2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
71 static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
74 struct ext4_extent_tail *et;
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
86 static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
89 struct ext4_extent_tail *et;
91 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
95 et = find_ext4_extent_tail(eh);
96 et->et_checksum = ext4_extent_block_csum(inode, eh);
99 static int ext4_split_extent(handle_t *handle,
101 struct ext4_ext_path *path,
102 struct ext4_map_blocks *map,
106 static int ext4_split_extent_at(handle_t *handle,
108 struct ext4_ext_path *path,
113 static int ext4_find_delayed_extent(struct inode *inode,
114 struct extent_status *newes);
116 static int ext4_ext_truncate_extend_restart(handle_t *handle,
122 if (!ext4_handle_valid(handle))
124 if (handle->h_buffer_credits > needed)
126 err = ext4_journal_extend(handle, needed);
129 err = ext4_truncate_restart_trans(handle, inode, needed);
141 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
142 struct ext4_ext_path *path)
145 /* path points to block */
146 BUFFER_TRACE(path->p_bh, "get_write_access");
147 return ext4_journal_get_write_access(handle, path->p_bh);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
165 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
167 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
168 /* path points to block */
169 err = __ext4_handle_dirty_metadata(where, line, handle,
172 /* path points to leaf/index in inode body */
173 err = ext4_mark_inode_dirty(handle, inode);
178 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
179 struct ext4_ext_path *path,
183 int depth = path->p_depth;
184 struct ext4_extent *ex;
187 * Try to predict block placement assuming that we are
188 * filling in a file which will eventually be
189 * non-sparse --- i.e., in the case of libbfd writing
190 * an ELF object sections out-of-order but in a way
191 * the eventually results in a contiguous object or
192 * executable file, or some database extending a table
193 * space file. However, this is actually somewhat
194 * non-ideal if we are writing a sparse file such as
195 * qemu or KVM writing a raw image file that is going
196 * to stay fairly sparse, since it will end up
197 * fragmenting the file system's free space. Maybe we
198 * should have some hueristics or some way to allow
199 * userspace to pass a hint to file system,
200 * especially if the latter case turns out to be
203 ex = path[depth].p_ext;
205 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
206 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
208 if (block > ext_block)
209 return ext_pblk + (block - ext_block);
211 return ext_pblk - (ext_block - block);
214 /* it looks like index is empty;
215 * try to find starting block from index itself */
216 if (path[depth].p_bh)
217 return path[depth].p_bh->b_blocknr;
220 /* OK. use inode's group */
221 return ext4_inode_to_goal_block(inode);
225 * Allocation for a meta data block
228 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
229 struct ext4_ext_path *path,
230 struct ext4_extent *ex, int *err, unsigned int flags)
232 ext4_fsblk_t goal, newblock;
234 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
235 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
240 static inline int ext4_ext_space_block(struct inode *inode, int check)
244 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
245 / sizeof(struct ext4_extent);
246 #ifdef AGGRESSIVE_TEST
247 if (!check && size > 6)
253 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
257 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
258 / sizeof(struct ext4_extent_idx);
259 #ifdef AGGRESSIVE_TEST
260 if (!check && size > 5)
266 static inline int ext4_ext_space_root(struct inode *inode, int check)
270 size = sizeof(EXT4_I(inode)->i_data);
271 size -= sizeof(struct ext4_extent_header);
272 size /= sizeof(struct ext4_extent);
273 #ifdef AGGRESSIVE_TEST
274 if (!check && size > 3)
280 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
284 size = sizeof(EXT4_I(inode)->i_data);
285 size -= sizeof(struct ext4_extent_header);
286 size /= sizeof(struct ext4_extent_idx);
287 #ifdef AGGRESSIVE_TEST
288 if (!check && size > 4)
295 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
296 struct ext4_ext_path *path, ext4_lblk_t lblk,
299 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
301 return ext4_split_extent_at(handle, inode, path, lblk, unwritten ?
302 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
303 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
304 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
308 * Calculate the number of metadata blocks needed
309 * to allocate @blocks
310 * Worse case is one block per extent
312 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
314 struct ext4_inode_info *ei = EXT4_I(inode);
317 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
318 / sizeof(struct ext4_extent_idx));
321 * If the new delayed allocation block is contiguous with the
322 * previous da block, it can share index blocks with the
323 * previous block, so we only need to allocate a new index
324 * block every idxs leaf blocks. At ldxs**2 blocks, we need
325 * an additional index block, and at ldxs**3 blocks, yet
326 * another index blocks.
328 if (ei->i_da_metadata_calc_len &&
329 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
332 if ((ei->i_da_metadata_calc_len % idxs) == 0)
334 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
336 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
338 ei->i_da_metadata_calc_len = 0;
340 ei->i_da_metadata_calc_len++;
341 ei->i_da_metadata_calc_last_lblock++;
346 * In the worst case we need a new set of index blocks at
347 * every level of the inode's extent tree.
349 ei->i_da_metadata_calc_len = 1;
350 ei->i_da_metadata_calc_last_lblock = lblock;
351 return ext_depth(inode) + 1;
355 ext4_ext_max_entries(struct inode *inode, int depth)
359 if (depth == ext_depth(inode)) {
361 max = ext4_ext_space_root(inode, 1);
363 max = ext4_ext_space_root_idx(inode, 1);
366 max = ext4_ext_space_block(inode, 1);
368 max = ext4_ext_space_block_idx(inode, 1);
374 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
376 ext4_fsblk_t block = ext4_ext_pblock(ext);
377 int len = ext4_ext_get_actual_len(ext);
378 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
379 ext4_lblk_t last = lblock + len - 1;
383 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
386 static int ext4_valid_extent_idx(struct inode *inode,
387 struct ext4_extent_idx *ext_idx)
389 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
391 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
394 static int ext4_valid_extent_entries(struct inode *inode,
395 struct ext4_extent_header *eh,
398 unsigned short entries;
399 if (eh->eh_entries == 0)
402 entries = le16_to_cpu(eh->eh_entries);
406 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
407 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
408 ext4_fsblk_t pblock = 0;
409 ext4_lblk_t lblock = 0;
410 ext4_lblk_t prev = 0;
413 if (!ext4_valid_extent(inode, ext))
416 /* Check for overlapping extents */
417 lblock = le32_to_cpu(ext->ee_block);
418 len = ext4_ext_get_actual_len(ext);
419 if ((lblock <= prev) && prev) {
420 pblock = ext4_ext_pblock(ext);
421 es->s_last_error_block = cpu_to_le64(pblock);
426 prev = lblock + len - 1;
429 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
431 if (!ext4_valid_extent_idx(inode, ext_idx))
440 static int __ext4_ext_check(const char *function, unsigned int line,
441 struct inode *inode, struct ext4_extent_header *eh,
442 int depth, ext4_fsblk_t pblk)
444 const char *error_msg;
447 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
448 error_msg = "invalid magic";
451 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
452 error_msg = "unexpected eh_depth";
455 if (unlikely(eh->eh_max == 0)) {
456 error_msg = "invalid eh_max";
459 max = ext4_ext_max_entries(inode, depth);
460 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
461 error_msg = "too large eh_max";
464 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
465 error_msg = "invalid eh_entries";
468 if (!ext4_valid_extent_entries(inode, eh, depth)) {
469 error_msg = "invalid extent entries";
472 /* Verify checksum on non-root extent tree nodes */
473 if (ext_depth(inode) != depth &&
474 !ext4_extent_block_csum_verify(inode, eh)) {
475 error_msg = "extent tree corrupted";
481 ext4_error_inode(inode, function, line, 0,
482 "pblk %llu bad header/extent: %s - magic %x, "
483 "entries %u, max %u(%u), depth %u(%u)",
484 (unsigned long long) pblk, error_msg,
485 le16_to_cpu(eh->eh_magic),
486 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
487 max, le16_to_cpu(eh->eh_depth), depth);
491 #define ext4_ext_check(inode, eh, depth, pblk) \
492 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
494 int ext4_ext_check_inode(struct inode *inode)
496 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
499 static struct buffer_head *
500 __read_extent_tree_block(const char *function, unsigned int line,
501 struct inode *inode, ext4_fsblk_t pblk, int depth,
504 struct buffer_head *bh;
507 bh = sb_getblk(inode->i_sb, pblk);
509 return ERR_PTR(-ENOMEM);
511 if (!bh_uptodate_or_lock(bh)) {
512 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
513 err = bh_submit_read(bh);
517 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
519 err = __ext4_ext_check(function, line, inode,
520 ext_block_hdr(bh), depth, pblk);
523 set_buffer_verified(bh);
525 * If this is a leaf block, cache all of its entries
527 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
528 struct ext4_extent_header *eh = ext_block_hdr(bh);
529 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
530 ext4_lblk_t prev = 0;
533 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
534 unsigned int status = EXTENT_STATUS_WRITTEN;
535 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
536 int len = ext4_ext_get_actual_len(ex);
538 if (prev && (prev != lblk))
539 ext4_es_cache_extent(inode, prev,
543 if (ext4_ext_is_unwritten(ex))
544 status = EXTENT_STATUS_UNWRITTEN;
545 ext4_es_cache_extent(inode, lblk, len,
546 ext4_ext_pblock(ex), status);
557 #define read_extent_tree_block(inode, pblk, depth, flags) \
558 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
562 * This function is called to cache a file's extent information in the
565 int ext4_ext_precache(struct inode *inode)
567 struct ext4_inode_info *ei = EXT4_I(inode);
568 struct ext4_ext_path *path = NULL;
569 struct buffer_head *bh;
570 int i = 0, depth, ret = 0;
572 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
573 return 0; /* not an extent-mapped inode */
575 down_read(&ei->i_data_sem);
576 depth = ext_depth(inode);
578 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
581 up_read(&ei->i_data_sem);
585 /* Don't cache anything if there are no external extent blocks */
588 path[0].p_hdr = ext_inode_hdr(inode);
589 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
592 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
595 * If this is a leaf block or we've reached the end of
596 * the index block, go up
599 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
600 brelse(path[i].p_bh);
605 bh = read_extent_tree_block(inode,
606 ext4_idx_pblock(path[i].p_idx++),
608 EXT4_EX_FORCE_CACHE);
615 path[i].p_hdr = ext_block_hdr(bh);
616 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
618 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
620 up_read(&ei->i_data_sem);
621 ext4_ext_drop_refs(path);
627 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
629 int k, l = path->p_depth;
632 for (k = 0; k <= l; k++, path++) {
634 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
635 ext4_idx_pblock(path->p_idx));
636 } else if (path->p_ext) {
637 ext_debug(" %d:[%d]%d:%llu ",
638 le32_to_cpu(path->p_ext->ee_block),
639 ext4_ext_is_unwritten(path->p_ext),
640 ext4_ext_get_actual_len(path->p_ext),
641 ext4_ext_pblock(path->p_ext));
648 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
650 int depth = ext_depth(inode);
651 struct ext4_extent_header *eh;
652 struct ext4_extent *ex;
658 eh = path[depth].p_hdr;
659 ex = EXT_FIRST_EXTENT(eh);
661 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
663 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
664 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
665 ext4_ext_is_unwritten(ex),
666 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
671 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
672 ext4_fsblk_t newblock, int level)
674 int depth = ext_depth(inode);
675 struct ext4_extent *ex;
677 if (depth != level) {
678 struct ext4_extent_idx *idx;
679 idx = path[level].p_idx;
680 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
681 ext_debug("%d: move %d:%llu in new index %llu\n", level,
682 le32_to_cpu(idx->ei_block),
683 ext4_idx_pblock(idx),
691 ex = path[depth].p_ext;
692 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
693 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
694 le32_to_cpu(ex->ee_block),
696 ext4_ext_is_unwritten(ex),
697 ext4_ext_get_actual_len(ex),
704 #define ext4_ext_show_path(inode, path)
705 #define ext4_ext_show_leaf(inode, path)
706 #define ext4_ext_show_move(inode, path, newblock, level)
709 void ext4_ext_drop_refs(struct ext4_ext_path *path)
711 int depth = path->p_depth;
714 for (i = 0; i <= depth; i++, path++)
722 * ext4_ext_binsearch_idx:
723 * binary search for the closest index of the given block
724 * the header must be checked before calling this
727 ext4_ext_binsearch_idx(struct inode *inode,
728 struct ext4_ext_path *path, ext4_lblk_t block)
730 struct ext4_extent_header *eh = path->p_hdr;
731 struct ext4_extent_idx *r, *l, *m;
734 ext_debug("binsearch for %u(idx): ", block);
736 l = EXT_FIRST_INDEX(eh) + 1;
737 r = EXT_LAST_INDEX(eh);
740 if (block < le32_to_cpu(m->ei_block))
744 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
745 m, le32_to_cpu(m->ei_block),
746 r, le32_to_cpu(r->ei_block));
750 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
751 ext4_idx_pblock(path->p_idx));
753 #ifdef CHECK_BINSEARCH
755 struct ext4_extent_idx *chix, *ix;
758 chix = ix = EXT_FIRST_INDEX(eh);
759 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
761 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
762 printk(KERN_DEBUG "k=%d, ix=0x%p, "
764 ix, EXT_FIRST_INDEX(eh));
765 printk(KERN_DEBUG "%u <= %u\n",
766 le32_to_cpu(ix->ei_block),
767 le32_to_cpu(ix[-1].ei_block));
769 BUG_ON(k && le32_to_cpu(ix->ei_block)
770 <= le32_to_cpu(ix[-1].ei_block));
771 if (block < le32_to_cpu(ix->ei_block))
775 BUG_ON(chix != path->p_idx);
782 * ext4_ext_binsearch:
783 * binary search for closest extent of the given block
784 * the header must be checked before calling this
787 ext4_ext_binsearch(struct inode *inode,
788 struct ext4_ext_path *path, ext4_lblk_t block)
790 struct ext4_extent_header *eh = path->p_hdr;
791 struct ext4_extent *r, *l, *m;
793 if (eh->eh_entries == 0) {
795 * this leaf is empty:
796 * we get such a leaf in split/add case
801 ext_debug("binsearch for %u: ", block);
803 l = EXT_FIRST_EXTENT(eh) + 1;
804 r = EXT_LAST_EXTENT(eh);
808 if (block < le32_to_cpu(m->ee_block))
812 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
813 m, le32_to_cpu(m->ee_block),
814 r, le32_to_cpu(r->ee_block));
818 ext_debug(" -> %d:%llu:[%d]%d ",
819 le32_to_cpu(path->p_ext->ee_block),
820 ext4_ext_pblock(path->p_ext),
821 ext4_ext_is_unwritten(path->p_ext),
822 ext4_ext_get_actual_len(path->p_ext));
824 #ifdef CHECK_BINSEARCH
826 struct ext4_extent *chex, *ex;
829 chex = ex = EXT_FIRST_EXTENT(eh);
830 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
831 BUG_ON(k && le32_to_cpu(ex->ee_block)
832 <= le32_to_cpu(ex[-1].ee_block));
833 if (block < le32_to_cpu(ex->ee_block))
837 BUG_ON(chex != path->p_ext);
843 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
845 struct ext4_extent_header *eh;
847 eh = ext_inode_hdr(inode);
850 eh->eh_magic = EXT4_EXT_MAGIC;
851 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
852 ext4_mark_inode_dirty(handle, inode);
856 struct ext4_ext_path *
857 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
858 struct ext4_ext_path *path, int flags)
860 struct ext4_extent_header *eh;
861 struct buffer_head *bh;
862 short int depth, i, ppos = 0, alloc = 0;
865 eh = ext_inode_hdr(inode);
866 depth = ext_depth(inode);
868 /* account possible depth increase */
870 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
873 return ERR_PTR(-ENOMEM);
880 /* walk through the tree */
882 ext_debug("depth %d: num %d, max %d\n",
883 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
885 ext4_ext_binsearch_idx(inode, path + ppos, block);
886 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
887 path[ppos].p_depth = i;
888 path[ppos].p_ext = NULL;
890 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
892 if (unlikely(IS_ERR(bh))) {
897 eh = ext_block_hdr(bh);
899 if (unlikely(ppos > depth)) {
901 EXT4_ERROR_INODE(inode,
902 "ppos %d > depth %d", ppos, depth);
906 path[ppos].p_bh = bh;
907 path[ppos].p_hdr = eh;
910 path[ppos].p_depth = i;
911 path[ppos].p_ext = NULL;
912 path[ppos].p_idx = NULL;
915 ext4_ext_binsearch(inode, path + ppos, block);
916 /* if not an empty leaf */
917 if (path[ppos].p_ext)
918 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
920 ext4_ext_show_path(inode, path);
925 ext4_ext_drop_refs(path);
932 * ext4_ext_insert_index:
933 * insert new index [@logical;@ptr] into the block at @curp;
934 * check where to insert: before @curp or after @curp
936 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
937 struct ext4_ext_path *curp,
938 int logical, ext4_fsblk_t ptr)
940 struct ext4_extent_idx *ix;
943 err = ext4_ext_get_access(handle, inode, curp);
947 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
948 EXT4_ERROR_INODE(inode,
949 "logical %d == ei_block %d!",
950 logical, le32_to_cpu(curp->p_idx->ei_block));
954 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
955 >= le16_to_cpu(curp->p_hdr->eh_max))) {
956 EXT4_ERROR_INODE(inode,
957 "eh_entries %d >= eh_max %d!",
958 le16_to_cpu(curp->p_hdr->eh_entries),
959 le16_to_cpu(curp->p_hdr->eh_max));
963 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
965 ext_debug("insert new index %d after: %llu\n", logical, ptr);
966 ix = curp->p_idx + 1;
969 ext_debug("insert new index %d before: %llu\n", logical, ptr);
973 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
976 ext_debug("insert new index %d: "
977 "move %d indices from 0x%p to 0x%p\n",
978 logical, len, ix, ix + 1);
979 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
982 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
983 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
987 ix->ei_block = cpu_to_le32(logical);
988 ext4_idx_store_pblock(ix, ptr);
989 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
991 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
992 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
996 err = ext4_ext_dirty(handle, inode, curp);
997 ext4_std_error(inode->i_sb, err);
1004 * inserts new subtree into the path, using free index entry
1006 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1007 * - makes decision where to split
1008 * - moves remaining extents and index entries (right to the split point)
1009 * into the newly allocated blocks
1010 * - initializes subtree
1012 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1014 struct ext4_ext_path *path,
1015 struct ext4_extent *newext, int at)
1017 struct buffer_head *bh = NULL;
1018 int depth = ext_depth(inode);
1019 struct ext4_extent_header *neh;
1020 struct ext4_extent_idx *fidx;
1021 int i = at, k, m, a;
1022 ext4_fsblk_t newblock, oldblock;
1024 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1027 /* make decision: where to split? */
1028 /* FIXME: now decision is simplest: at current extent */
1030 /* if current leaf will be split, then we should use
1031 * border from split point */
1032 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1033 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1036 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1037 border = path[depth].p_ext[1].ee_block;
1038 ext_debug("leaf will be split."
