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 **orig_path, int flags)
860 struct ext4_extent_header *eh;
861 struct buffer_head *bh;
862 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
863 short int depth, i, ppos = 0;
864 short free_on_err = (flags & EXT4_EX_NOFREE_ON_ERR) == 0;
867 eh = ext_inode_hdr(inode);
868 depth = ext_depth(inode);
870 /* account possible depth increase */
872 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
875 return ERR_PTR(-ENOMEM);
882 /* walk through the tree */
884 ext_debug("depth %d: num %d, max %d\n",
885 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
887 ext4_ext_binsearch_idx(inode, path + ppos, block);
888 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
889 path[ppos].p_depth = i;
890 path[ppos].p_ext = NULL;
892 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
894 if (unlikely(IS_ERR(bh))) {
899 eh = ext_block_hdr(bh);
901 if (unlikely(ppos > depth)) {
903 EXT4_ERROR_INODE(inode,
904 "ppos %d > depth %d", ppos, depth);
908 path[ppos].p_bh = bh;
909 path[ppos].p_hdr = eh;
912 path[ppos].p_depth = i;
913 path[ppos].p_ext = NULL;
914 path[ppos].p_idx = NULL;
917 ext4_ext_binsearch(inode, path + ppos, block);
918 /* if not an empty leaf */
919 if (path[ppos].p_ext)
920 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
922 ext4_ext_show_path(inode, path);
927 ext4_ext_drop_refs(path);
937 * ext4_ext_insert_index:
938 * insert new index [@logical;@ptr] into the block at @curp;
939 * check where to insert: before @curp or after @curp
941 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
942 struct ext4_ext_path *curp,
943 int logical, ext4_fsblk_t ptr)
945 struct ext4_extent_idx *ix;
948 err = ext4_ext_get_access(handle, inode, curp);
952 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
953 EXT4_ERROR_INODE(inode,
954 "logical %d == ei_block %d!",
955 logical, le32_to_cpu(curp->p_idx->ei_block));
959 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
960 >= le16_to_cpu(curp->p_hdr->eh_max))) {
961 EXT4_ERROR_INODE(inode,
962 "eh_entries %d >= eh_max %d!",
963 le16_to_cpu(curp->p_hdr->eh_entries),
964 le16_to_cpu(curp->p_hdr->eh_max));
968 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
970 ext_debug("insert new index %d after: %llu\n", logical, ptr);
971 ix = curp->p_idx + 1;
974 ext_debug("insert new index %d before: %llu\n", logical, ptr);
978 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
981 ext_debug("insert new index %d: "
982 "move %d indices from 0x%p to 0x%p\n",
983 logical, len, ix, ix + 1);
984 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
987 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
988 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
992 ix->ei_block = cpu_to_le32(logical);
993 ext4_idx_store_pblock(ix, ptr);
994 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
996 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
997 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1001 err = ext4_ext_dirty(handle, inode, curp);
1002 ext4_std_error(inode->i_sb, err);
1009 * inserts new subtree into the path, using free index entry
1011 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1012 * - makes decision where to split
1013 * - moves remaining extents and index entries (right to the split point)
1014 * into the newly allocated blocks
1015 * - initializes subtree
1017 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1019 struct ext4_ext_path *path,
1020 struct ext4_extent *newext, int at)
1022 struct buffer_head *bh = NULL;
1023 int depth = ext_depth(inode);
1024 struct ext4_extent_header *neh;
1025 struct ext4_extent_idx *fidx;
1026 int i = at, k, m, a;
1027 ext4_fsblk_t newblock, oldblock;
1029 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1032 /* make decision: where to split? */
1033 /* FIXME: now decision is simplest: at current extent */
1035 /* if current leaf will be split, then we should use
1036 * border from split point */
1037 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1038 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1041 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1042 border = path[depth].p_ext[1].ee_block;
1043 ext_debug("leaf will be split."
1044 " next leaf starts at %d\n",
1045 le32_to_cpu(border));
1047 border = newext->ee_block;
1048 ext_debug("leaf will be added."
1049 " next leaf starts at %d\n",
1050 le32_to_cpu(border));
1054 * If error occurs, then we break processing
1055 * and mark filesystem read-only. index won't
1056 * be inserted and tree will be in consistent
1057 * state. Next mount will repair buffers too.
1061 * Get array to track all allocated blocks.
1062 * We need this to handle errors and free blocks
1065 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1069 /* allocate all needed blocks */
1070 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1071 for (a = 0; a < depth - at; a++) {
1072 newblock = ext4_ext_new_meta_block(handle, inode, path,
1073 newext, &err, flags);
1076 ablocks[a] = newblock;
1079 /* initialize new leaf */
1080 newblock = ablocks[--a];
1081 if (unlikely(newblock == 0)) {
1082 EXT4_ERROR_INODE(inode, "newblock == 0!");
1086 bh = sb_getblk(inode->i_sb, newblock);
1087 if (unlikely(!bh)) {
1093 err = ext4_journal_get_create_access(handle, bh);
1097 neh = ext_block_hdr(bh);
1098 neh->eh_entries = 0;
1099 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1100 neh->eh_magic = EXT4_EXT_MAGIC;
1103 /* move remainder of path[depth] to the new leaf */
1104 if (unlikely(path[depth].p_hdr->eh_entries !=
1105 path[depth].p_hdr->eh_max)) {
1106 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1107 path[depth].p_hdr->eh_entries,
1108 path[depth].p_hdr->eh_max);
1112 /* start copy from next extent */
1113 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1114 ext4_ext_show_move(inode, path, newblock, depth);
1116 struct ext4_extent *ex;
1117 ex = EXT_FIRST_EXTENT(neh);
1118 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1119 le16_add_cpu(&neh->eh_entries, m);
1122 ext4_extent_block_csum_set(inode, neh);
1123 set_buffer_uptodate(bh);
1126 err = ext4_handle_dirty_metadata(handle, inode, bh);
1132 /* correct old leaf */
1134 err = ext4_ext_get_access(handle, inode, path + depth);
1137 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1138 err = ext4_ext_dirty(handle, inode, path + depth);
1144 /* create intermediate indexes */
1146 if (unlikely(k < 0)) {
1147 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1152 ext_debug("create %d intermediate indices\n", k);
1153 /* insert new index into current index block */
1154 /* current depth stored in i var */
1157 oldblock = newblock;
1158 newblock = ablocks[--a];
1159 bh = sb_getblk(inode->i_sb, newblock);
1160 if (unlikely(!bh)) {
1166 err = ext4_journal_get_create_access(handle, bh);
1170 neh = ext_block_hdr(bh);
1171 neh->eh_entries = cpu_to_le16(1);
1172 neh->eh_magic = EXT4_EXT_MAGIC;
1173 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1174 neh->eh_depth = cpu_to_le16(depth - i);
1175 fidx = EXT_FIRST_INDEX(neh);
1176 fidx->ei_block = border;
1177 ext4_idx_store_pblock(fidx, oldblock);
1179 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1180 i, newblock, le32_to_cpu(border), oldblock);
1182 /* move remainder of path[i] to the new index block */
1183 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1184 EXT_LAST_INDEX(path[i].p_hdr))) {
1185 EXT4_ERROR_INODE(inode,
1186 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1187 le32_to_cpu(path[i].p_ext->ee_block));
1191 /* start copy indexes */
1192 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1193 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1194 EXT_MAX_INDEX(path[i].p_hdr));
1195 ext4_ext_show_move(inode, path, newblock, i);
1197 memmove(++fidx, path[i].p_idx,
1198 sizeof(struct ext4_extent_idx) * m);
1199 le16_add_cpu(&neh->eh_entries, m);
1201 ext4_extent_block_csum_set(inode, neh);
1202 set_buffer_uptodate(bh);
1205 err = ext4_handle_dirty_metadata(handle, inode, bh);
1211 /* correct old index */
1213 err = ext4_ext_get_access(handle, inode, path + i);
1216 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1217 err = ext4_ext_dirty(handle, inode, path + i);
1225 /* insert new index */
1226 err = ext4_ext_insert_index(handle, inode, path + at,
1227 le32_to_cpu(border), newblock);
1231 if (buffer_locked(bh))
1237 /* free all allocated blocks in error case */
1238 for (i = 0; i < depth; i++) {
1241 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1242 EXT4_FREE_BLOCKS_METADATA);
1251 * ext4_ext_grow_indepth:
1252 * implements tree growing procedure:
1253 * - allocates new block
1254 * - moves top-level data (index block or leaf) into the new block
1255 * - initializes new top-level, creating index that points to the
1256 * just created block
1258 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1260 struct ext4_extent *newext)
1262 struct ext4_extent_header *neh;
1263 struct buffer_head *bh;
1264 ext4_fsblk_t newblock;
1267 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1268 newext, &err, flags);
1272 bh = sb_getblk(inode->i_sb, newblock);
1277 err = ext4_journal_get_create_access(handle, bh);
1283 /* move top-level index/leaf into new block */
1284 memmove(bh->b_data, EXT4_I(inode)->i_data,
1285 sizeof(EXT4_I(inode)->i_data));
1287 /* set size of new block */
1288 neh = ext_block_hdr(bh);
1289 /* old root could have indexes or leaves
1290 * so calculate e_max right way */
1291 if (ext_depth(inode))
1292 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1294 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1295 neh->eh_magic = EXT4_EXT_MAGIC;
1296 ext4_extent_block_csum_set(inode, neh);
1297 set_buffer_uptodate(bh);
1300 err = ext4_handle_dirty_metadata(handle, inode, bh);
1304 /* Update top-level index: num,max,pointer */
1305 neh = ext_inode_hdr(inode);
1306 neh->eh_entries = cpu_to_le16(1);
1307 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1308 if (neh->eh_depth == 0) {
1309 /* Root extent block becomes index block */
1310 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1311 EXT_FIRST_INDEX(neh)->ei_block =
1312 EXT_FIRST_EXTENT(neh)->ee_block;
1314 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1315 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1316 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1317 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1319 le16_add_cpu(&neh->eh_depth, 1);
1320 ext4_mark_inode_dirty(handle, inode);
1328 * ext4_ext_create_new_leaf:
1329 * finds empty index and adds new leaf.
1330 * if no free index is found, then it requests in-depth growing.
1332 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1333 unsigned int mb_flags,
1334 unsigned int gb_flags,
1335 struct ext4_ext_path *path,
1336 struct ext4_extent *newext)
1338 struct ext4_ext_path *curp;
1339 int depth, i, err = 0;
1342 i = depth = ext_depth(inode);
1344 /* walk up to the tree and look for free index entry */
1345 curp = path + depth;
1346 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1351 /* we use already allocated block for index block,
1352 * so subsequent data blocks should be contiguous */
1353 if (EXT_HAS_FREE_INDEX(curp)) {
1354 /* if we found index with free entry, then use that
1355 * entry: create all needed subtree and add new leaf */
1356 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1361 ext4_ext_drop_refs(path);
1362 path = ext4_ext_find_extent(inode,
1363 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1364 &path, gb_flags | EXT4_EX_NOFREE_ON_ERR);
1366 err = PTR_ERR(path);
1368 /* tree is full, time to grow in depth */
1369 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1374 ext4_ext_drop_refs(path);
1375 path = ext4_ext_find_extent(inode,
1376 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1377 &path, gb_flags | EXT4_EX_NOFREE_ON_ERR);
1379 err = PTR_ERR(path);
1384 * only first (depth 0 -> 1) produces free space;
1385 * in all other cases we have to split the grown tree
1387 depth = ext_depth(inode);
1388 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1389 /* now we need to split */
1399 * search the closest allocated block to the left for *logical
1400 * and returns it at @logical + it's physical address at @phys
1401 * if *logical is the smallest allocated block, the function
1402 * returns 0 at @phys
1403 * return value contains 0 (success) or error code
1405 static int ext4_ext_search_left(struct inode *inode,
1406 struct ext4_ext_path *path,
1407 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1409 struct ext4_extent_idx *ix;
1410 struct ext4_extent *ex;
1413 if (unlikely(path == NULL)) {
1414 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1417 depth = path->p_depth;
1420 if (depth == 0 && path->p_ext == NULL)
1423 /* usually extent in the path covers blocks smaller
1424 * then *logical, but it can be that extent is the
1425 * first one in the file */
1427 ex = path[depth].p_ext;
1428 ee_len = ext4_ext_get_actual_len(ex);
1429 if (*logical < le32_to_cpu(ex->ee_block)) {
1430 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1431 EXT4_ERROR_INODE(inode,
1432 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1433 *logical, le32_to_cpu(ex->ee_block));
1436 while (--depth >= 0) {
1437 ix = path[depth].p_idx;
1438 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1439 EXT4_ERROR_INODE(inode,
1440 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1441 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1442 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1443 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1451 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1452 EXT4_ERROR_INODE(inode,
1453 "logical %d < ee_block %d + ee_len %d!",
1454 *logical, le32_to_cpu(ex->ee_block), ee_len);
1458 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1459 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1464 * search the closest allocated block to the right for *logical
1465 * and returns it at @logical + it's physical address at @phys
1466 * if *logical is the largest allocated block, the function
1467 * returns 0 at @phys
1468 * return value contains 0 (success) or error code
1470 static int ext4_ext_search_right(struct inode *inode,
1471 struct ext4_ext_path *path,
1472 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1473 struct ext4_extent **ret_ex)
1475 struct buffer_head *bh = NULL;
1476 struct ext4_extent_header *eh;
1477 struct ext4_extent_idx *ix;
1478 struct ext4_extent *ex;
1480 int depth; /* Note, NOT eh_depth; depth from top of tree */
1483 if (unlikely(path == NULL)) {
1484 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1487 depth = path->p_depth;
1490 if (depth == 0 && path->p_ext == NULL)
1493 /* usually extent in the path covers blocks smaller
1494 * then *logical, but it can be that extent is the
1495 * first one in the file */
1497 ex = path[depth].p_ext;
1498 ee_len = ext4_ext_get_actual_len(ex);
1499 if (*logical < le32_to_cpu(ex->ee_block)) {
1500 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1501 EXT4_ERROR_INODE(inode,
1502 "first_extent(path[%d].p_hdr) != ex",
1506 while (--depth >= 0) {
1507 ix = path[depth].p_idx;
1508 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1509 EXT4_ERROR_INODE(inode,
1510 "ix != EXT_FIRST_INDEX *logical %d!",
1518 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1519 EXT4_ERROR_INODE(inode,
1520 "logical %d < ee_block %d + ee_len %d!",
1521 *logical, le32_to_cpu(ex->ee_block), ee_len);
1525 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1526 /* next allocated block in this leaf */
1531 /* go up and search for index to the right */
1532 while (--depth >= 0) {
1533 ix = path[depth].p_idx;
1534 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1538 /* we've gone up to the root and found no index to the right */
1542 /* we've found index to the right, let's
1543 * follow it and find the closest allocated
1544 * block to the right */
1546 block = ext4_idx_pblock(ix);
1547 while (++depth < path->p_depth) {
1548 /* subtract from p_depth to get proper eh_depth */
1549 bh = read_extent_tree_block(inode, block,
1550 path->p_depth - depth, 0);
1553 eh = ext_block_hdr(bh);
1554 ix = EXT_FIRST_INDEX(eh);
1555 block = ext4_idx_pblock(ix);
1559 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1562 eh = ext_block_hdr(bh);
1563 ex = EXT_FIRST_EXTENT(eh);
1565 *logical = le32_to_cpu(ex->ee_block);
1566 *phys = ext4_ext_pblock(ex);
1574 * ext4_ext_next_allocated_block:
1575 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1576 * NOTE: it considers block number from index entry as
1577 * allocated block. Thus, index entries have to be consistent
1581 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1585 BUG_ON(path == NULL);
1586 depth = path->p_depth;
1588 if (depth == 0 && path->p_ext == NULL)
1589 return EXT_MAX_BLOCKS;
1591 while (depth >= 0) {
1592 if (depth == path->p_depth) {
1594 if (path[depth].p_ext &&
1595 path[depth].p_ext !=
1596 EXT_LAST_EXTENT(path[depth].p_hdr))
1597 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1600 if (path[depth].p_idx !=
1601 EXT_LAST_INDEX(path[depth].p_hdr))
1602 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1607 return EXT_MAX_BLOCKS;
1611 * ext4_ext_next_leaf_block:
1612 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1614 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1618 BUG_ON(path == NULL);
1619 depth = path->p_depth;
1621 /* zero-tree has no leaf blocks at all */
1623 return EXT_MAX_BLOCKS;
1625 /* go to index block */
1628 while (depth >= 0) {
1629 if (path[depth].p_idx !=
1630 EXT_LAST_INDEX(path[depth].p_hdr))
1631 return (ext4_lblk_t)
1632 le32_to_cpu(path[depth].p_idx[1].ei_block);
1636 return EXT_MAX_BLOCKS;
1640 * ext4_ext_correct_indexes:
1641 * if leaf gets modified and modified extent is first in the leaf,
1642 * then we have to correct all indexes above.
