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 <asm/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 /* Verify checksum on non-root extent tree nodes */
482 if (ext_depth(inode) != depth &&
483 !ext4_extent_block_csum_verify(inode, eh)) {
484 error_msg = "extent tree corrupted";
491 ext4_error_inode(inode, function, line, 0,
492 "pblk %llu bad header/extent: %s - magic %x, "
493 "entries %u, max %u(%u), depth %u(%u)",
494 (unsigned long long) pblk, error_msg,
495 le16_to_cpu(eh->eh_magic),
496 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
497 max, le16_to_cpu(eh->eh_depth), depth);
501 #define ext4_ext_check(inode, eh, depth, pblk) \
502 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
504 int ext4_ext_check_inode(struct inode *inode)
506 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
509 static struct buffer_head *
510 __read_extent_tree_block(const char *function, unsigned int line,
511 struct inode *inode, ext4_fsblk_t pblk, int depth,
514 struct buffer_head *bh;
517 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
519 return ERR_PTR(-ENOMEM);
521 if (!bh_uptodate_or_lock(bh)) {
522 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
523 err = bh_submit_read(bh);
527 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
529 err = __ext4_ext_check(function, line, inode,
530 ext_block_hdr(bh), depth, pblk);
533 set_buffer_verified(bh);
535 * If this is a leaf block, cache all of its entries
537 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
538 struct ext4_extent_header *eh = ext_block_hdr(bh);
539 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
540 ext4_lblk_t prev = 0;
543 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
544 unsigned int status = EXTENT_STATUS_WRITTEN;
545 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
546 int len = ext4_ext_get_actual_len(ex);
548 if (prev && (prev != lblk))
549 ext4_es_cache_extent(inode, prev,
553 if (ext4_ext_is_unwritten(ex))
554 status = EXTENT_STATUS_UNWRITTEN;
555 ext4_es_cache_extent(inode, lblk, len,
556 ext4_ext_pblock(ex), status);
567 #define read_extent_tree_block(inode, pblk, depth, flags) \
568 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
572 * This function is called to cache a file's extent information in the
575 int ext4_ext_precache(struct inode *inode)
577 struct ext4_inode_info *ei = EXT4_I(inode);
578 struct ext4_ext_path *path = NULL;
579 struct buffer_head *bh;
580 int i = 0, depth, ret = 0;
582 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
583 return 0; /* not an extent-mapped inode */
585 down_read(&ei->i_data_sem);
586 depth = ext_depth(inode);
588 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
591 up_read(&ei->i_data_sem);
595 /* Don't cache anything if there are no external extent blocks */
598 path[0].p_hdr = ext_inode_hdr(inode);
599 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
602 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
605 * If this is a leaf block or we've reached the end of
606 * the index block, go up
609 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
610 brelse(path[i].p_bh);
615 bh = read_extent_tree_block(inode,
616 ext4_idx_pblock(path[i].p_idx++),
618 EXT4_EX_FORCE_CACHE);
625 path[i].p_hdr = ext_block_hdr(bh);
626 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
628 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
630 up_read(&ei->i_data_sem);
631 ext4_ext_drop_refs(path);
637 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
639 int k, l = path->p_depth;
642 for (k = 0; k <= l; k++, path++) {
644 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
645 ext4_idx_pblock(path->p_idx));
646 } else if (path->p_ext) {
647 ext_debug(" %d:[%d]%d:%llu ",
648 le32_to_cpu(path->p_ext->ee_block),
649 ext4_ext_is_unwritten(path->p_ext),
650 ext4_ext_get_actual_len(path->p_ext),
651 ext4_ext_pblock(path->p_ext));
658 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
660 int depth = ext_depth(inode);
661 struct ext4_extent_header *eh;
662 struct ext4_extent *ex;
668 eh = path[depth].p_hdr;
669 ex = EXT_FIRST_EXTENT(eh);
671 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
673 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
674 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
675 ext4_ext_is_unwritten(ex),
676 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
681 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
682 ext4_fsblk_t newblock, int level)
684 int depth = ext_depth(inode);
685 struct ext4_extent *ex;
687 if (depth != level) {
688 struct ext4_extent_idx *idx;
689 idx = path[level].p_idx;
690 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
691 ext_debug("%d: move %d:%llu in new index %llu\n", level,
692 le32_to_cpu(idx->ei_block),
693 ext4_idx_pblock(idx),
701 ex = path[depth].p_ext;
702 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
703 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
704 le32_to_cpu(ex->ee_block),
706 ext4_ext_is_unwritten(ex),
707 ext4_ext_get_actual_len(ex),
714 #define ext4_ext_show_path(inode, path)
715 #define ext4_ext_show_leaf(inode, path)
716 #define ext4_ext_show_move(inode, path, newblock, level)
719 void ext4_ext_drop_refs(struct ext4_ext_path *path)
725 depth = path->p_depth;
726 for (i = 0; i <= depth; i++, path++)
734 * ext4_ext_binsearch_idx:
735 * binary search for the closest index of the given block
736 * the header must be checked before calling this
739 ext4_ext_binsearch_idx(struct inode *inode,
740 struct ext4_ext_path *path, ext4_lblk_t block)
742 struct ext4_extent_header *eh = path->p_hdr;
743 struct ext4_extent_idx *r, *l, *m;
746 ext_debug("binsearch for %u(idx): ", block);
748 l = EXT_FIRST_INDEX(eh) + 1;
749 r = EXT_LAST_INDEX(eh);
752 if (block < le32_to_cpu(m->ei_block))
756 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
757 m, le32_to_cpu(m->ei_block),
758 r, le32_to_cpu(r->ei_block));
762 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
763 ext4_idx_pblock(path->p_idx));
765 #ifdef CHECK_BINSEARCH
767 struct ext4_extent_idx *chix, *ix;
770 chix = ix = EXT_FIRST_INDEX(eh);
771 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
773 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
774 printk(KERN_DEBUG "k=%d, ix=0x%p, "
776 ix, EXT_FIRST_INDEX(eh));
777 printk(KERN_DEBUG "%u <= %u\n",
778 le32_to_cpu(ix->ei_block),
779 le32_to_cpu(ix[-1].ei_block));
781 BUG_ON(k && le32_to_cpu(ix->ei_block)
782 <= le32_to_cpu(ix[-1].ei_block));
783 if (block < le32_to_cpu(ix->ei_block))
787 BUG_ON(chix != path->p_idx);
794 * ext4_ext_binsearch:
795 * binary search for closest extent of the given block
796 * the header must be checked before calling this
799 ext4_ext_binsearch(struct inode *inode,
800 struct ext4_ext_path *path, ext4_lblk_t block)
802 struct ext4_extent_header *eh = path->p_hdr;
803 struct ext4_extent *r, *l, *m;
805 if (eh->eh_entries == 0) {
807 * this leaf is empty:
808 * we get such a leaf in split/add case
813 ext_debug("binsearch for %u: ", block);
815 l = EXT_FIRST_EXTENT(eh) + 1;
816 r = EXT_LAST_EXTENT(eh);
820 if (block < le32_to_cpu(m->ee_block))
824 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
825 m, le32_to_cpu(m->ee_block),
826 r, le32_to_cpu(r->ee_block));
830 ext_debug(" -> %d:%llu:[%d]%d ",
831 le32_to_cpu(path->p_ext->ee_block),
832 ext4_ext_pblock(path->p_ext),
833 ext4_ext_is_unwritten(path->p_ext),
834 ext4_ext_get_actual_len(path->p_ext));
836 #ifdef CHECK_BINSEARCH
838 struct ext4_extent *chex, *ex;
841 chex = ex = EXT_FIRST_EXTENT(eh);
842 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
843 BUG_ON(k && le32_to_cpu(ex->ee_block)
844 <= le32_to_cpu(ex[-1].ee_block));
845 if (block < le32_to_cpu(ex->ee_block))
849 BUG_ON(chex != path->p_ext);
855 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
857 struct ext4_extent_header *eh;
859 eh = ext_inode_hdr(inode);
862 eh->eh_magic = EXT4_EXT_MAGIC;
863 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
864 ext4_mark_inode_dirty(handle, inode);
868 struct ext4_ext_path *
869 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
870 struct ext4_ext_path **orig_path, int flags)
872 struct ext4_extent_header *eh;
873 struct buffer_head *bh;
874 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
875 short int depth, i, ppos = 0;
878 eh = ext_inode_hdr(inode);
879 depth = ext_depth(inode);
882 ext4_ext_drop_refs(path);
883 if (depth > path[0].p_maxdepth) {
885 *orig_path = path = NULL;
889 /* account possible depth increase */
890 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
893 return ERR_PTR(-ENOMEM);
894 path[0].p_maxdepth = depth + 1;
900 /* walk through the tree */
902 ext_debug("depth %d: num %d, max %d\n",
903 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
905 ext4_ext_binsearch_idx(inode, path + ppos, block);
906 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
907 path[ppos].p_depth = i;
908 path[ppos].p_ext = NULL;
910 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
917 eh = ext_block_hdr(bh);
919 path[ppos].p_bh = bh;
920 path[ppos].p_hdr = eh;
923 path[ppos].p_depth = i;
924 path[ppos].p_ext = NULL;
925 path[ppos].p_idx = NULL;
928 ext4_ext_binsearch(inode, path + ppos, block);
929 /* if not an empty leaf */
930 if (path[ppos].p_ext)
931 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
933 ext4_ext_show_path(inode, path);
938 ext4_ext_drop_refs(path);
946 * ext4_ext_insert_index:
947 * insert new index [@logical;@ptr] into the block at @curp;
948 * check where to insert: before @curp or after @curp
950 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
951 struct ext4_ext_path *curp,
952 int logical, ext4_fsblk_t ptr)
954 struct ext4_extent_idx *ix;
957 err = ext4_ext_get_access(handle, inode, curp);
961 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
962 EXT4_ERROR_INODE(inode,
963 "logical %d == ei_block %d!",
964 logical, le32_to_cpu(curp->p_idx->ei_block));
965 return -EFSCORRUPTED;
968 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
969 >= le16_to_cpu(curp->p_hdr->eh_max))) {
970 EXT4_ERROR_INODE(inode,
971 "eh_entries %d >= eh_max %d!",
972 le16_to_cpu(curp->p_hdr->eh_entries),
973 le16_to_cpu(curp->p_hdr->eh_max));
974 return -EFSCORRUPTED;
977 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
979 ext_debug("insert new index %d after: %llu\n", logical, ptr);
980 ix = curp->p_idx + 1;
983 ext_debug("insert new index %d before: %llu\n", logical, ptr);
987 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
990 ext_debug("insert new index %d: "
991 "move %d indices from 0x%p to 0x%p\n",
992 logical, len, ix, ix + 1);
993 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
996 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
997 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
998 return -EFSCORRUPTED;
1001 ix->ei_block = cpu_to_le32(logical);
1002 ext4_idx_store_pblock(ix, ptr);
1003 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1005 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1006 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1007 return -EFSCORRUPTED;
1010 err = ext4_ext_dirty(handle, inode, curp);
1011 ext4_std_error(inode->i_sb, err);
1018 * inserts new subtree into the path, using free index entry
1020 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1021 * - makes decision where to split
1022 * - moves remaining extents and index entries (right to the split point)
1023 * into the newly allocated blocks
1024 * - initializes subtree
1026 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1028 struct ext4_ext_path *path,
1029 struct ext4_extent *newext, int at)
1031 struct buffer_head *bh = NULL;
1032 int depth = ext_depth(inode);
1033 struct ext4_extent_header *neh;
1034 struct ext4_extent_idx *fidx;
1035 int i = at, k, m, a;
1036 ext4_fsblk_t newblock, oldblock;
1038 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1041 /* make decision: where to split? */
1042 /* FIXME: now decision is simplest: at current extent */
1044 /* if current leaf will be split, then we should use
1045 * border from split point */
1046 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1047 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1048 return -EFSCORRUPTED;
1050 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1051 border = path[depth].p_ext[1].ee_block;
1052 ext_debug("leaf will be split."
1053 " next leaf starts at %d\n",
1054 le32_to_cpu(border));
1056 border = newext->ee_block;
1057 ext_debug("leaf will be added."
1058 " next leaf starts at %d\n",
1059 le32_to_cpu(border));
1063 * If error occurs, then we break processing
1064 * and mark filesystem read-only. index won't
1065 * be inserted and tree will be in consistent
1066 * state. Next mount will repair buffers too.
1070 * Get array to track all allocated blocks.
1071 * We need this to handle errors and free blocks
1074 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1078 /* allocate all needed blocks */
1079 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1080 for (a = 0; a < depth - at; a++) {
1081 newblock = ext4_ext_new_meta_block(handle, inode, path,
1082 newext, &err, flags);
1085 ablocks[a] = newblock;
1088 /* initialize new leaf */
1089 newblock = ablocks[--a];
1090 if (unlikely(newblock == 0)) {
1091 EXT4_ERROR_INODE(inode, "newblock == 0!");
1092 err = -EFSCORRUPTED;
1095 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1096 if (unlikely(!bh)) {
1102 err = ext4_journal_get_create_access(handle, bh);
1106 neh = ext_block_hdr(bh);
1107 neh->eh_entries = 0;
1108 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1109 neh->eh_magic = EXT4_EXT_MAGIC;
1112 /* move remainder of path[depth] to the new leaf */
1113 if (unlikely(path[depth].p_hdr->eh_entries !=
1114 path[depth].p_hdr->eh_max)) {
1115 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1116 path[depth].p_hdr->eh_entries,
1117 path[depth].p_hdr->eh_max);
1118 err = -EFSCORRUPTED;
1121 /* start copy from next extent */
1122 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1123 ext4_ext_show_move(inode, path, newblock, depth);
1125 struct ext4_extent *ex;
1126 ex = EXT_FIRST_EXTENT(neh);
1127 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1128 le16_add_cpu(&neh->eh_entries, m);
1131 ext4_extent_block_csum_set(inode, neh);
1132 set_buffer_uptodate(bh);
1135 err = ext4_handle_dirty_metadata(handle, inode, bh);
1141 /* correct old leaf */
1143 err = ext4_ext_get_access(handle, inode, path + depth);
1146 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1147 err = ext4_ext_dirty(handle, inode, path + depth);
1153 /* create intermediate indexes */
1155 if (unlikely(k < 0)) {
1156 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1157 err = -EFSCORRUPTED;
1161 ext_debug("create %d intermediate indices\n", k);
1162 /* insert new index into current index block */
1163 /* current depth stored in i var */
1166 oldblock = newblock;
1167 newblock = ablocks[--a];
1168 bh = sb_getblk(inode->i_sb, newblock);
1169 if (unlikely(!bh)) {
1175 err = ext4_journal_get_create_access(handle, bh);
1179 neh = ext_block_hdr(bh);
1180 neh->eh_entries = cpu_to_le16(1);
1181 neh->eh_magic = EXT4_EXT_MAGIC;
1182 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1183 neh->eh_depth = cpu_to_le16(depth - i);
1184 fidx = EXT_FIRST_INDEX(neh);
1185 fidx->ei_block = border;
1186 ext4_idx_store_pblock(fidx, oldblock);
1188 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1189 i, newblock, le32_to_cpu(border), oldblock);
1191 /* move remainder of path[i] to the new index block */
1192 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1193 EXT_LAST_INDEX(path[i].p_hdr))) {
1194 EXT4_ERROR_INODE(inode,
1195 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1196 le32_to_cpu(path[i].p_ext->ee_block));
1197 err = -EFSCORRUPTED;
1200 /* start copy indexes */
1201 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1202 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1203 EXT_MAX_INDEX(path[i].p_hdr));
1204 ext4_ext_show_move(inode, path, newblock, i);
1206 memmove(++fidx, path[i].p_idx,
1207 sizeof(struct ext4_extent_idx) * m);
1208 le16_add_cpu(&neh->eh_entries, m);
1210 ext4_extent_block_csum_set(inode, neh);
1211 set_buffer_uptodate(bh);
1214 err = ext4_handle_dirty_metadata(handle, inode, bh);
1220 /* correct old index */
1222 err = ext4_ext_get_access(handle, inode, path + i);
1225 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1226 err = ext4_ext_dirty(handle, inode, path + i);
1234 /* insert new index */
1235 err = ext4_ext_insert_index(handle, inode, path + at,
1236 le32_to_cpu(border), newblock);
1240 if (buffer_locked(bh))
1246 /* free all allocated blocks in error case */
1247 for (i = 0; i < depth; i++) {
1250 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1251 EXT4_FREE_BLOCKS_METADATA);
1260 * ext4_ext_grow_indepth:
1261 * implements tree growing procedure:
1262 * - allocates new block
1263 * - moves top-level data (index block or leaf) into the new block
1264 * - initializes new top-level, creating index that points to the
1265 * just created block
1267 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1270 struct ext4_extent_header *neh;
1271 struct buffer_head *bh;
1272 ext4_fsblk_t newblock, goal = 0;
1273 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1276 /* Try to prepend new index to old one */
1277 if (ext_depth(inode))
1278 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1279 if (goal > le32_to_cpu(es->s_first_data_block)) {
1280 flags |= EXT4_MB_HINT_TRY_GOAL;
1283 goal = ext4_inode_to_goal_block(inode);
1284 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1289 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1294 err = ext4_journal_get_create_access(handle, bh);
1300 /* move top-level index/leaf into new block */
1301 memmove(bh->b_data, EXT4_I(inode)->i_data,
1302 sizeof(EXT4_I(inode)->i_data));
1304 /* set size of new block */
1305 neh = ext_block_hdr(bh);
1306 /* old root could have indexes or leaves
1307 * so calculate e_max right way */
1308 if (ext_depth(inode))
1309 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1311 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1312 neh->eh_magic = EXT4_EXT_MAGIC;
1313 ext4_extent_block_csum_set(inode, neh);
1314 set_buffer_uptodate(bh);
1317 err = ext4_handle_dirty_metadata(handle, inode, bh);
1321 /* Update top-level index: num,max,pointer */
1322 neh = ext_inode_hdr(inode);
1323 neh->eh_entries = cpu_to_le16(1);
1324 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1325 if (neh->eh_depth == 0) {
1326 /* Root extent block becomes index block */
1327 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1328 EXT_FIRST_INDEX(neh)->ei_block =
1329 EXT_FIRST_EXTENT(neh)->ee_block;
1331 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1332 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1333 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1334 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1336 le16_add_cpu(&neh->eh_depth, 1);
1337 ext4_mark_inode_dirty(handle, inode);
1345 * ext4_ext_create_new_leaf:
1346 * finds empty index and adds new leaf.
