1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
6 * Architecture independence:
7 * Copyright (c) 2005, Bull S.A.
8 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
12 * Extents support for EXT4
15 * - ext4*_error() should be used in some situations
16 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17 * - smart tree reduction
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/iomap.h>
31 #include <linux/sched/mm.h>
32 #include "ext4_jbd2.h"
33 #include "ext4_extents.h"
36 #include <trace/events/ext4.h>
39 * used by extent splitting.
41 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
43 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
44 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
46 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
47 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
49 static __le32 ext4_extent_block_csum(struct inode *inode,
50 struct ext4_extent_header *eh)
52 struct ext4_inode_info *ei = EXT4_I(inode);
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
57 EXT4_EXTENT_TAIL_OFFSET(eh));
58 return cpu_to_le32(csum);
61 static int ext4_extent_block_csum_verify(struct inode *inode,
62 struct ext4_extent_header *eh)
64 struct ext4_extent_tail *et;
66 if (!ext4_has_metadata_csum(inode->i_sb))
69 et = find_ext4_extent_tail(eh);
70 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
75 static void ext4_extent_block_csum_set(struct inode *inode,
76 struct ext4_extent_header *eh)
78 struct ext4_extent_tail *et;
80 if (!ext4_has_metadata_csum(inode->i_sb))
83 et = find_ext4_extent_tail(eh);
84 et->et_checksum = ext4_extent_block_csum(inode, eh);
87 static int ext4_split_extent_at(handle_t *handle,
89 struct ext4_ext_path **ppath,
94 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
97 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
98 * moment, get_block can be called only for blocks inside i_size since
99 * page cache has been already dropped and writes are blocked by
100 * i_rwsem. So we can safely drop the i_data_sem here.
102 BUG_ON(EXT4_JOURNAL(inode) == NULL);
103 ext4_discard_preallocations(inode, 0);
104 up_write(&EXT4_I(inode)->i_data_sem);
110 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
111 * transaction with 'restart_cred' credits. The function drops i_data_sem
112 * when restarting transaction and gets it after transaction is restarted.
114 * The function returns 0 on success, 1 if transaction had to be restarted,
115 * and < 0 in case of fatal error.
117 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
118 int check_cred, int restart_cred,
124 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
125 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
127 down_write(&EXT4_I(inode)->i_data_sem);
136 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
137 struct ext4_ext_path *path)
142 /* path points to block */
143 BUFFER_TRACE(path->p_bh, "get_write_access");
144 err = ext4_journal_get_write_access(handle, inode->i_sb,
145 path->p_bh, EXT4_JTR_NONE);
147 * The extent buffer's verified bit will be set again in
148 * __ext4_ext_dirty(). We could leave an inconsistent
149 * buffer if the extents updating procudure break off du
150 * to some error happens, force to check it again.
153 clear_buffer_verified(path->p_bh);
155 /* path points to leaf/index in inode body */
156 /* we use in-core data, no need to protect them */
166 static int __ext4_ext_dirty(const char *where, unsigned int line,
167 handle_t *handle, struct inode *inode,
168 struct ext4_ext_path *path)
172 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
174 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
175 /* path points to block */
176 err = __ext4_handle_dirty_metadata(where, line, handle,
178 /* Extents updating done, re-set verified flag */
180 set_buffer_verified(path->p_bh);
182 /* path points to leaf/index in inode body */
183 err = ext4_mark_inode_dirty(handle, inode);
188 #define ext4_ext_dirty(handle, inode, path) \
189 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
191 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
192 struct ext4_ext_path *path,
196 int depth = path->p_depth;
197 struct ext4_extent *ex;
200 * Try to predict block placement assuming that we are
201 * filling in a file which will eventually be
202 * non-sparse --- i.e., in the case of libbfd writing
203 * an ELF object sections out-of-order but in a way
204 * the eventually results in a contiguous object or
205 * executable file, or some database extending a table
206 * space file. However, this is actually somewhat
207 * non-ideal if we are writing a sparse file such as
208 * qemu or KVM writing a raw image file that is going
209 * to stay fairly sparse, since it will end up
210 * fragmenting the file system's free space. Maybe we
211 * should have some hueristics or some way to allow
212 * userspace to pass a hint to file system,
213 * especially if the latter case turns out to be
216 ex = path[depth].p_ext;
218 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
219 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
221 if (block > ext_block)
222 return ext_pblk + (block - ext_block);
224 return ext_pblk - (ext_block - block);
227 /* it looks like index is empty;
228 * try to find starting block from index itself */
229 if (path[depth].p_bh)
230 return path[depth].p_bh->b_blocknr;
233 /* OK. use inode's group */
234 return ext4_inode_to_goal_block(inode);
238 * Allocation for a meta data block
241 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
242 struct ext4_ext_path *path,
243 struct ext4_extent *ex, int *err, unsigned int flags)
245 ext4_fsblk_t goal, newblock;
247 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
248 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
253 static inline int ext4_ext_space_block(struct inode *inode, int check)
257 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
258 / sizeof(struct ext4_extent);
259 #ifdef AGGRESSIVE_TEST
260 if (!check && size > 6)
266 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
270 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
271 / sizeof(struct ext4_extent_idx);
272 #ifdef AGGRESSIVE_TEST
273 if (!check && size > 5)
279 static inline int ext4_ext_space_root(struct inode *inode, int check)
283 size = sizeof(EXT4_I(inode)->i_data);
284 size -= sizeof(struct ext4_extent_header);
285 size /= sizeof(struct ext4_extent);
286 #ifdef AGGRESSIVE_TEST
287 if (!check && size > 3)
293 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
297 size = sizeof(EXT4_I(inode)->i_data);
298 size -= sizeof(struct ext4_extent_header);
299 size /= sizeof(struct ext4_extent_idx);
300 #ifdef AGGRESSIVE_TEST
301 if (!check && size > 4)
308 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
309 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
312 struct ext4_ext_path *path = *ppath;
313 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
314 int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
317 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
319 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
320 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
325 ext4_ext_max_entries(struct inode *inode, int depth)
329 if (depth == ext_depth(inode)) {
331 max = ext4_ext_space_root(inode, 1);
333 max = ext4_ext_space_root_idx(inode, 1);
336 max = ext4_ext_space_block(inode, 1);
338 max = ext4_ext_space_block_idx(inode, 1);
344 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
346 ext4_fsblk_t block = ext4_ext_pblock(ext);
347 int len = ext4_ext_get_actual_len(ext);
348 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
353 * - overflow/wrap-around
355 if (lblock + len <= lblock)
357 return ext4_inode_block_valid(inode, block, len);
360 static int ext4_valid_extent_idx(struct inode *inode,
361 struct ext4_extent_idx *ext_idx)
363 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
365 return ext4_inode_block_valid(inode, block, 1);
368 static int ext4_valid_extent_entries(struct inode *inode,
369 struct ext4_extent_header *eh,
370 ext4_lblk_t lblk, ext4_fsblk_t *pblk,
373 unsigned short entries;
374 ext4_lblk_t lblock = 0;
375 ext4_lblk_t prev = 0;
377 if (eh->eh_entries == 0)
380 entries = le16_to_cpu(eh->eh_entries);
384 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
387 * The logical block in the first entry should equal to
388 * the number in the index block.
390 if (depth != ext_depth(inode) &&
391 lblk != le32_to_cpu(ext->ee_block))
394 if (!ext4_valid_extent(inode, ext))
397 /* Check for overlapping extents */
398 lblock = le32_to_cpu(ext->ee_block);
399 if ((lblock <= prev) && prev) {
400 *pblk = ext4_ext_pblock(ext);
403 prev = lblock + ext4_ext_get_actual_len(ext) - 1;
408 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
411 * The logical block in the first entry should equal to
412 * the number in the parent index block.
414 if (depth != ext_depth(inode) &&
415 lblk != le32_to_cpu(ext_idx->ei_block))
418 if (!ext4_valid_extent_idx(inode, ext_idx))
421 /* Check for overlapping index extents */
422 lblock = le32_to_cpu(ext_idx->ei_block);
423 if ((lblock <= prev) && prev) {
424 *pblk = ext4_idx_pblock(ext_idx);
435 static int __ext4_ext_check(const char *function, unsigned int line,
436 struct inode *inode, struct ext4_extent_header *eh,
437 int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
439 const char *error_msg;
440 int max = 0, err = -EFSCORRUPTED;
442 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
443 error_msg = "invalid magic";
446 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
447 error_msg = "unexpected eh_depth";
450 if (unlikely(eh->eh_max == 0)) {
451 error_msg = "invalid eh_max";
454 max = ext4_ext_max_entries(inode, depth);
455 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
456 error_msg = "too large eh_max";
459 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
460 error_msg = "invalid eh_entries";
463 if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
464 error_msg = "invalid extent entries";
467 if (unlikely(depth > 32)) {
468 error_msg = "too large eh_depth";
471 /* Verify checksum on non-root extent tree nodes */
472 if (ext_depth(inode) != depth &&
473 !ext4_extent_block_csum_verify(inode, eh)) {
474 error_msg = "extent tree corrupted";
481 ext4_error_inode_err(inode, function, line, 0, -err,
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),
487 le16_to_cpu(eh->eh_max),
488 max, le16_to_cpu(eh->eh_depth), depth);
492 #define ext4_ext_check(inode, eh, depth, pblk) \
493 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
495 int ext4_ext_check_inode(struct inode *inode)
497 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
500 static void ext4_cache_extents(struct inode *inode,
501 struct ext4_extent_header *eh)
503 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
504 ext4_lblk_t prev = 0;
507 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
508 unsigned int status = EXTENT_STATUS_WRITTEN;
509 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
510 int len = ext4_ext_get_actual_len(ex);
512 if (prev && (prev != lblk))
513 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
516 if (ext4_ext_is_unwritten(ex))
517 status = EXTENT_STATUS_UNWRITTEN;
518 ext4_es_cache_extent(inode, lblk, len,
519 ext4_ext_pblock(ex), status);
524 static struct buffer_head *
525 __read_extent_tree_block(const char *function, unsigned int line,
526 struct inode *inode, struct ext4_extent_idx *idx,
527 int depth, int flags)
529 struct buffer_head *bh;
531 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
534 if (flags & EXT4_EX_NOFAIL)
535 gfp_flags |= __GFP_NOFAIL;
537 pblk = ext4_idx_pblock(idx);
538 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
540 return ERR_PTR(-ENOMEM);
542 if (!bh_uptodate_or_lock(bh)) {
543 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
544 err = ext4_read_bh(bh, 0, NULL);
548 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
550 err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
551 depth, pblk, le32_to_cpu(idx->ei_block));
554 set_buffer_verified(bh);
556 * If this is a leaf block, cache all of its entries
558 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
559 struct ext4_extent_header *eh = ext_block_hdr(bh);
560 ext4_cache_extents(inode, eh);
569 #define read_extent_tree_block(inode, idx, depth, flags) \
570 __read_extent_tree_block(__func__, __LINE__, (inode), (idx), \
574 * This function is called to cache a file's extent information in the
577 int ext4_ext_precache(struct inode *inode)
579 struct ext4_inode_info *ei = EXT4_I(inode);
580 struct ext4_ext_path *path = NULL;
581 struct buffer_head *bh;
582 int i = 0, depth, ret = 0;
584 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
585 return 0; /* not an extent-mapped inode */
587 down_read(&ei->i_data_sem);
588 depth = ext_depth(inode);
590 /* Don't cache anything if there are no external extent blocks */
592 up_read(&ei->i_data_sem);
596 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
599 up_read(&ei->i_data_sem);
603 path[0].p_hdr = ext_inode_hdr(inode);
604 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
607 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
610 * If this is a leaf block or we've reached the end of
611 * the index block, go up
614 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
615 brelse(path[i].p_bh);
620 bh = read_extent_tree_block(inode, path[i].p_idx++,
622 EXT4_EX_FORCE_CACHE);
629 path[i].p_hdr = ext_block_hdr(bh);
630 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
632 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
634 up_read(&ei->i_data_sem);
635 ext4_ext_drop_refs(path);
641 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
643 int k, l = path->p_depth;
645 ext_debug(inode, "path:");
646 for (k = 0; k <= l; k++, path++) {
648 ext_debug(inode, " %d->%llu",
649 le32_to_cpu(path->p_idx->ei_block),
650 ext4_idx_pblock(path->p_idx));
651 } else if (path->p_ext) {
652 ext_debug(inode, " %d:[%d]%d:%llu ",
653 le32_to_cpu(path->p_ext->ee_block),
654 ext4_ext_is_unwritten(path->p_ext),
655 ext4_ext_get_actual_len(path->p_ext),
656 ext4_ext_pblock(path->p_ext));
658 ext_debug(inode, " []");
660 ext_debug(inode, "\n");
663 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
665 int depth = ext_depth(inode);
666 struct ext4_extent_header *eh;
667 struct ext4_extent *ex;
673 eh = path[depth].p_hdr;
674 ex = EXT_FIRST_EXTENT(eh);
676 ext_debug(inode, "Displaying leaf extents\n");
678 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
679 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
680 ext4_ext_is_unwritten(ex),
681 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
683 ext_debug(inode, "\n");
686 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
687 ext4_fsblk_t newblock, int level)
689 int depth = ext_depth(inode);
690 struct ext4_extent *ex;
692 if (depth != level) {
693 struct ext4_extent_idx *idx;
694 idx = path[level].p_idx;
695 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
696 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
697 level, le32_to_cpu(idx->ei_block),
698 ext4_idx_pblock(idx), newblock);
705 ex = path[depth].p_ext;
706 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
707 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
708 le32_to_cpu(ex->ee_block),
710 ext4_ext_is_unwritten(ex),
711 ext4_ext_get_actual_len(ex),
718 #define ext4_ext_show_path(inode, path)
719 #define ext4_ext_show_leaf(inode, path)
720 #define ext4_ext_show_move(inode, path, newblock, level)
723 void ext4_ext_drop_refs(struct ext4_ext_path *path)
729 depth = path->p_depth;
730 for (i = 0; i <= depth; i++, path++) {
737 * ext4_ext_binsearch_idx:
738 * binary search for the closest index of the given block
739 * the header must be checked before calling this
742 ext4_ext_binsearch_idx(struct inode *inode,
743 struct ext4_ext_path *path, ext4_lblk_t block)
745 struct ext4_extent_header *eh = path->p_hdr;
746 struct ext4_extent_idx *r, *l, *m;
749 ext_debug(inode, "binsearch for %u(idx): ", block);
751 l = EXT_FIRST_INDEX(eh) + 1;
752 r = EXT_LAST_INDEX(eh);
755 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
756 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
757 r, le32_to_cpu(r->ei_block));
759 if (block < le32_to_cpu(m->ei_block))
766 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
767 ext4_idx_pblock(path->p_idx));
769 #ifdef CHECK_BINSEARCH
771 struct ext4_extent_idx *chix, *ix;
774 chix = ix = EXT_FIRST_INDEX(eh);
775 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
776 if (k != 0 && le32_to_cpu(ix->ei_block) <=
777 le32_to_cpu(ix[-1].ei_block)) {
778 printk(KERN_DEBUG "k=%d, ix=0x%p, "
780 ix, EXT_FIRST_INDEX(eh));
781 printk(KERN_DEBUG "%u <= %u\n",
782 le32_to_cpu(ix->ei_block),
783 le32_to_cpu(ix[-1].ei_block));
785 BUG_ON(k && le32_to_cpu(ix->ei_block)
786 <= le32_to_cpu(ix[-1].ei_block));
787 if (block < le32_to_cpu(ix->ei_block))
791 BUG_ON(chix != path->p_idx);
798 * ext4_ext_binsearch:
799 * binary search for closest extent of the given block
800 * the header must be checked before calling this
803 ext4_ext_binsearch(struct inode *inode,
804 struct ext4_ext_path *path, ext4_lblk_t block)
806 struct ext4_extent_header *eh = path->p_hdr;
807 struct ext4_extent *r, *l, *m;
809 if (eh->eh_entries == 0) {
811 * this leaf is empty:
812 * we get such a leaf in split/add case
817 ext_debug(inode, "binsearch for %u: ", block);
819 l = EXT_FIRST_EXTENT(eh) + 1;
820 r = EXT_LAST_EXTENT(eh);
824 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
825 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
826 r, le32_to_cpu(r->ee_block));
828 if (block < le32_to_cpu(m->ee_block))
835 ext_debug(inode, " -> %d:%llu:[%d]%d ",
836 le32_to_cpu(path->p_ext->ee_block),
837 ext4_ext_pblock(path->p_ext),
838 ext4_ext_is_unwritten(path->p_ext),
839 ext4_ext_get_actual_len(path->p_ext));
841 #ifdef CHECK_BINSEARCH
843 struct ext4_extent *chex, *ex;
846 chex = ex = EXT_FIRST_EXTENT(eh);
847 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
848 BUG_ON(k && le32_to_cpu(ex->ee_block)
849 <= le32_to_cpu(ex[-1].ee_block));
850 if (block < le32_to_cpu(ex->ee_block))
854 BUG_ON(chex != path->p_ext);
860 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
862 struct ext4_extent_header *eh;
864 eh = ext_inode_hdr(inode);
867 eh->eh_magic = EXT4_EXT_MAGIC;
868 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
869 eh->eh_generation = 0;
870 ext4_mark_inode_dirty(handle, inode);
873 struct ext4_ext_path *
874 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
875 struct ext4_ext_path **orig_path, int flags)
877 struct ext4_extent_header *eh;
878 struct buffer_head *bh;
879 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
880 short int depth, i, ppos = 0;
882 gfp_t gfp_flags = GFP_NOFS;
884 if (flags & EXT4_EX_NOFAIL)
885 gfp_flags |= __GFP_NOFAIL;
887 eh = ext_inode_hdr(inode);
888 depth = ext_depth(inode);
889 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
890 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
897 ext4_ext_drop_refs(path);
898 if (depth > path[0].p_maxdepth) {
900 *orig_path = path = NULL;
904 /* account possible depth increase */
905 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
908 return ERR_PTR(-ENOMEM);
909 path[0].p_maxdepth = depth + 1;
915 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
916 ext4_cache_extents(inode, eh);
917 /* walk through the tree */
919 ext_debug(inode, "depth %d: num %d, max %d\n",
920 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
922 ext4_ext_binsearch_idx(inode, path + ppos, block);
923 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
924 path[ppos].p_depth = i;
925 path[ppos].p_ext = NULL;
927 bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
933 eh = ext_block_hdr(bh);
935 path[ppos].p_bh = bh;
936 path[ppos].p_hdr = eh;
939 path[ppos].p_depth = i;
940 path[ppos].p_ext = NULL;
941 path[ppos].p_idx = NULL;
944 ext4_ext_binsearch(inode, path + ppos, block);
945 /* if not an empty leaf */
946 if (path[ppos].p_ext)
947 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
949 ext4_ext_show_path(inode, path);
954 ext4_ext_drop_refs(path);
962 * ext4_ext_insert_index:
963 * insert new index [@logical;@ptr] into the block at @curp;
964 * check where to insert: before @curp or after @curp
966 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
967 struct ext4_ext_path *curp,
968 int logical, ext4_fsblk_t ptr)
970 struct ext4_extent_idx *ix;
973 err = ext4_ext_get_access(handle, inode, curp);
977 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
978 EXT4_ERROR_INODE(inode,
979 "logical %d == ei_block %d!",
980 logical, le32_to_cpu(curp->p_idx->ei_block));
981 return -EFSCORRUPTED;
984 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
985 >= le16_to_cpu(curp->p_hdr->eh_max))) {
986 EXT4_ERROR_INODE(inode,
987 "eh_entries %d >= eh_max %d!",
988 le16_to_cpu(curp->p_hdr->eh_entries),
989 le16_to_cpu(curp->p_hdr->eh_max));
990 return -EFSCORRUPTED;
993 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
995 ext_debug(inode, "insert new index %d after: %llu\n",
997 ix = curp->p_idx + 1;
1000 ext_debug(inode, "insert new index %d before: %llu\n",
1005 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1008 ext_debug(inode, "insert new index %d: "
1009 "move %d indices from 0x%p to 0x%p\n",
1010 logical, len, ix, ix + 1);
1011 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1014 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1015 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1016 return -EFSCORRUPTED;
1019 ix->ei_block = cpu_to_le32(logical);
1020 ext4_idx_store_pblock(ix, ptr);
1021 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1023 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1024 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1025 return -EFSCORRUPTED;
1028 err = ext4_ext_dirty(handle, inode, curp);
1029 ext4_std_error(inode->i_sb, err);
1036 * inserts new subtree into the path, using free index entry
1038 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1039 * - makes decision where to split
1040 * - moves remaining extents and index entries (right to the split point)
1041 * into the newly allocated blocks
1042 * - initializes subtree
1044 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1046 struct ext4_ext_path *path,
1047 struct ext4_extent *newext, int at)
1049 struct buffer_head *bh = NULL;
1050 int depth = ext_depth(inode);
1051 struct ext4_extent_header *neh;
1052 struct ext4_extent_idx *fidx;
1053 int i = at, k, m, a;
1054 ext4_fsblk_t newblock, oldblock;
1056 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1057 gfp_t gfp_flags = GFP_NOFS;
1059 size_t ext_size = 0;
1061 if (flags & EXT4_EX_NOFAIL)
1062 gfp_flags |= __GFP_NOFAIL;
1064 /* make decision: where to split? */
1065 /* FIXME: now decision is simplest: at current extent */
1067 /* if current leaf will be split, then we should use
1068 * border from split point */
1069 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1070 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1071 return -EFSCORRUPTED;
1073 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1074 border = path[depth].p_ext[1].ee_block;
1075 ext_debug(inode, "leaf will be split."