1643 * TODO: do we need to correct tree in all cases?
1645 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1646 struct ext4_ext_path *path)
1648 struct ext4_extent_header *eh;
1649 int depth = ext_depth(inode);
1650 struct ext4_extent *ex;
1654 eh = path[depth].p_hdr;
1655 ex = path[depth].p_ext;
1657 if (unlikely(ex == NULL || eh == NULL)) {
1658 EXT4_ERROR_INODE(inode,
1659 "ex %p == NULL or eh %p == NULL", ex, eh);
1664 /* there is no tree at all */
1668 if (ex != EXT_FIRST_EXTENT(eh)) {
1669 /* we correct tree if first leaf got modified only */
1674 * TODO: we need correction if border is smaller than current one
1677 border = path[depth].p_ext->ee_block;
1678 err = ext4_ext_get_access(handle, inode, path + k);
1681 path[k].p_idx->ei_block = border;
1682 err = ext4_ext_dirty(handle, inode, path + k);
1687 /* change all left-side indexes */
1688 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1690 err = ext4_ext_get_access(handle, inode, path + k);
1693 path[k].p_idx->ei_block = border;
1694 err = ext4_ext_dirty(handle, inode, path + k);
1703 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1704 struct ext4_extent *ex2)
1706 unsigned short ext1_ee_len, ext2_ee_len;
1709 * Make sure that both extents are initialized. We don't merge
1710 * unwritten extents so that we can be sure that end_io code has
1711 * the extent that was written properly split out and conversion to
1712 * initialized is trivial.
1714 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1717 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1718 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1720 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1721 le32_to_cpu(ex2->ee_block))
1725 * To allow future support for preallocated extents to be added
1726 * as an RO_COMPAT feature, refuse to merge to extents if
1727 * this can result in the top bit of ee_len being set.
1729 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1731 if (ext4_ext_is_unwritten(ex1) &&
1732 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1733 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1734 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1736 #ifdef AGGRESSIVE_TEST
1737 if (ext1_ee_len >= 4)
1741 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1747 * This function tries to merge the "ex" extent to the next extent in the tree.
1748 * It always tries to merge towards right. If you want to merge towards
1749 * left, pass "ex - 1" as argument instead of "ex".
1750 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1751 * 1 if they got merged.
1753 static int ext4_ext_try_to_merge_right(struct inode *inode,
1754 struct ext4_ext_path *path,
1755 struct ext4_extent *ex)
1757 struct ext4_extent_header *eh;
1758 unsigned int depth, len;
1759 int merge_done = 0, unwritten;
1761 depth = ext_depth(inode);
1762 BUG_ON(path[depth].p_hdr == NULL);
1763 eh = path[depth].p_hdr;
1765 while (ex < EXT_LAST_EXTENT(eh)) {
1766 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1768 /* merge with next extent! */
1769 unwritten = ext4_ext_is_unwritten(ex);
1770 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1771 + ext4_ext_get_actual_len(ex + 1));
1773 ext4_ext_mark_unwritten(ex);
1775 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1776 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1777 * sizeof(struct ext4_extent);
1778 memmove(ex + 1, ex + 2, len);
1780 le16_add_cpu(&eh->eh_entries, -1);
1782 WARN_ON(eh->eh_entries == 0);
1783 if (!eh->eh_entries)
1784 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1791 * This function does a very simple check to see if we can collapse
1792 * an extent tree with a single extent tree leaf block into the inode.
1794 static void ext4_ext_try_to_merge_up(handle_t *handle,
1795 struct inode *inode,
1796 struct ext4_ext_path *path)
1799 unsigned max_root = ext4_ext_space_root(inode, 0);
1802 if ((path[0].p_depth != 1) ||
1803 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1804 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1808 * We need to modify the block allocation bitmap and the block
1809 * group descriptor to release the extent tree block. If we
1810 * can't get the journal credits, give up.
1812 if (ext4_journal_extend(handle, 2))
1816 * Copy the extent data up to the inode
1818 blk = ext4_idx_pblock(path[0].p_idx);
1819 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1820 sizeof(struct ext4_extent_idx);
1821 s += sizeof(struct ext4_extent_header);
1823 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1824 path[0].p_depth = 0;
1825 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1826 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1827 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1829 brelse(path[1].p_bh);
1830 ext4_free_blocks(handle, inode, NULL, blk, 1,
1831 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1835 * This function tries to merge the @ex extent to neighbours in the tree.
1836 * return 1 if merge left else 0.
1838 static void ext4_ext_try_to_merge(handle_t *handle,
1839 struct inode *inode,
1840 struct ext4_ext_path *path,
1841 struct ext4_extent *ex) {
1842 struct ext4_extent_header *eh;
1846 depth = ext_depth(inode);
1847 BUG_ON(path[depth].p_hdr == NULL);
1848 eh = path[depth].p_hdr;
1850 if (ex > EXT_FIRST_EXTENT(eh))
1851 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1854 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1856 ext4_ext_try_to_merge_up(handle, inode, path);
1860 * check if a portion of the "newext" extent overlaps with an
1863 * If there is an overlap discovered, it updates the length of the newext
1864 * such that there will be no overlap, and then returns 1.
1865 * If there is no overlap found, it returns 0.
1867 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1868 struct inode *inode,
1869 struct ext4_extent *newext,
1870 struct ext4_ext_path *path)
1873 unsigned int depth, len1;
1874 unsigned int ret = 0;
1876 b1 = le32_to_cpu(newext->ee_block);
1877 len1 = ext4_ext_get_actual_len(newext);
1878 depth = ext_depth(inode);
1879 if (!path[depth].p_ext)
1881 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1884 * get the next allocated block if the extent in the path
1885 * is before the requested block(s)
1888 b2 = ext4_ext_next_allocated_block(path);
1889 if (b2 == EXT_MAX_BLOCKS)
1891 b2 = EXT4_LBLK_CMASK(sbi, b2);
1894 /* check for wrap through zero on extent logical start block*/
1895 if (b1 + len1 < b1) {
1896 len1 = EXT_MAX_BLOCKS - b1;
1897 newext->ee_len = cpu_to_le16(len1);
1901 /* check for overlap */
1902 if (b1 + len1 > b2) {
1903 newext->ee_len = cpu_to_le16(b2 - b1);
1911 * ext4_ext_insert_extent:
1912 * tries to merge requsted extent into the existing extent or
1913 * inserts requested extent as new one into the tree,
1914 * creating new leaf in the no-space case.
1916 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1917 struct ext4_ext_path *path,
1918 struct ext4_extent *newext, int gb_flags)
1920 struct ext4_extent_header *eh;
1921 struct ext4_extent *ex, *fex;
1922 struct ext4_extent *nearex; /* nearest extent */
1923 struct ext4_ext_path *npath = NULL;
1924 int depth, len, err;
1926 int mb_flags = 0, unwritten;
1928 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1929 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1932 depth = ext_depth(inode);
1933 ex = path[depth].p_ext;
1934 eh = path[depth].p_hdr;
1935 if (unlikely(path[depth].p_hdr == NULL)) {
1936 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1940 /* try to insert block into found extent and return */
1941 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1944 * Try to see whether we should rather test the extent on
1945 * right from ex, or from the left of ex. This is because
1946 * ext4_ext_find_extent() can return either extent on the
1947 * left, or on the right from the searched position. This
1948 * will make merging more effective.
1950 if (ex < EXT_LAST_EXTENT(eh) &&
1951 (le32_to_cpu(ex->ee_block) +
1952 ext4_ext_get_actual_len(ex) <
1953 le32_to_cpu(newext->ee_block))) {
1956 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1957 (le32_to_cpu(newext->ee_block) +
1958 ext4_ext_get_actual_len(newext) <
1959 le32_to_cpu(ex->ee_block)))
1962 /* Try to append newex to the ex */
1963 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1964 ext_debug("append [%d]%d block to %u:[%d]%d"
1966 ext4_ext_is_unwritten(newext),
1967 ext4_ext_get_actual_len(newext),
1968 le32_to_cpu(ex->ee_block),
1969 ext4_ext_is_unwritten(ex),
1970 ext4_ext_get_actual_len(ex),
1971 ext4_ext_pblock(ex));
1972 err = ext4_ext_get_access(handle, inode,
1976 unwritten = ext4_ext_is_unwritten(ex);
1977 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1978 + ext4_ext_get_actual_len(newext));
1980 ext4_ext_mark_unwritten(ex);
1981 eh = path[depth].p_hdr;
1987 /* Try to prepend newex to the ex */
1988 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1989 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1991 le32_to_cpu(newext->ee_block),
1992 ext4_ext_is_unwritten(newext),
1993 ext4_ext_get_actual_len(newext),
1994 le32_to_cpu(ex->ee_block),
1995 ext4_ext_is_unwritten(ex),
1996 ext4_ext_get_actual_len(ex),
1997 ext4_ext_pblock(ex));
1998 err = ext4_ext_get_access(handle, inode,
2003 unwritten = ext4_ext_is_unwritten(ex);
2004 ex->ee_block = newext->ee_block;
2005 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2006 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2007 + ext4_ext_get_actual_len(newext));
2009 ext4_ext_mark_unwritten(ex);
2010 eh = path[depth].p_hdr;
2016 depth = ext_depth(inode);
2017 eh = path[depth].p_hdr;
2018 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2021 /* probably next leaf has space for us? */
2022 fex = EXT_LAST_EXTENT(eh);
2023 next = EXT_MAX_BLOCKS;
2024 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2025 next = ext4_ext_next_leaf_block(path);
2026 if (next != EXT_MAX_BLOCKS) {
2027 ext_debug("next leaf block - %u\n", next);
2028 BUG_ON(npath != NULL);
2029 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2031 return PTR_ERR(npath);
2032 BUG_ON(npath->p_depth != path->p_depth);
2033 eh = npath[depth].p_hdr;
2034 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2035 ext_debug("next leaf isn't full(%d)\n",
2036 le16_to_cpu(eh->eh_entries));
2040 ext_debug("next leaf has no free space(%d,%d)\n",
2041 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2045 * There is no free space in the found leaf.
2046 * We're gonna add a new leaf in the tree.
2048 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2049 mb_flags = EXT4_MB_USE_RESERVED;
2050 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2054 depth = ext_depth(inode);
2055 eh = path[depth].p_hdr;
2058 nearex = path[depth].p_ext;
2060 err = ext4_ext_get_access(handle, inode, path + depth);
2065 /* there is no extent in this leaf, create first one */
2066 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2067 le32_to_cpu(newext->ee_block),
2068 ext4_ext_pblock(newext),
2069 ext4_ext_is_unwritten(newext),
2070 ext4_ext_get_actual_len(newext));
2071 nearex = EXT_FIRST_EXTENT(eh);
2073 if (le32_to_cpu(newext->ee_block)
2074 > le32_to_cpu(nearex->ee_block)) {
2076 ext_debug("insert %u:%llu:[%d]%d before: "
2078 le32_to_cpu(newext->ee_block),
2079 ext4_ext_pblock(newext),
2080 ext4_ext_is_unwritten(newext),
2081 ext4_ext_get_actual_len(newext),
2086 BUG_ON(newext->ee_block == nearex->ee_block);
2087 ext_debug("insert %u:%llu:[%d]%d after: "
2089 le32_to_cpu(newext->ee_block),
2090 ext4_ext_pblock(newext),
2091 ext4_ext_is_unwritten(newext),
2092 ext4_ext_get_actual_len(newext),
2095 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2097 ext_debug("insert %u:%llu:[%d]%d: "
2098 "move %d extents from 0x%p to 0x%p\n",
2099 le32_to_cpu(newext->ee_block),
2100 ext4_ext_pblock(newext),
2101 ext4_ext_is_unwritten(newext),
2102 ext4_ext_get_actual_len(newext),
2103 len, nearex, nearex + 1);
2104 memmove(nearex + 1, nearex,
2105 len * sizeof(struct ext4_extent));
2109 le16_add_cpu(&eh->eh_entries, 1);
2110 path[depth].p_ext = nearex;
2111 nearex->ee_block = newext->ee_block;
2112 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2113 nearex->ee_len = newext->ee_len;
2116 /* try to merge extents */
2117 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2118 ext4_ext_try_to_merge(handle, inode, path, nearex);
2121 /* time to correct all indexes above */
2122 err = ext4_ext_correct_indexes(handle, inode, path);
2126 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2130 ext4_ext_drop_refs(npath);
2136 static int ext4_fill_fiemap_extents(struct inode *inode,
2137 ext4_lblk_t block, ext4_lblk_t num,
2138 struct fiemap_extent_info *fieinfo)
2140 struct ext4_ext_path *path = NULL;
2141 struct ext4_extent *ex;
2142 struct extent_status es;
2143 ext4_lblk_t next, next_del, start = 0, end = 0;
2144 ext4_lblk_t last = block + num;
2145 int exists, depth = 0, err = 0;
2146 unsigned int flags = 0;
2147 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2149 while (block < last && block != EXT_MAX_BLOCKS) {
2151 /* find extent for this block */
2152 down_read(&EXT4_I(inode)->i_data_sem);
2154 if (path && ext_depth(inode) != depth) {
2155 /* depth was changed. we have to realloc path */
2160 path = ext4_ext_find_extent(inode, block, &path, 0);
2162 up_read(&EXT4_I(inode)->i_data_sem);
2163 err = PTR_ERR(path);
2168 depth = ext_depth(inode);
2169 if (unlikely(path[depth].p_hdr == NULL)) {
2170 up_read(&EXT4_I(inode)->i_data_sem);
2171 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2175 ex = path[depth].p_ext;
2176 next = ext4_ext_next_allocated_block(path);
2177 ext4_ext_drop_refs(path);
2182 /* there is no extent yet, so try to allocate
2183 * all requested space */
2186 } else if (le32_to_cpu(ex->ee_block) > block) {
2187 /* need to allocate space before found extent */
2189 end = le32_to_cpu(ex->ee_block);
2190 if (block + num < end)
2192 } else if (block >= le32_to_cpu(ex->ee_block)
2193 + ext4_ext_get_actual_len(ex)) {
2194 /* need to allocate space after found extent */
2199 } else if (block >= le32_to_cpu(ex->ee_block)) {
2201 * some part of requested space is covered
2205 end = le32_to_cpu(ex->ee_block)
2206 + ext4_ext_get_actual_len(ex);
2207 if (block + num < end)
2213 BUG_ON(end <= start);
2217 es.es_len = end - start;
2220 es.es_lblk = le32_to_cpu(ex->ee_block);
2221 es.es_len = ext4_ext_get_actual_len(ex);
2222 es.es_pblk = ext4_ext_pblock(ex);
2223 if (ext4_ext_is_unwritten(ex))
2224 flags |= FIEMAP_EXTENT_UNWRITTEN;
2228 * Find delayed extent and update es accordingly. We call
2229 * it even in !exists case to find out whether es is the
2230 * last existing extent or not.
