be written on success.
3668 static int ext4_split_convert_extents(handle_t *handle,
3669 struct inode *inode,
3670 struct ext4_map_blocks *map,
3671 struct ext4_ext_path **ppath,
3674 struct ext4_ext_path *path = *ppath;
3675 ext4_lblk_t eof_block;
3676 ext4_lblk_t ee_block;
3677 struct ext4_extent *ex;
3678 unsigned int ee_len;
3679 int split_flag = 0, depth;
3681 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3682 __func__, inode->i_ino,
3683 (unsigned long long)map->m_lblk, map->m_len);
3685 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3686 inode->i_sb->s_blocksize_bits;
3687 if (eof_block < map->m_lblk + map->m_len)
3688 eof_block = map->m_lblk + map->m_len;
3690 * It is safe to convert extent to initialized via explicit
3691 * zeroout only if extent is fully insde i_size or new_size.
3693 depth = ext_depth(inode);
3694 ex = path[depth].p_ext;
3695 ee_block = le32_to_cpu(ex->ee_block);
3696 ee_len = ext4_ext_get_actual_len(ex);
3698 /* Convert to unwritten */
3699 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3700 split_flag |= EXT4_EXT_DATA_VALID1;
3701 /* Convert to initialized */
3702 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3703 split_flag |= ee_block + ee_len <= eof_block ?
3704 EXT4_EXT_MAY_ZEROOUT : 0;
3705 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3707 flags |= EXT4_GET_BLOCKS_PRE_IO;
3708 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3711 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3712 struct inode *inode,
3713 struct ext4_map_blocks *map,
3714 struct ext4_ext_path **ppath)
3716 struct ext4_ext_path *path = *ppath;
3717 struct ext4_extent *ex;
3718 ext4_lblk_t ee_block;
3719 unsigned int ee_len;
3723 depth = ext_depth(inode);
3724 ex = path[depth].p_ext;
3725 ee_block = le32_to_cpu(ex->ee_block);
3726 ee_len = ext4_ext_get_actual_len(ex);
3728 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3729 "block %llu, max_blocks %u\n", inode->i_ino,
3730 (unsigned long long)ee_block, ee_len);
3732 /* If extent is larger than requested it is a clear sign that we still
3733 * have some extent state machine issues left. So extent_split is still
3735 * TODO: Once all related issues will be fixed this situation should be
3738 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3740 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3741 " len %u; IO logical block %llu, len %u",
3742 inode->i_ino, (unsigned long long)ee_block, ee_len,
3743 (unsigned long long)map->m_lblk, map->m_len);
3745 err = ext4_split_convert_extents(handle, inode, map, ppath,
3746 EXT4_GET_BLOCKS_CONVERT);
3749 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3751 return PTR_ERR(path);
3752 depth = ext_depth(inode);
3753 ex = path[depth].p_ext;
3756 err = ext4_ext_get_access(handle, inode, path + depth);
3759 /* first mark the extent as initialized */
3760 ext4_ext_mark_initialized(ex);
3762 /* note: ext4_ext_correct_indexes() isn't needed here because
3763 * borders are not changed
3765 ext4_ext_try_to_merge(handle, inode, path, ex);
3767 /* Mark modified extent as dirty */
3768 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3770 ext4_ext_show_leaf(inode, path);
3775 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3777 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3779 struct ext4_ext_path *path,
3783 struct ext4_extent_header *eh;
3784 struct ext4_extent *last_ex;
3786 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3789 depth = ext_depth(inode);
3790 eh = path[depth].p_hdr;
3793 * We're going to remove EOFBLOCKS_FL entirely in future so we
3794 * do not care for this case anymore. Simply remove the flag
3795 * if there are no extents.
3797 if (unlikely(!eh->eh_entries))
3799 last_ex = EXT_LAST_EXTENT(eh);
3801 * We should clear the EOFBLOCKS_FL flag if we are writing the
3802 * last block in the last extent in the file. We test this by
3803 * first checking to see if the caller to
3804 * ext4_ext_get_blocks() was interested in the last block (or
3805 * a block beyond the last block) in the current extent. If
3806 * this turns out to be false, we can bail out from this
3807 * function immediately.
3809 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3810 ext4_ext_get_actual_len(last_ex))
3813 * If the caller does appear to be planning to write at or
3814 * beyond the end of the current extent, we then test to see
3815 * if the current extent is the last extent in the file, by
3816 * checking to make sure it was reached via the rightmost node
3817 * at each level of the tree.
3819 for (i = depth-1; i >= 0; i--)
3820 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3823 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3824 return ext4_mark_inode_dirty(handle, inode);
3828 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3830 * Return 1 if there is a delalloc block in the range, otherwise 0.
3832 int ext4_find_delalloc_range(struct inode *inode,
3833 ext4_lblk_t lblk_start,
3834 ext4_lblk_t lblk_end)
3836 struct extent_status es;
3838 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3840 return 0; /* there is no delay extent in this tree */
3841 else if (es.es_lblk <= lblk_start &&
3842 lblk_start < es.es_lblk + es.es_len)
3844 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3850 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3852 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3853 ext4_lblk_t lblk_start, lblk_end;
3854 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3855 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3857 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3861 * Determines how many complete clusters (out of those specified by the 'map')
3862 * are under delalloc and were reserved quota for.
3863 * This function is called when we are writing out the blocks that were
3864 * originally written with their allocation delayed, but then the space was
3865 * allocated using fallocate() before the delayed allocation could be resolved.
3866 * The cases to look for are:
3867 * ('=' indicated delayed allocated blocks
3868 * '-' indicates non-delayed allocated blocks)
3869 * (a) partial clusters towards beginning and/or end outside of allocated range
3870 * are not delalloc'ed.
3872 * |----c---=|====c====|====c====|===-c----|
3873 * |++++++ allocated ++++++|
3874 * ==> 4 complete clusters in above example
3876 * (b) partial cluster (outside of allocated range) towards either end is
3877 * marked for delayed allocation. In this case, we will exclude that
3880 * |----====c========|========c========|
3881 * |++++++ allocated ++++++|
3882 * ==> 1 complete clusters in above example
3885 * |================c================|
3886 * |++++++ allocated ++++++|
3887 * ==> 0 complete clusters in above example
3889 * The ext4_da_update_reserve_space will be called only if we
3890 * determine here that there were some "entire" clusters that span
3891 * this 'allocated' range.
3892 * In the non-bigalloc case, this function will just end up returning num_blks
3893 * without ever calling ext4_find_delalloc_range.