1076 " next leaf starts at %d\n",
1077 le32_to_cpu(border));
1079 border = newext->ee_block;
1080 ext_debug(inode, "leaf will be added."
1081 " next leaf starts at %d\n",
1082 le32_to_cpu(border));
1086 * If error occurs, then we break processing
1087 * and mark filesystem read-only. index won't
1088 * be inserted and tree will be in consistent
1089 * state. Next mount will repair buffers too.
1093 * Get array to track all allocated blocks.
1094 * We need this to handle errors and free blocks
1097 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1101 /* allocate all needed blocks */
1102 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1103 for (a = 0; a < depth - at; a++) {
1104 newblock = ext4_ext_new_meta_block(handle, inode, path,
1105 newext, &err, flags);
1108 ablocks[a] = newblock;
1111 /* initialize new leaf */
1112 newblock = ablocks[--a];
1113 if (unlikely(newblock == 0)) {
1114 EXT4_ERROR_INODE(inode, "newblock == 0!");
1115 err = -EFSCORRUPTED;
1118 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1119 if (unlikely(!bh)) {
1125 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1130 neh = ext_block_hdr(bh);
1131 neh->eh_entries = 0;
1132 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1133 neh->eh_magic = EXT4_EXT_MAGIC;
1135 neh->eh_generation = 0;
1137 /* move remainder of path[depth] to the new leaf */
1138 if (unlikely(path[depth].p_hdr->eh_entries !=
1139 path[depth].p_hdr->eh_max)) {
1140 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1141 path[depth].p_hdr->eh_entries,
1142 path[depth].p_hdr->eh_max);
1143 err = -EFSCORRUPTED;
1146 /* start copy from next extent */
1147 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1148 ext4_ext_show_move(inode, path, newblock, depth);
1150 struct ext4_extent *ex;
1151 ex = EXT_FIRST_EXTENT(neh);
1152 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1153 le16_add_cpu(&neh->eh_entries, m);
1156 /* zero out unused area in the extent block */
1157 ext_size = sizeof(struct ext4_extent_header) +
1158 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1159 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1160 ext4_extent_block_csum_set(inode, neh);
1161 set_buffer_uptodate(bh);
1164 err = ext4_handle_dirty_metadata(handle, inode, bh);
1170 /* correct old leaf */
1172 err = ext4_ext_get_access(handle, inode, path + depth);
1175 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1176 err = ext4_ext_dirty(handle, inode, path + depth);
1182 /* create intermediate indexes */
1184 if (unlikely(k < 0)) {
1185 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1186 err = -EFSCORRUPTED;
1190 ext_debug(inode, "create %d intermediate indices\n", k);
1191 /* insert new index into current index block */
1192 /* current depth stored in i var */
1195 oldblock = newblock;
1196 newblock = ablocks[--a];
1197 bh = sb_getblk(inode->i_sb, newblock);
1198 if (unlikely(!bh)) {
1204 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1209 neh = ext_block_hdr(bh);
1210 neh->eh_entries = cpu_to_le16(1);
1211 neh->eh_magic = EXT4_EXT_MAGIC;
1212 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1213 neh->eh_depth = cpu_to_le16(depth - i);
1214 neh->eh_generation = 0;
1215 fidx = EXT_FIRST_INDEX(neh);
1216 fidx->ei_block = border;
1217 ext4_idx_store_pblock(fidx, oldblock);
1219 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1220 i, newblock, le32_to_cpu(border), oldblock);
1222 /* move remainder of path[i] to the new index block */
1223 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1224 EXT_LAST_INDEX(path[i].p_hdr))) {
1225 EXT4_ERROR_INODE(inode,
1226 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1227 le32_to_cpu(path[i].p_ext->ee_block));
1228 err = -EFSCORRUPTED;
1231 /* start copy indexes */
1232 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1233 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1234 EXT_MAX_INDEX(path[i].p_hdr));
1235 ext4_ext_show_move(inode, path, newblock, i);
1237 memmove(++fidx, path[i].p_idx,
1238 sizeof(struct ext4_extent_idx) * m);
1239 le16_add_cpu(&neh->eh_entries, m);
1241 /* zero out unused area in the extent block */
1242 ext_size = sizeof(struct ext4_extent_header) +
1243 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1244 memset(bh->b_data + ext_size, 0,
1245 inode->i_sb->s_blocksize - ext_size);
1246 ext4_extent_block_csum_set(inode, neh);
1247 set_buffer_uptodate(bh);
1250 err = ext4_handle_dirty_metadata(handle, inode, bh);
1256 /* correct old index */
1258 err = ext4_ext_get_access(handle, inode, path + i);
1261 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1262 err = ext4_ext_dirty(handle, inode, path + i);
1270 /* insert new index */
1271 err = ext4_ext_insert_index(handle, inode, path + at,
1272 le32_to_cpu(border), newblock);
1276 if (buffer_locked(bh))
1282 /* free all allocated blocks in error case */
1283 for (i = 0; i < depth; i++) {
1286 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1287 EXT4_FREE_BLOCKS_METADATA);
1296 * ext4_ext_grow_indepth:
1297 * implements tree growing procedure:
1298 * - allocates new block
1299 * - moves top-level data (index block or leaf) into the new block
1300 * - initializes new top-level, creating index that points to the
1301 * just created block
1303 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1306 struct ext4_extent_header *neh;
1307 struct buffer_head *bh;
1308 ext4_fsblk_t newblock, goal = 0;
1309 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1311 size_t ext_size = 0;
1313 /* Try to prepend new index to old one */
1314 if (ext_depth(inode))
1315 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1316 if (goal > le32_to_cpu(es->s_first_data_block)) {
1317 flags |= EXT4_MB_HINT_TRY_GOAL;
1320 goal = ext4_inode_to_goal_block(inode);
1321 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1326 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1331 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1338 ext_size = sizeof(EXT4_I(inode)->i_data);
1339 /* move top-level index/leaf into new block */
1340 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1341 /* zero out unused area in the extent block */
1342 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1344 /* set size of new block */
1345 neh = ext_block_hdr(bh);
1346 /* old root could have indexes or leaves
1347 * so calculate e_max right way */
1348 if (ext_depth(inode))
1349 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1351 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1352 neh->eh_magic = EXT4_EXT_MAGIC;
1353 ext4_extent_block_csum_set(inode, neh);
1354 set_buffer_uptodate(bh);
1355 set_buffer_verified(bh);
1358 err = ext4_handle_dirty_metadata(handle, inode, bh);
1362 /* Update top-level index: num,max,pointer */
1363 neh = ext_inode_hdr(inode);
1364 neh->eh_entries = cpu_to_le16(1);
1365 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1366 if (neh->eh_depth == 0) {
1367 /* Root extent block becomes index block */
1368 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1369 EXT_FIRST_INDEX(neh)->ei_block =
1370 EXT_FIRST_EXTENT(neh)->ee_block;
1372 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1373 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1374 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1375 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1377 le16_add_cpu(&neh->eh_depth, 1);
1378 err = ext4_mark_inode_dirty(handle, inode);
1386 * ext4_ext_create_new_leaf:
1387 * finds empty index and adds new leaf.
1388 * if no free index is found, then it requests in-depth growing.
1390 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1391 unsigned int mb_flags,
1392 unsigned int gb_flags,
1393 struct ext4_ext_path **ppath,
1394 struct ext4_extent *newext)
1396 struct ext4_ext_path *path = *ppath;
1397 struct ext4_ext_path *curp;
1398 int depth, i, err = 0;
1401 i = depth = ext_depth(inode);
1403 /* walk up to the tree and look for free index entry */
1404 curp = path + depth;
1405 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1410 /* we use already allocated block for index block,
1411 * so subsequent data blocks should be contiguous */
1412 if (EXT_HAS_FREE_INDEX(curp)) {
1413 /* if we found index with free entry, then use that
1414 * entry: create all needed subtree and add new leaf */
1415 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1420 path = ext4_find_extent(inode,
1421 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1424 err = PTR_ERR(path);
1426 /* tree is full, time to grow in depth */
1427 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1432 path = ext4_find_extent(inode,
1433 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1436 err = PTR_ERR(path);
1441 * only first (depth 0 -> 1) produces free space;
1442 * in all other cases we have to split the grown tree
1444 depth = ext_depth(inode);
1445 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1446 /* now we need to split */
1456 * search the closest allocated block to the left for *logical
1457 * and returns it at @logical + it's physical address at @phys
1458 * if *logical is the smallest allocated block, the function
1459 * returns 0 at @phys
1460 * return value contains 0 (success) or error code
1462 static int ext4_ext_search_left(struct inode *inode,
1463 struct ext4_ext_path *path,
1464 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1466 struct ext4_extent_idx *ix;
1467 struct ext4_extent *ex;
1470 if (unlikely(path == NULL)) {
1471 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1472 return -EFSCORRUPTED;
1474 depth = path->p_depth;
1477 if (depth == 0 && path->p_ext == NULL)
1480 /* usually extent in the path covers blocks smaller
1481 * then *logical, but it can be that extent is the
1482 * first one in the file */
1484 ex = path[depth].p_ext;
1485 ee_len = ext4_ext_get_actual_len(ex);
1486 if (*logical < le32_to_cpu(ex->ee_block)) {
1487 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1488 EXT4_ERROR_INODE(inode,
1489 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1490 *logical, le32_to_cpu(ex->ee_block));
1491 return -EFSCORRUPTED;
1493 while (--depth >= 0) {
1494 ix = path[depth].p_idx;
1495 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1496 EXT4_ERROR_INODE(inode,
1497 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1498 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1499 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block),
1501 return -EFSCORRUPTED;
1507 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1508 EXT4_ERROR_INODE(inode,
1509 "logical %d < ee_block %d + ee_len %d!",
1510 *logical, le32_to_cpu(ex->ee_block), ee_len);
1511 return -EFSCORRUPTED;
1514 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1515 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1520 * Search the closest allocated block to the right for *logical
1521 * and returns it at @logical + it's physical address at @phys.
1522 * If not exists, return 0 and @phys is set to 0. We will return
1523 * 1 which means we found an allocated block and ret_ex is valid.
1524 * Or return a (< 0) error code.
1526 static int ext4_ext_search_right(struct inode *inode,
1527 struct ext4_ext_path *path,
1528 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1529 struct ext4_extent *ret_ex)
1531 struct buffer_head *bh = NULL;
1532 struct ext4_extent_header *eh;
1533 struct ext4_extent_idx *ix;
1534 struct ext4_extent *ex;
1535 int depth; /* Note, NOT eh_depth; depth from top of tree */
1538 if (unlikely(path == NULL)) {
1539 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1540 return -EFSCORRUPTED;
1542 depth = path->p_depth;
1545 if (depth == 0 && path->p_ext == NULL)
1548 /* usually extent in the path covers blocks smaller
1549 * then *logical, but it can be that extent is the
1550 * first one in the file */
1552 ex = path[depth].p_ext;
1553 ee_len = ext4_ext_get_actual_len(ex);
1554 if (*logical < le32_to_cpu(ex->ee_block)) {
1555 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1556 EXT4_ERROR_INODE(inode,
1557 "first_extent(path[%d].p_hdr) != ex",
1559 return -EFSCORRUPTED;
1561 while (--depth >= 0) {
1562 ix = path[depth].p_idx;
1563 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1564 EXT4_ERROR_INODE(inode,
1565 "ix != EXT_FIRST_INDEX *logical %d!",
1567 return -EFSCORRUPTED;
1573 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1574 EXT4_ERROR_INODE(inode,
1575 "logical %d < ee_block %d + ee_len %d!",
1576 *logical, le32_to_cpu(ex->ee_block), ee_len);
1577 return -EFSCORRUPTED;
1580 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1581 /* next allocated block in this leaf */
1586 /* go up and search for index to the right */
1587 while (--depth >= 0) {
1588 ix = path[depth].p_idx;
1589 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1593 /* we've gone up to the root and found no index to the right */
1597 /* we've found index to the right, let's
1598 * follow it and find the closest allocated
1599 * block to the right */
1601 while (++depth < path->p_depth) {
1602 /* subtract from p_depth to get proper eh_depth */
1603 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1606 eh = ext_block_hdr(bh);
1607 ix = EXT_FIRST_INDEX(eh);
1611 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1614 eh = ext_block_hdr(bh);
1615 ex = EXT_FIRST_EXTENT(eh);
1617 *logical = le32_to_cpu(ex->ee_block);
1618 *phys = ext4_ext_pblock(ex);
1627 * ext4_ext_next_allocated_block:
1628 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1629 * NOTE: it considers block number from index entry as
1630 * allocated block. Thus, index entries have to be consistent
1634 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1638 BUG_ON(path == NULL);
1639 depth = path->p_depth;
1641 if (depth == 0 && path->p_ext == NULL)
1642 return EXT_MAX_BLOCKS;
1644 while (depth >= 0) {
1645 struct ext4_ext_path *p = &path[depth];
1647 if (depth == path->p_depth) {
1649 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1650 return le32_to_cpu(p->p_ext[1].ee_block);
1653 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1654 return le32_to_cpu(p->p_idx[1].ei_block);
1659 return EXT_MAX_BLOCKS;
1663 * ext4_ext_next_leaf_block:
1664 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1666 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1670 BUG_ON(path == NULL);
1671 depth = path->p_depth;
1673 /* zero-tree has no leaf blocks at all */
1675 return EXT_MAX_BLOCKS;
1677 /* go to index block */
1680 while (depth >= 0) {
1681 if (path[depth].p_idx !=
1682 EXT_LAST_INDEX(path[depth].p_hdr))
1683 return (ext4_lblk_t)
1684 le32_to_cpu(path[depth].p_idx[1].ei_block);
1688 return EXT_MAX_BLOCKS;
1692 * ext4_ext_correct_indexes:
1693 * if leaf gets modified and modified extent is first in the leaf,
1694 * then we have to correct all indexes above.
1695 * TODO: do we need to correct tree in all cases?