2232 next_del = ext4_find_delayed_extent(inode, &es);
2233 if (!exists && next_del) {
2235 flags |= (FIEMAP_EXTENT_DELALLOC |
2236 FIEMAP_EXTENT_UNKNOWN);
2238 up_read(&EXT4_I(inode)->i_data_sem);
2240 if (unlikely(es.es_len == 0)) {
2241 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2247 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2248 * we need to check next == EXT_MAX_BLOCKS because it is
2249 * possible that an extent is with unwritten and delayed
2250 * status due to when an extent is delayed allocated and
2251 * is allocated by fallocate status tree will track both of
2254 * So we could return a unwritten and delayed extent, and
2255 * its block is equal to 'next'.
2257 if (next == next_del && next == EXT_MAX_BLOCKS) {
2258 flags |= FIEMAP_EXTENT_LAST;
2259 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2260 next != EXT_MAX_BLOCKS)) {
2261 EXT4_ERROR_INODE(inode,
2262 "next extent == %u, next "
2263 "delalloc extent = %u",
2271 err = fiemap_fill_next_extent(fieinfo,
2272 (__u64)es.es_lblk << blksize_bits,
2273 (__u64)es.es_pblk << blksize_bits,
2274 (__u64)es.es_len << blksize_bits,
2284 block = es.es_lblk + es.es_len;
2288 ext4_ext_drop_refs(path);
2296 * ext4_ext_put_gap_in_cache:
2297 * calculate boundaries of the gap that the requested block fits into
2298 * and cache this gap
2301 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2304 int depth = ext_depth(inode);
2305 unsigned long len = 0;
2306 ext4_lblk_t lblock = 0;
2307 struct ext4_extent *ex;
2309 ex = path[depth].p_ext;
2312 * there is no extent yet, so gap is [0;-] and we
2315 ext_debug("cache gap(whole file):");
2316 } else if (block < le32_to_cpu(ex->ee_block)) {
2318 len = le32_to_cpu(ex->ee_block) - block;
2319 ext_debug("cache gap(before): %u [%u:%u]",
2321 le32_to_cpu(ex->ee_block),
2322 ext4_ext_get_actual_len(ex));
2323 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2324 ext4_es_insert_extent(inode, lblock, len, ~0,
2325 EXTENT_STATUS_HOLE);
2326 } else if (block >= le32_to_cpu(ex->ee_block)
2327 + ext4_ext_get_actual_len(ex)) {
2329 lblock = le32_to_cpu(ex->ee_block)
2330 + ext4_ext_get_actual_len(ex);
2332 next = ext4_ext_next_allocated_block(path);
2333 ext_debug("cache gap(after): [%u:%u] %u",
2334 le32_to_cpu(ex->ee_block),
2335 ext4_ext_get_actual_len(ex),
2337 BUG_ON(next == lblock);
2338 len = next - lblock;
2339 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2340 ext4_es_insert_extent(inode, lblock, len, ~0,
2341 EXTENT_STATUS_HOLE);
2346 ext_debug(" -> %u:%lu\n", lblock, len);
2351 * removes index from the index block.
2353 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2354 struct ext4_ext_path *path, int depth)
2359 /* free index block */
2361 path = path + depth;
2362 leaf = ext4_idx_pblock(path->p_idx);
2363 if (unlikely(path->p_hdr->eh_entries == 0)) {
2364 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2367 err = ext4_ext_get_access(handle, inode, path);
2371 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2372 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2373 len *= sizeof(struct ext4_extent_idx);
2374 memmove(path->p_idx, path->p_idx + 1, len);
2377 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2378 err = ext4_ext_dirty(handle, inode, path);
2381 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2382 trace_ext4_ext_rm_idx(inode, leaf);
2384 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2385 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2387 while (--depth >= 0) {
2388 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2391 err = ext4_ext_get_access(handle, inode, path);
2394 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2395 err = ext4_ext_dirty(handle, inode, path);
2403 * ext4_ext_calc_credits_for_single_extent:
2404 * This routine returns max. credits that needed to insert an extent
2405 * to the extent tree.
2406 * When pass the actual path, the caller should calculate credits
2409 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2410 struct ext4_ext_path *path)
2413 int depth = ext_depth(inode);
2416 /* probably there is space in leaf? */
2417 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2418 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2421 * There are some space in the leaf tree, no
2422 * need to account for leaf block credit
2424 * bitmaps and block group descriptor blocks
2425 * and other metadata blocks still need to be
2428 /* 1 bitmap, 1 block group descriptor */
2429 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2434 return ext4_chunk_trans_blocks(inode, nrblocks);
2438 * How many index/leaf blocks need to change/allocate to add @extents extents?
2440 * If we add a single extent, then in the worse case, each tree level
2441 * index/leaf need to be changed in case of the tree split.
2443 * If more extents are inserted, they could cause the whole tree split more
2444 * than once, but this is really rare.
2446 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2451 /* If we are converting the inline data, only one is needed here. */
2452 if (ext4_has_inline_data(inode))
2455 depth = ext_depth(inode);
2465 static inline int get_default_free_blocks_flags(struct inode *inode)
2467 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2468 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2469 else if (ext4_should_journal_data(inode))
2470 return EXT4_FREE_BLOCKS_FORGET;
2474 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2475 struct ext4_extent *ex,
2476 long long *partial_cluster,
2477 ext4_lblk_t from, ext4_lblk_t to)
2479 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2480 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2482 int flags = get_default_free_blocks_flags(inode);
2485 * For bigalloc file systems, we never free a partial cluster
2486 * at the beginning of the extent. Instead, we make a note
2487 * that we tried freeing the cluster, and check to see if we
2488 * need to free it on a subsequent call to ext4_remove_blocks,
2489 * or at the end of the ext4_truncate() operation.
2491 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2493 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2495 * If we have a partial cluster, and it's different from the
2496 * cluster of the last block, we need to explicitly free the
2497 * partial cluster here.
2499 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2500 if ((*partial_cluster > 0) &&
2501 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2502 ext4_free_blocks(handle, inode, NULL,
2503 EXT4_C2B(sbi, *partial_cluster),
2504 sbi->s_cluster_ratio, flags);
2505 *partial_cluster = 0;
2508 #ifdef EXTENTS_STATS
2510 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2511 spin_lock(&sbi->s_ext_stats_lock);
2512 sbi->s_ext_blocks += ee_len;
2513 sbi->s_ext_extents++;
2514 if (ee_len < sbi->s_ext_min)
2515 sbi->s_ext_min = ee_len;
2516 if (ee_len > sbi->s_ext_max)
2517 sbi->s_ext_max = ee_len;
2518 if (ext_depth(inode) > sbi->s_depth_max)
2519 sbi->s_depth_max = ext_depth(inode);
2520 spin_unlock(&sbi->s_ext_stats_lock);
2523 if (from >= le32_to_cpu(ex->ee_block)
2524 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2527 unsigned int unaligned;
2529 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2530 pblk = ext4_ext_pblock(ex) + ee_len - num;
2532 * Usually we want to free partial cluster at the end of the
2533 * extent, except for the situation when the cluster is still
2534 * used by any other extent (partial_cluster is negative).
2536 if (*partial_cluster < 0 &&
2537 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2538 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2540 ext_debug("free last %u blocks starting %llu partial %lld\n",
2541 num, pblk, *partial_cluster);
2542 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2544 * If the block range to be freed didn't start at the
2545 * beginning of a cluster, and we removed the entire
2546 * extent and the cluster is not used by any other extent,
2547 * save the partial cluster here, since we might need to
2548 * delete if we determine that the truncate operation has
2549 * removed all of the blocks in the cluster.
2551 * On the other hand, if we did not manage to free the whole
2552 * extent, we have to mark the cluster as used (store negative
2553 * cluster number in partial_cluster).
2555 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2556 if (unaligned && (ee_len == num) &&
2557 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2558 *partial_cluster = EXT4_B2C(sbi, pblk);
2560 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2561 else if (*partial_cluster > 0)
2562 *partial_cluster = 0;
2564 ext4_error(sbi->s_sb, "strange request: removal(2) "
2565 "%u-%u from %u:%u\n",
2566 from, to, le32_to_cpu(ex->ee_block), ee_len);
2572 * ext4_ext_rm_leaf() Removes the extents associated with the
2573 * blocks appearing between "start" and "end", and splits the extents
2574 * if "start" and "end" appear in the same extent
2576 * @handle: The journal handle
2577 * @inode: The files inode
2578 * @path: The path to the leaf
2579 * @partial_cluster: The cluster which we'll have to free if all extents
2580 * has been released from it. It gets negative in case
2581 * that the cluster is still used.
2582 * @start: The first block to remove
2583 * @end: The last block to remove
2586 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2587 struct ext4_ext_path *path,
2588 long long *partial_cluster,
2589 ext4_lblk_t start, ext4_lblk_t end)
2591 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2592 int err = 0, correct_index = 0;
2593 int depth = ext_depth(inode), credits;
2594 struct ext4_extent_header *eh;
2597 ext4_lblk_t ex_ee_block;
2598 unsigned short ex_ee_len;
2599 unsigned unwritten = 0;
2600 struct ext4_extent *ex;
2603 /* the header must be checked already in ext4_ext_remove_space() */
2604 ext_debug("truncate since %u in leaf to %u\n", start, end);
2605 if (!path[depth].p_hdr)
2606 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2607 eh = path[depth].p_hdr;
2608 if (unlikely(path[depth].p_hdr == NULL)) {
2609 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2612 /* find where to start removing */
2613 ex = path[depth].p_ext;
2615 ex = EXT_LAST_EXTENT(eh);
2617 ex_ee_block = le32_to_cpu(ex->ee_block);
2618 ex_ee_len = ext4_ext_get_actual_len(ex);
2621 * If we're starting with an extent other than the last one in the
2622 * node, we need to see if it shares a cluster with the extent to
2623 * the right (towards the end of the file). If its leftmost cluster
2624 * is this extent's rightmost cluster and it is not cluster aligned,
2625 * we'll mark it as a partial that is not to be deallocated.
2628 if (ex != EXT_LAST_EXTENT(eh)) {
2629 ext4_fsblk_t current_pblk, right_pblk;
2630 long long current_cluster, right_cluster;
2632 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2633 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2634 right_pblk = ext4_ext_pblock(ex + 1);
2635 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2636 if (current_cluster == right_cluster &&
2637 EXT4_PBLK_COFF(sbi, right_pblk))
2638 *partial_cluster = -right_cluster;
2641 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2643 while (ex >= EXT_FIRST_EXTENT(eh) &&
2644 ex_ee_block + ex_ee_len > start) {
2646 if (ext4_ext_is_unwritten(ex))
2651 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2652 unwritten, ex_ee_len);
2653 path[depth].p_ext = ex;
2655 a = ex_ee_block > start ? ex_ee_block : start;
2656 b = ex_ee_block+ex_ee_len - 1 < end ?
2657 ex_ee_block+ex_ee_len - 1 : end;
2659 ext_debug(" border %u:%u\n", a, b);
2661 /* If this extent is beyond the end of the hole, skip it */
2662 if (end < ex_ee_block) {
2664 * We're going to skip this extent and move to another,
2665 * so if this extent is not cluster aligned we have
2666 * to mark the current cluster as used to avoid
2667 * accidentally freeing it later on
2669 pblk = ext4_ext_pblock(ex);
2670 if (EXT4_PBLK_COFF(sbi, pblk))
2672 -((long long)EXT4_B2C(sbi, pblk));
2674 ex_ee_block = le32_to_cpu(ex->ee_block);
2675 ex_ee_len = ext4_ext_get_actual_len(ex);
2677 } else if (b != ex_ee_block + ex_ee_len - 1) {
2678 EXT4_ERROR_INODE(inode,
2679 "can not handle truncate %u:%u "
2681 start, end, ex_ee_block,
2682 ex_ee_block + ex_ee_len - 1);
2685 } else if (a != ex_ee_block) {
2686 /* remove tail of the extent */
2687 num = a - ex_ee_block;
2689 /* remove whole extent: excellent! */
2693 * 3 for leaf, sb, and inode plus 2 (bmap and group
2694 * descriptor) for each block group; assume two block
2695 * groups plus ex_ee_len/blocks_per_block_group for
2698 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2699 if (ex == EXT_FIRST_EXTENT(eh)) {
2701 credits += (ext_depth(inode)) + 1;
2703 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2705 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2709 err = ext4_ext_get_access(handle, inode, path + depth);
2713 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2719 /* this extent is removed; mark slot entirely unused */
2720 ext4_ext_store_pblock(ex, 0);
2722 ex->ee_len = cpu_to_le16(num);
2724 * Do not mark unwritten if all the blocks in the
2725 * extent have been removed.
