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;
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);
400 *pblk = ext4_ext_pblock(ext);
403 cur = lblock + ext4_ext_get_actual_len(ext);
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);
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 (unlikely((eh->eh_entries == 0) && (depth > 0))) {
464 error_msg = "eh_entries is 0 but eh_depth is > 0";
467 if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
468 error_msg = "invalid extent entries";
471 if (unlikely(depth > 32)) {
472 error_msg = "too large eh_depth";
475 /* Verify checksum on non-root extent tree nodes */
476 if (ext_depth(inode) != depth &&
477 !ext4_extent_block_csum_verify(inode, eh)) {
478 error_msg = "extent tree corrupted";
485 ext4_error_inode_err(inode, function, line, 0, -err,
486 "pblk %llu bad header/extent: %s - magic %x, "
487 "entries %u, max %u(%u), depth %u(%u)",
488 (unsigned long long) pblk, error_msg,
489 le16_to_cpu(eh->eh_magic),
490 le16_to_cpu(eh->eh_entries),
491 le16_to_cpu(eh->eh_max),
492 max, le16_to_cpu(eh->eh_depth), depth);
496 #define ext4_ext_check(inode, eh, depth, pblk) \
497 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
499 int ext4_ext_check_inode(struct inode *inode)
501 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
504 static void ext4_cache_extents(struct inode *inode,
505 struct ext4_extent_header *eh)
507 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
508 ext4_lblk_t prev = 0;
511 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
512 unsigned int status = EXTENT_STATUS_WRITTEN;
513 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
514 int len = ext4_ext_get_actual_len(ex);
516 if (prev && (prev != lblk))
517 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
520 if (ext4_ext_is_unwritten(ex))
521 status = EXTENT_STATUS_UNWRITTEN;
522 ext4_es_cache_extent(inode, lblk, len,
523 ext4_ext_pblock(ex), status);
528 static struct buffer_head *
529 __read_extent_tree_block(const char *function, unsigned int line,
530 struct inode *inode, struct ext4_extent_idx *idx,
531 int depth, int flags)
533 struct buffer_head *bh;
535 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
538 if (flags & EXT4_EX_NOFAIL)
539 gfp_flags |= __GFP_NOFAIL;
541 pblk = ext4_idx_pblock(idx);
542 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
544 return ERR_PTR(-ENOMEM);
546 if (!bh_uptodate_or_lock(bh)) {
547 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
548 err = ext4_read_bh(bh, 0, NULL);
552 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
554 err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
555 depth, pblk, le32_to_cpu(idx->ei_block));
558 set_buffer_verified(bh);
560 * If this is a leaf block, cache all of its entries
562 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
563 struct ext4_extent_header *eh = ext_block_hdr(bh);
564 ext4_cache_extents(inode, eh);
573 #define read_extent_tree_block(inode, idx, depth, flags) \
574 __read_extent_tree_block(__func__, __LINE__, (inode), (idx), \
578 * This function is called to cache a file's extent information in the
581 int ext4_ext_precache(struct inode *inode)
583 struct ext4_inode_info *ei = EXT4_I(inode);
584 struct ext4_ext_path *path = NULL;
585 struct buffer_head *bh;
586 int i = 0, depth, ret = 0;
588 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
589 return 0; /* not an extent-mapped inode */
591 down_read(&ei->i_data_sem);
592 depth = ext_depth(inode);
594 /* Don't cache anything if there are no external extent blocks */
596 up_read(&ei->i_data_sem);
600 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
603 up_read(&ei->i_data_sem);
607 path[0].p_hdr = ext_inode_hdr(inode);
608 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
611 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
614 * If this is a leaf block or we've reached the end of
615 * the index block, go up
618 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
619 brelse(path[i].p_bh);
624 bh = read_extent_tree_block(inode, path[i].p_idx++,
626 EXT4_EX_FORCE_CACHE);
633 path[i].p_hdr = ext_block_hdr(bh);
634 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
636 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
638 up_read(&ei->i_data_sem);
639 ext4_ext_drop_refs(path);
645 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
647 int k, l = path->p_depth;
649 ext_debug(inode, "path:");
650 for (k = 0; k <= l; k++, path++) {
652 ext_debug(inode, " %d->%llu",
653 le32_to_cpu(path->p_idx->ei_block),
654 ext4_idx_pblock(path->p_idx));
655 } else if (path->p_ext) {
656 ext_debug(inode, " %d:[%d]%d:%llu ",
657 le32_to_cpu(path->p_ext->ee_block),
658 ext4_ext_is_unwritten(path->p_ext),
659 ext4_ext_get_actual_len(path->p_ext),
660 ext4_ext_pblock(path->p_ext));