1347 * if no free index is found, then it requests in-depth growing.
1349 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1350 unsigned int mb_flags,
1351 unsigned int gb_flags,
1352 struct ext4_ext_path **ppath,
1353 struct ext4_extent *newext)
1355 struct ext4_ext_path *path = *ppath;
1356 struct ext4_ext_path *curp;
1357 int depth, i, err = 0;
1360 i = depth = ext_depth(inode);
1362 /* walk up to the tree and look for free index entry */
1363 curp = path + depth;
1364 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1369 /* we use already allocated block for index block,
1370 * so subsequent data blocks should be contiguous */
1371 if (EXT_HAS_FREE_INDEX(curp)) {
1372 /* if we found index with free entry, then use that
1373 * entry: create all needed subtree and add new leaf */
1374 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1379 path = ext4_find_extent(inode,
1380 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1383 err = PTR_ERR(path);
1385 /* tree is full, time to grow in depth */
1386 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1391 path = ext4_find_extent(inode,
1392 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1395 err = PTR_ERR(path);
1400 * only first (depth 0 -> 1) produces free space;
1401 * in all other cases we have to split the grown tree
1403 depth = ext_depth(inode);
1404 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1405 /* now we need to split */
1415 * search the closest allocated block to the left for *logical
1416 * and returns it at @logical + it's physical address at @phys
1417 * if *logical is the smallest allocated block, the function
1418 * returns 0 at @phys
1419 * return value contains 0 (success) or error code
1421 static int ext4_ext_search_left(struct inode *inode,
1422 struct ext4_ext_path *path,
1423 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1425 struct ext4_extent_idx *ix;
1426 struct ext4_extent *ex;
1429 if (unlikely(path == NULL)) {
1430 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1431 return -EFSCORRUPTED;
1433 depth = path->p_depth;
1436 if (depth == 0 && path->p_ext == NULL)
1439 /* usually extent in the path covers blocks smaller
1440 * then *logical, but it can be that extent is the
1441 * first one in the file */
1443 ex = path[depth].p_ext;
1444 ee_len = ext4_ext_get_actual_len(ex);
1445 if (*logical < le32_to_cpu(ex->ee_block)) {
1446 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1447 EXT4_ERROR_INODE(inode,
1448 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1449 *logical, le32_to_cpu(ex->ee_block));
1450 return -EFSCORRUPTED;
1452 while (--depth >= 0) {
1453 ix = path[depth].p_idx;
1454 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1455 EXT4_ERROR_INODE(inode,
1456 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1457 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1458 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1459 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1461 return -EFSCORRUPTED;
1467 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1468 EXT4_ERROR_INODE(inode,
1469 "logical %d < ee_block %d + ee_len %d!",
1470 *logical, le32_to_cpu(ex->ee_block), ee_len);
1471 return -EFSCORRUPTED;
1474 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1475 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1480 * search the closest allocated block to the right for *logical
1481 * and returns it at @logical + it's physical address at @phys
1482 * if *logical is the largest allocated block, the function
1483 * returns 0 at @phys
1484 * return value contains 0 (success) or error code
1486 static int ext4_ext_search_right(struct inode *inode,
1487 struct ext4_ext_path *path,
1488 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1489 struct ext4_extent **ret_ex)
1491 struct buffer_head *bh = NULL;
1492 struct ext4_extent_header *eh;
1493 struct ext4_extent_idx *ix;
1494 struct ext4_extent *ex;
1496 int depth; /* Note, NOT eh_depth; depth from top of tree */
1499 if (unlikely(path == NULL)) {
1500 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1501 return -EFSCORRUPTED;
1503 depth = path->p_depth;
1506 if (depth == 0 && path->p_ext == NULL)
1509 /* usually extent in the path covers blocks smaller
1510 * then *logical, but it can be that extent is the
1511 * first one in the file */
1513 ex = path[depth].p_ext;
1514 ee_len = ext4_ext_get_actual_len(ex);
1515 if (*logical < le32_to_cpu(ex->ee_block)) {
1516 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1517 EXT4_ERROR_INODE(inode,
1518 "first_extent(path[%d].p_hdr) != ex",
1520 return -EFSCORRUPTED;
1522 while (--depth >= 0) {
1523 ix = path[depth].p_idx;
1524 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1525 EXT4_ERROR_INODE(inode,
1526 "ix != EXT_FIRST_INDEX *logical %d!",
1528 return -EFSCORRUPTED;
1534 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1535 EXT4_ERROR_INODE(inode,
1536 "logical %d < ee_block %d + ee_len %d!",
1537 *logical, le32_to_cpu(ex->ee_block), ee_len);
1538 return -EFSCORRUPTED;
1541 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1542 /* next allocated block in this leaf */
1547 /* go up and search for index to the right */
1548 while (--depth >= 0) {
1549 ix = path[depth].p_idx;
1550 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1554 /* we've gone up to the root and found no index to the right */
1558 /* we've found index to the right, let's
1559 * follow it and find the closest allocated
1560 * block to the right */
1562 block = ext4_idx_pblock(ix);
1563 while (++depth < path->p_depth) {
1564 /* subtract from p_depth to get proper eh_depth */
1565 bh = read_extent_tree_block(inode, block,
1566 path->p_depth - depth, 0);
1569 eh = ext_block_hdr(bh);
1570 ix = EXT_FIRST_INDEX(eh);
1571 block = ext4_idx_pblock(ix);
1575 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1578 eh = ext_block_hdr(bh);
1579 ex = EXT_FIRST_EXTENT(eh);
1581 *logical = le32_to_cpu(ex->ee_block);
1582 *phys = ext4_ext_pblock(ex);
1590 * ext4_ext_next_allocated_block:
1591 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1592 * NOTE: it considers block number from index entry as
1593 * allocated block. Thus, index entries have to be consistent
1597 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1601 BUG_ON(path == NULL);
1602 depth = path->p_depth;
1604 if (depth == 0 && path->p_ext == NULL)
1605 return EXT_MAX_BLOCKS;
1607 while (depth >= 0) {
1608 if (depth == path->p_depth) {
1610 if (path[depth].p_ext &&
1611 path[depth].p_ext !=
1612 EXT_LAST_EXTENT(path[depth].p_hdr))
1613 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1616 if (path[depth].p_idx !=
1617 EXT_LAST_INDEX(path[depth].p_hdr))
1618 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1623 return EXT_MAX_BLOCKS;
1627 * ext4_ext_next_leaf_block:
1628 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1630 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1634 BUG_ON(path == NULL);
1635 depth = path->p_depth;
1637 /* zero-tree has no leaf blocks at all */
1639 return EXT_MAX_BLOCKS;
1641 /* go to index block */
1644 while (depth >= 0) {
1645 if (path[depth].p_idx !=
1646 EXT_LAST_INDEX(path[depth].p_hdr))
1647 return (ext4_lblk_t)
1648 le32_to_cpu(path[depth].p_idx[1].ei_block);
1652 return EXT_MAX_BLOCKS;
1656 * ext4_ext_correct_indexes:
1657 * if leaf gets modified and modified extent is first in the leaf,
1658 * then we have to correct all indexes above.
1659 * TODO: do we need to correct tree in all cases?
1661 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1662 struct ext4_ext_path *path)
1664 struct ext4_extent_header *eh;
1665 int depth = ext_depth(inode);
1666 struct ext4_extent *ex;
1670 eh = path[depth].p_hdr;
1671 ex = path[depth].p_ext;
1673 if (unlikely(ex == NULL || eh == NULL)) {
1674 EXT4_ERROR_INODE(inode,
1675 "ex %p == NULL or eh %p == NULL", ex, eh);
1676 return -EFSCORRUPTED;
1680 /* there is no tree at all */
1684 if (ex != EXT_FIRST_EXTENT(eh)) {
1685 /* we correct tree if first leaf got modified only */
1690 * TODO: we need correction if border is smaller than current one
1693 border = path[depth].p_ext->ee_block;
1694 err = ext4_ext_get_access(handle, inode, path + k);
1697 path[k].p_idx->ei_block = border;
1698 err = ext4_ext_dirty(handle, inode, path + k);
1703 /* change all left-side indexes */
1704 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1706 err = ext4_ext_get_access(handle, inode, path + k);
1709 path[k].p_idx->ei_block = border;
1710 err = ext4_ext_dirty(handle, inode, path + k);
1719 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1720 struct ext4_extent *ex2)
1722 unsigned short ext1_ee_len, ext2_ee_len;
1724 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1727 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1728 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1730 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1731 le32_to_cpu(ex2->ee_block))
1735 * To allow future support for preallocated extents to be added
1736 * as an RO_COMPAT feature, refuse to merge to extents if
1737 * this can result in the top bit of ee_len being set.
1739 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1742 * The check for IO to unwritten extent is somewhat racy as we
1743 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1744 * dropping i_data_sem. But reserved blocks should save us in that
1747 if (ext4_ext_is_unwritten(ex1) &&
1748 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1749 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1750 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1752 #ifdef AGGRESSIVE_TEST
1753 if (ext1_ee_len >= 4)
1757 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1763 * This function tries to merge the "ex" extent to the next extent in the tree.
1764 * It always tries to merge towards right. If you want to merge towards
1765 * left, pass "ex - 1" as argument instead of "ex".
1766 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1767 * 1 if they got merged.
1769 static int ext4_ext_try_to_merge_right(struct inode *inode,
1770 struct ext4_ext_path *path,
1771 struct ext4_extent *ex)
1773 struct ext4_extent_header *eh;
1774 unsigned int depth, len;
1775 int merge_done = 0, unwritten;
1777 depth = ext_depth(inode);
1778 BUG_ON(path[depth].p_hdr == NULL);
1779 eh = path[depth].p_hdr;
1781 while (ex < EXT_LAST_EXTENT(eh)) {
1782 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1784 /* merge with next extent! */
1785 unwritten = ext4_ext_is_unwritten(ex);
1786 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1787 + ext4_ext_get_actual_len(ex + 1));
1789 ext4_ext_mark_unwritten(ex);
1791 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1792 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1793 * sizeof(struct ext4_extent);
1794 memmove(ex + 1, ex + 2, len);
1796 le16_add_cpu(&eh->eh_entries, -1);
1798 WARN_ON(eh->eh_entries == 0);
1799 if (!eh->eh_entries)
1800 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1807 * This function does a very simple check to see if we can collapse
1808 * an extent tree with a single extent tree leaf block into the inode.
1810 static void ext4_ext_try_to_merge_up(handle_t *handle,
1811 struct inode *inode,
1812 struct ext4_ext_path *path)
1815 unsigned max_root = ext4_ext_space_root(inode, 0);
1818 if ((path[0].p_depth != 1) ||
1819 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1820 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1824 * We need to modify the block allocation bitmap and the block
1825 * group descriptor to release the extent tree block. If we
1826 * can't get the journal credits, give up.
1828 if (ext4_journal_extend(handle, 2))
1832 * Copy the extent data up to the inode
1834 blk = ext4_idx_pblock(path[0].p_idx);
1835 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1836 sizeof(struct ext4_extent_idx);
1837 s += sizeof(struct ext4_extent_header);
1839 path[1].p_maxdepth = path[0].p_maxdepth;
1840 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1841 path[0].p_depth = 0;
1842 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1843 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1844 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1846 brelse(path[1].p_bh);
1847 ext4_free_blocks(handle, inode, NULL, blk, 1,
1848 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1852 * This function tries to merge the @ex extent to neighbours in the tree.
1853 * return 1 if merge left else 0.
1855 static void ext4_ext_try_to_merge(handle_t *handle,
1856 struct inode *inode,
1857 struct ext4_ext_path *path,
1858 struct ext4_extent *ex) {
1859 struct ext4_extent_header *eh;
1863 depth = ext_depth(inode);
1864 BUG_ON(path[depth].p_hdr == NULL);
1865 eh = path[depth].p_hdr;
1867 if (ex > EXT_FIRST_EXTENT(eh))
1868 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1871 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1873 ext4_ext_try_to_merge_up(handle, inode, path);
1877 * check if a portion of the "newext" extent overlaps with an
1880 * If there is an overlap discovered, it updates the length of the newext
1881 * such that there will be no overlap, and then returns 1.
1882 * If there is no overlap found, it returns 0.
1884 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1885 struct inode *inode,
1886 struct ext4_extent *newext,
1887 struct ext4_ext_path *path)
1890 unsigned int depth, len1;
1891 unsigned int ret = 0;
1893 b1 = le32_to_cpu(newext->ee_block);
1894 len1 = ext4_ext_get_actual_len(newext);
1895 depth = ext_depth(inode);
1896 if (!path[depth].p_ext)
1898 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1901 * get the next allocated block if the extent in the path
1902 * is before the requested block(s)
1905 b2 = ext4_ext_next_allocated_block(path);
1906 if (b2 == EXT_MAX_BLOCKS)
1908 b2 = EXT4_LBLK_CMASK(sbi, b2);
1911 /* check for wrap through zero on extent logical start block*/
1912 if (b1 + len1 < b1) {
1913 len1 = EXT_MAX_BLOCKS - b1;
1914 newext->ee_len = cpu_to_le16(len1);
1918 /* check for overlap */
1919 if (b1 + len1 > b2) {
1920 newext->ee_len = cpu_to_le16(b2 - b1);
1928 * ext4_ext_insert_extent:
1929 * tries to merge requsted extent into the existing extent or
1930 * inserts requested extent as new one into the tree,
1931 * creating new leaf in the no-space case.
1933 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1934 struct ext4_ext_path **ppath,
1935 struct ext4_extent *newext, int gb_flags)
1937 struct ext4_ext_path *path = *ppath;
1938 struct ext4_extent_header *eh;
1939 struct ext4_extent *ex, *fex;
1940 struct ext4_extent *nearex; /* nearest extent */
1941 struct ext4_ext_path *npath = NULL;
1942 int depth, len, err;
1944 int mb_flags = 0, unwritten;
1946 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1947 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1948 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1949 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1950 return -EFSCORRUPTED;
1952 depth = ext_depth(inode);
1953 ex = path[depth].p_ext;
1954 eh = path[depth].p_hdr;
1955 if (unlikely(path[depth].p_hdr == NULL)) {
1956 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1957 return -EFSCORRUPTED;
1960 /* try to insert block into found extent and return */
1961 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1964 * Try to see whether we should rather test the extent on
1965 * right from ex, or from the left of ex. This is because
1966 * ext4_find_extent() can return either extent on the
1967 * left, or on the right from the searched position. This
1968 * will make merging more effective.