1039 " next leaf starts at %d\n",
1040 le32_to_cpu(border));
1042 border = newext->ee_block;
1043 ext_debug("leaf will be added."
1044 " next leaf starts at %d\n",
1045 le32_to_cpu(border));
1049 * If error occurs, then we break processing
1050 * and mark filesystem read-only. index won't
1051 * be inserted and tree will be in consistent
1052 * state. Next mount will repair buffers too.
1056 * Get array to track all allocated blocks.
1057 * We need this to handle errors and free blocks
1060 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1064 /* allocate all needed blocks */
1065 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1066 for (a = 0; a < depth - at; a++) {
1067 newblock = ext4_ext_new_meta_block(handle, inode, path,
1068 newext, &err, flags);
1071 ablocks[a] = newblock;
1074 /* initialize new leaf */
1075 newblock = ablocks[--a];
1076 if (unlikely(newblock == 0)) {
1077 EXT4_ERROR_INODE(inode, "newblock == 0!");
1081 bh = sb_getblk(inode->i_sb, newblock);
1082 if (unlikely(!bh)) {
1088 err = ext4_journal_get_create_access(handle, bh);
1092 neh = ext_block_hdr(bh);
1093 neh->eh_entries = 0;
1094 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1095 neh->eh_magic = EXT4_EXT_MAGIC;
1098 /* move remainder of path[depth] to the new leaf */
1099 if (unlikely(path[depth].p_hdr->eh_entries !=
1100 path[depth].p_hdr->eh_max)) {
1101 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1102 path[depth].p_hdr->eh_entries,
1103 path[depth].p_hdr->eh_max);
1107 /* start copy from next extent */
1108 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1109 ext4_ext_show_move(inode, path, newblock, depth);
1111 struct ext4_extent *ex;
1112 ex = EXT_FIRST_EXTENT(neh);
1113 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1114 le16_add_cpu(&neh->eh_entries, m);
1117 ext4_extent_block_csum_set(inode, neh);
1118 set_buffer_uptodate(bh);
1121 err = ext4_handle_dirty_metadata(handle, inode, bh);
1127 /* correct old leaf */
1129 err = ext4_ext_get_access(handle, inode, path + depth);
1132 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1133 err = ext4_ext_dirty(handle, inode, path + depth);
1139 /* create intermediate indexes */
1141 if (unlikely(k < 0)) {
1142 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1147 ext_debug("create %d intermediate indices\n", k);
1148 /* insert new index into current index block */
1149 /* current depth stored in i var */
1152 oldblock = newblock;
1153 newblock = ablocks[--a];
1154 bh = sb_getblk(inode->i_sb, newblock);
1155 if (unlikely(!bh)) {
1161 err = ext4_journal_get_create_access(handle, bh);
1165 neh = ext_block_hdr(bh);
1166 neh->eh_entries = cpu_to_le16(1);
1167 neh->eh_magic = EXT4_EXT_MAGIC;
1168 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1169 neh->eh_depth = cpu_to_le16(depth - i);
1170 fidx = EXT_FIRST_INDEX(neh);
1171 fidx->ei_block = border;
1172 ext4_idx_store_pblock(fidx, oldblock);
1174 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1175 i, newblock, le32_to_cpu(border), oldblock);
1177 /* move remainder of path[i] to the new index block */
1178 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1179 EXT_LAST_INDEX(path[i].p_hdr))) {
1180 EXT4_ERROR_INODE(inode,
1181 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1182 le32_to_cpu(path[i].p_ext->ee_block));
1186 /* start copy indexes */
1187 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1188 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1189 EXT_MAX_INDEX(path[i].p_hdr));
1190 ext4_ext_show_move(inode, path, newblock, i);
1192 memmove(++fidx, path[i].p_idx,
1193 sizeof(struct ext4_extent_idx) * m);
1194 le16_add_cpu(&neh->eh_entries, m);
1196 ext4_extent_block_csum_set(inode, neh);
1197 set_buffer_uptodate(bh);
1200 err = ext4_handle_dirty_metadata(handle, inode, bh);
1206 /* correct old index */
1208 err = ext4_ext_get_access(handle, inode, path + i);
1211 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1212 err = ext4_ext_dirty(handle, inode, path + i);
1220 /* insert new index */
1221 err = ext4_ext_insert_index(handle, inode, path + at,
1222 le32_to_cpu(border), newblock);
1226 if (buffer_locked(bh))
1232 /* free all allocated blocks in error case */
1233 for (i = 0; i < depth; i++) {
1236 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1237 EXT4_FREE_BLOCKS_METADATA);
1246 * ext4_ext_grow_indepth:
1247 * implements tree growing procedure:
1248 * - allocates new block
1249 * - moves top-level data (index block or leaf) into the new block
1250 * - initializes new top-level, creating index that points to the
1251 * just created block
1253 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1255 struct ext4_extent *newext)
1257 struct ext4_extent_header *neh;
1258 struct buffer_head *bh;
1259 ext4_fsblk_t newblock;
1262 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1263 newext, &err, flags);
1267 bh = sb_getblk(inode->i_sb, newblock);
1272 err = ext4_journal_get_create_access(handle, bh);
1278 /* move top-level index/leaf into new block */
1279 memmove(bh->b_data, EXT4_I(inode)->i_data,
1280 sizeof(EXT4_I(inode)->i_data));
1282 /* set size of new block */
1283 neh = ext_block_hdr(bh);
1284 /* old root could have indexes or leaves
1285 * so calculate e_max right way */
1286 if (ext_depth(inode))
1287 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1289 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1290 neh->eh_magic = EXT4_EXT_MAGIC;
1291 ext4_extent_block_csum_set(inode, neh);
1292 set_buffer_uptodate(bh);
1295 err = ext4_handle_dirty_metadata(handle, inode, bh);
1299 /* Update top-level index: num,max,pointer */
1300 neh = ext_inode_hdr(inode);
1301 neh->eh_entries = cpu_to_le16(1);
1302 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1303 if (neh->eh_depth == 0) {
1304 /* Root extent block becomes index block */
1305 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1306 EXT_FIRST_INDEX(neh)->ei_block =
1307 EXT_FIRST_EXTENT(neh)->ee_block;
1309 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1310 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1311 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1312 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1314 le16_add_cpu(&neh->eh_depth, 1);
1315 ext4_mark_inode_dirty(handle, inode);
1323 * ext4_ext_create_new_leaf:
1324 * finds empty index and adds new leaf.
1325 * if no free index is found, then it requests in-depth growing.
1327 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1328 unsigned int mb_flags,
1329 unsigned int gb_flags,
1330 struct ext4_ext_path *path,
1331 struct ext4_extent *newext)
1333 struct ext4_ext_path *curp;
1334 int depth, i, err = 0;
1337 i = depth = ext_depth(inode);
1339 /* walk up to the tree and look for free index entry */
1340 curp = path + depth;
1341 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1346 /* we use already allocated block for index block,
1347 * so subsequent data blocks should be contiguous */
1348 if (EXT_HAS_FREE_INDEX(curp)) {
1349 /* if we found index with free entry, then use that
1350 * entry: create all needed subtree and add new leaf */
1351 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1356 ext4_ext_drop_refs(path);
1357 path = ext4_ext_find_extent(inode,
1358 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1361 err = PTR_ERR(path);
1363 /* tree is full, time to grow in depth */
1364 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1369 ext4_ext_drop_refs(path);
1370 path = ext4_ext_find_extent(inode,
1371 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1374 err = PTR_ERR(path);
1379 * only first (depth 0 -> 1) produces free space;
1380 * in all other cases we have to split the grown tree
1382 depth = ext_depth(inode);
1383 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1384 /* now we need to split */
1394 * search the closest allocated block to the left for *logical
1395 * and returns it at @logical + it's physical address at @phys
1396 * if *logical is the smallest allocated block, the function
1397 * returns 0 at @phys
1398 * return value contains 0 (success) or error code
1400 static int ext4_ext_search_left(struct inode *inode,
1401 struct ext4_ext_path *path,
1402 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1404 struct ext4_extent_idx *ix;
1405 struct ext4_extent *ex;
1408 if (unlikely(path == NULL)) {
1409 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1412 depth = path->p_depth;
1415 if (depth == 0 && path->p_ext == NULL)
1418 /* usually extent in the path covers blocks smaller
1419 * then *logical, but it can be that extent is the
1420 * first one in the file */
1422 ex = path[depth].p_ext;
1423 ee_len = ext4_ext_get_actual_len(ex);
1424 if (*logical < le32_to_cpu(ex->ee_block)) {
1425 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1426 EXT4_ERROR_INODE(inode,
1427 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1428 *logical, le32_to_cpu(ex->ee_block));
1431 while (--depth >= 0) {
1432 ix = path[depth].p_idx;
1433 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1434 EXT4_ERROR_INODE(inode,
1435 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1436 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1437 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1438 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1446 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1447 EXT4_ERROR_INODE(inode,
1448 "logical %d < ee_block %d + ee_len %d!",
1449 *logical, le32_to_cpu(ex->ee_block), ee_len);
1453 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1454 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1459 * search the closest allocated block to the right for *logical
1460 * and returns it at @logical + it's physical address at @phys
1461 * if *logical is the largest allocated block, the function
1462 * returns 0 at @phys
1463 * return value contains 0 (success) or error code
1465 static int ext4_ext_search_right(struct inode *inode,
1466 struct ext4_ext_path *path,
1467 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1468 struct ext4_extent **ret_ex)
1470 struct buffer_head *bh = NULL;
1471 struct ext4_extent_header *eh;
1472 struct ext4_extent_idx *ix;
1473 struct ext4_extent *ex;
1475 int depth; /* Note, NOT eh_depth; depth from top of tree */
1478 if (unlikely(path == NULL)) {
1479 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1482 depth = path->p_depth;
1485 if (depth == 0 && path->p_ext == NULL)
1488 /* usually extent in the path covers blocks smaller
1489 * then *logical, but it can be that extent is the
1490 * first one in the file */
1492 ex = path[depth].p_ext;
1493 ee_len = ext4_ext_get_actual_len(ex);
1494 if (*logical < le32_to_cpu(ex->ee_block)) {
1495 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1496 EXT4_ERROR_INODE(inode,
1497 "first_extent(path[%d].p_hdr) != ex",
1501 while (--depth >= 0) {
1502 ix = path[depth].p_idx;
1503 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1504 EXT4_ERROR_INODE(inode,
1505 "ix != EXT_FIRST_INDEX *logical %d!",
1513 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1514 EXT4_ERROR_INODE(inode,
1515 "logical %d < ee_block %d + ee_len %d!",
1516 *logical, le32_to_cpu(ex->ee_block), ee_len);
1520 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1521 /* next allocated block in this leaf */
1526 /* go up and search for index to the right */
1527 while (--depth >= 0) {
1528 ix = path[depth].p_idx;
1529 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1533 /* we've gone up to the root and found no index to the right */
1537 /* we've found index to the right, let's
1538 * follow it and find the closest allocated
1539 * block to the right */
1541 block = ext4_idx_pblock(ix);
1542 while (++depth < path->p_depth) {
1543 /* subtract from p_depth to get proper eh_depth */
1544 bh = read_extent_tree_block(inode, block,
1545 path->p_depth - depth, 0);
1548 eh = ext_block_hdr(bh);
1549 ix = EXT_FIRST_INDEX(eh);
1550 block = ext4_idx_pblock(ix);
1554 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1557 eh = ext_block_hdr(bh);
1558 ex = EXT_FIRST_EXTENT(eh);
1560 *logical = le32_to_cpu(ex->ee_block);
1561 *phys = ext4_ext_pblock(ex);
1569 * ext4_ext_next_allocated_block:
1570 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1571 * NOTE: it considers block number from index entry as
1572 * allocated block. Thus, index entries have to be consistent
1576 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1580 BUG_ON(path == NULL);
1581 depth = path->p_depth;
1583 if (depth == 0 && path->p_ext == NULL)
1584 return EXT_MAX_BLOCKS;
1586 while (depth >= 0) {
1587 if (depth == path->p_depth) {
1589 if (path[depth].p_ext &&
1590 path[depth].p_ext !=
1591 EXT_LAST_EXTENT(path[depth].p_hdr))
1592 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1595 if (path[depth].p_idx !=
1596 EXT_LAST_INDEX(path[depth].p_hdr))
1597 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1602 return EXT_MAX_BLOCKS;
1606 * ext4_ext_next_leaf_block:
1607 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1609 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1613 BUG_ON(path == NULL);
1614 depth = path->p_depth;
1616 /* zero-tree has no leaf blocks at all */
1618 return EXT_MAX_BLOCKS;
1620 /* go to index block */
1623 while (depth >= 0) {
1624 if (path[depth].p_idx !=
1625 EXT_LAST_INDEX(path[depth].p_hdr))
1626 return (ext4_lblk_t)
1627 le32_to_cpu(path[depth].p_idx[1].ei_block);
1631 return EXT_MAX_BLOCKS;
1635 * ext4_ext_correct_indexes:
1636 * if leaf gets modified and modified extent is first in the leaf,
1637 * then we have to correct all indexes above.
1638 * TODO: do we need to correct tree in all cases?
1640 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1641 struct ext4_ext_path *path)
1643 struct ext4_extent_header *eh;
1644 int depth = ext_depth(inode);
1645 struct ext4_extent *ex;
1649 eh = path[depth].p_hdr;
1650 ex = path[depth].p_ext;
1652 if (unlikely(ex == NULL || eh == NULL)) {
1653 EXT4_ERROR_INODE(inode,
1654 "ex %p == NULL or eh %p == NULL", ex, eh);
1659 /* there is no tree at all */
1663 if (ex != EXT_FIRST_EXTENT(eh)) {
1664 /* we correct tree if first leaf got modified only */
1669 * TODO: we need correction if border is smaller than current one
1672 border = path[depth].p_ext->ee_block;
1673 err = ext4_ext_get_access(handle, inode, path + k);
1676 path[k].p_idx->ei_block = border;
1677 err = ext4_ext_dirty(handle, inode, path + k);
1682 /* change all left-side indexes */
1683 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1685 err = ext4_ext_get_access(handle, inode, path + k);
1688 path[k].p_idx->ei_block = border;
1689 err = ext4_ext_dirty(handle, inode, path + k);
1698 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1699 struct ext4_extent *ex2)
1701 unsigned short ext1_ee_len, ext2_ee_len;
1704 * Make sure that both extents are initialized. We don't merge
1705 * unwritten extents so that we can be sure that end_io code has
1706 * the extent that was written properly split out and conversion to
1707 * initialized is trivial.