3896 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3897 unsigned int num_blks)
3899 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3900 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3901 ext4_lblk_t lblk_from, lblk_to, c_offset;
3902 unsigned int allocated_clusters = 0;
3904 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3905 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3907 /* max possible clusters for this allocation */
3908 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3910 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3912 /* Check towards left side */
3913 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3915 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3916 lblk_to = lblk_from + c_offset - 1;
3918 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3919 allocated_clusters--;
3922 /* Now check towards right. */
3923 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3924 if (allocated_clusters && c_offset) {
3925 lblk_from = lblk_start + num_blks;
3926 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3928 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3929 allocated_clusters--;
3932 return allocated_clusters;
3936 convert_initialized_extent(handle_t *handle, struct inode *inode,
3937 struct ext4_map_blocks *map,
3938 struct ext4_ext_path **ppath,
3939 unsigned int allocated)
3941 struct ext4_ext_path *path = *ppath;
3942 struct ext4_extent *ex;
3943 ext4_lblk_t ee_block;
3944 unsigned int ee_len;
3949 * Make sure that the extent is no bigger than we support with
3952 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3953 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3955 depth = ext_depth(inode);
3956 ex = path[depth].p_ext;
3957 ee_block = le32_to_cpu(ex->ee_block);
3958 ee_len = ext4_ext_get_actual_len(ex);
3960 ext_debug("%s: inode %lu, logical"
3961 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3962 (unsigned long long)ee_block, ee_len);
3964 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3965 err = ext4_split_convert_extents(handle, inode, map, ppath,
3966 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3969 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3971 return PTR_ERR(path);
3972 depth = ext_depth(inode);
3973 ex = path[depth].p_ext;
3975 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3976 (unsigned long) map->m_lblk);
3977 return -EFSCORRUPTED;
3981 err = ext4_ext_get_access(handle, inode, path + depth);
3984 /* first mark the extent as unwritten */
3985 ext4_ext_mark_unwritten(ex);
3987 /* note: ext4_ext_correct_indexes() isn't needed here because
3988 * borders are not changed
3990 ext4_ext_try_to_merge(handle, inode, path, ex);
3992 /* Mark modified extent as dirty */
3993 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3996 ext4_ext_show_leaf(inode, path);
3998 ext4_update_inode_fsync_trans(handle, inode, 1);
3999 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4002 map->m_flags |= EXT4_MAP_UNWRITTEN;
4003 if (allocated > map->m_len)
4004 allocated = map->m_len;
4005 map->m_len = allocated;
4010 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4011 struct ext4_map_blocks *map,
4012 struct ext4_ext_path **ppath, int flags,
4013 unsigned int allocated, ext4_fsblk_t newblock)
4015 struct ext4_ext_path *path = *ppath;
4019 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4020 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4021 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4023 ext4_ext_show_leaf(inode, path);
4026 * When writing into unwritten space, we should not fail to
4027 * allocate metadata blocks for the new extent block if needed.
4029 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4031 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4032 allocated, newblock);
4034 /* get_block() before submit the IO, split the extent */
4035 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4036 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4037 flags | EXT4_GET_BLOCKS_CONVERT);
4040 map->m_flags |= EXT4_MAP_UNWRITTEN;
4043 /* IO end_io complete, convert the filled extent to written */
4044 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4045 if (flags & EXT4_GET_BLOCKS_ZERO) {
4046 if (allocated > map->m_len)
4047 allocated = map->m_len;
4048 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4053 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4056 ext4_update_inode_fsync_trans(handle, inode, 1);
4057 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4061 map->m_flags |= EXT4_MAP_MAPPED;
4062 map->m_pblk = newblock;
4063 if (allocated > map->m_len)
4064 allocated = map->m_len;
4065 map->m_len = allocated;
4068 /* buffered IO case */
4070 * repeat fallocate creation request
4071 * we already have an unwritten extent
4073 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4074 map->m_flags |= EXT4_MAP_UNWRITTEN;
4078 /* buffered READ or buffered write_begin() lookup */
4079 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4081 * We have blocks reserved already. We
4082 * return allocated blocks so that delalloc
4083 * won't do block reservation for us. But
4084 * the buffer head will be unmapped so that
4085 * a read from the block returns 0s.
4087 map->m_flags |= EXT4_MAP_UNWRITTEN;
4091 /* buffered write, writepage time, convert*/
4092 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4094 ext4_update_inode_fsync_trans(handle, inode, 1);
4101 map->m_flags |= EXT4_MAP_NEW;
4103 * if we allocated more blocks than requested
4104 * we need to make sure we unmap the extra block
4105 * allocated. The actual needed block will get
4106 * unmapped later when we find the buffer_head marked
4109 if (allocated > map->m_len) {
4110 clean_bdev_aliases(inode->i_sb->s_bdev, newblock + map->m_len,
4111 allocated - map->m_len);
4112 allocated = map->m_len;
4114 map->m_len = allocated;
4117 * If we have done fallocate with the offset that is already
4118 * delayed allocated, we would have block reservation
4119 * and quota reservation done in the delayed write path.
4120 * But fallocate would have already updated quota and block
4121 * count for this offset. So cancel these reservation
4123 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4124 unsigned int reserved_clusters;
4125 reserved_clusters = get_reserved_cluster_alloc(inode,
4126 map->m_lblk, map->m_len);
4127 if (reserved_clusters)
4128 ext4_da_update_reserve_space(inode,
4134 map->m_flags |= EXT4_MAP_MAPPED;
4135 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4136 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4142 if (allocated > map->m_len)
4143 allocated = map->m_len;
4144 ext4_ext_show_leaf(inode, path);
4145 map->m_pblk = newblock;
4146 map->m_len = allocated;
4148 return err ? err : allocated;
4152 * get_implied_cluster_alloc - check to see if the requested
4153 * allocation (in the map structure) overlaps with a cluster already
4154 * allocated in an extent.
4155 * @sb The filesystem superblock structure
4156 * @map The requested lblk->pblk mapping
4157 * @ex The extent structure which might contain an implied
4158 * cluster allocation
4160 * This function is called by ext4_ext_map_blocks() after we failed to
4161 * find blocks that were already in the inode's extent tree. Hence,
4162 * we know that the beginning of the requested region cannot overlap
4163 * the extent from the inode's extent tree. There are three cases we
4164 * want to catch. The first is this case:
4166 * |--- cluster # N--|
4167 * |--- extent ---| |---- requested region ---|
4170 * The second case that we need to test for is this one:
4172 * |--------- cluster # N ----------------|
4173 * |--- requested region --| |------- extent ----|
4174 * |=======================|
4176 * The third case is when the requested region lies between two extents
4177 * within the same cluster:
4178 * |------------- cluster # N-------------|
4179 * |----- ex -----| |---- ex_right ----|
4180 * |------ requested region ------|
4181 * |================|
4183 * In each of the above cases, we need to set the map->m_pblk and
4184 * map->m_len so it corresponds to the return the extent labelled as
4185 * "|====|" from cluster #N, since it is already in use for data in
4186 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4187 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4188 * as a new "allocated" block region. Otherwise, we will return 0 and
4189 * ext4_ext_map_blocks() will then allocate one or more new clusters
4190 * by calling ext4_mb_new_blocks().
