2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <linux/falloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/fiemap.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_UNINIT1 0x2 /* mark first half uninitialized */
55 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
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_RO_COMPAT_FEATURE(inode->i_sb,
78 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
81 et = find_ext4_extent_tail(eh);
82 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
87 static void ext4_extent_block_csum_set(struct inode *inode,
88 struct ext4_extent_header *eh)
90 struct ext4_extent_tail *et;
92 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
93 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
96 et = find_ext4_extent_tail(eh);
97 et->et_checksum = ext4_extent_block_csum(inode, eh);
100 static int ext4_split_extent(handle_t *handle,
102 struct ext4_ext_path *path,
103 struct ext4_map_blocks *map,
107 static int ext4_split_extent_at(handle_t *handle,
109 struct ext4_ext_path *path,
114 static int ext4_find_delayed_extent(struct inode *inode,
115 struct extent_status *newes);
117 static int ext4_ext_truncate_extend_restart(handle_t *handle,
123 if (!ext4_handle_valid(handle))
125 if (handle->h_buffer_credits > needed)
127 err = ext4_journal_extend(handle, needed);
130 err = ext4_truncate_restart_trans(handle, inode, needed);
142 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
143 struct ext4_ext_path *path)
146 /* path points to block */
147 return ext4_journal_get_write_access(handle, path->p_bh);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
165 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
166 /* path points to block */
167 err = __ext4_handle_dirty_metadata(where, line, handle,
170 /* path points to leaf/index in inode body */
171 err = ext4_mark_inode_dirty(handle, inode);
176 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
177 struct ext4_ext_path *path,
181 int depth = path->p_depth;
182 struct ext4_extent *ex;
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
198 * especially if the latter case turns out to be
201 ex = path[depth].p_ext;
203 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
204 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
206 if (block > ext_block)
207 return ext_pblk + (block - ext_block);
209 return ext_pblk - (ext_block - block);
212 /* it looks like index is empty;
213 * try to find starting block from index itself */
214 if (path[depth].p_bh)
215 return path[depth].p_bh->b_blocknr;
218 /* OK. use inode's group */
219 return ext4_inode_to_goal_block(inode);
223 * Allocation for a meta data block
226 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
227 struct ext4_ext_path *path,
228 struct ext4_extent *ex, int *err, unsigned int flags)
230 ext4_fsblk_t goal, newblock;
232 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
233 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
238 static inline int ext4_ext_space_block(struct inode *inode, int check)
242 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent);
244 #ifdef AGGRESSIVE_TEST
245 if (!check && size > 6)
251 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
255 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
256 / sizeof(struct ext4_extent_idx);
257 #ifdef AGGRESSIVE_TEST
258 if (!check && size > 5)
264 static inline int ext4_ext_space_root(struct inode *inode, int check)
268 size = sizeof(EXT4_I(inode)->i_data);
269 size -= sizeof(struct ext4_extent_header);
270 size /= sizeof(struct ext4_extent);
271 #ifdef AGGRESSIVE_TEST
272 if (!check && size > 3)
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
285 #ifdef AGGRESSIVE_TEST
286 if (!check && size > 4)
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
297 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
299 struct ext4_inode_info *ei = EXT4_I(inode);
302 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
303 / sizeof(struct ext4_extent_idx));
306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
313 if (ei->i_da_metadata_calc_len &&
314 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
317 if ((ei->i_da_metadata_calc_len % idxs) == 0)
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
323 ei->i_da_metadata_calc_len = 0;
325 ei->i_da_metadata_calc_len++;
326 ei->i_da_metadata_calc_last_lblock++;
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
334 ei->i_da_metadata_calc_len = 1;
335 ei->i_da_metadata_calc_last_lblock = lblock;
336 return ext_depth(inode) + 1;
340 ext4_ext_max_entries(struct inode *inode, int depth)
344 if (depth == ext_depth(inode)) {
346 max = ext4_ext_space_root(inode, 1);
348 max = ext4_ext_space_root_idx(inode, 1);
351 max = ext4_ext_space_block(inode, 1);
353 max = ext4_ext_space_block_idx(inode, 1);
359 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
361 ext4_fsblk_t block = ext4_ext_pblock(ext);
362 int len = ext4_ext_get_actual_len(ext);
366 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
369 static int ext4_valid_extent_idx(struct inode *inode,
370 struct ext4_extent_idx *ext_idx)
372 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
374 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
377 static int ext4_valid_extent_entries(struct inode *inode,
378 struct ext4_extent_header *eh,
381 unsigned short entries;
382 if (eh->eh_entries == 0)
385 entries = le16_to_cpu(eh->eh_entries);
389 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
391 if (!ext4_valid_extent(inode, ext))
397 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
399 if (!ext4_valid_extent_idx(inode, ext_idx))
408 static int __ext4_ext_check(const char *function, unsigned int line,
409 struct inode *inode, struct ext4_extent_header *eh,
410 int depth, ext4_fsblk_t pblk)
412 const char *error_msg;
415 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
416 error_msg = "invalid magic";
419 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
420 error_msg = "unexpected eh_depth";
423 if (unlikely(eh->eh_max == 0)) {
424 error_msg = "invalid eh_max";
427 max = ext4_ext_max_entries(inode, depth);
428 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
429 error_msg = "too large eh_max";
432 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
433 error_msg = "invalid eh_entries";
436 if (!ext4_valid_extent_entries(inode, eh, depth)) {
437 error_msg = "invalid extent entries";
440 /* Verify checksum on non-root extent tree nodes */
441 if (ext_depth(inode) != depth &&
442 !ext4_extent_block_csum_verify(inode, eh)) {
443 error_msg = "extent tree corrupted";
449 ext4_error_inode(inode, function, line, 0,
450 "pblk %llu bad header/extent: %s - magic %x, "
451 "entries %u, max %u(%u), depth %u(%u)",
452 (unsigned long long) pblk, error_msg,
453 le16_to_cpu(eh->eh_magic),
454 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
455 max, le16_to_cpu(eh->eh_depth), depth);
459 #define ext4_ext_check(inode, eh, depth, pblk) \
460 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
462 int ext4_ext_check_inode(struct inode *inode)
464 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
467 static struct buffer_head *
468 __read_extent_tree_block(const char *function, unsigned int line,
469 struct inode *inode, ext4_fsblk_t pblk, int depth,
472 struct buffer_head *bh;
475 bh = sb_getblk(inode->i_sb, pblk);
477 return ERR_PTR(-ENOMEM);
479 if (!bh_uptodate_or_lock(bh)) {
480 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
481 err = bh_submit_read(bh);
485 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
487 err = __ext4_ext_check(function, line, inode,
488 ext_block_hdr(bh), depth, pblk);
491 set_buffer_verified(bh);
493 * If this is a leaf block, cache all of its entries
495 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
496 struct ext4_extent_header *eh = ext_block_hdr(bh);
497 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
498 ext4_lblk_t prev = 0;
501 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
502 unsigned int status = EXTENT_STATUS_WRITTEN;
503 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
504 int len = ext4_ext_get_actual_len(ex);
506 if (prev && (prev != lblk))
507 ext4_es_cache_extent(inode, prev,
511 if (ext4_ext_is_uninitialized(ex))
512 status = EXTENT_STATUS_UNWRITTEN;
513 ext4_es_cache_extent(inode, lblk, len,
514 ext4_ext_pblock(ex), status);
525 #define read_extent_tree_block(inode, pblk, depth, flags) \
526 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
530 * This function is called to cache a file's extent information in the
533 int ext4_ext_precache(struct inode *inode)
535 struct ext4_inode_info *ei = EXT4_I(inode);
536 struct ext4_ext_path *path = NULL;
537 struct buffer_head *bh;
538 int i = 0, depth, ret = 0;
540 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
541 return 0; /* not an extent-mapped inode */
543 down_read(&ei->i_data_sem);
544 depth = ext_depth(inode);
546 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
549 up_read(&ei->i_data_sem);
553 /* Don't cache anything if there are no external extent blocks */
556 path[0].p_hdr = ext_inode_hdr(inode);
557 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
560 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
563 * If this is a leaf block or we've reached the end of
564 * the index block, go up
567 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
568 brelse(path[i].p_bh);
573 bh = read_extent_tree_block(inode,
574 ext4_idx_pblock(path[i].p_idx++),
576 EXT4_EX_FORCE_CACHE);
583 path[i].p_hdr = ext_block_hdr(bh);
584 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
586 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
588 up_read(&ei->i_data_sem);
589 ext4_ext_drop_refs(path);
595 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
597 int k, l = path->p_depth;
600 for (k = 0; k <= l; k++, path++) {
602 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
603 ext4_idx_pblock(path->p_idx));
604 } else if (path->p_ext) {
605 ext_debug(" %d:[%d]%d:%llu ",
606 le32_to_cpu(path->p_ext->ee_block),
607 ext4_ext_is_uninitialized(path->p_ext),
608 ext4_ext_get_actual_len(path->p_ext),
609 ext4_ext_pblock(path->p_ext));
616 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
618 int depth = ext_depth(inode);
619 struct ext4_extent_header *eh;
620 struct ext4_extent *ex;
626 eh = path[depth].p_hdr;
627 ex = EXT_FIRST_EXTENT(eh);
629 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
631 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
632 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
633 ext4_ext_is_uninitialized(ex),
634 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
639 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
640 ext4_fsblk_t newblock, int level)
642 int depth = ext_depth(inode);
643 struct ext4_extent *ex;
645 if (depth != level) {
646 struct ext4_extent_idx *idx;
647 idx = path[level].p_idx;
648 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
649 ext_debug("%d: move %d:%llu in new index %llu\n", level,
650 le32_to_cpu(idx->ei_block),
651 ext4_idx_pblock(idx),
659 ex = path[depth].p_ext;
660 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
661 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
662 le32_to_cpu(ex->ee_block),
664 ext4_ext_is_uninitialized(ex),
665 ext4_ext_get_actual_len(ex),
672 #define ext4_ext_show_path(inode, path)
673 #define ext4_ext_show_leaf(inode, path)
674 #define ext4_ext_show_move(inode, path, newblock, level)
677 void ext4_ext_drop_refs(struct ext4_ext_path *path)
679 int depth = path->p_depth;
682 for (i = 0; i <= depth; i++, path++)
690 * ext4_ext_binsearch_idx:
691 * binary search for the closest index of the given block
692 * the header must be checked before calling this
695 ext4_ext_binsearch_idx(struct inode *inode,
696 struct ext4_ext_path *path, ext4_lblk_t block)
698 struct ext4_extent_header *eh = path->p_hdr;
699 struct ext4_extent_idx *r, *l, *m;
702 ext_debug("binsearch for %u(idx): ", block);
704 l = EXT_FIRST_INDEX(eh) + 1;
705 r = EXT_LAST_INDEX(eh);
708 if (block < le32_to_cpu(m->ei_block))
712 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
713 m, le32_to_cpu(m->ei_block),
714 r, le32_to_cpu(r->ei_block));
718 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
719 ext4_idx_pblock(path->p_idx));
721 #ifdef CHECK_BINSEARCH
723 struct ext4_extent_idx *chix, *ix;
726 chix = ix = EXT_FIRST_INDEX(eh);
727 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
729 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
730 printk(KERN_DEBUG "k=%d, ix=0x%p, "
732 ix, EXT_FIRST_INDEX(eh));
733 printk(KERN_DEBUG "%u <= %u\n",
734 le32_to_cpu(ix->ei_block),
735 le32_to_cpu(ix[-1].ei_block));
737 BUG_ON(k && le32_to_cpu(ix->ei_block)
738 <= le32_to_cpu(ix[-1].ei_block));
739 if (block < le32_to_cpu(ix->ei_block))
743 BUG_ON(chix != path->p_idx);
750 * ext4_ext_binsearch:
751 * binary search for closest extent of the given block
752 * the header must be checked before calling this
755 ext4_ext_binsearch(struct inode *inode,
756 struct ext4_ext_path *path, ext4_lblk_t block)
758 struct ext4_extent_header *eh = path->p_hdr;
759 struct ext4_extent *r, *l, *m;
761 if (eh->eh_entries == 0) {
763 * this leaf is empty:
764 * we get such a leaf in split/add case
769 ext_debug("binsearch for %u: ", block);
771 l = EXT_FIRST_EXTENT(eh) + 1;
772 r = EXT_LAST_EXTENT(eh);
776 if (block < le32_to_cpu(m->ee_block))
780 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
781 m, le32_to_cpu(m->ee_block),
782 r, le32_to_cpu(r->ee_block));
786 ext_debug(" -> %d:%llu:[%d]%d ",
787 le32_to_cpu(path->p_ext->ee_block),
788 ext4_ext_pblock(path->p_ext),
789 ext4_ext_is_uninitialized(path->p_ext),
790 ext4_ext_get_actual_len(path->p_ext));
792 #ifdef CHECK_BINSEARCH
794 struct ext4_extent *chex, *ex;
797 chex = ex = EXT_FIRST_EXTENT(eh);
798 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
799 BUG_ON(k && le32_to_cpu(ex->ee_block)
800 <= le32_to_cpu(ex[-1].ee_block));
801 if (block < le32_to_cpu(ex->ee_block))
805 BUG_ON(chex != path->p_ext);
811 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
813 struct ext4_extent_header *eh;
815 eh = ext_inode_hdr(inode);
818 eh->eh_magic = EXT4_EXT_MAGIC;
819 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
820 ext4_mark_inode_dirty(handle, inode);
824 struct ext4_ext_path *
825 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
826 struct ext4_ext_path *path, int flags)
828 struct ext4_extent_header *eh;
829 struct buffer_head *bh;
830 short int depth, i, ppos = 0, alloc = 0;
833 eh = ext_inode_hdr(inode);
834 depth = ext_depth(inode);
836 /* account possible depth increase */
838 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
841 return ERR_PTR(-ENOMEM);
848 /* walk through the tree */
850 ext_debug("depth %d: num %d, max %d\n",
851 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
853 ext4_ext_binsearch_idx(inode, path + ppos, block);
854 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
855 path[ppos].p_depth = i;
856 path[ppos].p_ext = NULL;
858 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
865 eh = ext_block_hdr(bh);
867 if (unlikely(ppos > depth)) {
869 EXT4_ERROR_INODE(inode,
870 "ppos %d > depth %d", ppos, depth);
874 path[ppos].p_bh = bh;
875 path[ppos].p_hdr = eh;
878 path[ppos].p_depth = i;
879 path[ppos].p_ext = NULL;
880 path[ppos].p_idx = NULL;
883 ext4_ext_binsearch(inode, path + ppos, block);
884 /* if not an empty leaf */
885 if (path[ppos].p_ext)
886 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
888 ext4_ext_show_path(inode, path);
893 ext4_ext_drop_refs(path);
900 * ext4_ext_insert_index:
901 * insert new index [@logical;@ptr] into the block at @curp;
902 * check where to insert: before @curp or after @curp
904 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
905 struct ext4_ext_path *curp,
906 int logical, ext4_fsblk_t ptr)
908 struct ext4_extent_idx *ix;
911 err = ext4_ext_get_access(handle, inode, curp);
915 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
916 EXT4_ERROR_INODE(inode,
917 "logical %d == ei_block %d!",
918 logical, le32_to_cpu(curp->p_idx->ei_block));
922 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
923 >= le16_to_cpu(curp->p_hdr->eh_max))) {
924 EXT4_ERROR_INODE(inode,
925 "eh_entries %d >= eh_max %d!",
926 le16_to_cpu(curp->p_hdr->eh_entries),
927 le16_to_cpu(curp->p_hdr->eh_max));
931 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
933 ext_debug("insert new index %d after: %llu\n", logical, ptr);
934 ix = curp->p_idx + 1;
937 ext_debug("insert new index %d before: %llu\n", logical, ptr);
941 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
944 ext_debug("insert new index %d: "
945 "move %d indices from 0x%p to 0x%p\n",
946 logical, len, ix, ix + 1);
947 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
950 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
951 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
955 ix->ei_block = cpu_to_le32(logical);
956 ext4_idx_store_pblock(ix, ptr);
957 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
959 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
960 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
964 err = ext4_ext_dirty(handle, inode, curp);
965 ext4_std_error(inode->i_sb, err);
972 * inserts new subtree into the path, using free index entry
974 * - allocates all needed blocks (new leaf and all intermediate index blocks)
975 * - makes decision where to split
976 * - moves remaining extents and index entries (right to the split point)
977 * into the newly allocated blocks
978 * - initializes subtree
980 static int ext4_ext_split(handle_t *handle, struct inode *inode,
982 struct ext4_ext_path *path,
983 struct ext4_extent *newext, int at)
985 struct buffer_head *bh = NULL;
986 int depth = ext_depth(inode);
987 struct ext4_extent_header *neh;
988 struct ext4_extent_idx *fidx;
990 ext4_fsblk_t newblock, oldblock;
992 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
995 /* make decision: where to split? */
996 /* FIXME: now decision is simplest: at current extent */
998 /* if current leaf will be split, then we should use
999 * border from split point */
1000 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1001 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1004 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1005 border = path[depth].p_ext[1].ee_block;
1006 ext_debug("leaf will be split."