1709 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1712 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1713 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1715 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1716 le32_to_cpu(ex2->ee_block))
1720 * To allow future support for preallocated extents to be added
1721 * as an RO_COMPAT feature, refuse to merge to extents if
1722 * this can result in the top bit of ee_len being set.
1724 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1726 if (ext4_ext_is_unwritten(ex1) &&
1727 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1728 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1729 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1731 #ifdef AGGRESSIVE_TEST
1732 if (ext1_ee_len >= 4)
1736 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1742 * This function tries to merge the "ex" extent to the next extent in the tree.
1743 * It always tries to merge towards right. If you want to merge towards
1744 * left, pass "ex - 1" as argument instead of "ex".
1745 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1746 * 1 if they got merged.
1748 static int ext4_ext_try_to_merge_right(struct inode *inode,
1749 struct ext4_ext_path *path,
1750 struct ext4_extent *ex)
1752 struct ext4_extent_header *eh;
1753 unsigned int depth, len;
1754 int merge_done = 0, unwritten;
1756 depth = ext_depth(inode);
1757 BUG_ON(path[depth].p_hdr == NULL);
1758 eh = path[depth].p_hdr;
1760 while (ex < EXT_LAST_EXTENT(eh)) {
1761 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1763 /* merge with next extent! */
1764 unwritten = ext4_ext_is_unwritten(ex);
1765 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1766 + ext4_ext_get_actual_len(ex + 1));
1768 ext4_ext_mark_unwritten(ex);
1770 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1771 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1772 * sizeof(struct ext4_extent);
1773 memmove(ex + 1, ex + 2, len);
1775 le16_add_cpu(&eh->eh_entries, -1);
1777 WARN_ON(eh->eh_entries == 0);
1778 if (!eh->eh_entries)
1779 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1786 * This function does a very simple check to see if we can collapse
1787 * an extent tree with a single extent tree leaf block into the inode.
1789 static void ext4_ext_try_to_merge_up(handle_t *handle,
1790 struct inode *inode,
1791 struct ext4_ext_path *path)
1794 unsigned max_root = ext4_ext_space_root(inode, 0);
1797 if ((path[0].p_depth != 1) ||
1798 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1799 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1803 * We need to modify the block allocation bitmap and the block
1804 * group descriptor to release the extent tree block. If we
1805 * can't get the journal credits, give up.
1807 if (ext4_journal_extend(handle, 2))
1811 * Copy the extent data up to the inode
1813 blk = ext4_idx_pblock(path[0].p_idx);
1814 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1815 sizeof(struct ext4_extent_idx);
1816 s += sizeof(struct ext4_extent_header);
1818 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1819 path[0].p_depth = 0;
1820 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1821 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1822 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1824 brelse(path[1].p_bh);
1825 ext4_free_blocks(handle, inode, NULL, blk, 1,
1826 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1830 * This function tries to merge the @ex extent to neighbours in the tree.
1831 * return 1 if merge left else 0.
1833 static void ext4_ext_try_to_merge(handle_t *handle,
1834 struct inode *inode,
1835 struct ext4_ext_path *path,
1836 struct ext4_extent *ex) {
1837 struct ext4_extent_header *eh;
1841 depth = ext_depth(inode);
1842 BUG_ON(path[depth].p_hdr == NULL);
1843 eh = path[depth].p_hdr;
1845 if (ex > EXT_FIRST_EXTENT(eh))
1846 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1849 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1851 ext4_ext_try_to_merge_up(handle, inode, path);
1855 * check if a portion of the "newext" extent overlaps with an
1858 * If there is an overlap discovered, it updates the length of the newext
1859 * such that there will be no overlap, and then returns 1.
1860 * If there is no overlap found, it returns 0.
1862 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1863 struct inode *inode,
1864 struct ext4_extent *newext,
1865 struct ext4_ext_path *path)
1868 unsigned int depth, len1;
1869 unsigned int ret = 0;
1871 b1 = le32_to_cpu(newext->ee_block);
1872 len1 = ext4_ext_get_actual_len(newext);
1873 depth = ext_depth(inode);
1874 if (!path[depth].p_ext)
1876 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1879 * get the next allocated block if the extent in the path
1880 * is before the requested block(s)
1883 b2 = ext4_ext_next_allocated_block(path);
1884 if (b2 == EXT_MAX_BLOCKS)
1886 b2 = EXT4_LBLK_CMASK(sbi, b2);
1889 /* check for wrap through zero on extent logical start block*/
1890 if (b1 + len1 < b1) {
1891 len1 = EXT_MAX_BLOCKS - b1;
1892 newext->ee_len = cpu_to_le16(len1);
1896 /* check for overlap */
1897 if (b1 + len1 > b2) {
1898 newext->ee_len = cpu_to_le16(b2 - b1);
1906 * ext4_ext_insert_extent:
1907 * tries to merge requsted extent into the existing extent or
1908 * inserts requested extent as new one into the tree,
1909 * creating new leaf in the no-space case.
1911 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1912 struct ext4_ext_path *path,
1913 struct ext4_extent *newext, int gb_flags)
1915 struct ext4_extent_header *eh;
1916 struct ext4_extent *ex, *fex;
1917 struct ext4_extent *nearex; /* nearest extent */
1918 struct ext4_ext_path *npath = NULL;
1919 int depth, len, err;
1921 int mb_flags = 0, unwritten;
1923 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1924 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1927 depth = ext_depth(inode);
1928 ex = path[depth].p_ext;
1929 eh = path[depth].p_hdr;
1930 if (unlikely(path[depth].p_hdr == NULL)) {
1931 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1935 /* try to insert block into found extent and return */
1936 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1939 * Try to see whether we should rather test the extent on
1940 * right from ex, or from the left of ex. This is because
1941 * ext4_ext_find_extent() can return either extent on the
1942 * left, or on the right from the searched position. This
1943 * will make merging more effective.
1945 if (ex < EXT_LAST_EXTENT(eh) &&
1946 (le32_to_cpu(ex->ee_block) +
1947 ext4_ext_get_actual_len(ex) <
1948 le32_to_cpu(newext->ee_block))) {
1951 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1952 (le32_to_cpu(newext->ee_block) +
1953 ext4_ext_get_actual_len(newext) <
1954 le32_to_cpu(ex->ee_block)))
1957 /* Try to append newex to the ex */
1958 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1959 ext_debug("append [%d]%d block to %u:[%d]%d"
1961 ext4_ext_is_unwritten(newext),
1962 ext4_ext_get_actual_len(newext),
1963 le32_to_cpu(ex->ee_block),
1964 ext4_ext_is_unwritten(ex),
1965 ext4_ext_get_actual_len(ex),
1966 ext4_ext_pblock(ex));
1967 err = ext4_ext_get_access(handle, inode,
1971 unwritten = ext4_ext_is_unwritten(ex);
1972 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1973 + ext4_ext_get_actual_len(newext));
1975 ext4_ext_mark_unwritten(ex);
1976 eh = path[depth].p_hdr;
1982 /* Try to prepend newex to the ex */
1983 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1984 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1986 le32_to_cpu(newext->ee_block),
1987 ext4_ext_is_unwritten(newext),
1988 ext4_ext_get_actual_len(newext),
1989 le32_to_cpu(ex->ee_block),
1990 ext4_ext_is_unwritten(ex),
1991 ext4_ext_get_actual_len(ex),
1992 ext4_ext_pblock(ex));
1993 err = ext4_ext_get_access(handle, inode,
1998 unwritten = ext4_ext_is_unwritten(ex);
1999 ex->ee_block = newext->ee_block;
2000 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2001 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2002 + ext4_ext_get_actual_len(newext));
2004 ext4_ext_mark_unwritten(ex);
2005 eh = path[depth].p_hdr;
2011 depth = ext_depth(inode);
2012 eh = path[depth].p_hdr;
2013 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2016 /* probably next leaf has space for us? */
2017 fex = EXT_LAST_EXTENT(eh);
2018 next = EXT_MAX_BLOCKS;
2019 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2020 next = ext4_ext_next_leaf_block(path);
2021 if (next != EXT_MAX_BLOCKS) {
2022 ext_debug("next leaf block - %u\n", next);
2023 BUG_ON(npath != NULL);
2024 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2026 return PTR_ERR(npath);
2027 BUG_ON(npath->p_depth != path->p_depth);
2028 eh = npath[depth].p_hdr;
2029 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2030 ext_debug("next leaf isn't full(%d)\n",
2031 le16_to_cpu(eh->eh_entries));
2035 ext_debug("next leaf has no free space(%d,%d)\n",
2036 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2040 * There is no free space in the found leaf.
2041 * We're gonna add a new leaf in the tree.
2043 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2044 mb_flags = EXT4_MB_USE_RESERVED;
2045 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2049 depth = ext_depth(inode);
2050 eh = path[depth].p_hdr;
2053 nearex = path[depth].p_ext;
2055 err = ext4_ext_get_access(handle, inode, path + depth);
2060 /* there is no extent in this leaf, create first one */
2061 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2062 le32_to_cpu(newext->ee_block),
2063 ext4_ext_pblock(newext),
2064 ext4_ext_is_unwritten(newext),
2065 ext4_ext_get_actual_len(newext));
2066 nearex = EXT_FIRST_EXTENT(eh);
2068 if (le32_to_cpu(newext->ee_block)
2069 > le32_to_cpu(nearex->ee_block)) {
2071 ext_debug("insert %u:%llu:[%d]%d before: "
2073 le32_to_cpu(newext->ee_block),
2074 ext4_ext_pblock(newext),
2075 ext4_ext_is_unwritten(newext),
2076 ext4_ext_get_actual_len(newext),
2081 BUG_ON(newext->ee_block == nearex->ee_block);
2082 ext_debug("insert %u:%llu:[%d]%d after: "
2084 le32_to_cpu(newext->ee_block),
2085 ext4_ext_pblock(newext),
2086 ext4_ext_is_unwritten(newext),
2087 ext4_ext_get_actual_len(newext),
2090 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2092 ext_debug("insert %u:%llu:[%d]%d: "
2093 "move %d extents from 0x%p to 0x%p\n",
2094 le32_to_cpu(newext->ee_block),
2095 ext4_ext_pblock(newext),
2096 ext4_ext_is_unwritten(newext),
2097 ext4_ext_get_actual_len(newext),
2098 len, nearex, nearex + 1);
2099 memmove(nearex + 1, nearex,
2100 len * sizeof(struct ext4_extent));
2104 le16_add_cpu(&eh->eh_entries, 1);
2105 path[depth].p_ext = nearex;
2106 nearex->ee_block = newext->ee_block;
2107 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2108 nearex->ee_len = newext->ee_len;
2111 /* try to merge extents */
2112 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2113 ext4_ext_try_to_merge(handle, inode, path, nearex);
2116 /* time to correct all indexes above */
2117 err = ext4_ext_correct_indexes(handle, inode, path);
2121 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2125 ext4_ext_drop_refs(npath);
2131 static int ext4_fill_fiemap_extents(struct inode *inode,
2132 ext4_lblk_t block, ext4_lblk_t num,
2133 struct fiemap_extent_info *fieinfo)
2135 struct ext4_ext_path *path = NULL;
2136 struct ext4_extent *ex;
2137 struct extent_status es;
2138 ext4_lblk_t next, next_del, start = 0, end = 0;
2139 ext4_lblk_t last = block + num;
2140 int exists, depth = 0, err = 0;
2141 unsigned int flags = 0;
2142 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2144 while (block < last && block != EXT_MAX_BLOCKS) {
2146 /* find extent for this block */
2147 down_read(&EXT4_I(inode)->i_data_sem);
2149 if (path && ext_depth(inode) != depth) {
2150 /* depth was changed. we have to realloc path */
2155 path = ext4_ext_find_extent(inode, block, path, 0);
2157 up_read(&EXT4_I(inode)->i_data_sem);
2158 err = PTR_ERR(path);
2163 depth = ext_depth(inode);
2164 if (unlikely(path[depth].p_hdr == NULL)) {
2165 up_read(&EXT4_I(inode)->i_data_sem);
2166 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2170 ex = path[depth].p_ext;
2171 next = ext4_ext_next_allocated_block(path);
2172 ext4_ext_drop_refs(path);
2177 /* there is no extent yet, so try to allocate
2178 * all requested space */
2181 } else if (le32_to_cpu(ex->ee_block) > block) {
2182 /* need to allocate space before found extent */
2184 end = le32_to_cpu(ex->ee_block);
2185 if (block + num < end)
2187 } else if (block >= le32_to_cpu(ex->ee_block)
2188 + ext4_ext_get_actual_len(ex)) {
2189 /* need to allocate space after found extent */
2194 } else if (block >= le32_to_cpu(ex->ee_block)) {
2196 * some part of requested space is covered
2200 end = le32_to_cpu(ex->ee_block)
2201 + ext4_ext_get_actual_len(ex);
2202 if (block + num < end)
2208 BUG_ON(end <= start);
2212 es.es_len = end - start;
2215 es.es_lblk = le32_to_cpu(ex->ee_block);
2216 es.es_len = ext4_ext_get_actual_len(ex);
2217 es.es_pblk = ext4_ext_pblock(ex);
2218 if (ext4_ext_is_unwritten(ex))
2219 flags |= FIEMAP_EXTENT_UNWRITTEN;
2223 * Find delayed extent and update es accordingly. We call
2224 * it even in !exists case to find out whether es is the
2225 * last existing extent or not.
2227 next_del = ext4_find_delayed_extent(inode, &es);
2228 if (!exists && next_del) {
2230 flags |= (FIEMAP_EXTENT_DELALLOC |
2231 FIEMAP_EXTENT_UNKNOWN);
2233 up_read(&EXT4_I(inode)->i_data_sem);
2235 if (unlikely(es.es_len == 0)) {
2236 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2242 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2243 * we need to check next == EXT_MAX_BLOCKS because it is
2244 * possible that an extent is with unwritten and delayed
2245 * status due to when an extent is delayed allocated and
2246 * is allocated by fallocate status tree will track both of
2249 * So we could return a unwritten and delayed extent, and
2250 * its block is equal to 'next'.
2252 if (next == next_del && next == EXT_MAX_BLOCKS) {
2253 flags |= FIEMAP_EXTENT_LAST;
2254 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2255 next != EXT_MAX_BLOCKS)) {
2256 EXT4_ERROR_INODE(inode,
2257 "next extent == %u, next "
2258 "delalloc extent = %u",
2266 err = fiemap_fill_next_extent(fieinfo,
2267 (__u64)es.es_lblk << blksize_bits,
2268 (__u64)es.es_pblk << blksize_bits,
2269 (__u64)es.es_len << blksize_bits,
2279 block = es.es_lblk + es.es_len;
2283 ext4_ext_drop_refs(path);
2291 * ext4_ext_put_gap_in_cache:
2292 * calculate boundaries of the gap that the requested block fits into
2293 * and cache this gap
2296 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2299 int depth = ext_depth(inode);
2300 unsigned long len = 0;
2301 ext4_lblk_t lblock = 0;
2302 struct ext4_extent *ex;
2304 ex = path[depth].p_ext;
2307 * there is no extent yet, so gap is [0;-] and we
2310 ext_debug("cache gap(whole file):");
2311 } else if (block < le32_to_cpu(ex->ee_block)) {
2313 len = le32_to_cpu(ex->ee_block) - block;
2314 ext_debug("cache gap(before): %u [%u:%u]",
2316 le32_to_cpu(ex->ee_block),
2317 ext4_ext_get_actual_len(ex));
2318 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2319 ext4_es_insert_extent(inode, lblock, len, ~0,
2320 EXTENT_STATUS_HOLE);
2321 } else if (block >= le32_to_cpu(ex->ee_block)
2322 + ext4_ext_get_actual_len(ex)) {
2324 lblock = le32_to_cpu(ex->ee_block)
2325 + ext4_ext_get_actual_len(ex);
2327 next = ext4_ext_next_allocated_block(path);
2328 ext_debug("cache gap(after): [%u:%u] %u",
2329 le32_to_cpu(ex->ee_block),
2330 ext4_ext_get_actual_len(ex),
2332 BUG_ON(next == lblock);
2333 len = next - lblock;
2334 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2335 ext4_es_insert_extent(inode, lblock, len, ~0,
2336 EXTENT_STATUS_HOLE);
2341 ext_debug(" -> %u:%lu\n", lblock, len);
2346 * removes index from the index block.