2727 if (unwritten && num)
2728 ext4_ext_mark_unwritten(ex);
2730 * If the extent was completely released,
2731 * we need to remove it from the leaf
2734 if (end != EXT_MAX_BLOCKS - 1) {
2736 * For hole punching, we need to scoot all the
2737 * extents up when an extent is removed so that
2738 * we dont have blank extents in the middle
2740 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2741 sizeof(struct ext4_extent));
2743 /* Now get rid of the one at the end */
2744 memset(EXT_LAST_EXTENT(eh), 0,
2745 sizeof(struct ext4_extent));
2747 le16_add_cpu(&eh->eh_entries, -1);
2748 } else if (*partial_cluster > 0)
2749 *partial_cluster = 0;
2751 err = ext4_ext_dirty(handle, inode, path + depth);
2755 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2756 ext4_ext_pblock(ex));
2758 ex_ee_block = le32_to_cpu(ex->ee_block);
2759 ex_ee_len = ext4_ext_get_actual_len(ex);
2762 if (correct_index && eh->eh_entries)
2763 err = ext4_ext_correct_indexes(handle, inode, path);
2766 * If there's a partial cluster and at least one extent remains in
2767 * the leaf, free the partial cluster if it isn't shared with the
2768 * current extent. If there's a partial cluster and no extents
2769 * remain in the leaf, it can't be freed here. It can only be
2770 * freed when it's possible to determine if it's not shared with
2771 * any other extent - when the next leaf is processed or when space
2772 * removal is complete.
2774 if (*partial_cluster > 0 && eh->eh_entries &&
2775 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2776 *partial_cluster)) {
2777 int flags = get_default_free_blocks_flags(inode);
2779 ext4_free_blocks(handle, inode, NULL,
2780 EXT4_C2B(sbi, *partial_cluster),
2781 sbi->s_cluster_ratio, flags);
2782 *partial_cluster = 0;
2785 /* if this leaf is free, then we should
2786 * remove it from index block above */
2787 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2788 err = ext4_ext_rm_idx(handle, inode, path, depth);
2795 * ext4_ext_more_to_rm:
2796 * returns 1 if current index has to be freed (even partial)
2799 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2801 BUG_ON(path->p_idx == NULL);
2803 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2807 * if truncate on deeper level happened, it wasn't partial,
2808 * so we have to consider current index for truncation
2810 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2815 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2818 struct super_block *sb = inode->i_sb;
2819 int depth = ext_depth(inode);
2820 struct ext4_ext_path *path = NULL;
2821 long long partial_cluster = 0;
2825 ext_debug("truncate since %u to %u\n", start, end);
2827 /* probably first extent we're gonna free will be last in block */
2828 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2830 return PTR_ERR(handle);
2833 trace_ext4_ext_remove_space(inode, start, end, depth);
2836 * Check if we are removing extents inside the extent tree. If that
2837 * is the case, we are going to punch a hole inside the extent tree
2838 * so we have to check whether we need to split the extent covering
2839 * the last block to remove so we can easily remove the part of it
2840 * in ext4_ext_rm_leaf().
2842 if (end < EXT_MAX_BLOCKS - 1) {
2843 struct ext4_extent *ex;
2844 ext4_lblk_t ee_block;
2846 /* find extent for this block */
2847 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2849 ext4_journal_stop(handle);
2850 return PTR_ERR(path);
2852 depth = ext_depth(inode);
2853 /* Leaf not may not exist only if inode has no blocks at all */
2854 ex = path[depth].p_ext;
2857 EXT4_ERROR_INODE(inode,
2858 "path[%d].p_hdr == NULL",
2865 ee_block = le32_to_cpu(ex->ee_block);
2868 * See if the last block is inside the extent, if so split
2869 * the extent at 'end' block so we can easily remove the
2870 * tail of the first part of the split extent in
2871 * ext4_ext_rm_leaf().
2873 if (end >= ee_block &&
2874 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2876 * Split the extent in two so that 'end' is the last
2877 * block in the first new extent. Also we should not
2878 * fail removing space due to ENOSPC so try to use
2879 * reserved block if that happens.
2881 err = ext4_force_split_extent_at(handle, inode, path,
2888 * We start scanning from right side, freeing all the blocks
2889 * after i_size and walking into the tree depth-wise.
2891 depth = ext_depth(inode);
2896 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2898 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2901 ext4_journal_stop(handle);
2904 path[0].p_depth = depth;
2905 path[0].p_hdr = ext_inode_hdr(inode);
2908 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2915 while (i >= 0 && err == 0) {
2917 /* this is leaf block */
2918 err = ext4_ext_rm_leaf(handle, inode, path,
2919 &partial_cluster, start,
2921 /* root level has p_bh == NULL, brelse() eats this */
2922 brelse(path[i].p_bh);
2923 path[i].p_bh = NULL;
2928 /* this is index block */
2929 if (!path[i].p_hdr) {
2930 ext_debug("initialize header\n");
2931 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2934 if (!path[i].p_idx) {
2935 /* this level hasn't been touched yet */
2936 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2937 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2938 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2940 le16_to_cpu(path[i].p_hdr->eh_entries));
2942 /* we were already here, see at next index */
2946 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2947 i, EXT_FIRST_INDEX(path[i].p_hdr),
2949 if (ext4_ext_more_to_rm(path + i)) {
2950 struct buffer_head *bh;
2951 /* go to the next level */
2952 ext_debug("move to level %d (block %llu)\n",
2953 i + 1, ext4_idx_pblock(path[i].p_idx));
2954 memset(path + i + 1, 0, sizeof(*path));
2955 bh = read_extent_tree_block(inode,
2956 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2959 /* should we reset i_size? */
2963 /* Yield here to deal with large extent trees.
2964 * Should be a no-op if we did IO above. */
2966 if (WARN_ON(i + 1 > depth)) {
2970 path[i + 1].p_bh = bh;
2972 /* save actual number of indexes since this
2973 * number is changed at the next iteration */
2974 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2977 /* we finished processing this index, go up */
2978 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2979 /* index is empty, remove it;
2980 * handle must be already prepared by the
2981 * truncatei_leaf() */
2982 err = ext4_ext_rm_idx(handle, inode, path, i);
2984 /* root level has p_bh == NULL, brelse() eats this */
2985 brelse(path[i].p_bh);
2986 path[i].p_bh = NULL;
2988 ext_debug("return to level %d\n", i);
2992 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2993 partial_cluster, path->p_hdr->eh_entries);
2995 /* If we still have something in the partial cluster and we have removed
2996 * even the first extent, then we should free the blocks in the partial
2997 * cluster as well. */
2998 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2999 int flags = get_default_free_blocks_flags(inode);
3001 ext4_free_blocks(handle, inode, NULL,
3002 EXT4_C2B(EXT4_SB(sb), partial_cluster),
3003 EXT4_SB(sb)->s_cluster_ratio, flags);
3004 partial_cluster = 0;
3007 /* TODO: flexible tree reduction should be here */
3008 if (path->p_hdr->eh_entries == 0) {
3010 * truncate to zero freed all the tree,
3011 * so we need to correct eh_depth
3013 err = ext4_ext_get_access(handle, inode, path);
3015 ext_inode_hdr(inode)->eh_depth = 0;
3016 ext_inode_hdr(inode)->eh_max =
3017 cpu_to_le16(ext4_ext_space_root(inode, 0));
3018 err = ext4_ext_dirty(handle, inode, path);
3022 ext4_ext_drop_refs(path);
3024 if (err == -EAGAIN) {
3028 ext4_journal_stop(handle);
3034 * called at mount time
3036 void ext4_ext_init(struct super_block *sb)
3039 * possible initialization would be here
3042 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3043 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3044 printk(KERN_INFO "EXT4-fs: file extents enabled"
3045 #ifdef AGGRESSIVE_TEST
3046 ", aggressive tests"
3048 #ifdef CHECK_BINSEARCH
3051 #ifdef EXTENTS_STATS
3056 #ifdef EXTENTS_STATS
3057 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3058 EXT4_SB(sb)->s_ext_min = 1 << 30;
3059 EXT4_SB(sb)->s_ext_max = 0;
3065 * called at umount time
3067 void ext4_ext_release(struct super_block *sb)
3069 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3072 #ifdef EXTENTS_STATS
3073 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3074 struct ext4_sb_info *sbi = EXT4_SB(sb);
3075 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3076 sbi->s_ext_blocks, sbi->s_ext_extents,
3077 sbi->s_ext_blocks / sbi->s_ext_extents);
3078 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3079 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3084 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3086 ext4_lblk_t ee_block;
3087 ext4_fsblk_t ee_pblock;
3088 unsigned int ee_len;
3090 ee_block = le32_to_cpu(ex->ee_block);
3091 ee_len = ext4_ext_get_actual_len(ex);
3092 ee_pblock = ext4_ext_pblock(ex);
3097 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3098 EXTENT_STATUS_WRITTEN);
3101 /* FIXME!! we need to try to merge to left or right after zero-out */
3102 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3104 ext4_fsblk_t ee_pblock;
3105 unsigned int ee_len;
3108 ee_len = ext4_ext_get_actual_len(ex);
3109 ee_pblock = ext4_ext_pblock(ex);
3111 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3119 * ext4_split_extent_at() splits an extent at given block.
3121 * @handle: the journal handle
3122 * @inode: the file inode
3123 * @path: the path to the extent
3124 * @split: the logical block where the extent is splitted.
3125 * @split_flags: indicates if the extent could be zeroout if split fails, and
3126 * the states(init or unwritten) of new extents.
3127 * @flags: flags used to insert new extent to extent tree.
3130 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3131 * of which are deterimined by split_flag.
3133 * There are two cases:
3134 * a> the extent are splitted into two extent.
3135 * b> split is not needed, and just mark the extent.
3137 * return 0 on success.
3139 static int ext4_split_extent_at(handle_t *handle,
3140 struct inode *inode,
3141 struct ext4_ext_path *path,
3146 ext4_fsblk_t newblock;
3147 ext4_lblk_t ee_block;
3148 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3149 struct ext4_extent *ex2 = NULL;
3150 unsigned int ee_len, depth;
3153 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3154 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3156 ext_debug("ext4_split_extents_at: inode %lu, logical"
3157 "block %llu\n", inode->i_ino, (unsigned long long)split);
3159 ext4_ext_show_leaf(inode, path);
3161 depth = ext_depth(inode);
3162 ex = path[depth].p_ext;
3163 ee_block = le32_to_cpu(ex->ee_block);
3164 ee_len = ext4_ext_get_actual_len(ex);
3165 newblock = split - ee_block + ext4_ext_pblock(ex);
3167 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3168 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3169 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3170 EXT4_EXT_MARK_UNWRIT1 |
3171 EXT4_EXT_MARK_UNWRIT2));
3173 err = ext4_ext_get_access(handle, inode, path + depth);
3177 if (split == ee_block) {
3179 * case b: block @split is the block that the extent begins with
3180 * then we just change the state of the extent, and splitting
3183 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3184 ext4_ext_mark_unwritten(ex);
3186 ext4_ext_mark_initialized(ex);
3188 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3189 ext4_ext_try_to_merge(handle, inode, path, ex);
3191 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3196 memcpy(&orig_ex, ex, sizeof(orig_ex));
3197 ex->ee_len = cpu_to_le16(split - ee_block);
3198 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3199 ext4_ext_mark_unwritten(ex);
3202 * path may lead to new leaf, not to original leaf any more
3203 * after ext4_ext_insert_extent() returns,
3205 err = ext4_ext_dirty(handle, inode, path + depth);
3207 goto fix_extent_len;
3210 ex2->ee_block = cpu_to_le32(split);
3211 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3212 ext4_ext_store_pblock(ex2, newblock);
3213 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3214 ext4_ext_mark_unwritten(ex2);
3216 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3217 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3218 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3219 if (split_flag & EXT4_EXT_DATA_VALID1) {
3220 err = ext4_ext_zeroout(inode, ex2);
3221 zero_ex.ee_block = ex2->ee_block;
3222 zero_ex.ee_len = cpu_to_le16(
3223 ext4_ext_get_actual_len(ex2));
3224 ext4_ext_store_pblock(&zero_ex,
3225 ext4_ext_pblock(ex2));
3227 err = ext4_ext_zeroout(inode, ex);
3228 zero_ex.ee_block = ex->ee_block;
3229 zero_ex.ee_len = cpu_to_le16(
3230 ext4_ext_get_actual_len(ex));
3231 ext4_ext_store_pblock(&zero_ex,
3232 ext4_ext_pblock(ex));
3235 err = ext4_ext_zeroout(inode, &orig_ex);
3236 zero_ex.ee_block = orig_ex.ee_block;
3237 zero_ex.ee_len = cpu_to_le16(
3238 ext4_ext_get_actual_len(&orig_ex));
3239 ext4_ext_store_pblock(&zero_ex,
3240 ext4_ext_pblock(&orig_ex));
3244 goto fix_extent_len;
3245 /* update the extent length and mark as initialized */
3246 ex->ee_len = cpu_to_le16(ee_len);
3247 ext4_ext_try_to_merge(handle, inode, path, ex);
3248 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3250 goto fix_extent_len;
3252 /* update extent status tree */
3253 err = ext4_zeroout_es(inode, &zero_ex);
3257 goto fix_extent_len;
3260 ext4_ext_show_leaf(inode, path);
3264 ex->ee_len = orig_ex.ee_len;
3265 ext4_ext_dirty(handle, inode, path + path->p_depth);
3270 * ext4_split_extents() splits an extent and mark extent which is covered
3271 * by @map as split_flags indicates
3273 * It may result in splitting the extent into multiple extents (up to three)
3274 * There are three possibilities:
3275 * a> There is no split required
3276 * b> Splits in two extents: Split is happening at either end of the extent
3277 * c> Splits in three extents: Somone is splitting in middle of the extent
3280 static int ext4_split_extent(handle_t *handle,
3281 struct inode *inode,
3282 struct ext4_ext_path *path,
3283 struct ext4_map_blocks *map,
3287 ext4_lblk_t ee_block;
3288 struct ext4_extent *ex;
3289 unsigned int ee_len, depth;
3292 int split_flag1, flags1;
3293 int allocated = map->m_len;
3295 depth = ext_depth(inode);
3296 ex = path[depth].p_ext;
3297 ee_block = le32_to_cpu(ex->ee_block);
3298 ee_len = ext4_ext_get_actual_len(ex);
3299 unwritten = ext4_ext_is_unwritten(ex);
3301 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3302 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3303 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3305 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3306 EXT4_EXT_MARK_UNWRIT2;
3307 if (split_flag & EXT4_EXT_DATA_VALID2)
3308 split_flag1 |= EXT4_EXT_DATA_VALID1;
3309 err = ext4_split_extent_at(handle, inode, path,
3310 map->m_lblk + map->m_len, split_flag1, flags1);
3314 allocated = ee_len - (map->m_lblk - ee_block);
3317 * Update path is required because previous ext4_split_extent_at() may
3318 * result in split of original leaf or extent zeroout.