662 ext_debug(inode, " []");
664 ext_debug(inode, "\n");
667 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
669 int depth = ext_depth(inode);
670 struct ext4_extent_header *eh;
671 struct ext4_extent *ex;
677 eh = path[depth].p_hdr;
678 ex = EXT_FIRST_EXTENT(eh);
680 ext_debug(inode, "Displaying leaf extents\n");
682 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
683 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
684 ext4_ext_is_unwritten(ex),
685 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
687 ext_debug(inode, "\n");
690 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
691 ext4_fsblk_t newblock, int level)
693 int depth = ext_depth(inode);
694 struct ext4_extent *ex;
696 if (depth != level) {
697 struct ext4_extent_idx *idx;
698 idx = path[level].p_idx;
699 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
700 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
701 level, le32_to_cpu(idx->ei_block),
702 ext4_idx_pblock(idx), newblock);
709 ex = path[depth].p_ext;
710 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
711 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
712 le32_to_cpu(ex->ee_block),
714 ext4_ext_is_unwritten(ex),
715 ext4_ext_get_actual_len(ex),
722 #define ext4_ext_show_path(inode, path)
723 #define ext4_ext_show_leaf(inode, path)
724 #define ext4_ext_show_move(inode, path, newblock, level)
727 void ext4_ext_drop_refs(struct ext4_ext_path *path)
733 depth = path->p_depth;
734 for (i = 0; i <= depth; i++, path++) {
741 * ext4_ext_binsearch_idx:
742 * binary search for the closest index of the given block
743 * the header must be checked before calling this
746 ext4_ext_binsearch_idx(struct inode *inode,
747 struct ext4_ext_path *path, ext4_lblk_t block)
749 struct ext4_extent_header *eh = path->p_hdr;
750 struct ext4_extent_idx *r, *l, *m;
753 ext_debug(inode, "binsearch for %u(idx): ", block);
755 l = EXT_FIRST_INDEX(eh) + 1;
756 r = EXT_LAST_INDEX(eh);
759 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
760 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
761 r, le32_to_cpu(r->ei_block));
763 if (block < le32_to_cpu(m->ei_block))
770 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
771 ext4_idx_pblock(path->p_idx));
773 #ifdef CHECK_BINSEARCH
775 struct ext4_extent_idx *chix, *ix;
778 chix = ix = EXT_FIRST_INDEX(eh);
779 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
780 if (k != 0 && le32_to_cpu(ix->ei_block) <=
781 le32_to_cpu(ix[-1].ei_block)) {
782 printk(KERN_DEBUG "k=%d, ix=0x%p, "
784 ix, EXT_FIRST_INDEX(eh));
785 printk(KERN_DEBUG "%u <= %u\n",
786 le32_to_cpu(ix->ei_block),
787 le32_to_cpu(ix[-1].ei_block));
789 BUG_ON(k && le32_to_cpu(ix->ei_block)
790 <= le32_to_cpu(ix[-1].ei_block));
791 if (block < le32_to_cpu(ix->ei_block))
795 BUG_ON(chix != path->p_idx);
802 * ext4_ext_binsearch:
803 * binary search for closest extent of the given block
804 * the header must be checked before calling this
807 ext4_ext_binsearch(struct inode *inode,
808 struct ext4_ext_path *path, ext4_lblk_t block)
810 struct ext4_extent_header *eh = path->p_hdr;
811 struct ext4_extent *r, *l, *m;
813 if (eh->eh_entries == 0) {
815 * this leaf is empty:
816 * we get such a leaf in split/add case
821 ext_debug(inode, "binsearch for %u: ", block);
823 l = EXT_FIRST_EXTENT(eh) + 1;
824 r = EXT_LAST_EXTENT(eh);
828 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
829 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
830 r, le32_to_cpu(r->ee_block));
832 if (block < le32_to_cpu(m->ee_block))
839 ext_debug(inode, " -> %d:%llu:[%d]%d ",
840 le32_to_cpu(path->p_ext->ee_block),
841 ext4_ext_pblock(path->p_ext),
842 ext4_ext_is_unwritten(path->p_ext),
843 ext4_ext_get_actual_len(path->p_ext));
845 #ifdef CHECK_BINSEARCH
847 struct ext4_extent *chex, *ex;
850 chex = ex = EXT_FIRST_EXTENT(eh);
851 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
852 BUG_ON(k && le32_to_cpu(ex->ee_block)
853 <= le32_to_cpu(ex[-1].ee_block));
854 if (block < le32_to_cpu(ex->ee_block))
858 BUG_ON(chex != path->p_ext);
864 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
866 struct ext4_extent_header *eh;
868 eh = ext_inode_hdr(inode);
871 eh->eh_magic = EXT4_EXT_MAGIC;
872 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
873 eh->eh_generation = 0;
874 ext4_mark_inode_dirty(handle, inode);
877 struct ext4_ext_path *
878 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
879 struct ext4_ext_path **orig_path, int flags)
881 struct ext4_extent_header *eh;
882 struct buffer_head *bh;
883 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
884 short int depth, i, ppos = 0;
886 gfp_t gfp_flags = GFP_NOFS;
888 if (flags & EXT4_EX_NOFAIL)
889 gfp_flags |= __GFP_NOFAIL;
891 eh = ext_inode_hdr(inode);
892 depth = ext_depth(inode);