2348 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2349 struct ext4_ext_path *path, int depth)
2354 /* free index block */
2356 path = path + depth;
2357 leaf = ext4_idx_pblock(path->p_idx);
2358 if (unlikely(path->p_hdr->eh_entries == 0)) {
2359 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2362 err = ext4_ext_get_access(handle, inode, path);
2366 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2367 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2368 len *= sizeof(struct ext4_extent_idx);
2369 memmove(path->p_idx, path->p_idx + 1, len);
2372 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2373 err = ext4_ext_dirty(handle, inode, path);
2376 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2377 trace_ext4_ext_rm_idx(inode, leaf);
2379 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2380 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2382 while (--depth >= 0) {
2383 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2386 err = ext4_ext_get_access(handle, inode, path);
2389 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2390 err = ext4_ext_dirty(handle, inode, path);
2398 * ext4_ext_calc_credits_for_single_extent:
2399 * This routine returns max. credits that needed to insert an extent
2400 * to the extent tree.
2401 * When pass the actual path, the caller should calculate credits
2404 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2405 struct ext4_ext_path *path)
2408 int depth = ext_depth(inode);
2411 /* probably there is space in leaf? */
2412 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2413 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2416 * There are some space in the leaf tree, no
2417 * need to account for leaf block credit
2419 * bitmaps and block group descriptor blocks
2420 * and other metadata blocks still need to be
2423 /* 1 bitmap, 1 block group descriptor */
2424 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2429 return ext4_chunk_trans_blocks(inode, nrblocks);
2433 * How many index/leaf blocks need to change/allocate to add @extents extents?
2435 * If we add a single extent, then in the worse case, each tree level
2436 * index/leaf need to be changed in case of the tree split.
2438 * If more extents are inserted, they could cause the whole tree split more
2439 * than once, but this is really rare.
2441 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2446 /* If we are converting the inline data, only one is needed here. */
2447 if (ext4_has_inline_data(inode))
2450 depth = ext_depth(inode);
2460 static inline int get_default_free_blocks_flags(struct inode *inode)
2462 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2463 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2464 else if (ext4_should_journal_data(inode))
2465 return EXT4_FREE_BLOCKS_FORGET;
2469 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2470 struct ext4_extent *ex,
2471 long long *partial_cluster,
2472 ext4_lblk_t from, ext4_lblk_t to)
2474 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2475 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2477 int flags = get_default_free_blocks_flags(inode);
2480 * For bigalloc file systems, we never free a partial cluster
2481 * at the beginning of the extent. Instead, we make a note
2482 * that we tried freeing the cluster, and check to see if we
2483 * need to free it on a subsequent call to ext4_remove_blocks,
2484 * or at the end of the ext4_truncate() operation.
2486 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2488 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2490 * If we have a partial cluster, and it's different from the
2491 * cluster of the last block, we need to explicitly free the
2492 * partial cluster here.
2494 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2495 if ((*partial_cluster > 0) &&
2496 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2497 ext4_free_blocks(handle, inode, NULL,
2498 EXT4_C2B(sbi, *partial_cluster),
2499 sbi->s_cluster_ratio, flags);
2500 *partial_cluster = 0;
2503 #ifdef EXTENTS_STATS
2505 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2506 spin_lock(&sbi->s_ext_stats_lock);
2507 sbi->s_ext_blocks += ee_len;
2508 sbi->s_ext_extents++;
2509 if (ee_len < sbi->s_ext_min)
2510 sbi->s_ext_min = ee_len;
2511 if (ee_len > sbi->s_ext_max)
2512 sbi->s_ext_max = ee_len;
2513 if (ext_depth(inode) > sbi->s_depth_max)
2514 sbi->s_depth_max = ext_depth(inode);
2515 spin_unlock(&sbi->s_ext_stats_lock);
2518 if (from >= le32_to_cpu(ex->ee_block)
2519 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2522 unsigned int unaligned;
2524 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2525 pblk = ext4_ext_pblock(ex) + ee_len - num;
2527 * Usually we want to free partial cluster at the end of the
2528 * extent, except for the situation when the cluster is still
2529 * used by any other extent (partial_cluster is negative).
2531 if (*partial_cluster < 0 &&
2532 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2533 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2535 ext_debug("free last %u blocks starting %llu partial %lld\n",
2536 num, pblk, *partial_cluster);
2537 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2539 * If the block range to be freed didn't start at the
2540 * beginning of a cluster, and we removed the entire
2541 * extent and the cluster is not used by any other extent,
2542 * save the partial cluster here, since we might need to
2543 * delete if we determine that the truncate operation has
2544 * removed all of the blocks in the cluster.
2546 * On the other hand, if we did not manage to free the whole
2547 * extent, we have to mark the cluster as used (store negative
2548 * cluster number in partial_cluster).
2550 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2551 if (unaligned && (ee_len == num) &&
2552 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2553 *partial_cluster = EXT4_B2C(sbi, pblk);
2555 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2556 else if (*partial_cluster > 0)
2557 *partial_cluster = 0;
2559 ext4_error(sbi->s_sb, "strange request: removal(2) "
2560 "%u-%u from %u:%u\n",
2561 from, to, le32_to_cpu(ex->ee_block), ee_len);
2567 * ext4_ext_rm_leaf() Removes the extents associated with the
2568 * blocks appearing between "start" and "end", and splits the extents
2569 * if "start" and "end" appear in the same extent
2571 * @handle: The journal handle
2572 * @inode: The files inode
2573 * @path: The path to the leaf
2574 * @partial_cluster: The cluster which we'll have to free if all extents
2575 * has been released from it. It gets negative in case
2576 * that the cluster is still used.
2577 * @start: The first block to remove
2578 * @end: The last block to remove
2581 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2582 struct ext4_ext_path *path,
2583 long long *partial_cluster,
2584 ext4_lblk_t start, ext4_lblk_t end)
2586 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2587 int err = 0, correct_index = 0;
2588 int depth = ext_depth(inode), credits;
2589 struct ext4_extent_header *eh;
2592 ext4_lblk_t ex_ee_block;
2593 unsigned short ex_ee_len;
2594 unsigned unwritten = 0;
2595 struct ext4_extent *ex;
2598 /* the header must be checked already in ext4_ext_remove_space() */
2599 ext_debug("truncate since %u in leaf to %u\n", start, end);
2600 if (!path[depth].p_hdr)
2601 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2602 eh = path[depth].p_hdr;
2603 if (unlikely(path[depth].p_hdr == NULL)) {
2604 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2607 /* find where to start removing */
2608 ex = path[depth].p_ext;
2610 ex = EXT_LAST_EXTENT(eh);
2612 ex_ee_block = le32_to_cpu(ex->ee_block);
2613 ex_ee_len = ext4_ext_get_actual_len(ex);
2616 * If we're starting with an extent other than the last one in the
2617 * node, we need to see if it shares a cluster with the extent to
2618 * the right (towards the end of the file). If its leftmost cluster
2619 * is this extent's rightmost cluster and it is not cluster aligned,
2620 * we'll mark it as a partial that is not to be deallocated.
2623 if (ex != EXT_LAST_EXTENT(eh)) {
2624 ext4_fsblk_t current_pblk, right_pblk;
2625 long long current_cluster, right_cluster;
2627 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2628 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2629 right_pblk = ext4_ext_pblock(ex + 1);
2630 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2631 if (current_cluster == right_cluster &&
2632 EXT4_PBLK_COFF(sbi, right_pblk))
2633 *partial_cluster = -right_cluster;
2636 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2638 while (ex >= EXT_FIRST_EXTENT(eh) &&
2639 ex_ee_block + ex_ee_len > start) {
2641 if (ext4_ext_is_unwritten(ex))
2646 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2647 unwritten, ex_ee_len);
2648 path[depth].p_ext = ex;
2650 a = ex_ee_block > start ? ex_ee_block : start;
2651 b = ex_ee_block+ex_ee_len - 1 < end ?
2652 ex_ee_block+ex_ee_len - 1 : end;
2654 ext_debug(" border %u:%u\n", a, b);
2656 /* If this extent is beyond the end of the hole, skip it */
2657 if (end < ex_ee_block) {
2659 * We're going to skip this extent and move to another,
2660 * so if this extent is not cluster aligned we have
2661 * to mark the current cluster as used to avoid
2662 * accidentally freeing it later on
2664 pblk = ext4_ext_pblock(ex);
2665 if (EXT4_PBLK_COFF(sbi, pblk))
2667 -((long long)EXT4_B2C(sbi, pblk));
2669 ex_ee_block = le32_to_cpu(ex->ee_block);
2670 ex_ee_len = ext4_ext_get_actual_len(ex);
2672 } else if (b != ex_ee_block + ex_ee_len - 1) {
2673 EXT4_ERROR_INODE(inode,
2674 "can not handle truncate %u:%u "
2676 start, end, ex_ee_block,
2677 ex_ee_block + ex_ee_len - 1);
2680 } else if (a != ex_ee_block) {
2681 /* remove tail of the extent */
2682 num = a - ex_ee_block;
2684 /* remove whole extent: excellent! */
2688 * 3 for leaf, sb, and inode plus 2 (bmap and group
2689 * descriptor) for each block group; assume two block
2690 * groups plus ex_ee_len/blocks_per_block_group for
2693 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2694 if (ex == EXT_FIRST_EXTENT(eh)) {
2696 credits += (ext_depth(inode)) + 1;
2698 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2700 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2704 err = ext4_ext_get_access(handle, inode, path + depth);
2708 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2714 /* this extent is removed; mark slot entirely unused */
2715 ext4_ext_store_pblock(ex, 0);
2717 ex->ee_len = cpu_to_le16(num);
2719 * Do not mark unwritten if all the blocks in the
2720 * extent have been removed.
2722 if (unwritten && num)
2723 ext4_ext_mark_unwritten(ex);
2725 * If the extent was completely released,
2726 * we need to remove it from the leaf
2729 if (end != EXT_MAX_BLOCKS - 1) {
2731 * For hole punching, we need to scoot all the
2732 * extents up when an extent is removed so that
2733 * we dont have blank extents in the middle
2735 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2736 sizeof(struct ext4_extent));
2738 /* Now get rid of the one at the end */
2739 memset(EXT_LAST_EXTENT(eh), 0,
2740 sizeof(struct ext4_extent));
2742 le16_add_cpu(&eh->eh_entries, -1);
2743 } else if (*partial_cluster > 0)
2744 *partial_cluster = 0;
2746 err = ext4_ext_dirty(handle, inode, path + depth);
2750 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2751 ext4_ext_pblock(ex));
2753 ex_ee_block = le32_to_cpu(ex->ee_block);
2754 ex_ee_len = ext4_ext_get_actual_len(ex);
2757 if (correct_index && eh->eh_entries)
2758 err = ext4_ext_correct_indexes(handle, inode, path);
2761 * If there's a partial cluster and at least one extent remains in
2762 * the leaf, free the partial cluster if it isn't shared with the
2763 * current extent. If there's a partial cluster and no extents
2764 * remain in the leaf, it can't be freed here. It can only be
2765 * freed when it's possible to determine if it's not shared with
2766 * any other extent - when the next leaf is processed or when space
2767 * removal is complete.
2769 if (*partial_cluster > 0 && eh->eh_entries &&
2770 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2771 *partial_cluster)) {
2772 int flags = get_default_free_blocks_flags(inode);
2774 ext4_free_blocks(handle, inode, NULL,
2775 EXT4_C2B(sbi, *partial_cluster),
2776 sbi->s_cluster_ratio, flags);
2777 *partial_cluster = 0;
2780 /* if this leaf is free, then we should
2781 * remove it from index block above */
2782 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2783 err = ext4_ext_rm_idx(handle, inode, path, depth);
2790 * ext4_ext_more_to_rm:
2791 * returns 1 if current index has to be freed (even partial)
2794 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2796 BUG_ON(path->p_idx == NULL);
2798 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2802 * if truncate on deeper level happened, it wasn't partial,
2803 * so we have to consider current index for truncation
2805 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2810 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2813 struct super_block *sb = inode->i_sb;
2814 int depth = ext_depth(inode);
2815 struct ext4_ext_path *path = NULL;
2816 long long partial_cluster = 0;
2820 ext_debug("truncate since %u to %u\n", start, end);
2822 /* probably first extent we're gonna free will be last in block */
2823 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2825 return PTR_ERR(handle);
2828 trace_ext4_ext_remove_space(inode, start, end, depth);
2831 * Check if we are removing extents inside the extent tree. If that
2832 * is the case, we are going to punch a hole inside the extent tree
2833 * so we have to check whether we need to split the extent covering
2834 * the last block to remove so we can easily remove the part of it
2835 * in ext4_ext_rm_leaf().
2837 if (end < EXT_MAX_BLOCKS - 1) {
2838 struct ext4_extent *ex;
2839 ext4_lblk_t ee_block;
2841 /* find extent for this block */
2842 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2844 ext4_journal_stop(handle);
2845 return PTR_ERR(path);
2847 depth = ext_depth(inode);
2848 /* Leaf not may not exist only if inode has no blocks at all */
2849 ex = path[depth].p_ext;
2852 EXT4_ERROR_INODE(inode,
2853 "path[%d].p_hdr == NULL",
2860 ee_block = le32_to_cpu(ex->ee_block);
2863 * See if the last block is inside the extent, if so split
2864 * the extent at 'end' block so we can easily remove the
2865 * tail of the first part of the split extent in
2866 * ext4_ext_rm_leaf().
2868 if (end >= ee_block &&
2869 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2871 * Split the extent in two so that 'end' is the last
2872 * block in the first new extent. Also we should not
2873 * fail removing space due to ENOSPC so try to use
2874 * reserved block if that happens.
2876 err = ext4_force_split_extent_at(handle, inode, path,
2883 * We start scanning from right side, freeing all the blocks
2884 * after i_size and walking into the tree depth-wise.
2886 depth = ext_depth(inode);
2891 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2893 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2896 ext4_journal_stop(handle);
2899 path[0].p_depth = depth;
2900 path[0].p_hdr = ext_inode_hdr(inode);
2903 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2910 while (i >= 0 && err == 0) {
2912 /* this is leaf block */
2913 err = ext4_ext_rm_leaf(handle, inode, path,
2914 &partial_cluster, start,
2916 /* root level has p_bh == NULL, brelse() eats this */
2917 brelse(path[i].p_bh);
2918 path[i].p_bh = NULL;
2923 /* this is index block */
2924 if (!path[i].p_hdr) {
2925 ext_debug("initialize header\n");
2926 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2929 if (!path[i].p_idx) {
2930 /* this level hasn't been touched yet */
2931 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2932 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2933 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2935 le16_to_cpu(path[i].p_hdr->eh_entries));
2937 /* we were already here, see at next index */
2941 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2942 i, EXT_FIRST_INDEX(path[i].p_hdr),
2944 if (ext4_ext_more_to_rm(path + i)) {
2945 struct buffer_head *bh;
2946 /* go to the next level */
2947 ext_debug("move to level %d (block %llu)\n",
2948 i + 1, ext4_idx_pblock(path[i].p_idx));
2949 memset(path + i + 1, 0, sizeof(*path));
2950 bh = read_extent_tree_block(inode,
2951 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2954 /* should we reset i_size? */
2958 /* Yield here to deal with large extent trees.