3320 ext4_ext_drop_refs(path);
3321 path = ext4_ext_find_extent(inode, map->m_lblk, &path,
3322 EXT4_EX_NOFREE_ON_ERR);
3324 return PTR_ERR(path);
3325 depth = ext_depth(inode);
3326 ex = path[depth].p_ext;
3328 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3329 (unsigned long) map->m_lblk);
3332 unwritten = ext4_ext_is_unwritten(ex);
3335 if (map->m_lblk >= ee_block) {
3336 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3338 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3339 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3340 EXT4_EXT_MARK_UNWRIT2);
3342 err = ext4_split_extent_at(handle, inode, path,
3343 map->m_lblk, split_flag1, flags);
3348 ext4_ext_show_leaf(inode, path);
3350 return err ? err : allocated;
3354 * This function is called by ext4_ext_map_blocks() if someone tries to write
3355 * to an unwritten extent. It may result in splitting the unwritten
3356 * extent into multiple extents (up to three - one initialized and two
3358 * There are three possibilities:
3359 * a> There is no split required: Entire extent should be initialized
3360 * b> Splits in two extents: Write is happening at either end of the extent
3361 * c> Splits in three extents: Somone is writing in middle of the extent
3364 * - The extent pointed to by 'path' is unwritten.
3365 * - The extent pointed to by 'path' contains a superset
3366 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3368 * Post-conditions on success:
3369 * - the returned value is the number of blocks beyond map->l_lblk
3370 * that are allocated and initialized.
3371 * It is guaranteed to be >= map->m_len.
3373 static int ext4_ext_convert_to_initialized(handle_t *handle,
3374 struct inode *inode,
3375 struct ext4_map_blocks *map,
3376 struct ext4_ext_path *path,
3379 struct ext4_sb_info *sbi;
3380 struct ext4_extent_header *eh;
3381 struct ext4_map_blocks split_map;
3382 struct ext4_extent zero_ex;
3383 struct ext4_extent *ex, *abut_ex;
3384 ext4_lblk_t ee_block, eof_block;
3385 unsigned int ee_len, depth, map_len = map->m_len;
3386 int allocated = 0, max_zeroout = 0;
3390 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3391 "block %llu, max_blocks %u\n", inode->i_ino,
3392 (unsigned long long)map->m_lblk, map_len);
3394 sbi = EXT4_SB(inode->i_sb);
3395 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3396 inode->i_sb->s_blocksize_bits;
3397 if (eof_block < map->m_lblk + map_len)
3398 eof_block = map->m_lblk + map_len;
3400 depth = ext_depth(inode);
3401 eh = path[depth].p_hdr;
3402 ex = path[depth].p_ext;
3403 ee_block = le32_to_cpu(ex->ee_block);
3404 ee_len = ext4_ext_get_actual_len(ex);
3407 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3409 /* Pre-conditions */
3410 BUG_ON(!ext4_ext_is_unwritten(ex));
3411 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3414 * Attempt to transfer newly initialized blocks from the currently
3415 * unwritten extent to its neighbor. This is much cheaper
3416 * than an insertion followed by a merge as those involve costly
3417 * memmove() calls. Transferring to the left is the common case in
3418 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3419 * followed by append writes.
3421 * Limitations of the current logic:
3422 * - L1: we do not deal with writes covering the whole extent.
3423 * This would require removing the extent if the transfer
3425 * - L2: we only attempt to merge with an extent stored in the
3426 * same extent tree node.
3428 if ((map->m_lblk == ee_block) &&
3429 /* See if we can merge left */
3430 (map_len < ee_len) && /*L1*/
3431 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3432 ext4_lblk_t prev_lblk;
3433 ext4_fsblk_t prev_pblk, ee_pblk;
3434 unsigned int prev_len;
3437 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3438 prev_len = ext4_ext_get_actual_len(abut_ex);
3439 prev_pblk = ext4_ext_pblock(abut_ex);
3440 ee_pblk = ext4_ext_pblock(ex);
3443 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3444 * upon those conditions:
3445 * - C1: abut_ex is initialized,
3446 * - C2: abut_ex is logically abutting ex,
3447 * - C3: abut_ex is physically abutting ex,
3448 * - C4: abut_ex can receive the additional blocks without
3449 * overflowing the (initialized) length limit.
3451 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3452 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3453 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3454 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3455 err = ext4_ext_get_access(handle, inode, path + depth);
3459 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3462 /* Shift the start of ex by 'map_len' blocks */
3463 ex->ee_block = cpu_to_le32(ee_block + map_len);
3464 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3465 ex->ee_len = cpu_to_le16(ee_len - map_len);
3466 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3468 /* Extend abut_ex by 'map_len' blocks */
3469 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3471 /* Result: number of initialized blocks past m_lblk */
3472 allocated = map_len;
3474 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3475 (map_len < ee_len) && /*L1*/
3476 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3477 /* See if we can merge right */
3478 ext4_lblk_t next_lblk;
3479 ext4_fsblk_t next_pblk, ee_pblk;
3480 unsigned int next_len;
3483 next_lblk = le32_to_cpu(abut_ex->ee_block);
3484 next_len = ext4_ext_get_actual_len(abut_ex);
3485 next_pblk = ext4_ext_pblock(abut_ex);
3486 ee_pblk = ext4_ext_pblock(ex);
3489 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3490 * upon those conditions:
3491 * - C1: abut_ex is initialized,
3492 * - C2: abut_ex is logically abutting ex,
3493 * - C3: abut_ex is physically abutting ex,
3494 * - C4: abut_ex can receive the additional blocks without
3495 * overflowing the (initialized) length limit.
3497 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3498 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3499 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3500 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3501 err = ext4_ext_get_access(handle, inode, path + depth);
3505 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3508 /* Shift the start of abut_ex by 'map_len' blocks */
3509 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3510 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3511 ex->ee_len = cpu_to_le16(ee_len - map_len);
3512 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3514 /* Extend abut_ex by 'map_len' blocks */
3515 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3517 /* Result: number of initialized blocks past m_lblk */
3518 allocated = map_len;
3522 /* Mark the block containing both extents as dirty */
3523 ext4_ext_dirty(handle, inode, path + depth);
3525 /* Update path to point to the right extent */
3526 path[depth].p_ext = abut_ex;
3529 allocated = ee_len - (map->m_lblk - ee_block);
3531 WARN_ON(map->m_lblk < ee_block);
3533 * It is safe to convert extent to initialized via explicit
3534 * zeroout only if extent is fully inside i_size or new_size.
3536 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3538 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3539 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3540 (inode->i_sb->s_blocksize_bits - 10);
3542 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3543 if (max_zeroout && (ee_len <= max_zeroout)) {
3544 err = ext4_ext_zeroout(inode, ex);
3547 zero_ex.ee_block = ex->ee_block;
3548 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3549 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3551 err = ext4_ext_get_access(handle, inode, path + depth);
3554 ext4_ext_mark_initialized(ex);
3555 ext4_ext_try_to_merge(handle, inode, path, ex);
3556 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3562 * 1. split the extent into three extents.
3563 * 2. split the extent into two extents, zeroout the first half.
3564 * 3. split the extent into two extents, zeroout the second half.
3565 * 4. split the extent into two extents with out zeroout.
3567 split_map.m_lblk = map->m_lblk;
3568 split_map.m_len = map->m_len;
3570 if (max_zeroout && (allocated > map->m_len)) {
3571 if (allocated <= max_zeroout) {
3574 cpu_to_le32(map->m_lblk);
3575 zero_ex.ee_len = cpu_to_le16(allocated);
3576 ext4_ext_store_pblock(&zero_ex,
3577 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3578 err = ext4_ext_zeroout(inode, &zero_ex);
3581 split_map.m_lblk = map->m_lblk;
3582 split_map.m_len = allocated;
3583 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3585 if (map->m_lblk != ee_block) {
3586 zero_ex.ee_block = ex->ee_block;
3587 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3589 ext4_ext_store_pblock(&zero_ex,
3590 ext4_ext_pblock(ex));
3591 err = ext4_ext_zeroout(inode, &zero_ex);
3596 split_map.m_lblk = ee_block;
3597 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3598 allocated = map->m_len;
3602 allocated = ext4_split_extent(handle, inode, path,
3603 &split_map, split_flag, flags);
3608 /* If we have gotten a failure, don't zero out status tree */
3610 err = ext4_zeroout_es(inode, &zero_ex);
3611 return err ? err : allocated;
3615 * This function is called by ext4_ext_map_blocks() from
3616 * ext4_get_blocks_dio_write() when DIO to write
3617 * to an unwritten extent.
3619 * Writing to an unwritten extent may result in splitting the unwritten
3620 * extent into multiple initialized/unwritten extents (up to three)
3621 * There are three possibilities:
3622 * a> There is no split required: Entire extent should be unwritten
3623 * b> Splits in two extents: Write is happening at either end of the extent
3624 * c> Splits in three extents: Somone is writing in middle of the extent
3626 * This works the same way in the case of initialized -> unwritten conversion.
3628 * One of more index blocks maybe needed if the extent tree grow after
3629 * the unwritten extent split. To prevent ENOSPC occur at the IO
3630 * complete, we need to split the unwritten extent before DIO submit
3631 * the IO. The unwritten extent called at this time will be split
3632 * into three unwritten extent(at most). After IO complete, the part
3633 * being filled will be convert to initialized by the end_io callback function
3634 * via ext4_convert_unwritten_extents().
3636 * Returns the size of unwritten extent to be written on success.
3638 static int ext4_split_convert_extents(handle_t *handle,
3639 struct inode *inode,
3640 struct ext4_map_blocks *map,
3641 struct ext4_ext_path *path,
3644 ext4_lblk_t eof_block;
3645 ext4_lblk_t ee_block;
3646 struct ext4_extent *ex;
3647 unsigned int ee_len;
3648 int split_flag = 0, depth;
3650 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3651 __func__, inode->i_ino,
3652 (unsigned long long)map->m_lblk, map->m_len);
3654 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3655 inode->i_sb->s_blocksize_bits;
3656 if (eof_block < map->m_lblk + map->m_len)
3657 eof_block = map->m_lblk + map->m_len;
3659 * It is safe to convert extent to initialized via explicit
3660 * zeroout only if extent is fully insde i_size or new_size.
3662 depth = ext_depth(inode);
3663 ex = path[depth].p_ext;
3664 ee_block = le32_to_cpu(ex->ee_block);
3665 ee_len = ext4_ext_get_actual_len(ex);
3667 /* Convert to unwritten */
3668 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3669 split_flag |= EXT4_EXT_DATA_VALID1;
3670 /* Convert to initialized */
3671 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3672 split_flag |= ee_block + ee_len <= eof_block ?
3673 EXT4_EXT_MAY_ZEROOUT : 0;
3674 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3676 flags |= EXT4_GET_BLOCKS_PRE_IO;
3677 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3680 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3681 struct inode *inode,
3682 struct ext4_map_blocks *map,
3683 struct ext4_ext_path *path)
3685 struct ext4_extent *ex;
3686 ext4_lblk_t ee_block;
3687 unsigned int ee_len;
3691 depth = ext_depth(inode);
3692 ex = path[depth].p_ext;
3693 ee_block = le32_to_cpu(ex->ee_block);
3694 ee_len = ext4_ext_get_actual_len(ex);
3696 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3697 "block %llu, max_blocks %u\n", inode->i_ino,
3698 (unsigned long long)ee_block, ee_len);
3700 /* If extent is larger than requested it is a clear sign that we still
3701 * have some extent state machine issues left. So extent_split is still
3703 * TODO: Once all related issues will be fixed this situation should be
3706 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3708 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3709 " len %u; IO logical block %llu, len %u\n",
3710 inode->i_ino, (unsigned long long)ee_block, ee_len,
3711 (unsigned long long)map->m_lblk, map->m_len);
3713 err = ext4_split_convert_extents(handle, inode, map, path,
3714 EXT4_GET_BLOCKS_CONVERT);
3717 ext4_ext_drop_refs(path);
3718 path = ext4_ext_find_extent(inode, map->m_lblk, &path,
3719 EXT4_EX_NOFREE_ON_ERR);
3721 err = PTR_ERR(path);
3724 depth = ext_depth(inode);
3725 ex = path[depth].p_ext;
3728 err = ext4_ext_get_access(handle, inode, path + depth);
3731 /* first mark the extent as initialized */
3732 ext4_ext_mark_initialized(ex);
3734 /* note: ext4_ext_correct_indexes() isn't needed here because
3735 * borders are not changed
3737 ext4_ext_try_to_merge(handle, inode, path, ex);
3739 /* Mark modified extent as dirty */
3740 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3742 ext4_ext_show_leaf(inode, path);
3746 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3747 sector_t block, int count)
3750 for (i = 0; i < count; i++)
3751 unmap_underlying_metadata(bdev, block + i);
3755 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3757 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3759 struct ext4_ext_path *path,
3763 struct ext4_extent_header *eh;
3764 struct ext4_extent *last_ex;
3766 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3769 depth = ext_depth(inode);
3770 eh = path[depth].p_hdr;
3773 * We're going to remove EOFBLOCKS_FL entirely in future so we
3774 * do not care for this case anymore. Simply remove the flag
3775 * if there are no extents.
3777 if (unlikely(!eh->eh_entries))
3779 last_ex = EXT_LAST_EXTENT(eh);
3781 * We should clear the EOFBLOCKS_FL flag if we are writing the
3782 * last block in the last extent in the file. We test this by
3783 * first checking to see if the caller to
3784 * ext4_ext_get_blocks() was interested in the last block (or
3785 * a block beyond the last block) in the current extent. If
3786 * this turns out to be false, we can bail out from this
3787 * function immediately.
3789 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3790 ext4_ext_get_actual_len(last_ex))
3793 * If the caller does appear to be planning to write at or
3794 * beyond the end of the current extent, we then test to see
3795 * if the current extent is the last extent in the file, by
3796 * checking to make sure it was reached via the rightmost node
3797 * at each level of the tree.
3799 for (i = depth-1; i >= 0; i--)
3800 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3803 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3804 return ext4_mark_inode_dirty(handle, inode);
3808 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3810 * Return 1 if there is a delalloc block in the range, otherwise 0.