4192 static int get_implied_cluster_alloc(struct super_block *sb,
4193 struct ext4_map_blocks *map,
4194 struct ext4_extent *ex,
4195 struct ext4_ext_path *path)
4197 struct ext4_sb_info *sbi = EXT4_SB(sb);
4198 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4199 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4200 ext4_lblk_t rr_cluster_start;
4201 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4202 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4203 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4205 /* The extent passed in that we are trying to match */
4206 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4207 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4209 /* The requested region passed into ext4_map_blocks() */
4210 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4212 if ((rr_cluster_start == ex_cluster_end) ||
4213 (rr_cluster_start == ex_cluster_start)) {
4214 if (rr_cluster_start == ex_cluster_end)
4215 ee_start += ee_len - 1;
4216 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4217 map->m_len = min(map->m_len,
4218 (unsigned) sbi->s_cluster_ratio - c_offset);
4220 * Check for and handle this case:
4222 * |--------- cluster # N-------------|
4223 * |------- extent ----|
4224 * |--- requested region ---|
4228 if (map->m_lblk < ee_block)
4229 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4232 * Check for the case where there is already another allocated
4233 * block to the right of 'ex' but before the end of the cluster.
4235 * |------------- cluster # N-------------|
4236 * |----- ex -----| |---- ex_right ----|
4237 * |------ requested region ------|
4238 * |================|
4240 if (map->m_lblk > ee_block) {
4241 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4242 map->m_len = min(map->m_len, next - map->m_lblk);
4245 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4249 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4255 * Block allocation/map/preallocation routine for extents based files
4258 * Need to be called with
4259 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4260 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4262 * return > 0, number of of blocks already mapped/allocated
4263 * if create == 0 and these are pre-allocated blocks
4264 * buffer head is unmapped
4265 * otherwise blocks are mapped
4267 * return = 0, if plain look up failed (blocks have not been allocated)
4268 * buffer head is unmapped
4270 * return < 0, error case.
4272 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4273 struct ext4_map_blocks *map, int flags)
4275 struct ext4_ext_path *path = NULL;
4276 struct ext4_extent newex, *ex, *ex2;
4277 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4278 ext4_fsblk_t newblock = 0;
4279 int free_on_err = 0, err = 0, depth, ret;
4280 unsigned int allocated = 0, offset = 0;
4281 unsigned int allocated_clusters = 0;
4282 struct ext4_allocation_request ar;
4283 ext4_lblk_t cluster_offset;
4284 bool map_from_cluster = false;
4286 ext_debug("blocks %u/%u requested for inode %lu\n",
4287 map->m_lblk, map->m_len, inode->i_ino);
4288 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4290 /* find extent for this block */
4291 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4293 err = PTR_ERR(path);
4298 depth = ext_depth(inode);
4301 * consistent leaf must not be empty;
4302 * this situation is possible, though, _during_ tree modification;
4303 * this is why assert can't be put in ext4_find_extent()
4305 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4306 EXT4_ERROR_INODE(inode, "bad extent address "
4307 "lblock: %lu, depth: %d pblock %lld",
4308 (unsigned long) map->m_lblk, depth,
4309 path[depth].p_block);
4310 err = -EFSCORRUPTED;
4314 ex = path[depth].p_ext;
4316 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4317 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4318 unsigned short ee_len;
4322 * unwritten extents are treated as holes, except that
4323 * we split out initialized portions during a write.
4325 ee_len = ext4_ext_get_actual_len(ex);
4327 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4329 /* if found extent covers block, simply return it */
4330 if (in_range(map->m_lblk, ee_block, ee_len)) {
4331 newblock = map->m_lblk - ee_block + ee_start;
4332 /* number of remaining blocks in the extent */
4333 allocated = ee_len - (map->m_lblk - ee_block);
4334 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4335 ee_block, ee_len, newblock);
4338 * If the extent is initialized check whether the
4339 * caller wants to convert it to unwritten.
4341 if ((!ext4_ext_is_unwritten(ex)) &&
4342 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4343 allocated = convert_initialized_extent(
4344 handle, inode, map, &path,
4347 } else if (!ext4_ext_is_unwritten(ex))
4350 ret = ext4_ext_handle_unwritten_extents(
4351 handle, inode, map, &path, flags,
4352 allocated, newblock);
4362 * requested block isn't allocated yet;
4363 * we couldn't try to create block if create flag is zero
4365 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4366 ext4_lblk_t hole_start, hole_len;
4368 hole_start = map->m_lblk;
4369 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4371 * put just found gap into cache to speed up
4372 * subsequent requests
4374 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4376 /* Update hole_len to reflect hole size after map->m_lblk */
4377 if (hole_start != map->m_lblk)
4378 hole_len -= map->m_lblk - hole_start;
4380 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4386 * Okay, we need to do block allocation.
4388 newex.ee_block = cpu_to_le32(map->m_lblk);
4389 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4392 * If we are doing bigalloc, check to see if the extent returned
4393 * by ext4_find_extent() implies a cluster we can use.
4395 if (cluster_offset && ex &&
4396 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4397 ar.len = allocated = map->m_len;
4398 newblock = map->m_pblk;
4399 map_from_cluster = true;
4400 goto got_allocated_blocks;
4403 /* find neighbour allocated blocks */
4404 ar.lleft = map->m_lblk;
4405 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4408 ar.lright = map->m_lblk;
4410 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4414 /* Check if the extent after searching to the right implies a
4415 * cluster we can use. */
4416 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4417 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4418 ar.len = allocated = map->m_len;
4419 newblock = map->m_pblk;
4420 map_from_cluster = true;
4421 goto got_allocated_blocks;
4425 * See if request is beyond maximum number of blocks we can have in
4426 * a single extent. For an initialized extent this limit is
4427 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4428 * EXT_UNWRITTEN_MAX_LEN.
4430 if (map->m_len > EXT_INIT_MAX_LEN &&
4431 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4432 map->m_len = EXT_INIT_MAX_LEN;
4433 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4434 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4435 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4437 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4438 newex.ee_len = cpu_to_le16(map->m_len);
4439 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4441 allocated = ext4_ext_get_actual_len(&newex);
4443 allocated = map->m_len;
4445 /* allocate new block */
4447 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4448 ar.logical = map->m_lblk;
4450 * We calculate the offset from the beginning of the cluster
4451 * for the logical block number, since when we allocate a
4452 * physical cluster, the physical block should start at the
4453 * same offset from the beginning of the cluster. This is
4454 * needed so that future calls to get_implied_cluster_alloc()
4457 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4458 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4460 ar.logical -= offset;
4461 if (S_ISREG(inode->i_mode))
4462 ar.flags = EXT4_MB_HINT_DATA;
4464 /* disable in-core preallocation for non-regular files */
4466 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4467 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4468 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4469 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4470 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4471 ar.flags |= EXT4_MB_USE_RESERVED;
4472 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4475 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4476 ar.goal, newblock, allocated);
4478 allocated_clusters = ar.len;
4479 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4480 if (ar.len > allocated)
4483 got_allocated_blocks:
4484 /* try to insert new extent into found leaf and return */
4485 ext4_ext_store_pblock(&newex, newblock + offset);
4486 newex.ee_len = cpu_to_le16(ar.len);
4487 /* Mark unwritten */
4488 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4489 ext4_ext_mark_unwritten(&newex);
4490 map->m_flags |= EXT4_MAP_UNWRITTEN;
4494 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4495 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4498 err = ext4_ext_insert_extent(handle, inode, &path,
4501 if (err && free_on_err) {
4502 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4503 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4504 /* free data blocks we just allocated */
4505 /* not a good idea to call discard here directly,
4506 * but otherwise we'd need to call it every free() */
4507 ext4_discard_preallocations(inode);
4508 ext4_free_blocks(handle, inode, NULL, newblock,
4509 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4513 /* previous routine could use block we allocated */
4514 newblock = ext4_ext_pblock(&newex);
4515 allocated = ext4_ext_get_actual_len(&newex);
4516 if (allocated > map->m_len)
4517 allocated = map->m_len;
4518 map->m_flags |= EXT4_MAP_NEW;
4521 * Update reserved blocks/metadata blocks after successful
4522 * block allocation which had been deferred till now.