1697 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1698 struct ext4_ext_path *path)
1700 struct ext4_extent_header *eh;
1701 int depth = ext_depth(inode);
1702 struct ext4_extent *ex;
1706 eh = path[depth].p_hdr;
1707 ex = path[depth].p_ext;
1709 if (unlikely(ex == NULL || eh == NULL)) {
1710 EXT4_ERROR_INODE(inode,
1711 "ex %p == NULL or eh %p == NULL", ex, eh);
1712 return -EFSCORRUPTED;
1716 /* there is no tree at all */
1720 if (ex != EXT_FIRST_EXTENT(eh)) {
1721 /* we correct tree if first leaf got modified only */
1726 * TODO: we need correction if border is smaller than current one
1729 border = path[depth].p_ext->ee_block;
1730 err = ext4_ext_get_access(handle, inode, path + k);
1733 path[k].p_idx->ei_block = border;
1734 err = ext4_ext_dirty(handle, inode, path + k);
1739 /* change all left-side indexes */
1740 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1742 err = ext4_ext_get_access(handle, inode, path + k);
1745 path[k].p_idx->ei_block = border;
1746 err = ext4_ext_dirty(handle, inode, path + k);
1754 static int ext4_can_extents_be_merged(struct inode *inode,
1755 struct ext4_extent *ex1,
1756 struct ext4_extent *ex2)
1758 unsigned short ext1_ee_len, ext2_ee_len;
1760 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1763 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1764 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1766 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1767 le32_to_cpu(ex2->ee_block))
1770 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1773 if (ext4_ext_is_unwritten(ex1) &&
1774 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1776 #ifdef AGGRESSIVE_TEST
1777 if (ext1_ee_len >= 4)
1781 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1787 * This function tries to merge the "ex" extent to the next extent in the tree.
1788 * It always tries to merge towards right. If you want to merge towards
1789 * left, pass "ex - 1" as argument instead of "ex".
1790 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1791 * 1 if they got merged.
1793 static int ext4_ext_try_to_merge_right(struct inode *inode,
1794 struct ext4_ext_path *path,
1795 struct ext4_extent *ex)
1797 struct ext4_extent_header *eh;
1798 unsigned int depth, len;
1799 int merge_done = 0, unwritten;
1801 depth = ext_depth(inode);
1802 BUG_ON(path[depth].p_hdr == NULL);
1803 eh = path[depth].p_hdr;
1805 while (ex < EXT_LAST_EXTENT(eh)) {
1806 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1808 /* merge with next extent! */
1809 unwritten = ext4_ext_is_unwritten(ex);
1810 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1811 + ext4_ext_get_actual_len(ex + 1));
1813 ext4_ext_mark_unwritten(ex);
1815 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1816 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1817 * sizeof(struct ext4_extent);
1818 memmove(ex + 1, ex + 2, len);
1820 le16_add_cpu(&eh->eh_entries, -1);
1822 WARN_ON(eh->eh_entries == 0);
1823 if (!eh->eh_entries)
1824 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1831 * This function does a very simple check to see if we can collapse
1832 * an extent tree with a single extent tree leaf block into the inode.
1834 static void ext4_ext_try_to_merge_up(handle_t *handle,
1835 struct inode *inode,
1836 struct ext4_ext_path *path)
1839 unsigned max_root = ext4_ext_space_root(inode, 0);
1842 if ((path[0].p_depth != 1) ||
1843 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1844 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1848 * We need to modify the block allocation bitmap and the block
1849 * group descriptor to release the extent tree block. If we
1850 * can't get the journal credits, give up.
1852 if (ext4_journal_extend(handle, 2,
1853 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1857 * Copy the extent data up to the inode
1859 blk = ext4_idx_pblock(path[0].p_idx);
1860 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1861 sizeof(struct ext4_extent_idx);
1862 s += sizeof(struct ext4_extent_header);
1864 path[1].p_maxdepth = path[0].p_maxdepth;
1865 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1866 path[0].p_depth = 0;
1867 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1868 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1869 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1871 brelse(path[1].p_bh);
1872 ext4_free_blocks(handle, inode, NULL, blk, 1,
1873 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1877 * This function tries to merge the @ex extent to neighbours in the tree, then
1878 * tries to collapse the extent tree into the inode.
1880 static void ext4_ext_try_to_merge(handle_t *handle,
1881 struct inode *inode,
1882 struct ext4_ext_path *path,
1883 struct ext4_extent *ex)
1885 struct ext4_extent_header *eh;
1889 depth = ext_depth(inode);
1890 BUG_ON(path[depth].p_hdr == NULL);
1891 eh = path[depth].p_hdr;
1893 if (ex > EXT_FIRST_EXTENT(eh))
1894 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1897 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1899 ext4_ext_try_to_merge_up(handle, inode, path);
1903 * check if a portion of the "newext" extent overlaps with an
1906 * If there is an overlap discovered, it updates the length of the newext
1907 * such that there will be no overlap, and then returns 1.
1908 * If there is no overlap found, it returns 0.
1910 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1911 struct inode *inode,
1912 struct ext4_extent *newext,
1913 struct ext4_ext_path *path)
1916 unsigned int depth, len1;
1917 unsigned int ret = 0;
1919 b1 = le32_to_cpu(newext->ee_block);
1920 len1 = ext4_ext_get_actual_len(newext);
1921 depth = ext_depth(inode);
1922 if (!path[depth].p_ext)
1924 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1927 * get the next allocated block if the extent in the path
1928 * is before the requested block(s)
1931 b2 = ext4_ext_next_allocated_block(path);
1932 if (b2 == EXT_MAX_BLOCKS)
1934 b2 = EXT4_LBLK_CMASK(sbi, b2);
1937 /* check for wrap through zero on extent logical start block*/
1938 if (b1 + len1 < b1) {
1939 len1 = EXT_MAX_BLOCKS - b1;
1940 newext->ee_len = cpu_to_le16(len1);
1944 /* check for overlap */
1945 if (b1 + len1 > b2) {
1946 newext->ee_len = cpu_to_le16(b2 - b1);
1954 * ext4_ext_insert_extent:
1955 * tries to merge requested extent into the existing extent or
1956 * inserts requested extent as new one into the tree,
1957 * creating new leaf in the no-space case.
1959 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1960 struct ext4_ext_path **ppath,
1961 struct ext4_extent *newext, int gb_flags)
1963 struct ext4_ext_path *path = *ppath;
1964 struct ext4_extent_header *eh;
1965 struct ext4_extent *ex, *fex;
1966 struct ext4_extent *nearex; /* nearest extent */
1967 struct ext4_ext_path *npath = NULL;
1968 int depth, len, err;
1970 int mb_flags = 0, unwritten;
1972 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1973 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1974 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1975 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1976 return -EFSCORRUPTED;
1978 depth = ext_depth(inode);
1979 ex = path[depth].p_ext;
1980 eh = path[depth].p_hdr;
1981 if (unlikely(path[depth].p_hdr == NULL)) {
1982 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1983 return -EFSCORRUPTED;
1986 /* try to insert block into found extent and return */
1987 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1990 * Try to see whether we should rather test the extent on
1991 * right from ex, or from the left of ex. This is because
1992 * ext4_find_extent() can return either extent on the
1993 * left, or on the right from the searched position. This
1994 * will make merging more effective.
1996 if (ex < EXT_LAST_EXTENT(eh) &&
1997 (le32_to_cpu(ex->ee_block) +
1998 ext4_ext_get_actual_len(ex) <
1999 le32_to_cpu(newext->ee_block))) {
2002 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2003 (le32_to_cpu(newext->ee_block) +
2004 ext4_ext_get_actual_len(newext) <
2005 le32_to_cpu(ex->ee_block)))
2008 /* Try to append newex to the ex */
2009 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2010 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2012 ext4_ext_is_unwritten(newext),
2013 ext4_ext_get_actual_len(newext),
2014 le32_to_cpu(ex->ee_block),
2015 ext4_ext_is_unwritten(ex),
2016 ext4_ext_get_actual_len(ex),
2017 ext4_ext_pblock(ex));
2018 err = ext4_ext_get_access(handle, inode,
2022 unwritten = ext4_ext_is_unwritten(ex);
2023 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2024 + ext4_ext_get_actual_len(newext));
2026 ext4_ext_mark_unwritten(ex);
2032 /* Try to prepend newex to the ex */
2033 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2034 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2036 le32_to_cpu(newext->ee_block),
2037 ext4_ext_is_unwritten(newext),
2038 ext4_ext_get_actual_len(newext),
2039 le32_to_cpu(ex->ee_block),
2040 ext4_ext_is_unwritten(ex),
2041 ext4_ext_get_actual_len(ex),
2042 ext4_ext_pblock(ex));
2043 err = ext4_ext_get_access(handle, inode,
2048 unwritten = ext4_ext_is_unwritten(ex);
2049 ex->ee_block = newext->ee_block;
2050 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2051 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2052 + ext4_ext_get_actual_len(newext));
2054 ext4_ext_mark_unwritten(ex);
2060 depth = ext_depth(inode);
2061 eh = path[depth].p_hdr;
2062 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2065 /* probably next leaf has space for us? */
2066 fex = EXT_LAST_EXTENT(eh);
2067 next = EXT_MAX_BLOCKS;
2068 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2069 next = ext4_ext_next_leaf_block(path);
2070 if (next != EXT_MAX_BLOCKS) {
2071 ext_debug(inode, "next leaf block - %u\n", next);
2072 BUG_ON(npath != NULL);
2073 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2075 return PTR_ERR(npath);
2076 BUG_ON(npath->p_depth != path->p_depth);
2077 eh = npath[depth].p_hdr;
2078 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2079 ext_debug(inode, "next leaf isn't full(%d)\n",
2080 le16_to_cpu(eh->eh_entries));
2084 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2085 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2089 * There is no free space in the found leaf.
2090 * We're gonna add a new leaf in the tree.
2092 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2093 mb_flags |= EXT4_MB_USE_RESERVED;
2094 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2098 depth = ext_depth(inode);
2099 eh = path[depth].p_hdr;
2102 nearex = path[depth].p_ext;
2104 err = ext4_ext_get_access(handle, inode, path + depth);
2109 /* there is no extent in this leaf, create first one */
2110 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2111 le32_to_cpu(newext->ee_block),
2112 ext4_ext_pblock(newext),
2113 ext4_ext_is_unwritten(newext),
2114 ext4_ext_get_actual_len(newext));
2115 nearex = EXT_FIRST_EXTENT(eh);
2117 if (le32_to_cpu(newext->ee_block)
2118 > le32_to_cpu(nearex->ee_block)) {
2120 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2122 le32_to_cpu(newext->ee_block),
2123 ext4_ext_pblock(newext),
2124 ext4_ext_is_unwritten(newext),
2125 ext4_ext_get_actual_len(newext),
2130 BUG_ON(newext->ee_block == nearex->ee_block);
2131 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2133 le32_to_cpu(newext->ee_block),
2134 ext4_ext_pblock(newext),
2135 ext4_ext_is_unwritten(newext),
2136 ext4_ext_get_actual_len(newext),
2139 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2141 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2142 "move %d extents from 0x%p to 0x%p\n",
2143 le32_to_cpu(newext->ee_block),
2144 ext4_ext_pblock(newext),
2145 ext4_ext_is_unwritten(newext),
2146 ext4_ext_get_actual_len(newext),
2147 len, nearex, nearex + 1);
2148 memmove(nearex + 1, nearex,
2149 len * sizeof(struct ext4_extent));
2153 le16_add_cpu(&eh->eh_entries, 1);
2154 path[depth].p_ext = nearex;
2155 nearex->ee_block = newext->ee_block;
2156 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2157 nearex->ee_len = newext->ee_len;
2160 /* try to merge extents */
2161 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2162 ext4_ext_try_to_merge(handle, inode, path, nearex);
2165 /* time to correct all indexes above */
2166 err = ext4_ext_correct_indexes(handle, inode, path);
2170 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2173 ext4_ext_drop_refs(npath);
2178 static int ext4_fill_es_cache_info(struct inode *inode,
2179 ext4_lblk_t block, ext4_lblk_t num,
2180 struct fiemap_extent_info *fieinfo)
2182 ext4_lblk_t next, end = block + num - 1;
2183 struct extent_status es;
2184 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2188 while (block <= end) {
2191 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2193 if (ext4_es_is_unwritten(&es))
2194 flags |= FIEMAP_EXTENT_UNWRITTEN;
2195 if (ext4_es_is_delayed(&es))
2196 flags |= (FIEMAP_EXTENT_DELALLOC |
2197 FIEMAP_EXTENT_UNKNOWN);
2198 if (ext4_es_is_hole(&es))
2199 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2201 flags |= FIEMAP_EXTENT_LAST;
2202 if (flags & (FIEMAP_EXTENT_DELALLOC|
2203 EXT4_FIEMAP_EXTENT_HOLE))
2206 es.es_pblk = ext4_es_pblock(&es);
2207 err = fiemap_fill_next_extent(fieinfo,
2208 (__u64)es.es_lblk << blksize_bits,
2209 (__u64)es.es_pblk << blksize_bits,
2210 (__u64)es.es_len << blksize_bits,
2225 * ext4_ext_determine_hole - determine hole around given block
2226 * @inode: inode we lookup in
2227 * @path: path in extent tree to @lblk
2228 * @lblk: pointer to logical block around which we want to determine hole
2230 * Determine hole length (and start if easily possible) around given logical
2231 * block. We don't try too hard to find the beginning of the hole but @path
2232 * actually points to extent before @lblk, we provide it.
2234 * The function returns the length of a hole starting at @lblk. We update @lblk
2235 * to the beginning of the hole if we managed to find it.
2237 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2238 struct ext4_ext_path *path,
2241 int depth = ext_depth(inode);
2242 struct ext4_extent *ex;
2245 ex = path[depth].p_ext;
2247 /* there is no extent yet, so gap is [0;-] */
2249 len = EXT_MAX_BLOCKS;
2250 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2251 len = le32_to_cpu(ex->ee_block) - *lblk;
2252 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2253 + ext4_ext_get_actual_len(ex)) {
2256 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2257 next = ext4_ext_next_allocated_block(path);
2258 BUG_ON(next == *lblk);
2267 * ext4_ext_put_gap_in_cache:
2268 * calculate boundaries of the gap that the requested block fits into
2269 * and cache this gap
2272 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2273 ext4_lblk_t hole_len)
2275 struct extent_status es;
2277 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2278 hole_start + hole_len - 1, &es);
2280 /* There's delayed extent containing lblock? */
2281 if (es.es_lblk <= hole_start)
2283 hole_len = min(es.es_lblk - hole_start, hole_len);
2285 ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2286 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2287 EXTENT_STATUS_HOLE);
2292 * removes index from the index block.
2294 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2295 struct ext4_ext_path *path, int depth)
2300 /* free index block */
2302 path = path + depth;
2303 leaf = ext4_idx_pblock(path->p_idx);
2304 if (unlikely(path->p_hdr->eh_entries == 0)) {
2305 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2306 return -EFSCORRUPTED;
2308 err = ext4_ext_get_access(handle, inode, path);
2312 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2313 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2314 len *= sizeof(struct ext4_extent_idx);
2315 memmove(path->p_idx, path->p_idx + 1, len);
2318 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2319 err = ext4_ext_dirty(handle, inode, path);
2322 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2323 trace_ext4_ext_rm_idx(inode, leaf);
2325 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2326 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2328 while (--depth >= 0) {
2329 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2332 err = ext4_ext_get_access(handle, inode, path);
2335 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2336 err = ext4_ext_dirty(handle, inode, path);
2344 * ext4_ext_calc_credits_for_single_extent:
2345 * This routine returns max. credits that needed to insert an extent
2346 * to the extent tree.
2347 * When pass the actual path, the caller should calculate credits
2350 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2351 struct ext4_ext_path *path)
2354 int depth = ext_depth(inode);
2357 /* probably there is space in leaf? */
2358 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2359 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2362 * There are some space in the leaf tree, no
2363 * need to account for leaf block credit
2365 * bitmaps and block group descriptor blocks
2366 * and other metadata blocks still need to be
2369 /* 1 bitmap, 1 block group descriptor */
2370 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2375 return ext4_chunk_trans_blocks(inode, nrblocks);
2379 * How many index/leaf blocks need to change/allocate to add @extents extents?
2381 * If we add a single extent, then in the worse case, each tree level
2382 * index/leaf need to be changed in case of the tree split.
2384 * If more extents are inserted, they could cause the whole tree split more
2385 * than once, but this is really rare.
2387 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2392 /* If we are converting the inline data, only one is needed here. */
2393 if (ext4_has_inline_data(inode))
2396 depth = ext_depth(inode);
2406 static inline int get_default_free_blocks_flags(struct inode *inode)
2408 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2409 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2410 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2411 else if (ext4_should_journal_data(inode))
2412 return EXT4_FREE_BLOCKS_FORGET;
2417 * ext4_rereserve_cluster - increment the reserved cluster count when
2418 * freeing a cluster with a pending reservation
2420 * @inode - file containing the cluster
2421 * @lblk - logical block in cluster to be reserved
2423 * Increments the reserved cluster count and adjusts quota in a bigalloc
2424 * file system when freeing a partial cluster containing at least one
2425 * delayed and unwritten block. A partial cluster meeting that
2426 * requirement will have a pending reservation. If so, the
2427 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2428 * defer reserved and allocated space accounting to a subsequent call
2431 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2433 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2434 struct ext4_inode_info *ei = EXT4_I(inode);
2436 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2438 spin_lock(&ei->i_block_reservation_lock);
2439 ei->i_reserved_data_blocks++;
2440 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2441 spin_unlock(&ei->i_block_reservation_lock);
2443 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2444 ext4_remove_pending(inode, lblk);
2447 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2448 struct ext4_extent *ex,
2449 struct partial_cluster *partial,
2450 ext4_lblk_t from, ext4_lblk_t to)
2452 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2453 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2454 ext4_fsblk_t last_pblk, pblk;
2458 /* only extent tail removal is allowed */
2459 if (from < le32_to_cpu(ex->ee_block) ||
2460 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2461 ext4_error(sbi->s_sb,
2462 "strange request: removal(2) %u-%u from %u:%u",
2463 from, to, le32_to_cpu(ex->ee_block), ee_len);
2467 #ifdef EXTENTS_STATS
2468 spin_lock(&sbi->s_ext_stats_lock);
2469 sbi->s_ext_blocks += ee_len;
2470 sbi->s_ext_extents++;
2471 if (ee_len < sbi->s_ext_min)
2472 sbi->s_ext_min = ee_len;
2473 if (ee_len > sbi->s_ext_max)
2474 sbi->s_ext_max = ee_len;
2475 if (ext_depth(inode) > sbi->s_depth_max)
2476 sbi->s_depth_max = ext_depth(inode);
2477 spin_unlock(&sbi->s_ext_stats_lock);
2480 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2483 * if we have a partial cluster, and it's different from the
2484 * cluster of the last block in the extent, we free it
2486 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2488 if (partial->state != initial &&
2489 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2490 if (partial->state == tofree) {
2491 flags = get_default_free_blocks_flags(inode);
2492 if (ext4_is_pending(inode, partial->lblk))
2493 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2494 ext4_free_blocks(handle, inode, NULL,
2495 EXT4_C2B(sbi, partial->pclu),
2496 sbi->s_cluster_ratio, flags);
2497 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2498 ext4_rereserve_cluster(inode, partial->lblk);
2500 partial->state = initial;
2503 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2504 pblk = ext4_ext_pblock(ex) + ee_len - num;
2507 * We free the partial cluster at the end of the extent (if any),
2508 * unless the cluster is used by another extent (partial_cluster
2509 * state is nofree). If a partial cluster exists here, it must be
2510 * shared with the last block in the extent.