1007 " next leaf starts at %d\n",
1008 le32_to_cpu(border));
1010 border = newext->ee_block;
1011 ext_debug("leaf will be added."
1012 " next leaf starts at %d\n",
1013 le32_to_cpu(border));
1017 * If error occurs, then we break processing
1018 * and mark filesystem read-only. index won't
1019 * be inserted and tree will be in consistent
1020 * state. Next mount will repair buffers too.
1024 * Get array to track all allocated blocks.
1025 * We need this to handle errors and free blocks
1028 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1032 /* allocate all needed blocks */
1033 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1034 for (a = 0; a < depth - at; a++) {
1035 newblock = ext4_ext_new_meta_block(handle, inode, path,
1036 newext, &err, flags);
1039 ablocks[a] = newblock;
1042 /* initialize new leaf */
1043 newblock = ablocks[--a];
1044 if (unlikely(newblock == 0)) {
1045 EXT4_ERROR_INODE(inode, "newblock == 0!");
1049 bh = sb_getblk(inode->i_sb, newblock);
1050 if (unlikely(!bh)) {
1056 err = ext4_journal_get_create_access(handle, bh);
1060 neh = ext_block_hdr(bh);
1061 neh->eh_entries = 0;
1062 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1063 neh->eh_magic = EXT4_EXT_MAGIC;
1066 /* move remainder of path[depth] to the new leaf */
1067 if (unlikely(path[depth].p_hdr->eh_entries !=
1068 path[depth].p_hdr->eh_max)) {
1069 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1070 path[depth].p_hdr->eh_entries,
1071 path[depth].p_hdr->eh_max);
1075 /* start copy from next extent */
1076 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1077 ext4_ext_show_move(inode, path, newblock, depth);
1079 struct ext4_extent *ex;
1080 ex = EXT_FIRST_EXTENT(neh);
1081 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1082 le16_add_cpu(&neh->eh_entries, m);
1085 ext4_extent_block_csum_set(inode, neh);
1086 set_buffer_uptodate(bh);
1089 err = ext4_handle_dirty_metadata(handle, inode, bh);
1095 /* correct old leaf */
1097 err = ext4_ext_get_access(handle, inode, path + depth);
1100 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1101 err = ext4_ext_dirty(handle, inode, path + depth);
1107 /* create intermediate indexes */
1109 if (unlikely(k < 0)) {
1110 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1115 ext_debug("create %d intermediate indices\n", k);
1116 /* insert new index into current index block */
1117 /* current depth stored in i var */
1120 oldblock = newblock;
1121 newblock = ablocks[--a];
1122 bh = sb_getblk(inode->i_sb, newblock);
1123 if (unlikely(!bh)) {
1129 err = ext4_journal_get_create_access(handle, bh);
1133 neh = ext_block_hdr(bh);
1134 neh->eh_entries = cpu_to_le16(1);
1135 neh->eh_magic = EXT4_EXT_MAGIC;
1136 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1137 neh->eh_depth = cpu_to_le16(depth - i);
1138 fidx = EXT_FIRST_INDEX(neh);
1139 fidx->ei_block = border;
1140 ext4_idx_store_pblock(fidx, oldblock);
1142 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1143 i, newblock, le32_to_cpu(border), oldblock);
1145 /* move remainder of path[i] to the new index block */
1146 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1147 EXT_LAST_INDEX(path[i].p_hdr))) {
1148 EXT4_ERROR_INODE(inode,
1149 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1150 le32_to_cpu(path[i].p_ext->ee_block));
1154 /* start copy indexes */
1155 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1156 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1157 EXT_MAX_INDEX(path[i].p_hdr));
1158 ext4_ext_show_move(inode, path, newblock, i);
1160 memmove(++fidx, path[i].p_idx,
1161 sizeof(struct ext4_extent_idx) * m);
1162 le16_add_cpu(&neh->eh_entries, m);
1164 ext4_extent_block_csum_set(inode, neh);
1165 set_buffer_uptodate(bh);
1168 err = ext4_handle_dirty_metadata(handle, inode, bh);
1174 /* correct old index */
1176 err = ext4_ext_get_access(handle, inode, path + i);
1179 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1180 err = ext4_ext_dirty(handle, inode, path + i);
1188 /* insert new index */
1189 err = ext4_ext_insert_index(handle, inode, path + at,
1190 le32_to_cpu(border), newblock);
1194 if (buffer_locked(bh))
1200 /* free all allocated blocks in error case */
1201 for (i = 0; i < depth; i++) {
1204 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1205 EXT4_FREE_BLOCKS_METADATA);
1214 * ext4_ext_grow_indepth:
1215 * implements tree growing procedure:
1216 * - allocates new block
1217 * - moves top-level data (index block or leaf) into the new block
1218 * - initializes new top-level, creating index that points to the
1219 * just created block
1221 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1223 struct ext4_extent *newext)
1225 struct ext4_extent_header *neh;
1226 struct buffer_head *bh;
1227 ext4_fsblk_t newblock;
1230 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1231 newext, &err, flags);
1235 bh = sb_getblk(inode->i_sb, newblock);
1240 err = ext4_journal_get_create_access(handle, bh);
1246 /* move top-level index/leaf into new block */
1247 memmove(bh->b_data, EXT4_I(inode)->i_data,
1248 sizeof(EXT4_I(inode)->i_data));
1250 /* set size of new block */
1251 neh = ext_block_hdr(bh);
1252 /* old root could have indexes or leaves
1253 * so calculate e_max right way */
1254 if (ext_depth(inode))
1255 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1257 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1258 neh->eh_magic = EXT4_EXT_MAGIC;
1259 ext4_extent_block_csum_set(inode, neh);
1260 set_buffer_uptodate(bh);
1263 err = ext4_handle_dirty_metadata(handle, inode, bh);
1267 /* Update top-level index: num,max,pointer */
1268 neh = ext_inode_hdr(inode);
1269 neh->eh_entries = cpu_to_le16(1);
1270 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1271 if (neh->eh_depth == 0) {
1272 /* Root extent block becomes index block */
1273 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1274 EXT_FIRST_INDEX(neh)->ei_block =
1275 EXT_FIRST_EXTENT(neh)->ee_block;
1277 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1278 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1279 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1280 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1282 le16_add_cpu(&neh->eh_depth, 1);
1283 ext4_mark_inode_dirty(handle, inode);
1291 * ext4_ext_create_new_leaf:
1292 * finds empty index and adds new leaf.
1293 * if no free index is found, then it requests in-depth growing.
1295 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1296 unsigned int mb_flags,
1297 unsigned int gb_flags,
1298 struct ext4_ext_path *path,
1299 struct ext4_extent *newext)
1301 struct ext4_ext_path *curp;
1302 int depth, i, err = 0;
1305 i = depth = ext_depth(inode);
1307 /* walk up to the tree and look for free index entry */
1308 curp = path + depth;
1309 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1314 /* we use already allocated block for index block,
1315 * so subsequent data blocks should be contiguous */
1316 if (EXT_HAS_FREE_INDEX(curp)) {
1317 /* if we found index with free entry, then use that
1318 * entry: create all needed subtree and add new leaf */
1319 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1324 ext4_ext_drop_refs(path);
1325 path = ext4_ext_find_extent(inode,
1326 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1329 err = PTR_ERR(path);
1331 /* tree is full, time to grow in depth */
1332 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1337 ext4_ext_drop_refs(path);
1338 path = ext4_ext_find_extent(inode,
1339 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1342 err = PTR_ERR(path);
1347 * only first (depth 0 -> 1) produces free space;
1348 * in all other cases we have to split the grown tree
1350 depth = ext_depth(inode);
1351 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1352 /* now we need to split */
1362 * search the closest allocated block to the left for *logical
1363 * and returns it at @logical + it's physical address at @phys
1364 * if *logical is the smallest allocated block, the function
1365 * returns 0 at @phys
1366 * return value contains 0 (success) or error code
1368 static int ext4_ext_search_left(struct inode *inode,
1369 struct ext4_ext_path *path,
1370 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1372 struct ext4_extent_idx *ix;
1373 struct ext4_extent *ex;
1376 if (unlikely(path == NULL)) {
1377 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1380 depth = path->p_depth;
1383 if (depth == 0 && path->p_ext == NULL)
1386 /* usually extent in the path covers blocks smaller
1387 * then *logical, but it can be that extent is the
1388 * first one in the file */
1390 ex = path[depth].p_ext;
1391 ee_len = ext4_ext_get_actual_len(ex);
1392 if (*logical < le32_to_cpu(ex->ee_block)) {
1393 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1394 EXT4_ERROR_INODE(inode,
1395 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1396 *logical, le32_to_cpu(ex->ee_block));
1399 while (--depth >= 0) {
1400 ix = path[depth].p_idx;
1401 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1402 EXT4_ERROR_INODE(inode,
1403 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1404 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1405 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1406 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1414 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1415 EXT4_ERROR_INODE(inode,
1416 "logical %d < ee_block %d + ee_len %d!",
1417 *logical, le32_to_cpu(ex->ee_block), ee_len);
1421 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1422 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1427 * search the closest allocated block to the right for *logical
1428 * and returns it at @logical + it's physical address at @phys
1429 * if *logical is the largest allocated block, the function
1430 * returns 0 at @phys
1431 * return value contains 0 (success) or error code
1433 static int ext4_ext_search_right(struct inode *inode,
1434 struct ext4_ext_path *path,
1435 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1436 struct ext4_extent **ret_ex)
1438 struct buffer_head *bh = NULL;
1439 struct ext4_extent_header *eh;
1440 struct ext4_extent_idx *ix;
1441 struct ext4_extent *ex;
1443 int depth; /* Note, NOT eh_depth; depth from top of tree */
1446 if (unlikely(path == NULL)) {
1447 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1450 depth = path->p_depth;
1453 if (depth == 0 && path->p_ext == NULL)
1456 /* usually extent in the path covers blocks smaller
1457 * then *logical, but it can be that extent is the
1458 * first one in the file */
1460 ex = path[depth].p_ext;
1461 ee_len = ext4_ext_get_actual_len(ex);
1462 if (*logical < le32_to_cpu(ex->ee_block)) {
1463 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1464 EXT4_ERROR_INODE(inode,
1465 "first_extent(path[%d].p_hdr) != ex",
1469 while (--depth >= 0) {
1470 ix = path[depth].p_idx;
1471 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1472 EXT4_ERROR_INODE(inode,
1473 "ix != EXT_FIRST_INDEX *logical %d!",
1481 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1482 EXT4_ERROR_INODE(inode,
1483 "logical %d < ee_block %d + ee_len %d!",
1484 *logical, le32_to_cpu(ex->ee_block), ee_len);
1488 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1489 /* next allocated block in this leaf */
1494 /* go up and search for index to the right */
1495 while (--depth >= 0) {
1496 ix = path[depth].p_idx;
1497 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1501 /* we've gone up to the root and found no index to the right */
1505 /* we've found index to the right, let's
1506 * follow it and find the closest allocated
1507 * block to the right */
1509 block = ext4_idx_pblock(ix);
1510 while (++depth < path->p_depth) {
1511 /* subtract from p_depth to get proper eh_depth */
1512 bh = read_extent_tree_block(inode, block,
1513 path->p_depth - depth, 0);
1516 eh = ext_block_hdr(bh);
1517 ix = EXT_FIRST_INDEX(eh);
1518 block = ext4_idx_pblock(ix);
1522 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1525 eh = ext_block_hdr(bh);
1526 ex = EXT_FIRST_EXTENT(eh);
1528 *logical = le32_to_cpu(ex->ee_block);
1529 *phys = ext4_ext_pblock(ex);
1537 * ext4_ext_next_allocated_block:
1538 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1539 * NOTE: it considers block number from index entry as
1540 * allocated block. Thus, index entries have to be consistent
1544 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1548 BUG_ON(path == NULL);
1549 depth = path->p_depth;
1551 if (depth == 0 && path->p_ext == NULL)
1552 return EXT_MAX_BLOCKS;
1554 while (depth >= 0) {
1555 if (depth == path->p_depth) {
1557 if (path[depth].p_ext &&
1558 path[depth].p_ext !=
1559 EXT_LAST_EXTENT(path[depth].p_hdr))
1560 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1563 if (path[depth].p_idx !=
1564 EXT_LAST_INDEX(path[depth].p_hdr))
1565 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1570 return EXT_MAX_BLOCKS;
1574 * ext4_ext_next_leaf_block:
1575 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1577 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1581 BUG_ON(path == NULL);
1582 depth = path->p_depth;
1584 /* zero-tree has no leaf blocks at all */
1586 return EXT_MAX_BLOCKS;
1588 /* go to index block */
1591 while (depth >= 0) {
1592 if (path[depth].p_idx !=
1593 EXT_LAST_INDEX(path[depth].p_hdr))
1594 return (ext4_lblk_t)
1595 le32_to_cpu(path[depth].p_idx[1].ei_block);
1599 return EXT_MAX_BLOCKS;
1603 * ext4_ext_correct_indexes:
1604 * if leaf gets modified and modified extent is first in the leaf,
1605 * then we have to correct all indexes above.