4524 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4525 unsigned int reserved_clusters;
4527 * Check how many clusters we had reserved this allocated range
4529 reserved_clusters = get_reserved_cluster_alloc(inode,
4530 map->m_lblk, allocated);
4531 if (!map_from_cluster) {
4532 BUG_ON(allocated_clusters < reserved_clusters);
4533 if (reserved_clusters < allocated_clusters) {
4534 struct ext4_inode_info *ei = EXT4_I(inode);
4535 int reservation = allocated_clusters -
4538 * It seems we claimed few clusters outside of
4539 * the range of this allocation. We should give
4540 * it back to the reservation pool. This can
4541 * happen in the following case:
4543 * * Suppose s_cluster_ratio is 4 (i.e., each
4544 * cluster has 4 blocks. Thus, the clusters
4545 * are [0-3],[4-7],[8-11]...
4546 * * First comes delayed allocation write for
4547 * logical blocks 10 & 11. Since there were no
4548 * previous delayed allocated blocks in the
4549 * range [8-11], we would reserve 1 cluster
4551 * * Next comes write for logical blocks 3 to 8.
4552 * In this case, we will reserve 2 clusters
4553 * (for [0-3] and [4-7]; and not for [8-11] as
4554 * that range has a delayed allocated blocks.
4555 * Thus total reserved clusters now becomes 3.
4556 * * Now, during the delayed allocation writeout
4557 * time, we will first write blocks [3-8] and
4558 * allocate 3 clusters for writing these
4559 * blocks. Also, we would claim all these
4560 * three clusters above.
4561 * * Now when we come here to writeout the
4562 * blocks [10-11], we would expect to claim
4563 * the reservation of 1 cluster we had made
4564 * (and we would claim it since there are no
4565 * more delayed allocated blocks in the range
4566 * [8-11]. But our reserved cluster count had
4567 * already gone to 0.
4569 * Thus, at the step 4 above when we determine
4570 * that there are still some unwritten delayed
4571 * allocated blocks outside of our current
4572 * block range, we should increment the
4573 * reserved clusters count so that when the
4574 * remaining blocks finally gets written, we
4577 dquot_reserve_block(inode,
4578 EXT4_C2B(sbi, reservation));
4579 spin_lock(&ei->i_block_reservation_lock);
4580 ei->i_reserved_data_blocks += reservation;
4581 spin_unlock(&ei->i_block_reservation_lock);
4584 * We will claim quota for all newly allocated blocks.
4585 * We're updating the reserved space *after* the
4586 * correction above so we do not accidentally free
4587 * all the metadata reservation because we might
4588 * actually need it later on.
4590 ext4_da_update_reserve_space(inode, allocated_clusters,
4596 * Cache the extent and update transaction to commit on fdatasync only
4597 * when it is _not_ an unwritten extent.
4599 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4600 ext4_update_inode_fsync_trans(handle, inode, 1);
4602 ext4_update_inode_fsync_trans(handle, inode, 0);
4604 if (allocated > map->m_len)
4605 allocated = map->m_len;
4606 ext4_ext_show_leaf(inode, path);
4607 map->m_flags |= EXT4_MAP_MAPPED;
4608 map->m_pblk = newblock;
4609 map->m_len = allocated;
4611 ext4_ext_drop_refs(path);
4614 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4615 err ? err : allocated);
4616 return err ? err : allocated;
4619 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4621 struct super_block *sb = inode->i_sb;
4622 ext4_lblk_t last_block;
4626 * TODO: optimization is possible here.
4627 * Probably we need not scan at all,
4628 * because page truncation is enough.
4631 /* we have to know where to truncate from in crash case */
4632 EXT4_I(inode)->i_disksize = inode->i_size;
4633 err = ext4_mark_inode_dirty(handle, inode);
4637 last_block = (inode->i_size + sb->s_blocksize - 1)
4638 >> EXT4_BLOCK_SIZE_BITS(sb);
4640 err = ext4_es_remove_extent(inode, last_block,
4641 EXT_MAX_BLOCKS - last_block);
4642 if (err == -ENOMEM) {
4644 congestion_wait(BLK_RW_ASYNC, HZ/50);
4649 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4652 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4653 ext4_lblk_t len, loff_t new_size,
4654 int flags, int mode)
4656 struct inode *inode = file_inode(file);
4662 struct ext4_map_blocks map;
4663 unsigned int credits;
4666 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4667 map.m_lblk = offset;
4670 * Don't normalize the request if it can fit in one extent so
4671 * that it doesn't get unnecessarily split into multiple
4674 if (len <= EXT_UNWRITTEN_MAX_LEN)
4675 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4678 * credits to insert 1 extent into extent tree
4680 credits = ext4_chunk_trans_blocks(inode, len);
4681 depth = ext_depth(inode);
4684 while (ret >= 0 && len) {
4686 * Recalculate credits when extent tree depth changes.