2512 flags = get_default_free_blocks_flags(inode);
2514 /* partial, left end cluster aligned, right end unaligned */
2515 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2516 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2517 (partial->state != nofree)) {
2518 if (ext4_is_pending(inode, to))
2519 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2520 ext4_free_blocks(handle, inode, NULL,
2521 EXT4_PBLK_CMASK(sbi, last_pblk),
2522 sbi->s_cluster_ratio, flags);
2523 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2524 ext4_rereserve_cluster(inode, to);
2525 partial->state = initial;
2526 flags = get_default_free_blocks_flags(inode);
2529 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2532 * For bigalloc file systems, we never free a partial cluster
2533 * at the beginning of the extent. Instead, we check to see if we
2534 * need to free it on a subsequent call to ext4_remove_blocks,
2535 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2537 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2538 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2540 /* reset the partial cluster if we've freed past it */
2541 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2542 partial->state = initial;
2545 * If we've freed the entire extent but the beginning is not left
2546 * cluster aligned and is not marked as ineligible for freeing we
2547 * record the partial cluster at the beginning of the extent. It
2548 * wasn't freed by the preceding ext4_free_blocks() call, and we
2549 * need to look farther to the left to determine if it's to be freed
2550 * (not shared with another extent). Else, reset the partial
2551 * cluster - we're either done freeing or the beginning of the
2552 * extent is left cluster aligned.
2554 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2555 if (partial->state == initial) {
2556 partial->pclu = EXT4_B2C(sbi, pblk);
2557 partial->lblk = from;
2558 partial->state = tofree;
2561 partial->state = initial;
2568 * ext4_ext_rm_leaf() Removes the extents associated with the
2569 * blocks appearing between "start" and "end". Both "start"
2570 * and "end" must appear in the same extent or EIO is returned.
2572 * @handle: The journal handle
2573 * @inode: The files inode
2574 * @path: The path to the leaf
2575 * @partial_cluster: The cluster which we'll have to free if all extents
2576 * has been released from it. However, if this value is
2577 * negative, it's a cluster just to the right of the
2578 * punched region and it must not be freed.
2579 * @start: The first block to remove
2580 * @end: The last block to remove
2583 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2584 struct ext4_ext_path *path,
2585 struct partial_cluster *partial,
2586 ext4_lblk_t start, ext4_lblk_t end)
2588 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2589 int err = 0, correct_index = 0;
2590 int depth = ext_depth(inode), credits, revoke_credits;
2591 struct ext4_extent_header *eh;
2594 ext4_lblk_t ex_ee_block;
2595 unsigned short ex_ee_len;
2596 unsigned unwritten = 0;
2597 struct ext4_extent *ex;
2600 /* the header must be checked already in ext4_ext_remove_space() */
2601 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2602 if (!path[depth].p_hdr)
2603 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2604 eh = path[depth].p_hdr;
2605 if (unlikely(path[depth].p_hdr == NULL)) {
2606 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2607 return -EFSCORRUPTED;
2609 /* find where to start removing */
2610 ex = path[depth].p_ext;
2612 ex = EXT_LAST_EXTENT(eh);
2614 ex_ee_block = le32_to_cpu(ex->ee_block);
2615 ex_ee_len = ext4_ext_get_actual_len(ex);
2617 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2619 while (ex >= EXT_FIRST_EXTENT(eh) &&
2620 ex_ee_block + ex_ee_len > start) {
2622 if (ext4_ext_is_unwritten(ex))
2627 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2628 unwritten, ex_ee_len);
2629 path[depth].p_ext = ex;
2631 a = ex_ee_block > start ? ex_ee_block : start;
2632 b = ex_ee_block+ex_ee_len - 1 < end ?
2633 ex_ee_block+ex_ee_len - 1 : end;
2635 ext_debug(inode, " border %u:%u\n", a, b);
2637 /* If this extent is beyond the end of the hole, skip it */
2638 if (end < ex_ee_block) {
2640 * We're going to skip this extent and move to another,
2641 * so note that its first cluster is in use to avoid
2642 * freeing it when removing blocks. Eventually, the
2643 * right edge of the truncated/punched region will
2644 * be just to the left.
2646 if (sbi->s_cluster_ratio > 1) {
2647 pblk = ext4_ext_pblock(ex);
2648 partial->pclu = EXT4_B2C(sbi, pblk);
2649 partial->state = nofree;
2652 ex_ee_block = le32_to_cpu(ex->ee_block);
2653 ex_ee_len = ext4_ext_get_actual_len(ex);
2655 } else if (b != ex_ee_block + ex_ee_len - 1) {
2656 EXT4_ERROR_INODE(inode,
2657 "can not handle truncate %u:%u "
2659 start, end, ex_ee_block,
2660 ex_ee_block + ex_ee_len - 1);
2661 err = -EFSCORRUPTED;
2663 } else if (a != ex_ee_block) {
2664 /* remove tail of the extent */
2665 num = a - ex_ee_block;
2667 /* remove whole extent: excellent! */
2671 * 3 for leaf, sb, and inode plus 2 (bmap and group
2672 * descriptor) for each block group; assume two block
2673 * groups plus ex_ee_len/blocks_per_block_group for
2676 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2677 if (ex == EXT_FIRST_EXTENT(eh)) {
2679 credits += (ext_depth(inode)) + 1;
2681 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2683 * We may end up freeing some index blocks and data from the
2684 * punched range. Note that partial clusters are accounted for
2685 * by ext4_free_data_revoke_credits().
2688 ext4_free_metadata_revoke_credits(inode->i_sb,
2690 ext4_free_data_revoke_credits(inode, b - a + 1);
2692 err = ext4_datasem_ensure_credits(handle, inode, credits,
2693 credits, revoke_credits);
2700 err = ext4_ext_get_access(handle, inode, path + depth);
2704 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2709 /* this extent is removed; mark slot entirely unused */
2710 ext4_ext_store_pblock(ex, 0);
2712 ex->ee_len = cpu_to_le16(num);
2714 * Do not mark unwritten if all the blocks in the
2715 * extent have been removed.
2717 if (unwritten && num)
2718 ext4_ext_mark_unwritten(ex);
2720 * If the extent was completely released,
2721 * we need to remove it from the leaf
2724 if (end != EXT_MAX_BLOCKS - 1) {
2726 * For hole punching, we need to scoot all the
2727 * extents up when an extent is removed so that
2728 * we dont have blank extents in the middle
2730 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2731 sizeof(struct ext4_extent));
2733 /* Now get rid of the one at the end */
2734 memset(EXT_LAST_EXTENT(eh), 0,
2735 sizeof(struct ext4_extent));
2737 le16_add_cpu(&eh->eh_entries, -1);
2740 err = ext4_ext_dirty(handle, inode, path + depth);
2744 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2745 ext4_ext_pblock(ex));
2747 ex_ee_block = le32_to_cpu(ex->ee_block);
2748 ex_ee_len = ext4_ext_get_actual_len(ex);
2751 if (correct_index && eh->eh_entries)
2752 err = ext4_ext_correct_indexes(handle, inode, path);
2755 * If there's a partial cluster and at least one extent remains in
2756 * the leaf, free the partial cluster if it isn't shared with the
2757 * current extent. If it is shared with the current extent
2758 * we reset the partial cluster because we've reached the start of the
2759 * truncated/punched region and we're done removing blocks.
2761 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2762 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2763 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2764 int flags = get_default_free_blocks_flags(inode);
2766 if (ext4_is_pending(inode, partial->lblk))
2767 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2768 ext4_free_blocks(handle, inode, NULL,
2769 EXT4_C2B(sbi, partial->pclu),
2770 sbi->s_cluster_ratio, flags);
2771 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2772 ext4_rereserve_cluster(inode, partial->lblk);
2774 partial->state = initial;
2777 /* if this leaf is free, then we should
2778 * remove it from index block above */
2779 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2780 err = ext4_ext_rm_idx(handle, inode, path, depth);
2787 * ext4_ext_more_to_rm:
2788 * returns 1 if current index has to be freed (even partial)
2791 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2793 BUG_ON(path->p_idx == NULL);
2795 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2799 * if truncate on deeper level happened, it wasn't partial,
2800 * so we have to consider current index for truncation
2802 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2807 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2810 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2811 int depth = ext_depth(inode);
2812 struct ext4_ext_path *path = NULL;
2813 struct partial_cluster partial;
2819 partial.state = initial;
2821 ext_debug(inode, "truncate since %u to %u\n", start, end);
2823 /* probably first extent we're gonna free will be last in block */
2824 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2826 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2828 return PTR_ERR(handle);
2831 trace_ext4_ext_remove_space(inode, start, end, depth);
2834 * Check if we are removing extents inside the extent tree. If that
2835 * is the case, we are going to punch a hole inside the extent tree
2836 * so we have to check whether we need to split the extent covering
2837 * the last block to remove so we can easily remove the part of it
2838 * in ext4_ext_rm_leaf().
2840 if (end < EXT_MAX_BLOCKS - 1) {
2841 struct ext4_extent *ex;
2842 ext4_lblk_t ee_block, ex_end, lblk;
2845 /* find extent for or closest extent to this block */
2846 path = ext4_find_extent(inode, end, NULL,
2847 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
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",
2860 err = -EFSCORRUPTED;
2865 ee_block = le32_to_cpu(ex->ee_block);
2866 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2869 * See if the last block is inside the extent, if so split
2870 * the extent at 'end' block so we can easily remove the
2871 * tail of the first part of the split extent in
2872 * ext4_ext_rm_leaf().
2874 if (end >= ee_block && end < ex_end) {
2877 * If we're going to split the extent, note that
2878 * the cluster containing the block after 'end' is
2879 * in use to avoid freeing it when removing blocks.
2881 if (sbi->s_cluster_ratio > 1) {
2882 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2883 partial.pclu = EXT4_B2C(sbi, pblk);
2884 partial.state = nofree;
2888 * Split the extent in two so that 'end' is the last
2889 * block in the first new extent. Also we should not
2890 * fail removing space due to ENOSPC so try to use
2891 * reserved block if that happens.
2893 err = ext4_force_split_extent_at(handle, inode, &path,
2898 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2899 partial.state == initial) {
2901 * If we're punching, there's an extent to the right.
2902 * If the partial cluster hasn't been set, set it to
2903 * that extent's first cluster and its state to nofree
2904 * so it won't be freed should it contain blocks to be
2905 * removed. If it's already set (tofree/nofree), we're
2906 * retrying and keep the original partial cluster info
2907 * so a cluster marked tofree as a result of earlier
2908 * extent removal is not lost.
2911 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2916 partial.pclu = EXT4_B2C(sbi, pblk);
2917 partial.state = nofree;
2922 * We start scanning from right side, freeing all the blocks
2923 * after i_size and walking into the tree depth-wise.
2925 depth = ext_depth(inode);
2930 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2932 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2933 GFP_NOFS | __GFP_NOFAIL);
2935 ext4_journal_stop(handle);
2938 path[0].p_maxdepth = path[0].p_depth = depth;
2939 path[0].p_hdr = ext_inode_hdr(inode);
2942 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2943 err = -EFSCORRUPTED;
2949 while (i >= 0 && err == 0) {
2951 /* this is leaf block */
2952 err = ext4_ext_rm_leaf(handle, inode, path,
2953 &partial, start, end);
2954 /* root level has p_bh == NULL, brelse() eats this */
2955 brelse(path[i].p_bh);
2956 path[i].p_bh = NULL;
2961 /* this is index block */
2962 if (!path[i].p_hdr) {
2963 ext_debug(inode, "initialize header\n");
2964 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2967 if (!path[i].p_idx) {
2968 /* this level hasn't been touched yet */
2969 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2970 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2971 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2973 le16_to_cpu(path[i].p_hdr->eh_entries));
2975 /* we were already here, see at next index */
2979 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2980 i, EXT_FIRST_INDEX(path[i].p_hdr),
2982 if (ext4_ext_more_to_rm(path + i)) {
2983 struct buffer_head *bh;
2984 /* go to the next level */
2985 ext_debug(inode, "move to level %d (block %llu)\n",
2986 i + 1, ext4_idx_pblock(path[i].p_idx));
2987 memset(path + i + 1, 0, sizeof(*path));
2988 bh = read_extent_tree_block(inode, path[i].p_idx,
2992 /* should we reset i_size? */
2996 /* Yield here to deal with large extent trees.
2997 * Should be a no-op if we did IO above. */
2999 if (WARN_ON(i + 1 > depth)) {
3000 err = -EFSCORRUPTED;
3003 path[i + 1].p_bh = bh;
3005 /* save actual number of indexes since this
3006 * number is changed at the next iteration */
3007 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3010 /* we finished processing this index, go up */
3011 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3012 /* index is empty, remove it;
3013 * handle must be already prepared by the
3014 * truncatei_leaf() */
3015 err = ext4_ext_rm_idx(handle, inode, path, i);
3017 /* root level has p_bh == NULL, brelse() eats this */
3018 brelse(path[i].p_bh);
3019 path[i].p_bh = NULL;
3021 ext_debug(inode, "return to level %d\n", i);
3025 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3026 path->p_hdr->eh_entries);
3029 * if there's a partial cluster and we have removed the first extent
3030 * in the file, then we also free the partial cluster, if any
3032 if (partial.state == tofree && err == 0) {
3033 int flags = get_default_free_blocks_flags(inode);
3035 if (ext4_is_pending(inode, partial.lblk))
3036 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3037 ext4_free_blocks(handle, inode, NULL,
3038 EXT4_C2B(sbi, partial.pclu),
3039 sbi->s_cluster_ratio, flags);
3040 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3041 ext4_rereserve_cluster(inode, partial.lblk);
3042 partial.state = initial;
3045 /* TODO: flexible tree reduction should be here */
3046 if (path->p_hdr->eh_entries == 0) {
3048 * truncate to zero freed all the tree,
3049 * so we need to correct eh_depth
3051 err = ext4_ext_get_access(handle, inode, path);
3053 ext_inode_hdr(inode)->eh_depth = 0;
3054 ext_inode_hdr(inode)->eh_max =
3055 cpu_to_le16(ext4_ext_space_root(inode, 0));
3056 err = ext4_ext_dirty(handle, inode, path);
3060 ext4_ext_drop_refs(path);
3065 ext4_journal_stop(handle);
3071 * called at mount time
3073 void ext4_ext_init(struct super_block *sb)
3076 * possible initialization would be here
3079 if (ext4_has_feature_extents(sb)) {
3080 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3081 printk(KERN_INFO "EXT4-fs: file extents enabled"
3082 #ifdef AGGRESSIVE_TEST
3083 ", aggressive tests"
3085 #ifdef CHECK_BINSEARCH
3088 #ifdef EXTENTS_STATS
3093 #ifdef EXTENTS_STATS
3094 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3095 EXT4_SB(sb)->s_ext_min = 1 << 30;
3096 EXT4_SB(sb)->s_ext_max = 0;
3102 * called at umount time
3104 void ext4_ext_release(struct super_block *sb)
3106 if (!ext4_has_feature_extents(sb))
3109 #ifdef EXTENTS_STATS
3110 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3111 struct ext4_sb_info *sbi = EXT4_SB(sb);
3112 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3113 sbi->s_ext_blocks, sbi->s_ext_extents,
3114 sbi->s_ext_blocks / sbi->s_ext_extents);
3115 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3116 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3121 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3123 ext4_lblk_t ee_block;
3124 ext4_fsblk_t ee_pblock;
3125 unsigned int ee_len;
3127 ee_block = le32_to_cpu(ex->ee_block);
3128 ee_len = ext4_ext_get_actual_len(ex);
3129 ee_pblock = ext4_ext_pblock(ex);
3134 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3135 EXTENT_STATUS_WRITTEN);
3138 /* FIXME!! we need to try to merge to left or right after zero-out */
3139 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3141 ext4_fsblk_t ee_pblock;
3142 unsigned int ee_len;
3144 ee_len = ext4_ext_get_actual_len(ex);
3145 ee_pblock = ext4_ext_pblock(ex);
3146 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3151 * ext4_split_extent_at() splits an extent at given block.
3153 * @handle: the journal handle
3154 * @inode: the file inode
3155 * @path: the path to the extent
3156 * @split: the logical block where the extent is splitted.
3157 * @split_flags: indicates if the extent could be zeroout if split fails, and
3158 * the states(init or unwritten) of new extents.
3159 * @flags: flags used to insert new extent to extent tree.
3162 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3163 * of which are determined by split_flag.