2959 * Should be a no-op if we did IO above. */
2961 if (WARN_ON(i + 1 > depth)) {
2965 path[i + 1].p_bh = bh;
2967 /* save actual number of indexes since this
2968 * number is changed at the next iteration */
2969 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2972 /* we finished processing this index, go up */
2973 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2974 /* index is empty, remove it;
2975 * handle must be already prepared by the
2976 * truncatei_leaf() */
2977 err = ext4_ext_rm_idx(handle, inode, path, i);
2979 /* root level has p_bh == NULL, brelse() eats this */
2980 brelse(path[i].p_bh);
2981 path[i].p_bh = NULL;
2983 ext_debug("return to level %d\n", i);
2987 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2988 partial_cluster, path->p_hdr->eh_entries);
2990 /* If we still have something in the partial cluster and we have removed
2991 * even the first extent, then we should free the blocks in the partial
2992 * cluster as well. */
2993 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2994 int flags = get_default_free_blocks_flags(inode);
2996 ext4_free_blocks(handle, inode, NULL,
2997 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2998 EXT4_SB(sb)->s_cluster_ratio, flags);
2999 partial_cluster = 0;
3002 /* TODO: flexible tree reduction should be here */
3003 if (path->p_hdr->eh_entries == 0) {
3005 * truncate to zero freed all the tree,
3006 * so we need to correct eh_depth
3008 err = ext4_ext_get_access(handle, inode, path);
3010 ext_inode_hdr(inode)->eh_depth = 0;
3011 ext_inode_hdr(inode)->eh_max =
3012 cpu_to_le16(ext4_ext_space_root(inode, 0));
3013 err = ext4_ext_dirty(handle, inode, path);
3017 ext4_ext_drop_refs(path);
3019 if (err == -EAGAIN) {
3023 ext4_journal_stop(handle);
3029 * called at mount time
3031 void ext4_ext_init(struct super_block *sb)
3034 * possible initialization would be here
3037 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3038 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3039 printk(KERN_INFO "EXT4-fs: file extents enabled"
3040 #ifdef AGGRESSIVE_TEST
3041 ", aggressive tests"
3043 #ifdef CHECK_BINSEARCH
3046 #ifdef EXTENTS_STATS
3051 #ifdef EXTENTS_STATS
3052 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3053 EXT4_SB(sb)->s_ext_min = 1 << 30;
3054 EXT4_SB(sb)->s_ext_max = 0;
3060 * called at umount time
3062 void ext4_ext_release(struct super_block *sb)
3064 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3067 #ifdef EXTENTS_STATS
3068 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3069 struct ext4_sb_info *sbi = EXT4_SB(sb);
3070 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3071 sbi->s_ext_blocks, sbi->s_ext_extents,
3072 sbi->s_ext_blocks / sbi->s_ext_extents);
3073 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3074 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3079 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3081 ext4_lblk_t ee_block;
3082 ext4_fsblk_t ee_pblock;
3083 unsigned int ee_len;
3085 ee_block = le32_to_cpu(ex->ee_block);
3086 ee_len = ext4_ext_get_actual_len(ex);
3087 ee_pblock = ext4_ext_pblock(ex);
3092 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3093 EXTENT_STATUS_WRITTEN);
3096 /* FIXME!! we need to try to merge to left or right after zero-out */
3097 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3099 ext4_fsblk_t ee_pblock;
3100 unsigned int ee_len;
3103 ee_len = ext4_ext_get_actual_len(ex);
3104 ee_pblock = ext4_ext_pblock(ex);
3106 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3114 * ext4_split_extent_at() splits an extent at given block.
3116 * @handle: the journal handle
3117 * @inode: the file inode
3118 * @path: the path to the extent
3119 * @split: the logical block where the extent is splitted.
3120 * @split_flags: indicates if the extent could be zeroout if split fails, and
3121 * the states(init or unwritten) of new extents.
3122 * @flags: flags used to insert new extent to extent tree.
3125 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3126 * of which are deterimined by split_flag.
3128 * There are two cases:
3129 * a> the extent are splitted into two extent.
3130 * b> split is not needed, and just mark the extent.
3132 * return 0 on success.
3134 static int ext4_split_extent_at(handle_t *handle,
3135 struct inode *inode,
3136 struct ext4_ext_path *path,
3141 ext4_fsblk_t newblock;
3142 ext4_lblk_t ee_block;
3143 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3144 struct ext4_extent *ex2 = NULL;
3145 unsigned int ee_len, depth;
3148 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3149 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3151 ext_debug("ext4_split_extents_at: inode %lu, logical"
3152 "block %llu\n", inode->i_ino, (unsigned long long)split);
3154 ext4_ext_show_leaf(inode, path);
3156 depth = ext_depth(inode);
3157 ex = path[depth].p_ext;
3158 ee_block = le32_to_cpu(ex->ee_block);
3159 ee_len = ext4_ext_get_actual_len(ex);
3160 newblock = split - ee_block + ext4_ext_pblock(ex);
3162 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3163 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3164 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3165 EXT4_EXT_MARK_UNWRIT1 |
3166 EXT4_EXT_MARK_UNWRIT2));
3168 err = ext4_ext_get_access(handle, inode, path + depth);
3172 if (split == ee_block) {
3174 * case b: block @split is the block that the extent begins with
3175 * then we just change the state of the extent, and splitting
3178 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3179 ext4_ext_mark_unwritten(ex);
3181 ext4_ext_mark_initialized(ex);
3183 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3184 ext4_ext_try_to_merge(handle, inode, path, ex);
3186 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3191 memcpy(&orig_ex, ex, sizeof(orig_ex));
3192 ex->ee_len = cpu_to_le16(split - ee_block);
3193 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3194 ext4_ext_mark_unwritten(ex);
3197 * path may lead to new leaf, not to original leaf any more
3198 * after ext4_ext_insert_extent() returns,
3200 err = ext4_ext_dirty(handle, inode, path + depth);
3202 goto fix_extent_len;
3205 ex2->ee_block = cpu_to_le32(split);
3206 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3207 ext4_ext_store_pblock(ex2, newblock);
3208 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3209 ext4_ext_mark_unwritten(ex2);
3211 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3212 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3213 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3214 if (split_flag & EXT4_EXT_DATA_VALID1) {
3215 err = ext4_ext_zeroout(inode, ex2);
3216 zero_ex.ee_block = ex2->ee_block;
3217 zero_ex.ee_len = cpu_to_le16(
3218 ext4_ext_get_actual_len(ex2));
3219 ext4_ext_store_pblock(&zero_ex,
3220 ext4_ext_pblock(ex2));
3222 err = ext4_ext_zeroout(inode, ex);
3223 zero_ex.ee_block = ex->ee_block;
3224 zero_ex.ee_len = cpu_to_le16(
3225 ext4_ext_get_actual_len(ex));
3226 ext4_ext_store_pblock(&zero_ex,
3227 ext4_ext_pblock(ex));
3230 err = ext4_ext_zeroout(inode, &orig_ex);
3231 zero_ex.ee_block = orig_ex.ee_block;
3232 zero_ex.ee_len = cpu_to_le16(
3233 ext4_ext_get_actual_len(&orig_ex));
3234 ext4_ext_store_pblock(&zero_ex,
3235 ext4_ext_pblock(&orig_ex));
3239 goto fix_extent_len;
3240 /* update the extent length and mark as initialized */
3241 ex->ee_len = cpu_to_le16(ee_len);
3242 ext4_ext_try_to_merge(handle, inode, path, ex);
3243 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3245 goto fix_extent_len;
3247 /* update extent status tree */
3248 err = ext4_zeroout_es(inode, &zero_ex);
3252 goto fix_extent_len;
3255 ext4_ext_show_leaf(inode, path);
3259 ex->ee_len = orig_ex.ee_len;
3260 ext4_ext_dirty(handle, inode, path + path->p_depth);
3265 * ext4_split_extents() splits an extent and mark extent which is covered
3266 * by @map as split_flags indicates
3268 * It may result in splitting the extent into multiple extents (up to three)
3269 * There are three possibilities:
3270 * a> There is no split required
3271 * b> Splits in two extents: Split is happening at either end of the extent
3272 * c> Splits in three extents: Somone is splitting in middle of the extent
3275 static int ext4_split_extent(handle_t *handle,
3276 struct inode *inode,
3277 struct ext4_ext_path *path,
3278 struct ext4_map_blocks *map,
3282 ext4_lblk_t ee_block;
3283 struct ext4_extent *ex;
3284 unsigned int ee_len, depth;
3287 int split_flag1, flags1;
3288 int allocated = map->m_len;
3290 depth = ext_depth(inode);
3291 ex = path[depth].p_ext;
3292 ee_block = le32_to_cpu(ex->ee_block);
3293 ee_len = ext4_ext_get_actual_len(ex);
3294 unwritten = ext4_ext_is_unwritten(ex);
3296 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3297 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3298 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3300 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3301 EXT4_EXT_MARK_UNWRIT2;
3302 if (split_flag & EXT4_EXT_DATA_VALID2)
3303 split_flag1 |= EXT4_EXT_DATA_VALID1;
3304 err = ext4_split_extent_at(handle, inode, path,
3305 map->m_lblk + map->m_len, split_flag1, flags1);
3309 allocated = ee_len - (map->m_lblk - ee_block);
3312 * Update path is required because previous ext4_split_extent_at() may
3313 * result in split of original leaf or extent zeroout.
3315 ext4_ext_drop_refs(path);
3316 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3318 return PTR_ERR(path);
3319 depth = ext_depth(inode);
3320 ex = path[depth].p_ext;
3322 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3323 (unsigned long) map->m_lblk);
3326 unwritten = ext4_ext_is_unwritten(ex);
3329 if (map->m_lblk >= ee_block) {
3330 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3332 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3333 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3334 EXT4_EXT_MARK_UNWRIT2);
3336 err = ext4_split_extent_at(handle, inode, path,
3337 map->m_lblk, split_flag1, flags);
3342 ext4_ext_show_leaf(inode, path);
3344 return err ? err : allocated;
3348 * This function is called by ext4_ext_map_blocks() if someone tries to write
3349 * to an unwritten extent. It may result in splitting the unwritten
3350 * extent into multiple extents (up to three - one initialized and two
3352 * There are three possibilities:
3353 * a> There is no split required: Entire extent should be initialized
3354 * b> Splits in two extents: Write is happening at either end of the extent
3355 * c> Splits in three extents: Somone is writing in middle of the extent
3358 * - The extent pointed to by 'path' is unwritten.
3359 * - The extent pointed to by 'path' contains a superset
3360 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3362 * Post-conditions on success:
3363 * - the returned value is the number of blocks beyond map->l_lblk
3364 * that are allocated and initialized.
3365 * It is guaranteed to be >= map->m_len.
3367 static int ext4_ext_convert_to_initialized(handle_t *handle,
3368 struct inode *inode,
3369 struct ext4_map_blocks *map,
3370 struct ext4_ext_path *path,
3373 struct ext4_sb_info *sbi;
3374 struct ext4_extent_header *eh;
3375 struct ext4_map_blocks split_map;
3376 struct ext4_extent zero_ex;
3377 struct ext4_extent *ex, *abut_ex;
3378 ext4_lblk_t ee_block, eof_block;
3379 unsigned int ee_len, depth, map_len = map->m_len;
3380 int allocated = 0, max_zeroout = 0;
3384 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3385 "block %llu, max_blocks %u\n", inode->i_ino,
3386 (unsigned long long)map->m_lblk, map_len);
3388 sbi = EXT4_SB(inode->i_sb);
3389 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3390 inode->i_sb->s_blocksize_bits;
3391 if (eof_block < map->m_lblk + map_len)
3392 eof_block = map->m_lblk + map_len;
3394 depth = ext_depth(inode);
3395 eh = path[depth].p_hdr;
3396 ex = path[depth].p_ext;
3397 ee_block = le32_to_cpu(ex->ee_block);
3398 ee_len = ext4_ext_get_actual_len(ex);
3401 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3403 /* Pre-conditions */
3404 BUG_ON(!ext4_ext_is_unwritten(ex));
3405 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3408 * Attempt to transfer newly initialized blocks from the currently
3409 * unwritten extent to its neighbor. This is much cheaper
3410 * than an insertion followed by a merge as those involve costly
3411 * memmove() calls. Transferring to the left is the common case in
3412 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3413 * followed by append writes.
3415 * Limitations of the current logic:
3416 * - L1: we do not deal with writes covering the whole extent.
3417 * This would require removing the extent if the transfer
3419 * - L2: we only attempt to merge with an extent stored in the
3420 * same extent tree node.
3422 if ((map->m_lblk == ee_block) &&
3423 /* See if we can merge left */
3424 (map_len < ee_len) && /*L1*/
3425 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3426 ext4_lblk_t prev_lblk;
3427 ext4_fsblk_t prev_pblk, ee_pblk;
3428 unsigned int prev_len;
3431 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3432 prev_len = ext4_ext_get_actual_len(abut_ex);
3433 prev_pblk = ext4_ext_pblock(abut_ex);
3434 ee_pblk = ext4_ext_pblock(ex);
3437 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3438 * upon those conditions:
3439 * - C1: abut_ex is initialized,
3440 * - C2: abut_ex is logically abutting ex,
3441 * - C3: abut_ex is physically abutting ex,
3442 * - C4: abut_ex can receive the additional blocks without
3443 * overflowing the (initialized) length limit.
3445 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3446 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3447 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3448 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3449 err = ext4_ext_get_access(handle, inode, path + depth);
3453 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3456 /* Shift the start of ex by 'map_len' blocks */
3457 ex->ee_block = cpu_to_le32(ee_block + map_len);
3458 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3459 ex->ee_len = cpu_to_le16(ee_len - map_len);
3460 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3462 /* Extend abut_ex by 'map_len' blocks */
3463 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3465 /* Result: number of initialized blocks past m_lblk */
3466 allocated = map_len;
3468 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3469 (map_len < ee_len) && /*L1*/
3470 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3471 /* See if we can merge right */
3472 ext4_lblk_t next_lblk;
3473 ext4_fsblk_t next_pblk, ee_pblk;
3474 unsigned int next_len;
3477 next_lblk = le32_to_cpu(abut_ex->ee_block);
3478 next_len = ext4_ext_get_actual_len(abut_ex);
3479 next_pblk = ext4_ext_pblock(abut_ex);
3480 ee_pblk = ext4_ext_pblock(ex);
3483 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3484 * upon those conditions:
3485 * - C1: abut_ex is initialized,
3486 * - C2: abut_ex is logically abutting ex,
3487 * - C3: abut_ex is physically abutting ex,
3488 * - C4: abut_ex can receive the additional blocks without
3489 * overflowing the (initialized) length limit.
3491 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3492 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3493 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3494 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3495 err = ext4_ext_get_access(handle, inode, path + depth);
3499 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3502 /* Shift the start of abut_ex by 'map_len' blocks */
3503 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3504 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3505 ex->ee_len = cpu_to_le16(ee_len - map_len);
3506 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3508 /* Extend abut_ex by 'map_len' blocks */
3509 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3511 /* Result: number of initialized blocks past m_lblk */
3512 allocated = map_len;
3516 /* Mark the block containing both extents as dirty */
3517 ext4_ext_dirty(handle, inode, path + depth);
3519 /* Update path to point to the right extent */
3520 path[depth].p_ext = abut_ex;
3523 allocated = ee_len - (map->m_lblk - ee_block);
3525 WARN_ON(map->m_lblk < ee_block);
3527 * It is safe to convert extent to initialized via explicit
3528 * zeroout only if extent is fully inside i_size or new_size.
3530 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3532 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3533 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3534 (inode->i_sb->s_blocksize_bits - 10);
3536 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3537 if (max_zeroout && (ee_len <= max_zeroout)) {
3538 err = ext4_ext_zeroout(inode, ex);
3541 zero_ex.ee_block = ex->ee_block;
3542 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3543 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3545 err = ext4_ext_get_access(handle, inode, path + depth);
3548 ext4_ext_mark_initialized(ex);
3549 ext4_ext_try_to_merge(handle, inode, path, ex);
3550 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3556 * 1. split the extent into three extents.
3557 * 2. split the extent into two extents, zeroout the first half.
3558 * 3. split the extent into two extents, zeroout the second half.
3559 * 4. split the extent into two extents with out zeroout.
3561 split_map.m_lblk = map->m_lblk;
3562 split_map.m_len = map->m_len;
3564 if (max_zeroout && (allocated > map->m_len)) {
3565 if (allocated <= max_zeroout) {
3568 cpu_to_le32(map->m_lblk);
3569 zero_ex.ee_len = cpu_to_le16(allocated);
3570 ext4_ext_store_pblock(&zero_ex,
3571 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3572 err = ext4_ext_zeroout(inode, &zero_ex);
3575 split_map.m_lblk = map->m_lblk;
3576 split_map.m_len = allocated;
3577 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3579 if (map->m_lblk != ee_block) {
3580 zero_ex.ee_block = ex->ee_block;
3581 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3583 ext4_ext_store_pblock(&zero_ex,
3584 ext4_ext_pblock(ex));
3585 err = ext4_ext_zeroout(inode, &zero_ex);
3590 split_map.m_lblk = ee_block;
3591 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3592 allocated = map->m_len;
3596 allocated = ext4_split_extent(handle, inode, path,
3597 &split_map, split_flag, flags);
3602 /* If we have gotten a failure, don't zero out status tree */
3604 err = ext4_zeroout_es(inode, &zero_ex);
3605 return err ? err : allocated;
3609 * This function is called by ext4_ext_map_blocks() from
3610 * ext4_get_blocks_dio_write() when DIO to write
3611 * to an unwritten extent.
3613 * Writing to an unwritten extent may result in splitting the unwritten
3614 * extent into multiple initialized/unwritten extents (up to three)
3615 * There are three possibilities:
3616 * a> There is no split required: Entire extent should be unwritten
3617 * b> Splits in two extents: Write is happening at either end of the extent
3618 * c> Splits in three extents: Somone is writing in middle of the extent
3620 * This works the same way in the case of initialized -> unwritten conversion.