1606 * TODO: do we need to correct tree in all cases?
1608 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1609 struct ext4_ext_path *path)
1611 struct ext4_extent_header *eh;
1612 int depth = ext_depth(inode);
1613 struct ext4_extent *ex;
1617 eh = path[depth].p_hdr;
1618 ex = path[depth].p_ext;
1620 if (unlikely(ex == NULL || eh == NULL)) {
1621 EXT4_ERROR_INODE(inode,
1622 "ex %p == NULL or eh %p == NULL", ex, eh);
1627 /* there is no tree at all */
1631 if (ex != EXT_FIRST_EXTENT(eh)) {
1632 /* we correct tree if first leaf got modified only */
1637 * TODO: we need correction if border is smaller than current one
1640 border = path[depth].p_ext->ee_block;
1641 err = ext4_ext_get_access(handle, inode, path + k);
1644 path[k].p_idx->ei_block = border;
1645 err = ext4_ext_dirty(handle, inode, path + k);
1650 /* change all left-side indexes */
1651 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1653 err = ext4_ext_get_access(handle, inode, path + k);
1656 path[k].p_idx->ei_block = border;
1657 err = ext4_ext_dirty(handle, inode, path + k);
1666 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1667 struct ext4_extent *ex2)
1669 unsigned short ext1_ee_len, ext2_ee_len;
1672 * Make sure that both extents are initialized. We don't merge
1673 * uninitialized extents so that we can be sure that end_io code has
1674 * the extent that was written properly split out and conversion to
1675 * initialized is trivial.
1677 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
1680 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1681 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1683 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1684 le32_to_cpu(ex2->ee_block))
1688 * To allow future support for preallocated extents to be added
1689 * as an RO_COMPAT feature, refuse to merge to extents if
1690 * this can result in the top bit of ee_len being set.
1692 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1694 #ifdef AGGRESSIVE_TEST
1695 if (ext1_ee_len >= 4)
1699 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1705 * This function tries to merge the "ex" extent to the next extent in the tree.
1706 * It always tries to merge towards right. If you want to merge towards
1707 * left, pass "ex - 1" as argument instead of "ex".
1708 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1709 * 1 if they got merged.
1711 static int ext4_ext_try_to_merge_right(struct inode *inode,
1712 struct ext4_ext_path *path,
1713 struct ext4_extent *ex)
1715 struct ext4_extent_header *eh;
1716 unsigned int depth, len;
1718 int uninitialized = 0;
1720 depth = ext_depth(inode);
1721 BUG_ON(path[depth].p_hdr == NULL);
1722 eh = path[depth].p_hdr;
1724 while (ex < EXT_LAST_EXTENT(eh)) {
1725 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1727 /* merge with next extent! */
1728 if (ext4_ext_is_uninitialized(ex))
1730 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1731 + ext4_ext_get_actual_len(ex + 1));
1733 ext4_ext_mark_uninitialized(ex);
1735 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1736 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1737 * sizeof(struct ext4_extent);
1738 memmove(ex + 1, ex + 2, len);
1740 le16_add_cpu(&eh->eh_entries, -1);
1742 WARN_ON(eh->eh_entries == 0);
1743 if (!eh->eh_entries)
1744 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1751 * This function does a very simple check to see if we can collapse
1752 * an extent tree with a single extent tree leaf block into the inode.
1754 static void ext4_ext_try_to_merge_up(handle_t *handle,
1755 struct inode *inode,
1756 struct ext4_ext_path *path)
1759 unsigned max_root = ext4_ext_space_root(inode, 0);
1762 if ((path[0].p_depth != 1) ||
1763 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1764 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1768 * We need to modify the block allocation bitmap and the block
1769 * group descriptor to release the extent tree block. If we
1770 * can't get the journal credits, give up.
1772 if (ext4_journal_extend(handle, 2))
1776 * Copy the extent data up to the inode
1778 blk = ext4_idx_pblock(path[0].p_idx);
1779 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1780 sizeof(struct ext4_extent_idx);
1781 s += sizeof(struct ext4_extent_header);
1783 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1784 path[0].p_depth = 0;
1785 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1786 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1787 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1789 brelse(path[1].p_bh);
1790 ext4_free_blocks(handle, inode, NULL, blk, 1,
1791 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET |
1792 EXT4_FREE_BLOCKS_RESERVE);
1796 * This function tries to merge the @ex extent to neighbours in the tree.
1797 * return 1 if merge left else 0.
1799 static void ext4_ext_try_to_merge(handle_t *handle,
1800 struct inode *inode,
1801 struct ext4_ext_path *path,
1802 struct ext4_extent *ex) {
1803 struct ext4_extent_header *eh;
1807 depth = ext_depth(inode);
1808 BUG_ON(path[depth].p_hdr == NULL);
1809 eh = path[depth].p_hdr;
1811 if (ex > EXT_FIRST_EXTENT(eh))
1812 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1815 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1817 ext4_ext_try_to_merge_up(handle, inode, path);
1821 * check if a portion of the "newext" extent overlaps with an
1824 * If there is an overlap discovered, it updates the length of the newext
1825 * such that there will be no overlap, and then returns 1.
1826 * If there is no overlap found, it returns 0.
1828 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1829 struct inode *inode,
1830 struct ext4_extent *newext,
1831 struct ext4_ext_path *path)
1834 unsigned int depth, len1;
1835 unsigned int ret = 0;
1837 b1 = le32_to_cpu(newext->ee_block);
1838 len1 = ext4_ext_get_actual_len(newext);
1839 depth = ext_depth(inode);
1840 if (!path[depth].p_ext)
1842 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1843 b2 &= ~(sbi->s_cluster_ratio - 1);
1846 * get the next allocated block if the extent in the path
1847 * is before the requested block(s)
1850 b2 = ext4_ext_next_allocated_block(path);
1851 if (b2 == EXT_MAX_BLOCKS)
1853 b2 &= ~(sbi->s_cluster_ratio - 1);
1856 /* check for wrap through zero on extent logical start block*/
1857 if (b1 + len1 < b1) {
1858 len1 = EXT_MAX_BLOCKS - b1;
1859 newext->ee_len = cpu_to_le16(len1);
1863 /* check for overlap */
1864 if (b1 + len1 > b2) {
1865 newext->ee_len = cpu_to_le16(b2 - b1);
1873 * ext4_ext_insert_extent:
1874 * tries to merge requsted extent into the existing extent or
1875 * inserts requested extent as new one into the tree,
1876 * creating new leaf in the no-space case.
1878 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1879 struct ext4_ext_path *path,
1880 struct ext4_extent *newext, int gb_flags)
1882 struct ext4_extent_header *eh;
1883 struct ext4_extent *ex, *fex;
1884 struct ext4_extent *nearex; /* nearest extent */
1885 struct ext4_ext_path *npath = NULL;
1886 int depth, len, err;
1888 unsigned uninitialized = 0;
1891 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1892 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1895 depth = ext_depth(inode);
1896 ex = path[depth].p_ext;
1897 eh = path[depth].p_hdr;
1898 if (unlikely(path[depth].p_hdr == NULL)) {
1899 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1903 /* try to insert block into found extent and return */
1904 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1907 * Try to see whether we should rather test the extent on
1908 * right from ex, or from the left of ex. This is because
1909 * ext4_ext_find_extent() can return either extent on the
1910 * left, or on the right from the searched position. This
1911 * will make merging more effective.
1913 if (ex < EXT_LAST_EXTENT(eh) &&
1914 (le32_to_cpu(ex->ee_block) +
1915 ext4_ext_get_actual_len(ex) <
1916 le32_to_cpu(newext->ee_block))) {
1919 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1920 (le32_to_cpu(newext->ee_block) +
1921 ext4_ext_get_actual_len(newext) <
1922 le32_to_cpu(ex->ee_block)))
1925 /* Try to append newex to the ex */
1926 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1927 ext_debug("append [%d]%d block to %u:[%d]%d"
1929 ext4_ext_is_uninitialized(newext),
1930 ext4_ext_get_actual_len(newext),
1931 le32_to_cpu(ex->ee_block),
1932 ext4_ext_is_uninitialized(ex),
1933 ext4_ext_get_actual_len(ex),
1934 ext4_ext_pblock(ex));
1935 err = ext4_ext_get_access(handle, inode,
1941 * ext4_can_extents_be_merged should have checked
1942 * that either both extents are uninitialized, or
1943 * both aren't. Thus we need to check only one of
1946 if (ext4_ext_is_uninitialized(ex))
1948 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1949 + ext4_ext_get_actual_len(newext));
1951 ext4_ext_mark_uninitialized(ex);
1952 eh = path[depth].p_hdr;
1958 /* Try to prepend newex to the ex */
1959 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1960 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1962 le32_to_cpu(newext->ee_block),
1963 ext4_ext_is_uninitialized(newext),
1964 ext4_ext_get_actual_len(newext),
1965 le32_to_cpu(ex->ee_block),
1966 ext4_ext_is_uninitialized(ex),
1967 ext4_ext_get_actual_len(ex),
1968 ext4_ext_pblock(ex));
1969 err = ext4_ext_get_access(handle, inode,
1975 * ext4_can_extents_be_merged should have checked
1976 * that either both extents are uninitialized, or
1977 * both aren't. Thus we need to check only one of
1980 if (ext4_ext_is_uninitialized(ex))
1982 ex->ee_block = newext->ee_block;
1983 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1984 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1985 + ext4_ext_get_actual_len(newext));
1987 ext4_ext_mark_uninitialized(ex);
1988 eh = path[depth].p_hdr;
1994 depth = ext_depth(inode);
1995 eh = path[depth].p_hdr;
1996 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1999 /* probably next leaf has space for us? */
2000 fex = EXT_LAST_EXTENT(eh);
2001 next = EXT_MAX_BLOCKS;
2002 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2003 next = ext4_ext_next_leaf_block(path);
2004 if (next != EXT_MAX_BLOCKS) {
2005 ext_debug("next leaf block - %u\n", next);
2006 BUG_ON(npath != NULL);
2007 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2009 return PTR_ERR(npath);
2010 BUG_ON(npath->p_depth != path->p_depth);
2011 eh = npath[depth].p_hdr;
2012 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2013 ext_debug("next leaf isn't full(%d)\n",
2014 le16_to_cpu(eh->eh_entries));
2018 ext_debug("next leaf has no free space(%d,%d)\n",
2019 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2023 * There is no free space in the found leaf.
2024 * We're gonna add a new leaf in the tree.