893 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
894 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
901 ext4_ext_drop_refs(path);
902 if (depth > path[0].p_maxdepth) {
904 *orig_path = path = NULL;
908 /* account possible depth increase */
909 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
912 return ERR_PTR(-ENOMEM);
913 path[0].p_maxdepth = depth + 1;
919 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
920 ext4_cache_extents(inode, eh);
921 /* walk through the tree */
923 ext_debug(inode, "depth %d: num %d, max %d\n",
924 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
926 ext4_ext_binsearch_idx(inode, path + ppos, block);
927 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
928 path[ppos].p_depth = i;
929 path[ppos].p_ext = NULL;
931 bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
937 eh = ext_block_hdr(bh);
939 path[ppos].p_bh = bh;
940 path[ppos].p_hdr = eh;
943 path[ppos].p_depth = i;
944 path[ppos].p_ext = NULL;
945 path[ppos].p_idx = NULL;
948 ext4_ext_binsearch(inode, path + ppos, block);
949 /* if not an empty leaf */
950 if (path[ppos].p_ext)
951 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
953 ext4_ext_show_path(inode, path);
958 ext4_ext_drop_refs(path);
966 * ext4_ext_insert_index:
967 * insert new index [@logical;@ptr] into the block at @curp;
968 * check where to insert: before @curp or after @curp
970 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
971 struct ext4_ext_path *curp,
972 int logical, ext4_fsblk_t ptr)
974 struct ext4_extent_idx *ix;
977 err = ext4_ext_get_access(handle, inode, curp);
981 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
982 EXT4_ERROR_INODE(inode,
983 "logical %d == ei_block %d!",
984 logical, le32_to_cpu(curp->p_idx->ei_block));
985 return -EFSCORRUPTED;
988 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
989 >= le16_to_cpu(curp->p_hdr->eh_max))) {
990 EXT4_ERROR_INODE(inode,
991 "eh_entries %d >= eh_max %d!",
992 le16_to_cpu(curp->p_hdr->eh_entries),
993 le16_to_cpu(curp->p_hdr->eh_max));
994 return -EFSCORRUPTED;
997 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
999 ext_debug(inode, "insert new index %d after: %llu\n",
1001 ix = curp->p_idx + 1;
1004 ext_debug(inode, "insert new index %d before: %llu\n",
1009 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1012 ext_debug(inode, "insert new index %d: "
1013 "move %d indices from 0x%p to 0x%p\n",
1014 logical, len, ix, ix + 1);
1015 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1018 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1019 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1020 return -EFSCORRUPTED;
1023 ix->ei_block = cpu_to_le32(logical);
1024 ext4_idx_store_pblock(ix, ptr);
1025 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1027 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1028 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1029 return -EFSCORRUPTED;
1032 err = ext4_ext_dirty(handle, inode, curp);
1033 ext4_std_error(inode->i_sb, err);
1040 * inserts new subtree into the path, using free index entry
1042 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1043 * - makes decision where to split
1044 * - moves remaining extents and index entries (right to the split point)
1045 * into the newly allocated blocks
1046 * - initializes subtree
1048 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1050 struct ext4_ext_path *path,
1051 struct ext4_extent *newext, int at)
1053 struct buffer_head *bh = NULL;
1054 int depth = ext_depth(inode);
1055 struct ext4_extent_header *neh;
1056 struct ext4_extent_idx *fidx;
1057 int i = at, k, m, a;
1058 ext4_fsblk_t newblock, oldblock;
1060 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1061 gfp_t gfp_flags = GFP_NOFS;
1063 size_t ext_size = 0;
1065 if (flags & EXT4_EX_NOFAIL)
1066 gfp_flags |= __GFP_NOFAIL;
1068 /* make decision: where to split? */
1069 /* FIXME: now decision is simplest: at current extent */
1071 /* if current leaf will be split, then we should use
1072 * border from split point */
1073 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1074 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1075 return -EFSCORRUPTED;
1077 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1078 border = path[depth].p_ext[1].ee_block;
1079 ext_debug(inode, "leaf will be split."
1080 " next leaf starts at %d\n",
1081 le32_to_cpu(border));
1083 border = newext->ee_block;
1084 ext_debug(inode, "leaf will be added."
1085 " next leaf starts at %d\n",
1086 le32_to_cpu(border));
1090 * If error occurs, then we break processing
1091 * and mark filesystem read-only. index won't
1092 * be inserted and tree will be in consistent
1093 * state. Next mount will repair buffers too.
1097 * Get array to track all allocated blocks.