1970 if (ex < EXT_LAST_EXTENT(eh) &&
1971 (le32_to_cpu(ex->ee_block) +
1972 ext4_ext_get_actual_len(ex) <
1973 le32_to_cpu(newext->ee_block))) {
1976 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1977 (le32_to_cpu(newext->ee_block) +
1978 ext4_ext_get_actual_len(newext) <
1979 le32_to_cpu(ex->ee_block)))
1982 /* Try to append newex to the ex */
1983 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1984 ext_debug("append [%d]%d block to %u:[%d]%d"
1986 ext4_ext_is_unwritten(newext),
1987 ext4_ext_get_actual_len(newext),
1988 le32_to_cpu(ex->ee_block),
1989 ext4_ext_is_unwritten(ex),
1990 ext4_ext_get_actual_len(ex),
1991 ext4_ext_pblock(ex));
1992 err = ext4_ext_get_access(handle, inode,
1996 unwritten = ext4_ext_is_unwritten(ex);
1997 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1998 + ext4_ext_get_actual_len(newext));
2000 ext4_ext_mark_unwritten(ex);
2001 eh = path[depth].p_hdr;
2007 /* Try to prepend newex to the ex */
2008 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2009 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2011 le32_to_cpu(newext->ee_block),
2012 ext4_ext_is_unwritten(newext),
2013 ext4_ext_get_actual_len(newext),
2014 le32_to_cpu(ex->ee_block),
2015 ext4_ext_is_unwritten(ex),
2016 ext4_ext_get_actual_len(ex),
2017 ext4_ext_pblock(ex));
2018 err = ext4_ext_get_access(handle, inode,
2023 unwritten = ext4_ext_is_unwritten(ex);
2024 ex->ee_block = newext->ee_block;
2025 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2026 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2027 + ext4_ext_get_actual_len(newext));
2029 ext4_ext_mark_unwritten(ex);
2030 eh = path[depth].p_hdr;
2036 depth = ext_depth(inode);
2037 eh = path[depth].p_hdr;
2038 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2041 /* probably next leaf has space for us? */
2042 fex = EXT_LAST_EXTENT(eh);
2043 next = EXT_MAX_BLOCKS;
2044 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2045 next = ext4_ext_next_leaf_block(path);
2046 if (next != EXT_MAX_BLOCKS) {
2047 ext_debug("next leaf block - %u\n", next);
2048 BUG_ON(npath != NULL);
2049 npath = ext4_find_extent(inode, next, NULL, 0);
2051 return PTR_ERR(npath);
2052 BUG_ON(npath->p_depth != path->p_depth);
2053 eh = npath[depth].p_hdr;
2054 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2055 ext_debug("next leaf isn't full(%d)\n",
2056 le16_to_cpu(eh->eh_entries));
2060 ext_debug("next leaf has no free space(%d,%d)\n",
2061 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2065 * There is no free space in the found leaf.
2066 * We're gonna add a new leaf in the tree.
2068 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2069 mb_flags |= EXT4_MB_USE_RESERVED;
2070 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2074 depth = ext_depth(inode);
2075 eh = path[depth].p_hdr;
2078 nearex = path[depth].p_ext;
2080 err = ext4_ext_get_access(handle, inode, path + depth);
2085 /* there is no extent in this leaf, create first one */
2086 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2087 le32_to_cpu(newext->ee_block),
2088 ext4_ext_pblock(newext),
2089 ext4_ext_is_unwritten(newext),
2090 ext4_ext_get_actual_len(newext));
2091 nearex = EXT_FIRST_EXTENT(eh);
2093 if (le32_to_cpu(newext->ee_block)
2094 > le32_to_cpu(nearex->ee_block)) {
2096 ext_debug("insert %u:%llu:[%d]%d before: "
2098 le32_to_cpu(newext->ee_block),
2099 ext4_ext_pblock(newext),
2100 ext4_ext_is_unwritten(newext),
2101 ext4_ext_get_actual_len(newext),
2106 BUG_ON(newext->ee_block == nearex->ee_block);
2107 ext_debug("insert %u:%llu:[%d]%d after: "
2109 le32_to_cpu(newext->ee_block),
2110 ext4_ext_pblock(newext),
2111 ext4_ext_is_unwritten(newext),
2112 ext4_ext_get_actual_len(newext),
2115 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2117 ext_debug("insert %u:%llu:[%d]%d: "
2118 "move %d extents from 0x%p to 0x%p\n",
2119 le32_to_cpu(newext->ee_block),
2120 ext4_ext_pblock(newext),
2121 ext4_ext_is_unwritten(newext),
2122 ext4_ext_get_actual_len(newext),
2123 len, nearex, nearex + 1);
2124 memmove(nearex + 1, nearex,
2125 len * sizeof(struct ext4_extent));
2129 le16_add_cpu(&eh->eh_entries, 1);
2130 path[depth].p_ext = nearex;
2131 nearex->ee_block = newext->ee_block;
2132 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2133 nearex->ee_len = newext->ee_len;
2136 /* try to merge extents */
2137 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2138 ext4_ext_try_to_merge(handle, inode, path, nearex);
2141 /* time to correct all indexes above */
2142 err = ext4_ext_correct_indexes(handle, inode, path);
2146 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2149 ext4_ext_drop_refs(npath);
2154 static int ext4_fill_fiemap_extents(struct inode *inode,
2155 ext4_lblk_t block, ext4_lblk_t num,
2156 struct fiemap_extent_info *fieinfo)
2158 struct ext4_ext_path *path = NULL;
2159 struct ext4_extent *ex;
2160 struct extent_status es;
2161 ext4_lblk_t next, next_del, start = 0, end = 0;
2162 ext4_lblk_t last = block + num;
2163 int exists, depth = 0, err = 0;
2164 unsigned int flags = 0;
2165 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2167 while (block < last && block != EXT_MAX_BLOCKS) {
2169 /* find extent for this block */
2170 down_read(&EXT4_I(inode)->i_data_sem);
2172 path = ext4_find_extent(inode, block, &path, 0);
2174 up_read(&EXT4_I(inode)->i_data_sem);
2175 err = PTR_ERR(path);
2180 depth = ext_depth(inode);
2181 if (unlikely(path[depth].p_hdr == NULL)) {
2182 up_read(&EXT4_I(inode)->i_data_sem);
2183 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2184 err = -EFSCORRUPTED;
2187 ex = path[depth].p_ext;
2188 next = ext4_ext_next_allocated_block(path);
2193 /* there is no extent yet, so try to allocate
2194 * all requested space */
2197 } else if (le32_to_cpu(ex->ee_block) > block) {
2198 /* need to allocate space before found extent */
2200 end = le32_to_cpu(ex->ee_block);
2201 if (block + num < end)
2203 } else if (block >= le32_to_cpu(ex->ee_block)
2204 + ext4_ext_get_actual_len(ex)) {
2205 /* need to allocate space after found extent */
2210 } else if (block >= le32_to_cpu(ex->ee_block)) {
2212 * some part of requested space is covered
2216 end = le32_to_cpu(ex->ee_block)
2217 + ext4_ext_get_actual_len(ex);
2218 if (block + num < end)
2224 BUG_ON(end <= start);
2228 es.es_len = end - start;
2231 es.es_lblk = le32_to_cpu(ex->ee_block);
2232 es.es_len = ext4_ext_get_actual_len(ex);
2233 es.es_pblk = ext4_ext_pblock(ex);
2234 if (ext4_ext_is_unwritten(ex))
2235 flags |= FIEMAP_EXTENT_UNWRITTEN;
2239 * Find delayed extent and update es accordingly. We call
2240 * it even in !exists case to find out whether es is the
2241 * last existing extent or not.
2243 next_del = ext4_find_delayed_extent(inode, &es);
2244 if (!exists && next_del) {
2246 flags |= (FIEMAP_EXTENT_DELALLOC |
2247 FIEMAP_EXTENT_UNKNOWN);
2249 up_read(&EXT4_I(inode)->i_data_sem);
2251 if (unlikely(es.es_len == 0)) {
2252 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2253 err = -EFSCORRUPTED;
2258 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2259 * we need to check next == EXT_MAX_BLOCKS because it is
2260 * possible that an extent is with unwritten and delayed
2261 * status due to when an extent is delayed allocated and
2262 * is allocated by fallocate status tree will track both of
2265 * So we could return a unwritten and delayed extent, and
2266 * its block is equal to 'next'.
2268 if (next == next_del && next == EXT_MAX_BLOCKS) {
2269 flags |= FIEMAP_EXTENT_LAST;
2270 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2271 next != EXT_MAX_BLOCKS)) {
2272 EXT4_ERROR_INODE(inode,
2273 "next extent == %u, next "
2274 "delalloc extent = %u",
2276 err = -EFSCORRUPTED;
2282 err = fiemap_fill_next_extent(fieinfo,
2283 (__u64)es.es_lblk << blksize_bits,
2284 (__u64)es.es_pblk << blksize_bits,
2285 (__u64)es.es_len << blksize_bits,
2295 block = es.es_lblk + es.es_len;
2298 ext4_ext_drop_refs(path);
2304 * ext4_ext_determine_hole - determine hole around given block
2305 * @inode: inode we lookup in
2306 * @path: path in extent tree to @lblk
2307 * @lblk: pointer to logical block around which we want to determine hole
2309 * Determine hole length (and start if easily possible) around given logical
2310 * block. We don't try too hard to find the beginning of the hole but @path
2311 * actually points to extent before @lblk, we provide it.
2313 * The function returns the length of a hole starting at @lblk. We update @lblk
2314 * to the beginning of the hole if we managed to find it.
2316 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2317 struct ext4_ext_path *path,
2320 int depth = ext_depth(inode);
2321 struct ext4_extent *ex;
2324 ex = path[depth].p_ext;
2326 /* there is no extent yet, so gap is [0;-] */
2328 len = EXT_MAX_BLOCKS;
2329 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2330 len = le32_to_cpu(ex->ee_block) - *lblk;
2331 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2332 + ext4_ext_get_actual_len(ex)) {
2335 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2336 next = ext4_ext_next_allocated_block(path);
2337 BUG_ON(next == *lblk);
2346 * ext4_ext_put_gap_in_cache:
2347 * calculate boundaries of the gap that the requested block fits into
2348 * and cache this gap
2351 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2352 ext4_lblk_t hole_len)
2354 struct extent_status es;
2356 ext4_es_find_delayed_extent_range(inode, hole_start,
2357 hole_start + hole_len - 1, &es);
2359 /* There's delayed extent containing lblock? */
2360 if (es.es_lblk <= hole_start)
2362 hole_len = min(es.es_lblk - hole_start, hole_len);
2364 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2365 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2366 EXTENT_STATUS_HOLE);
2371 * removes index from the index block.
2373 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2374 struct ext4_ext_path *path, int depth)
2379 /* free index block */
2381 path = path + depth;
2382 leaf = ext4_idx_pblock(path->p_idx);
2383 if (unlikely(path->p_hdr->eh_entries == 0)) {
2384 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2385 return -EFSCORRUPTED;
2387 err = ext4_ext_get_access(handle, inode, path);
2391 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2392 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2393 len *= sizeof(struct ext4_extent_idx);
2394 memmove(path->p_idx, path->p_idx + 1, len);
2397 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2398 err = ext4_ext_dirty(handle, inode, path);
2401 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2402 trace_ext4_ext_rm_idx(inode, leaf);
2404 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2405 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2407 while (--depth >= 0) {
2408 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2411 err = ext4_ext_get_access(handle, inode, path);
2414 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2415 err = ext4_ext_dirty(handle, inode, path);
2423 * ext4_ext_calc_credits_for_single_extent:
2424 * This routine returns max. credits that needed to insert an extent
2425 * to the extent tree.
2426 * When pass the actual path, the caller should calculate credits
2429 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2430 struct ext4_ext_path *path)
2433 int depth = ext_depth(inode);
2436 /* probably there is space in leaf? */
2437 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2438 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2441 * There are some space in the leaf tree, no
2442 * need to account for leaf block credit
2444 * bitmaps and block group descriptor blocks
2445 * and other metadata blocks still need to be
2448 /* 1 bitmap, 1 block group descriptor */
2449 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2454 return ext4_chunk_trans_blocks(inode, nrblocks);
2458 * How many index/leaf blocks need to change/allocate to add @extents extents?
2460 * If we add a single extent, then in the worse case, each tree level
2461 * index/leaf need to be changed in case of the tree split.
2463 * If more extents are inserted, they could cause the whole tree split more
2464 * than once, but this is really rare.
2466 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2471 /* If we are converting the inline data, only one is needed here. */
2472 if (ext4_has_inline_data(inode))
2475 depth = ext_depth(inode);
2485 static inline int get_default_free_blocks_flags(struct inode *inode)
2487 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2488 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2489 else if (ext4_should_journal_data(inode))
2490 return EXT4_FREE_BLOCKS_FORGET;
2494 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2495 struct ext4_extent *ex,
2496 long long *partial_cluster,
2497 ext4_lblk_t from, ext4_lblk_t to)
2499 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2500 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2502 int flags = get_default_free_blocks_flags(inode);
2505 * For bigalloc file systems, we never free a partial cluster
2506 * at the beginning of the extent. Instead, we make a note
2507 * that we tried freeing the cluster, and check to see if we
2508 * need to free it on a subsequent call to ext4_remove_blocks,
2509 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2511 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2513 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2515 * If we have a partial cluster, and it's different from the
2516 * cluster of the last block, we need to explicitly free the
2517 * partial cluster here.
2519 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2520 if (*partial_cluster > 0 &&
2521 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2522 ext4_free_blocks(handle, inode, NULL,
2523 EXT4_C2B(sbi, *partial_cluster),
2524 sbi->s_cluster_ratio, flags);
2525 *partial_cluster = 0;
2528 #ifdef EXTENTS_STATS
2530 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2531 spin_lock(&sbi->s_ext_stats_lock);
2532 sbi->s_ext_blocks += ee_len;
2533 sbi->s_ext_extents++;
2534 if (ee_len < sbi->s_ext_min)
2535 sbi->s_ext_min = ee_len;
2536 if (ee_len > sbi->s_ext_max)
2537 sbi->s_ext_max = ee_len;
2538 if (ext_depth(inode) > sbi->s_depth_max)
2539 sbi->s_depth_max = ext_depth(inode);
2540 spin_unlock(&sbi->s_ext_stats_lock);
2543 if (from >= le32_to_cpu(ex->ee_block)
2544 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2547 long long first_cluster;
2549 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2550 pblk = ext4_ext_pblock(ex) + ee_len - num;
2552 * Usually we want to free partial cluster at the end of the
2553 * extent, except for the situation when the cluster is still
2554 * used by any other extent (partial_cluster is negative).
2556 if (*partial_cluster < 0 &&
2557 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2558 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2560 ext_debug("free last %u blocks starting %llu partial %lld\n",
2561 num, pblk, *partial_cluster);
2562 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2564 * If the block range to be freed didn't start at the
2565 * beginning of a cluster, and we removed the entire
2566 * extent and the cluster is not used by any other extent,
2567 * save the partial cluster here, since we might need to
2568 * delete if we determine that the truncate or punch hole
2569 * operation has removed all of the blocks in the cluster.
2570 * If that cluster is used by another extent, preserve its
2571 * negative value so it isn't freed later on.
2573 * If the whole extent wasn't freed, we've reached the
2574 * start of the truncated/punched region and have finished
2575 * removing blocks. If there's a partial cluster here it's
2576 * shared with the remainder of the extent and is no longer
2577 * a candidate for removal.
2579 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2580 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2581 if (first_cluster != -*partial_cluster)
2582 *partial_cluster = first_cluster;
2584 *partial_cluster = 0;
2587 ext4_error(sbi->s_sb, "strange request: removal(2) "
2589 from, to, le32_to_cpu(ex->ee_block), ee_len);
2595 * ext4_ext_rm_leaf() Removes the extents associated with the
2596 * blocks appearing between "start" and "end". Both "start"
2597 * and "end" must appear in the same extent or EIO is returned.
2599 * @handle: The journal handle
2600 * @inode: The files inode
2601 * @path: The path to the leaf
2602 * @partial_cluster: The cluster which we'll have to free if all extents
2603 * has been released from it. However, if this value is
2604 * negative, it's a cluster just to the right of the
2605 * punched region and it must not be freed.