2026 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2027 mb_flags = EXT4_MB_USE_RESERVED;
2028 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2032 depth = ext_depth(inode);
2033 eh = path[depth].p_hdr;
2036 nearex = path[depth].p_ext;
2038 err = ext4_ext_get_access(handle, inode, path + depth);
2043 /* there is no extent in this leaf, create first one */
2044 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2045 le32_to_cpu(newext->ee_block),
2046 ext4_ext_pblock(newext),
2047 ext4_ext_is_uninitialized(newext),
2048 ext4_ext_get_actual_len(newext));
2049 nearex = EXT_FIRST_EXTENT(eh);
2051 if (le32_to_cpu(newext->ee_block)
2052 > le32_to_cpu(nearex->ee_block)) {
2054 ext_debug("insert %u:%llu:[%d]%d before: "
2056 le32_to_cpu(newext->ee_block),
2057 ext4_ext_pblock(newext),
2058 ext4_ext_is_uninitialized(newext),
2059 ext4_ext_get_actual_len(newext),
2064 BUG_ON(newext->ee_block == nearex->ee_block);
2065 ext_debug("insert %u:%llu:[%d]%d after: "
2067 le32_to_cpu(newext->ee_block),
2068 ext4_ext_pblock(newext),
2069 ext4_ext_is_uninitialized(newext),
2070 ext4_ext_get_actual_len(newext),
2073 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2075 ext_debug("insert %u:%llu:[%d]%d: "
2076 "move %d extents from 0x%p to 0x%p\n",
2077 le32_to_cpu(newext->ee_block),
2078 ext4_ext_pblock(newext),
2079 ext4_ext_is_uninitialized(newext),
2080 ext4_ext_get_actual_len(newext),
2081 len, nearex, nearex + 1);
2082 memmove(nearex + 1, nearex,
2083 len * sizeof(struct ext4_extent));
2087 le16_add_cpu(&eh->eh_entries, 1);
2088 path[depth].p_ext = nearex;
2089 nearex->ee_block = newext->ee_block;
2090 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2091 nearex->ee_len = newext->ee_len;
2094 /* try to merge extents */
2095 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2096 ext4_ext_try_to_merge(handle, inode, path, nearex);
2099 /* time to correct all indexes above */
2100 err = ext4_ext_correct_indexes(handle, inode, path);
2104 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2108 ext4_ext_drop_refs(npath);
2114 static int ext4_fill_fiemap_extents(struct inode *inode,
2115 ext4_lblk_t block, ext4_lblk_t num,
2116 struct fiemap_extent_info *fieinfo)
2118 struct ext4_ext_path *path = NULL;
2119 struct ext4_extent *ex;
2120 struct extent_status es;
2121 ext4_lblk_t next, next_del, start = 0, end = 0;
2122 ext4_lblk_t last = block + num;
2123 int exists, depth = 0, err = 0;
2124 unsigned int flags = 0;
2125 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2127 while (block < last && block != EXT_MAX_BLOCKS) {
2129 /* find extent for this block */
2130 down_read(&EXT4_I(inode)->i_data_sem);
2132 if (path && ext_depth(inode) != depth) {
2133 /* depth was changed. we have to realloc path */
2138 path = ext4_ext_find_extent(inode, block, path, 0);
2140 up_read(&EXT4_I(inode)->i_data_sem);
2141 err = PTR_ERR(path);
2146 depth = ext_depth(inode);
2147 if (unlikely(path[depth].p_hdr == NULL)) {
2148 up_read(&EXT4_I(inode)->i_data_sem);
2149 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2153 ex = path[depth].p_ext;
2154 next = ext4_ext_next_allocated_block(path);
2155 ext4_ext_drop_refs(path);
2160 /* there is no extent yet, so try to allocate
2161 * all requested space */
2164 } else if (le32_to_cpu(ex->ee_block) > block) {
2165 /* need to allocate space before found extent */
2167 end = le32_to_cpu(ex->ee_block);
2168 if (block + num < end)
2170 } else if (block >= le32_to_cpu(ex->ee_block)
2171 + ext4_ext_get_actual_len(ex)) {
2172 /* need to allocate space after found extent */
2177 } else if (block >= le32_to_cpu(ex->ee_block)) {
2179 * some part of requested space is covered
2183 end = le32_to_cpu(ex->ee_block)
2184 + ext4_ext_get_actual_len(ex);
2185 if (block + num < end)
2191 BUG_ON(end <= start);
2195 es.es_len = end - start;
2198 es.es_lblk = le32_to_cpu(ex->ee_block);
2199 es.es_len = ext4_ext_get_actual_len(ex);
2200 es.es_pblk = ext4_ext_pblock(ex);
2201 if (ext4_ext_is_uninitialized(ex))
2202 flags |= FIEMAP_EXTENT_UNWRITTEN;
2206 * Find delayed extent and update es accordingly. We call
2207 * it even in !exists case to find out whether es is the
2208 * last existing extent or not.
2210 next_del = ext4_find_delayed_extent(inode, &es);
2211 if (!exists && next_del) {
2213 flags |= (FIEMAP_EXTENT_DELALLOC |
2214 FIEMAP_EXTENT_UNKNOWN);
2216 up_read(&EXT4_I(inode)->i_data_sem);
2218 if (unlikely(es.es_len == 0)) {
2219 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2225 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2226 * we need to check next == EXT_MAX_BLOCKS because it is
2227 * possible that an extent is with unwritten and delayed
2228 * status due to when an extent is delayed allocated and
2229 * is allocated by fallocate status tree will track both of
2232 * So we could return a unwritten and delayed extent, and
2233 * its block is equal to 'next'.
2235 if (next == next_del && next == EXT_MAX_BLOCKS) {
2236 flags |= FIEMAP_EXTENT_LAST;
2237 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2238 next != EXT_MAX_BLOCKS)) {
2239 EXT4_ERROR_INODE(inode,
2240 "next extent == %u, next "
2241 "delalloc extent = %u",
2249 err = fiemap_fill_next_extent(fieinfo,
2250 (__u64)es.es_lblk << blksize_bits,
2251 (__u64)es.es_pblk << blksize_bits,
2252 (__u64)es.es_len << blksize_bits,
2262 block = es.es_lblk + es.es_len;
2266 ext4_ext_drop_refs(path);
2274 * ext4_ext_put_gap_in_cache:
2275 * calculate boundaries of the gap that the requested block fits into
2276 * and cache this gap
2279 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2282 int depth = ext_depth(inode);
2283 unsigned long len = 0;
2284 ext4_lblk_t lblock = 0;
2285 struct ext4_extent *ex;
2287 ex = path[depth].p_ext;
2290 * there is no extent yet, so gap is [0;-] and we
2293 ext_debug("cache gap(whole file):");
2294 } else if (block < le32_to_cpu(ex->ee_block)) {
2296 len = le32_to_cpu(ex->ee_block) - block;
2297 ext_debug("cache gap(before): %u [%u:%u]",
2299 le32_to_cpu(ex->ee_block),
2300 ext4_ext_get_actual_len(ex));
2301 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2302 ext4_es_insert_extent(inode, lblock, len, ~0,
2303 EXTENT_STATUS_HOLE);
2304 } else if (block >= le32_to_cpu(ex->ee_block)
2305 + ext4_ext_get_actual_len(ex)) {
2307 lblock = le32_to_cpu(ex->ee_block)
2308 + ext4_ext_get_actual_len(ex);
2310 next = ext4_ext_next_allocated_block(path);
2311 ext_debug("cache gap(after): [%u:%u] %u",
2312 le32_to_cpu(ex->ee_block),
2313 ext4_ext_get_actual_len(ex),
2315 BUG_ON(next == lblock);
2316 len = next - lblock;
2317 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2318 ext4_es_insert_extent(inode, lblock, len, ~0,
2319 EXTENT_STATUS_HOLE);
2324 ext_debug(" -> %u:%lu\n", lblock, len);
2329 * removes index from the index block.
2331 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2332 struct ext4_ext_path *path, int depth)
2337 /* free index block */
2339 path = path + depth;
2340 leaf = ext4_idx_pblock(path->p_idx);
2341 if (unlikely(path->p_hdr->eh_entries == 0)) {
2342 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2345 err = ext4_ext_get_access(handle, inode, path);
2349 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2350 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2351 len *= sizeof(struct ext4_extent_idx);
2352 memmove(path->p_idx, path->p_idx + 1, len);
2355 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2356 err = ext4_ext_dirty(handle, inode, path);
2359 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2360 trace_ext4_ext_rm_idx(inode, leaf);
2362 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2363 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2365 while (--depth >= 0) {
2366 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2369 err = ext4_ext_get_access(handle, inode, path);
2372 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2373 err = ext4_ext_dirty(handle, inode, path);
2381 * ext4_ext_calc_credits_for_single_extent:
2382 * This routine returns max. credits that needed to insert an extent
2383 * to the extent tree.
2384 * When pass the actual path, the caller should calculate credits
2387 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2388 struct ext4_ext_path *path)
2391 int depth = ext_depth(inode);
2394 /* probably there is space in leaf? */
2395 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2396 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2399 * There are some space in the leaf tree, no
2400 * need to account for leaf block credit
2402 * bitmaps and block group descriptor blocks
2403 * and other metadata blocks still need to be
2406 /* 1 bitmap, 1 block group descriptor */
2407 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2412 return ext4_chunk_trans_blocks(inode, nrblocks);
2416 * How many index/leaf blocks need to change/allocate to add @extents extents?
2418 * If we add a single extent, then in the worse case, each tree level
2419 * index/leaf need to be changed in case of the tree split.
2421 * If more extents are inserted, they could cause the whole tree split more
2422 * than once, but this is really rare.
2424 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2429 /* If we are converting the inline data, only one is needed here. */
2430 if (ext4_has_inline_data(inode))
2433 depth = ext_depth(inode);
2443 static inline int get_default_free_blocks_flags(struct inode *inode)
2445 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2446 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2447 else if (ext4_should_journal_data(inode))
2448 return EXT4_FREE_BLOCKS_FORGET;
2452 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2453 struct ext4_extent *ex,
2454 long long *partial_cluster,
2455 ext4_lblk_t from, ext4_lblk_t to)
2457 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2458 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2460 int flags = get_default_free_blocks_flags(inode);
2463 * For bigalloc file systems, we never free a partial cluster
2464 * at the beginning of the extent. Instead, we make a note
2465 * that we tried freeing the cluster, and check to see if we
2466 * need to free it on a subsequent call to ext4_remove_blocks,
2467 * or at the end of the ext4_truncate() operation.
2469 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2471 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2473 * If we have a partial cluster, and it's different from the
2474 * cluster of the last block, we need to explicitly free the
2475 * partial cluster here.
2477 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2478 if ((*partial_cluster > 0) &&
2479 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2480 ext4_free_blocks(handle, inode, NULL,
2481 EXT4_C2B(sbi, *partial_cluster),
2482 sbi->s_cluster_ratio, flags);
2483 *partial_cluster = 0;
2486 #ifdef EXTENTS_STATS
2488 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2489 spin_lock(&sbi->s_ext_stats_lock);
2490 sbi->s_ext_blocks += ee_len;
2491 sbi->s_ext_extents++;
2492 if (ee_len < sbi->s_ext_min)
2493 sbi->s_ext_min = ee_len;
2494 if (ee_len > sbi->s_ext_max)
2495 sbi->s_ext_max = ee_len;
2496 if (ext_depth(inode) > sbi->s_depth_max)
2497 sbi->s_depth_max = ext_depth(inode);
2498 spin_unlock(&sbi->s_ext_stats_lock);
2501 if (from >= le32_to_cpu(ex->ee_block)
2502 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2505 unsigned int unaligned;
2507 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2508 pblk = ext4_ext_pblock(ex) + ee_len - num;
2510 * Usually we want to free partial cluster at the end of the
2511 * extent, except for the situation when the cluster is still
2512 * used by any other extent (partial_cluster is negative).
2514 if (*partial_cluster < 0 &&
2515 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2516 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2518 ext_debug("free last %u blocks starting %llu partial %lld\n",
2519 num, pblk, *partial_cluster);
2520 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2522 * If the block range to be freed didn't start at the
2523 * beginning of a cluster, and we removed the entire
2524 * extent and the cluster is not used by any other extent,
2525 * save the partial cluster here, since we might need to
2526 * delete if we determine that the truncate operation has
2527 * removed all of the blocks in the cluster.
2529 * On the other hand, if we did not manage to free the whole
2530 * extent, we have to mark the cluster as used (store negative
2531 * cluster number in partial_cluster).
2533 unaligned = pblk & (sbi->s_cluster_ratio - 1);
2534 if (unaligned && (ee_len == num) &&
2535 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2536 *partial_cluster = EXT4_B2C(sbi, pblk);
2538 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2539 else if (*partial_cluster > 0)
2540 *partial_cluster = 0;
2542 ext4_error(sbi->s_sb, "strange request: removal(2) "
2543 "%u-%u from %u:%u\n",
2544 from, to, le32_to_cpu(ex->ee_block), ee_len);
2550 * ext4_ext_rm_leaf() Removes the extents associated with the
2551 * blocks appearing between "start" and "end", and splits the extents
2552 * if "start" and "end" appear in the same extent
2554 * @handle: The journal handle
2555 * @inode: The files inode
2556 * @path: The path to the leaf
2557 * @partial_cluster: The cluster which we'll have to free if all extents
2558 * has been released from it. It gets negative in case
2559 * that the cluster is still used.
2560 * @start: The first block to remove
2561 * @end: The last block to remove
2564 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2565 struct ext4_ext_path *path,
2566 long long *partial_cluster,
2567 ext4_lblk_t start, ext4_lblk_t end)
2569 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2570 int err = 0, correct_index = 0;
2571 int depth = ext_depth(inode), credits;
2572 struct ext4_extent_header *eh;
2575 ext4_lblk_t ex_ee_block;
2576 unsigned short ex_ee_len;
2577 unsigned uninitialized = 0;
2578 struct ext4_extent *ex;
2581 /* the header must be checked already in ext4_ext_remove_space() */
2582 ext_debug("truncate since %u in leaf to %u\n", start, end);
2583 if (!path[depth].p_hdr)
2584 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2585 eh = path[depth].p_hdr;
2586 if (unlikely(path[depth].p_hdr == NULL)) {
2587 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2590 /* find where to start removing */
2591 ex = path[depth].p_ext;
2593 ex = EXT_LAST_EXTENT(eh);
2595 ex_ee_block = le32_to_cpu(ex->ee_block);
2596 ex_ee_len = ext4_ext_get_actual_len(ex);
2598 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2600 while (ex >= EXT_FIRST_EXTENT(eh) &&
2601 ex_ee_block + ex_ee_len > start) {
2603 if (ext4_ext_is_uninitialized(ex))
2608 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2609 uninitialized, ex_ee_len);
2610 path[depth].p_ext = ex;
2612 a = ex_ee_block > start ? ex_ee_block : start;
2613 b = ex_ee_block+ex_ee_len - 1 < end ?
2614 ex_ee_block+ex_ee_len - 1 : end;
2616 ext_debug(" border %u:%u\n", a, b);
2618 /* If this extent is beyond the end of the hole, skip it */
2619 if (end < ex_ee_block) {
2621 * We're going to skip this extent and move to another,
2622 * so if this extent is not cluster aligned we have
2623 * to mark the current cluster as used to avoid
2624 * accidentally freeing it later on
2626 pblk = ext4_ext_pblock(ex);
2627 if (pblk & (sbi->s_cluster_ratio - 1))
2629 -((long long)EXT4_B2C(sbi, pblk));
2631 ex_ee_block = le32_to_cpu(ex->ee_block);
2632 ex_ee_len = ext4_ext_get_actual_len(ex);
2634 } else if (b != ex_ee_block + ex_ee_len - 1) {
2635 EXT4_ERROR_INODE(inode,
2636 "can not handle truncate %u:%u "
2638 start, end, ex_ee_block,
2639 ex_ee_block + ex_ee_len - 1);
2642 } else if (a != ex_ee_block) {
2643 /* remove tail of the extent */
2644 num = a - ex_ee_block;
2646 /* remove whole extent: excellent! */
2650 * 3 for leaf, sb, and inode plus 2 (bmap and group
2651 * descriptor) for each block group; assume two block
2652 * groups plus ex_ee_len/blocks_per_block_group for
2655 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2656 if (ex == EXT_FIRST_EXTENT(eh)) {
2658 credits += (ext_depth(inode)) + 1;
2660 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2662 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2666 err = ext4_ext_get_access(handle, inode, path + depth);
2670 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2676 /* this extent is removed; mark slot entirely unused */
2677 ext4_ext_store_pblock(ex, 0);
2679 ex->ee_len = cpu_to_le16(num);
2681 * Do not mark uninitialized if all the blocks in the
2682 * extent have been removed.