3812 int ext4_find_delalloc_range(struct inode *inode,
3813 ext4_lblk_t lblk_start,
3814 ext4_lblk_t lblk_end)
3816 struct extent_status es;
3818 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3820 return 0; /* there is no delay extent in this tree */
3821 else if (es.es_lblk <= lblk_start &&
3822 lblk_start < es.es_lblk + es.es_len)
3824 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3830 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3832 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3833 ext4_lblk_t lblk_start, lblk_end;
3834 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3835 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3837 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3841 * Determines how many complete clusters (out of those specified by the 'map')
3842 * are under delalloc and were reserved quota for.
3843 * This function is called when we are writing out the blocks that were
3844 * originally written with their allocation delayed, but then the space was
3845 * allocated using fallocate() before the delayed allocation could be resolved.
3846 * The cases to look for are:
3847 * ('=' indicated delayed allocated blocks
3848 * '-' indicates non-delayed allocated blocks)
3849 * (a) partial clusters towards beginning and/or end outside of allocated range
3850 * are not delalloc'ed.
3852 * |----c---=|====c====|====c====|===-c----|
3853 * |++++++ allocated ++++++|
3854 * ==> 4 complete clusters in above example
3856 * (b) partial cluster (outside of allocated range) towards either end is
3857 * marked for delayed allocation. In this case, we will exclude that
3860 * |----====c========|========c========|
3861 * |++++++ allocated ++++++|
3862 * ==> 1 complete clusters in above example
3865 * |================c================|
3866 * |++++++ allocated ++++++|
3867 * ==> 0 complete clusters in above example
3869 * The ext4_da_update_reserve_space will be called only if we
3870 * determine here that there were some "entire" clusters that span
3871 * this 'allocated' range.
3872 * In the non-bigalloc case, this function will just end up returning num_blks
3873 * without ever calling ext4_find_delalloc_range.
3876 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3877 unsigned int num_blks)
3879 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3880 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3881 ext4_lblk_t lblk_from, lblk_to, c_offset;
3882 unsigned int allocated_clusters = 0;
3884 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3885 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3887 /* max possible clusters for this allocation */
3888 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3890 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3892 /* Check towards left side */
3893 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3895 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3896 lblk_to = lblk_from + c_offset - 1;
3898 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3899 allocated_clusters--;
3902 /* Now check towards right. */
3903 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3904 if (allocated_clusters && c_offset) {
3905 lblk_from = lblk_start + num_blks;
3906 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3908 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3909 allocated_clusters--;
3912 return allocated_clusters;
3916 convert_initialized_extent(handle_t *handle, struct inode *inode,
3917 struct ext4_map_blocks *map,
3918 struct ext4_ext_path *path, int flags,
3919 unsigned int allocated, ext4_fsblk_t newblock)
3921 struct ext4_extent *ex;
3922 ext4_lblk_t ee_block;
3923 unsigned int ee_len;
3928 * Make sure that the extent is no bigger than we support with
3931 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3932 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3934 depth = ext_depth(inode);
3935 ex = path[depth].p_ext;
3936 ee_block = le32_to_cpu(ex->ee_block);
3937 ee_len = ext4_ext_get_actual_len(ex);
3939 ext_debug("%s: inode %lu, logical"
3940 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3941 (unsigned long long)ee_block, ee_len);
3943 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3944 err = ext4_split_convert_extents(handle, inode, map, path,
3945 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3948 ext4_ext_drop_refs(path);
3949 path = ext4_ext_find_extent(inode, map->m_lblk, &path,
3950 EXT4_EX_NOFREE_ON_ERR);
3952 return PTR_ERR(path);
3953 depth = ext_depth(inode);
3954 ex = path[depth].p_ext;
3956 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3957 (unsigned long) map->m_lblk);
3962 err = ext4_ext_get_access(handle, inode, path + depth);
3965 /* first mark the extent as unwritten */
3966 ext4_ext_mark_unwritten(ex);
3968 /* note: ext4_ext_correct_indexes() isn't needed here because
3969 * borders are not changed
3971 ext4_ext_try_to_merge(handle, inode, path, ex);
3973 /* Mark modified extent as dirty */
3974 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3977 ext4_ext_show_leaf(inode, path);
3979 ext4_update_inode_fsync_trans(handle, inode, 1);
3980 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3983 map->m_flags |= EXT4_MAP_UNWRITTEN;
3984 if (allocated > map->m_len)
3985 allocated = map->m_len;
3986 map->m_len = allocated;
3991 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3992 struct ext4_map_blocks *map,
3993 struct ext4_ext_path *path, int flags,
3994 unsigned int allocated, ext4_fsblk_t newblock)
3998 ext4_io_end_t *io = ext4_inode_aio(inode);
4000 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4001 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4002 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4004 ext4_ext_show_leaf(inode, path);
4007 * When writing into unwritten space, we should not fail to
4008 * allocate metadata blocks for the new extent block if needed.
4010 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4012 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4013 allocated, newblock);
4015 /* get_block() before submit the IO, split the extent */
4016 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4017 ret = ext4_split_convert_extents(handle, inode, map,
4018 path, flags | EXT4_GET_BLOCKS_CONVERT);
4022 * Flag the inode(non aio case) or end_io struct (aio case)
4023 * that this IO needs to conversion to written when IO is
4027 ext4_set_io_unwritten_flag(inode, io);
4029 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4030 map->m_flags |= EXT4_MAP_UNWRITTEN;
4033 /* IO end_io complete, convert the filled extent to written */
4034 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4035 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4038 ext4_update_inode_fsync_trans(handle, inode, 1);
4039 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4043 map->m_flags |= EXT4_MAP_MAPPED;
4044 map->m_pblk = newblock;
4045 if (allocated > map->m_len)
4046 allocated = map->m_len;
4047 map->m_len = allocated;
4050 /* buffered IO case */
4052 * repeat fallocate creation request
4053 * we already have an unwritten extent
4055 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4056 map->m_flags |= EXT4_MAP_UNWRITTEN;
4060 /* buffered READ or buffered write_begin() lookup */
4061 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4063 * We have blocks reserved already. We
4064 * return allocated blocks so that delalloc
4065 * won't do block reservation for us. But
4066 * the buffer head will be unmapped so that
4067 * a read from the block returns 0s.
4069 map->m_flags |= EXT4_MAP_UNWRITTEN;
4073 /* buffered write, writepage time, convert*/
4074 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
4076 ext4_update_inode_fsync_trans(handle, inode, 1);
4083 map->m_flags |= EXT4_MAP_NEW;
4085 * if we allocated more blocks than requested
4086 * we need to make sure we unmap the extra block
4087 * allocated. The actual needed block will get
4088 * unmapped later when we find the buffer_head marked
4091 if (allocated > map->m_len) {
4092 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4093 newblock + map->m_len,
4094 allocated - map->m_len);
4095 allocated = map->m_len;
4097 map->m_len = allocated;
4100 * If we have done fallocate with the offset that is already
4101 * delayed allocated, we would have block reservation
4102 * and quota reservation done in the delayed write path.
4103 * But fallocate would have already updated quota and block
4104 * count for this offset. So cancel these reservation
4106 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4107 unsigned int reserved_clusters;
4108 reserved_clusters = get_reserved_cluster_alloc(inode,
4109 map->m_lblk, map->m_len);
4110 if (reserved_clusters)
4111 ext4_da_update_reserve_space(inode,
4117 map->m_flags |= EXT4_MAP_MAPPED;
4118 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4119 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4125 if (allocated > map->m_len)
4126 allocated = map->m_len;
4127 ext4_ext_show_leaf(inode, path);
4128 map->m_pblk = newblock;
4129 map->m_len = allocated;
4131 return err ? err : allocated;
4135 * get_implied_cluster_alloc - check to see if the requested
4136 * allocation (in the map structure) overlaps with a cluster already
4137 * allocated in an extent.
4138 * @sb The filesystem superblock structure
4139 * @map The requested lblk->pblk mapping
4140 * @ex The extent structure which might contain an implied
4141 * cluster allocation
4143 * This function is called by ext4_ext_map_blocks() after we failed to
4144 * find blocks that were already in the inode's extent tree. Hence,
4145 * we know that the beginning of the requested region cannot overlap
4146 * the extent from the inode's extent tree. There are three cases we
4147 * want to catch. The first is this case:
4149 * |--- cluster # N--|
4150 * |--- extent ---| |---- requested region ---|
4153 * The second case that we need to test for is this one:
4155 * |--------- cluster # N ----------------|
4156 * |--- requested region --| |------- extent ----|
4157 * |=======================|
4159 * The third case is when the requested region lies between two extents
4160 * within the same cluster:
4161 * |------------- cluster # N-------------|
4162 * |----- ex -----| |---- ex_right ----|
4163 * |------ requested region ------|
4164 * |================|
4166 * In each of the above cases, we need to set the map->m_pblk and
4167 * map->m_len so it corresponds to the return the extent labelled as
4168 * "|====|" from cluster #N, since it is already in use for data in
4169 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4170 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4171 * as a new "allocated" block region. Otherwise, we will return 0 and
4172 * ext4_ext_map_blocks() will then allocate one or more new clusters
4173 * by calling ext4_mb_new_blocks().
4175 static int get_implied_cluster_alloc(struct super_block *sb,
4176 struct ext4_map_blocks *map,
4177 struct ext4_extent *ex,
4178 struct ext4_ext_path *path)
4180 struct ext4_sb_info *sbi = EXT4_SB(sb);
4181 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4182 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4183 ext4_lblk_t rr_cluster_start;
4184 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4185 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4186 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4188 /* The extent passed in that we are trying to match */
4189 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4190 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4192 /* The requested region passed into ext4_map_blocks() */
4193 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4195 if ((rr_cluster_start == ex_cluster_end) ||
4196 (rr_cluster_start == ex_cluster_start)) {
4197 if (rr_cluster_start == ex_cluster_end)
4198 ee_start += ee_len - 1;
4199 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4200 map->m_len = min(map->m_len,
4201 (unsigned) sbi->s_cluster_ratio - c_offset);
4203 * Check for and handle this case:
4205 * |--------- cluster # N-------------|
4206 * |------- extent ----|
4207 * |--- requested region ---|
4211 if (map->m_lblk < ee_block)
4212 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4215 * Check for the case where there is already another allocated
4216 * block to the right of 'ex' but before the end of the cluster.
4218 * |------------- cluster # N-------------|
4219 * |----- ex -----| |---- ex_right ----|
4220 * |------ requested region ------|
4221 * |================|
4223 if (map->m_lblk > ee_block) {
4224 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4225 map->m_len = min(map->m_len, next - map->m_lblk);
4228 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4232 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4238 * Block allocation/map/preallocation routine for extents based files
4241 * Need to be called with
4242 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4243 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4245 * return > 0, number of of blocks already mapped/allocated
4246 * if create == 0 and these are pre-allocated blocks
4247 * buffer head is unmapped
4248 * otherwise blocks are mapped
4250 * return = 0, if plain look up failed (blocks have not been allocated)
4251 * buffer head is unmapped
4253 * return < 0, error case.
4255 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4256 struct ext4_map_blocks *map, int flags)
4258 struct ext4_ext_path *path = NULL;
4259 struct ext4_extent newex, *ex, *ex2;
4260 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4261 ext4_fsblk_t newblock = 0;
4262 int free_on_err = 0, err = 0, depth, ret;
4263 unsigned int allocated = 0, offset = 0;
4264 unsigned int allocated_clusters = 0;
4265 struct ext4_allocation_request ar;
4266 ext4_io_end_t *io = ext4_inode_aio(inode);
4267 ext4_lblk_t cluster_offset;
4268 int set_unwritten = 0;
4270 ext_debug("blocks %u/%u requested for inode %lu\n",
4271 map->m_lblk, map->m_len, inode->i_ino);
4272 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4274 /* find extent for this block */
4275 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4277 err = PTR_ERR(path);
4282 depth = ext_depth(inode);
4285 * consistent leaf must not be empty;
4286 * this situation is possible, though, _during_ tree modification;
4287 * this is why assert can't be put in ext4_ext_find_extent()
4289 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4290 EXT4_ERROR_INODE(inode, "bad extent address "
4291 "lblock: %lu, depth: %d pblock %lld",
4292 (unsigned long) map->m_lblk, depth,
4293 path[depth].p_block);
4298 ex = path[depth].p_ext;
4300 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4301 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4302 unsigned short ee_len;
4306 * unwritten extents are treated as holes, except that
4307 * we split out initialized portions during a write.
4309 ee_len = ext4_ext_get_actual_len(ex);
4311 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4313 /* if found extent covers block, simply return it */
4314 if (in_range(map->m_lblk, ee_block, ee_len)) {
4315 newblock = map->m_lblk - ee_block + ee_start;
4316 /* number of remaining blocks in the extent */
4317 allocated = ee_len - (map->m_lblk - ee_block);
4318 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4319 ee_block, ee_len, newblock);
4322 * If the extent is initialized check whether the
4323 * caller wants to convert it to unwritten.
4325 if ((!ext4_ext_is_unwritten(ex)) &&
4326 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4327 allocated = convert_initialized_extent(
4328 handle, inode, map, path, flags,
4329 allocated, newblock);
4331 } else if (!ext4_ext_is_unwritten(ex))
4334 ret = ext4_ext_handle_unwritten_extents(
4335 handle, inode, map, path, flags,
4336 allocated, newblock);
4345 if ((sbi->s_cluster_ratio > 1) &&
4346 ext4_find_delalloc_cluster(inode, map->m_lblk))
4347 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4350 * requested block isn't allocated yet;
4351 * we couldn't try to create block if create flag is zero
4353 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4355 * put just found gap into cache to speed up
4356 * subsequent requests
4358 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4359 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4364 * Okay, we need to do block allocation.
4366 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4367 newex.ee_block = cpu_to_le32(map->m_lblk);
4368 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4371 * If we are doing bigalloc, check to see if the extent returned
4372 * by ext4_ext_find_extent() implies a cluster we can use.
4374 if (cluster_offset && ex &&
4375 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4376 ar.len = allocated = map->m_len;
4377 newblock = map->m_pblk;
4378 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4379 goto got_allocated_blocks;
4382 /* find neighbour allocated blocks */
4383 ar.lleft = map->m_lblk;
4384 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4387 ar.lright = map->m_lblk;
4389 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4393 /* Check if the extent after searching to the right implies a
4394 * cluster we can use. */
4395 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4396 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4397 ar.len = allocated = map->m_len;
4398 newblock = map->m_pblk;
4399 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4400 goto got_allocated_blocks;
4404 * See if request is beyond maximum number of blocks we can have in
4405 * a single extent. For an initialized extent this limit is
4406 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4407 * EXT_UNWRITTEN_MAX_LEN.