3165 * There are two cases:
3166 * a> the extent are splitted into two extent.
3167 * b> split is not needed, and just mark the extent.
3169 * return 0 on success.
3171 static int ext4_split_extent_at(handle_t *handle,
3172 struct inode *inode,
3173 struct ext4_ext_path **ppath,
3178 struct ext4_ext_path *path = *ppath;
3179 ext4_fsblk_t newblock;
3180 ext4_lblk_t ee_block;
3181 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3182 struct ext4_extent *ex2 = NULL;
3183 unsigned int ee_len, depth;
3186 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3187 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3189 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3191 ext4_ext_show_leaf(inode, path);
3193 depth = ext_depth(inode);
3194 ex = path[depth].p_ext;
3195 ee_block = le32_to_cpu(ex->ee_block);
3196 ee_len = ext4_ext_get_actual_len(ex);
3197 newblock = split - ee_block + ext4_ext_pblock(ex);
3199 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3200 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3201 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3202 EXT4_EXT_MARK_UNWRIT1 |
3203 EXT4_EXT_MARK_UNWRIT2));
3205 err = ext4_ext_get_access(handle, inode, path + depth);
3209 if (split == ee_block) {
3211 * case b: block @split is the block that the extent begins with
3212 * then we just change the state of the extent, and splitting
3215 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3216 ext4_ext_mark_unwritten(ex);
3218 ext4_ext_mark_initialized(ex);
3220 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3221 ext4_ext_try_to_merge(handle, inode, path, ex);
3223 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3228 memcpy(&orig_ex, ex, sizeof(orig_ex));
3229 ex->ee_len = cpu_to_le16(split - ee_block);
3230 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3231 ext4_ext_mark_unwritten(ex);
3234 * path may lead to new leaf, not to original leaf any more
3235 * after ext4_ext_insert_extent() returns,
3237 err = ext4_ext_dirty(handle, inode, path + depth);
3239 goto fix_extent_len;
3242 ex2->ee_block = cpu_to_le32(split);
3243 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3244 ext4_ext_store_pblock(ex2, newblock);
3245 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3246 ext4_ext_mark_unwritten(ex2);
3248 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3249 if (err != -ENOSPC && err != -EDQUOT)
3252 if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3253 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3254 if (split_flag & EXT4_EXT_DATA_VALID1) {
3255 err = ext4_ext_zeroout(inode, ex2);
3256 zero_ex.ee_block = ex2->ee_block;
3257 zero_ex.ee_len = cpu_to_le16(
3258 ext4_ext_get_actual_len(ex2));
3259 ext4_ext_store_pblock(&zero_ex,
3260 ext4_ext_pblock(ex2));
3262 err = ext4_ext_zeroout(inode, ex);
3263 zero_ex.ee_block = ex->ee_block;
3264 zero_ex.ee_len = cpu_to_le16(
3265 ext4_ext_get_actual_len(ex));
3266 ext4_ext_store_pblock(&zero_ex,
3267 ext4_ext_pblock(ex));
3270 err = ext4_ext_zeroout(inode, &orig_ex);
3271 zero_ex.ee_block = orig_ex.ee_block;
3272 zero_ex.ee_len = cpu_to_le16(
3273 ext4_ext_get_actual_len(&orig_ex));
3274 ext4_ext_store_pblock(&zero_ex,
3275 ext4_ext_pblock(&orig_ex));
3279 /* update the extent length and mark as initialized */
3280 ex->ee_len = cpu_to_le16(ee_len);
3281 ext4_ext_try_to_merge(handle, inode, path, ex);
3282 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3284 /* update extent status tree */
3285 err = ext4_zeroout_es(inode, &zero_ex);
3286 /* If we failed at this point, we don't know in which
3287 * state the extent tree exactly is so don't try to fix
3288 * length of the original extent as it may do even more
3296 ex->ee_len = orig_ex.ee_len;
3298 * Ignore ext4_ext_dirty return value since we are already in error path
3299 * and err is a non-zero error code.
3301 ext4_ext_dirty(handle, inode, path + path->p_depth);
3304 ext4_ext_show_leaf(inode, path);
3309 * ext4_split_extents() splits an extent and mark extent which is covered
3310 * by @map as split_flags indicates
3312 * It may result in splitting the extent into multiple extents (up to three)
3313 * There are three possibilities:
3314 * a> There is no split required
3315 * b> Splits in two extents: Split is happening at either end of the extent
3316 * c> Splits in three extents: Somone is splitting in middle of the extent
3319 static int ext4_split_extent(handle_t *handle,
3320 struct inode *inode,
3321 struct ext4_ext_path **ppath,
3322 struct ext4_map_blocks *map,
3326 struct ext4_ext_path *path = *ppath;
3327 ext4_lblk_t ee_block;
3328 struct ext4_extent *ex;
3329 unsigned int ee_len, depth;
3332 int split_flag1, flags1;
3333 int allocated = map->m_len;
3335 depth = ext_depth(inode);
3336 ex = path[depth].p_ext;
3337 ee_block = le32_to_cpu(ex->ee_block);
3338 ee_len = ext4_ext_get_actual_len(ex);
3339 unwritten = ext4_ext_is_unwritten(ex);
3341 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3342 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3343 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3345 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3346 EXT4_EXT_MARK_UNWRIT2;
3347 if (split_flag & EXT4_EXT_DATA_VALID2)
3348 split_flag1 |= EXT4_EXT_DATA_VALID1;
3349 err = ext4_split_extent_at(handle, inode, ppath,
3350 map->m_lblk + map->m_len, split_flag1, flags1);
3354 allocated = ee_len - (map->m_lblk - ee_block);
3357 * Update path is required because previous ext4_split_extent_at() may
3358 * result in split of original leaf or extent zeroout.
3360 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3362 return PTR_ERR(path);
3363 depth = ext_depth(inode);
3364 ex = path[depth].p_ext;
3366 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3367 (unsigned long) map->m_lblk);
3368 return -EFSCORRUPTED;
3370 unwritten = ext4_ext_is_unwritten(ex);
3372 if (map->m_lblk >= ee_block) {
3373 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3375 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3376 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3377 EXT4_EXT_MARK_UNWRIT2);
3379 err = ext4_split_extent_at(handle, inode, ppath,
3380 map->m_lblk, split_flag1, flags);
3385 ext4_ext_show_leaf(inode, path);
3387 return err ? err : allocated;
3391 * This function is called by ext4_ext_map_blocks() if someone tries to write
3392 * to an unwritten extent. It may result in splitting the unwritten
3393 * extent into multiple extents (up to three - one initialized and two
3395 * There are three possibilities:
3396 * a> There is no split required: Entire extent should be initialized
3397 * b> Splits in two extents: Write is happening at either end of the extent
3398 * c> Splits in three extents: Somone is writing in middle of the extent
3401 * - The extent pointed to by 'path' is unwritten.
3402 * - The extent pointed to by 'path' contains a superset
3403 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3405 * Post-conditions on success:
3406 * - the returned value is the number of blocks beyond map->l_lblk
3407 * that are allocated and initialized.
3408 * It is guaranteed to be >= map->m_len.
3410 static int ext4_ext_convert_to_initialized(handle_t *handle,
3411 struct inode *inode,
3412 struct ext4_map_blocks *map,
3413 struct ext4_ext_path **ppath,
3416 struct ext4_ext_path *path = *ppath;
3417 struct ext4_sb_info *sbi;
3418 struct ext4_extent_header *eh;
3419 struct ext4_map_blocks split_map;
3420 struct ext4_extent zero_ex1, zero_ex2;
3421 struct ext4_extent *ex, *abut_ex;
3422 ext4_lblk_t ee_block, eof_block;
3423 unsigned int ee_len, depth, map_len = map->m_len;
3424 int allocated = 0, max_zeroout = 0;
3426 int split_flag = EXT4_EXT_DATA_VALID2;
3428 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3429 (unsigned long long)map->m_lblk, map_len);
3431 sbi = EXT4_SB(inode->i_sb);
3432 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3433 >> inode->i_sb->s_blocksize_bits;
3434 if (eof_block < map->m_lblk + map_len)
3435 eof_block = map->m_lblk + map_len;
3437 depth = ext_depth(inode);
3438 eh = path[depth].p_hdr;
3439 ex = path[depth].p_ext;
3440 ee_block = le32_to_cpu(ex->ee_block);
3441 ee_len = ext4_ext_get_actual_len(ex);
3442 zero_ex1.ee_len = 0;
3443 zero_ex2.ee_len = 0;
3445 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3447 /* Pre-conditions */
3448 BUG_ON(!ext4_ext_is_unwritten(ex));
3449 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3452 * Attempt to transfer newly initialized blocks from the currently
3453 * unwritten extent to its neighbor. This is much cheaper
3454 * than an insertion followed by a merge as those involve costly
3455 * memmove() calls. Transferring to the left is the common case in
3456 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3457 * followed by append writes.
3459 * Limitations of the current logic:
3460 * - L1: we do not deal with writes covering the whole extent.
3461 * This would require removing the extent if the transfer
3463 * - L2: we only attempt to merge with an extent stored in the
3464 * same extent tree node.
3466 if ((map->m_lblk == ee_block) &&
3467 /* See if we can merge left */
3468 (map_len < ee_len) && /*L1*/
3469 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3470 ext4_lblk_t prev_lblk;
3471 ext4_fsblk_t prev_pblk, ee_pblk;
3472 unsigned int prev_len;
3475 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3476 prev_len = ext4_ext_get_actual_len(abut_ex);
3477 prev_pblk = ext4_ext_pblock(abut_ex);
3478 ee_pblk = ext4_ext_pblock(ex);
3481 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3482 * upon those conditions:
3483 * - C1: abut_ex is initialized,
3484 * - C2: abut_ex is logically abutting ex,
3485 * - C3: abut_ex is physically abutting ex,
3486 * - C4: abut_ex can receive the additional blocks without
3487 * overflowing the (initialized) length limit.
3489 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3490 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3491 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3492 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3493 err = ext4_ext_get_access(handle, inode, path + depth);
3497 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3500 /* Shift the start of ex by 'map_len' blocks */
3501 ex->ee_block = cpu_to_le32(ee_block + map_len);
3502 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3503 ex->ee_len = cpu_to_le16(ee_len - map_len);
3504 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3506 /* Extend abut_ex by 'map_len' blocks */
3507 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3509 /* Result: number of initialized blocks past m_lblk */
3510 allocated = map_len;
3512 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3513 (map_len < ee_len) && /*L1*/
3514 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3515 /* See if we can merge right */
3516 ext4_lblk_t next_lblk;
3517 ext4_fsblk_t next_pblk, ee_pblk;
3518 unsigned int next_len;
3521 next_lblk = le32_to_cpu(abut_ex->ee_block);
3522 next_len = ext4_ext_get_actual_len(abut_ex);
3523 next_pblk = ext4_ext_pblock(abut_ex);
3524 ee_pblk = ext4_ext_pblock(ex);
3527 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3528 * upon those conditions:
3529 * - C1: abut_ex is initialized,
3530 * - C2: abut_ex is logically abutting ex,
3531 * - C3: abut_ex is physically abutting ex,
3532 * - C4: abut_ex can receive the additional blocks without
3533 * overflowing the (initialized) length limit.
3535 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3536 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3537 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3538 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3539 err = ext4_ext_get_access(handle, inode, path + depth);
3543 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3546 /* Shift the start of abut_ex by 'map_len' blocks */
3547 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3548 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3549 ex->ee_len = cpu_to_le16(ee_len - map_len);
3550 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3552 /* Extend abut_ex by 'map_len' blocks */
3553 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3555 /* Result: number of initialized blocks past m_lblk */
3556 allocated = map_len;
3560 /* Mark the block containing both extents as dirty */
3561 err = ext4_ext_dirty(handle, inode, path + depth);
3563 /* Update path to point to the right extent */
3564 path[depth].p_ext = abut_ex;
3567 allocated = ee_len - (map->m_lblk - ee_block);
3569 WARN_ON(map->m_lblk < ee_block);
3571 * It is safe to convert extent to initialized via explicit
3572 * zeroout only if extent is fully inside i_size or new_size.
3574 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3576 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3577 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3578 (inode->i_sb->s_blocksize_bits - 10);
3582 * 1. split the extent into three extents.
3583 * 2. split the extent into two extents, zeroout the head of the first
3585 * 3. split the extent into two extents, zeroout the tail of the second
3587 * 4. split the extent into two extents with out zeroout.
3588 * 5. no splitting needed, just possibly zeroout the head and / or the
3589 * tail of the extent.
3591 split_map.m_lblk = map->m_lblk;
3592 split_map.m_len = map->m_len;
3594 if (max_zeroout && (allocated > split_map.m_len)) {
3595 if (allocated <= max_zeroout) {
3598 cpu_to_le32(split_map.m_lblk +
3601 cpu_to_le16(allocated - split_map.m_len);
3602 ext4_ext_store_pblock(&zero_ex1,
3603 ext4_ext_pblock(ex) + split_map.m_lblk +
3604 split_map.m_len - ee_block);
3605 err = ext4_ext_zeroout(inode, &zero_ex1);
3608 split_map.m_len = allocated;
3610 if (split_map.m_lblk - ee_block + split_map.m_len <
3613 if (split_map.m_lblk != ee_block) {
3614 zero_ex2.ee_block = ex->ee_block;
3615 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3617 ext4_ext_store_pblock(&zero_ex2,
3618 ext4_ext_pblock(ex));
3619 err = ext4_ext_zeroout(inode, &zero_ex2);
3624 split_map.m_len += split_map.m_lblk - ee_block;
3625 split_map.m_lblk = ee_block;
3626 allocated = map->m_len;
3631 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3636 /* If we have gotten a failure, don't zero out status tree */
3638 err = ext4_zeroout_es(inode, &zero_ex1);
3640 err = ext4_zeroout_es(inode, &zero_ex2);
3642 return err ? err : allocated;
3646 * This function is called by ext4_ext_map_blocks() from
3647 * ext4_get_blocks_dio_write() when DIO to write
3648 * to an unwritten extent.
3650 * Writing to an unwritten extent may result in splitting the unwritten
3651 * extent into multiple initialized/unwritten extents (up to three)
3652 * There are three possibilities:
3653 * a> There is no split required: Entire extent should be unwritten
3654 * b> Splits in two extents: Write is happening at either end of the extent
3655 * c> Splits in three extents: Somone is writing in middle of the extent
3657 * This works the same way in the case of initialized -> unwritten conversion.
3659 * One of more index blocks maybe needed if the extent tree grow after
3660 * the unwritten extent split. To prevent ENOSPC occur at the IO
3661 * complete, we need to split the unwritten extent before DIO submit
3662 * the IO. The unwritten extent called at this time will be split
3663 * into three unwritten extent(at most). After IO complete, the part
3664 * being filled will be convert to initialized by the end_io callback function
3665 * via ext4_convert_unwritten_extents().
3667 * Returns the size of unwritten extent to be written on success.
3669 static int ext4_split_convert_extents(handle_t *handle,
3670 struct inode *inode,
3671 struct ext4_map_blocks *map,
3672 struct ext4_ext_path **ppath,
3675 struct ext4_ext_path *path = *ppath;
3676 ext4_lblk_t eof_block;
3677 ext4_lblk_t ee_block;
3678 struct ext4_extent *ex;
3679 unsigned int ee_len;
3680 int split_flag = 0, depth;
3682 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3683 (unsigned long long)map->m_lblk, map->m_len);
3685 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3686 >> inode->i_sb->s_blocksize_bits;
3687 if (eof_block < map->m_lblk + map->m_len)
3688 eof_block = map->m_lblk + map->m_len;
3690 * It is safe to convert extent to initialized via explicit
3691 * zeroout only if extent is fully inside i_size or new_size.
3693 depth = ext_depth(inode);
3694 ex = path[depth].p_ext;
3695 ee_block = le32_to_cpu(ex->ee_block);
3696 ee_len = ext4_ext_get_actual_len(ex);
3698 /* Convert to unwritten */
3699 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3700 split_flag |= EXT4_EXT_DATA_VALID1;
3701 /* Convert to initialized */
3702 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3703 split_flag |= ee_block + ee_len <= eof_block ?