1098 * We need this to handle errors and free blocks
1101 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1105 /* allocate all needed blocks */
1106 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1107 for (a = 0; a < depth - at; a++) {
1108 newblock = ext4_ext_new_meta_block(handle, inode, path,
1109 newext, &err, flags);
1112 ablocks[a] = newblock;
1115 /* initialize new leaf */
1116 newblock = ablocks[--a];
1117 if (unlikely(newblock == 0)) {
1118 EXT4_ERROR_INODE(inode, "newblock == 0!");
1119 err = -EFSCORRUPTED;
1122 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1123 if (unlikely(!bh)) {
1129 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1134 neh = ext_block_hdr(bh);
1135 neh->eh_entries = 0;
1136 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1137 neh->eh_magic = EXT4_EXT_MAGIC;
1139 neh->eh_generation = 0;
1141 /* move remainder of path[depth] to the new leaf */
1142 if (unlikely(path[depth].p_hdr->eh_entries !=
1143 path[depth].p_hdr->eh_max)) {
1144 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1145 path[depth].p_hdr->eh_entries,
1146 path[depth].p_hdr->eh_max);
1147 err = -EFSCORRUPTED;
1150 /* start copy from next extent */
1151 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1152 ext4_ext_show_move(inode, path, newblock, depth);
1154 struct ext4_extent *ex;
1155 ex = EXT_FIRST_EXTENT(neh);
1156 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1157 le16_add_cpu(&neh->eh_entries, m);
1160 /* zero out unused area in the extent block */
1161 ext_size = sizeof(struct ext4_extent_header) +
1162 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1163 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1164 ext4_extent_block_csum_set(inode, neh);
1165 set_buffer_uptodate(bh);
1168 err = ext4_handle_dirty_metadata(handle, inode, bh);
1174 /* correct old leaf */
1176 err = ext4_ext_get_access(handle, inode, path + depth);
1179 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1180 err = ext4_ext_dirty(handle, inode, path + depth);
1186 /* create intermediate indexes */
1188 if (unlikely(k < 0)) {
1189 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1190 err = -EFSCORRUPTED;
1194 ext_debug(inode, "create %d intermediate indices\n", k);
1195 /* insert new index into current index block */
1196 /* current depth stored in i var */
1199 oldblock = newblock;
1200 newblock = ablocks[--a];
1201 bh = sb_getblk(inode->i_sb, newblock);
1202 if (unlikely(!bh)) {
1208 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1213 neh = ext_block_hdr(bh);
1214 neh->eh_entries = cpu_to_le16(1);
1215 neh->eh_magic = EXT4_EXT_MAGIC;
1216 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1217 neh->eh_depth = cpu_to_le16(depth - i);
1218 neh->eh_generation = 0;
1219 fidx = EXT_FIRST_INDEX(neh);
1220 fidx->ei_block = border;
1221 ext4_idx_store_pblock(fidx, oldblock);
1223 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1224 i, newblock, le32_to_cpu(border), oldblock);
1226 /* move remainder of path[i] to the new index block */
1227 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1228 EXT_LAST_INDEX(path[i].p_hdr))) {
1229 EXT4_ERROR_INODE(inode,
1230 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1231 le32_to_cpu(path[i].p_ext->ee_block));
1232 err = -EFSCORRUPTED;
1235 /* start copy indexes */
1236 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1237 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1238 EXT_MAX_INDEX(path[i].p_hdr));
1239 ext4_ext_show_move(inode, path, newblock, i);
1241 memmove(++fidx, path[i].p_idx,
1242 sizeof(struct ext4_extent_idx) * m);
1243 le16_add_cpu(&neh->eh_entries, m);
1245 /* zero out unused area in the extent block */
1246 ext_size = sizeof(struct ext4_extent_header) +
1247 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1248 memset(bh->b_data + ext_size, 0,
1249 inode->i_sb->s_blocksize - ext_size);
1250 ext4_extent_block_csum_set(inode, neh);
1251 set_buffer_uptodate(bh);
1254 err = ext4_handle_dirty_metadata(handle, inode, bh);
1260 /* correct old index */
1262 err = ext4_ext_get_access(handle, inode, path + i);
1265 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1266 err = ext4_ext_dirty(handle, inode, path + i);
1274 /* insert new index */
1275 err = ext4_ext_insert_index(handle, inode, path + at,
1276 le32_to_cpu(border), newblock);
1280 if (buffer_locked(bh))
1286 /* free all allocated blocks in error case */
1287 for (i = 0; i < depth; i++) {
1290 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1291 EXT4_FREE_BLOCKS_METADATA);
1300 * ext4_ext_grow_indepth:
1301 * implements tree growing procedure:
1302 * - allocates new block
1303 * - moves top-level data (index block or leaf) into the new block
1304 * - initializes new top-level, creating index that points to the
1305 * just created block
1307 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1310 struct ext4_extent_header *neh;
1311 struct buffer_head *bh;
1312 ext4_fsblk_t newblock, goal = 0;
1313 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1315 size_t ext_size = 0;
1317 /* Try to prepend new index to old one */
1318 if (ext_depth(inode))
1319 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1320 if (goal > le32_to_cpu(es->s_first_data_block)) {
1321 flags |= EXT4_MB_HINT_TRY_GOAL;
1324 goal = ext4_inode_to_goal_block(inode);
1325 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1330 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1335 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1342 ext_size = sizeof(EXT4_I(inode)->i_data);
1343 /* move top-level index/leaf into new block */
1344 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1345 /* zero out unused area in the extent block */
1346 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1348 /* set size of new block */
1349 neh = ext_block_hdr(bh);
1350 /* old root could have indexes or leaves
1351 * so calculate e_max right way */
1352 if (ext_depth(inode))
1353 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1355 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1356 neh->eh_magic = EXT4_EXT_MAGIC;
1357 ext4_extent_block_csum_set(inode, neh);
1358 set_buffer_uptodate(bh);
1359 set_buffer_verified(bh);
1362 err = ext4_handle_dirty_metadata(handle, inode, bh);
1366 /* Update top-level index: num,max,pointer */
1367 neh = ext_inode_hdr(inode);
1368 neh->eh_entries = cpu_to_le16(1);
1369 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1370 if (neh->eh_depth == 0) {
1371 /* Root extent block becomes index block */
1372 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1373 EXT_FIRST_INDEX(neh)->ei_block =
1374 EXT_FIRST_EXTENT(neh)->ee_block;
1376 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1377 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1378 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1379 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1381 le16_add_cpu(&neh->eh_depth, 1);
1382 err = ext4_mark_inode_dirty(handle, inode);
1390 * ext4_ext_create_new_leaf:
1391 * finds empty index and adds new leaf.