3622 * One of more index blocks maybe needed if the extent tree grow after
3623 * the unwritten extent split. To prevent ENOSPC occur at the IO
3624 * complete, we need to split the unwritten extent before DIO submit
3625 * the IO. The unwritten extent called at this time will be split
3626 * into three unwritten extent(at most). After IO complete, the part
3627 * being filled will be convert to initialized by the end_io callback function
3628 * via ext4_convert_unwritten_extents().
3630 * Returns the size of unwritten extent to be written on success.
3632 static int ext4_split_convert_extents(handle_t *handle,
3633 struct inode *inode,
3634 struct ext4_map_blocks *map,
3635 struct ext4_ext_path *path,
3638 ext4_lblk_t eof_block;
3639 ext4_lblk_t ee_block;
3640 struct ext4_extent *ex;
3641 unsigned int ee_len;
3642 int split_flag = 0, depth;
3644 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3645 __func__, inode->i_ino,
3646 (unsigned long long)map->m_lblk, map->m_len);
3648 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3649 inode->i_sb->s_blocksize_bits;
3650 if (eof_block < map->m_lblk + map->m_len)
3651 eof_block = map->m_lblk + map->m_len;
3653 * It is safe to convert extent to initialized via explicit
3654 * zeroout only if extent is fully insde i_size or new_size.
3656 depth = ext_depth(inode);
3657 ex = path[depth].p_ext;
3658 ee_block = le32_to_cpu(ex->ee_block);
3659 ee_len = ext4_ext_get_actual_len(ex);
3661 /* Convert to unwritten */
3662 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3663 split_flag |= EXT4_EXT_DATA_VALID1;
3664 /* Convert to initialized */
3665 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3666 split_flag |= ee_block + ee_len <= eof_block ?
3667 EXT4_EXT_MAY_ZEROOUT : 0;
3668 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3670 flags |= EXT4_GET_BLOCKS_PRE_IO;
3671 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3674 static int ext4_convert_initialized_extents(handle_t *handle,
3675 struct inode *inode,
3676 struct ext4_map_blocks *map,
3677 struct ext4_ext_path *path)
3679 struct ext4_extent *ex;
3680 ext4_lblk_t ee_block;
3681 unsigned int ee_len;
3685 depth = ext_depth(inode);
3686 ex = path[depth].p_ext;
3687 ee_block = le32_to_cpu(ex->ee_block);
3688 ee_len = ext4_ext_get_actual_len(ex);
3690 ext_debug("%s: inode %lu, logical"
3691 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3692 (unsigned long long)ee_block, ee_len);
3694 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3695 err = ext4_split_convert_extents(handle, inode, map, path,
3696 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3699 ext4_ext_drop_refs(path);
3700 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3702 err = PTR_ERR(path);
3705 depth = ext_depth(inode);
3706 ex = path[depth].p_ext;
3708 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3709 (unsigned long) map->m_lblk);
3715 err = ext4_ext_get_access(handle, inode, path + depth);
3718 /* first mark the extent as unwritten */
3719 ext4_ext_mark_unwritten(ex);
3721 /* note: ext4_ext_correct_indexes() isn't needed here because
3722 * borders are not changed
3724 ext4_ext_try_to_merge(handle, inode, path, ex);
3726 /* Mark modified extent as dirty */
3727 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3729 ext4_ext_show_leaf(inode, path);
3734 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3735 struct inode *inode,
3736 struct ext4_map_blocks *map,
3737 struct ext4_ext_path *path)
3739 struct ext4_extent *ex;
3740 ext4_lblk_t ee_block;
3741 unsigned int ee_len;
3745 depth = ext_depth(inode);
3746 ex = path[depth].p_ext;
3747 ee_block = le32_to_cpu(ex->ee_block);
3748 ee_len = ext4_ext_get_actual_len(ex);
3750 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3751 "block %llu, max_blocks %u\n", inode->i_ino,
3752 (unsigned long long)ee_block, ee_len);
3754 /* If extent is larger than requested it is a clear sign that we still
3755 * have some extent state machine issues left. So extent_split is still
3757 * TODO: Once all related issues will be fixed this situation should be
3760 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3762 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3763 " len %u; IO logical block %llu, len %u\n",
3764 inode->i_ino, (unsigned long long)ee_block, ee_len,
3765 (unsigned long long)map->m_lblk, map->m_len);
3767 err = ext4_split_convert_extents(handle, inode, map, path,
3768 EXT4_GET_BLOCKS_CONVERT);
3771 ext4_ext_drop_refs(path);
3772 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3774 err = PTR_ERR(path);
3777 depth = ext_depth(inode);
3778 ex = path[depth].p_ext;
3781 err = ext4_ext_get_access(handle, inode, path + depth);
3784 /* first mark the extent as initialized */
3785 ext4_ext_mark_initialized(ex);
3787 /* note: ext4_ext_correct_indexes() isn't needed here because
3788 * borders are not changed
3790 ext4_ext_try_to_merge(handle, inode, path, ex);
3792 /* Mark modified extent as dirty */
3793 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3795 ext4_ext_show_leaf(inode, path);
3799 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3800 sector_t block, int count)
3803 for (i = 0; i < count; i++)
3804 unmap_underlying_metadata(bdev, block + i);
3808 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3810 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3812 struct ext4_ext_path *path,
3816 struct ext4_extent_header *eh;
3817 struct ext4_extent *last_ex;
3819 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3822 depth = ext_depth(inode);
3823 eh = path[depth].p_hdr;
3826 * We're going to remove EOFBLOCKS_FL entirely in future so we
3827 * do not care for this case anymore. Simply remove the flag
3828 * if there are no extents.
3830 if (unlikely(!eh->eh_entries))
3832 last_ex = EXT_LAST_EXTENT(eh);
3834 * We should clear the EOFBLOCKS_FL flag if we are writing the
3835 * last block in the last extent in the file. We test this by
3836 * first checking to see if the caller to
3837 * ext4_ext_get_blocks() was interested in the last block (or
3838 * a block beyond the last block) in the current extent. If
3839 * this turns out to be false, we can bail out from this
3840 * function immediately.
3842 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3843 ext4_ext_get_actual_len(last_ex))
3846 * If the caller does appear to be planning to write at or
3847 * beyond the end of the current extent, we then test to see
3848 * if the current extent is the last extent in the file, by
3849 * checking to make sure it was reached via the rightmost node
3850 * at each level of the tree.
3852 for (i = depth-1; i >= 0; i--)
3853 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3856 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3857 return ext4_mark_inode_dirty(handle, inode);
3861 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3863 * Return 1 if there is a delalloc block in the range, otherwise 0.
3865 int ext4_find_delalloc_range(struct inode *inode,
3866 ext4_lblk_t lblk_start,
3867 ext4_lblk_t lblk_end)
3869 struct extent_status es;
3871 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3873 return 0; /* there is no delay extent in this tree */
3874 else if (es.es_lblk <= lblk_start &&
3875 lblk_start < es.es_lblk + es.es_len)
3877 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3883 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3885 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3886 ext4_lblk_t lblk_start, lblk_end;
3887 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3888 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3890 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3894 * Determines how many complete clusters (out of those specified by the 'map')
3895 * are under delalloc and were reserved quota for.
3896 * This function is called when we are writing out the blocks that were
3897 * originally written with their allocation delayed, but then the space was
3898 * allocated using fallocate() before the delayed allocation could be resolved.
3899 * The cases to look for are:
3900 * ('=' indicated delayed allocated blocks
3901 * '-' indicates non-delayed allocated blocks)
3902 * (a) partial clusters towards beginning and/or end outside of allocated range
3903 * are not delalloc'ed.
3905 * |----c---=|====c====|====c====|===-c----|
3906 * |++++++ allocated ++++++|
3907 * ==> 4 complete clusters in above example
3909 * (b) partial cluster (outside of allocated range) towards either end is
3910 * marked for delayed allocation. In this case, we will exclude that
3913 * |----====c========|========c========|
3914 * |++++++ allocated ++++++|
3915 * ==> 1 complete clusters in above example
3918 * |================c================|
3919 * |++++++ allocated ++++++|
3920 * ==> 0 complete clusters in above example
3922 * The ext4_da_update_reserve_space will be called only if we
3923 * determine here that there were some "entire" clusters that span
3924 * this 'allocated' range.
3925 * In the non-bigalloc case, this function will just end up returning num_blks
3926 * without ever calling ext4_find_delalloc_range.
3929 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3930 unsigned int num_blks)
3932 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3933 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3934 ext4_lblk_t lblk_from, lblk_to, c_offset;
3935 unsigned int allocated_clusters = 0;
3937 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3938 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3940 /* max possible clusters for this allocation */
3941 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3943 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3945 /* Check towards left side */
3946 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3948 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3949 lblk_to = lblk_from + c_offset - 1;
3951 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3952 allocated_clusters--;
3955 /* Now check towards right. */
3956 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3957 if (allocated_clusters && c_offset) {
3958 lblk_from = lblk_start + num_blks;
3959 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3961 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3962 allocated_clusters--;
3965 return allocated_clusters;
3969 ext4_ext_convert_initialized_extent(handle_t *handle, struct inode *inode,
3970 struct ext4_map_blocks *map,
3971 struct ext4_ext_path *path, int flags,
3972 unsigned int allocated, ext4_fsblk_t newblock)
3978 * Make sure that the extent is no bigger than we support with
3981 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3982 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3984 ret = ext4_convert_initialized_extents(handle, inode, map,
3987 ext4_update_inode_fsync_trans(handle, inode, 1);
3988 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3992 map->m_flags |= EXT4_MAP_UNWRITTEN;
3993 if (allocated > map->m_len)
3994 allocated = map->m_len;
3995 map->m_len = allocated;
3997 return err ? err : allocated;
4001 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4002 struct ext4_map_blocks *map,
4003 struct ext4_ext_path *path, int flags,
4004 unsigned int allocated, ext4_fsblk_t newblock)
4008 ext4_io_end_t *io = ext4_inode_aio(inode);
4010 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4011 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4012 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4014 ext4_ext_show_leaf(inode, path);
4017 * When writing into unwritten space, we should not fail to
4018 * allocate metadata blocks for the new extent block if needed.
4020 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4022 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4023 allocated, newblock);
4025 /* get_block() before submit the IO, split the extent */
4026 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4027 ret = ext4_split_convert_extents(handle, inode, map,
4028 path, flags | EXT4_GET_BLOCKS_CONVERT);
4032 * Flag the inode(non aio case) or end_io struct (aio case)
4033 * that this IO needs to conversion to written when IO is
4037 ext4_set_io_unwritten_flag(inode, io);
4039 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4040 map->m_flags |= EXT4_MAP_UNWRITTEN;
4043 /* IO end_io complete, convert the filled extent to written */
4044 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4045 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4048 ext4_update_inode_fsync_trans(handle, inode, 1);
4049 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4053 map->m_flags |= EXT4_MAP_MAPPED;
4054 map->m_pblk = newblock;
4055 if (allocated > map->m_len)
4056 allocated = map->m_len;
4057 map->m_len = allocated;
4060 /* buffered IO case */
4062 * repeat fallocate creation request
4063 * we already have an unwritten extent
4065 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4066 map->m_flags |= EXT4_MAP_UNWRITTEN;
4070 /* buffered READ or buffered write_begin() lookup */
4071 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4073 * We have blocks reserved already. We
4074 * return allocated blocks so that delalloc
4075 * won't do block reservation for us. But
4076 * the buffer head will be unmapped so that
4077 * a read from the block returns 0s.
4079 map->m_flags |= EXT4_MAP_UNWRITTEN;
4083 /* buffered write, writepage time, convert*/
4084 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
4086 ext4_update_inode_fsync_trans(handle, inode, 1);
4093 map->m_flags |= EXT4_MAP_NEW;
4095 * if we allocated more blocks than requested
4096 * we need to make sure we unmap the extra block
4097 * allocated. The actual needed block will get
4098 * unmapped later when we find the buffer_head marked
4101 if (allocated > map->m_len) {
4102 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4103 newblock + map->m_len,
4104 allocated - map->m_len);
4105 allocated = map->m_len;
4107 map->m_len = allocated;
4110 * If we have done fallocate with the offset that is already
4111 * delayed allocated, we would have block reservation
4112 * and quota reservation done in the delayed write path.
4113 * But fallocate would have already updated quota and block
4114 * count for this offset. So cancel these reservation
4116 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4117 unsigned int reserved_clusters;
4118 reserved_clusters = get_reserved_cluster_alloc(inode,
4119 map->m_lblk, map->m_len);
4120 if (reserved_clusters)
4121 ext4_da_update_reserve_space(inode,
4127 map->m_flags |= EXT4_MAP_MAPPED;
4128 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4129 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4135 if (allocated > map->m_len)
4136 allocated = map->m_len;
4137 ext4_ext_show_leaf(inode, path);
4138 map->m_pblk = newblock;
4139 map->m_len = allocated;
4141 return err ? err : allocated;
4145 * get_implied_cluster_alloc - check to see if the requested
4146 * allocation (in the map structure) overlaps with a cluster already
4147 * allocated in an extent.
4148 * @sb The filesystem superblock structure
4149 * @map The requested lblk->pblk mapping
4150 * @ex The extent structure which might contain an implied
4151 * cluster allocation
4153 * This function is called by ext4_ext_map_blocks() after we failed to
4154 * find blocks that were already in the inode's extent tree. Hence,
4155 * we know that the beginning of the requested region cannot overlap
4156 * the extent from the inode's extent tree. There are three cases we
4157 * want to catch. The first is this case:
4159 * |--- cluster # N--|
4160 * |--- extent ---| |---- requested region ---|
4163 * The second case that we need to test for is this one:
4165 * |--------- cluster # N ----------------|
4166 * |--- requested region --| |------- extent ----|
4167 * |=======================|
4169 * The third case is when the requested region lies between two extents
4170 * within the same cluster:
4171 * |------------- cluster # N-------------|
4172 * |----- ex -----| |---- ex_right ----|
4173 * |------ requested region ------|
4174 * |================|
4176 * In each of the above cases, we need to set the map->m_pblk and
4177 * map->m_len so it corresponds to the return the extent labelled as
4178 * "|====|" from cluster #N, since it is already in use for data in
4179 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4180 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4181 * as a new "allocated" block region. Otherwise, we will return 0 and
4182 * ext4_ext_map_blocks() will then allocate one or more new clusters
4183 * by calling ext4_mb_new_blocks().
4185 static int get_implied_cluster_alloc(struct super_block *sb,
4186 struct ext4_map_blocks *map,
4187 struct ext4_extent *ex,
4188 struct ext4_ext_path *path)
4190 struct ext4_sb_info *sbi = EXT4_SB(sb);
4191 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4192 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4193 ext4_lblk_t rr_cluster_start;
4194 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4195 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4196 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4198 /* The extent passed in that we are trying to match */
4199 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4200 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4202 /* The requested region passed into ext4_map_blocks() */
4203 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4205 if ((rr_cluster_start == ex_cluster_end) ||
4206 (rr_cluster_start == ex_cluster_start)) {
4207 if (rr_cluster_start == ex_cluster_end)
4208 ee_start += ee_len - 1;
4209 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4210 map->m_len = min(map->m_len,
4211 (unsigned) sbi->s_cluster_ratio - c_offset);
4213 * Check for and handle this case:
4215 * |--------- cluster # N-------------|
4216 * |------- extent ----|
4217 * |--- requested region ---|
4221 if (map->m_lblk < ee_block)
4222 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4225 * Check for the case where there is already another allocated
4226 * block to the right of 'ex' but before the end of the cluster.
4228 * |------------- cluster # N-------------|
4229 * |----- ex -----| |---- ex_right ----|
4230 * |------ requested region ------|
4231 * |================|
4233 if (map->m_lblk > ee_block) {
4234 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4235 map->m_len = min(map->m_len, next - map->m_lblk);
4238 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4242 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4248 * Block allocation/map/preallocation routine for extents based files
4251 * Need to be called with
4252 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4253 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4255 * return > 0, number of of blocks already mapped/allocated
4256 * if create == 0 and these are pre-allocated blocks
4257 * buffer head is unmapped
4258 * otherwise blocks are mapped
4260 * return = 0, if plain look up failed (blocks have not been allocated)
4261 * buffer head is unmapped
4263 * return < 0, error case.