2684 if (uninitialized && num)
2685 ext4_ext_mark_uninitialized(ex);
2687 * If the extent was completely released,
2688 * we need to remove it from the leaf
2691 if (end != EXT_MAX_BLOCKS - 1) {
2693 * For hole punching, we need to scoot all the
2694 * extents up when an extent is removed so that
2695 * we dont have blank extents in the middle
2697 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2698 sizeof(struct ext4_extent));
2700 /* Now get rid of the one at the end */
2701 memset(EXT_LAST_EXTENT(eh), 0,
2702 sizeof(struct ext4_extent));
2704 le16_add_cpu(&eh->eh_entries, -1);
2705 } else if (*partial_cluster > 0)
2706 *partial_cluster = 0;
2708 err = ext4_ext_dirty(handle, inode, path + depth);
2712 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2713 ext4_ext_pblock(ex));
2715 ex_ee_block = le32_to_cpu(ex->ee_block);
2716 ex_ee_len = ext4_ext_get_actual_len(ex);
2719 if (correct_index && eh->eh_entries)
2720 err = ext4_ext_correct_indexes(handle, inode, path);
2723 * Free the partial cluster only if the current extent does not
2724 * reference it. Otherwise we might free used cluster.
2726 if (*partial_cluster > 0 &&
2727 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2728 *partial_cluster)) {
2729 int flags = get_default_free_blocks_flags(inode);
2731 ext4_free_blocks(handle, inode, NULL,
2732 EXT4_C2B(sbi, *partial_cluster),
2733 sbi->s_cluster_ratio, flags);
2734 *partial_cluster = 0;
2737 /* if this leaf is free, then we should
2738 * remove it from index block above */
2739 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2740 err = ext4_ext_rm_idx(handle, inode, path, depth);
2747 * ext4_ext_more_to_rm:
2748 * returns 1 if current index has to be freed (even partial)
2751 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2753 BUG_ON(path->p_idx == NULL);
2755 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2759 * if truncate on deeper level happened, it wasn't partial,
2760 * so we have to consider current index for truncation
2762 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2767 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2770 struct super_block *sb = inode->i_sb;
2771 int depth = ext_depth(inode);
2772 struct ext4_ext_path *path = NULL;
2773 long long partial_cluster = 0;
2777 ext_debug("truncate since %u to %u\n", start, end);
2779 /* probably first extent we're gonna free will be last in block */
2780 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2782 return PTR_ERR(handle);
2785 trace_ext4_ext_remove_space(inode, start, end, depth);
2788 * Check if we are removing extents inside the extent tree. If that
2789 * is the case, we are going to punch a hole inside the extent tree
2790 * so we have to check whether we need to split the extent covering
2791 * the last block to remove so we can easily remove the part of it
2792 * in ext4_ext_rm_leaf().
2794 if (end < EXT_MAX_BLOCKS - 1) {
2795 struct ext4_extent *ex;
2796 ext4_lblk_t ee_block;
2798 /* find extent for this block */
2799 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2801 ext4_journal_stop(handle);
2802 return PTR_ERR(path);
2804 depth = ext_depth(inode);
2805 /* Leaf not may not exist only if inode has no blocks at all */
2806 ex = path[depth].p_ext;
2809 EXT4_ERROR_INODE(inode,
2810 "path[%d].p_hdr == NULL",
2817 ee_block = le32_to_cpu(ex->ee_block);
2820 * See if the last block is inside the extent, if so split
2821 * the extent at 'end' block so we can easily remove the
2822 * tail of the first part of the split extent in
2823 * ext4_ext_rm_leaf().
2825 if (end >= ee_block &&
2826 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2829 if (ext4_ext_is_uninitialized(ex))
2830 split_flag = EXT4_EXT_MARK_UNINIT1 |
2831 EXT4_EXT_MARK_UNINIT2;
2834 * Split the extent in two so that 'end' is the last
2835 * block in the first new extent. Also we should not
2836 * fail removing space due to ENOSPC so try to use
2837 * reserved block if that happens.
2839 err = ext4_split_extent_at(handle, inode, path,
2840 end + 1, split_flag,
2842 EXT4_GET_BLOCKS_PRE_IO |
2843 EXT4_GET_BLOCKS_METADATA_NOFAIL);
2850 * We start scanning from right side, freeing all the blocks
2851 * after i_size and walking into the tree depth-wise.
2853 depth = ext_depth(inode);
2858 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2860 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2863 ext4_journal_stop(handle);
2866 path[0].p_depth = depth;
2867 path[0].p_hdr = ext_inode_hdr(inode);
2870 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2877 while (i >= 0 && err == 0) {
2879 /* this is leaf block */
2880 err = ext4_ext_rm_leaf(handle, inode, path,
2881 &partial_cluster, start,
2883 /* root level has p_bh == NULL, brelse() eats this */
2884 brelse(path[i].p_bh);
2885 path[i].p_bh = NULL;
2890 /* this is index block */
2891 if (!path[i].p_hdr) {
2892 ext_debug("initialize header\n");
2893 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2896 if (!path[i].p_idx) {
2897 /* this level hasn't been touched yet */
2898 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2899 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2900 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2902 le16_to_cpu(path[i].p_hdr->eh_entries));
2904 /* we were already here, see at next index */
2908 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2909 i, EXT_FIRST_INDEX(path[i].p_hdr),
2911 if (ext4_ext_more_to_rm(path + i)) {
2912 struct buffer_head *bh;
2913 /* go to the next level */
2914 ext_debug("move to level %d (block %llu)\n",
2915 i + 1, ext4_idx_pblock(path[i].p_idx));
2916 memset(path + i + 1, 0, sizeof(*path));
2917 bh = read_extent_tree_block(inode,
2918 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2921 /* should we reset i_size? */
2925 /* Yield here to deal with large extent trees.
2926 * Should be a no-op if we did IO above. */
2928 if (WARN_ON(i + 1 > depth)) {
2932 path[i + 1].p_bh = bh;
2934 /* save actual number of indexes since this
2935 * number is changed at the next iteration */
2936 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2939 /* we finished processing this index, go up */
2940 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2941 /* index is empty, remove it;
2942 * handle must be already prepared by the
2943 * truncatei_leaf() */
2944 err = ext4_ext_rm_idx(handle, inode, path, i);
2946 /* root level has p_bh == NULL, brelse() eats this */
2947 brelse(path[i].p_bh);
2948 path[i].p_bh = NULL;
2950 ext_debug("return to level %d\n", i);
2954 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2955 partial_cluster, path->p_hdr->eh_entries);
2957 /* If we still have something in the partial cluster and we have removed
2958 * even the first extent, then we should free the blocks in the partial
2959 * cluster as well. */
2960 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2961 int flags = get_default_free_blocks_flags(inode);
2963 ext4_free_blocks(handle, inode, NULL,
2964 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2965 EXT4_SB(sb)->s_cluster_ratio, flags);
2966 partial_cluster = 0;
2969 /* TODO: flexible tree reduction should be here */
2970 if (path->p_hdr->eh_entries == 0) {
2972 * truncate to zero freed all the tree,
2973 * so we need to correct eh_depth
2975 err = ext4_ext_get_access(handle, inode, path);
2977 ext_inode_hdr(inode)->eh_depth = 0;
2978 ext_inode_hdr(inode)->eh_max =
2979 cpu_to_le16(ext4_ext_space_root(inode, 0));
2980 err = ext4_ext_dirty(handle, inode, path);
2984 ext4_ext_drop_refs(path);
2986 if (err == -EAGAIN) {
2990 ext4_journal_stop(handle);
2996 * called at mount time
2998 void ext4_ext_init(struct super_block *sb)
3001 * possible initialization would be here
3004 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3005 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3006 printk(KERN_INFO "EXT4-fs: file extents enabled"
3007 #ifdef AGGRESSIVE_TEST
3008 ", aggressive tests"
3010 #ifdef CHECK_BINSEARCH
3013 #ifdef EXTENTS_STATS
3018 #ifdef EXTENTS_STATS
3019 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3020 EXT4_SB(sb)->s_ext_min = 1 << 30;
3021 EXT4_SB(sb)->s_ext_max = 0;
3027 * called at umount time
3029 void ext4_ext_release(struct super_block *sb)
3031 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3034 #ifdef EXTENTS_STATS
3035 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3036 struct ext4_sb_info *sbi = EXT4_SB(sb);
3037 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3038 sbi->s_ext_blocks, sbi->s_ext_extents,
3039 sbi->s_ext_blocks / sbi->s_ext_extents);
3040 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3041 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3046 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3048 ext4_lblk_t ee_block;
3049 ext4_fsblk_t ee_pblock;
3050 unsigned int ee_len;
3052 ee_block = le32_to_cpu(ex->ee_block);
3053 ee_len = ext4_ext_get_actual_len(ex);
3054 ee_pblock = ext4_ext_pblock(ex);
3059 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3060 EXTENT_STATUS_WRITTEN);
3063 /* FIXME!! we need to try to merge to left or right after zero-out */
3064 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3066 ext4_fsblk_t ee_pblock;
3067 unsigned int ee_len;
3070 ee_len = ext4_ext_get_actual_len(ex);
3071 ee_pblock = ext4_ext_pblock(ex);
3073 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3081 * ext4_split_extent_at() splits an extent at given block.
3083 * @handle: the journal handle
3084 * @inode: the file inode
3085 * @path: the path to the extent
3086 * @split: the logical block where the extent is splitted.
3087 * @split_flags: indicates if the extent could be zeroout if split fails, and
3088 * the states(init or uninit) of new extents.
3089 * @flags: flags used to insert new extent to extent tree.
3092 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3093 * of which are deterimined by split_flag.
3095 * There are two cases:
3096 * a> the extent are splitted into two extent.
3097 * b> split is not needed, and just mark the extent.
3099 * return 0 on success.
3101 static int ext4_split_extent_at(handle_t *handle,
3102 struct inode *inode,
3103 struct ext4_ext_path *path,
3108 ext4_fsblk_t newblock;
3109 ext4_lblk_t ee_block;
3110 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3111 struct ext4_extent *ex2 = NULL;
3112 unsigned int ee_len, depth;
3115 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3116 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3118 ext_debug("ext4_split_extents_at: inode %lu, logical"
3119 "block %llu\n", inode->i_ino, (unsigned long long)split);
3121 ext4_ext_show_leaf(inode, path);
3123 depth = ext_depth(inode);
3124 ex = path[depth].p_ext;
3125 ee_block = le32_to_cpu(ex->ee_block);
3126 ee_len = ext4_ext_get_actual_len(ex);
3127 newblock = split - ee_block + ext4_ext_pblock(ex);
3129 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3130 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3131 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3132 EXT4_EXT_MARK_UNINIT1 |
3133 EXT4_EXT_MARK_UNINIT2));
3135 err = ext4_ext_get_access(handle, inode, path + depth);
3139 if (split == ee_block) {
3141 * case b: block @split is the block that the extent begins with
3142 * then we just change the state of the extent, and splitting
3145 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3146 ext4_ext_mark_uninitialized(ex);
3148 ext4_ext_mark_initialized(ex);
3150 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3151 ext4_ext_try_to_merge(handle, inode, path, ex);
3153 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3158 memcpy(&orig_ex, ex, sizeof(orig_ex));
3159 ex->ee_len = cpu_to_le16(split - ee_block);
3160 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3161 ext4_ext_mark_uninitialized(ex);
3164 * path may lead to new leaf, not to original leaf any more
3165 * after ext4_ext_insert_extent() returns,
3167 err = ext4_ext_dirty(handle, inode, path + depth);
3169 goto fix_extent_len;
3172 ex2->ee_block = cpu_to_le32(split);
3173 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3174 ext4_ext_store_pblock(ex2, newblock);
3175 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3176 ext4_ext_mark_uninitialized(ex2);
3178 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3179 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3180 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3181 if (split_flag & EXT4_EXT_DATA_VALID1) {
3182 err = ext4_ext_zeroout(inode, ex2);
3183 zero_ex.ee_block = ex2->ee_block;
3184 zero_ex.ee_len = cpu_to_le16(
3185 ext4_ext_get_actual_len(ex2));
3186 ext4_ext_store_pblock(&zero_ex,
3187 ext4_ext_pblock(ex2));
3189 err = ext4_ext_zeroout(inode, ex);
3190 zero_ex.ee_block = ex->ee_block;
3191 zero_ex.ee_len = cpu_to_le16(
3192 ext4_ext_get_actual_len(ex));
3193 ext4_ext_store_pblock(&zero_ex,
3194 ext4_ext_pblock(ex));
3197 err = ext4_ext_zeroout(inode, &orig_ex);
3198 zero_ex.ee_block = orig_ex.ee_block;
3199 zero_ex.ee_len = cpu_to_le16(
3200 ext4_ext_get_actual_len(&orig_ex));
3201 ext4_ext_store_pblock(&zero_ex,
3202 ext4_ext_pblock(&orig_ex));
3206 goto fix_extent_len;
3207 /* update the extent length and mark as initialized */
3208 ex->ee_len = cpu_to_le16(ee_len);
3209 ext4_ext_try_to_merge(handle, inode, path, ex);
3210 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3212 goto fix_extent_len;
3214 /* update extent status tree */
3215 err = ext4_zeroout_es(inode, &zero_ex);
3219 goto fix_extent_len;
3222 ext4_ext_show_leaf(inode, path);
3226 ex->ee_len = orig_ex.ee_len;
3227 ext4_ext_dirty(handle, inode, path + depth);
3232 * ext4_split_extents() splits an extent and mark extent which is covered
3233 * by @map as split_flags indicates
3235 * It may result in splitting the extent into multiple extents (up to three)
3236 * There are three possibilities:
3237 * a> There is no split required
3238 * b> Splits in two extents: Split is happening at either end of the extent
3239 * c> Splits in three extents: Somone is splitting in middle of the extent
3242 static int ext4_split_extent(handle_t *handle,
3243 struct inode *inode,
3244 struct ext4_ext_path *path,
3245 struct ext4_map_blocks *map,
3249 ext4_lblk_t ee_block;
3250 struct ext4_extent *ex;
3251 unsigned int ee_len, depth;
3254 int split_flag1, flags1;
3255 int allocated = map->m_len;
3257 depth = ext_depth(inode);
3258 ex = path[depth].p_ext;
3259 ee_block = le32_to_cpu(ex->ee_block);
3260 ee_len = ext4_ext_get_actual_len(ex);
3261 uninitialized = ext4_ext_is_uninitialized(ex);
3263 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3264 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3265 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3267 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3268 EXT4_EXT_MARK_UNINIT2;
3269 if (split_flag & EXT4_EXT_DATA_VALID2)
3270 split_flag1 |= EXT4_EXT_DATA_VALID1;
3271 err = ext4_split_extent_at(handle, inode, path,
3272 map->m_lblk + map->m_len, split_flag1, flags1);
3276 allocated = ee_len - (map->m_lblk - ee_block);
3279 * Update path is required because previous ext4_split_extent_at() may
3280 * result in split of original leaf or extent zeroout.