2606 * @start: The first block to remove
2607 * @end: The last block to remove
2610 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2611 struct ext4_ext_path *path,
2612 long long *partial_cluster,
2613 ext4_lblk_t start, ext4_lblk_t end)
2615 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2616 int err = 0, correct_index = 0;
2617 int depth = ext_depth(inode), credits;
2618 struct ext4_extent_header *eh;
2621 ext4_lblk_t ex_ee_block;
2622 unsigned short ex_ee_len;
2623 unsigned unwritten = 0;
2624 struct ext4_extent *ex;
2627 /* the header must be checked already in ext4_ext_remove_space() */
2628 ext_debug("truncate since %u in leaf to %u\n", start, end);
2629 if (!path[depth].p_hdr)
2630 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2631 eh = path[depth].p_hdr;
2632 if (unlikely(path[depth].p_hdr == NULL)) {
2633 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2634 return -EFSCORRUPTED;
2636 /* find where to start removing */
2637 ex = path[depth].p_ext;
2639 ex = EXT_LAST_EXTENT(eh);
2641 ex_ee_block = le32_to_cpu(ex->ee_block);
2642 ex_ee_len = ext4_ext_get_actual_len(ex);
2644 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2646 while (ex >= EXT_FIRST_EXTENT(eh) &&
2647 ex_ee_block + ex_ee_len > start) {
2649 if (ext4_ext_is_unwritten(ex))
2654 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2655 unwritten, ex_ee_len);
2656 path[depth].p_ext = ex;
2658 a = ex_ee_block > start ? ex_ee_block : start;
2659 b = ex_ee_block+ex_ee_len - 1 < end ?
2660 ex_ee_block+ex_ee_len - 1 : end;
2662 ext_debug(" border %u:%u\n", a, b);
2664 /* If this extent is beyond the end of the hole, skip it */
2665 if (end < ex_ee_block) {
2667 * We're going to skip this extent and move to another,
2668 * so note that its first cluster is in use to avoid
2669 * freeing it when removing blocks. Eventually, the
2670 * right edge of the truncated/punched region will
2671 * be just to the left.
2673 if (sbi->s_cluster_ratio > 1) {
2674 pblk = ext4_ext_pblock(ex);
2676 -(long long) EXT4_B2C(sbi, pblk);
2679 ex_ee_block = le32_to_cpu(ex->ee_block);
2680 ex_ee_len = ext4_ext_get_actual_len(ex);
2682 } else if (b != ex_ee_block + ex_ee_len - 1) {
2683 EXT4_ERROR_INODE(inode,
2684 "can not handle truncate %u:%u "
2686 start, end, ex_ee_block,
2687 ex_ee_block + ex_ee_len - 1);
2688 err = -EFSCORRUPTED;
2690 } else if (a != ex_ee_block) {
2691 /* remove tail of the extent */
2692 num = a - ex_ee_block;
2694 /* remove whole extent: excellent! */
2698 * 3 for leaf, sb, and inode plus 2 (bmap and group
2699 * descriptor) for each block group; assume two block
2700 * groups plus ex_ee_len/blocks_per_block_group for
2703 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2704 if (ex == EXT_FIRST_EXTENT(eh)) {
2706 credits += (ext_depth(inode)) + 1;
2708 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2710 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2714 err = ext4_ext_get_access(handle, inode, path + depth);
2718 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2724 /* this extent is removed; mark slot entirely unused */
2725 ext4_ext_store_pblock(ex, 0);
2727 ex->ee_len = cpu_to_le16(num);
2729 * Do not mark unwritten if all the blocks in the
2730 * extent have been removed.
2732 if (unwritten && num)
2733 ext4_ext_mark_unwritten(ex);
2735 * If the extent was completely released,
2736 * we need to remove it from the leaf
2739 if (end != EXT_MAX_BLOCKS - 1) {
2741 * For hole punching, we need to scoot all the
2742 * extents up when an extent is removed so that
2743 * we dont have blank extents in the middle
2745 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2746 sizeof(struct ext4_extent));
2748 /* Now get rid of the one at the end */
2749 memset(EXT_LAST_EXTENT(eh), 0,
2750 sizeof(struct ext4_extent));
2752 le16_add_cpu(&eh->eh_entries, -1);
2755 err = ext4_ext_dirty(handle, inode, path + depth);
2759 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2760 ext4_ext_pblock(ex));
2762 ex_ee_block = le32_to_cpu(ex->ee_block);
2763 ex_ee_len = ext4_ext_get_actual_len(ex);
2766 if (correct_index && eh->eh_entries)
2767 err = ext4_ext_correct_indexes(handle, inode, path);
2770 * If there's a partial cluster and at least one extent remains in
2771 * the leaf, free the partial cluster if it isn't shared with the
2772 * current extent. If it is shared with the current extent
2773 * we zero partial_cluster because we've reached the start of the
2774 * truncated/punched region and we're done removing blocks.
2776 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2777 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2778 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2779 ext4_free_blocks(handle, inode, NULL,
2780 EXT4_C2B(sbi, *partial_cluster),
2781 sbi->s_cluster_ratio,
2782 get_default_free_blocks_flags(inode));
2784 *partial_cluster = 0;
2787 /* if this leaf is free, then we should
2788 * remove it from index block above */
2789 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2790 err = ext4_ext_rm_idx(handle, inode, path, depth);
2797 * ext4_ext_more_to_rm:
2798 * returns 1 if current index has to be freed (even partial)
2801 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2803 BUG_ON(path->p_idx == NULL);
2805 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2809 * if truncate on deeper level happened, it wasn't partial,
2810 * so we have to consider current index for truncation
2812 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2817 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2820 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2821 int depth = ext_depth(inode);
2822 struct ext4_ext_path *path = NULL;
2823 long long partial_cluster = 0;
2827 ext_debug("truncate since %u to %u\n", start, end);
2829 /* probably first extent we're gonna free will be last in block */
2830 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2832 return PTR_ERR(handle);
2835 trace_ext4_ext_remove_space(inode, start, end, depth);
2838 * Check if we are removing extents inside the extent tree. If that
2839 * is the case, we are going to punch a hole inside the extent tree
2840 * so we have to check whether we need to split the extent covering
2841 * the last block to remove so we can easily remove the part of it
2842 * in ext4_ext_rm_leaf().
2844 if (end < EXT_MAX_BLOCKS - 1) {
2845 struct ext4_extent *ex;
2846 ext4_lblk_t ee_block, ex_end, lblk;
2849 /* find extent for or closest extent to this block */
2850 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2852 ext4_journal_stop(handle);
2853 return PTR_ERR(path);
2855 depth = ext_depth(inode);
2856 /* Leaf not may not exist only if inode has no blocks at all */
2857 ex = path[depth].p_ext;
2860 EXT4_ERROR_INODE(inode,
2861 "path[%d].p_hdr == NULL",
2863 err = -EFSCORRUPTED;
2868 ee_block = le32_to_cpu(ex->ee_block);
2869 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2872 * See if the last block is inside the extent, if so split
2873 * the extent at 'end' block so we can easily remove the
2874 * tail of the first part of the split extent in
2875 * ext4_ext_rm_leaf().
2877 if (end >= ee_block && end < ex_end) {
2880 * If we're going to split the extent, note that
2881 * the cluster containing the block after 'end' is
2882 * in use to avoid freeing it when removing blocks.
2884 if (sbi->s_cluster_ratio > 1) {
2885 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2887 -(long long) EXT4_B2C(sbi, pblk);
2891 * Split the extent in two so that 'end' is the last
2892 * block in the first new extent. Also we should not
2893 * fail removing space due to ENOSPC so try to use
2894 * reserved block if that happens.
2896 err = ext4_force_split_extent_at(handle, inode, &path,
2901 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2903 * If there's an extent to the right its first cluster
2904 * contains the immediate right boundary of the
2905 * truncated/punched region. Set partial_cluster to
2906 * its negative value so it won't be freed if shared
2907 * with the current extent. The end < ee_block case
2908 * is handled in ext4_ext_rm_leaf().
2911 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2917 -(long long) EXT4_B2C(sbi, pblk);
2921 * We start scanning from right side, freeing all the blocks
2922 * after i_size and walking into the tree depth-wise.
2924 depth = ext_depth(inode);
2929 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2931 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2934 ext4_journal_stop(handle);
2937 path[0].p_maxdepth = path[0].p_depth = depth;
2938 path[0].p_hdr = ext_inode_hdr(inode);
2941 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2942 err = -EFSCORRUPTED;
2948 while (i >= 0 && err == 0) {
2950 /* this is leaf block */
2951 err = ext4_ext_rm_leaf(handle, inode, path,
2952 &partial_cluster, start,
2954 /* root level has p_bh == NULL, brelse() eats this */
2955 brelse(path[i].p_bh);
2956 path[i].p_bh = NULL;
2961 /* this is index block */
2962 if (!path[i].p_hdr) {
2963 ext_debug("initialize header\n");
2964 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2967 if (!path[i].p_idx) {
2968 /* this level hasn't been touched yet */
2969 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2970 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2971 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2973 le16_to_cpu(path[i].p_hdr->eh_entries));
2975 /* we were already here, see at next index */
2979 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2980 i, EXT_FIRST_INDEX(path[i].p_hdr),
2982 if (ext4_ext_more_to_rm(path + i)) {
2983 struct buffer_head *bh;
2984 /* go to the next level */
2985 ext_debug("move to level %d (block %llu)\n",
2986 i + 1, ext4_idx_pblock(path[i].p_idx));
2987 memset(path + i + 1, 0, sizeof(*path));
2988 bh = read_extent_tree_block(inode,
2989 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2992 /* should we reset i_size? */
2996 /* Yield here to deal with large extent trees.
2997 * Should be a no-op if we did IO above. */
2999 if (WARN_ON(i + 1 > depth)) {
3000 err = -EFSCORRUPTED;
3003 path[i + 1].p_bh = bh;
3005 /* save actual number of indexes since this
3006 * number is changed at the next iteration */
3007 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3010 /* we finished processing this index, go up */
3011 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3012 /* index is empty, remove it;
3013 * handle must be already prepared by the
3014 * truncatei_leaf() */
3015 err = ext4_ext_rm_idx(handle, inode, path, i);
3017 /* root level has p_bh == NULL, brelse() eats this */
3018 brelse(path[i].p_bh);
3019 path[i].p_bh = NULL;
3021 ext_debug("return to level %d\n", i);
3025 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3026 partial_cluster, path->p_hdr->eh_entries);
3029 * If we still have something in the partial cluster and we have removed
3030 * even the first extent, then we should free the blocks in the partial
3031 * cluster as well. (This code will only run when there are no leaves
3032 * to the immediate left of the truncated/punched region.)
3034 if (partial_cluster > 0 && err == 0) {
3035 /* don't zero partial_cluster since it's not used afterwards */
3036 ext4_free_blocks(handle, inode, NULL,
3037 EXT4_C2B(sbi, partial_cluster),
3038 sbi->s_cluster_ratio,
3039 get_default_free_blocks_flags(inode));
3042 /* TODO: flexible tree reduction should be here */
3043 if (path->p_hdr->eh_entries == 0) {
3045 * truncate to zero freed all the tree,
3046 * so we need to correct eh_depth
3048 err = ext4_ext_get_access(handle, inode, path);
3050 ext_inode_hdr(inode)->eh_depth = 0;
3051 ext_inode_hdr(inode)->eh_max =
3052 cpu_to_le16(ext4_ext_space_root(inode, 0));
3053 err = ext4_ext_dirty(handle, inode, path);
3057 ext4_ext_drop_refs(path);
3062 ext4_journal_stop(handle);
3068 * called at mount time
3070 void ext4_ext_init(struct super_block *sb)
3073 * possible initialization would be here
3076 if (ext4_has_feature_extents(sb)) {
3077 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3078 printk(KERN_INFO "EXT4-fs: file extents enabled"
3079 #ifdef AGGRESSIVE_TEST
3080 ", aggressive tests"
3082 #ifdef CHECK_BINSEARCH
3085 #ifdef EXTENTS_STATS
3090 #ifdef EXTENTS_STATS
3091 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3092 EXT4_SB(sb)->s_ext_min = 1 << 30;
3093 EXT4_SB(sb)->s_ext_max = 0;
3099 * called at umount time
3101 void ext4_ext_release(struct super_block *sb)
3103 if (!ext4_has_feature_extents(sb))
3106 #ifdef EXTENTS_STATS
3107 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3108 struct ext4_sb_info *sbi = EXT4_SB(sb);
3109 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3110 sbi->s_ext_blocks, sbi->s_ext_extents,
3111 sbi->s_ext_blocks / sbi->s_ext_extents);
3112 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3113 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3118 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3120 ext4_lblk_t ee_block;
3121 ext4_fsblk_t ee_pblock;
3122 unsigned int ee_len;
3124 ee_block = le32_to_cpu(ex->ee_block);
3125 ee_len = ext4_ext_get_actual_len(ex);
3126 ee_pblock = ext4_ext_pblock(ex);
3131 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3132 EXTENT_STATUS_WRITTEN);
3135 /* FIXME!! we need to try to merge to left or right after zero-out */
3136 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3138 ext4_fsblk_t ee_pblock;
3139 unsigned int ee_len;
3141 ee_len = ext4_ext_get_actual_len(ex);
3142 ee_pblock = ext4_ext_pblock(ex);
3143 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3148 * ext4_split_extent_at() splits an extent at given block.
3150 * @handle: the journal handle
3151 * @inode: the file inode
3152 * @path: the path to the extent
3153 * @split: the logical block where the extent is splitted.
3154 * @split_flags: indicates if the extent could be zeroout if split fails, and
3155 * the states(init or unwritten) of new extents.
3156 * @flags: flags used to insert new extent to extent tree.
3159 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3160 * of which are deterimined by split_flag.
3162 * There are two cases:
3163 * a> the extent are splitted into two extent.
3164 * b> split is not needed, and just mark the extent.
3166 * return 0 on success.
3168 static int ext4_split_extent_at(handle_t *handle,
3169 struct inode *inode,
3170 struct ext4_ext_path **ppath,
3175 struct ext4_ext_path *path = *ppath;
3176 ext4_fsblk_t newblock;
3177 ext4_lblk_t ee_block;
3178 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3179 struct ext4_extent *ex2 = NULL;
3180 unsigned int ee_len, depth;
3183 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3184 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3186 ext_debug("ext4_split_extents_at: inode %lu, logical"
3187 "block %llu\n", inode->i_ino, (unsigned long long)split);
3189 ext4_ext_show_leaf(inode, path);
3191 depth = ext_depth(inode);
3192 ex = path[depth].p_ext;
3193 ee_block = le32_to_cpu(ex->ee_block);
3194 ee_len = ext4_ext_get_actual_len(ex);
3195 newblock = split - ee_block + ext4_ext_pblock(ex);
3197 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3198 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3199 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3200 EXT4_EXT_MARK_UNWRIT1 |
3201 EXT4_EXT_MARK_UNWRIT2));
3203 err = ext4_ext_get_access(handle, inode, path + depth);
3207 if (split == ee_block) {
3209 * case b: block @split is the block that the extent begins with
3210 * then we just change the state of the extent, and splitting
3213 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3214 ext4_ext_mark_unwritten(ex);
3216 ext4_ext_mark_initialized(ex);
3218 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3219 ext4_ext_try_to_merge(handle, inode, path, ex);
3221 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3226 memcpy(&orig_ex, ex, sizeof(orig_ex));
3227 ex->ee_len = cpu_to_le16(split - ee_block);
3228 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3229 ext4_ext_mark_unwritten(ex);
3232 * path may lead to new leaf, not to original leaf any more
3233 * after ext4_ext_insert_extent() returns,
3235 err = ext4_ext_dirty(handle, inode, path + depth);
3237 goto fix_extent_len;
3240 ex2->ee_block = cpu_to_le32(split);
3241 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3242 ext4_ext_store_pblock(ex2, newblock);
3243 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3244 ext4_ext_mark_unwritten(ex2);
3246 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3247 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3248 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3249 if (split_flag & EXT4_EXT_DATA_VALID1) {
3250 err = ext4_ext_zeroout(inode, ex2);
3251 zero_ex.ee_block = ex2->ee_block;
3252 zero_ex.ee_len = cpu_to_le16(
3253 ext4_ext_get_actual_len(ex2));
3254 ext4_ext_store_pblock(&zero_ex,
3255 ext4_ext_pblock(ex2));
3257 err = ext4_ext_zeroout(inode, ex);
3258 zero_ex.ee_block = ex->ee_block;
3259 zero_ex.ee_len = cpu_to_le16(
3260 ext4_ext_get_actual_len(ex));
3261 ext4_ext_store_pblock(&zero_ex,
3262 ext4_ext_pblock(ex));
3265 err = ext4_ext_zeroout(inode, &orig_ex);
3266 zero_ex.ee_block = orig_ex.ee_block;
3267 zero_ex.ee_len = cpu_to_le16(
3268 ext4_ext_get_actual_len(&orig_ex));
3269 ext4_ext_store_pblock(&zero_ex,
3270 ext4_ext_pblock(&orig_ex));
3274 goto fix_extent_len;
3275 /* update the extent length and mark as initialized */
3276 ex->ee_len = cpu_to_le16(ee_len);
3277 ext4_ext_try_to_merge(handle, inode, path, ex);
3278 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3280 goto fix_extent_len;
3282 /* update extent status tree */
3283 err = ext4_zeroout_es(inode, &zero_ex);
3287 goto fix_extent_len;
3290 ext4_ext_show_leaf(inode, path);
3294 ex->ee_len = orig_ex.ee_len;
3295 ext4_ext_dirty(handle, inode, path + path->p_depth);
3300 * ext4_split_extents() splits an extent and mark extent which is covered
3301 * by @map as split_flags indicates
3303 * It may result in splitting the extent into multiple extents (up to three)
3304 * There are three possibilities:
3305 * a> There is no split required
3306 * b> Splits in two extents: Split is happening at either end of the extent
3307 * c> Splits in three extents: Somone is splitting in middle of the extent
3310 static int ext4_split_extent(handle_t *handle,
3311 struct inode *inode,
3312 struct ext4_ext_path **ppath,
3313 struct ext4_map_blocks *map,
3317 struct ext4_ext_path *path = *ppath;
3318 ext4_lblk_t ee_block;
3319 struct ext4_extent *ex;
3320 unsigned int ee_len, depth;
3323 int split_flag1, flags1;
3324 int allocated = map->m_len;
3326 depth = ext_depth(inode);
3327 ex = path[depth].p_ext;
3328 ee_block = le32_to_cpu(ex->ee_block);
3329 ee_len = ext4_ext_get_actual_len(ex);
3330 unwritten = ext4_ext_is_unwritten(ex);
3332 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3333 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3334 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3336 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3337 EXT4_EXT_MARK_UNWRIT2;
3338 if (split_flag & EXT4_EXT_DATA_VALID2)
3339 split_flag1 |= EXT4_EXT_DATA_VALID1;
3340 err = ext4_split_extent_at(handle, inode, ppath,
3341 map->m_lblk + map->m_len, split_flag1, flags1);
3345 allocated = ee_len - (map->m_lblk - ee_block);
3348 * Update path is required because previous ext4_split_extent_at() may
3349 * result in split of original leaf or extent zeroout.