4409 if (map->m_len > EXT_INIT_MAX_LEN &&
4410 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4411 map->m_len = EXT_INIT_MAX_LEN;
4412 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4413 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4414 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4416 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4417 newex.ee_len = cpu_to_le16(map->m_len);
4418 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4420 allocated = ext4_ext_get_actual_len(&newex);
4422 allocated = map->m_len;
4424 /* allocate new block */
4426 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4427 ar.logical = map->m_lblk;
4429 * We calculate the offset from the beginning of the cluster
4430 * for the logical block number, since when we allocate a
4431 * physical cluster, the physical block should start at the
4432 * same offset from the beginning of the cluster. This is
4433 * needed so that future calls to get_implied_cluster_alloc()
4436 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4437 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4439 ar.logical -= offset;
4440 if (S_ISREG(inode->i_mode))
4441 ar.flags = EXT4_MB_HINT_DATA;
4443 /* disable in-core preallocation for non-regular files */
4445 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4446 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4447 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4450 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4451 ar.goal, newblock, allocated);
4453 allocated_clusters = ar.len;
4454 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4455 if (ar.len > allocated)
4458 got_allocated_blocks:
4459 /* try to insert new extent into found leaf and return */
4460 ext4_ext_store_pblock(&newex, newblock + offset);
4461 newex.ee_len = cpu_to_le16(ar.len);
4462 /* Mark unwritten */
4463 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4464 ext4_ext_mark_unwritten(&newex);
4465 map->m_flags |= EXT4_MAP_UNWRITTEN;
4467 * io_end structure was created for every IO write to an
4468 * unwritten extent. To avoid unnecessary conversion,
4469 * here we flag the IO that really needs the conversion.
4470 * For non asycn direct IO case, flag the inode state
4471 * that we need to perform conversion when IO is done.
4473 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4478 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4479 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4482 err = ext4_ext_insert_extent(handle, inode, path,
4485 if (!err && set_unwritten) {
4487 ext4_set_io_unwritten_flag(inode, io);
4489 ext4_set_inode_state(inode,
4490 EXT4_STATE_DIO_UNWRITTEN);
4493 if (err && free_on_err) {
4494 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4495 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4496 /* free data blocks we just allocated */
4497 /* not a good idea to call discard here directly,
4498 * but otherwise we'd need to call it every free() */
4499 ext4_discard_preallocations(inode);
4500 ext4_free_blocks(handle, inode, NULL, newblock,
4501 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4505 /* previous routine could use block we allocated */
4506 newblock = ext4_ext_pblock(&newex);
4507 allocated = ext4_ext_get_actual_len(&newex);
4508 if (allocated > map->m_len)
4509 allocated = map->m_len;
4510 map->m_flags |= EXT4_MAP_NEW;
4513 * Update reserved blocks/metadata blocks after successful
4514 * block allocation which had been deferred till now.
4516 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4517 unsigned int reserved_clusters;
4519 * Check how many clusters we had reserved this allocated range
4521 reserved_clusters = get_reserved_cluster_alloc(inode,
4522 map->m_lblk, allocated);
4523 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4524 if (reserved_clusters) {
4526 * We have clusters reserved for this range.
4527 * But since we are not doing actual allocation
4528 * and are simply using blocks from previously
4529 * allocated cluster, we should release the
4530 * reservation and not claim quota.
4532 ext4_da_update_reserve_space(inode,
4533 reserved_clusters, 0);
4536 BUG_ON(allocated_clusters < reserved_clusters);
4537 if (reserved_clusters < allocated_clusters) {
4538 struct ext4_inode_info *ei = EXT4_I(inode);
4539 int reservation = allocated_clusters -
4542 * It seems we claimed few clusters outside of
4543 * the range of this allocation. We should give
4544 * it back to the reservation pool. This can
4545 * happen in the following case:
4547 * * Suppose s_cluster_ratio is 4 (i.e., each
4548 * cluster has 4 blocks. Thus, the clusters
4549 * are [0-3],[4-7],[8-11]...
4550 * * First comes delayed allocation write for
4551 * logical blocks 10 & 11. Since there were no
4552 * previous delayed allocated blocks in the
4553 * range [8-11], we would reserve 1 cluster
4555 * * Next comes write for logical blocks 3 to 8.
4556 * In this case, we will reserve 2 clusters
4557 * (for [0-3] and [4-7]; and not for [8-11] as
4558 * that range has a delayed allocated blocks.
4559 * Thus total reserved clusters now becomes 3.
4560 * * Now, during the delayed allocation writeout
4561 * time, we will first write blocks [3-8] and
4562 * allocate 3 clusters for writing these
4563 * blocks. Also, we would claim all these
4564 * three clusters above.
4565 * * Now when we come here to writeout the
4566 * blocks [10-11], we would expect to claim
4567 * the reservation of 1 cluster we had made
4568 * (and we would claim it since there are no
4569 * more delayed allocated blocks in the range
4570 * [8-11]. But our reserved cluster count had
4571 * already gone to 0.
4573 * Thus, at the step 4 above when we determine
4574 * that there are still some unwritten delayed
4575 * allocated blocks outside of our current
4576 * block range, we should increment the
4577 * reserved clusters count so that when the
4578 * remaining blocks finally gets written, we
4581 dquot_reserve_block(inode,
4582 EXT4_C2B(sbi, reservation));
4583 spin_lock(&ei->i_block_reservation_lock);
4584 ei->i_reserved_data_blocks += reservation;
4585 spin_unlock(&ei->i_block_reservation_lock);
4588 * We will claim quota for all newly allocated blocks.
4589 * We're updating the reserved space *after* the
4590 * correction above so we do not accidentally free
4591 * all the metadata reservation because we might
4592 * actually need it later on.
4594 ext4_da_update_reserve_space(inode, allocated_clusters,
4600 * Cache the extent and update transaction to commit on fdatasync only
4601 * when it is _not_ an unwritten extent.
4603 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4604 ext4_update_inode_fsync_trans(handle, inode, 1);
4606 ext4_update_inode_fsync_trans(handle, inode, 0);
4608 if (allocated > map->m_len)
4609 allocated = map->m_len;
4610 ext4_ext_show_leaf(inode, path);
4611 map->m_flags |= EXT4_MAP_MAPPED;
4612 map->m_pblk = newblock;
4613 map->m_len = allocated;
4616 ext4_ext_drop_refs(path);
4620 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4621 err ? err : allocated);
4622 ext4_es_lru_add(inode);
4623 return err ? err : allocated;
4626 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4628 struct super_block *sb = inode->i_sb;
4629 ext4_lblk_t last_block;
4633 * TODO: optimization is possible here.
4634 * Probably we need not scan at all,
4635 * because page truncation is enough.
4638 /* we have to know where to truncate from in crash case */
4639 EXT4_I(inode)->i_disksize = inode->i_size;
4640 ext4_mark_inode_dirty(handle, inode);
4642 last_block = (inode->i_size + sb->s_blocksize - 1)
4643 >> EXT4_BLOCK_SIZE_BITS(sb);
4645 err = ext4_es_remove_extent(inode, last_block,
4646 EXT_MAX_BLOCKS - last_block);
4647 if (err == -ENOMEM) {
4649 congestion_wait(BLK_RW_ASYNC, HZ/50);
4653 ext4_std_error(inode->i_sb, err);
4656 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4657 ext4_std_error(inode->i_sb, err);
4660 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4661 ext4_lblk_t len, loff_t new_size,
4662 int flags, int mode)
4664 struct inode *inode = file_inode(file);
4669 struct ext4_map_blocks map;
4670 unsigned int credits;
4673 map.m_lblk = offset;
4676 * Don't normalize the request if it can fit in one extent so
4677 * that it doesn't get unnecessarily split into multiple
4680 if (len <= EXT_UNWRITTEN_MAX_LEN)
4681 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4684 * credits to insert 1 extent into extent tree
4686 credits = ext4_chunk_trans_blocks(inode, len);
4689 while (ret >= 0 && len) {
4690 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4692 if (IS_ERR(handle)) {
4693 ret = PTR_ERR(handle);
4696 ret = ext4_map_blocks(handle, inode, &map, flags);
4698 ext4_debug("inode #%lu: block %u: len %u: "
4699 "ext4_ext_map_blocks returned %d",
4700 inode->i_ino, map.m_lblk,
4702 ext4_mark_inode_dirty(handle, inode);
4703 ret2 = ext4_journal_stop(handle);
4707 map.m_len = len = len - ret;
4708 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4709 inode->i_ctime = ext4_current_time(inode);
4711 if (epos > new_size)
4713 if (ext4_update_inode_size(inode, epos) & 0x1)
4714 inode->i_mtime = inode->i_ctime;
4716 if (epos > inode->i_size)
4717 ext4_set_inode_flag(inode,
4718 EXT4_INODE_EOFBLOCKS);
4720 ext4_mark_inode_dirty(handle, inode);
4721 ret2 = ext4_journal_stop(handle);
4725 if (ret == -ENOSPC &&
4726 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4731 return ret > 0 ? ret2 : ret;
4734 static long ext4_zero_range(struct file *file, loff_t offset,
4735 loff_t len, int mode)
4737 struct inode *inode = file_inode(file);
4738 handle_t *handle = NULL;
4739 unsigned int max_blocks;
4740 loff_t new_size = 0;
4744 int partial_begin, partial_end;
4747 struct address_space *mapping = inode->i_mapping;
4748 unsigned int blkbits = inode->i_blkbits;
4750 trace_ext4_zero_range(inode, offset, len, mode);
4752 if (!S_ISREG(inode->i_mode))
4755 /* Call ext4_force_commit to flush all data in case of data=journal. */
4756 if (ext4_should_journal_data(inode)) {
4757 ret = ext4_force_commit(inode->i_sb);
4763 * Write out all dirty pages to avoid race conditions
4764 * Then release them.
4766 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4767 ret = filemap_write_and_wait_range(mapping, offset,
4774 * Round up offset. This is not fallocate, we neet to zero out
4775 * blocks, so convert interior block aligned part of the range to
4776 * unwritten and possibly manually zero out unaligned parts of the
4779 start = round_up(offset, 1 << blkbits);
4780 end = round_down((offset + len), 1 << blkbits);
4782 if (start < offset || end > offset + len)
4784 partial_begin = offset & ((1 << blkbits) - 1);
4785 partial_end = (offset + len) & ((1 << blkbits) - 1);
4787 lblk = start >> blkbits;
4788 max_blocks = (end >> blkbits);
4789 if (max_blocks < lblk)
4794 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4795 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4797 if (mode & FALLOC_FL_KEEP_SIZE)
4798 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4800 mutex_lock(&inode->i_mutex);
4803 * Indirect files do not support unwritten extnets
4805 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4810 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4811 offset + len > i_size_read(inode)) {
4812 new_size = offset + len;
4813 ret = inode_newsize_ok(inode, new_size);
4817 * If we have a partial block after EOF we have to allocate
4824 if (max_blocks > 0) {
4826 /* Now release the pages and zero block aligned part of pages*/
4827 truncate_pagecache_range(inode, start, end - 1);
4828 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4830 /* Wait all existing dio workers, newcomers will block on i_mutex */
4831 ext4_inode_block_unlocked_dio(inode);
4832 inode_dio_wait(inode);
4834 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4839 * Remove entire range from the extent status tree.
4841 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4842 * NOT sufficient. I'm not sure why this is the case,
4843 * but let's be conservative and remove the extent
4844 * status tree for the entire inode. There should be
4845 * no outstanding delalloc extents thanks to the
4846 * filemap_write_and_wait_range() call above.
4848 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4852 if (!partial_begin && !partial_end)
4856 * In worst case we have to writeout two nonadjacent unwritten
4857 * blocks and update the inode
4859 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4860 if (ext4_should_journal_data(inode))
4862 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4863 if (IS_ERR(handle)) {
4864 ret = PTR_ERR(handle);
4865 ext4_std_error(inode->i_sb, ret);
4869 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4871 ext4_update_inode_size(inode, new_size);
4874 * Mark that we allocate beyond EOF so the subsequent truncate
4875 * can proceed even if the new size is the same as i_size.
4877 if ((offset + len) > i_size_read(inode))
4878 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4880 ext4_mark_inode_dirty(handle, inode);
4882 /* Zero out partial block at the edges of the range */
4883 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4885 if (file->f_flags & O_SYNC)
4886 ext4_handle_sync(handle);
4888 ext4_journal_stop(handle);
4890 ext4_inode_resume_unlocked_dio(inode);
4892 mutex_unlock(&inode->i_mutex);
4897 * preallocate space for a file. This implements ext4's fallocate file
4898 * operation, which gets called from sys_fallocate system call.
4899 * For block-mapped files, posix_fallocate should fall back to the method
4900 * of writing zeroes to the required new blocks (the same behavior which is
4901 * expected for file systems which do not support fallocate() system call).
4903 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4905 struct inode *inode = file_inode(file);
4906 loff_t new_size = 0;
4907 unsigned int max_blocks;
4911 unsigned int blkbits = inode->i_blkbits;
4913 /* Return error if mode is not supported */
4914 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4915 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4918 if (mode & FALLOC_FL_PUNCH_HOLE)
4919 return ext4_punch_hole(inode, offset, len);
4921 ret = ext4_convert_inline_data(inode);
4926 * currently supporting (pre)allocate mode for extent-based
4929 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4932 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4933 return ext4_collapse_range(inode, offset, len);
4935 if (mode & FALLOC_FL_ZERO_RANGE)
4936 return ext4_zero_range(file, offset, len, mode);
4938 trace_ext4_fallocate_enter(inode, offset, len, mode);
4939 lblk = offset >> blkbits;
4941 * We can't just convert len to max_blocks because
4942 * If blocksize = 4096 offset = 3072 and len = 2048
4944 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4947 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4948 if (mode & FALLOC_FL_KEEP_SIZE)
4949 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4951 mutex_lock(&inode->i_mutex);
4953 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4954 offset + len > i_size_read(inode)) {
4955 new_size = offset + len;
4956 ret = inode_newsize_ok(inode, new_size);
4961 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4966 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4967 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4968 EXT4_I(inode)->i_sync_tid);
4971 mutex_unlock(&inode->i_mutex);
4972 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4977 * This function convert a range of blocks to written extents
4978 * The caller of this function will pass the start offset and the size.
4979 * all unwritten extents within this range will be converted to
4982 * This function is called from the direct IO end io call back
4983 * function, to convert the fallocated extents after IO is completed.
4984 * Returns 0 on success.
4986 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4987 loff_t offset, ssize_t len)
4989 unsigned int max_blocks;
4992 struct ext4_map_blocks map;
4993 unsigned int credits, blkbits = inode->i_blkbits;
4995 map.m_lblk = offset >> blkbits;
4997 * We can't just convert len to max_blocks because
4998 * If blocksize = 4096 offset = 3072 and len = 2048
5000 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5003 * This is somewhat ugly but the idea is clear: When transaction is
5004 * reserved, everything goes into it. Otherwise we rather start several
5005 * smaller transactions for conversion of each extent separately.