3282 ext4_ext_drop_refs(path);
3283 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3285 return PTR_ERR(path);
3286 depth = ext_depth(inode);
3287 ex = path[depth].p_ext;
3288 uninitialized = ext4_ext_is_uninitialized(ex);
3291 if (map->m_lblk >= ee_block) {
3292 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3293 if (uninitialized) {
3294 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3295 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3296 EXT4_EXT_MARK_UNINIT2);
3298 err = ext4_split_extent_at(handle, inode, path,
3299 map->m_lblk, split_flag1, flags);
3304 ext4_ext_show_leaf(inode, path);
3306 return err ? err : allocated;
3310 * This function is called by ext4_ext_map_blocks() if someone tries to write
3311 * to an uninitialized extent. It may result in splitting the uninitialized
3312 * extent into multiple extents (up to three - one initialized and two
3314 * There are three possibilities:
3315 * a> There is no split required: Entire extent should be initialized
3316 * b> Splits in two extents: Write is happening at either end of the extent
3317 * c> Splits in three extents: Somone is writing in middle of the extent
3320 * - The extent pointed to by 'path' is uninitialized.
3321 * - The extent pointed to by 'path' contains a superset
3322 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3324 * Post-conditions on success:
3325 * - the returned value is the number of blocks beyond map->l_lblk
3326 * that are allocated and initialized.
3327 * It is guaranteed to be >= map->m_len.
3329 static int ext4_ext_convert_to_initialized(handle_t *handle,
3330 struct inode *inode,
3331 struct ext4_map_blocks *map,
3332 struct ext4_ext_path *path,
3335 struct ext4_sb_info *sbi;
3336 struct ext4_extent_header *eh;
3337 struct ext4_map_blocks split_map;
3338 struct ext4_extent zero_ex;
3339 struct ext4_extent *ex, *abut_ex;
3340 ext4_lblk_t ee_block, eof_block;
3341 unsigned int ee_len, depth, map_len = map->m_len;
3342 int allocated = 0, max_zeroout = 0;
3346 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3347 "block %llu, max_blocks %u\n", inode->i_ino,
3348 (unsigned long long)map->m_lblk, map_len);
3350 sbi = EXT4_SB(inode->i_sb);
3351 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3352 inode->i_sb->s_blocksize_bits;
3353 if (eof_block < map->m_lblk + map_len)
3354 eof_block = map->m_lblk + map_len;
3356 depth = ext_depth(inode);
3357 eh = path[depth].p_hdr;
3358 ex = path[depth].p_ext;
3359 ee_block = le32_to_cpu(ex->ee_block);
3360 ee_len = ext4_ext_get_actual_len(ex);
3363 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3365 /* Pre-conditions */
3366 BUG_ON(!ext4_ext_is_uninitialized(ex));
3367 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3370 * Attempt to transfer newly initialized blocks from the currently
3371 * uninitialized extent to its neighbor. This is much cheaper
3372 * than an insertion followed by a merge as those involve costly
3373 * memmove() calls. Transferring to the left is the common case in
3374 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3375 * followed by append writes.
3377 * Limitations of the current logic:
3378 * - L1: we do not deal with writes covering the whole extent.
3379 * This would require removing the extent if the transfer
3381 * - L2: we only attempt to merge with an extent stored in the
3382 * same extent tree node.
3384 if ((map->m_lblk == ee_block) &&
3385 /* See if we can merge left */
3386 (map_len < ee_len) && /*L1*/
3387 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3388 ext4_lblk_t prev_lblk;
3389 ext4_fsblk_t prev_pblk, ee_pblk;
3390 unsigned int prev_len;
3393 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3394 prev_len = ext4_ext_get_actual_len(abut_ex);
3395 prev_pblk = ext4_ext_pblock(abut_ex);
3396 ee_pblk = ext4_ext_pblock(ex);
3399 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3400 * upon those conditions:
3401 * - C1: abut_ex is initialized,
3402 * - C2: abut_ex is logically abutting ex,
3403 * - C3: abut_ex is physically abutting ex,
3404 * - C4: abut_ex can receive the additional blocks without
3405 * overflowing the (initialized) length limit.
3407 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3408 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3409 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3410 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3411 err = ext4_ext_get_access(handle, inode, path + depth);
3415 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3418 /* Shift the start of ex by 'map_len' blocks */
3419 ex->ee_block = cpu_to_le32(ee_block + map_len);
3420 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3421 ex->ee_len = cpu_to_le16(ee_len - map_len);
3422 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3424 /* Extend abut_ex by 'map_len' blocks */
3425 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3427 /* Result: number of initialized blocks past m_lblk */
3428 allocated = map_len;
3430 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3431 (map_len < ee_len) && /*L1*/
3432 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3433 /* See if we can merge right */
3434 ext4_lblk_t next_lblk;
3435 ext4_fsblk_t next_pblk, ee_pblk;
3436 unsigned int next_len;
3439 next_lblk = le32_to_cpu(abut_ex->ee_block);
3440 next_len = ext4_ext_get_actual_len(abut_ex);
3441 next_pblk = ext4_ext_pblock(abut_ex);
3442 ee_pblk = ext4_ext_pblock(ex);
3445 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3446 * upon those conditions:
3447 * - C1: abut_ex is initialized,
3448 * - C2: abut_ex is logically abutting ex,
3449 * - C3: abut_ex is physically abutting ex,
3450 * - C4: abut_ex can receive the additional blocks without
3451 * overflowing the (initialized) length limit.
3453 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3454 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3455 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3456 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3457 err = ext4_ext_get_access(handle, inode, path + depth);
3461 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3464 /* Shift the start of abut_ex by 'map_len' blocks */
3465 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3466 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3467 ex->ee_len = cpu_to_le16(ee_len - map_len);
3468 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3470 /* Extend abut_ex by 'map_len' blocks */
3471 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3473 /* Result: number of initialized blocks past m_lblk */
3474 allocated = map_len;
3478 /* Mark the block containing both extents as dirty */
3479 ext4_ext_dirty(handle, inode, path + depth);
3481 /* Update path to point to the right extent */
3482 path[depth].p_ext = abut_ex;
3485 allocated = ee_len - (map->m_lblk - ee_block);
3487 WARN_ON(map->m_lblk < ee_block);
3489 * It is safe to convert extent to initialized via explicit
3490 * zeroout only if extent is fully insde i_size or new_size.
3492 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3494 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3495 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3496 (inode->i_sb->s_blocksize_bits - 10);
3498 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3499 if (max_zeroout && (ee_len <= max_zeroout)) {
3500 err = ext4_ext_zeroout(inode, ex);
3503 zero_ex.ee_block = ex->ee_block;
3504 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3505 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3507 err = ext4_ext_get_access(handle, inode, path + depth);
3510 ext4_ext_mark_initialized(ex);
3511 ext4_ext_try_to_merge(handle, inode, path, ex);
3512 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3518 * 1. split the extent into three extents.
3519 * 2. split the extent into two extents, zeroout the first half.
3520 * 3. split the extent into two extents, zeroout the second half.
3521 * 4. split the extent into two extents with out zeroout.
3523 split_map.m_lblk = map->m_lblk;
3524 split_map.m_len = map->m_len;
3526 if (max_zeroout && (allocated > map->m_len)) {
3527 if (allocated <= max_zeroout) {
3530 cpu_to_le32(map->m_lblk);
3531 zero_ex.ee_len = cpu_to_le16(allocated);
3532 ext4_ext_store_pblock(&zero_ex,
3533 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3534 err = ext4_ext_zeroout(inode, &zero_ex);
3537 split_map.m_lblk = map->m_lblk;
3538 split_map.m_len = allocated;
3539 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3541 if (map->m_lblk != ee_block) {
3542 zero_ex.ee_block = ex->ee_block;
3543 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3545 ext4_ext_store_pblock(&zero_ex,
3546 ext4_ext_pblock(ex));
3547 err = ext4_ext_zeroout(inode, &zero_ex);
3552 split_map.m_lblk = ee_block;
3553 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3554 allocated = map->m_len;
3558 allocated = ext4_split_extent(handle, inode, path,
3559 &split_map, split_flag, flags);
3564 /* If we have gotten a failure, don't zero out status tree */
3566 err = ext4_zeroout_es(inode, &zero_ex);
3567 return err ? err : allocated;
3571 * This function is called by ext4_ext_map_blocks() from
3572 * ext4_get_blocks_dio_write() when DIO to write
3573 * to an uninitialized extent.
3575 * Writing to an uninitialized extent may result in splitting the uninitialized
3576 * extent into multiple initialized/uninitialized extents (up to three)
3577 * There are three possibilities:
3578 * a> There is no split required: Entire extent should be uninitialized
3579 * b> Splits in two extents: Write is happening at either end of the extent
3580 * c> Splits in three extents: Somone is writing in middle of the extent
3582 * One of more index blocks maybe needed if the extent tree grow after
3583 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3584 * complete, we need to split the uninitialized extent before DIO submit
3585 * the IO. The uninitialized extent called at this time will be split
3586 * into three uninitialized extent(at most). After IO complete, the part
3587 * being filled will be convert to initialized by the end_io callback function
3588 * via ext4_convert_unwritten_extents().
3590 * Returns the size of uninitialized extent to be written on success.
3592 static int ext4_split_unwritten_extents(handle_t *handle,
3593 struct inode *inode,
3594 struct ext4_map_blocks *map,
3595 struct ext4_ext_path *path,
3598 ext4_lblk_t eof_block;
3599 ext4_lblk_t ee_block;
3600 struct ext4_extent *ex;
3601 unsigned int ee_len;
3602 int split_flag = 0, depth;
3604 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3605 "block %llu, max_blocks %u\n", inode->i_ino,
3606 (unsigned long long)map->m_lblk, map->m_len);
3608 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3609 inode->i_sb->s_blocksize_bits;
3610 if (eof_block < map->m_lblk + map->m_len)
3611 eof_block = map->m_lblk + map->m_len;
3613 * It is safe to convert extent to initialized via explicit
3614 * zeroout only if extent is fully insde i_size or new_size.
3616 depth = ext_depth(inode);
3617 ex = path[depth].p_ext;
3618 ee_block = le32_to_cpu(ex->ee_block);
3619 ee_len = ext4_ext_get_actual_len(ex);
3621 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3622 split_flag |= EXT4_EXT_MARK_UNINIT2;
3623 if (flags & EXT4_GET_BLOCKS_CONVERT)
3624 split_flag |= EXT4_EXT_DATA_VALID2;
3625 flags |= EXT4_GET_BLOCKS_PRE_IO;
3626 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3629 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3630 struct inode *inode,
3631 struct ext4_map_blocks *map,
3632 struct ext4_ext_path *path)
3634 struct ext4_extent *ex;
3635 ext4_lblk_t ee_block;
3636 unsigned int ee_len;
3640 depth = ext_depth(inode);
3641 ex = path[depth].p_ext;
3642 ee_block = le32_to_cpu(ex->ee_block);
3643 ee_len = ext4_ext_get_actual_len(ex);
3645 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3646 "block %llu, max_blocks %u\n", inode->i_ino,
3647 (unsigned long long)ee_block, ee_len);
3649 /* If extent is larger than requested it is a clear sign that we still
3650 * have some extent state machine issues left. So extent_split is still
3652 * TODO: Once all related issues will be fixed this situation should be
3655 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3657 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3658 " len %u; IO logical block %llu, len %u\n",
3659 inode->i_ino, (unsigned long long)ee_block, ee_len,
3660 (unsigned long long)map->m_lblk, map->m_len);
3662 err = ext4_split_unwritten_extents(handle, inode, map, path,
3663 EXT4_GET_BLOCKS_CONVERT);
3666 ext4_ext_drop_refs(path);
3667 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3669 err = PTR_ERR(path);
3672 depth = ext_depth(inode);
3673 ex = path[depth].p_ext;
3676 err = ext4_ext_get_access(handle, inode, path + depth);
3679 /* first mark the extent as initialized */
3680 ext4_ext_mark_initialized(ex);
3682 /* note: ext4_ext_correct_indexes() isn't needed here because
3683 * borders are not changed
3685 ext4_ext_try_to_merge(handle, inode, path, ex);
3687 /* Mark modified extent as dirty */
3688 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3690 ext4_ext_show_leaf(inode, path);
3694 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3695 sector_t block, int count)
3698 for (i = 0; i < count; i++)
3699 unmap_underlying_metadata(bdev, block + i);
3703 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3705 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3707 struct ext4_ext_path *path,
3711 struct ext4_extent_header *eh;
3712 struct ext4_extent *last_ex;
3714 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3717 depth = ext_depth(inode);
3718 eh = path[depth].p_hdr;
3721 * We're going to remove EOFBLOCKS_FL entirely in future so we
3722 * do not care for this case anymore. Simply remove the flag
3723 * if there are no extents.
3725 if (unlikely(!eh->eh_entries))
3727 last_ex = EXT_LAST_EXTENT(eh);
3729 * We should clear the EOFBLOCKS_FL flag if we are writing the
3730 * last block in the last extent in the file. We test this by
3731 * first checking to see if the caller to
3732 * ext4_ext_get_blocks() was interested in the last block (or
3733 * a block beyond the last block) in the current extent. If
3734 * this turns out to be false, we can bail out from this
3735 * function immediately.