3351 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3353 return PTR_ERR(path);
3354 depth = ext_depth(inode);
3355 ex = path[depth].p_ext;
3357 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3358 (unsigned long) map->m_lblk);
3359 return -EFSCORRUPTED;
3361 unwritten = ext4_ext_is_unwritten(ex);
3364 if (map->m_lblk >= ee_block) {
3365 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3367 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3368 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3369 EXT4_EXT_MARK_UNWRIT2);
3371 err = ext4_split_extent_at(handle, inode, ppath,
3372 map->m_lblk, split_flag1, flags);
3377 ext4_ext_show_leaf(inode, path);
3379 return err ? err : allocated;
3383 * This function is called by ext4_ext_map_blocks() if someone tries to write
3384 * to an unwritten extent. It may result in splitting the unwritten
3385 * extent into multiple extents (up to three - one initialized and two
3387 * There are three possibilities:
3388 * a> There is no split required: Entire extent should be initialized
3389 * b> Splits in two extents: Write is happening at either end of the extent
3390 * c> Splits in three extents: Somone is writing in middle of the extent
3393 * - The extent pointed to by 'path' is unwritten.
3394 * - The extent pointed to by 'path' contains a superset
3395 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3397 * Post-conditions on success:
3398 * - the returned value is the number of blocks beyond map->l_lblk
3399 * that are allocated and initialized.
3400 * It is guaranteed to be >= map->m_len.
3402 static int ext4_ext_convert_to_initialized(handle_t *handle,
3403 struct inode *inode,
3404 struct ext4_map_blocks *map,
3405 struct ext4_ext_path **ppath,
3408 struct ext4_ext_path *path = *ppath;
3409 struct ext4_sb_info *sbi;
3410 struct ext4_extent_header *eh;
3411 struct ext4_map_blocks split_map;
3412 struct ext4_extent zero_ex;
3413 struct ext4_extent *ex, *abut_ex;
3414 ext4_lblk_t ee_block, eof_block;
3415 unsigned int ee_len, depth, map_len = map->m_len;
3416 int allocated = 0, max_zeroout = 0;
3420 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3421 "block %llu, max_blocks %u\n", inode->i_ino,
3422 (unsigned long long)map->m_lblk, map_len);
3424 sbi = EXT4_SB(inode->i_sb);
3425 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3426 inode->i_sb->s_blocksize_bits;
3427 if (eof_block < map->m_lblk + map_len)
3428 eof_block = map->m_lblk + map_len;
3430 depth = ext_depth(inode);
3431 eh = path[depth].p_hdr;
3432 ex = path[depth].p_ext;
3433 ee_block = le32_to_cpu(ex->ee_block);
3434 ee_len = ext4_ext_get_actual_len(ex);
3437 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3439 /* Pre-conditions */
3440 BUG_ON(!ext4_ext_is_unwritten(ex));
3441 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3444 * Attempt to transfer newly initialized blocks from the currently
3445 * unwritten extent to its neighbor. This is much cheaper
3446 * than an insertion followed by a merge as those involve costly
3447 * memmove() calls. Transferring to the left is the common case in
3448 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3449 * followed by append writes.
3451 * Limitations of the current logic:
3452 * - L1: we do not deal with writes covering the whole extent.
3453 * This would require removing the extent if the transfer
3455 * - L2: we only attempt to merge with an extent stored in the
3456 * same extent tree node.
3458 if ((map->m_lblk == ee_block) &&
3459 /* See if we can merge left */
3460 (map_len < ee_len) && /*L1*/
3461 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3462 ext4_lblk_t prev_lblk;
3463 ext4_fsblk_t prev_pblk, ee_pblk;
3464 unsigned int prev_len;
3467 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3468 prev_len = ext4_ext_get_actual_len(abut_ex);
3469 prev_pblk = ext4_ext_pblock(abut_ex);
3470 ee_pblk = ext4_ext_pblock(ex);
3473 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3474 * upon those conditions:
3475 * - C1: abut_ex is initialized,
3476 * - C2: abut_ex is logically abutting ex,
3477 * - C3: abut_ex is physically abutting ex,
3478 * - C4: abut_ex can receive the additional blocks without
3479 * overflowing the (initialized) length limit.
3481 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3482 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3483 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3484 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3485 err = ext4_ext_get_access(handle, inode, path + depth);
3489 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3492 /* Shift the start of ex by 'map_len' blocks */
3493 ex->ee_block = cpu_to_le32(ee_block + map_len);
3494 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3495 ex->ee_len = cpu_to_le16(ee_len - map_len);
3496 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3498 /* Extend abut_ex by 'map_len' blocks */
3499 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3501 /* Result: number of initialized blocks past m_lblk */
3502 allocated = map_len;
3504 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3505 (map_len < ee_len) && /*L1*/
3506 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3507 /* See if we can merge right */
3508 ext4_lblk_t next_lblk;
3509 ext4_fsblk_t next_pblk, ee_pblk;
3510 unsigned int next_len;
3513 next_lblk = le32_to_cpu(abut_ex->ee_block);
3514 next_len = ext4_ext_get_actual_len(abut_ex);
3515 next_pblk = ext4_ext_pblock(abut_ex);
3516 ee_pblk = ext4_ext_pblock(ex);
3519 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3520 * upon those conditions:
3521 * - C1: abut_ex is initialized,
3522 * - C2: abut_ex is logically abutting ex,
3523 * - C3: abut_ex is physically abutting ex,
3524 * - C4: abut_ex can receive the additional blocks without
3525 * overflowing the (initialized) length limit.
3527 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3528 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3529 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3530 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3531 err = ext4_ext_get_access(handle, inode, path + depth);
3535 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3538 /* Shift the start of abut_ex by 'map_len' blocks */
3539 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3540 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3541 ex->ee_len = cpu_to_le16(ee_len - map_len);
3542 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3544 /* Extend abut_ex by 'map_len' blocks */
3545 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3547 /* Result: number of initialized blocks past m_lblk */
3548 allocated = map_len;
3552 /* Mark the block containing both extents as dirty */
3553 ext4_ext_dirty(handle, inode, path + depth);
3555 /* Update path to point to the right extent */
3556 path[depth].p_ext = abut_ex;
3559 allocated = ee_len - (map->m_lblk - ee_block);
3561 WARN_ON(map->m_lblk < ee_block);
3563 * It is safe to convert extent to initialized via explicit
3564 * zeroout only if extent is fully inside i_size or new_size.
3566 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3568 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3569 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3570 (inode->i_sb->s_blocksize_bits - 10);
3572 if (ext4_encrypted_inode(inode))
3575 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3576 if (max_zeroout && (ee_len <= max_zeroout)) {
3577 err = ext4_ext_zeroout(inode, ex);
3580 zero_ex.ee_block = ex->ee_block;
3581 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3582 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3584 err = ext4_ext_get_access(handle, inode, path + depth);
3587 ext4_ext_mark_initialized(ex);
3588 ext4_ext_try_to_merge(handle, inode, path, ex);
3589 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3595 * 1. split the extent into three extents.
3596 * 2. split the extent into two extents, zeroout the first half.
3597 * 3. split the extent into two extents, zeroout the second half.
3598 * 4. split the extent into two extents with out zeroout.
3600 split_map.m_lblk = map->m_lblk;
3601 split_map.m_len = map->m_len;
3603 if (max_zeroout && (allocated > map->m_len)) {
3604 if (allocated <= max_zeroout) {
3607 cpu_to_le32(map->m_lblk);
3608 zero_ex.ee_len = cpu_to_le16(allocated);
3609 ext4_ext_store_pblock(&zero_ex,
3610 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3611 err = ext4_ext_zeroout(inode, &zero_ex);
3614 split_map.m_lblk = map->m_lblk;
3615 split_map.m_len = allocated;
3616 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3618 if (map->m_lblk != ee_block) {
3619 zero_ex.ee_block = ex->ee_block;
3620 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3622 ext4_ext_store_pblock(&zero_ex,
3623 ext4_ext_pblock(ex));
3624 err = ext4_ext_zeroout(inode, &zero_ex);
3629 split_map.m_lblk = ee_block;
3630 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3631 allocated = map->m_len;
3635 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3640 /* If we have gotten a failure, don't zero out status tree */
3642 err = ext4_zeroout_es(inode, &zero_ex);
3643 return err ? err : allocated;
3647 * This function is called by ext4_ext_map_blocks() from
3648 * ext4_get_blocks_dio_write() when DIO to write
3649 * to an unwritten extent.
3651 * Writing to an unwritten extent may result in splitting the unwritten
3652 * extent into multiple initialized/unwritten extents (up to three)
3653 * There are three possibilities:
3654 * a> There is no split required: Entire extent should be unwritten
3655 * b> Splits in two extents: Write is happening at either end of the extent
3656 * c> Splits in three extents: Somone is writing in middle of the extent
3658 * This works the same way in the case of initialized -> unwritten conversion.
3660 * One of more index blocks maybe needed if the extent tree grow after
3661 * the unwritten extent split. To prevent ENOSPC occur at the IO
3662 * complete, we need to split the unwritten extent before DIO submit
3663 * the IO. The unwritten extent called at this time will be split
3664 * into three unwritten extent(at most). After IO complete, the part
3665 * being filled will be convert to initialized by the end_io callback function
3666 * via ext4_convert_unwritten_extents().
3668 * Returns the size of unwritten extent to be written on success.
3670 static int ext4_split_convert_extents(handle_t *handle,
3671 struct inode *inode,
3672 struct ext4_map_blocks *map,
3673 struct ext4_ext_path **ppath,
3676 struct ext4_ext_path *path = *ppath;
3677 ext4_lblk_t eof_block;
3678 ext4_lblk_t ee_block;
3679 struct ext4_extent *ex;
3680 unsigned int ee_len;
3681 int split_flag = 0, depth;
3683 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3684 __func__, inode->i_ino,
3685 (unsigned long long)map->m_lblk, map->m_len);
3687 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3688 inode->i_sb->s_blocksize_bits;
3689 if (eof_block < map->m_lblk + map->m_len)
3690 eof_block = map->m_lblk + map->m_len;
3692 * It is safe to convert extent to initialized via explicit
3693 * zeroout only if extent is fully insde i_size or new_size.
3695 depth = ext_depth(inode);
3696 ex = path[depth].p_ext;
3697 ee_block = le32_to_cpu(ex->ee_block);
3698 ee_len = ext4_ext_get_actual_len(ex);
3700 /* Convert to unwritten */
3701 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3702 split_flag |= EXT4_EXT_DATA_VALID1;
3703 /* Convert to initialized */
3704 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3705 split_flag |= ee_block + ee_len <= eof_block ?
3706 EXT4_EXT_MAY_ZEROOUT : 0;
3707 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3709 flags |= EXT4_GET_BLOCKS_PRE_IO;
3710 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3713 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3714 struct inode *inode,
3715 struct ext4_map_blocks *map,
3716 struct ext4_ext_path **ppath)
3718 struct ext4_ext_path *path = *ppath;
3719 struct ext4_extent *ex;
3720 ext4_lblk_t ee_block;
3721 unsigned int ee_len;
3725 depth = ext_depth(inode);
3726 ex = path[depth].p_ext;
3727 ee_block = le32_to_cpu(ex->ee_block);
3728 ee_len = ext4_ext_get_actual_len(ex);
3730 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3731 "block %llu, max_blocks %u\n", inode->i_ino,
3732 (unsigned long long)ee_block, ee_len);
3734 /* If extent is larger than requested it is a clear sign that we still
3735 * have some extent state machine issues left. So extent_split is still
3737 * TODO: Once all related issues will be fixed this situation should be
3740 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3742 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3743 " len %u; IO logical block %llu, len %u",
3744 inode->i_ino, (unsigned long long)ee_block, ee_len,
3745 (unsigned long long)map->m_lblk, map->m_len);
3747 err = ext4_split_convert_extents(handle, inode, map, ppath,
3748 EXT4_GET_BLOCKS_CONVERT);
3751 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3753 return PTR_ERR(path);
3754 depth = ext_depth(inode);
3755 ex = path[depth].p_ext;
3758 err = ext4_ext_get_access(handle, inode, path + depth);
3761 /* first mark the extent as initialized */
3762 ext4_ext_mark_initialized(ex);
3764 /* note: ext4_ext_correct_indexes() isn't needed here because
3765 * borders are not changed
3767 ext4_ext_try_to_merge(handle, inode, path, ex);
3769 /* Mark modified extent as dirty */
3770 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3772 ext4_ext_show_leaf(inode, path);
3776 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3777 sector_t block, int count)
3780 for (i = 0; i < count; i++)
3781 unmap_underlying_metadata(bdev, block + i);
3785 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3787 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3789 struct ext4_ext_path *path,
3793 struct ext4_extent_header *eh;
3794 struct ext4_extent *last_ex;
3796 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3799 depth = ext_depth(inode);
3800 eh = path[depth].p_hdr;
3803 * We're going to remove EOFBLOCKS_FL entirely in future so we
3804 * do not care for this case anymore. Simply remove the flag
3805 * if there are no extents.
3807 if (unlikely(!eh->eh_entries))
3809 last_ex = EXT_LAST_EXTENT(eh);
3811 * We should clear the EOFBLOCKS_FL flag if we are writing the
3812 * last block in the last extent in the file. We test this by
3813 * first checking to see if the caller to
3814 * ext4_ext_get_blocks() was interested in the last block (or
3815 * a block beyond the last block) in the current extent. If
3816 * this turns out to be false, we can bail out from this
3817 * function immediately.
3819 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3820 ext4_ext_get_actual_len(last_ex))
3823 * If the caller does appear to be planning to write at or
3824 * beyond the end of the current extent, we then test to see
3825 * if the current extent is the last extent in the file, by
3826 * checking to make sure it was reached via the rightmost node
3827 * at each level of the tree.
3829 for (i = depth-1; i >= 0; i--)
3830 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3833 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3834 return ext4_mark_inode_dirty(handle, inode);
3838 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3840 * Return 1 if there is a delalloc block in the range, otherwise 0.
3842 int ext4_find_delalloc_range(struct inode *inode,
3843 ext4_lblk_t lblk_start,
3844 ext4_lblk_t lblk_end)
3846 struct extent_status es;
3848 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3850 return 0; /* there is no delay extent in this tree */
3851 else if (es.es_lblk <= lblk_start &&
3852 lblk_start < es.es_lblk + es.es_len)
3854 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3860 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3862 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3863 ext4_lblk_t lblk_start, lblk_end;
3864 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3865 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3867 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3871 * Determines how many complete clusters (out of those specified by the 'map')
3872 * are under delalloc and were reserved quota for.