1392 * if no free index is found, then it requests in-depth growing.
1394 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1395 unsigned int mb_flags,
1396 unsigned int gb_flags,
1397 struct ext4_ext_path **ppath,
1398 struct ext4_extent *newext)
1400 struct ext4_ext_path *path = *ppath;
1401 struct ext4_ext_path *curp;
1402 int depth, i, err = 0;
1405 i = depth = ext_depth(inode);
1407 /* walk up to the tree and look for free index entry */
1408 curp = path + depth;
1409 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1414 /* we use already allocated block for index block,
1415 * so subsequent data blocks should be contiguous */
1416 if (EXT_HAS_FREE_INDEX(curp)) {
1417 /* if we found index with free entry, then use that
1418 * entry: create all needed subtree and add new leaf */
1419 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1424 path = ext4_find_extent(inode,
1425 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1428 err = PTR_ERR(path);
1430 /* tree is full, time to grow in depth */
1431 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1436 path = ext4_find_extent(inode,
1437 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1440 err = PTR_ERR(path);
1445 * only first (depth 0 -> 1) produces free space;
1446 * in all other cases we have to split the grown tree
1448 depth = ext_depth(inode);
1449 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1450 /* now we need to split */
1460 * search the closest allocated block to the left for *logical
1461 * and returns it at @logical + it's physical address at @phys
1462 * if *logical is the smallest allocated block, the function
1463 * returns 0 at @phys
1464 * return value contains 0 (success) or error code
1466 static int ext4_ext_search_left(struct inode *inode,
1467 struct ext4_ext_path *path,
1468 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1470 struct ext4_extent_idx *ix;
1471 struct ext4_extent *ex;
1474 if (unlikely(path == NULL)) {
1475 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1476 return -EFSCORRUPTED;
1478 depth = path->p_depth;
1481 if (depth == 0 && path->p_ext == NULL)
1484 /* usually extent in the path covers blocks smaller
1485 * then *logical, but it can be that extent is the
1486 * first one in the file */
1488 ex = path[depth].p_ext;
1489 ee_len = ext4_ext_get_actual_len(ex);
1490 if (*logical < le32_to_cpu(ex->ee_block)) {
1491 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1492 EXT4_ERROR_INODE(inode,
1493 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1494 *logical, le32_to_cpu(ex->ee_block));
1495 return -EFSCORRUPTED;
1497 while (--depth >= 0) {
1498 ix = path[depth].p_idx;
1499 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1500 EXT4_ERROR_INODE(inode,
1501 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1502 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1503 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block),
1505 return -EFSCORRUPTED;
1511 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1512 EXT4_ERROR_INODE(inode,
1513 "logical %d < ee_block %d + ee_len %d!",
1514 *logical, le32_to_cpu(ex->ee_block), ee_len);
1515 return -EFSCORRUPTED;
1518 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1519 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1524 * Search the closest allocated block to the right for *logical
1525 * and returns it at @logical + it's physical address at @phys.
1526 * If not exists, return 0 and @phys is set to 0. We will return
1527 * 1 which means we found an allocated block and ret_ex is valid.
1528 * Or return a (< 0) error code.
1530 static int ext4_ext_search_right(struct inode *inode,
1531 struct ext4_ext_path *path,
1532 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1533 struct ext4_extent *ret_ex)
1535 struct buffer_head *bh = NULL;
1536 struct ext4_extent_header *eh;
1537 struct ext4_extent_idx *ix;
1538 struct ext4_extent *ex;
1539 int depth; /* Note, NOT eh_depth; depth from top of tree */
1542 if (unlikely(path == NULL)) {
1543 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1544 return -EFSCORRUPTED;
1546 depth = path->p_depth;
1549 if (depth == 0 && path->p_ext == NULL)
1552 /* usually extent in the path covers blocks smaller
1553 * then *logical, but it can be that extent is the
1554 * first one in the file */
1556 ex = path[depth].p_ext;
1557 ee_len = ext4_ext_get_actual_len(ex);
1558 if (*logical < le32_to_cpu(ex->ee_block)) {
1559 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1560 EXT4_ERROR_INODE(inode,
1561 "first_extent(path[%d].p_hdr) != ex",
1563 return -EFSCORRUPTED;
1565 while (--depth >= 0) {
1566 ix = path[depth].p_idx;
1567 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1568 EXT4_ERROR_INODE(inode,
1569 "ix != EXT_FIRST_INDEX *logical %d!",
1571 return -EFSCORRUPTED;
1577 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1578 EXT4_ERROR_INODE(inode,
1579 "logical %d < ee_block %d + ee_len %d!",
1580 *logical, le32_to_cpu(ex->ee_block), ee_len);
1581 return -EFSCORRUPTED;
1584 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1585 /* next allocated block in this leaf */
1590 /* go up and search for index to the right */
1591 while (--depth >= 0) {
1592 ix = path[depth].p_idx;