4688 if (depth != ext_depth(inode)) {
4689 credits = ext4_chunk_trans_blocks(inode, len);
4690 depth = ext_depth(inode);
4693 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4695 if (IS_ERR(handle)) {
4696 ret = PTR_ERR(handle);
4699 ret = ext4_map_blocks(handle, inode, &map, flags);
4701 ext4_debug("inode #%lu: block %u: len %u: "
4702 "ext4_ext_map_blocks returned %d",
4703 inode->i_ino, map.m_lblk,
4705 ext4_mark_inode_dirty(handle, inode);
4706 ret2 = ext4_journal_stop(handle);
4710 map.m_len = len = len - ret;
4711 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4712 inode->i_ctime = current_time(inode);
4714 if (epos > new_size)
4716 if (ext4_update_inode_size(inode, epos) & 0x1)
4717 inode->i_mtime = inode->i_ctime;
4719 if (epos > inode->i_size)
4720 ext4_set_inode_flag(inode,
4721 EXT4_INODE_EOFBLOCKS);
4723 ext4_mark_inode_dirty(handle, inode);
4724 ret2 = ext4_journal_stop(handle);
4728 if (ret == -ENOSPC &&
4729 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4734 return ret > 0 ? ret2 : ret;
4737 static long ext4_zero_range(struct file *file, loff_t offset,
4738 loff_t len, int mode)
4740 struct inode *inode = file_inode(file);
4741 handle_t *handle = NULL;
4742 unsigned int max_blocks;
4743 loff_t new_size = 0;
4747 int partial_begin, partial_end;
4750 unsigned int blkbits = inode->i_blkbits;
4752 trace_ext4_zero_range(inode, offset, len, mode);
4754 if (!S_ISREG(inode->i_mode))
4757 /* Call ext4_force_commit to flush all data in case of data=journal. */
4758 if (ext4_should_journal_data(inode)) {
4759 ret = ext4_force_commit(inode->i_sb);
4765 * Round up offset. This is not fallocate, we neet to zero out
4766 * blocks, so convert interior block aligned part of the range to
4767 * unwritten and possibly manually zero out unaligned parts of the
4770 start = round_up(offset, 1 << blkbits);
4771 end = round_down((offset + len), 1 << blkbits);
4773 if (start < offset || end > offset + len)
4775 partial_begin = offset & ((1 << blkbits) - 1);
4776 partial_end = (offset + len) & ((1 << blkbits) - 1);
4778 lblk = start >> blkbits;
4779 max_blocks = (end >> blkbits);
4780 if (max_blocks < lblk)
4788 * Indirect files do not support unwritten extnets
4790 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4795 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4796 offset + len > i_size_read(inode)) {
4797 new_size = offset + len;
4798 ret = inode_newsize_ok(inode, new_size);
4803 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4804 if (mode & FALLOC_FL_KEEP_SIZE)
4805 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4807 /* Wait all existing dio workers, newcomers will block on i_mutex */
4808 ext4_inode_block_unlocked_dio(inode);
4809 inode_dio_wait(inode);
4811 /* Preallocate the range including the unaligned edges */
4812 if (partial_begin || partial_end) {
4813 ret = ext4_alloc_file_blocks(file,
4814 round_down(offset, 1 << blkbits) >> blkbits,
4815 (round_up((offset + len), 1 << blkbits) -
4816 round_down(offset, 1 << blkbits)) >> blkbits,
4817 new_size, flags, mode);
4823 /* Zero range excluding the unaligned edges */
4824 if (max_blocks > 0) {
4825 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4829 * Prevent page faults from reinstantiating pages we have
4830 * released from page cache.
4832 down_write(&EXT4_I(inode)->i_mmap_sem);
4833 ret = ext4_update_disksize_before_punch(inode, offset, len);
4835 up_write(&EXT4_I(inode)->i_mmap_sem);
4838 /* Now release the pages and zero block aligned part of pages */
4839 truncate_pagecache_range(inode, start, end - 1);
4840 inode->i_mtime = inode->i_ctime = current_time(inode);
4842 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4844 up_write(&EXT4_I(inode)->i_mmap_sem);
4848 if (!partial_begin && !partial_end)
4852 * In worst case we have to writeout two nonadjacent unwritten
4853 * blocks and update the inode
4855 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4856 if (ext4_should_journal_data(inode))
4858 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4859 if (IS_ERR(handle)) {
4860 ret = PTR_ERR(handle);
4861 ext4_std_error(inode->i_sb, ret);
4865 inode->i_mtime = inode->i_ctime = current_time(inode);
4867 ext4_update_inode_size(inode, new_size);
4870 * Mark that we allocate beyond EOF so the subsequent truncate
4871 * can proceed even if the new size is the same as i_size.
4873 if ((offset + len) > i_size_read(inode))
4874 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4876 ext4_mark_inode_dirty(handle, inode);
4878 /* Zero out partial block at the edges of the range */
4879 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4881 ext4_update_inode_fsync_trans(handle, inode, 1);
4883 if (file->f_flags & O_SYNC)
4884 ext4_handle_sync(handle);
4886 ext4_journal_stop(handle);
4888 ext4_inode_resume_unlocked_dio(inode);
4890 inode_unlock(inode);
4895 * preallocate space for a file. This implements ext4's fallocate file
4896 * operation, which gets called from sys_fallocate system call.
4897 * For block-mapped files, posix_fallocate should fall back to the method
4898 * of writing zeroes to the required new blocks (the same behavior which is
4899 * expected for file systems which do not support fallocate() system call).
4901 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4903 struct inode *inode = file_inode(file);
4904 loff_t new_size = 0;
4905 unsigned int max_blocks;
4909 unsigned int blkbits = inode->i_blkbits;
4912 * Encrypted inodes can't handle collapse range or insert
4913 * range since we would need to re-encrypt blocks with a
4914 * different IV or XTS tweak (which are based on the logical
4917 * XXX It's not clear why zero range isn't working, but we'll
4918 * leave it disabled for encrypted inodes for now. This is a
4919 * bug we should fix....
4921 if (ext4_encrypted_inode(inode) &&
4922 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4923 FALLOC_FL_ZERO_RANGE)))
4926 /* Return error if mode is not supported */
4927 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4928 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4929 FALLOC_FL_INSERT_RANGE))
4932 if (mode & FALLOC_FL_PUNCH_HOLE)
4933 return ext4_punch_hole(inode, offset, len);
4935 ret = ext4_convert_inline_data(inode);
4939 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4940 return ext4_collapse_range(inode, offset, len);
4942 if (mode & FALLOC_FL_INSERT_RANGE)
4943 return ext4_insert_range(inode, offset, len);
4945 if (mode & FALLOC_FL_ZERO_RANGE)
4946 return ext4_zero_range(file, offset, len, mode);
4948 trace_ext4_fallocate_enter(inode, offset, len, mode);
4949 lblk = offset >> blkbits;
4951 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4952 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4953 if (mode & FALLOC_FL_KEEP_SIZE)
4954 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4959 * We only support preallocation for extent-based files only
4961 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4966 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4967 offset + len > i_size_read(inode)) {
4968 new_size = offset + len;
4969 ret = inode_newsize_ok(inode, new_size);
4974 /* Wait all existing dio workers, newcomers will block on i_mutex */
4975 ext4_inode_block_unlocked_dio(inode);
4976 inode_dio_wait(inode);
4978 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4980 ext4_inode_resume_unlocked_dio(inode);
4984 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4985 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4986 EXT4_I(inode)->i_sync_tid);
4989 inode_unlock(inode);
4990 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4995 * This function convert a range of blocks to written extents
4996 * The caller of this function will pass the start offset and the size.
4997 * all unwritten extents within this range will be converted to
5000 * This function is called from the direct IO end io call back
5001 * function, to convert the fallocated extents after IO is completed.
5002 * Returns 0 on success.
5004 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5005 loff_t offset, ssize_t len)
5007 unsigned int max_blocks;
5010 struct ext4_map_blocks map;
5011 unsigned int credits, blkbits = inode->i_blkbits;
5013 map.m_lblk = offset >> blkbits;
5014 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
5017 * This is somewhat ugly but the idea is clear: When transaction is
5018 * reserved, everything goes into it. Otherwise we rather start several
5019 * smaller transactions for conversion of each extent separately.