4265 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4266 struct ext4_map_blocks *map, int flags)
4268 struct ext4_ext_path *path = NULL;
4269 struct ext4_extent newex, *ex, *ex2;
4270 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4271 ext4_fsblk_t newblock = 0;
4272 int free_on_err = 0, err = 0, depth, ret;
4273 unsigned int allocated = 0, offset = 0;
4274 unsigned int allocated_clusters = 0;
4275 struct ext4_allocation_request ar;
4276 ext4_io_end_t *io = ext4_inode_aio(inode);
4277 ext4_lblk_t cluster_offset;
4278 int set_unwritten = 0;
4280 ext_debug("blocks %u/%u requested for inode %lu\n",
4281 map->m_lblk, map->m_len, inode->i_ino);
4282 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4284 /* find extent for this block */
4285 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4287 err = PTR_ERR(path);
4292 depth = ext_depth(inode);
4295 * consistent leaf must not be empty;
4296 * this situation is possible, though, _during_ tree modification;
4297 * this is why assert can't be put in ext4_ext_find_extent()
4299 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4300 EXT4_ERROR_INODE(inode, "bad extent address "
4301 "lblock: %lu, depth: %d pblock %lld",
4302 (unsigned long) map->m_lblk, depth,
4303 path[depth].p_block);
4308 ex = path[depth].p_ext;
4310 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4311 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4312 unsigned short ee_len;
4316 * unwritten extents are treated as holes, except that
4317 * we split out initialized portions during a write.
4319 ee_len = ext4_ext_get_actual_len(ex);
4321 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4323 /* if found extent covers block, simply return it */
4324 if (in_range(map->m_lblk, ee_block, ee_len)) {
4325 newblock = map->m_lblk - ee_block + ee_start;
4326 /* number of remaining blocks in the extent */
4327 allocated = ee_len - (map->m_lblk - ee_block);
4328 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4329 ee_block, ee_len, newblock);
4332 * If the extent is initialized check whether the
4333 * caller wants to convert it to unwritten.
4335 if ((!ext4_ext_is_unwritten(ex)) &&
4336 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4337 allocated = ext4_ext_convert_initialized_extent(
4338 handle, inode, map, path, flags,
4339 allocated, newblock);
4341 } else if (!ext4_ext_is_unwritten(ex))
4344 ret = ext4_ext_handle_unwritten_extents(
4345 handle, inode, map, path, flags,
4346 allocated, newblock);
4355 if ((sbi->s_cluster_ratio > 1) &&
4356 ext4_find_delalloc_cluster(inode, map->m_lblk))
4357 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4360 * requested block isn't allocated yet;
4361 * we couldn't try to create block if create flag is zero
4363 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4365 * put just found gap into cache to speed up
4366 * subsequent requests
4368 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4369 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4374 * Okay, we need to do block allocation.
4376 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4377 newex.ee_block = cpu_to_le32(map->m_lblk);
4378 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4381 * If we are doing bigalloc, check to see if the extent returned
4382 * by ext4_ext_find_extent() implies a cluster we can use.
4384 if (cluster_offset && ex &&
4385 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4386 ar.len = allocated = map->m_len;
4387 newblock = map->m_pblk;
4388 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4389 goto got_allocated_blocks;
4392 /* find neighbour allocated blocks */
4393 ar.lleft = map->m_lblk;
4394 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4397 ar.lright = map->m_lblk;
4399 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4403 /* Check if the extent after searching to the right implies a
4404 * cluster we can use. */
4405 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4406 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4407 ar.len = allocated = map->m_len;
4408 newblock = map->m_pblk;
4409 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4410 goto got_allocated_blocks;
4414 * See if request is beyond maximum number of blocks we can have in
4415 * a single extent. For an initialized extent this limit is
4416 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4417 * EXT_UNWRITTEN_MAX_LEN.
4419 if (map->m_len > EXT_INIT_MAX_LEN &&
4420 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4421 map->m_len = EXT_INIT_MAX_LEN;
4422 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4423 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4424 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4426 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4427 newex.ee_len = cpu_to_le16(map->m_len);
4428 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4430 allocated = ext4_ext_get_actual_len(&newex);
4432 allocated = map->m_len;
4434 /* allocate new block */
4436 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4437 ar.logical = map->m_lblk;
4439 * We calculate the offset from the beginning of the cluster
4440 * for the logical block number, since when we allocate a
4441 * physical cluster, the physical block should start at the
4442 * same offset from the beginning of the cluster. This is
4443 * needed so that future calls to get_implied_cluster_alloc()
4446 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4447 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4449 ar.logical -= offset;
4450 if (S_ISREG(inode->i_mode))
4451 ar.flags = EXT4_MB_HINT_DATA;
4453 /* disable in-core preallocation for non-regular files */
4455 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4456 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4457 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4460 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4461 ar.goal, newblock, allocated);
4463 allocated_clusters = ar.len;
4464 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4465 if (ar.len > allocated)
4468 got_allocated_blocks:
4469 /* try to insert new extent into found leaf and return */
4470 ext4_ext_store_pblock(&newex, newblock + offset);
4471 newex.ee_len = cpu_to_le16(ar.len);
4472 /* Mark unwritten */
4473 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4474 ext4_ext_mark_unwritten(&newex);
4475 map->m_flags |= EXT4_MAP_UNWRITTEN;
4477 * io_end structure was created for every IO write to an
4478 * unwritten extent. To avoid unnecessary conversion,
4479 * here we flag the IO that really needs the conversion.
4480 * For non asycn direct IO case, flag the inode state
4481 * that we need to perform conversion when IO is done.
4483 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4488 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4489 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4492 err = ext4_ext_insert_extent(handle, inode, path,
4495 if (!err && set_unwritten) {
4497 ext4_set_io_unwritten_flag(inode, io);
4499 ext4_set_inode_state(inode,
4500 EXT4_STATE_DIO_UNWRITTEN);
4503 if (err && free_on_err) {
4504 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4505 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4506 /* free data blocks we just allocated */
4507 /* not a good idea to call discard here directly,
4508 * but otherwise we'd need to call it every free() */
4509 ext4_discard_preallocations(inode);
4510 ext4_free_blocks(handle, inode, NULL, newblock,
4511 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4515 /* previous routine could use block we allocated */
4516 newblock = ext4_ext_pblock(&newex);
4517 allocated = ext4_ext_get_actual_len(&newex);
4518 if (allocated > map->m_len)
4519 allocated = map->m_len;
4520 map->m_flags |= EXT4_MAP_NEW;
4523 * Update reserved blocks/metadata blocks after successful
4524 * block allocation which had been deferred till now.
4526 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4527 unsigned int reserved_clusters;
4529 * Check how many clusters we had reserved this allocated range
4531 reserved_clusters = get_reserved_cluster_alloc(inode,
4532 map->m_lblk, allocated);
4533 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4534 if (reserved_clusters) {
4536 * We have clusters reserved for this range.
4537 * But since we are not doing actual allocation
4538 * and are simply using blocks from previously
4539 * allocated cluster, we should release the
4540 * reservation and not claim quota.
4542 ext4_da_update_reserve_space(inode,
4543 reserved_clusters, 0);
4546 BUG_ON(allocated_clusters < reserved_clusters);
4547 if (reserved_clusters < allocated_clusters) {
4548 struct ext4_inode_info *ei = EXT4_I(inode);
4549 int reservation = allocated_clusters -
4552 * It seems we claimed few clusters outside of
4553 * the range of this allocation. We should give
4554 * it back to the reservation pool. This can
4555 * happen in the following case:
4557 * * Suppose s_cluster_ratio is 4 (i.e., each
4558 * cluster has 4 blocks. Thus, the clusters
4559 * are [0-3],[4-7],[8-11]...
4560 * * First comes delayed allocation write for
4561 * logical blocks 10 & 11. Since there were no
4562 * previous delayed allocated blocks in the
4563 * range [8-11], we would reserve 1 cluster
4565 * * Next comes write for logical blocks 3 to 8.
4566 * In this case, we will reserve 2 clusters
4567 * (for [0-3] and [4-7]; and not for [8-11] as
4568 * that range has a delayed allocated blocks.
4569 * Thus total reserved clusters now becomes 3.
4570 * * Now, during the delayed allocation writeout
4571 * time, we will first write blocks [3-8] and
4572 * allocate 3 clusters for writing these
4573 * blocks. Also, we would claim all these
4574 * three clusters above.
4575 * * Now when we come here to writeout the
4576 * blocks [10-11], we would expect to claim
4577 * the reservation of 1 cluster we had made
4578 * (and we would claim it since there are no
4579 * more delayed allocated blocks in the range
4580 * [8-11]. But our reserved cluster count had
4581 * already gone to 0.
4583 * Thus, at the step 4 above when we determine
4584 * that there are still some unwritten delayed
4585 * allocated blocks outside of our current
4586 * block range, we should increment the
4587 * reserved clusters count so that when the
4588 * remaining blocks finally gets written, we
4591 dquot_reserve_block(inode,
4592 EXT4_C2B(sbi, reservation));
4593 spin_lock(&ei->i_block_reservation_lock);
4594 ei->i_reserved_data_blocks += reservation;
4595 spin_unlock(&ei->i_block_reservation_lock);
4598 * We will claim quota for all newly allocated blocks.
4599 * We're updating the reserved space *after* the
4600 * correction above so we do not accidentally free
4601 * all the metadata reservation because we might
4602 * actually need it later on.
4604 ext4_da_update_reserve_space(inode, allocated_clusters,
4610 * Cache the extent and update transaction to commit on fdatasync only
4611 * when it is _not_ an unwritten extent.
4613 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4614 ext4_update_inode_fsync_trans(handle, inode, 1);
4616 ext4_update_inode_fsync_trans(handle, inode, 0);
4618 if (allocated > map->m_len)
4619 allocated = map->m_len;
4620 ext4_ext_show_leaf(inode, path);
4621 map->m_flags |= EXT4_MAP_MAPPED;
4622 map->m_pblk = newblock;
4623 map->m_len = allocated;
4626 ext4_ext_drop_refs(path);
4630 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4631 err ? err : allocated);
4632 ext4_es_lru_add(inode);
4633 return err ? err : allocated;
4636 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4638 struct super_block *sb = inode->i_sb;
4639 ext4_lblk_t last_block;
4643 * TODO: optimization is possible here.
4644 * Probably we need not scan at all,
4645 * because page truncation is enough.
4648 /* we have to know where to truncate from in crash case */
4649 EXT4_I(inode)->i_disksize = inode->i_size;
4650 ext4_mark_inode_dirty(handle, inode);
4652 last_block = (inode->i_size + sb->s_blocksize - 1)
4653 >> EXT4_BLOCK_SIZE_BITS(sb);
4655 err = ext4_es_remove_extent(inode, last_block,
4656 EXT_MAX_BLOCKS - last_block);
4657 if (err == -ENOMEM) {
4659 congestion_wait(BLK_RW_ASYNC, HZ/50);
4663 ext4_std_error(inode->i_sb, err);
4666 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4667 ext4_std_error(inode->i_sb, err);
4670 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4671 ext4_lblk_t len, loff_t new_size,
4672 int flags, int mode)
4674 struct inode *inode = file_inode(file);
4679 struct ext4_map_blocks map;
4680 unsigned int credits;
4683 map.m_lblk = offset;
4686 * Don't normalize the request if it can fit in one extent so
4687 * that it doesn't get unnecessarily split into multiple
4690 if (len <= EXT_UNWRITTEN_MAX_LEN)
4691 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4694 * credits to insert 1 extent into extent tree
4696 credits = ext4_chunk_trans_blocks(inode, len);
4699 while (ret >= 0 && len) {
4700 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4702 if (IS_ERR(handle)) {
4703 ret = PTR_ERR(handle);
4706 ret = ext4_map_blocks(handle, inode, &map, flags);
4708 ext4_debug("inode #%lu: block %u: len %u: "
4709 "ext4_ext_map_blocks returned %d",
4710 inode->i_ino, map.m_lblk,
4712 ext4_mark_inode_dirty(handle, inode);
4713 ret2 = ext4_journal_stop(handle);
4717 map.m_len = len = len - ret;
4718 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4719 inode->i_ctime = ext4_current_time(inode);
4721 if (epos > new_size)
4723 if (ext4_update_inode_size(inode, epos) & 0x1)
4724 inode->i_mtime = inode->i_ctime;
4726 if (epos > inode->i_size)
4727 ext4_set_inode_flag(inode,
4728 EXT4_INODE_EOFBLOCKS);
4730 ext4_mark_inode_dirty(handle, inode);
4731 ret2 = ext4_journal_stop(handle);
4735 if (ret == -ENOSPC &&
4736 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4741 return ret > 0 ? ret2 : ret;
4744 static long ext4_zero_range(struct file *file, loff_t offset,
4745 loff_t len, int mode)
4747 struct inode *inode = file_inode(file);
4748 handle_t *handle = NULL;
4749 unsigned int max_blocks;
4750 loff_t new_size = 0;
4754 int partial_begin, partial_end;
4757 struct address_space *mapping = inode->i_mapping;
4758 unsigned int blkbits = inode->i_blkbits;
4760 trace_ext4_zero_range(inode, offset, len, mode);
4762 if (!S_ISREG(inode->i_mode))
4765 /* Call ext4_force_commit to flush all data in case of data=journal. */
4766 if (ext4_should_journal_data(inode)) {
4767 ret = ext4_force_commit(inode->i_sb);
4773 * Write out all dirty pages to avoid race conditions
4774 * Then release them.
4776 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4777 ret = filemap_write_and_wait_range(mapping, offset,
4784 * Round up offset. This is not fallocate, we neet to zero out
4785 * blocks, so convert interior block aligned part of the range to
4786 * unwritten and possibly manually zero out unaligned parts of the
4789 start = round_up(offset, 1 << blkbits);
4790 end = round_down((offset + len), 1 << blkbits);
4792 if (start < offset || end > offset + len)
4794 partial_begin = offset & ((1 << blkbits) - 1);
4795 partial_end = (offset + len) & ((1 << blkbits) - 1);
4797 lblk = start >> blkbits;
4798 max_blocks = (end >> blkbits);
4799 if (max_blocks < lblk)
4804 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4805 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4807 if (mode & FALLOC_FL_KEEP_SIZE)
4808 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4810 mutex_lock(&inode->i_mutex);
4813 * Indirect files do not support unwritten extnets
4815 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4820 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4821 offset + len > i_size_read(inode)) {
4822 new_size = offset + len;
4823 ret = inode_newsize_ok(inode, new_size);
4827 * If we have a partial block after EOF we have to allocate
4834 if (max_blocks > 0) {
4836 /* Now release the pages and zero block aligned part of pages*/
4837 truncate_pagecache_range(inode, start, end - 1);
4838 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4840 /* Wait all existing dio workers, newcomers will block on i_mutex */
4841 ext4_inode_block_unlocked_dio(inode);
4842 inode_dio_wait(inode);
4844 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4849 * Remove entire range from the extent status tree.
4851 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4852 * NOT sufficient. I'm not sure why this is the case,
4853 * but let's be conservative and remove the extent
4854 * status tree for the entire inode. There should be
4855 * no outstanding delalloc extents thanks to the
4856 * filemap_write_and_wait_range() call above.
4858 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4862 if (!partial_begin && !partial_end)
4866 * In worst case we have to writeout two nonadjacent unwritten
4867 * blocks and update the inode
4869 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4870 if (ext4_should_journal_data(inode))
4872 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4873 if (IS_ERR(handle)) {
4874 ret = PTR_ERR(handle);
4875 ext4_std_error(inode->i_sb, ret);
4879 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4881 ext4_update_inode_size(inode, new_size);
4884 * Mark that we allocate beyond EOF so the subsequent truncate
4885 * can proceed even if the new size is the same as i_size.
4887 if ((offset + len) > i_size_read(inode))
4888 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4890 ext4_mark_inode_dirty(handle, inode);
4892 /* Zero out partial block at the edges of the range */
4893 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4895 if (file->f_flags & O_SYNC)
4896 ext4_handle_sync(handle);
4898 ext4_journal_stop(handle);
4900 ext4_inode_resume_unlocked_dio(inode);
4902 mutex_unlock(&inode->i_mutex);
4907 * preallocate space for a file. This implements ext4's fallocate file
4908 * operation, which gets called from sys_fallocate system call.
4909 * For block-mapped files, posix_fallocate should fall back to the method
4910 * of writing zeroes to the required new blocks (the same behavior which is
4911 * expected for file systems which do not support fallocate() system call).