1593 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1597 /* we've gone up to the root and found no index to the right */
1601 /* we've found index to the right, let's
1602 * follow it and find the closest allocated
1603 * block to the right */
1605 while (++depth < path->p_depth) {
1606 /* subtract from p_depth to get proper eh_depth */
1607 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1610 eh = ext_block_hdr(bh);
1611 ix = EXT_FIRST_INDEX(eh);
1615 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1618 eh = ext_block_hdr(bh);
1619 ex = EXT_FIRST_EXTENT(eh);
1621 *logical = le32_to_cpu(ex->ee_block);
1622 *phys = ext4_ext_pblock(ex);
1631 * ext4_ext_next_allocated_block:
1632 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1633 * NOTE: it considers block number from index entry as
1634 * allocated block. Thus, index entries have to be consistent
1638 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1642 BUG_ON(path == NULL);
1643 depth = path->p_depth;
1645 if (depth == 0 && path->p_ext == NULL)
1646 return EXT_MAX_BLOCKS;
1648 while (depth >= 0) {
1649 struct ext4_ext_path *p = &path[depth];
1651 if (depth == path->p_depth) {
1653 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1654 return le32_to_cpu(p->p_ext[1].ee_block);
1657 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1658 return le32_to_cpu(p->p_idx[1].ei_block);
1663 return EXT_MAX_BLOCKS;
1667 * ext4_ext_next_leaf_block:
1668 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1670 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1674 BUG_ON(path == NULL);
1675 depth = path->p_depth;
1677 /* zero-tree has no leaf blocks at all */
1679 return EXT_MAX_BLOCKS;
1681 /* go to index block */
1684 while (depth >= 0) {
1685 if (path[depth].p_idx !=
1686 EXT_LAST_INDEX(path[depth].p_hdr))
1687 return (ext4_lblk_t)
1688 le32_to_cpu(path[depth].p_idx[1].ei_block);
1692 return EXT_MAX_BLOCKS;
1696 * ext4_ext_correct_indexes:
1697 * if leaf gets modified and modified extent is first in the leaf,
1698 * then we have to correct all indexes above.
1699 * TODO: do we need to correct tree in all cases?
1701 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1702 struct ext4_ext_path *path)
1704 struct ext4_extent_header *eh;
1705 int depth = ext_depth(inode);
1706 struct ext4_extent *ex;
1710 eh = path[depth].p_hdr;
1711 ex = path[depth].p_ext;
1713 if (unlikely(ex == NULL || eh == NULL)) {
1714 EXT4_ERROR_INODE(inode,
1715 "ex %p == NULL or eh %p == NULL", ex, eh);
1716 return -EFSCORRUPTED;
1720 /* there is no tree at all */
1724 if (ex != EXT_FIRST_EXTENT(eh)) {
1725 /* we correct tree if first leaf got modified only */
1730 * TODO: we need correction if border is smaller than current one
1733 border = path[depth].p_ext->ee_block;
1734 err = ext4_ext_get_access(handle, inode, path + k);
1737 path[k].p_idx->ei_block = border;
1738 err = ext4_ext_dirty(handle, inode, path + k);
1743 /* change all left-side indexes */
1744 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1746 err = ext4_ext_get_access(handle, inode, path + k);
1749 path[k].p_idx->ei_block = border;
1750 err = ext4_ext_dirty(handle, inode, path + k);
1758 static int ext4_can_extents_be_merged(struct inode *inode,
1759 struct ext4_extent *ex1,
1760 struct ext4_extent *ex2)
1762 unsigned short ext1_ee_len, ext2_ee_len;
1764 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1767 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1768 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1770 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1771 le32_to_cpu(ex2->ee_block))
1774 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1777 if (ext4_ext_is_unwritten(ex1) &&
1778 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1780 #ifdef AGGRESSIVE_TEST
1781 if (ext1_ee_len >= 4)
1785 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1791 * This function tries to merge the "ex" extent to the next extent in the tree.
1792 * It always tries to merge towards right. If you want to merge towards
1793 * left, pass "ex - 1" as argument instead of "ex".
1794 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1795 * 1 if they got merged.
1797 static int ext4_ext_try_to_merge_right(struct inode *inode,
1798 struct ext4_ext_path *path,
1799 struct ext4_extent *ex)
1801 struct ext4_extent_header *eh;
1802 unsigned int depth, len;
1803 int merge_done = 0, unwritten;
1805 depth = ext_depth(inode);
1806 BUG_ON(path[depth].p_hdr == NULL);
1807 eh = path[depth].p_hdr;
1809 while (ex < EXT_LAST_EXTENT(eh)) {
1810 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1812 /* merge with next extent! */
1813 unwritten = ext4_ext_is_unwritten(ex);
1814 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1815 + ext4_ext_get_actual_len(ex + 1));
1817 ext4_ext_mark_unwritten(ex);
1819 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1820 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1821 * sizeof(struct ext4_extent);
1822 memmove(ex + 1, ex + 2, len);
1824 le16_add_cpu(&eh->eh_entries, -1);
1826 WARN_ON(eh->eh_entries == 0);
1827 if (!eh->eh_entries)
1828 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1835 * This function does a very simple check to see if we can collapse
1836 * an extent tree with a single extent tree leaf block into the inode.