5022 handle = ext4_journal_start_reserved(handle,
5023 EXT4_HT_EXT_CONVERT);
5025 return PTR_ERR(handle);
5029 * credits to insert 1 extent into extent tree
5031 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5033 while (ret >= 0 && ret < max_blocks) {
5035 map.m_len = (max_blocks -= ret);
5037 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5039 if (IS_ERR(handle)) {
5040 ret = PTR_ERR(handle);
5044 ret = ext4_map_blocks(handle, inode, &map,
5045 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5047 ext4_warning(inode->i_sb,
5048 "inode #%lu: block %u: len %u: "
5049 "ext4_ext_map_blocks returned %d",
5050 inode->i_ino, map.m_lblk,
5052 ext4_mark_inode_dirty(handle, inode);
5054 ret2 = ext4_journal_stop(handle);
5055 if (ret <= 0 || ret2)
5059 ret2 = ext4_journal_stop(handle);
5060 return ret > 0 ? ret2 : ret;
5064 * If newes is not existing extent (newes->ec_pblk equals zero) find
5065 * delayed extent at start of newes and update newes accordingly and
5066 * return start of the next delayed extent.
5068 * If newes is existing extent (newes->ec_pblk is not equal zero)
5069 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5070 * extent found. Leave newes unmodified.
5072 static int ext4_find_delayed_extent(struct inode *inode,
5073 struct extent_status *newes)
5075 struct extent_status es;
5076 ext4_lblk_t block, next_del;
5078 if (newes->es_pblk == 0) {
5079 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5080 newes->es_lblk + newes->es_len - 1, &es);
5083 * No extent in extent-tree contains block @newes->es_pblk,
5084 * then the block may stay in 1)a hole or 2)delayed-extent.
5090 if (es.es_lblk > newes->es_lblk) {
5092 newes->es_len = min(es.es_lblk - newes->es_lblk,
5097 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5100 block = newes->es_lblk + newes->es_len;
5101 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5103 next_del = EXT_MAX_BLOCKS;
5105 next_del = es.es_lblk;
5109 /* fiemap flags we can handle specified here */
5110 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5112 static int ext4_xattr_fiemap(struct inode *inode,
5113 struct fiemap_extent_info *fieinfo)
5117 __u32 flags = FIEMAP_EXTENT_LAST;
5118 int blockbits = inode->i_sb->s_blocksize_bits;
5122 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5123 struct ext4_iloc iloc;
5124 int offset; /* offset of xattr in inode */
5126 error = ext4_get_inode_loc(inode, &iloc);
5129 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5130 offset = EXT4_GOOD_OLD_INODE_SIZE +
5131 EXT4_I(inode)->i_extra_isize;
5133 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5134 flags |= FIEMAP_EXTENT_DATA_INLINE;
5136 } else { /* external block */
5137 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5138 length = inode->i_sb->s_blocksize;
5142 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5144 return (error < 0 ? error : 0);
5147 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5148 __u64 start, __u64 len)
5150 ext4_lblk_t start_blk;
5153 if (ext4_has_inline_data(inode)) {
5156 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5163 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5164 error = ext4_ext_precache(inode);
5169 /* fallback to generic here if not in extents fmt */
5170 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5171 return generic_block_fiemap(inode, fieinfo, start, len,
5174 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5177 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5178 error = ext4_xattr_fiemap(inode, fieinfo);
5180 ext4_lblk_t len_blks;
5183 start_blk = start >> inode->i_sb->s_blocksize_bits;
5184 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5185 if (last_blk >= EXT_MAX_BLOCKS)
5186 last_blk = EXT_MAX_BLOCKS-1;
5187 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5190 * Walk the extent tree gathering extent information
5191 * and pushing extents back to the user.
5193 error = ext4_fill_fiemap_extents(inode, start_blk,
5201 * Function to access the path buffer for marking it dirty.
5202 * It also checks if there are sufficient credits left in the journal handle
5206 ext4_access_path(handle_t *handle, struct inode *inode,
5207 struct ext4_ext_path *path)
5211 if (!ext4_handle_valid(handle))
5215 * Check if need to extend journal credits
5216 * 3 for leaf, sb, and inode plus 2 (bmap and group
5217 * descriptor) for each block group; assume two block
5220 if (handle->h_buffer_credits < 7) {
5221 credits = ext4_writepage_trans_blocks(inode);
5222 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5223 /* EAGAIN is success */
5224 if (err && err != -EAGAIN)
5228 err = ext4_ext_get_access(handle, inode, path);
5233 * ext4_ext_shift_path_extents:
5234 * Shift the extents of a path structure lying between path[depth].p_ext
5235 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5236 * if it is right shift or left shift operation.
5239 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5240 struct inode *inode, handle_t *handle,
5241 enum SHIFT_DIRECTION SHIFT)
5244 struct ext4_extent *ex_start, *ex_last;
5246 depth = path->p_depth;
5248 while (depth >= 0) {
5249 if (depth == path->p_depth) {
5250 ex_start = path[depth].p_ext;
5252 return -EFSCORRUPTED;
5254 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5256 err = ext4_access_path(handle, inode, path + depth);
5260 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5263 while (ex_start <= ex_last) {
5264 if (SHIFT == SHIFT_LEFT) {
5265 le32_add_cpu(&ex_start->ee_block,
5267 /* Try to merge to the left. */
5269 EXT_FIRST_EXTENT(path[depth].p_hdr))
5271 ext4_ext_try_to_merge_right(inode,
5272 path, ex_start - 1))
5277 le32_add_cpu(&ex_last->ee_block, shift);
5278 ext4_ext_try_to_merge_right(inode, path,
5283 err = ext4_ext_dirty(handle, inode, path + depth);
5287 if (--depth < 0 || !update)
5291 /* Update index too */
5292 err = ext4_access_path(handle, inode, path + depth);
5296 if (SHIFT == SHIFT_LEFT)
5297 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5299 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5300 err = ext4_ext_dirty(handle, inode, path + depth);
5304 /* we are done if current index is not a starting index */
5305 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5316 * ext4_ext_shift_extents:
5317 * All the extents which lies in the range from @start to the last allocated
5318 * block for the @inode are shifted either towards left or right (depending
5319 * upon @SHIFT) by @shift blocks.
5320 * On success, 0 is returned, error otherwise.
5323 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5324 ext4_lblk_t start, ext4_lblk_t shift,
5325 enum SHIFT_DIRECTION SHIFT)
5327 struct ext4_ext_path *path;
5329 struct ext4_extent *extent;
5330 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5332 /* Let path point to the last extent */
5333 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5336 return PTR_ERR(path);
5338 depth = path->p_depth;
5339 extent = path[depth].p_ext;
5343 stop = le32_to_cpu(extent->ee_block);
5346 * In case of left shift, Don't start shifting extents until we make
5347 * sure the hole is big enough to accommodate the shift.