3737 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3738 ext4_ext_get_actual_len(last_ex))
3741 * If the caller does appear to be planning to write at or
3742 * beyond the end of the current extent, we then test to see
3743 * if the current extent is the last extent in the file, by
3744 * checking to make sure it was reached via the rightmost node
3745 * at each level of the tree.
3747 for (i = depth-1; i >= 0; i--)
3748 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3751 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3752 return ext4_mark_inode_dirty(handle, inode);
3756 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3758 * Return 1 if there is a delalloc block in the range, otherwise 0.
3760 int ext4_find_delalloc_range(struct inode *inode,
3761 ext4_lblk_t lblk_start,
3762 ext4_lblk_t lblk_end)
3764 struct extent_status es;
3766 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3768 return 0; /* there is no delay extent in this tree */
3769 else if (es.es_lblk <= lblk_start &&
3770 lblk_start < es.es_lblk + es.es_len)
3772 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3778 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3780 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3781 ext4_lblk_t lblk_start, lblk_end;
3782 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3783 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3785 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3789 * Determines how many complete clusters (out of those specified by the 'map')
3790 * are under delalloc and were reserved quota for.
3791 * This function is called when we are writing out the blocks that were
3792 * originally written with their allocation delayed, but then the space was
3793 * allocated using fallocate() before the delayed allocation could be resolved.
3794 * The cases to look for are:
3795 * ('=' indicated delayed allocated blocks
3796 * '-' indicates non-delayed allocated blocks)
3797 * (a) partial clusters towards beginning and/or end outside of allocated range
3798 * are not delalloc'ed.
3800 * |----c---=|====c====|====c====|===-c----|
3801 * |++++++ allocated ++++++|
3802 * ==> 4 complete clusters in above example
3804 * (b) partial cluster (outside of allocated range) towards either end is
3805 * marked for delayed allocation. In this case, we will exclude that
3808 * |----====c========|========c========|
3809 * |++++++ allocated ++++++|
3810 * ==> 1 complete clusters in above example
3813 * |================c================|
3814 * |++++++ allocated ++++++|
3815 * ==> 0 complete clusters in above example
3817 * The ext4_da_update_reserve_space will be called only if we
3818 * determine here that there were some "entire" clusters that span
3819 * this 'allocated' range.
3820 * In the non-bigalloc case, this function will just end up returning num_blks
3821 * without ever calling ext4_find_delalloc_range.
3824 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3825 unsigned int num_blks)
3827 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3828 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3829 ext4_lblk_t lblk_from, lblk_to, c_offset;
3830 unsigned int allocated_clusters = 0;
3832 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3833 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3835 /* max possible clusters for this allocation */
3836 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3838 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3840 /* Check towards left side */
3841 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3843 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3844 lblk_to = lblk_from + c_offset - 1;
3846 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3847 allocated_clusters--;
3850 /* Now check towards right. */
3851 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3852 if (allocated_clusters && c_offset) {
3853 lblk_from = lblk_start + num_blks;
3854 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3856 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3857 allocated_clusters--;
3860 return allocated_clusters;
3864 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3865 struct ext4_map_blocks *map,
3866 struct ext4_ext_path *path, int flags,
3867 unsigned int allocated, ext4_fsblk_t newblock)
3871 ext4_io_end_t *io = ext4_inode_aio(inode);
3873 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3874 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3875 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3877 ext4_ext_show_leaf(inode, path);
3880 * When writing into uninitialized space, we should not fail to
3881 * allocate metadata blocks for the new extent block if needed.
3883 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3885 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3886 allocated, newblock);
3888 /* get_block() before submit the IO, split the extent */
3889 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3890 ret = ext4_split_unwritten_extents(handle, inode, map,
3895 * Flag the inode(non aio case) or end_io struct (aio case)
3896 * that this IO needs to conversion to written when IO is
3900 ext4_set_io_unwritten_flag(inode, io);
3902 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3903 map->m_flags |= EXT4_MAP_UNWRITTEN;
3904 if (ext4_should_dioread_nolock(inode))
3905 map->m_flags |= EXT4_MAP_UNINIT;
3908 /* IO end_io complete, convert the filled extent to written */
3909 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3910 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3913 ext4_update_inode_fsync_trans(handle, inode, 1);
3914 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3918 map->m_flags |= EXT4_MAP_MAPPED;
3919 if (allocated > map->m_len)
3920 allocated = map->m_len;
3921 map->m_len = allocated;
3924 /* buffered IO case */
3926 * repeat fallocate creation request
3927 * we already have an unwritten extent
3929 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3930 map->m_flags |= EXT4_MAP_UNWRITTEN;
3934 /* buffered READ or buffered write_begin() lookup */
3935 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3937 * We have blocks reserved already. We
3938 * return allocated blocks so that delalloc
3939 * won't do block reservation for us. But
3940 * the buffer head will be unmapped so that
3941 * a read from the block returns 0s.
3943 map->m_flags |= EXT4_MAP_UNWRITTEN;
3947 /* buffered write, writepage time, convert*/
3948 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
3950 ext4_update_inode_fsync_trans(handle, inode, 1);
3957 map->m_flags |= EXT4_MAP_NEW;
3959 * if we allocated more blocks than requested
3960 * we need to make sure we unmap the extra block
3961 * allocated. The actual needed block will get
3962 * unmapped later when we find the buffer_head marked
3965 if (allocated > map->m_len) {
3966 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3967 newblock + map->m_len,
3968 allocated - map->m_len);
3969 allocated = map->m_len;
3971 map->m_len = allocated;
3974 * If we have done fallocate with the offset that is already
3975 * delayed allocated, we would have block reservation
3976 * and quota reservation done in the delayed write path.
3977 * But fallocate would have already updated quota and block
3978 * count for this offset. So cancel these reservation
3980 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3981 unsigned int reserved_clusters;
3982 reserved_clusters = get_reserved_cluster_alloc(inode,
3983 map->m_lblk, map->m_len);
3984 if (reserved_clusters)
3985 ext4_da_update_reserve_space(inode,
3991 map->m_flags |= EXT4_MAP_MAPPED;
3992 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3993 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3999 if (allocated > map->m_len)
4000 allocated = map->m_len;
4001 ext4_ext_show_leaf(inode, path);
4002 map->m_pblk = newblock;
4003 map->m_len = allocated;
4006 ext4_ext_drop_refs(path);
4009 return err ? err : allocated;
4013 * get_implied_cluster_alloc - check to see if the requested
4014 * allocation (in the map structure) overlaps with a cluster already
4015 * allocated in an extent.
4016 * @sb The filesystem superblock structure
4017 * @map The requested lblk->pblk mapping
4018 * @ex The extent structure which might contain an implied
4019 * cluster allocation
4021 * This function is called by ext4_ext_map_blocks() after we failed to
4022 * find blocks that were already in the inode's extent tree. Hence,
4023 * we know that the beginning of the requested region cannot overlap
4024 * the extent from the inode's extent tree. There are three cases we
4025 * want to catch. The first is this case:
4027 * |--- cluster # N--|
4028 * |--- extent ---| |---- requested region ---|
4031 * The second case that we need to test for is this one:
4033 * |--------- cluster # N ----------------|
4034 * |--- requested region --| |------- extent ----|
4035 * |=======================|
4037 * The third case is when the requested region lies between two extents
4038 * within the same cluster:
4039 * |------------- cluster # N-------------|
4040 * |----- ex -----| |---- ex_right ----|
4041 * |------ requested region ------|
4042 * |================|
4044 * In each of the above cases, we need to set the map->m_pblk and
4045 * map->m_len so it corresponds to the return the extent labelled as
4046 * "|====|" from cluster #N, since it is already in use for data in
4047 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4048 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4049 * as a new "allocated" block region. Otherwise, we will return 0 and
4050 * ext4_ext_map_blocks() will then allocate one or more new clusters
4051 * by calling ext4_mb_new_blocks().
4053 static int get_implied_cluster_alloc(struct super_block *sb,
4054 struct ext4_map_blocks *map,
4055 struct ext4_extent *ex,
4056 struct ext4_ext_path *path)
4058 struct ext4_sb_info *sbi = EXT4_SB(sb);
4059 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4060 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4061 ext4_lblk_t rr_cluster_start;
4062 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4063 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4064 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4066 /* The extent passed in that we are trying to match */
4067 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4068 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4070 /* The requested region passed into ext4_map_blocks() */
4071 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4073 if ((rr_cluster_start == ex_cluster_end) ||
4074 (rr_cluster_start == ex_cluster_start)) {
4075 if (rr_cluster_start == ex_cluster_end)
4076 ee_start += ee_len - 1;
4077 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
4079 map->m_len = min(map->m_len,
4080 (unsigned) sbi->s_cluster_ratio - c_offset);
4082 * Check for and handle this case:
4084 * |--------- cluster # N-------------|
4085 * |------- extent ----|
4086 * |--- requested region ---|
4090 if (map->m_lblk < ee_block)
4091 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4094 * Check for the case where there is already another allocated
4095 * block to the right of 'ex' but before the end of the cluster.
4097 * |------------- cluster # N-------------|
4098 * |----- ex -----| |---- ex_right ----|
4099 * |------ requested region ------|
4100 * |================|
4102 if (map->m_lblk > ee_block) {
4103 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4104 map->m_len = min(map->m_len, next - map->m_lblk);
4107 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4111 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4117 * Block allocation/map/preallocation routine for extents based files
4120 * Need to be called with
4121 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4122 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4124 * return > 0, number of of blocks already mapped/allocated
4125 * if create == 0 and these are pre-allocated blocks
4126 * buffer head is unmapped
4127 * otherwise blocks are mapped
4129 * return = 0, if plain look up failed (blocks have not been allocated)
4130 * buffer head is unmapped
4132 * return < 0, error case.
4134 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4135 struct ext4_map_blocks *map, int flags)
4137 struct ext4_ext_path *path = NULL;
4138 struct ext4_extent newex, *ex, *ex2;
4139 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4140 ext4_fsblk_t newblock = 0;
4141 int free_on_err = 0, err = 0, depth;
4142 unsigned int allocated = 0, offset = 0;
4143 unsigned int allocated_clusters = 0;
4144 struct ext4_allocation_request ar;
4145 ext4_io_end_t *io = ext4_inode_aio(inode);
4146 ext4_lblk_t cluster_offset;
4147 int set_unwritten = 0;
4149 ext_debug("blocks %u/%u requested for inode %lu\n",
4150 map->m_lblk, map->m_len, inode->i_ino);
4151 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4153 /* find extent for this block */
4154 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4156 err = PTR_ERR(path);
4161 depth = ext_depth(inode);
4164 * consistent leaf must not be empty;
4165 * this situation is possible, though, _during_ tree modification;
4166 * this is why assert can't be put in ext4_ext_find_extent()
4168 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4169 EXT4_ERROR_INODE(inode, "bad extent address "
4170 "lblock: %lu, depth: %d pblock %lld",
4171 (unsigned long) map->m_lblk, depth,
4172 path[depth].p_block);
4177 ex = path[depth].p_ext;
4179 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4180 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4181 unsigned short ee_len;
4184 * Uninitialized extents are treated as holes, except that
4185 * we split out initialized portions during a write.
4187 ee_len = ext4_ext_get_actual_len(ex);
4189 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4191 /* if found extent covers block, simply return it */
4192 if (in_range(map->m_lblk, ee_block, ee_len)) {
4193 newblock = map->m_lblk - ee_block + ee_start;
4194 /* number of remaining blocks in the extent */
4195 allocated = ee_len - (map->m_lblk - ee_block);
4196 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4197 ee_block, ee_len, newblock);
4199 if (!ext4_ext_is_uninitialized(ex))
4202 allocated = ext4_ext_handle_uninitialized_extents(
4203 handle, inode, map, path, flags,
4204 allocated, newblock);
4209 if ((sbi->s_cluster_ratio > 1) &&
4210 ext4_find_delalloc_cluster(inode, map->m_lblk))
4211 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4214 * requested block isn't allocated yet;
4215 * we couldn't try to create block if create flag is zero
4217 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4219 * put just found gap into cache to speed up
4220 * subsequent requests
4222 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4223 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4228 * Okay, we need to do block allocation.
4230 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4231 newex.ee_block = cpu_to_le32(map->m_lblk);
4232 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4235 * If we are doing bigalloc, check to see if the extent returned
4236 * by ext4_ext_find_extent() implies a cluster we can use.
4238 if (cluster_offset && ex &&
4239 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4240 ar.len = allocated = map->m_len;
4241 newblock = map->m_pblk;
4242 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4243 goto got_allocated_blocks;
4246 /* find neighbour allocated blocks */
4247 ar.lleft = map->m_lblk;
4248 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4251 ar.lright = map->m_lblk;
4253 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4257 /* Check if the extent after searching to the right implies a
4258 * cluster we can use. */
4259 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4260 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4261 ar.len = allocated = map->m_len;
4262 newblock = map->m_pblk;
4263 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4264 goto got_allocated_blocks;
4268 * See if request is beyond maximum number of blocks we can have in
4269 * a single extent. For an initialized extent this limit is
4270 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4271 * EXT_UNINIT_MAX_LEN.
4273 if (map->m_len > EXT_INIT_MAX_LEN &&
4274 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4275 map->m_len = EXT_INIT_MAX_LEN;
4276 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4277 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4278 map->m_len = EXT_UNINIT_MAX_LEN;
4280 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4281 newex.ee_len = cpu_to_le16(map->m_len);
4282 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4284 allocated = ext4_ext_get_actual_len(&newex);
4286 allocated = map->m_len;
4288 /* allocate new block */
4290 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4291 ar.logical = map->m_lblk;
4293 * We calculate the offset from the beginning of the cluster
4294 * for the logical block number, since when we allocate a
4295 * physical cluster, the physical block should start at the
4296 * same offset from the beginning of the cluster. This is
4297 * needed so that future calls to get_implied_cluster_alloc()
4300 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4301 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4303 ar.logical -= offset;
4304 if (S_ISREG(inode->i_mode))
4305 ar.flags = EXT4_MB_HINT_DATA;
4307 /* disable in-core preallocation for non-regular files */
4309 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4310 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4311 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4314 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4315 ar.goal, newblock, allocated);
4317 allocated_clusters = ar.len;
4318 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4319 if (ar.len > allocated)
4322 got_allocated_blocks:
4323 /* try to insert new extent into found leaf and return */
4324 ext4_ext_store_pblock(&newex, newblock + offset);
4325 newex.ee_len = cpu_to_le16(ar.len);
4326 /* Mark uninitialized */
4327 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4328 ext4_ext_mark_uninitialized(&newex);
4329 map->m_flags |= EXT4_MAP_UNWRITTEN;
4331 * io_end structure was created for every IO write to an
4332 * uninitialized extent. To avoid unnecessary conversion,
4333 * here we flag the IO that really needs the conversion.