3873 * This function is called when we are writing out the blocks that were
3874 * originally written with their allocation delayed, but then the space was
3875 * allocated using fallocate() before the delayed allocation could be resolved.
3876 * The cases to look for are:
3877 * ('=' indicated delayed allocated blocks
3878 * '-' indicates non-delayed allocated blocks)
3879 * (a) partial clusters towards beginning and/or end outside of allocated range
3880 * are not delalloc'ed.
3882 * |----c---=|====c====|====c====|===-c----|
3883 * |++++++ allocated ++++++|
3884 * ==> 4 complete clusters in above example
3886 * (b) partial cluster (outside of allocated range) towards either end is
3887 * marked for delayed allocation. In this case, we will exclude that
3890 * |----====c========|========c========|
3891 * |++++++ allocated ++++++|
3892 * ==> 1 complete clusters in above example
3895 * |================c================|
3896 * |++++++ allocated ++++++|
3897 * ==> 0 complete clusters in above example
3899 * The ext4_da_update_reserve_space will be called only if we
3900 * determine here that there were some "entire" clusters that span
3901 * this 'allocated' range.
3902 * In the non-bigalloc case, this function will just end up returning num_blks
3903 * without ever calling ext4_find_delalloc_range.
3906 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3907 unsigned int num_blks)
3909 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3910 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3911 ext4_lblk_t lblk_from, lblk_to, c_offset;
3912 unsigned int allocated_clusters = 0;
3914 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3915 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3917 /* max possible clusters for this allocation */
3918 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3920 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3922 /* Check towards left side */
3923 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3925 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3926 lblk_to = lblk_from + c_offset - 1;
3928 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3929 allocated_clusters--;
3932 /* Now check towards right. */
3933 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3934 if (allocated_clusters && c_offset) {
3935 lblk_from = lblk_start + num_blks;
3936 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3938 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3939 allocated_clusters--;
3942 return allocated_clusters;
3946 convert_initialized_extent(handle_t *handle, struct inode *inode,
3947 struct ext4_map_blocks *map,
3948 struct ext4_ext_path **ppath,
3949 unsigned int allocated)
3951 struct ext4_ext_path *path = *ppath;
3952 struct ext4_extent *ex;
3953 ext4_lblk_t ee_block;
3954 unsigned int ee_len;
3959 * Make sure that the extent is no bigger than we support with
3962 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3963 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3965 depth = ext_depth(inode);
3966 ex = path[depth].p_ext;
3967 ee_block = le32_to_cpu(ex->ee_block);
3968 ee_len = ext4_ext_get_actual_len(ex);
3970 ext_debug("%s: inode %lu, logical"
3971 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3972 (unsigned long long)ee_block, ee_len);
3974 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3975 err = ext4_split_convert_extents(handle, inode, map, ppath,
3976 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3979 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3981 return PTR_ERR(path);
3982 depth = ext_depth(inode);
3983 ex = path[depth].p_ext;
3985 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3986 (unsigned long) map->m_lblk);
3987 return -EFSCORRUPTED;
3991 err = ext4_ext_get_access(handle, inode, path + depth);
3994 /* first mark the extent as unwritten */
3995 ext4_ext_mark_unwritten(ex);
3997 /* note: ext4_ext_correct_indexes() isn't needed here because
3998 * borders are not changed
4000 ext4_ext_try_to_merge(handle, inode, path, ex);
4002 /* Mark modified extent as dirty */
4003 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4006 ext4_ext_show_leaf(inode, path);
4008 ext4_update_inode_fsync_trans(handle, inode, 1);
4009 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4012 map->m_flags |= EXT4_MAP_UNWRITTEN;
4013 if (allocated > map->m_len)
4014 allocated = map->m_len;
4015 map->m_len = allocated;
4020 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4021 struct ext4_map_blocks *map,
4022 struct ext4_ext_path **ppath, int flags,
4023 unsigned int allocated, ext4_fsblk_t newblock)
4025 struct ext4_ext_path *path = *ppath;
4029 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4030 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4031 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4033 ext4_ext_show_leaf(inode, path);
4036 * When writing into unwritten space, we should not fail to
4037 * allocate metadata blocks for the new extent block if needed.
4039 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4041 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4042 allocated, newblock);
4044 /* get_block() before submit the IO, split the extent */
4045 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4046 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4047 flags | EXT4_GET_BLOCKS_CONVERT);
4050 map->m_flags |= EXT4_MAP_UNWRITTEN;
4053 /* IO end_io complete, convert the filled extent to written */
4054 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4055 if (flags & EXT4_GET_BLOCKS_ZERO) {
4056 if (allocated > map->m_len)
4057 allocated = map->m_len;
4058 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4063 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4066 ext4_update_inode_fsync_trans(handle, inode, 1);
4067 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4071 map->m_flags |= EXT4_MAP_MAPPED;
4072 map->m_pblk = newblock;
4073 if (allocated > map->m_len)
4074 allocated = map->m_len;
4075 map->m_len = allocated;
4078 /* buffered IO case */
4080 * repeat fallocate creation request
4081 * we already have an unwritten extent
4083 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4084 map->m_flags |= EXT4_MAP_UNWRITTEN;
4088 /* buffered READ or buffered write_begin() lookup */
4089 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4091 * We have blocks reserved already. We
4092 * return allocated blocks so that delalloc
4093 * won't do block reservation for us. But
4094 * the buffer head will be unmapped so that
4095 * a read from the block returns 0s.
4097 map->m_flags |= EXT4_MAP_UNWRITTEN;
4101 /* buffered write, writepage time, convert*/
4102 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4104 ext4_update_inode_fsync_trans(handle, inode, 1);
4111 map->m_flags |= EXT4_MAP_NEW;
4113 * if we allocated more blocks than requested
4114 * we need to make sure we unmap the extra block
4115 * allocated. The actual needed block will get
4116 * unmapped later when we find the buffer_head marked
4119 if (allocated > map->m_len) {
4120 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4121 newblock + map->m_len,
4122 allocated - map->m_len);
4123 allocated = map->m_len;
4125 map->m_len = allocated;
4128 * If we have done fallocate with the offset that is already
4129 * delayed allocated, we would have block reservation
4130 * and quota reservation done in the delayed write path.
4131 * But fallocate would have already updated quota and block
4132 * count for this offset. So cancel these reservation
4134 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4135 unsigned int reserved_clusters;
4136 reserved_clusters = get_reserved_cluster_alloc(inode,
4137 map->m_lblk, map->m_len);
4138 if (reserved_clusters)
4139 ext4_da_update_reserve_space(inode,
4145 map->m_flags |= EXT4_MAP_MAPPED;
4146 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4147 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4153 if (allocated > map->m_len)
4154 allocated = map->m_len;
4155 ext4_ext_show_leaf(inode, path);
4156 map->m_pblk = newblock;
4157 map->m_len = allocated;
4159 return err ? err : allocated;
4163 * get_implied_cluster_alloc - check to see if the requested
4164 * allocation (in the map structure) overlaps with a cluster already
4165 * allocated in an extent.
4166 * @sb The filesystem superblock structure
4167 * @map The requested lblk->pblk mapping
4168 * @ex The extent structure which might contain an implied
4169 * cluster allocation
4171 * This function is called by ext4_ext_map_blocks() after we failed to
4172 * find blocks that were already in the inode's extent tree. Hence,
4173 * we know that the beginning of the requested region cannot overlap
4174 * the extent from the inode's extent tree. There are three cases we
4175 * want to catch. The first is this case:
4177 * |--- cluster # N--|
4178 * |--- extent ---| |---- requested region ---|
4181 * The second case that we need to test for is this one:
4183 * |--------- cluster # N ----------------|
4184 * |--- requested region --| |------- extent ----|
4185 * |=======================|
4187 * The third case is when the requested region lies between two extents
4188 * within the same cluster:
4189 * |------------- cluster # N-------------|
4190 * |----- ex -----| |---- ex_right ----|
4191 * |------ requested region ------|
4192 * |================|
4194 * In each of the above cases, we need to set the map->m_pblk and
4195 * map->m_len so it corresponds to the return the extent labelled as
4196 * "|====|" from cluster #N, since it is already in use for data in
4197 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4198 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4199 * as a new "allocated" block region. Otherwise, we will return 0 and
4200 * ext4_ext_map_blocks() will then allocate one or more new clusters
4201 * by calling ext4_mb_new_blocks().
4203 static int get_implied_cluster_alloc(struct super_block *sb,
4204 struct ext4_map_blocks *map,
4205 struct ext4_extent *ex,
4206 struct ext4_ext_path *path)
4208 struct ext4_sb_info *sbi = EXT4_SB(sb);
4209 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4210 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4211 ext4_lblk_t rr_cluster_start;
4212 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4213 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4214 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4216 /* The extent passed in that we are trying to match */
4217 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4218 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4220 /* The requested region passed into ext4_map_blocks() */
4221 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4223 if ((rr_cluster_start == ex_cluster_end) ||
4224 (rr_cluster_start == ex_cluster_start)) {
4225 if (rr_cluster_start == ex_cluster_end)
4226 ee_start += ee_len - 1;
4227 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4228 map->m_len = min(map->m_len,
4229 (unsigned) sbi->s_cluster_ratio - c_offset);
4231 * Check for and handle this case:
4233 * |--------- cluster # N-------------|
4234 * |------- extent ----|
4235 * |--- requested region ---|
4239 if (map->m_lblk < ee_block)
4240 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4243 * Check for the case where there is already another allocated
4244 * block to the right of 'ex' but before the end of the cluster.
4246 * |------------- cluster # N-------------|
4247 * |----- ex -----| |---- ex_right ----|
4248 * |------ requested region ------|
4249 * |================|
4251 if (map->m_lblk > ee_block) {
4252 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4253 map->m_len = min(map->m_len, next - map->m_lblk);
4256 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4260 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4266 * Block allocation/map/preallocation routine for extents based files
4269 * Need to be called with
4270 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4271 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4273 * return > 0, number of of blocks already mapped/allocated
4274 * if create == 0 and these are pre-allocated blocks
4275 * buffer head is unmapped
4276 * otherwise blocks are mapped
4278 * return = 0, if plain look up failed (blocks have not been allocated)
4279 * buffer head is unmapped
4281 * return < 0, error case.
4283 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4284 struct ext4_map_blocks *map, int flags)
4286 struct ext4_ext_path *path = NULL;
4287 struct ext4_extent newex, *ex, *ex2;
4288 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4289 ext4_fsblk_t newblock = 0;
4290 int free_on_err = 0, err = 0, depth, ret;
4291 unsigned int allocated = 0, offset = 0;
4292 unsigned int allocated_clusters = 0;
4293 struct ext4_allocation_request ar;
4294 ext4_lblk_t cluster_offset;
4295 bool map_from_cluster = false;
4297 ext_debug("blocks %u/%u requested for inode %lu\n",
4298 map->m_lblk, map->m_len, inode->i_ino);
4299 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4301 /* find extent for this block */
4302 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4304 err = PTR_ERR(path);
4309 depth = ext_depth(inode);
4312 * consistent leaf must not be empty;
4313 * this situation is possible, though, _during_ tree modification;
4314 * this is why assert can't be put in ext4_find_extent()
4316 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4317 EXT4_ERROR_INODE(inode, "bad extent address "
4318 "lblock: %lu, depth: %d pblock %lld",
4319 (unsigned long) map->m_lblk, depth,
4320 path[depth].p_block);
4321 err = -EFSCORRUPTED;
4325 ex = path[depth].p_ext;
4327 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4328 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4329 unsigned short ee_len;
4333 * unwritten extents are treated as holes, except that
4334 * we split out initialized portions during a write.
4336 ee_len = ext4_ext_get_actual_len(ex);
4338 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4340 /* if found extent covers block, simply return it */
4341 if (in_range(map->m_lblk, ee_block, ee_len)) {
4342 newblock = map->m_lblk - ee_block + ee_start;
4343 /* number of remaining blocks in the extent */
4344 allocated = ee_len - (map->m_lblk - ee_block);
4345 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4346 ee_block, ee_len, newblock);
4349 * If the extent is initialized check whether the
4350 * caller wants to convert it to unwritten.
4352 if ((!ext4_ext_is_unwritten(ex)) &&
4353 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4354 allocated = convert_initialized_extent(
4355 handle, inode, map, &path,
4358 } else if (!ext4_ext_is_unwritten(ex))
4361 ret = ext4_ext_handle_unwritten_extents(
4362 handle, inode, map, &path, flags,
4363 allocated, newblock);
4373 * requested block isn't allocated yet;
4374 * we couldn't try to create block if create flag is zero
4376 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4377 ext4_lblk_t hole_start, hole_len;
4379 hole_start = map->m_lblk;
4380 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4382 * put just found gap into cache to speed up
4383 * subsequent requests
4385 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4387 /* Update hole_len to reflect hole size after map->m_lblk */
4388 if (hole_start != map->m_lblk)
4389 hole_len -= map->m_lblk - hole_start;
4391 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4397 * Okay, we need to do block allocation.
4399 newex.ee_block = cpu_to_le32(map->m_lblk);
4400 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4403 * If we are doing bigalloc, check to see if the extent returned
4404 * by ext4_find_extent() implies a cluster we can use.
4406 if (cluster_offset && ex &&
4407 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4408 ar.len = allocated = map->m_len;
4409 newblock = map->m_pblk;
4410 map_from_cluster = true;
4411 goto got_allocated_blocks;
4414 /* find neighbour allocated blocks */
4415 ar.lleft = map->m_lblk;
4416 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4419 ar.lright = map->m_lblk;
4421 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4425 /* Check if the extent after searching to the right implies a
4426 * cluster we can use. */
4427 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4428 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4429 ar.len = allocated = map->m_len;
4430 newblock = map->m_pblk;
4431 map_from_cluster = true;
4432 goto got_allocated_blocks;
4436 * See if request is beyond maximum number of blocks we can have in
4437 * a single extent. For an initialized extent this limit is
4438 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4439 * EXT_UNWRITTEN_MAX_LEN.
4441 if (map->m_len > EXT_INIT_MAX_LEN &&
4442 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4443 map->m_len = EXT_INIT_MAX_LEN;
4444 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4445 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4446 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4448 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4449 newex.ee_len = cpu_to_le16(map->m_len);
4450 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4452 allocated = ext4_ext_get_actual_len(&newex);
4454 allocated = map->m_len;
4456 /* allocate new block */
4458 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4459 ar.logical = map->m_lblk;
4461 * We calculate the offset from the beginning of the cluster
4462 * for the logical block number, since when we allocate a
4463 * physical cluster, the physical block should start at the
4464 * same offset from the beginning of the cluster. This is
4465 * needed so that future calls to get_implied_cluster_alloc()
4468 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4469 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4471 ar.logical -= offset;
4472 if (S_ISREG(inode->i_mode))
4473 ar.flags = EXT4_MB_HINT_DATA;
4475 /* disable in-core preallocation for non-regular files */
4477 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4478 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4479 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4480 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4481 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4482 ar.flags |= EXT4_MB_USE_RESERVED;
4483 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4486 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4487 ar.goal, newblock, allocated);
4489 allocated_clusters = ar.len;
4490 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4491 if (ar.len > allocated)
4494 got_allocated_blocks:
4495 /* try to insert new extent into found leaf and return */
4496 ext4_ext_store_pblock(&newex, newblock + offset);
4497 newex.ee_len = cpu_to_le16(ar.len);
4498 /* Mark unwritten */
4499 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4500 ext4_ext_mark_unwritten(&newex);
4501 map->m_flags |= EXT4_MAP_UNWRITTEN;
4505 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4506 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4509 err = ext4_ext_insert_extent(handle, inode, &path,
4512 if (err && free_on_err) {
4513 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4514 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4515 /* free data blocks we just allocated */
4516 /* not a good idea to call discard here directly,
4517 * but otherwise we'd need to call it every free() */
4518 ext4_discard_preallocations(inode);
4519 ext4_free_blocks(handle, inode, NULL, newblock,
4520 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4524 /* previous routine could use block we allocated */
4525 newblock = ext4_ext_pblock(&newex);
4526 allocated = ext4_ext_get_actual_len(&newex);
4527 if (allocated > map->m_len)
4528 allocated = map->m_len;
4529 map->m_flags |= EXT4_MAP_NEW;
4532 * Update reserved blocks/metadata blocks after successful
4533 * block allocation which had been deferred till now.