5349 if (SHIFT == SHIFT_LEFT) {
5350 path = ext4_find_extent(inode, start - 1, &path,
5353 return PTR_ERR(path);
5354 depth = path->p_depth;
5355 extent = path[depth].p_ext;
5357 ex_start = le32_to_cpu(extent->ee_block);
5358 ex_end = le32_to_cpu(extent->ee_block) +
5359 ext4_ext_get_actual_len(extent);
5365 if ((start == ex_start && shift > ex_start) ||
5366 (shift > start - ex_end)) {
5367 ext4_ext_drop_refs(path);
5374 * In case of left shift, iterator points to start and it is increased
5375 * till we reach stop. In case of right shift, iterator points to stop
5376 * and it is decreased till we reach start.
5378 if (SHIFT == SHIFT_LEFT)
5384 * Its safe to start updating extents. Start and stop are unsigned, so
5385 * in case of right shift if extent with 0 block is reached, iterator
5386 * becomes NULL to indicate the end of the loop.
5388 while (iterator && start <= stop) {
5389 path = ext4_find_extent(inode, *iterator, &path,
5392 return PTR_ERR(path);
5393 depth = path->p_depth;
5394 extent = path[depth].p_ext;
5396 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5397 (unsigned long) *iterator);
5398 return -EFSCORRUPTED;
5400 if (SHIFT == SHIFT_LEFT && *iterator >
5401 le32_to_cpu(extent->ee_block)) {
5402 /* Hole, move to the next extent */
5403 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5404 path[depth].p_ext++;
5406 *iterator = ext4_ext_next_allocated_block(path);
5411 if (SHIFT == SHIFT_LEFT) {
5412 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5413 *iterator = le32_to_cpu(extent->ee_block) +
5414 ext4_ext_get_actual_len(extent);
5416 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5417 if (le32_to_cpu(extent->ee_block) > 0)
5418 *iterator = le32_to_cpu(extent->ee_block) - 1;
5420 /* Beginning is reached, end of the loop */
5422 /* Update path extent in case we need to stop */
5423 while (le32_to_cpu(extent->ee_block) < start)
5425 path[depth].p_ext = extent;
5427 ret = ext4_ext_shift_path_extents(path, shift, inode,
5433 ext4_ext_drop_refs(path);
5439 * ext4_collapse_range:
5440 * This implements the fallocate's collapse range functionality for ext4
5441 * Returns: 0 and non-zero on error.
5443 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5445 struct super_block *sb = inode->i_sb;
5446 ext4_lblk_t punch_start, punch_stop;
5448 unsigned int credits;
5449 loff_t new_size, ioffset;
5453 * We need to test this early because xfstests assumes that a
5454 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5455 * system does not support collapse range.
5457 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5460 /* Collapse range works only on fs block size aligned offsets. */
5461 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5462 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5465 if (!S_ISREG(inode->i_mode))
5468 trace_ext4_collapse_range(inode, offset, len);
5470 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5471 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5473 /* Call ext4_force_commit to flush all data in case of data=journal. */
5474 if (ext4_should_journal_data(inode)) {
5475 ret = ext4_force_commit(inode->i_sb);
5482 * There is no need to overlap collapse range with EOF, in which case
5483 * it is effectively a truncate operation
5485 if (offset + len >= i_size_read(inode)) {
5490 /* Currently just for extent based files */
5491 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5496 /* Wait for existing dio to complete */
5497 ext4_inode_block_unlocked_dio(inode);
5498 inode_dio_wait(inode);
5501 * Prevent page faults from reinstantiating pages we have released from
5504 down_write(&EXT4_I(inode)->i_mmap_sem);
5506 * Need to round down offset to be aligned with page size boundary
5507 * for page size > block size.
5509 ioffset = round_down(offset, PAGE_SIZE);
5511 * Write tail of the last page before removed range since it will get
5512 * removed from the page cache below.
5514 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5518 * Write data that will be shifted to preserve them when discarding
5519 * page cache below. We are also protected from pages becoming dirty
5522 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5526 truncate_pagecache(inode, ioffset);
5528 credits = ext4_writepage_trans_blocks(inode);
5529 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5530 if (IS_ERR(handle)) {
5531 ret = PTR_ERR(handle);
5535 down_write(&EXT4_I(inode)->i_data_sem);
5536 ext4_discard_preallocations(inode);
5538 ret = ext4_es_remove_extent(inode, punch_start,
5539 EXT_MAX_BLOCKS - punch_start);
5541 up_write(&EXT4_I(inode)->i_data_sem);
5545 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5547 up_write(&EXT4_I(inode)->i_data_sem);
5550 ext4_discard_preallocations(inode);
5552 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5553 punch_stop - punch_start, SHIFT_LEFT);
5555 up_write(&EXT4_I(inode)->i_data_sem);
5559 new_size = i_size_read(inode) - len;
5560 i_size_write(inode, new_size);
5561 EXT4_I(inode)->i_disksize = new_size;
5563 up_write(&EXT4_I(inode)->i_data_sem);
5565 ext4_handle_sync(handle);
5566 inode->i_mtime = inode->i_ctime = current_time(inode);
5567 ext4_mark_inode_dirty(handle, inode);
5568 ext4_update_inode_fsync_trans(handle, inode, 1);
5571 ext4_journal_stop(handle);
5573 up_write(&EXT4_I(inode)->i_mmap_sem);
5574 ext4_inode_resume_unlocked_dio(inode);
5576 inode_unlock(inode);
5581 * ext4_insert_range:
5582 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5583 * The data blocks starting from @offset to the EOF are shifted by @len
5584 * towards right to create a hole in the @inode. Inode size is increased
5586 * Returns 0 on success, error otherwise.
5588 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5590 struct super_block *sb = inode->i_sb;
5592 struct ext4_ext_path *path;
5593 struct ext4_extent *extent;
5594 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5595 unsigned int credits, ee_len;
5596 int ret = 0, depth, split_flag = 0;
5600 * We need to test this early because xfstests assumes that an
5601 * insert range of (0, 1) will return EOPNOTSUPP if the file
5602 * system does not support insert range.
5604 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5607 /* Insert range works only on fs block size aligned offsets. */
5608 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5609 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5612 if (!S_ISREG(inode->i_mode))
5615 trace_ext4_insert_range(inode, offset, len);
5617 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5618 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5620 /* Call ext4_force_commit to flush all data in case of data=journal */
5621 if (ext4_should_journal_data(inode)) {
5622 ret = ext4_force_commit(inode->i_sb);
5628 /* Currently just for extent based files */
5629 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5634 /* Check for wrap through zero */
5635 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5640 /* Offset should be less than i_size */
5641 if (offset >= i_size_read(inode)) {
5646 /* Wait for existing dio to complete */
5647 ext4_inode_block_unlocked_dio(inode);
5648 inode_dio_wait(inode);
5651 * Prevent page faults from reinstantiating pages we have released from
5654 down_write(&EXT4_I(inode)->i_mmap_sem);
5656 * Need to round down to align start offset to page size boundary
5657 * for page size > block size.