4334 * For non asycn direct IO case, flag the inode state
4335 * that we need to perform conversion when IO is done.
4337 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4339 if (ext4_should_dioread_nolock(inode))
4340 map->m_flags |= EXT4_MAP_UNINIT;
4344 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4345 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4348 err = ext4_ext_insert_extent(handle, inode, path,
4351 if (!err && set_unwritten) {
4353 ext4_set_io_unwritten_flag(inode, io);
4355 ext4_set_inode_state(inode,
4356 EXT4_STATE_DIO_UNWRITTEN);
4359 if (err && free_on_err) {
4360 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4361 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4362 /* free data blocks we just allocated */
4363 /* not a good idea to call discard here directly,
4364 * but otherwise we'd need to call it every free() */
4365 ext4_discard_preallocations(inode);
4366 ext4_free_blocks(handle, inode, NULL, newblock,
4367 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4371 /* previous routine could use block we allocated */
4372 newblock = ext4_ext_pblock(&newex);
4373 allocated = ext4_ext_get_actual_len(&newex);
4374 if (allocated > map->m_len)
4375 allocated = map->m_len;
4376 map->m_flags |= EXT4_MAP_NEW;
4379 * Update reserved blocks/metadata blocks after successful
4380 * block allocation which had been deferred till now.
4382 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4383 unsigned int reserved_clusters;
4385 * Check how many clusters we had reserved this allocated range
4387 reserved_clusters = get_reserved_cluster_alloc(inode,
4388 map->m_lblk, allocated);
4389 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4390 if (reserved_clusters) {
4392 * We have clusters reserved for this range.
4393 * But since we are not doing actual allocation
4394 * and are simply using blocks from previously
4395 * allocated cluster, we should release the
4396 * reservation and not claim quota.
4398 ext4_da_update_reserve_space(inode,
4399 reserved_clusters, 0);
4402 BUG_ON(allocated_clusters < reserved_clusters);
4403 if (reserved_clusters < allocated_clusters) {
4404 struct ext4_inode_info *ei = EXT4_I(inode);
4405 int reservation = allocated_clusters -
4408 * It seems we claimed few clusters outside of
4409 * the range of this allocation. We should give
4410 * it back to the reservation pool. This can
4411 * happen in the following case:
4413 * * Suppose s_cluster_ratio is 4 (i.e., each
4414 * cluster has 4 blocks. Thus, the clusters
4415 * are [0-3],[4-7],[8-11]...
4416 * * First comes delayed allocation write for
4417 * logical blocks 10 & 11. Since there were no
4418 * previous delayed allocated blocks in the
4419 * range [8-11], we would reserve 1 cluster
4421 * * Next comes write for logical blocks 3 to 8.
4422 * In this case, we will reserve 2 clusters
4423 * (for [0-3] and [4-7]; and not for [8-11] as
4424 * that range has a delayed allocated blocks.
4425 * Thus total reserved clusters now becomes 3.
4426 * * Now, during the delayed allocation writeout
4427 * time, we will first write blocks [3-8] and
4428 * allocate 3 clusters for writing these
4429 * blocks. Also, we would claim all these
4430 * three clusters above.
4431 * * Now when we come here to writeout the
4432 * blocks [10-11], we would expect to claim
4433 * the reservation of 1 cluster we had made
4434 * (and we would claim it since there are no
4435 * more delayed allocated blocks in the range
4436 * [8-11]. But our reserved cluster count had
4437 * already gone to 0.
4439 * Thus, at the step 4 above when we determine
4440 * that there are still some unwritten delayed
4441 * allocated blocks outside of our current
4442 * block range, we should increment the
4443 * reserved clusters count so that when the
4444 * remaining blocks finally gets written, we
4447 dquot_reserve_block(inode,
4448 EXT4_C2B(sbi, reservation));
4449 spin_lock(&ei->i_block_reservation_lock);
4450 ei->i_reserved_data_blocks += reservation;
4451 spin_unlock(&ei->i_block_reservation_lock);
4454 * We will claim quota for all newly allocated blocks.
4455 * We're updating the reserved space *after* the
4456 * correction above so we do not accidentally free
4457 * all the metadata reservation because we might
4458 * actually need it later on.
4460 ext4_da_update_reserve_space(inode, allocated_clusters,
4466 * Cache the extent and update transaction to commit on fdatasync only
4467 * when it is _not_ an uninitialized extent.
4469 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
4470 ext4_update_inode_fsync_trans(handle, inode, 1);
4472 ext4_update_inode_fsync_trans(handle, inode, 0);
4474 if (allocated > map->m_len)
4475 allocated = map->m_len;
4476 ext4_ext_show_leaf(inode, path);
4477 map->m_flags |= EXT4_MAP_MAPPED;
4478 map->m_pblk = newblock;
4479 map->m_len = allocated;
4482 ext4_ext_drop_refs(path);
4487 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4488 err ? err : allocated);
4489 ext4_es_lru_add(inode);
4490 return err ? err : allocated;
4493 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4495 struct super_block *sb = inode->i_sb;
4496 ext4_lblk_t last_block;
4500 * TODO: optimization is possible here.
4501 * Probably we need not scan at all,
4502 * because page truncation is enough.
4505 /* we have to know where to truncate from in crash case */
4506 EXT4_I(inode)->i_disksize = inode->i_size;
4507 ext4_mark_inode_dirty(handle, inode);
4509 last_block = (inode->i_size + sb->s_blocksize - 1)
4510 >> EXT4_BLOCK_SIZE_BITS(sb);
4512 err = ext4_es_remove_extent(inode, last_block,
4513 EXT_MAX_BLOCKS - last_block);
4514 if (err == -ENOMEM) {
4516 congestion_wait(BLK_RW_ASYNC, HZ/50);
4520 ext4_std_error(inode->i_sb, err);
4523 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4524 ext4_std_error(inode->i_sb, err);
4527 static void ext4_falloc_update_inode(struct inode *inode,
4528 int mode, loff_t new_size, int update_ctime)
4530 struct timespec now;
4533 now = current_fs_time(inode->i_sb);
4534 if (!timespec_equal(&inode->i_ctime, &now))
4535 inode->i_ctime = now;
4538 * Update only when preallocation was requested beyond
4541 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4542 if (new_size > i_size_read(inode))
4543 i_size_write(inode, new_size);
4544 if (new_size > EXT4_I(inode)->i_disksize)
4545 ext4_update_i_disksize(inode, new_size);
4548 * Mark that we allocate beyond EOF so the subsequent truncate
4549 * can proceed even if the new size is the same as i_size.
4551 if (new_size > i_size_read(inode))
4552 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4558 * preallocate space for a file. This implements ext4's fallocate file
4559 * operation, which gets called from sys_fallocate system call.
4560 * For block-mapped files, posix_fallocate should fall back to the method
4561 * of writing zeroes to the required new blocks (the same behavior which is
4562 * expected for file systems which do not support fallocate() system call).
4564 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4566 struct inode *inode = file_inode(file);
4569 unsigned int max_blocks;
4574 struct ext4_map_blocks map;
4575 unsigned int credits, blkbits = inode->i_blkbits;
4577 /* Return error if mode is not supported */
4578 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4581 if (mode & FALLOC_FL_PUNCH_HOLE)
4582 return ext4_punch_hole(inode, offset, len);
4584 ret = ext4_convert_inline_data(inode);
4589 * currently supporting (pre)allocate mode for extent-based
4592 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4595 trace_ext4_fallocate_enter(inode, offset, len, mode);
4596 map.m_lblk = offset >> blkbits;
4598 * We can't just convert len to max_blocks because
4599 * If blocksize = 4096 offset = 3072 and len = 2048
4601 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4604 * credits to insert 1 extent into extent tree
4606 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4607 mutex_lock(&inode->i_mutex);
4608 ret = inode_newsize_ok(inode, (len + offset));
4610 mutex_unlock(&inode->i_mutex);
4611 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4614 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4615 if (mode & FALLOC_FL_KEEP_SIZE)
4616 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4618 * Don't normalize the request if it can fit in one extent so
4619 * that it doesn't get unnecessarily split into multiple
4622 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4623 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4626 while (ret >= 0 && ret < max_blocks) {
4627 map.m_lblk = map.m_lblk + ret;
4628 map.m_len = max_blocks = max_blocks - ret;
4629 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4631 if (IS_ERR(handle)) {
4632 ret = PTR_ERR(handle);
4635 ret = ext4_map_blocks(handle, inode, &map, flags);
4638 ext4_warning(inode->i_sb,
4639 "inode #%lu: block %u: len %u: "
4640 "ext4_ext_map_blocks returned %d",
4641 inode->i_ino, map.m_lblk,
4644 ext4_mark_inode_dirty(handle, inode);
4645 ret2 = ext4_journal_stop(handle);
4648 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4649 blkbits) >> blkbits))
4650 new_size = offset + len;
4652 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4654 ext4_falloc_update_inode(inode, mode, new_size,
4655 (map.m_flags & EXT4_MAP_NEW));
4656 ext4_mark_inode_dirty(handle, inode);
4657 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4658 ext4_handle_sync(handle);
4659 ret2 = ext4_journal_stop(handle);
4663 if (ret == -ENOSPC &&
4664 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4668 mutex_unlock(&inode->i_mutex);
4669 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4670 ret > 0 ? ret2 : ret);
4671 return ret > 0 ? ret2 : ret;
4675 * This function convert a range of blocks to written extents
4676 * The caller of this function will pass the start offset and the size.
4677 * all unwritten extents within this range will be converted to
4680 * This function is called from the direct IO end io call back
4681 * function, to convert the fallocated extents after IO is completed.
4682 * Returns 0 on success.
4684 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4685 loff_t offset, ssize_t len)
4687 unsigned int max_blocks;
4690 struct ext4_map_blocks map;
4691 unsigned int credits, blkbits = inode->i_blkbits;
4693 map.m_lblk = offset >> blkbits;
4695 * We can't just convert len to max_blocks because
4696 * If blocksize = 4096 offset = 3072 and len = 2048
4698 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4701 * This is somewhat ugly but the idea is clear: When transaction is
4702 * reserved, everything goes into it. Otherwise we rather start several
4703 * smaller transactions for conversion of each extent separately.
4706 handle = ext4_journal_start_reserved(handle,
4707 EXT4_HT_EXT_CONVERT);
4709 return PTR_ERR(handle);
4713 * credits to insert 1 extent into extent tree
4715 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4717 while (ret >= 0 && ret < max_blocks) {
4719 map.m_len = (max_blocks -= ret);
4721 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4723 if (IS_ERR(handle)) {
4724 ret = PTR_ERR(handle);
4728 ret = ext4_map_blocks(handle, inode, &map,
4729 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4731 ext4_warning(inode->i_sb,
4732 "inode #%lu: block %u: len %u: "
4733 "ext4_ext_map_blocks returned %d",
4734 inode->i_ino, map.m_lblk,
4736 ext4_mark_inode_dirty(handle, inode);
4738 ret2 = ext4_journal_stop(handle);
4739 if (ret <= 0 || ret2)
4743 ret2 = ext4_journal_stop(handle);
4744 return ret > 0 ? ret2 : ret;
4748 * If newes is not existing extent (newes->ec_pblk equals zero) find
4749 * delayed extent at start of newes and update newes accordingly and
4750 * return start of the next delayed extent.
4752 * If newes is existing extent (newes->ec_pblk is not equal zero)
4753 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4754 * extent found. Leave newes unmodified.
4756 static int ext4_find_delayed_extent(struct inode *inode,
4757 struct extent_status *newes)
4759 struct extent_status es;
4760 ext4_lblk_t block, next_del;
4762 if (newes->es_pblk == 0) {
4763 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4764 newes->es_lblk + newes->es_len - 1, &es);
4767 * No extent in extent-tree contains block @newes->es_pblk,
4768 * then the block may stay in 1)a hole or 2)delayed-extent.
4774 if (es.es_lblk > newes->es_lblk) {
4776 newes->es_len = min(es.es_lblk - newes->es_lblk,
4781 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
4784 block = newes->es_lblk + newes->es_len;
4785 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
4787 next_del = EXT_MAX_BLOCKS;
4789 next_del = es.es_lblk;
4793 /* fiemap flags we can handle specified here */
4794 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4796 static int ext4_xattr_fiemap(struct inode *inode,
4797 struct fiemap_extent_info *fieinfo)
4801 __u32 flags = FIEMAP_EXTENT_LAST;
4802 int blockbits = inode->i_sb->s_blocksize_bits;
4806 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4807 struct ext4_iloc iloc;
4808 int offset; /* offset of xattr in inode */
4810 error = ext4_get_inode_loc(inode, &iloc);
4813 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4814 offset = EXT4_GOOD_OLD_INODE_SIZE +
4815 EXT4_I(inode)->i_extra_isize;
4817 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4818 flags |= FIEMAP_EXTENT_DATA_INLINE;
4820 } else { /* external block */
4821 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4822 length = inode->i_sb->s_blocksize;
4826 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4828 return (error < 0 ? error : 0);
4831 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4832 __u64 start, __u64 len)
4834 ext4_lblk_t start_blk;
4837 if (ext4_has_inline_data(inode)) {
4840 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4846 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4847 error = ext4_ext_precache(inode);
4852 /* fallback to generic here if not in extents fmt */
4853 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4854 return generic_block_fiemap(inode, fieinfo, start, len,
4857 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4860 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4861 error = ext4_xattr_fiemap(inode, fieinfo);
4863 ext4_lblk_t len_blks;
4866 start_blk = start >> inode->i_sb->s_blocksize_bits;
4867 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4868 if (last_blk >= EXT_MAX_BLOCKS)
4869 last_blk = EXT_MAX_BLOCKS-1;
4870 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4873 * Walk the extent tree gathering extent information
4874 * and pushing extents back to the user.
4876 error = ext4_fill_fiemap_extents(inode, start_blk,
4879 ext4_es_lru_add(inode);