4913 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4915 struct inode *inode = file_inode(file);
4916 loff_t new_size = 0;
4917 unsigned int max_blocks;
4921 unsigned int blkbits = inode->i_blkbits;
4923 /* Return error if mode is not supported */
4924 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4925 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4928 if (mode & FALLOC_FL_PUNCH_HOLE)
4929 return ext4_punch_hole(inode, offset, len);
4931 ret = ext4_convert_inline_data(inode);
4936 * currently supporting (pre)allocate mode for extent-based
4939 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4942 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4943 return ext4_collapse_range(inode, offset, len);
4945 if (mode & FALLOC_FL_ZERO_RANGE)
4946 return ext4_zero_range(file, offset, len, mode);
4948 trace_ext4_fallocate_enter(inode, offset, len, mode);
4949 lblk = offset >> blkbits;
4951 * We can't just convert len to max_blocks because
4952 * If blocksize = 4096 offset = 3072 and len = 2048
4954 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4957 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4958 if (mode & FALLOC_FL_KEEP_SIZE)
4959 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4961 mutex_lock(&inode->i_mutex);
4963 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4964 offset + len > i_size_read(inode)) {
4965 new_size = offset + len;
4966 ret = inode_newsize_ok(inode, new_size);
4971 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4976 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4977 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4978 EXT4_I(inode)->i_sync_tid);
4981 mutex_unlock(&inode->i_mutex);
4982 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4987 * This function convert a range of blocks to written extents
4988 * The caller of this function will pass the start offset and the size.
4989 * all unwritten extents within this range will be converted to
4992 * This function is called from the direct IO end io call back
4993 * function, to convert the fallocated extents after IO is completed.
4994 * Returns 0 on success.
4996 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4997 loff_t offset, ssize_t len)
4999 unsigned int max_blocks;
5002 struct ext4_map_blocks map;
5003 unsigned int credits, blkbits = inode->i_blkbits;
5005 map.m_lblk = offset >> blkbits;
5007 * We can't just convert len to max_blocks because
5008 * If blocksize = 4096 offset = 3072 and len = 2048
5010 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5013 * This is somewhat ugly but the idea is clear: When transaction is
5014 * reserved, everything goes into it. Otherwise we rather start several
5015 * smaller transactions for conversion of each extent separately.
5018 handle = ext4_journal_start_reserved(handle,
5019 EXT4_HT_EXT_CONVERT);
5021 return PTR_ERR(handle);
5025 * credits to insert 1 extent into extent tree
5027 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5029 while (ret >= 0 && ret < max_blocks) {
5031 map.m_len = (max_blocks -= ret);
5033 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5035 if (IS_ERR(handle)) {
5036 ret = PTR_ERR(handle);
5040 ret = ext4_map_blocks(handle, inode, &map,
5041 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5043 ext4_warning(inode->i_sb,
5044 "inode #%lu: block %u: len %u: "
5045 "ext4_ext_map_blocks returned %d",
5046 inode->i_ino, map.m_lblk,
5048 ext4_mark_inode_dirty(handle, inode);
5050 ret2 = ext4_journal_stop(handle);
5051 if (ret <= 0 || ret2)
5055 ret2 = ext4_journal_stop(handle);
5056 return ret > 0 ? ret2 : ret;
5060 * If newes is not existing extent (newes->ec_pblk equals zero) find
5061 * delayed extent at start of newes and update newes accordingly and
5062 * return start of the next delayed extent.
5064 * If newes is existing extent (newes->ec_pblk is not equal zero)
5065 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5066 * extent found. Leave newes unmodified.
5068 static int ext4_find_delayed_extent(struct inode *inode,
5069 struct extent_status *newes)
5071 struct extent_status es;
5072 ext4_lblk_t block, next_del;
5074 if (newes->es_pblk == 0) {
5075 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5076 newes->es_lblk + newes->es_len - 1, &es);
5079 * No extent in extent-tree contains block @newes->es_pblk,
5080 * then the block may stay in 1)a hole or 2)delayed-extent.
5086 if (es.es_lblk > newes->es_lblk) {
5088 newes->es_len = min(es.es_lblk - newes->es_lblk,
5093 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5096 block = newes->es_lblk + newes->es_len;
5097 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5099 next_del = EXT_MAX_BLOCKS;
5101 next_del = es.es_lblk;
5105 /* fiemap flags we can handle specified here */
5106 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5108 static int ext4_xattr_fiemap(struct inode *inode,
5109 struct fiemap_extent_info *fieinfo)
5113 __u32 flags = FIEMAP_EXTENT_LAST;
5114 int blockbits = inode->i_sb->s_blocksize_bits;
5118 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5119 struct ext4_iloc iloc;
5120 int offset; /* offset of xattr in inode */
5122 error = ext4_get_inode_loc(inode, &iloc);
5125 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5126 offset = EXT4_GOOD_OLD_INODE_SIZE +
5127 EXT4_I(inode)->i_extra_isize;
5129 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5130 flags |= FIEMAP_EXTENT_DATA_INLINE;
5132 } else { /* external block */
5133 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5134 length = inode->i_sb->s_blocksize;
5138 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5140 return (error < 0 ? error : 0);
5143 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5144 __u64 start, __u64 len)
5146 ext4_lblk_t start_blk;
5149 if (ext4_has_inline_data(inode)) {
5152 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5158 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5159 error = ext4_ext_precache(inode);
5164 /* fallback to generic here if not in extents fmt */
5165 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5166 return generic_block_fiemap(inode, fieinfo, start, len,
5169 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5172 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5173 error = ext4_xattr_fiemap(inode, fieinfo);
5175 ext4_lblk_t len_blks;
5178 start_blk = start >> inode->i_sb->s_blocksize_bits;
5179 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5180 if (last_blk >= EXT_MAX_BLOCKS)
5181 last_blk = EXT_MAX_BLOCKS-1;
5182 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5185 * Walk the extent tree gathering extent information
5186 * and pushing extents back to the user.
5188 error = ext4_fill_fiemap_extents(inode, start_blk,
5191 ext4_es_lru_add(inode);
5197 * Function to access the path buffer for marking it dirty.
5198 * It also checks if there are sufficient credits left in the journal handle
5202 ext4_access_path(handle_t *handle, struct inode *inode,
5203 struct ext4_ext_path *path)
5207 if (!ext4_handle_valid(handle))
5211 * Check if need to extend journal credits
5212 * 3 for leaf, sb, and inode plus 2 (bmap and group
5213 * descriptor) for each block group; assume two block
5216 if (handle->h_buffer_credits < 7) {
5217 credits = ext4_writepage_trans_blocks(inode);
5218 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5219 /* EAGAIN is success */
5220 if (err && err != -EAGAIN)
5224 err = ext4_ext_get_access(handle, inode, path);
5229 * ext4_ext_shift_path_extents:
5230 * Shift the extents of a path structure lying between path[depth].p_ext
5231 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5232 * from starting block for each extent.
5235 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5236 struct inode *inode, handle_t *handle,
5240 struct ext4_extent *ex_start, *ex_last;
5242 depth = path->p_depth;
5244 while (depth >= 0) {
5245 if (depth == path->p_depth) {
5246 ex_start = path[depth].p_ext;
5250 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5254 err = ext4_access_path(handle, inode, path + depth);
5258 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5261 *start = le32_to_cpu(ex_last->ee_block) +
5262 ext4_ext_get_actual_len(ex_last);
5264 while (ex_start <= ex_last) {
5265 le32_add_cpu(&ex_start->ee_block, -shift);
5266 /* Try to merge to the left. */
5268 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5269 ext4_ext_try_to_merge_right(inode,
5270 path, ex_start - 1))
5275 err = ext4_ext_dirty(handle, inode, path + depth);
5279 if (--depth < 0 || !update)
5283 /* Update index too */
5284 err = ext4_access_path(handle, inode, path + depth);
5288 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5289 err = ext4_ext_dirty(handle, inode, path + depth);
5293 /* we are done if current index is not a starting index */
5294 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5305 * ext4_ext_shift_extents:
5306 * All the extents which lies in the range from start to the last allocated
5307 * block for the file are shifted downwards by shift blocks.
5308 * On success, 0 is returned, error otherwise.
5311 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5312 ext4_lblk_t start, ext4_lblk_t shift)
5314 struct ext4_ext_path *path;
5316 struct ext4_extent *extent;
5317 ext4_lblk_t stop_block;
5318 ext4_lblk_t ex_start, ex_end;
5320 /* Let path point to the last extent */
5321 path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5323 return PTR_ERR(path);
5325 depth = path->p_depth;
5326 extent = path[depth].p_ext;
5328 ext4_ext_drop_refs(path);
5333 stop_block = le32_to_cpu(extent->ee_block) +
5334 ext4_ext_get_actual_len(extent);
5335 ext4_ext_drop_refs(path);
5338 /* Nothing to shift, if hole is at the end of file */
5339 if (start >= stop_block)
5343 * Don't start shifting extents until we make sure the hole is big
5344 * enough to accomodate the shift.
5346 path = ext4_ext_find_extent(inode, start - 1, NULL, 0);
5348 return PTR_ERR(path);
5349 depth = path->p_depth;
5350 extent = path[depth].p_ext;
5352 ex_start = le32_to_cpu(extent->ee_block);
5353 ex_end = le32_to_cpu(extent->ee_block) +
5354 ext4_ext_get_actual_len(extent);
5359 ext4_ext_drop_refs(path);
5362 if ((start == ex_start && shift > ex_start) ||
5363 (shift > start - ex_end))
5366 /* Its safe to start updating extents */
5367 while (start < stop_block) {
5368 path = ext4_ext_find_extent(inode, start, NULL, 0);
5370 return PTR_ERR(path);
5371 depth = path->p_depth;
5372 extent = path[depth].p_ext;
5374 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5375 (unsigned long) start);
5378 if (start > le32_to_cpu(extent->ee_block)) {
5379 /* Hole, move to the next extent */
5380 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5381 path[depth].p_ext++;
5383 start = ext4_ext_next_allocated_block(path);
5384 ext4_ext_drop_refs(path);
5389 ret = ext4_ext_shift_path_extents(path, shift, inode,
5391 ext4_ext_drop_refs(path);
5401 * ext4_collapse_range:
5402 * This implements the fallocate's collapse range functionality for ext4
5403 * Returns: 0 and non-zero on error.
5405 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5407 struct super_block *sb = inode->i_sb;
5408 ext4_lblk_t punch_start, punch_stop;
5410 unsigned int credits;
5411 loff_t new_size, ioffset;
5414 /* Collapse range works only on fs block size aligned offsets. */
5415 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5416 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5419 if (!S_ISREG(inode->i_mode))
5422 trace_ext4_collapse_range(inode, offset, len);
5424 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5425 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5427 /* Call ext4_force_commit to flush all data in case of data=journal. */
5428 if (ext4_should_journal_data(inode)) {
5429 ret = ext4_force_commit(inode->i_sb);
5435 * Need to round down offset to be aligned with page size boundary
5436 * for page size > block size.
5438 ioffset = round_down(offset, PAGE_SIZE);
5440 /* Write out all dirty pages */
5441 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5446 /* Take mutex lock */
5447 mutex_lock(&inode->i_mutex);
5450 * There is no need to overlap collapse range with EOF, in which case
5451 * it is effectively a truncate operation
5453 if (offset + len >= i_size_read(inode)) {
5458 /* Currently just for extent based files */
5459 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5464 truncate_pagecache(inode, ioffset);
5466 /* Wait for existing dio to complete */
5467 ext4_inode_block_unlocked_dio(inode);
5468 inode_dio_wait(inode);
5470 credits = ext4_writepage_trans_blocks(inode);
5471 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5472 if (IS_ERR(handle)) {
5473 ret = PTR_ERR(handle);
5477 down_write(&EXT4_I(inode)->i_data_sem);
5478 ext4_discard_preallocations(inode);
5480 ret = ext4_es_remove_extent(inode, punch_start,
5481 EXT_MAX_BLOCKS - punch_start);
5483 up_write(&EXT4_I(inode)->i_data_sem);
5487 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5489 up_write(&EXT4_I(inode)->i_data_sem);
5492 ext4_discard_preallocations(inode);
5494 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5495 punch_stop - punch_start);
5497 up_write(&EXT4_I(inode)->i_data_sem);
5501 new_size = i_size_read(inode) - len;
5502 i_size_write(inode, new_size);
5503 EXT4_I(inode)->i_disksize = new_size;
5505 up_write(&EXT4_I(inode)->i_data_sem);
5507 ext4_handle_sync(handle);
5508 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5509 ext4_mark_inode_dirty(handle, inode);
5512 ext4_journal_stop(handle);
5514 ext4_inode_resume_unlocked_dio(inode);
5516 mutex_unlock(&inode->i_mutex);
5521 * ext4_swap_extents - Swap extents between two inodes
5523 * @inode1: First inode
5524 * @inode2: Second inode
5525 * @lblk1: Start block for first inode
5526 * @lblk2: Start block for second inode
5527 * @count: Number of blocks to swap
5528 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5529 * @erp: Pointer to save error value
5531 * This helper routine does exactly what is promise "swap extents". All other
5532 * stuff such as page-cache locking consistency, bh mapping consistency or
5533 * extent's data copying must be performed by caller.
5535 * i_mutex is held for both inodes
5536 * i_data_sem is locked for write for both inodes
5538 * All pages from requested range are locked for both inodes
5541 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5542 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5543 ext4_lblk_t count, int unwritten, int *erp)
5545 struct ext4_ext_path *path1 = NULL;
5546 struct ext4_ext_path *path2 = NULL;
5547 int replaced_count = 0;
5549 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5550 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5551 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5552 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5554 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5557 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5562 struct ext4_extent *ex1, *ex2, tmp_ex;
5563 ext4_lblk_t e1_blk, e2_blk;
5564 int e1_len, e2_len, len;
5567 path1 = ext4_ext_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5568 if (unlikely(IS_ERR(path1))) {
5569 *erp = PTR_ERR(path1);
5575 path2 = ext4_ext_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5576 if (unlikely(IS_ERR(path2))) {
5577 *erp = PTR_ERR(path2);
5581 ex1 = path1[path1->p_depth].p_ext;
5582 ex2 = path2[path2->p_depth].p_ext;
5583 /* Do we have somthing to swap ? */
5584 if (unlikely(!ex2 || !ex1))
5587 e1_blk = le32_to_cpu(ex1->ee_block);
5588 e2_blk = le32_to_cpu(ex2->ee_block);
5589 e1_len = ext4_ext_get_actual_len(ex1);
5590 e2_len = ext4_ext_get_actual_len(ex2);
5593 if (!in_range(lblk1, e1_blk, e1_len) ||
5594 !in_range(lblk2, e2_blk, e2_len)) {
5595 ext4_lblk_t next1, next2;
5597 /* if hole after extent, then go to next extent */
5598 next1 = ext4_ext_next_allocated_block(path1);
5599 next2 = ext4_ext_next_allocated_block(path2);
5600 /* If hole before extent, then shift to that extent */
5605 /* Do we have something to swap */
5606 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5608 /* Move to the rightest boundary */
5609 len = next1 - lblk1;
5610 if (len < next2 - lblk2)
5611 len = next2 - lblk2;
5620 /* Prepare left boundary */
5621 if (e1_blk < lblk1) {
5623 *erp = ext4_force_split_extent_at(handle, inode1,
5628 if (e2_blk < lblk2) {
5630 *erp = ext4_force_split_extent_at(handle, inode2,
5635 /* ext4_split_extent_at() may retult in leaf extent split,
5636 * path must to be revalidated. */
5640 /* Prepare right boundary */
5642 if (len > e1_blk + e1_len - lblk1)
5643 len = e1_blk + e1_len - lblk1;
5644 if (len > e2_blk + e2_len - lblk2)
5645 len = e2_blk + e2_len - lblk2;
5647 if (len != e1_len) {
5649 *erp = ext4_force_split_extent_at(handle, inode1,
5650 path1, lblk1 + len, 0);
5654 if (len != e2_len) {
5656 *erp = ext4_force_split_extent_at(handle, inode2,
5657 path2, lblk2 + len, 0);
5661 /* ext4_split_extent_at() may retult in leaf extent split,
5662 * path must to be revalidated. */
5666 BUG_ON(e2_len != e1_len);
5667 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5670 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5674 /* Both extents are fully inside boundaries. Swap it now */
5676 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5677 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5678 ex1->ee_len = cpu_to_le16(e2_len);
5679 ex2->ee_len = cpu_to_le16(e1_len);
5681 ext4_ext_mark_unwritten(ex2);
5682 if (ext4_ext_is_unwritten(&tmp_ex))
5683 ext4_ext_mark_unwritten(ex1);
5685 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5686 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5687 *erp = ext4_ext_dirty(handle, inode2, path2 +
5691 *erp = ext4_ext_dirty(handle, inode1, path1 +
5694 * Looks scarry ah..? second inode already points to new blocks,
5695 * and it was successfully dirtied. But luckily error may happen
5696 * only due to journal error, so full transaction will be
5703 replaced_count += len;
5708 ext4_ext_drop_refs(path1);
5713 ext4_ext_drop_refs(path2);
5718 return replaced_count;