1838 static void ext4_ext_try_to_merge_up(handle_t *handle,
1839 struct inode *inode,
1840 struct ext4_ext_path *path)
1843 unsigned max_root = ext4_ext_space_root(inode, 0);
1846 if ((path[0].p_depth != 1) ||
1847 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1848 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1852 * We need to modify the block allocation bitmap and the block
1853 * group descriptor to release the extent tree block. If we
1854 * can't get the journal credits, give up.
1856 if (ext4_journal_extend(handle, 2,
1857 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1861 * Copy the extent data up to the inode
1863 blk = ext4_idx_pblock(path[0].p_idx);
1864 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1865 sizeof(struct ext4_extent_idx);
1866 s += sizeof(struct ext4_extent_header);
1868 path[1].p_maxdepth = path[0].p_maxdepth;
1869 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1870 path[0].p_depth = 0;
1871 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1872 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1873 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1875 brelse(path[1].p_bh);
1876 ext4_free_blocks(handle, inode, NULL, blk, 1,
1877 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1881 * This function tries to merge the @ex extent to neighbours in the tree, then
1882 * tries to collapse the extent tree into the inode.
1884 static void ext4_ext_try_to_merge(handle_t *handle,
1885 struct inode *inode,
1886 struct ext4_ext_path *path,
1887 struct ext4_extent *ex)
1889 struct ext4_extent_header *eh;
1893 depth = ext_depth(inode);
1894 BUG_ON(path[depth].p_hdr == NULL);
1895 eh = path[depth].p_hdr;
1897 if (ex > EXT_FIRST_EXTENT(eh))
1898 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1901 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1903 ext4_ext_try_to_merge_up(handle, inode, path);
1907 * check if a portion of the "newext" extent overlaps with an
1910 * If there is an overlap discovered, it updates the length of the newext
1911 * such that there will be no overlap, and then returns 1.
1912 * If there is no overlap found, it returns 0.
1914 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1915 struct inode *inode,
1916 struct ext4_extent *newext,
1917 struct ext4_ext_path *path)
1920 unsigned int depth, len1;
1921 unsigned int ret = 0;
1923 b1 = le32_to_cpu(newext->ee_block);
1924 len1 = ext4_ext_get_actual_len(newext);
1925 depth = ext_depth(inode);
1926 if (!path[depth].p_ext)
1928 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1931 * get the next allocated block if the extent in the path
1932 * is before the requested block(s)
1935 b2 = ext4_ext_next_allocated_block(path);
1936 if (b2 == EXT_MAX_BLOCKS)
1938 b2 = EXT4_LBLK_CMASK(sbi, b2);
1941 /* check for wrap through zero on extent logical start block*/
1942 if (b1 + len1 < b1) {
1943 len1 = EXT_MAX_BLOCKS - b1;
1944 newext->ee_len = cpu_to_le16(len1);
1948 /* check for overlap */
1949 if (b1 + len1 > b2) {
1950 newext->ee_len = cpu_to_le16(b2 - b1);
1958 * ext4_ext_insert_extent:
1959 * tries to merge requested extent into the existing extent or
1960 * inserts requested extent as new one into the tree,
1961 * creating new leaf in the no-space case.
1963 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1964 struct ext4_ext_path **ppath,
1965 struct ext4_extent *newext, int gb_flags)
1967 struct ext4_ext_path *path = *ppath;
1968 struct ext4_extent_header *eh;
1969 struct ext4_extent *ex, *fex;
1970 struct ext4_extent *nearex; /* nearest extent */
1971 struct ext4_ext_path *npath = NULL;
1972 int depth, len, err;
1974 int mb_flags = 0, unwritten;
1976 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1977 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1978 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1979 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1980 return -EFSCORRUPTED;
1982 depth = ext_depth(inode);
1983 ex = path[depth].p_ext;
1984 eh = path[depth].p_hdr;
1985 if (unlikely(path[depth].p_hdr == NULL)) {
1986 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1987 return -EFSCORRUPTED;
1990 /* try to insert block into found extent and return */
1991 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1994 * Try to see whether we should rather test the extent on
1995 * right from ex, or from the left of ex. This is because
1996 * ext4_find_extent() can return either extent on the
1997 * left, or on the right from the searched position. This
1998 * will make merging more effective.