4535 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4536 unsigned int reserved_clusters;
4538 * Check how many clusters we had reserved this allocated range
4540 reserved_clusters = get_reserved_cluster_alloc(inode,
4541 map->m_lblk, allocated);
4542 if (!map_from_cluster) {
4543 BUG_ON(allocated_clusters < reserved_clusters);
4544 if (reserved_clusters < allocated_clusters) {
4545 struct ext4_inode_info *ei = EXT4_I(inode);
4546 int reservation = allocated_clusters -
4549 * It seems we claimed few clusters outside of
4550 * the range of this allocation. We should give
4551 * it back to the reservation pool. This can
4552 * happen in the following case:
4554 * * Suppose s_cluster_ratio is 4 (i.e., each
4555 * cluster has 4 blocks. Thus, the clusters
4556 * are [0-3],[4-7],[8-11]...
4557 * * First comes delayed allocation write for
4558 * logical blocks 10 & 11. Since there were no
4559 * previous delayed allocated blocks in the
4560 * range [8-11], we would reserve 1 cluster
4562 * * Next comes write for logical blocks 3 to 8.
4563 * In this case, we will reserve 2 clusters
4564 * (for [0-3] and [4-7]; and not for [8-11] as
4565 * that range has a delayed allocated blocks.
4566 * Thus total reserved clusters now becomes 3.
4567 * * Now, during the delayed allocation writeout
4568 * time, we will first write blocks [3-8] and
4569 * allocate 3 clusters for writing these
4570 * blocks. Also, we would claim all these
4571 * three clusters above.
4572 * * Now when we come here to writeout the
4573 * blocks [10-11], we would expect to claim
4574 * the reservation of 1 cluster we had made
4575 * (and we would claim it since there are no
4576 * more delayed allocated blocks in the range
4577 * [8-11]. But our reserved cluster count had
4578 * already gone to 0.
4580 * Thus, at the step 4 above when we determine
4581 * that there are still some unwritten delayed
4582 * allocated blocks outside of our current
4583 * block range, we should increment the
4584 * reserved clusters count so that when the
4585 * remaining blocks finally gets written, we
4588 dquot_reserve_block(inode,
4589 EXT4_C2B(sbi, reservation));
4590 spin_lock(&ei->i_block_reservation_lock);
4591 ei->i_reserved_data_blocks += reservation;
4592 spin_unlock(&ei->i_block_reservation_lock);
4595 * We will claim quota for all newly allocated blocks.
4596 * We're updating the reserved space *after* the
4597 * correction above so we do not accidentally free
4598 * all the metadata reservation because we might
4599 * actually need it later on.
4601 ext4_da_update_reserve_space(inode, allocated_clusters,
4607 * Cache the extent and update transaction to commit on fdatasync only
4608 * when it is _not_ an unwritten extent.
4610 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4611 ext4_update_inode_fsync_trans(handle, inode, 1);
4613 ext4_update_inode_fsync_trans(handle, inode, 0);
4615 if (allocated > map->m_len)
4616 allocated = map->m_len;
4617 ext4_ext_show_leaf(inode, path);
4618 map->m_flags |= EXT4_MAP_MAPPED;
4619 map->m_pblk = newblock;
4620 map->m_len = allocated;
4622 ext4_ext_drop_refs(path);
4625 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4626 err ? err : allocated);
4627 return err ? err : allocated;
4630 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4632 struct super_block *sb = inode->i_sb;
4633 ext4_lblk_t last_block;
4637 * TODO: optimization is possible here.
4638 * Probably we need not scan at all,
4639 * because page truncation is enough.
4642 /* we have to know where to truncate from in crash case */
4643 EXT4_I(inode)->i_disksize = inode->i_size;
4644 ext4_mark_inode_dirty(handle, inode);
4646 last_block = (inode->i_size + sb->s_blocksize - 1)
4647 >> EXT4_BLOCK_SIZE_BITS(sb);
4649 err = ext4_es_remove_extent(inode, last_block,
4650 EXT_MAX_BLOCKS - last_block);
4651 if (err == -ENOMEM) {
4653 congestion_wait(BLK_RW_ASYNC, HZ/50);
4657 ext4_std_error(inode->i_sb, err);
4660 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4661 ext4_std_error(inode->i_sb, err);
4664 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4665 ext4_lblk_t len, loff_t new_size,
4666 int flags, int mode)
4668 struct inode *inode = file_inode(file);
4674 struct ext4_map_blocks map;
4675 unsigned int credits;
4678 map.m_lblk = offset;
4681 * Don't normalize the request if it can fit in one extent so
4682 * that it doesn't get unnecessarily split into multiple
4685 if (len <= EXT_UNWRITTEN_MAX_LEN)
4686 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4689 * credits to insert 1 extent into extent tree
4691 credits = ext4_chunk_trans_blocks(inode, len);
4693 * We can only call ext_depth() on extent based inodes
4695 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4696 depth = ext_depth(inode);
4701 while (ret >= 0 && len) {
4703 * Recalculate credits when extent tree depth changes.
4705 if (depth >= 0 && depth != ext_depth(inode)) {
4706 credits = ext4_chunk_trans_blocks(inode, len);
4707 depth = ext_depth(inode);
4710 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4712 if (IS_ERR(handle)) {
4713 ret = PTR_ERR(handle);
4716 ret = ext4_map_blocks(handle, inode, &map, flags);
4718 ext4_debug("inode #%lu: block %u: len %u: "
4719 "ext4_ext_map_blocks returned %d",
4720 inode->i_ino, map.m_lblk,
4722 ext4_mark_inode_dirty(handle, inode);
4723 ret2 = ext4_journal_stop(handle);
4727 map.m_len = len = len - ret;
4728 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4729 inode->i_ctime = ext4_current_time(inode);
4731 if (epos > new_size)
4733 if (ext4_update_inode_size(inode, epos) & 0x1)
4734 inode->i_mtime = inode->i_ctime;
4736 if (epos > inode->i_size)
4737 ext4_set_inode_flag(inode,
4738 EXT4_INODE_EOFBLOCKS);
4740 ext4_mark_inode_dirty(handle, inode);
4741 ret2 = ext4_journal_stop(handle);
4745 if (ret == -ENOSPC &&
4746 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4751 return ret > 0 ? ret2 : ret;
4754 static long ext4_zero_range(struct file *file, loff_t offset,
4755 loff_t len, int mode)
4757 struct inode *inode = file_inode(file);
4758 handle_t *handle = NULL;
4759 unsigned int max_blocks;
4760 loff_t new_size = 0;
4764 int partial_begin, partial_end;
4767 unsigned int blkbits = inode->i_blkbits;
4769 trace_ext4_zero_range(inode, offset, len, mode);
4771 if (!S_ISREG(inode->i_mode))
4774 /* Call ext4_force_commit to flush all data in case of data=journal. */
4775 if (ext4_should_journal_data(inode)) {
4776 ret = ext4_force_commit(inode->i_sb);
4782 * Round up offset. This is not fallocate, we neet to zero out
4783 * blocks, so convert interior block aligned part of the range to
4784 * unwritten and possibly manually zero out unaligned parts of the
4787 start = round_up(offset, 1 << blkbits);
4788 end = round_down((offset + len), 1 << blkbits);
4790 if (start < offset || end > offset + len)
4792 partial_begin = offset & ((1 << blkbits) - 1);
4793 partial_end = (offset + len) & ((1 << blkbits) - 1);
4795 lblk = start >> blkbits;
4796 max_blocks = (end >> blkbits);
4797 if (max_blocks < lblk)
4805 * Indirect files do not support unwritten extnets
4807 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4812 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4813 offset + len > i_size_read(inode)) {
4814 new_size = offset + len;
4815 ret = inode_newsize_ok(inode, new_size);
4820 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4821 if (mode & FALLOC_FL_KEEP_SIZE)
4822 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4824 /* Wait all existing dio workers, newcomers will block on i_mutex */
4825 ext4_inode_block_unlocked_dio(inode);
4826 inode_dio_wait(inode);
4828 /* Preallocate the range including the unaligned edges */
4829 if (partial_begin || partial_end) {
4830 ret = ext4_alloc_file_blocks(file,
4831 round_down(offset, 1 << blkbits) >> blkbits,
4832 (round_up((offset + len), 1 << blkbits) -
4833 round_down(offset, 1 << blkbits)) >> blkbits,
4834 new_size, flags, mode);
4840 /* Zero range excluding the unaligned edges */
4841 if (max_blocks > 0) {
4842 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4846 * Prevent page faults from reinstantiating pages we have
4847 * released from page cache.
4849 down_write(&EXT4_I(inode)->i_mmap_sem);
4850 ret = ext4_update_disksize_before_punch(inode, offset, len);
4852 up_write(&EXT4_I(inode)->i_mmap_sem);
4855 /* Now release the pages and zero block aligned part of pages */
4856 truncate_pagecache_range(inode, start, end - 1);
4857 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4859 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4861 up_write(&EXT4_I(inode)->i_mmap_sem);
4865 if (!partial_begin && !partial_end)
4869 * In worst case we have to writeout two nonadjacent unwritten
4870 * blocks and update the inode
4872 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4873 if (ext4_should_journal_data(inode))
4875 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4876 if (IS_ERR(handle)) {
4877 ret = PTR_ERR(handle);
4878 ext4_std_error(inode->i_sb, ret);
4882 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4884 ext4_update_inode_size(inode, new_size);
4887 * Mark that we allocate beyond EOF so the subsequent truncate
4888 * can proceed even if the new size is the same as i_size.
4890 if ((offset + len) > i_size_read(inode))
4891 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4893 ext4_mark_inode_dirty(handle, inode);
4895 /* Zero out partial block at the edges of the range */
4896 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4898 if (file->f_flags & O_SYNC)
4899 ext4_handle_sync(handle);
4901 ext4_journal_stop(handle);
4903 ext4_inode_resume_unlocked_dio(inode);
4905 inode_unlock(inode);
4910 * preallocate space for a file. This implements ext4's fallocate file
4911 * operation, which gets called from sys_fallocate system call.
4912 * For block-mapped files, posix_fallocate should fall back to the method
4913 * of writing zeroes to the required new blocks (the same behavior which is
4914 * expected for file systems which do not support fallocate() system call).
4916 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4918 struct inode *inode = file_inode(file);
4919 loff_t new_size = 0;
4920 unsigned int max_blocks;
4924 unsigned int blkbits = inode->i_blkbits;
4927 * Encrypted inodes can't handle collapse range or insert
4928 * range since we would need to re-encrypt blocks with a
4929 * different IV or XTS tweak (which are based on the logical
4932 * XXX It's not clear why zero range isn't working, but we'll
4933 * leave it disabled for encrypted inodes for now. This is a
4934 * bug we should fix....
4936 if (ext4_encrypted_inode(inode) &&
4937 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4938 FALLOC_FL_ZERO_RANGE)))
4941 /* Return error if mode is not supported */
4942 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4943 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4944 FALLOC_FL_INSERT_RANGE))
4947 if (mode & FALLOC_FL_PUNCH_HOLE)
4948 return ext4_punch_hole(inode, offset, len);
4950 ret = ext4_convert_inline_data(inode);
4954 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4955 return ext4_collapse_range(inode, offset, len);
4957 if (mode & FALLOC_FL_INSERT_RANGE)
4958 return ext4_insert_range(inode, offset, len);
4960 if (mode & FALLOC_FL_ZERO_RANGE)
4961 return ext4_zero_range(file, offset, len, mode);
4963 trace_ext4_fallocate_enter(inode, offset, len, mode);
4964 lblk = offset >> blkbits;
4966 * We can't just convert len to max_blocks because
4967 * If blocksize = 4096 offset = 3072 and len = 2048
4969 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4972 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4973 if (mode & FALLOC_FL_KEEP_SIZE)
4974 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4979 * We only support preallocation for extent-based files only
4981 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4986 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4987 offset + len > i_size_read(inode)) {
4988 new_size = offset + len;
4989 ret = inode_newsize_ok(inode, new_size);
4994 /* Wait all existing dio workers, newcomers will block on i_mutex */
4995 ext4_inode_block_unlocked_dio(inode);
4996 inode_dio_wait(inode);
4998 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
5000 ext4_inode_resume_unlocked_dio(inode);
5004 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
5005 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5006 EXT4_I(inode)->i_sync_tid);
5009 inode_unlock(inode);
5010 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
5015 * This function convert a range of blocks to written extents
5016 * The caller of this function will pass the start offset and the size.
5017 * all unwritten extents within this range will be converted to
5020 * This function is called from the direct IO end io call back
5021 * function, to convert the fallocated extents after IO is completed.
5022 * Returns 0 on success.
5024 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5025 loff_t offset, ssize_t len)
5027 unsigned int max_blocks;
5030 struct ext4_map_blocks map;
5031 unsigned int credits, blkbits = inode->i_blkbits;
5033 map.m_lblk = offset >> blkbits;
5035 * We can't just convert len to max_blocks because
5036 * If blocksize = 4096 offset = 3072 and len = 2048
5038 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5041 * This is somewhat ugly but the idea is clear: When transaction is
5042 * reserved, everything goes into it. Otherwise we rather start several
5043 * smaller transactions for conversion of each extent separately.
5046 handle = ext4_journal_start_reserved(handle,
5047 EXT4_HT_EXT_CONVERT);
5049 return PTR_ERR(handle);
5053 * credits to insert 1 extent into extent tree
5055 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5057 while (ret >= 0 && ret < max_blocks) {
5059 map.m_len = (max_blocks -= ret);
5061 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5063 if (IS_ERR(handle)) {
5064 ret = PTR_ERR(handle);
5068 ret = ext4_map_blocks(handle, inode, &map,
5069 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5071 ext4_warning(inode->i_sb,
5072 "inode #%lu: block %u: len %u: "
5073 "ext4_ext_map_blocks returned %d",
5074 inode->i_ino, map.m_lblk,
5076 ext4_mark_inode_dirty(handle, inode);
5078 ret2 = ext4_journal_stop(handle);
5079 if (ret <= 0 || ret2)
5083 ret2 = ext4_journal_stop(handle);
5084 return ret > 0 ? ret2 : ret;
5088 * If newes is not existing extent (newes->ec_pblk equals zero) find
5089 * delayed extent at start of newes and update newes accordingly and
5090 * return start of the next delayed extent.
5092 * If newes is existing extent (newes->ec_pblk is not equal zero)
5093 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5094 * extent found. Leave newes unmodified.
5096 static int ext4_find_delayed_extent(struct inode *inode,
5097 struct extent_status *newes)
5099 struct extent_status es;
5100 ext4_lblk_t block, next_del;
5102 if (newes->es_pblk == 0) {
5103 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5104 newes->es_lblk + newes->es_len - 1, &es);
5107 * No extent in extent-tree contains block @newes->es_pblk,
5108 * then the block may stay in 1)a hole or 2)delayed-extent.
5114 if (es.es_lblk > newes->es_lblk) {
5116 newes->es_len = min(es.es_lblk - newes->es_lblk,
5121 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5124 block = newes->es_lblk + newes->es_len;
5125 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5127 next_del = EXT_MAX_BLOCKS;
5129 next_del = es.es_lblk;
5133 /* fiemap flags we can handle specified here */
5134 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5136 static int ext4_xattr_fiemap(struct inode *inode,
5137 struct fiemap_extent_info *fieinfo)
5141 __u32 flags = FIEMAP_EXTENT_LAST;
5142 int blockbits = inode->i_sb->s_blocksize_bits;
5146 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5147 struct ext4_iloc iloc;
5148 int offset; /* offset of xattr in inode */
5150 error = ext4_get_inode_loc(inode, &iloc);
5153 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5154 offset = EXT4_GOOD_OLD_INODE_SIZE +
5155 EXT4_I(inode)->i_extra_isize;
5157 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5158 flags |= FIEMAP_EXTENT_DATA_INLINE;
5160 } else { /* external block */
5161 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5162 length = inode->i_sb->s_blocksize;
5166 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5168 return (error < 0 ? error : 0);
5171 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5172 __u64 start, __u64 len)
5174 ext4_lblk_t start_blk;
5177 if (ext4_has_inline_data(inode)) {
5180 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5187 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5188 error = ext4_ext_precache(inode);
5193 /* fallback to generic here if not in extents fmt */
5194 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5195 return generic_block_fiemap(inode, fieinfo, start, len,
5198 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5201 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5202 error = ext4_xattr_fiemap(inode, fieinfo);
5204 ext4_lblk_t len_blks;
5207 start_blk = start >> inode->i_sb->s_blocksize_bits;
5208 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5209 if (last_blk >= EXT_MAX_BLOCKS)
5210 last_blk = EXT_MAX_BLOCKS-1;
5211 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5214 * Walk the extent tree gathering extent information
5215 * and pushing extents back to the user.