3704 EXT4_EXT_MAY_ZEROOUT : 0;
3705 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3707 flags |= EXT4_GET_BLOCKS_PRE_IO;
3708 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3711 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3712 struct inode *inode,
3713 struct ext4_map_blocks *map,
3714 struct ext4_ext_path **ppath)
3716 struct ext4_ext_path *path = *ppath;
3717 struct ext4_extent *ex;
3718 ext4_lblk_t ee_block;
3719 unsigned int ee_len;
3723 depth = ext_depth(inode);
3724 ex = path[depth].p_ext;
3725 ee_block = le32_to_cpu(ex->ee_block);
3726 ee_len = ext4_ext_get_actual_len(ex);
3728 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3729 (unsigned long long)ee_block, ee_len);
3731 /* If extent is larger than requested it is a clear sign that we still
3732 * have some extent state machine issues left. So extent_split is still
3734 * TODO: Once all related issues will be fixed this situation should be
3737 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3738 #ifdef CONFIG_EXT4_DEBUG
3739 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3740 " len %u; IO logical block %llu, len %u",
3741 inode->i_ino, (unsigned long long)ee_block, ee_len,
3742 (unsigned long long)map->m_lblk, map->m_len);
3744 err = ext4_split_convert_extents(handle, inode, map, ppath,
3745 EXT4_GET_BLOCKS_CONVERT);
3748 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3750 return PTR_ERR(path);
3751 depth = ext_depth(inode);
3752 ex = path[depth].p_ext;
3755 err = ext4_ext_get_access(handle, inode, path + depth);
3758 /* first mark the extent as initialized */
3759 ext4_ext_mark_initialized(ex);
3761 /* note: ext4_ext_correct_indexes() isn't needed here because
3762 * borders are not changed
3764 ext4_ext_try_to_merge(handle, inode, path, ex);
3766 /* Mark modified extent as dirty */
3767 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3769 ext4_ext_show_leaf(inode, path);
3774 convert_initialized_extent(handle_t *handle, struct inode *inode,
3775 struct ext4_map_blocks *map,
3776 struct ext4_ext_path **ppath,
3777 unsigned int *allocated)
3779 struct ext4_ext_path *path = *ppath;
3780 struct ext4_extent *ex;
3781 ext4_lblk_t ee_block;
3782 unsigned int ee_len;
3787 * Make sure that the extent is no bigger than we support with
3790 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3791 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3793 depth = ext_depth(inode);
3794 ex = path[depth].p_ext;
3795 ee_block = le32_to_cpu(ex->ee_block);
3796 ee_len = ext4_ext_get_actual_len(ex);
3798 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3799 (unsigned long long)ee_block, ee_len);
3801 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3802 err = ext4_split_convert_extents(handle, inode, map, ppath,
3803 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3806 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3808 return PTR_ERR(path);
3809 depth = ext_depth(inode);
3810 ex = path[depth].p_ext;
3812 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3813 (unsigned long) map->m_lblk);
3814 return -EFSCORRUPTED;
3818 err = ext4_ext_get_access(handle, inode, path + depth);
3821 /* first mark the extent as unwritten */
3822 ext4_ext_mark_unwritten(ex);
3824 /* note: ext4_ext_correct_indexes() isn't needed here because
3825 * borders are not changed
3827 ext4_ext_try_to_merge(handle, inode, path, ex);
3829 /* Mark modified extent as dirty */
3830 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3833 ext4_ext_show_leaf(inode, path);
3835 ext4_update_inode_fsync_trans(handle, inode, 1);
3837 map->m_flags |= EXT4_MAP_UNWRITTEN;
3838 if (*allocated > map->m_len)
3839 *allocated = map->m_len;
3840 map->m_len = *allocated;
3845 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3846 struct ext4_map_blocks *map,
3847 struct ext4_ext_path **ppath, int flags,
3848 unsigned int allocated, ext4_fsblk_t newblock)
3850 struct ext4_ext_path __maybe_unused *path = *ppath;
3854 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3855 (unsigned long long)map->m_lblk, map->m_len, flags,
3857 ext4_ext_show_leaf(inode, path);
3860 * When writing into unwritten space, we should not fail to
3861 * allocate metadata blocks for the new extent block if needed.
3863 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3865 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3866 allocated, newblock);
3868 /* get_block() before submitting IO, split the extent */
3869 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3870 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3871 flags | EXT4_GET_BLOCKS_CONVERT);
3877 * shouldn't get a 0 return when splitting an extent unless
3878 * m_len is 0 (bug) or extent has been corrupted
3880 if (unlikely(ret == 0)) {
3881 EXT4_ERROR_INODE(inode,
3882 "unexpected ret == 0, m_len = %u",
3884 err = -EFSCORRUPTED;
3887 map->m_flags |= EXT4_MAP_UNWRITTEN;
3890 /* IO end_io complete, convert the filled extent to written */
3891 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3892 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3896 ext4_update_inode_fsync_trans(handle, inode, 1);
3899 /* buffered IO cases */
3901 * repeat fallocate creation request
3902 * we already have an unwritten extent
3904 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3905 map->m_flags |= EXT4_MAP_UNWRITTEN;
3909 /* buffered READ or buffered write_begin() lookup */
3910 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3912 * We have blocks reserved already. We
3913 * return allocated blocks so that delalloc
3914 * won't do block reservation for us. But
3915 * the buffer head will be unmapped so that
3916 * a read from the block returns 0s.
3918 map->m_flags |= EXT4_MAP_UNWRITTEN;
3923 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3924 * For buffered writes, at writepage time, etc. Convert a
3925 * discovered unwritten extent to written.
3927 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3932 ext4_update_inode_fsync_trans(handle, inode, 1);
3934 * shouldn't get a 0 return when converting an unwritten extent
3935 * unless m_len is 0 (bug) or extent has been corrupted
3937 if (unlikely(ret == 0)) {
3938 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3940 err = -EFSCORRUPTED;
3946 map->m_flags |= EXT4_MAP_NEW;
3948 map->m_flags |= EXT4_MAP_MAPPED;
3950 map->m_pblk = newblock;
3951 if (allocated > map->m_len)
3952 allocated = map->m_len;
3953 map->m_len = allocated;
3954 ext4_ext_show_leaf(inode, path);
3956 return err ? err : allocated;
3960 * get_implied_cluster_alloc - check to see if the requested
3961 * allocation (in the map structure) overlaps with a cluster already
3962 * allocated in an extent.
3963 * @sb The filesystem superblock structure
3964 * @map The requested lblk->pblk mapping
3965 * @ex The extent structure which might contain an implied
3966 * cluster allocation
3968 * This function is called by ext4_ext_map_blocks() after we failed to
3969 * find blocks that were already in the inode's extent tree. Hence,
3970 * we know that the beginning of the requested region cannot overlap
3971 * the extent from the inode's extent tree. There are three cases we
3972 * want to catch. The first is this case:
3974 * |--- cluster # N--|
3975 * |--- extent ---| |---- requested region ---|
3978 * The second case that we need to test for is this one:
3980 * |--------- cluster # N ----------------|
3981 * |--- requested region --| |------- extent ----|
3982 * |=======================|
3984 * The third case is when the requested region lies between two extents
3985 * within the same cluster:
3986 * |------------- cluster # N-------------|
3987 * |----- ex -----| |---- ex_right ----|
3988 * |------ requested region ------|
3989 * |================|
3991 * In each of the above cases, we need to set the map->m_pblk and
3992 * map->m_len so it corresponds to the return the extent labelled as
3993 * "|====|" from cluster #N, since it is already in use for data in
3994 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3995 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3996 * as a new "allocated" block region. Otherwise, we will return 0 and
3997 * ext4_ext_map_blocks() will then allocate one or more new clusters
3998 * by calling ext4_mb_new_blocks().
4000 static int get_implied_cluster_alloc(struct super_block *sb,
4001 struct ext4_map_blocks *map,
4002 struct ext4_extent *ex,
4003 struct ext4_ext_path *path)
4005 struct ext4_sb_info *sbi = EXT4_SB(sb);
4006 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4007 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4008 ext4_lblk_t rr_cluster_start;
4009 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4010 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4011 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4013 /* The extent passed in that we are trying to match */
4014 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4015 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4017 /* The requested region passed into ext4_map_blocks() */
4018 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4020 if ((rr_cluster_start == ex_cluster_end) ||
4021 (rr_cluster_start == ex_cluster_start)) {
4022 if (rr_cluster_start == ex_cluster_end)
4023 ee_start += ee_len - 1;
4024 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4025 map->m_len = min(map->m_len,
4026 (unsigned) sbi->s_cluster_ratio - c_offset);
4028 * Check for and handle this case:
4030 * |--------- cluster # N-------------|
4031 * |------- extent ----|
4032 * |--- requested region ---|
4036 if (map->m_lblk < ee_block)
4037 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4040 * Check for the case where there is already another allocated
4041 * block to the right of 'ex' but before the end of the cluster.
4043 * |------------- cluster # N-------------|
4044 * |----- ex -----| |---- ex_right ----|
4045 * |------ requested region ------|
4046 * |================|
4048 if (map->m_lblk > ee_block) {
4049 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4050 map->m_len = min(map->m_len, next - map->m_lblk);
4053 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4057 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4063 * Block allocation/map/preallocation routine for extents based files
4066 * Need to be called with
4067 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4068 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4070 * return > 0, number of blocks already mapped/allocated
4071 * if create == 0 and these are pre-allocated blocks
4072 * buffer head is unmapped
4073 * otherwise blocks are mapped
4075 * return = 0, if plain look up failed (blocks have not been allocated)
4076 * buffer head is unmapped
4078 * return < 0, error case.
4080 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4081 struct ext4_map_blocks *map, int flags)
4083 struct ext4_ext_path *path = NULL;
4084 struct ext4_extent newex, *ex, ex2;
4085 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4086 ext4_fsblk_t newblock = 0, pblk;
4087 int err = 0, depth, ret;
4088 unsigned int allocated = 0, offset = 0;
4089 unsigned int allocated_clusters = 0;
4090 struct ext4_allocation_request ar;
4091 ext4_lblk_t cluster_offset;
4093 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4094 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4096 /* find extent for this block */
4097 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4099 err = PTR_ERR(path);
4104 depth = ext_depth(inode);
4107 * consistent leaf must not be empty;
4108 * this situation is possible, though, _during_ tree modification;
4109 * this is why assert can't be put in ext4_find_extent()
4111 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4112 EXT4_ERROR_INODE(inode, "bad extent address "
4113 "lblock: %lu, depth: %d pblock %lld",
4114 (unsigned long) map->m_lblk, depth,
4115 path[depth].p_block);
4116 err = -EFSCORRUPTED;
4120 ex = path[depth].p_ext;
4122 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4123 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4124 unsigned short ee_len;
4128 * unwritten extents are treated as holes, except that
4129 * we split out initialized portions during a write.
4131 ee_len = ext4_ext_get_actual_len(ex);
4133 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4135 /* if found extent covers block, simply return it */
4136 if (in_range(map->m_lblk, ee_block, ee_len)) {
4137 newblock = map->m_lblk - ee_block + ee_start;
4138 /* number of remaining blocks in the extent */
4139 allocated = ee_len - (map->m_lblk - ee_block);
4140 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4141 map->m_lblk, ee_block, ee_len, newblock);
4144 * If the extent is initialized check whether the
4145 * caller wants to convert it to unwritten.
4147 if ((!ext4_ext_is_unwritten(ex)) &&
4148 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4149 err = convert_initialized_extent(handle,
4150 inode, map, &path, &allocated);
4152 } else if (!ext4_ext_is_unwritten(ex)) {
4153 map->m_flags |= EXT4_MAP_MAPPED;
4154 map->m_pblk = newblock;
4155 if (allocated > map->m_len)
4156 allocated = map->m_len;
4157 map->m_len = allocated;
4158 ext4_ext_show_leaf(inode, path);
4162 ret = ext4_ext_handle_unwritten_extents(
4163 handle, inode, map, &path, flags,
4164 allocated, newblock);
4174 * requested block isn't allocated yet;
4175 * we couldn't try to create block if create flag is zero
4177 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4178 ext4_lblk_t hole_start, hole_len;
4180 hole_start = map->m_lblk;
4181 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4183 * put just found gap into cache to speed up
4184 * subsequent requests
4186 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4188 /* Update hole_len to reflect hole size after map->m_lblk */
4189 if (hole_start != map->m_lblk)
4190 hole_len -= map->m_lblk - hole_start;
4192 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4198 * Okay, we need to do block allocation.
4200 newex.ee_block = cpu_to_le32(map->m_lblk);
4201 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4204 * If we are doing bigalloc, check to see if the extent returned
4205 * by ext4_find_extent() implies a cluster we can use.
4207 if (cluster_offset && ex &&
4208 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4209 ar.len = allocated = map->m_len;
4210 newblock = map->m_pblk;
4211 goto got_allocated_blocks;
4214 /* find neighbour allocated blocks */
4215 ar.lleft = map->m_lblk;
4216 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4219 ar.lright = map->m_lblk;
4220 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4224 /* Check if the extent after searching to the right implies a
4225 * cluster we can use. */
4226 if ((sbi->s_cluster_ratio > 1) && err &&
4227 get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4228 ar.len = allocated = map->m_len;
4229 newblock = map->m_pblk;
4230 goto got_allocated_blocks;
4234 * See if request is beyond maximum number of blocks we can have in
4235 * a single extent. For an initialized extent this limit is
4236 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4237 * EXT_UNWRITTEN_MAX_LEN.
4239 if (map->m_len > EXT_INIT_MAX_LEN &&
4240 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4241 map->m_len = EXT_INIT_MAX_LEN;
4242 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4243 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4244 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4246 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4247 newex.ee_len = cpu_to_le16(map->m_len);
4248 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4250 allocated = ext4_ext_get_actual_len(&newex);
4252 allocated = map->m_len;
4254 /* allocate new block */
4256 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4257 ar.logical = map->m_lblk;
4259 * We calculate the offset from the beginning of the cluster
4260 * for the logical block number, since when we allocate a
4261 * physical cluster, the physical block should start at the
4262 * same offset from the beginning of the cluster. This is
4263 * needed so that future calls to get_implied_cluster_alloc()
4266 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4267 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4269 ar.logical -= offset;
4270 if (S_ISREG(inode->i_mode))
4271 ar.flags = EXT4_MB_HINT_DATA;
4273 /* disable in-core preallocation for non-regular files */
4275 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4276 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4277 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4278 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4279 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4280 ar.flags |= EXT4_MB_USE_RESERVED;
4281 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4284 allocated_clusters = ar.len;
4285 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4286 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4287 ar.goal, newblock, ar.len, allocated);
4288 if (ar.len > allocated)
4291 got_allocated_blocks:
4292 /* try to insert new extent into found leaf and return */
4293 pblk = newblock + offset;
4294 ext4_ext_store_pblock(&newex, pblk);
4295 newex.ee_len = cpu_to_le16(ar.len);
4296 /* Mark unwritten */
4297 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4298 ext4_ext_mark_unwritten(&newex);
4299 map->m_flags |= EXT4_MAP_UNWRITTEN;
4302 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4304 if (allocated_clusters) {
4308 * free data blocks we just allocated.
4309 * not a good idea to call discard here directly,
4310 * but otherwise we'd need to call it every free().
4312 ext4_discard_preallocations(inode, 0);
4313 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4314 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4315 ext4_free_blocks(handle, inode, NULL, newblock,
4316 EXT4_C2B(sbi, allocated_clusters),
4323 * Reduce the reserved cluster count to reflect successful deferred
4324 * allocation of delayed allocated clusters or direct allocation of
4325 * clusters discovered to be delayed allocated. Once allocated, a
4326 * cluster is not included in the reserved count.
4328 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4329 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4331 * When allocating delayed allocated clusters, simply
4332 * reduce the reserved cluster count and claim quota
4334 ext4_da_update_reserve_space(inode, allocated_clusters,
4337 ext4_lblk_t lblk, len;
4341 * When allocating non-delayed allocated clusters
4342 * (from fallocate, filemap, DIO, or clusters
4343 * allocated when delalloc has been disabled by
4344 * ext4_nonda_switch), reduce the reserved cluster
4345 * count by the number of allocated clusters that
4346 * have previously been delayed allocated. Quota
4347 * has been claimed by ext4_mb_new_blocks() above,
4348 * so release the quota reservations made for any
4349 * previously delayed allocated clusters.
4351 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4352 len = allocated_clusters << sbi->s_cluster_bits;
4353 n = ext4_es_delayed_clu(inode, lblk, len);
4355 ext4_da_update_reserve_space(inode, (int) n, 0);
4360 * Cache the extent and update transaction to commit on fdatasync only
4361 * when it is _not_ an unwritten extent.
4363 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4364 ext4_update_inode_fsync_trans(handle, inode, 1);
4366 ext4_update_inode_fsync_trans(handle, inode, 0);
4368 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4370 map->m_len = ar.len;
4371 allocated = map->m_len;
4372 ext4_ext_show_leaf(inode, path);
4374 ext4_ext_drop_refs(path);
4377 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4378 err ? err : allocated);
4379 return err ? err : allocated;
4382 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4384 struct super_block *sb = inode->i_sb;
4385 ext4_lblk_t last_block;
4389 * TODO: optimization is possible here.
4390 * Probably we need not scan at all,
4391 * because page truncation is enough.
4394 /* we have to know where to truncate from in crash case */
4395 EXT4_I(inode)->i_disksize = inode->i_size;
4396 err = ext4_mark_inode_dirty(handle, inode);
4400 last_block = (inode->i_size + sb->s_blocksize - 1)
4401 >> EXT4_BLOCK_SIZE_BITS(sb);
4403 err = ext4_es_remove_extent(inode, last_block,
4404 EXT_MAX_BLOCKS - last_block);
4405 if (err == -ENOMEM) {
4406 memalloc_retry_wait(GFP_ATOMIC);
4412 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4413 if (err == -ENOMEM) {
4414 memalloc_retry_wait(GFP_ATOMIC);
4415 goto retry_remove_space;
4420 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4421 ext4_lblk_t len, loff_t new_size,
4424 struct inode *inode = file_inode(file);
4426 int ret = 0, ret2 = 0, ret3 = 0;
4429 struct ext4_map_blocks map;
4430 unsigned int credits;
4433 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4434 map.m_lblk = offset;
4437 * Don't normalize the request if it can fit in one extent so
4438 * that it doesn't get unnecessarily split into multiple
4441 if (len <= EXT_UNWRITTEN_MAX_LEN)
4442 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4445 * credits to insert 1 extent into extent tree
4447 credits = ext4_chunk_trans_blocks(inode, len);
4448 depth = ext_depth(inode);
4453 * Recalculate credits when extent tree depth changes.