5008 handle = ext4_journal_start_reserved(handle,
5009 EXT4_HT_EXT_CONVERT);
5011 return PTR_ERR(handle);
5015 * credits to insert 1 extent into extent tree
5017 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5019 while (ret >= 0 && ret < max_blocks) {
5021 map.m_len = (max_blocks -= ret);
5023 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5025 if (IS_ERR(handle)) {
5026 ret = PTR_ERR(handle);
5030 ret = ext4_map_blocks(handle, inode, &map,
5031 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5033 ext4_warning(inode->i_sb,
5034 "inode #%lu: block %u: len %u: "
5035 "ext4_ext_map_blocks returned %d",
5036 inode->i_ino, map.m_lblk,
5038 ext4_mark_inode_dirty(handle, inode);
5040 ret2 = ext4_journal_stop(handle);
5041 if (ret <= 0 || ret2)
5045 ret2 = ext4_journal_stop(handle);
5046 return ret > 0 ? ret2 : ret;
5050 * If newes is not existing extent (newes->ec_pblk equals zero) find
5051 * delayed extent at start of newes and update newes accordingly and
5052 * return start of the next delayed extent.
5054 * If newes is existing extent (newes->ec_pblk is not equal zero)
5055 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5056 * extent found. Leave newes unmodified.
5058 static int ext4_find_delayed_extent(struct inode *inode,
5059 struct extent_status *newes)
5061 struct extent_status es;
5062 ext4_lblk_t block, next_del;
5064 if (newes->es_pblk == 0) {
5065 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5066 newes->es_lblk + newes->es_len - 1, &es);
5069 * No extent in extent-tree contains block @newes->es_pblk,
5070 * then the block may stay in 1)a hole or 2)delayed-extent.
5076 if (es.es_lblk > newes->es_lblk) {
5078 newes->es_len = min(es.es_lblk - newes->es_lblk,
5083 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5086 block = newes->es_lblk + newes->es_len;
5087 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5089 next_del = EXT_MAX_BLOCKS;
5091 next_del = es.es_lblk;
5095 /* fiemap flags we can handle specified here */
5096 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5098 static int ext4_xattr_fiemap(struct inode *inode,
5099 struct fiemap_extent_info *fieinfo)
5103 __u32 flags = FIEMAP_EXTENT_LAST;
5104 int blockbits = inode->i_sb->s_blocksize_bits;
5108 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5109 struct ext4_iloc iloc;
5110 int offset; /* offset of xattr in inode */
5112 error = ext4_get_inode_loc(inode, &iloc);
5115 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5116 offset = EXT4_GOOD_OLD_INODE_SIZE +
5117 EXT4_I(inode)->i_extra_isize;
5119 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5120 flags |= FIEMAP_EXTENT_DATA_INLINE;
5122 } else { /* external block */
5123 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5124 length = inode->i_sb->s_blocksize;
5128 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5130 return (error < 0 ? error : 0);
5133 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5134 __u64 start, __u64 len)
5136 ext4_lblk_t start_blk;
5139 if (ext4_has_inline_data(inode)) {
5142 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5148 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5149 error = ext4_ext_precache(inode);
5154 /* fallback to generic here if not in extents fmt */
5155 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5156 return generic_block_fiemap(inode, fieinfo, start, len,
5159 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5162 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5163 error = ext4_xattr_fiemap(inode, fieinfo);
5165 ext4_lblk_t len_blks;
5168 start_blk = start >> inode->i_sb->s_blocksize_bits;
5169 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5170 if (last_blk >= EXT_MAX_BLOCKS)
5171 last_blk = EXT_MAX_BLOCKS-1;
5172 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5175 * Walk the extent tree gathering extent information
5176 * and pushing extents back to the user.
5178 error = ext4_fill_fiemap_extents(inode, start_blk,
5181 ext4_es_lru_add(inode);
5187 * Function to access the path buffer for marking it dirty.
5188 * It also checks if there are sufficient credits left in the journal handle
5192 ext4_access_path(handle_t *handle, struct inode *inode,
5193 struct ext4_ext_path *path)
5197 if (!ext4_handle_valid(handle))
5201 * Check if need to extend journal credits
5202 * 3 for leaf, sb, and inode plus 2 (bmap and group
5203 * descriptor) for each block group; assume two block
5206 if (handle->h_buffer_credits < 7) {
5207 credits = ext4_writepage_trans_blocks(inode);
5208 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5209 /* EAGAIN is success */
5210 if (err && err != -EAGAIN)
5214 err = ext4_ext_get_access(handle, inode, path);
5219 * ext4_ext_shift_path_extents:
5220 * Shift the extents of a path structure lying between path[depth].p_ext
5221 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5222 * from starting block for each extent.
5225 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5226 struct inode *inode, handle_t *handle,
5230 struct ext4_extent *ex_start, *ex_last;
5232 depth = path->p_depth;
5234 while (depth >= 0) {
5235 if (depth == path->p_depth) {
5236 ex_start = path[depth].p_ext;
5240 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5244 err = ext4_access_path(handle, inode, path + depth);
5248 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5251 *start = le32_to_cpu(ex_last->ee_block) +
5252 ext4_ext_get_actual_len(ex_last);
5254 while (ex_start <= ex_last) {
5255 le32_add_cpu(&ex_start->ee_block, -shift);
5256 /* Try to merge to the left. */
5258 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5259 ext4_ext_try_to_merge_right(inode,
5260 path, ex_start - 1))
5265 err = ext4_ext_dirty(handle, inode, path + depth);
5269 if (--depth < 0 || !update)
5273 /* Update index too */
5274 err = ext4_access_path(handle, inode, path + depth);
5278 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5279 err = ext4_ext_dirty(handle, inode, path + depth);
5283 /* we are done if current index is not a starting index */
5284 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5295 * ext4_ext_shift_extents:
5296 * All the extents which lies in the range from start to the last allocated
5297 * block for the file are shifted downwards by shift blocks.
5298 * On success, 0 is returned, error otherwise.
5301 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5302 ext4_lblk_t start, ext4_lblk_t shift)
5304 struct ext4_ext_path *path;
5306 struct ext4_extent *extent;
5307 ext4_lblk_t stop_block;
5308 ext4_lblk_t ex_start, ex_end;
5310 /* Let path point to the last extent */
5311 path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5313 return PTR_ERR(path);
5315 depth = path->p_depth;
5316 extent = path[depth].p_ext;
5318 ext4_ext_drop_refs(path);
5323 stop_block = le32_to_cpu(extent->ee_block) +
5324 ext4_ext_get_actual_len(extent);
5325 ext4_ext_drop_refs(path);
5328 /* Nothing to shift, if hole is at the end of file */
5329 if (start >= stop_block)
5333 * Don't start shifting extents until we make sure the hole is big
5334 * enough to accomodate the shift.
5336 path = ext4_ext_find_extent(inode, start - 1, NULL, 0);
5338 return PTR_ERR(path);
5339 depth = path->p_depth;
5340 extent = path[depth].p_ext;
5342 ex_start = le32_to_cpu(extent->ee_block);
5343 ex_end = le32_to_cpu(extent->ee_block) +
5344 ext4_ext_get_actual_len(extent);
5349 ext4_ext_drop_refs(path);
5352 if ((start == ex_start && shift > ex_start) ||
5353 (shift > start - ex_end))
5356 /* Its safe to start updating extents */
5357 while (start < stop_block) {
5358 path = ext4_ext_find_extent(inode, start, NULL, 0);
5360 return PTR_ERR(path);
5361 depth = path->p_depth;
5362 extent = path[depth].p_ext;
5364 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5365 (unsigned long) start);
5368 if (start > le32_to_cpu(extent->ee_block)) {
5369 /* Hole, move to the next extent */
5370 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5371 path[depth].p_ext++;
5373 start = ext4_ext_next_allocated_block(path);
5374 ext4_ext_drop_refs(path);
5379 ret = ext4_ext_shift_path_extents(path, shift, inode,
5381 ext4_ext_drop_refs(path);
5391 * ext4_collapse_range:
5392 * This implements the fallocate's collapse range functionality for ext4
5393 * Returns: 0 and non-zero on error.
5395 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5397 struct super_block *sb = inode->i_sb;
5398 ext4_lblk_t punch_start, punch_stop;
5400 unsigned int credits;
5401 loff_t new_size, ioffset;
5404 /* Collapse range works only on fs block size aligned offsets. */
5405 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5406 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5409 if (!S_ISREG(inode->i_mode))
5412 trace_ext4_collapse_range(inode, offset, len);
5414 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5415 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5417 /* Call ext4_force_commit to flush all data in case of data=journal. */
5418 if (ext4_should_journal_data(inode)) {
5419 ret = ext4_force_commit(inode->i_sb);
5425 * Need to round down offset to be aligned with page size boundary
5426 * for page size > block size.
5428 ioffset = round_down(offset, PAGE_SIZE);
5430 /* Write out all dirty pages */
5431 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5436 /* Take mutex lock */
5437 mutex_lock(&inode->i_mutex);
5440 * There is no need to overlap collapse range with EOF, in which case
5441 * it is effectively a truncate operation
5443 if (offset + len >= i_size_read(inode)) {
5448 /* Currently just for extent based files */
5449 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5454 truncate_pagecache(inode, ioffset);
5456 /* Wait for existing dio to complete */
5457 ext4_inode_block_unlocked_dio(inode);
5458 inode_dio_wait(inode);
5460 credits = ext4_writepage_trans_blocks(inode);
5461 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5462 if (IS_ERR(handle)) {
5463 ret = PTR_ERR(handle);
5467 down_write(&EXT4_I(inode)->i_data_sem);
5468 ext4_discard_preallocations(inode);
5470 ret = ext4_es_remove_extent(inode, punch_start,
5471 EXT_MAX_BLOCKS - punch_start);
5473 up_write(&EXT4_I(inode)->i_data_sem);
5477 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5479 up_write(&EXT4_I(inode)->i_data_sem);
5482 ext4_discard_preallocations(inode);
5484 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5485 punch_stop - punch_start);
5487 up_write(&EXT4_I(inode)->i_data_sem);
5491 new_size = i_size_read(inode) - len;
5492 i_size_write(inode, new_size);
5493 EXT4_I(inode)->i_disksize = new_size;
5495 up_write(&EXT4_I(inode)->i_data_sem);
5497 ext4_handle_sync(handle);
5498 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5499 ext4_mark_inode_dirty(handle, inode);
5502 ext4_journal_stop(handle);
5504 ext4_inode_resume_unlocked_dio(inode);
5506 mutex_unlock(&inode->i_mutex);
5511 * ext4_swap_extents - Swap extents between two inodes
5513 * @inode1: First inode
5514 * @inode2: Second inode
5515 * @lblk1: Start block for first inode
5516 * @lblk2: Start block for second inode
5517 * @count: Number of blocks to swap
5518 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5519 * @erp: Pointer to save error value
5521 * This helper routine does exactly what is promise "swap extents". All other
5522 * stuff such as page-cache locking consistency, bh mapping consistency or
5523 * extent's data copying must be performed by caller.
5525 * i_mutex is held for both inodes
5526 * i_data_sem is locked for write for both inodes
5528 * All pages from requested range are locked for both inodes
5531 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5532 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5533 ext4_lblk_t count, int unwritten, int *erp)
5535 struct ext4_ext_path *path1 = NULL;
5536 struct ext4_ext_path *path2 = NULL;
5537 int replaced_count = 0;
5539 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5540 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5541 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5542 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5544 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5547 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5552 struct ext4_extent *ex1, *ex2, tmp_ex;
5553 ext4_lblk_t e1_blk, e2_blk;
5554 int e1_len, e2_len, len;
5557 path1 = ext4_ext_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5558 if (unlikely(IS_ERR(path1))) {
5559 *erp = PTR_ERR(path1);
5565 path2 = ext4_ext_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5566 if (unlikely(IS_ERR(path2))) {
5567 *erp = PTR_ERR(path2);
5571 ex1 = path1[path1->p_depth].p_ext;
5572 ex2 = path2[path2->p_depth].p_ext;
5573 /* Do we have somthing to swap ? */
5574 if (unlikely(!ex2 || !ex1))
5577 e1_blk = le32_to_cpu(ex1->ee_block);
5578 e2_blk = le32_to_cpu(ex2->ee_block);
5579 e1_len = ext4_ext_get_actual_len(ex1);
5580 e2_len = ext4_ext_get_actual_len(ex2);
5583 if (!in_range(lblk1, e1_blk, e1_len) ||
5584 !in_range(lblk2, e2_blk, e2_len)) {
5585 ext4_lblk_t next1, next2;
5587 /* if hole after extent, then go to next extent */
5588 next1 = ext4_ext_next_allocated_block(path1);
5589 next2 = ext4_ext_next_allocated_block(path2);
5590 /* If hole before extent, then shift to that extent */
5595 /* Do we have something to swap */
5596 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5598 /* Move to the rightest boundary */
5599 len = next1 - lblk1;
5600 if (len < next2 - lblk2)
5601 len = next2 - lblk2;
5610 /* Prepare left boundary */
5611 if (e1_blk < lblk1) {
5613 *erp = ext4_force_split_extent_at(handle, inode1,
5618 if (e2_blk < lblk2) {
5620 *erp = ext4_force_split_extent_at(handle, inode2,
5625 /* ext4_split_extent_at() may retult in leaf extent split,
5626 * path must to be revalidated. */
5630 /* Prepare right boundary */
5632 if (len > e1_blk + e1_len - lblk1)
5633 len = e1_blk + e1_len - lblk1;
5634 if (len > e2_blk + e2_len - lblk2)
5635 len = e2_blk + e2_len - lblk2;
5637 if (len != e1_len) {
5639 *erp = ext4_force_split_extent_at(handle, inode1,
5640 path1, lblk1 + len, 0);
5644 if (len != e2_len) {
5646 *erp = ext4_force_split_extent_at(handle, inode2,
5647 path2, lblk2 + len, 0);
5651 /* ext4_split_extent_at() may retult in leaf extent split,
5652 * path must to be revalidated. */
5656 BUG_ON(e2_len != e1_len);
5657 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5660 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5664 /* Both extents are fully inside boundaries. Swap it now */
5666 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5667 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5668 ex1->ee_len = cpu_to_le16(e2_len);
5669 ex2->ee_len = cpu_to_le16(e1_len);
5671 ext4_ext_mark_unwritten(ex2);
5672 if (ext4_ext_is_unwritten(&tmp_ex))
5673 ext4_ext_mark_unwritten(ex1);
5675 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5676 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5677 *erp = ext4_ext_dirty(handle, inode2, path2 +
5681 *erp = ext4_ext_dirty(handle, inode1, path1 +
5684 * Looks scarry ah..? second inode already points to new blocks,
5685 * and it was successfully dirtied. But luckily error may happen
5686 * only due to journal error, so full transaction will be
5693 replaced_count += len;
5698 ext4_ext_drop_refs(path1);
5703 ext4_ext_drop_refs(path2);
5708 return replaced_count;