5217 error = ext4_fill_fiemap_extents(inode, start_blk,
5225 * Function to access the path buffer for marking it dirty.
5226 * It also checks if there are sufficient credits left in the journal handle
5230 ext4_access_path(handle_t *handle, struct inode *inode,
5231 struct ext4_ext_path *path)
5235 if (!ext4_handle_valid(handle))
5239 * Check if need to extend journal credits
5240 * 3 for leaf, sb, and inode plus 2 (bmap and group
5241 * descriptor) for each block group; assume two block
5244 if (handle->h_buffer_credits < 7) {
5245 credits = ext4_writepage_trans_blocks(inode);
5246 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5247 /* EAGAIN is success */
5248 if (err && err != -EAGAIN)
5252 err = ext4_ext_get_access(handle, inode, path);
5257 * ext4_ext_shift_path_extents:
5258 * Shift the extents of a path structure lying between path[depth].p_ext
5259 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5260 * if it is right shift or left shift operation.
5263 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5264 struct inode *inode, handle_t *handle,
5265 enum SHIFT_DIRECTION SHIFT)
5268 struct ext4_extent *ex_start, *ex_last;
5270 depth = path->p_depth;
5272 while (depth >= 0) {
5273 if (depth == path->p_depth) {
5274 ex_start = path[depth].p_ext;
5276 return -EFSCORRUPTED;
5278 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5280 err = ext4_access_path(handle, inode, path + depth);
5284 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5287 while (ex_start <= ex_last) {
5288 if (SHIFT == SHIFT_LEFT) {
5289 le32_add_cpu(&ex_start->ee_block,
5291 /* Try to merge to the left. */
5293 EXT_FIRST_EXTENT(path[depth].p_hdr))
5295 ext4_ext_try_to_merge_right(inode,
5296 path, ex_start - 1))
5301 le32_add_cpu(&ex_last->ee_block, shift);
5302 ext4_ext_try_to_merge_right(inode, path,
5307 err = ext4_ext_dirty(handle, inode, path + depth);
5311 if (--depth < 0 || !update)
5315 /* Update index too */
5316 err = ext4_access_path(handle, inode, path + depth);
5320 if (SHIFT == SHIFT_LEFT)
5321 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5323 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5324 err = ext4_ext_dirty(handle, inode, path + depth);
5328 /* we are done if current index is not a starting index */
5329 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5340 * ext4_ext_shift_extents:
5341 * All the extents which lies in the range from @start to the last allocated
5342 * block for the @inode are shifted either towards left or right (depending
5343 * upon @SHIFT) by @shift blocks.
5344 * On success, 0 is returned, error otherwise.
5347 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5348 ext4_lblk_t start, ext4_lblk_t shift,
5349 enum SHIFT_DIRECTION SHIFT)
5351 struct ext4_ext_path *path;
5353 struct ext4_extent *extent;
5354 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5356 /* Let path point to the last extent */
5357 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5359 return PTR_ERR(path);
5361 depth = path->p_depth;
5362 extent = path[depth].p_ext;
5366 stop = le32_to_cpu(extent->ee_block) +
5367 ext4_ext_get_actual_len(extent);
5370 * In case of left shift, Don't start shifting extents until we make
5371 * sure the hole is big enough to accommodate the shift.
5373 if (SHIFT == SHIFT_LEFT) {
5374 path = ext4_find_extent(inode, start - 1, &path, 0);
5376 return PTR_ERR(path);
5377 depth = path->p_depth;
5378 extent = path[depth].p_ext;
5380 ex_start = le32_to_cpu(extent->ee_block);
5381 ex_end = le32_to_cpu(extent->ee_block) +
5382 ext4_ext_get_actual_len(extent);
5388 if ((start == ex_start && shift > ex_start) ||
5389 (shift > start - ex_end)) {
5390 ext4_ext_drop_refs(path);
5397 * In case of left shift, iterator points to start and it is increased
5398 * till we reach stop. In case of right shift, iterator points to stop
5399 * and it is decreased till we reach start.
5401 if (SHIFT == SHIFT_LEFT)
5406 /* Its safe to start updating extents */
5407 while (start < stop) {
5408 path = ext4_find_extent(inode, *iterator, &path, 0);
5410 return PTR_ERR(path);
5411 depth = path->p_depth;
5412 extent = path[depth].p_ext;
5414 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5415 (unsigned long) *iterator);
5416 return -EFSCORRUPTED;
5418 if (SHIFT == SHIFT_LEFT && *iterator >
5419 le32_to_cpu(extent->ee_block)) {
5420 /* Hole, move to the next extent */
5421 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5422 path[depth].p_ext++;
5424 *iterator = ext4_ext_next_allocated_block(path);
5429 if (SHIFT == SHIFT_LEFT) {
5430 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5431 *iterator = le32_to_cpu(extent->ee_block) +
5432 ext4_ext_get_actual_len(extent);
5434 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5435 *iterator = le32_to_cpu(extent->ee_block) > 0 ?
5436 le32_to_cpu(extent->ee_block) - 1 : 0;
5437 /* Update path extent in case we need to stop */
5438 while (le32_to_cpu(extent->ee_block) < start)
5440 path[depth].p_ext = extent;
5442 ret = ext4_ext_shift_path_extents(path, shift, inode,
5448 ext4_ext_drop_refs(path);
5454 * ext4_collapse_range:
5455 * This implements the fallocate's collapse range functionality for ext4
5456 * Returns: 0 and non-zero on error.
5458 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5460 struct super_block *sb = inode->i_sb;
5461 ext4_lblk_t punch_start, punch_stop;
5463 unsigned int credits;
5464 loff_t new_size, ioffset;
5468 * We need to test this early because xfstests assumes that a
5469 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5470 * system does not support collapse range.
5472 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5475 /* Collapse range works only on fs block size aligned offsets. */
5476 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5477 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5480 if (!S_ISREG(inode->i_mode))
5483 trace_ext4_collapse_range(inode, offset, len);
5485 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5486 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5488 /* Call ext4_force_commit to flush all data in case of data=journal. */
5489 if (ext4_should_journal_data(inode)) {
5490 ret = ext4_force_commit(inode->i_sb);
5497 * There is no need to overlap collapse range with EOF, in which case
5498 * it is effectively a truncate operation
5500 if (offset + len >= i_size_read(inode)) {
5505 /* Currently just for extent based files */
5506 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5511 /* Wait for existing dio to complete */
5512 ext4_inode_block_unlocked_dio(inode);
5513 inode_dio_wait(inode);
5516 * Prevent page faults from reinstantiating pages we have released from
5519 down_write(&EXT4_I(inode)->i_mmap_sem);
5521 * Need to round down offset to be aligned with page size boundary
5522 * for page size > block size.
5524 ioffset = round_down(offset, PAGE_SIZE);
5526 * Write tail of the last page before removed range since it will get
5527 * removed from the page cache below.
5529 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5533 * Write data that will be shifted to preserve them when discarding
5534 * page cache below. We are also protected from pages becoming dirty
5537 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5541 truncate_pagecache(inode, ioffset);
5543 credits = ext4_writepage_trans_blocks(inode);
5544 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5545 if (IS_ERR(handle)) {
5546 ret = PTR_ERR(handle);
5550 down_write(&EXT4_I(inode)->i_data_sem);
5551 ext4_discard_preallocations(inode);
5553 ret = ext4_es_remove_extent(inode, punch_start,
5554 EXT_MAX_BLOCKS - punch_start);
5556 up_write(&EXT4_I(inode)->i_data_sem);
5560 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5562 up_write(&EXT4_I(inode)->i_data_sem);
5565 ext4_discard_preallocations(inode);
5567 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5568 punch_stop - punch_start, SHIFT_LEFT);
5570 up_write(&EXT4_I(inode)->i_data_sem);
5574 new_size = i_size_read(inode) - len;
5575 i_size_write(inode, new_size);
5576 EXT4_I(inode)->i_disksize = new_size;
5578 up_write(&EXT4_I(inode)->i_data_sem);
5580 ext4_handle_sync(handle);
5581 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5582 ext4_mark_inode_dirty(handle, inode);
5585 ext4_journal_stop(handle);
5587 up_write(&EXT4_I(inode)->i_mmap_sem);
5588 ext4_inode_resume_unlocked_dio(inode);
5590 inode_unlock(inode);
5595 * ext4_insert_range:
5596 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5597 * The data blocks starting from @offset to the EOF are shifted by @len
5598 * towards right to create a hole in the @inode. Inode size is increased
5600 * Returns 0 on success, error otherwise.
5602 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5604 struct super_block *sb = inode->i_sb;
5606 struct ext4_ext_path *path;
5607 struct ext4_extent *extent;
5608 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5609 unsigned int credits, ee_len;
5610 int ret = 0, depth, split_flag = 0;
5614 * We need to test this early because xfstests assumes that an
5615 * insert range of (0, 1) will return EOPNOTSUPP if the file
5616 * system does not support insert range.
5618 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5621 /* Insert range works only on fs block size aligned offsets. */
5622 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5623 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5626 if (!S_ISREG(inode->i_mode))
5629 trace_ext4_insert_range(inode, offset, len);
5631 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5632 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5634 /* Call ext4_force_commit to flush all data in case of data=journal */
5635 if (ext4_should_journal_data(inode)) {
5636 ret = ext4_force_commit(inode->i_sb);
5642 /* Currently just for extent based files */
5643 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5648 /* Check for wrap through zero */
5649 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5654 /* Offset should be less than i_size */
5655 if (offset >= i_size_read(inode)) {
5660 /* Wait for existing dio to complete */
5661 ext4_inode_block_unlocked_dio(inode);
5662 inode_dio_wait(inode);
5665 * Prevent page faults from reinstantiating pages we have released from
5668 down_write(&EXT4_I(inode)->i_mmap_sem);
5670 * Need to round down to align start offset to page size boundary
5671 * for page size > block size.
5673 ioffset = round_down(offset, PAGE_SIZE);
5674 /* Write out all dirty pages */
5675 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5679 truncate_pagecache(inode, ioffset);
5681 credits = ext4_writepage_trans_blocks(inode);
5682 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5683 if (IS_ERR(handle)) {
5684 ret = PTR_ERR(handle);
5688 /* Expand file to avoid data loss if there is error while shifting */
5689 inode->i_size += len;
5690 EXT4_I(inode)->i_disksize += len;
5691 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5692 ret = ext4_mark_inode_dirty(handle, inode);
5696 down_write(&EXT4_I(inode)->i_data_sem);
5697 ext4_discard_preallocations(inode);
5699 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5701 up_write(&EXT4_I(inode)->i_data_sem);
5705 depth = ext_depth(inode);
5706 extent = path[depth].p_ext;
5708 ee_start_lblk = le32_to_cpu(extent->ee_block);
5709 ee_len = ext4_ext_get_actual_len(extent);
5712 * If offset_lblk is not the starting block of extent, split
5713 * the extent @offset_lblk
5715 if ((offset_lblk > ee_start_lblk) &&
5716 (offset_lblk < (ee_start_lblk + ee_len))) {
5717 if (ext4_ext_is_unwritten(extent))
5718 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5719 EXT4_EXT_MARK_UNWRIT2;
5720 ret = ext4_split_extent_at(handle, inode, &path,
5721 offset_lblk, split_flag,
5723 EXT4_GET_BLOCKS_PRE_IO |
5724 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5727 ext4_ext_drop_refs(path);
5730 up_write(&EXT4_I(inode)->i_data_sem);
5735 ret = ext4_es_remove_extent(inode, offset_lblk,
5736 EXT_MAX_BLOCKS - offset_lblk);
5738 up_write(&EXT4_I(inode)->i_data_sem);
5743 * if offset_lblk lies in a hole which is at start of file, use
5744 * ee_start_lblk to shift extents
5746 ret = ext4_ext_shift_extents(inode, handle,
5747 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5748 len_lblk, SHIFT_RIGHT);
5750 up_write(&EXT4_I(inode)->i_data_sem);
5752 ext4_handle_sync(handle);
5755 ext4_journal_stop(handle);
5757 up_write(&EXT4_I(inode)->i_mmap_sem);
5758 ext4_inode_resume_unlocked_dio(inode);
5760 inode_unlock(inode);
5765 * ext4_swap_extents - Swap extents between two inodes
5767 * @inode1: First inode
5768 * @inode2: Second inode
5769 * @lblk1: Start block for first inode
5770 * @lblk2: Start block for second inode
5771 * @count: Number of blocks to swap
5772 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5773 * @erp: Pointer to save error value
5775 * This helper routine does exactly what is promise "swap extents". All other
5776 * stuff such as page-cache locking consistency, bh mapping consistency or
5777 * extent's data copying must be performed by caller.
5779 * i_mutex is held for both inodes
5780 * i_data_sem is locked for write for both inodes
5782 * All pages from requested range are locked for both inodes
5785 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5786 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5787 ext4_lblk_t count, int unwritten, int *erp)
5789 struct ext4_ext_path *path1 = NULL;
5790 struct ext4_ext_path *path2 = NULL;
5791 int replaced_count = 0;
5793 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5794 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5795 BUG_ON(!inode_is_locked(inode1));
5796 BUG_ON(!inode_is_locked(inode2));
5798 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5801 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5806 struct ext4_extent *ex1, *ex2, tmp_ex;
5807 ext4_lblk_t e1_blk, e2_blk;
5808 int e1_len, e2_len, len;
5811 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5812 if (IS_ERR(path1)) {
5813 *erp = PTR_ERR(path1);
5819 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5820 if (IS_ERR(path2)) {
5821 *erp = PTR_ERR(path2);
5825 ex1 = path1[path1->p_depth].p_ext;
5826 ex2 = path2[path2->p_depth].p_ext;
5827 /* Do we have somthing to swap ? */
5828 if (unlikely(!ex2 || !ex1))
5831 e1_blk = le32_to_cpu(ex1->ee_block);
5832 e2_blk = le32_to_cpu(ex2->ee_block);
5833 e1_len = ext4_ext_get_actual_len(ex1);
5834 e2_len = ext4_ext_get_actual_len(ex2);
5837 if (!in_range(lblk1, e1_blk, e1_len) ||
5838 !in_range(lblk2, e2_blk, e2_len)) {
5839 ext4_lblk_t next1, next2;
5841 /* if hole after extent, then go to next extent */
5842 next1 = ext4_ext_next_allocated_block(path1);
5843 next2 = ext4_ext_next_allocated_block(path2);
5844 /* If hole before extent, then shift to that extent */
5849 /* Do we have something to swap */
5850 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5852 /* Move to the rightest boundary */
5853 len = next1 - lblk1;
5854 if (len < next2 - lblk2)
5855 len = next2 - lblk2;
5864 /* Prepare left boundary */
5865 if (e1_blk < lblk1) {
5867 *erp = ext4_force_split_extent_at(handle, inode1,
5872 if (e2_blk < lblk2) {
5874 *erp = ext4_force_split_extent_at(handle, inode2,
5879 /* ext4_split_extent_at() may result in leaf extent split,
5880 * path must to be revalidated. */
5884 /* Prepare right boundary */
5886 if (len > e1_blk + e1_len - lblk1)
5887 len = e1_blk + e1_len - lblk1;
5888 if (len > e2_blk + e2_len - lblk2)
5889 len = e2_blk + e2_len - lblk2;
5891 if (len != e1_len) {
5893 *erp = ext4_force_split_extent_at(handle, inode1,
5894 &path1, lblk1 + len, 0);
5898 if (len != e2_len) {
5900 *erp = ext4_force_split_extent_at(handle, inode2,
5901 &path2, lblk2 + len, 0);
5905 /* ext4_split_extent_at() may result in leaf extent split,
5906 * path must to be revalidated. */
5910 BUG_ON(e2_len != e1_len);
5911 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5914 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5918 /* Both extents are fully inside boundaries. Swap it now */
5920 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5921 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5922 ex1->ee_len = cpu_to_le16(e2_len);
5923 ex2->ee_len = cpu_to_le16(e1_len);
5925 ext4_ext_mark_unwritten(ex2);
5926 if (ext4_ext_is_unwritten(&tmp_ex))
5927 ext4_ext_mark_unwritten(ex1);
5929 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5930 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5931 *erp = ext4_ext_dirty(handle, inode2, path2 +
5935 *erp = ext4_ext_dirty(handle, inode1, path1 +
5938 * Looks scarry ah..? second inode already points to new blocks,
5939 * and it was successfully dirtied. But luckily error may happen
5940 * only due to journal error, so full transaction will be
5947 replaced_count += len;
5951 ext4_ext_drop_refs(path1);
5953 ext4_ext_drop_refs(path2);
5955 path1 = path2 = NULL;
5957 return replaced_count;