4455 if (depth != ext_depth(inode)) {
4456 credits = ext4_chunk_trans_blocks(inode, len);
4457 depth = ext_depth(inode);
4460 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4462 if (IS_ERR(handle)) {
4463 ret = PTR_ERR(handle);
4466 ret = ext4_map_blocks(handle, inode, &map, flags);
4468 ext4_debug("inode #%lu: block %u: len %u: "
4469 "ext4_ext_map_blocks returned %d",
4470 inode->i_ino, map.m_lblk,
4472 ext4_mark_inode_dirty(handle, inode);
4473 ext4_journal_stop(handle);
4477 * allow a full retry cycle for any remaining allocations
4481 map.m_len = len = len - ret;
4482 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4483 inode->i_ctime = current_time(inode);
4485 if (epos > new_size)
4487 if (ext4_update_inode_size(inode, epos) & 0x1)
4488 inode->i_mtime = inode->i_ctime;
4490 ret2 = ext4_mark_inode_dirty(handle, inode);
4491 ext4_update_inode_fsync_trans(handle, inode, 1);
4492 ret3 = ext4_journal_stop(handle);
4493 ret2 = ret3 ? ret3 : ret2;
4497 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4500 return ret > 0 ? ret2 : ret;
4503 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4505 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4507 static long ext4_zero_range(struct file *file, loff_t offset,
4508 loff_t len, int mode)
4510 struct inode *inode = file_inode(file);
4511 struct address_space *mapping = file->f_mapping;
4512 handle_t *handle = NULL;
4513 unsigned int max_blocks;
4514 loff_t new_size = 0;
4518 int partial_begin, partial_end;
4521 unsigned int blkbits = inode->i_blkbits;
4523 trace_ext4_zero_range(inode, offset, len, mode);
4525 /* Call ext4_force_commit to flush all data in case of data=journal. */
4526 if (ext4_should_journal_data(inode)) {
4527 ret = ext4_force_commit(inode->i_sb);
4533 * Round up offset. This is not fallocate, we need to zero out
4534 * blocks, so convert interior block aligned part of the range to
4535 * unwritten and possibly manually zero out unaligned parts of the
4538 start = round_up(offset, 1 << blkbits);
4539 end = round_down((offset + len), 1 << blkbits);
4541 if (start < offset || end > offset + len)
4543 partial_begin = offset & ((1 << blkbits) - 1);
4544 partial_end = (offset + len) & ((1 << blkbits) - 1);
4546 lblk = start >> blkbits;
4547 max_blocks = (end >> blkbits);
4548 if (max_blocks < lblk)
4556 * Indirect files do not support unwritten extents
4558 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4563 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4564 (offset + len > inode->i_size ||
4565 offset + len > EXT4_I(inode)->i_disksize)) {
4566 new_size = offset + len;
4567 ret = inode_newsize_ok(inode, new_size);
4572 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4574 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4575 inode_dio_wait(inode);
4577 ret = file_modified(file);
4581 /* Preallocate the range including the unaligned edges */
4582 if (partial_begin || partial_end) {
4583 ret = ext4_alloc_file_blocks(file,
4584 round_down(offset, 1 << blkbits) >> blkbits,
4585 (round_up((offset + len), 1 << blkbits) -
4586 round_down(offset, 1 << blkbits)) >> blkbits,
4593 /* Zero range excluding the unaligned edges */
4594 if (max_blocks > 0) {
4595 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4599 * Prevent page faults from reinstantiating pages we have
4600 * released from page cache.
4602 filemap_invalidate_lock(mapping);
4604 ret = ext4_break_layouts(inode);
4606 filemap_invalidate_unlock(mapping);
4610 ret = ext4_update_disksize_before_punch(inode, offset, len);
4612 filemap_invalidate_unlock(mapping);
4615 /* Now release the pages and zero block aligned part of pages */
4616 truncate_pagecache_range(inode, start, end - 1);
4617 inode->i_mtime = inode->i_ctime = current_time(inode);
4619 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4621 filemap_invalidate_unlock(mapping);
4625 if (!partial_begin && !partial_end)
4629 * In worst case we have to writeout two nonadjacent unwritten
4630 * blocks and update the inode
4632 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4633 if (ext4_should_journal_data(inode))
4635 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4636 if (IS_ERR(handle)) {
4637 ret = PTR_ERR(handle);
4638 ext4_std_error(inode->i_sb, ret);
4642 inode->i_mtime = inode->i_ctime = current_time(inode);
4644 ext4_update_inode_size(inode, new_size);
4645 ret = ext4_mark_inode_dirty(handle, inode);
4648 /* Zero out partial block at the edges of the range */
4649 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4651 ext4_update_inode_fsync_trans(handle, inode, 1);
4653 if (file->f_flags & O_SYNC)
4654 ext4_handle_sync(handle);
4657 ext4_journal_stop(handle);
4659 inode_unlock(inode);
4664 * preallocate space for a file. This implements ext4's fallocate file
4665 * operation, which gets called from sys_fallocate system call.
4666 * For block-mapped files, posix_fallocate should fall back to the method
4667 * of writing zeroes to the required new blocks (the same behavior which is
4668 * expected for file systems which do not support fallocate() system call).
4670 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4672 struct inode *inode = file_inode(file);
4673 loff_t new_size = 0;
4674 unsigned int max_blocks;
4678 unsigned int blkbits = inode->i_blkbits;
4681 * Encrypted inodes can't handle collapse range or insert
4682 * range since we would need to re-encrypt blocks with a
4683 * different IV or XTS tweak (which are based on the logical
4686 if (IS_ENCRYPTED(inode) &&
4687 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4690 /* Return error if mode is not supported */
4691 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4692 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4693 FALLOC_FL_INSERT_RANGE))
4697 ret = ext4_convert_inline_data(inode);
4698 inode_unlock(inode);
4702 if (mode & FALLOC_FL_PUNCH_HOLE) {
4703 ret = ext4_punch_hole(file, offset, len);
4707 if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4708 ret = ext4_collapse_range(file, offset, len);
4712 if (mode & FALLOC_FL_INSERT_RANGE) {
4713 ret = ext4_insert_range(file, offset, len);
4717 if (mode & FALLOC_FL_ZERO_RANGE) {
4718 ret = ext4_zero_range(file, offset, len, mode);
4721 trace_ext4_fallocate_enter(inode, offset, len, mode);
4722 lblk = offset >> blkbits;
4724 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4725 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4730 * We only support preallocation for extent-based files only
4732 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4737 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4738 (offset + len > inode->i_size ||
4739 offset + len > EXT4_I(inode)->i_disksize)) {
4740 new_size = offset + len;
4741 ret = inode_newsize_ok(inode, new_size);
4746 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4747 inode_dio_wait(inode);
4749 ret = file_modified(file);
4753 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4757 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4758 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4759 EXT4_I(inode)->i_sync_tid);
4762 inode_unlock(inode);
4763 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4769 * This function convert a range of blocks to written extents
4770 * The caller of this function will pass the start offset and the size.
4771 * all unwritten extents within this range will be converted to
4774 * This function is called from the direct IO end io call back
4775 * function, to convert the fallocated extents after IO is completed.
4776 * Returns 0 on success.
4778 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4779 loff_t offset, ssize_t len)
4781 unsigned int max_blocks;
4782 int ret = 0, ret2 = 0, ret3 = 0;
4783 struct ext4_map_blocks map;
4784 unsigned int blkbits = inode->i_blkbits;
4785 unsigned int credits = 0;
4787 map.m_lblk = offset >> blkbits;
4788 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4792 * credits to insert 1 extent into extent tree
4794 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4796 while (ret >= 0 && ret < max_blocks) {
4798 map.m_len = (max_blocks -= ret);
4800 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4802 if (IS_ERR(handle)) {
4803 ret = PTR_ERR(handle);
4807 ret = ext4_map_blocks(handle, inode, &map,
4808 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4810 ext4_warning(inode->i_sb,
4811 "inode #%lu: block %u: len %u: "
4812 "ext4_ext_map_blocks returned %d",
4813 inode->i_ino, map.m_lblk,
4815 ret2 = ext4_mark_inode_dirty(handle, inode);
4817 ret3 = ext4_journal_stop(handle);
4822 if (ret <= 0 || ret2)
4825 return ret > 0 ? ret2 : ret;
4828 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4830 int ret = 0, err = 0;
4831 struct ext4_io_end_vec *io_end_vec;
4834 * This is somewhat ugly but the idea is clear: When transaction is
4835 * reserved, everything goes into it. Otherwise we rather start several
4836 * smaller transactions for conversion of each extent separately.
4839 handle = ext4_journal_start_reserved(handle,
4840 EXT4_HT_EXT_CONVERT);
4842 return PTR_ERR(handle);
4845 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4846 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4854 err = ext4_journal_stop(handle);
4856 return ret < 0 ? ret : err;
4859 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4863 int blockbits = inode->i_sb->s_blocksize_bits;
4868 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4869 struct ext4_iloc iloc;
4870 int offset; /* offset of xattr in inode */
4872 error = ext4_get_inode_loc(inode, &iloc);
4875 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4876 offset = EXT4_GOOD_OLD_INODE_SIZE +
4877 EXT4_I(inode)->i_extra_isize;
4879 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4881 iomap_type = IOMAP_INLINE;
4882 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4883 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4884 length = inode->i_sb->s_blocksize;
4885 iomap_type = IOMAP_MAPPED;
4887 /* no in-inode or external block for xattr, so return -ENOENT */
4892 iomap->addr = physical;
4894 iomap->length = length;
4895 iomap->type = iomap_type;
4901 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4902 loff_t length, unsigned flags,
4903 struct iomap *iomap, struct iomap *srcmap)
4907 error = ext4_iomap_xattr_fiemap(inode, iomap);
4908 if (error == 0 && (offset >= iomap->length))
4913 static const struct iomap_ops ext4_iomap_xattr_ops = {
4914 .iomap_begin = ext4_iomap_xattr_begin,
4917 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4921 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4922 maxbytes = inode->i_sb->s_maxbytes;
4924 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4928 if (start > maxbytes)
4932 * Shrink request scope to what the fs can actually handle.
4934 if (*len > maxbytes || (maxbytes - *len) < start)
4935 *len = maxbytes - start;
4939 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4944 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4945 error = ext4_ext_precache(inode);
4948 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4952 * For bitmap files the maximum size limit could be smaller than
4953 * s_maxbytes, so check len here manually instead of just relying on the
4956 error = ext4_fiemap_check_ranges(inode, start, &len);
4960 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4961 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4962 return iomap_fiemap(inode, fieinfo, start, len,
4963 &ext4_iomap_xattr_ops);
4966 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4969 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4970 __u64 start, __u64 len)
4972 ext4_lblk_t start_blk, len_blks;
4976 if (ext4_has_inline_data(inode)) {
4979 down_read(&EXT4_I(inode)->xattr_sem);
4980 has_inline = ext4_has_inline_data(inode);
4981 up_read(&EXT4_I(inode)->xattr_sem);
4986 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4987 error = ext4_ext_precache(inode);
4990 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4993 error = fiemap_prep(inode, fieinfo, start, &len, 0);
4997 error = ext4_fiemap_check_ranges(inode, start, &len);
5001 start_blk = start >> inode->i_sb->s_blocksize_bits;
5002 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5003 if (last_blk >= EXT_MAX_BLOCKS)
5004 last_blk = EXT_MAX_BLOCKS-1;
5005 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5008 * Walk the extent tree gathering extent information
5009 * and pushing extents back to the user.
5011 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5015 * ext4_ext_shift_path_extents:
5016 * Shift the extents of a path structure lying between path[depth].p_ext
5017 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5018 * if it is right shift or left shift operation.
5021 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5022 struct inode *inode, handle_t *handle,
5023 enum SHIFT_DIRECTION SHIFT)
5026 struct ext4_extent *ex_start, *ex_last;
5027 bool update = false;
5028 int credits, restart_credits;
5029 depth = path->p_depth;
5031 while (depth >= 0) {
5032 if (depth == path->p_depth) {
5033 ex_start = path[depth].p_ext;
5035 return -EFSCORRUPTED;
5037 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5038 /* leaf + sb + inode */
5040 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5042 /* extent tree + sb + inode */
5043 credits = depth + 2;
5046 restart_credits = ext4_writepage_trans_blocks(inode);
5047 err = ext4_datasem_ensure_credits(handle, inode, credits,
5048 restart_credits, 0);
5055 err = ext4_ext_get_access(handle, inode, path + depth);
5059 while (ex_start <= ex_last) {
5060 if (SHIFT == SHIFT_LEFT) {
5061 le32_add_cpu(&ex_start->ee_block,
5063 /* Try to merge to the left. */
5065 EXT_FIRST_EXTENT(path[depth].p_hdr))
5067 ext4_ext_try_to_merge_right(inode,
5068 path, ex_start - 1))
5073 le32_add_cpu(&ex_last->ee_block, shift);
5074 ext4_ext_try_to_merge_right(inode, path,
5079 err = ext4_ext_dirty(handle, inode, path + depth);
5083 if (--depth < 0 || !update)
5087 /* Update index too */
5088 err = ext4_ext_get_access(handle, inode, path + depth);
5092 if (SHIFT == SHIFT_LEFT)
5093 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5095 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5096 err = ext4_ext_dirty(handle, inode, path + depth);
5100 /* we are done if current index is not a starting index */
5101 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5112 * ext4_ext_shift_extents:
5113 * All the extents which lies in the range from @start to the last allocated
5114 * block for the @inode are shifted either towards left or right (depending
5115 * upon @SHIFT) by @shift blocks.
5116 * On success, 0 is returned, error otherwise.
5119 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5120 ext4_lblk_t start, ext4_lblk_t shift,
5121 enum SHIFT_DIRECTION SHIFT)
5123 struct ext4_ext_path *path;
5125 struct ext4_extent *extent;
5126 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5127 ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5129 /* Let path point to the last extent */
5130 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5133 return PTR_ERR(path);
5135 depth = path->p_depth;
5136 extent = path[depth].p_ext;
5140 stop = le32_to_cpu(extent->ee_block);
5143 * For left shifts, make sure the hole on the left is big enough to
5144 * accommodate the shift. For right shifts, make sure the last extent
5145 * won't be shifted beyond EXT_MAX_BLOCKS.
5147 if (SHIFT == SHIFT_LEFT) {
5148 path = ext4_find_extent(inode, start - 1, &path,
5151 return PTR_ERR(path);
5152 depth = path->p_depth;
5153 extent = path[depth].p_ext;
5155 ex_start = le32_to_cpu(extent->ee_block);
5156 ex_end = le32_to_cpu(extent->ee_block) +
5157 ext4_ext_get_actual_len(extent);
5163 if ((start == ex_start && shift > ex_start) ||
5164 (shift > start - ex_end)) {
5169 if (shift > EXT_MAX_BLOCKS -
5170 (stop + ext4_ext_get_actual_len(extent))) {
5177 * In case of left shift, iterator points to start and it is increased
5178 * till we reach stop. In case of right shift, iterator points to stop
5179 * and it is decreased till we reach start.
5182 if (SHIFT == SHIFT_LEFT)
5187 if (tmp != EXT_MAX_BLOCKS)
5191 * Its safe to start updating extents. Start and stop are unsigned, so
5192 * in case of right shift if extent with 0 block is reached, iterator
5193 * becomes NULL to indicate the end of the loop.
5195 while (iterator && start <= stop) {
5196 path = ext4_find_extent(inode, *iterator, &path,
5199 return PTR_ERR(path);
5200 depth = path->p_depth;
5201 extent = path[depth].p_ext;
5203 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5204 (unsigned long) *iterator);
5205 return -EFSCORRUPTED;
5207 if (SHIFT == SHIFT_LEFT && *iterator >
5208 le32_to_cpu(extent->ee_block)) {
5209 /* Hole, move to the next extent */
5210 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5211 path[depth].p_ext++;
5213 *iterator = ext4_ext_next_allocated_block(path);
5219 if (SHIFT == SHIFT_LEFT) {
5220 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5221 *iterator = le32_to_cpu(extent->ee_block) +
5222 ext4_ext_get_actual_len(extent);
5224 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5225 if (le32_to_cpu(extent->ee_block) > 0)
5226 *iterator = le32_to_cpu(extent->ee_block) - 1;
5228 /* Beginning is reached, end of the loop */
5230 /* Update path extent in case we need to stop */
5231 while (le32_to_cpu(extent->ee_block) < start)
5233 path[depth].p_ext = extent;
5235 ret = ext4_ext_shift_path_extents(path, shift, inode,
5237 /* iterator can be NULL which means we should break */
5244 ext4_ext_drop_refs(path);
5250 * ext4_collapse_range:
5251 * This implements the fallocate's collapse range functionality for ext4
5252 * Returns: 0 and non-zero on error.
5254 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5256 struct inode *inode = file_inode(file);
5257 struct super_block *sb = inode->i_sb;
5258 struct address_space *mapping = inode->i_mapping;
5259 ext4_lblk_t punch_start, punch_stop;
5261 unsigned int credits;
5262 loff_t new_size, ioffset;
5266 * We need to test this early because xfstests assumes that a
5267 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5268 * system does not support collapse range.
5270 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5273 /* Collapse range works only on fs cluster size aligned regions. */
5274 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5277 trace_ext4_collapse_range(inode, offset, len);
5279 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5280 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5282 /* Call ext4_force_commit to flush all data in case of data=journal. */
5283 if (ext4_should_journal_data(inode)) {
5284 ret = ext4_force_commit(inode->i_sb);
5291 * There is no need to overlap collapse range with EOF, in which case
5292 * it is effectively a truncate operation
5294 if (offset + len >= inode->i_size) {
5299 /* Currently just for extent based files */
5300 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5305 /* Wait for existing dio to complete */
5306 inode_dio_wait(inode);
5308 ret = file_modified(file);
5313 * Prevent page faults from reinstantiating pages we have released from
5316 filemap_invalidate_lock(mapping);
5318 ret = ext4_break_layouts(inode);
5323 * Need to round down offset to be aligned with page size boundary
5324 * for page size > block size.
5326 ioffset = round_down(offset, PAGE_SIZE);
5328 * Write tail of the last page before removed range since it will get
5329 * removed from the page cache below.
5331 ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5335 * Write data that will be shifted to preserve them when discarding
5336 * page cache below. We are also protected from pages becoming dirty
5337 * by i_rwsem and invalidate_lock.