5659 ioffset = round_down(offset, PAGE_SIZE);
5660 /* Write out all dirty pages */
5661 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5665 truncate_pagecache(inode, ioffset);
5667 credits = ext4_writepage_trans_blocks(inode);
5668 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5669 if (IS_ERR(handle)) {
5670 ret = PTR_ERR(handle);
5674 /* Expand file to avoid data loss if there is error while shifting */
5675 inode->i_size += len;
5676 EXT4_I(inode)->i_disksize += len;
5677 inode->i_mtime = inode->i_ctime = current_time(inode);
5678 ret = ext4_mark_inode_dirty(handle, inode);
5682 down_write(&EXT4_I(inode)->i_data_sem);
5683 ext4_discard_preallocations(inode);
5685 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5687 up_write(&EXT4_I(inode)->i_data_sem);
5691 depth = ext_depth(inode);
5692 extent = path[depth].p_ext;
5694 ee_start_lblk = le32_to_cpu(extent->ee_block);
5695 ee_len = ext4_ext_get_actual_len(extent);
5698 * If offset_lblk is not the starting block of extent, split
5699 * the extent @offset_lblk
5701 if ((offset_lblk > ee_start_lblk) &&
5702 (offset_lblk < (ee_start_lblk + ee_len))) {
5703 if (ext4_ext_is_unwritten(extent))
5704 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5705 EXT4_EXT_MARK_UNWRIT2;
5706 ret = ext4_split_extent_at(handle, inode, &path,
5707 offset_lblk, split_flag,
5709 EXT4_GET_BLOCKS_PRE_IO |
5710 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5713 ext4_ext_drop_refs(path);
5716 up_write(&EXT4_I(inode)->i_data_sem);
5720 ext4_ext_drop_refs(path);
5724 ret = ext4_es_remove_extent(inode, offset_lblk,
5725 EXT_MAX_BLOCKS - offset_lblk);
5727 up_write(&EXT4_I(inode)->i_data_sem);
5732 * if offset_lblk lies in a hole which is at start of file, use
5733 * ee_start_lblk to shift extents
5735 ret = ext4_ext_shift_extents(inode, handle,
5736 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5737 len_lblk, SHIFT_RIGHT);
5739 up_write(&EXT4_I(inode)->i_data_sem);
5741 ext4_handle_sync(handle);
5743 ext4_update_inode_fsync_trans(handle, inode, 1);
5746 ext4_journal_stop(handle);
5748 up_write(&EXT4_I(inode)->i_mmap_sem);
5749 ext4_inode_resume_unlocked_dio(inode);
5751 inode_unlock(inode);
5756 * ext4_swap_extents - Swap extents between two inodes
5758 * @inode1: First inode
5759 * @inode2: Second inode
5760 * @lblk1: Start block for first inode
5761 * @lblk2: Start block for second inode
5762 * @count: Number of blocks to swap
5763 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5764 * @erp: Pointer to save error value
5766 * This helper routine does exactly what is promise "swap extents". All other
5767 * stuff such as page-cache locking consistency, bh mapping consistency or
5768 * extent's data copying must be performed by caller.
5770 * i_mutex is held for both inodes
5771 * i_data_sem is locked for write for both inodes
5773 * All pages from requested range are locked for both inodes
5776 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5777 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5778 ext4_lblk_t count, int unwritten, int *erp)
5780 struct ext4_ext_path *path1 = NULL;
5781 struct ext4_ext_path *path2 = NULL;
5782 int replaced_count = 0;
5784 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5785 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5786 BUG_ON(!inode_is_locked(inode1));
5787 BUG_ON(!inode_is_locked(inode2));
5789 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5792 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5797 struct ext4_extent *ex1, *ex2, tmp_ex;
5798 ext4_lblk_t e1_blk, e2_blk;
5799 int e1_len, e2_len, len;
5802 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5803 if (IS_ERR(path1)) {
5804 *erp = PTR_ERR(path1);
5810 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5811 if (IS_ERR(path2)) {
5812 *erp = PTR_ERR(path2);
5816 ex1 = path1[path1->p_depth].p_ext;
5817 ex2 = path2[path2->p_depth].p_ext;
5818 /* Do we have somthing to swap ? */
5819 if (unlikely(!ex2 || !ex1))
5822 e1_blk = le32_to_cpu(ex1->ee_block);
5823 e2_blk = le32_to_cpu(ex2->ee_block);
5824 e1_len = ext4_ext_get_actual_len(ex1);
5825 e2_len = ext4_ext_get_actual_len(ex2);
5828 if (!in_range(lblk1, e1_blk, e1_len) ||
5829 !in_range(lblk2, e2_blk, e2_len)) {
5830 ext4_lblk_t next1, next2;
5832 /* if hole after extent, then go to next extent */
5833 next1 = ext4_ext_next_allocated_block(path1);
5834 next2 = ext4_ext_next_allocated_block(path2);
5835 /* If hole before extent, then shift to that extent */
5840 /* Do we have something to swap */
5841 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5843 /* Move to the rightest boundary */
5844 len = next1 - lblk1;
5845 if (len < next2 - lblk2)
5846 len = next2 - lblk2;
5855 /* Prepare left boundary */
5856 if (e1_blk < lblk1) {
5858 *erp = ext4_force_split_extent_at(handle, inode1,
5863 if (e2_blk < lblk2) {
5865 *erp = ext4_force_split_extent_at(handle, inode2,
5870 /* ext4_split_extent_at() may result in leaf extent split,
5871 * path must to be revalidated. */
5875 /* Prepare right boundary */
5877 if (len > e1_blk + e1_len - lblk1)
5878 len = e1_blk + e1_len - lblk1;
5879 if (len > e2_blk + e2_len - lblk2)
5880 len = e2_blk + e2_len - lblk2;
5882 if (len != e1_len) {
5884 *erp = ext4_force_split_extent_at(handle, inode1,
5885 &path1, lblk1 + len, 0);
5889 if (len != e2_len) {
5891 *erp = ext4_force_split_extent_at(handle, inode2,
5892 &path2, lblk2 + len, 0);
5896 /* ext4_split_extent_at() may result in leaf extent split,
5897 * path must to be revalidated. */
5901 BUG_ON(e2_len != e1_len);
5902 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5905 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5909 /* Both extents are fully inside boundaries. Swap it now */
5911 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5912 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5913 ex1->ee_len = cpu_to_le16(e2_len);
5914 ex2->ee_len = cpu_to_le16(e1_len);
5916 ext4_ext_mark_unwritten(ex2);
5917 if (ext4_ext_is_unwritten(&tmp_ex))
5918 ext4_ext_mark_unwritten(ex1);
5920 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5921 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5922 *erp = ext4_ext_dirty(handle, inode2, path2 +
5926 *erp = ext4_ext_dirty(handle, inode1, path1 +
5929 * Looks scarry ah..? second inode already points to new blocks,
5930 * and it was successfully dirtied. But luckily error may happen
5931 * only due to journal error, so full transaction will be
5938 replaced_count += len;
5942 ext4_ext_drop_refs(path1);
5944 ext4_ext_drop_refs(path2);
5946 path1 = path2 = NULL;
5948 return replaced_count;