5339 ret = filemap_write_and_wait_range(mapping, offset + len,
5343 truncate_pagecache(inode, ioffset);
5345 credits = ext4_writepage_trans_blocks(inode);
5346 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5347 if (IS_ERR(handle)) {
5348 ret = PTR_ERR(handle);
5351 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5353 down_write(&EXT4_I(inode)->i_data_sem);
5354 ext4_discard_preallocations(inode, 0);
5356 ret = ext4_es_remove_extent(inode, punch_start,
5357 EXT_MAX_BLOCKS - punch_start);
5359 up_write(&EXT4_I(inode)->i_data_sem);
5363 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5365 up_write(&EXT4_I(inode)->i_data_sem);
5368 ext4_discard_preallocations(inode, 0);
5370 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5371 punch_stop - punch_start, SHIFT_LEFT);
5373 up_write(&EXT4_I(inode)->i_data_sem);
5377 new_size = inode->i_size - len;
5378 i_size_write(inode, new_size);
5379 EXT4_I(inode)->i_disksize = new_size;
5381 up_write(&EXT4_I(inode)->i_data_sem);
5383 ext4_handle_sync(handle);
5384 inode->i_mtime = inode->i_ctime = current_time(inode);
5385 ret = ext4_mark_inode_dirty(handle, inode);
5386 ext4_update_inode_fsync_trans(handle, inode, 1);
5389 ext4_journal_stop(handle);
5391 filemap_invalidate_unlock(mapping);
5393 inode_unlock(inode);
5398 * ext4_insert_range:
5399 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5400 * The data blocks starting from @offset to the EOF are shifted by @len
5401 * towards right to create a hole in the @inode. Inode size is increased
5403 * Returns 0 on success, error otherwise.
5405 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5407 struct inode *inode = file_inode(file);
5408 struct super_block *sb = inode->i_sb;
5409 struct address_space *mapping = inode->i_mapping;
5411 struct ext4_ext_path *path;
5412 struct ext4_extent *extent;
5413 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5414 unsigned int credits, ee_len;
5415 int ret = 0, depth, split_flag = 0;
5419 * We need to test this early because xfstests assumes that an
5420 * insert range of (0, 1) will return EOPNOTSUPP if the file
5421 * system does not support insert range.
5423 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5426 /* Insert range works only on fs cluster size aligned regions. */
5427 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5430 trace_ext4_insert_range(inode, offset, len);
5432 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5433 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5435 /* Call ext4_force_commit to flush all data in case of data=journal */
5436 if (ext4_should_journal_data(inode)) {
5437 ret = ext4_force_commit(inode->i_sb);
5443 /* Currently just for extent based files */
5444 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5449 /* Check whether the maximum file size would be exceeded */
5450 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5455 /* Offset must be less than i_size */
5456 if (offset >= inode->i_size) {
5461 /* Wait for existing dio to complete */
5462 inode_dio_wait(inode);
5464 ret = file_modified(file);
5469 * Prevent page faults from reinstantiating pages we have released from
5472 filemap_invalidate_lock(mapping);
5474 ret = ext4_break_layouts(inode);
5479 * Need to round down to align start offset to page size boundary
5480 * for page size > block size.
5482 ioffset = round_down(offset, PAGE_SIZE);
5483 /* Write out all dirty pages */
5484 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5488 truncate_pagecache(inode, ioffset);
5490 credits = ext4_writepage_trans_blocks(inode);
5491 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5492 if (IS_ERR(handle)) {
5493 ret = PTR_ERR(handle);
5496 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5498 /* Expand file to avoid data loss if there is error while shifting */
5499 inode->i_size += len;
5500 EXT4_I(inode)->i_disksize += len;
5501 inode->i_mtime = inode->i_ctime = current_time(inode);
5502 ret = ext4_mark_inode_dirty(handle, inode);
5506 down_write(&EXT4_I(inode)->i_data_sem);
5507 ext4_discard_preallocations(inode, 0);
5509 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5511 up_write(&EXT4_I(inode)->i_data_sem);
5515 depth = ext_depth(inode);
5516 extent = path[depth].p_ext;
5518 ee_start_lblk = le32_to_cpu(extent->ee_block);
5519 ee_len = ext4_ext_get_actual_len(extent);
5522 * If offset_lblk is not the starting block of extent, split
5523 * the extent @offset_lblk
5525 if ((offset_lblk > ee_start_lblk) &&
5526 (offset_lblk < (ee_start_lblk + ee_len))) {
5527 if (ext4_ext_is_unwritten(extent))
5528 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5529 EXT4_EXT_MARK_UNWRIT2;
5530 ret = ext4_split_extent_at(handle, inode, &path,
5531 offset_lblk, split_flag,
5533 EXT4_GET_BLOCKS_PRE_IO |
5534 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5537 ext4_ext_drop_refs(path);
5540 up_write(&EXT4_I(inode)->i_data_sem);
5544 ext4_ext_drop_refs(path);
5548 ret = ext4_es_remove_extent(inode, offset_lblk,
5549 EXT_MAX_BLOCKS - offset_lblk);
5551 up_write(&EXT4_I(inode)->i_data_sem);
5556 * if offset_lblk lies in a hole which is at start of file, use
5557 * ee_start_lblk to shift extents
5559 ret = ext4_ext_shift_extents(inode, handle,
5560 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5561 len_lblk, SHIFT_RIGHT);
5563 up_write(&EXT4_I(inode)->i_data_sem);
5565 ext4_handle_sync(handle);
5567 ext4_update_inode_fsync_trans(handle, inode, 1);
5570 ext4_journal_stop(handle);
5572 filemap_invalidate_unlock(mapping);
5574 inode_unlock(inode);
5579 * ext4_swap_extents() - Swap extents between two inodes
5580 * @handle: handle for this transaction
5581 * @inode1: First inode
5582 * @inode2: Second inode
5583 * @lblk1: Start block for first inode
5584 * @lblk2: Start block for second inode
5585 * @count: Number of blocks to swap
5586 * @unwritten: Mark second inode's extents as unwritten after swap
5587 * @erp: Pointer to save error value
5589 * This helper routine does exactly what is promise "swap extents". All other
5590 * stuff such as page-cache locking consistency, bh mapping consistency or
5591 * extent's data copying must be performed by caller.
5593 * i_rwsem is held for both inodes
5594 * i_data_sem is locked for write for both inodes
5596 * All pages from requested range are locked for both inodes
5599 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5600 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5601 ext4_lblk_t count, int unwritten, int *erp)
5603 struct ext4_ext_path *path1 = NULL;
5604 struct ext4_ext_path *path2 = NULL;
5605 int replaced_count = 0;
5607 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5608 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5609 BUG_ON(!inode_is_locked(inode1));
5610 BUG_ON(!inode_is_locked(inode2));
5612 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5615 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5620 struct ext4_extent *ex1, *ex2, tmp_ex;
5621 ext4_lblk_t e1_blk, e2_blk;
5622 int e1_len, e2_len, len;
5625 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5626 if (IS_ERR(path1)) {
5627 *erp = PTR_ERR(path1);
5633 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5634 if (IS_ERR(path2)) {
5635 *erp = PTR_ERR(path2);
5639 ex1 = path1[path1->p_depth].p_ext;
5640 ex2 = path2[path2->p_depth].p_ext;
5641 /* Do we have something to swap ? */
5642 if (unlikely(!ex2 || !ex1))
5645 e1_blk = le32_to_cpu(ex1->ee_block);
5646 e2_blk = le32_to_cpu(ex2->ee_block);
5647 e1_len = ext4_ext_get_actual_len(ex1);
5648 e2_len = ext4_ext_get_actual_len(ex2);
5651 if (!in_range(lblk1, e1_blk, e1_len) ||
5652 !in_range(lblk2, e2_blk, e2_len)) {
5653 ext4_lblk_t next1, next2;
5655 /* if hole after extent, then go to next extent */
5656 next1 = ext4_ext_next_allocated_block(path1);
5657 next2 = ext4_ext_next_allocated_block(path2);
5658 /* If hole before extent, then shift to that extent */
5663 /* Do we have something to swap */
5664 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5666 /* Move to the rightest boundary */
5667 len = next1 - lblk1;
5668 if (len < next2 - lblk2)
5669 len = next2 - lblk2;
5678 /* Prepare left boundary */
5679 if (e1_blk < lblk1) {
5681 *erp = ext4_force_split_extent_at(handle, inode1,
5686 if (e2_blk < lblk2) {
5688 *erp = ext4_force_split_extent_at(handle, inode2,
5693 /* ext4_split_extent_at() may result in leaf extent split,
5694 * path must to be revalidated. */
5698 /* Prepare right boundary */
5700 if (len > e1_blk + e1_len - lblk1)
5701 len = e1_blk + e1_len - lblk1;
5702 if (len > e2_blk + e2_len - lblk2)
5703 len = e2_blk + e2_len - lblk2;
5705 if (len != e1_len) {
5707 *erp = ext4_force_split_extent_at(handle, inode1,
5708 &path1, lblk1 + len, 0);
5712 if (len != e2_len) {
5714 *erp = ext4_force_split_extent_at(handle, inode2,
5715 &path2, lblk2 + len, 0);
5719 /* ext4_split_extent_at() may result in leaf extent split,
5720 * path must to be revalidated. */
5724 BUG_ON(e2_len != e1_len);
5725 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5728 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5732 /* Both extents are fully inside boundaries. Swap it now */
5734 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5735 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5736 ex1->ee_len = cpu_to_le16(e2_len);
5737 ex2->ee_len = cpu_to_le16(e1_len);
5739 ext4_ext_mark_unwritten(ex2);
5740 if (ext4_ext_is_unwritten(&tmp_ex))
5741 ext4_ext_mark_unwritten(ex1);
5743 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5744 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5745 *erp = ext4_ext_dirty(handle, inode2, path2 +
5749 *erp = ext4_ext_dirty(handle, inode1, path1 +
5752 * Looks scarry ah..? second inode already points to new blocks,
5753 * and it was successfully dirtied. But luckily error may happen
5754 * only due to journal error, so full transaction will be
5761 replaced_count += len;
5765 ext4_ext_drop_refs(path1);
5767 ext4_ext_drop_refs(path2);
5769 path1 = path2 = NULL;
5771 return replaced_count;
5775 * ext4_clu_mapped - determine whether any block in a logical cluster has
5776 * been mapped to a physical cluster
5778 * @inode - file containing the logical cluster
5779 * @lclu - logical cluster of interest
5781 * Returns 1 if any block in the logical cluster is mapped, signifying
5782 * that a physical cluster has been allocated for it. Otherwise,
5783 * returns 0. Can also return negative error codes. Derived from
5784 * ext4_ext_map_blocks().
5786 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5788 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5789 struct ext4_ext_path *path;
5790 int depth, mapped = 0, err = 0;
5791 struct ext4_extent *extent;
5792 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5794 /* search for the extent closest to the first block in the cluster */
5795 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5797 err = PTR_ERR(path);
5802 depth = ext_depth(inode);
5805 * A consistent leaf must not be empty. This situation is possible,
5806 * though, _during_ tree modification, and it's why an assert can't
5807 * be put in ext4_find_extent().
5809 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5810 EXT4_ERROR_INODE(inode,
5811 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5812 (unsigned long) EXT4_C2B(sbi, lclu),
5813 depth, path[depth].p_block);
5814 err = -EFSCORRUPTED;
5818 extent = path[depth].p_ext;
5820 /* can't be mapped if the extent tree is empty */
5824 first_lblk = le32_to_cpu(extent->ee_block);
5825 first_lclu = EXT4_B2C(sbi, first_lblk);
5828 * Three possible outcomes at this point - found extent spanning
5829 * the target cluster, to the left of the target cluster, or to the
5830 * right of the target cluster. The first two cases are handled here.
5831 * The last case indicates the target cluster is not mapped.
5833 if (lclu >= first_lclu) {
5834 last_lclu = EXT4_B2C(sbi, first_lblk +
5835 ext4_ext_get_actual_len(extent) - 1);
5836 if (lclu <= last_lclu) {
5839 first_lblk = ext4_ext_next_allocated_block(path);
5840 first_lclu = EXT4_B2C(sbi, first_lblk);
5841 if (lclu == first_lclu)
5847 ext4_ext_drop_refs(path);
5850 return err ? err : mapped;
5854 * Updates physical block address and unwritten status of extent
5855 * starting at lblk start and of len. If such an extent doesn't exist,
5856 * this function splits the extent tree appropriately to create an
5857 * extent like this. This function is called in the fast commit
5858 * replay path. Returns 0 on success and error on failure.
5860 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5861 int len, int unwritten, ext4_fsblk_t pblk)
5863 struct ext4_ext_path *path = NULL, *ppath;
5864 struct ext4_extent *ex;
5867 path = ext4_find_extent(inode, start, NULL, 0);
5869 return PTR_ERR(path);
5870 ex = path[path->p_depth].p_ext;
5872 ret = -EFSCORRUPTED;
5876 if (le32_to_cpu(ex->ee_block) != start ||
5877 ext4_ext_get_actual_len(ex) != len) {
5878 /* We need to split this extent to match our extent first */
5880 down_write(&EXT4_I(inode)->i_data_sem);
5881 ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5882 up_write(&EXT4_I(inode)->i_data_sem);
5886 path = ext4_find_extent(inode, start, NULL, 0);
5890 ex = path[path->p_depth].p_ext;
5891 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5892 if (ext4_ext_get_actual_len(ex) != len) {
5893 down_write(&EXT4_I(inode)->i_data_sem);
5894 ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5896 up_write(&EXT4_I(inode)->i_data_sem);
5900 path = ext4_find_extent(inode, start, NULL, 0);
5903 ex = path[path->p_depth].p_ext;
5907 ext4_ext_mark_unwritten(ex);
5909 ext4_ext_mark_initialized(ex);
5910 ext4_ext_store_pblock(ex, pblk);
5911 down_write(&EXT4_I(inode)->i_data_sem);
5912 ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5913 up_write(&EXT4_I(inode)->i_data_sem);
5915 ext4_ext_drop_refs(path);
5917 ext4_mark_inode_dirty(NULL, inode);
5921 /* Try to shrink the extent tree */
5922 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5924 struct ext4_ext_path *path = NULL;
5925 struct ext4_extent *ex;
5926 ext4_lblk_t old_cur, cur = 0;
5929 path = ext4_find_extent(inode, cur, NULL, 0);
5932 ex = path[path->p_depth].p_ext;
5934 ext4_ext_drop_refs(path);
5936 ext4_mark_inode_dirty(NULL, inode);
5940 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5943 ext4_ext_try_to_merge(NULL, inode, path, ex);
5944 down_write(&EXT4_I(inode)->i_data_sem);
5945 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5946 up_write(&EXT4_I(inode)->i_data_sem);
5947 ext4_mark_inode_dirty(NULL, inode);
5948 ext4_ext_drop_refs(path);
5953 /* Check if *cur is a hole and if it is, skip it */
5954 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5957 struct ext4_map_blocks map;
5960 map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5962 ret = ext4_map_blocks(NULL, inode, &map, 0);
5967 *cur = *cur + map.m_len;
5971 /* Count number of blocks used by this inode and update i_blocks */
5972 int ext4_ext_replay_set_iblocks(struct inode *inode)
5974 struct ext4_ext_path *path = NULL, *path2 = NULL;
5975 struct ext4_extent *ex;
5976 ext4_lblk_t cur = 0, end;
5977 int numblks = 0, i, ret = 0;
5978 ext4_fsblk_t cmp1, cmp2;
5979 struct ext4_map_blocks map;
5981 /* Determin the size of the file first */
5982 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5985 return PTR_ERR(path);
5986 ex = path[path->p_depth].p_ext;
5988 ext4_ext_drop_refs(path);
5992 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5993 ext4_ext_drop_refs(path);
5996 /* Count the number of data blocks */
6000 map.m_len = end - cur;
6001 ret = ext4_map_blocks(NULL, inode, &map, 0);
6006 cur = cur + map.m_len;
6010 * Count the number of extent tree blocks. We do it by looking up
6011 * two successive extents and determining the difference between
6012 * their paths. When path is different for 2 successive extents
6013 * we compare the blocks in the path at each level and increment
6014 * iblocks by total number of differences found.
6017 ret = skip_hole(inode, &cur);
6020 path = ext4_find_extent(inode, cur, NULL, 0);
6023 numblks += path->p_depth;
6024 ext4_ext_drop_refs(path);
6027 path = ext4_find_extent(inode, cur, NULL, 0);
6030 ex = path[path->p_depth].p_ext;
6032 ext4_ext_drop_refs(path);
6036 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6037 ext4_ext_get_actual_len(ex));
6038 ret = skip_hole(inode, &cur);
6040 ext4_ext_drop_refs(path);
6044 path2 = ext4_find_extent(inode, cur, NULL, 0);
6045 if (IS_ERR(path2)) {
6046 ext4_ext_drop_refs(path);
6050 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6052 if (i <= path->p_depth)
6053 cmp1 = path[i].p_bh ?
6054 path[i].p_bh->b_blocknr : 0;
6055 if (i <= path2->p_depth)
6056 cmp2 = path2[i].p_bh ?
6057 path2[i].p_bh->b_blocknr : 0;
6058 if (cmp1 != cmp2 && cmp2 != 0)
6061 ext4_ext_drop_refs(path);
6062 ext4_ext_drop_refs(path2);
6068 inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6069 ext4_mark_inode_dirty(NULL, inode);
6073 int ext4_ext_clear_bb(struct inode *inode)
6075 struct ext4_ext_path *path = NULL;
6076 struct ext4_extent *ex;
6077 ext4_lblk_t cur = 0, end;
6079 struct ext4_map_blocks map;
6081 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6084 /* Determin the size of the file first */
6085 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6088 return PTR_ERR(path);
6089 ex = path[path->p_depth].p_ext;
6091 ext4_ext_drop_refs(path);
6095 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6096 ext4_ext_drop_refs(path);
6102 map.m_len = end - cur;
6103 ret = ext4_map_blocks(NULL, inode, &map, 0);
6107 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6108 if (!IS_ERR_OR_NULL(path)) {
6109 for (j = 0; j < path->p_depth; j++) {
6111 ext4_mb_mark_bb(inode->i_sb,
6112 path[j].p_block, 1, 0);
6113 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6114 0, path[j].p_block, 1, 1);
6116 ext4_ext_drop_refs(path);
6119 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6120 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6121 map.m_lblk, map.m_pblk, map.m_len, 1);
6123 cur = cur + map.m_len;