2000 if (ex < EXT_LAST_EXTENT(eh) &&
2001 (le32_to_cpu(ex->ee_block) +
2002 ext4_ext_get_actual_len(ex) <
2003 le32_to_cpu(newext->ee_block))) {
2006 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2007 (le32_to_cpu(newext->ee_block) +
2008 ext4_ext_get_actual_len(newext) <
2009 le32_to_cpu(ex->ee_block)))
2012 /* Try to append newex to the ex */
2013 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2014 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2016 ext4_ext_is_unwritten(newext),
2017 ext4_ext_get_actual_len(newext),
2018 le32_to_cpu(ex->ee_block),
2019 ext4_ext_is_unwritten(ex),
2020 ext4_ext_get_actual_len(ex),
2021 ext4_ext_pblock(ex));
2022 err = ext4_ext_get_access(handle, inode,
2026 unwritten = ext4_ext_is_unwritten(ex);
2027 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2028 + ext4_ext_get_actual_len(newext));
2030 ext4_ext_mark_unwritten(ex);
2036 /* Try to prepend newex to the ex */
2037 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2038 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2040 le32_to_cpu(newext->ee_block),
2041 ext4_ext_is_unwritten(newext),
2042 ext4_ext_get_actual_len(newext),
2043 le32_to_cpu(ex->ee_block),
2044 ext4_ext_is_unwritten(ex),
2045 ext4_ext_get_actual_len(ex),
2046 ext4_ext_pblock(ex));
2047 err = ext4_ext_get_access(handle, inode,
2052 unwritten = ext4_ext_is_unwritten(ex);
2053 ex->ee_block = newext->ee_block;
2054 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2055 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2056 + ext4_ext_get_actual_len(newext));
2058 ext4_ext_mark_unwritten(ex);
2064 depth = ext_depth(inode);
2065 eh = path[depth].p_hdr;
2066 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2069 /* probably next leaf has space for us? */
2070 fex = EXT_LAST_EXTENT(eh);
2071 next = EXT_MAX_BLOCKS;
2072 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2073 next = ext4_ext_next_leaf_block(path);
2074 if (next != EXT_MAX_BLOCKS) {
2075 ext_debug(inode, "next leaf block - %u\n", next);
2076 BUG_ON(npath != NULL);
2077 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2079 return PTR_ERR(npath);
2080 BUG_ON(npath->p_depth != path->p_depth);
2081 eh = npath[depth].p_hdr;
2082 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2083 ext_debug(inode, "next leaf isn't full(%d)\n",
2084 le16_to_cpu(eh->eh_entries));
2088 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2089 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2093 * There is no free space in the found leaf.
2094 * We're gonna add a new leaf in the tree.
2096 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2097 mb_flags |= EXT4_MB_USE_RESERVED;
2098 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2102 depth = ext_depth(inode);
2103 eh = path[depth].p_hdr;
2106 nearex = path[depth].p_ext;
2108 err = ext4_ext_get_access(handle, inode, path + depth);
2113 /* there is no extent in this leaf, create first one */
2114 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2115 le32_to_cpu(newext->ee_block),
2116 ext4_ext_pblock(newext),
2117 ext4_ext_is_unwritten(newext),
2118 ext4_ext_get_actual_len(newext));
2119 nearex = EXT_FIRST_EXTENT(eh);
2121 if (le32_to_cpu(newext->ee_block)
2122 > le32_to_cpu(nearex->ee_block)) {
2124 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2126 le32_to_cpu(newext->ee_block),
2127 ext4_ext_pblock(newext),
2128 ext4_ext_is_unwritten(newext),
2129 ext4_ext_get_actual_len(newext),
2134 BUG_ON(newext->ee_block == nearex->ee_block);
2135 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2137 le32_to_cpu(newext->ee_block),
2138 ext4_ext_pblock(newext),
2139 ext4_ext_is_unwritten(newext),
2140 ext4_ext_get_actual_len(newext),
2143 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2145 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2146 "move %d extents from 0x%p to 0x%p\n",
2147 le32_to_cpu(newext->ee_block),
2148 ext4_ext_pblock(newext),
2149 ext4_ext_is_unwritten(newext),
2150 ext4_ext_get_actual_len(newext),
2151 len, nearex, nearex + 1);
2152 memmove(nearex + 1, nearex,
2153 len * sizeof(struct ext4_extent));
2157 le16_add_cpu(&eh->eh_entries, 1);
2158 path[depth].p_ext = nearex;
2159 nearex->ee_block = newext->ee_block;
2160 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2161 nearex->ee_len = newext->ee_len;
2164 /* try to merge extents */
2165 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2166 ext4_ext_try_to_merge(handle, inode, path, nearex);
2169 /* time to correct all indexes above */
2170 err = ext4_ext_correct_indexes(handle, inode, path);
2174 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2177 ext4_ext_drop_refs(npath);
2182 static int ext4_fill_es_cache_info(struct inode *inode,
2183 ext4_lblk_t block, ext4_lblk_t num,
2184 struct fiemap_extent_info *fieinfo)
2186 ext4_lblk_t next, end = block + num - 1;
2187 struct extent_status es;
2188 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2192 while (block <= end) {
2195 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2197 if (ext4_es_is_unwritten(&es))
2198 flags |= FIEMAP_EXTENT_UNWRITTEN;
2199 if (ext4_es_is_delayed(&es))
2200 flags |= (FIEMAP_EXTENT_DELALLOC |
2201 FIEMAP_EXTENT_UNKNOWN);
2202 if (ext4_es_is_hole(&es))
2203 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2205 flags |= FIEMAP_EXTENT_LAST;
2206 if (flags & (FIEMAP_EXTENT_DELALLOC|
2207 EXT4_FIEMAP_EXTENT_HOLE))
2210 es.es_pblk = ext4_es_pblock(&es);
2211 err = fiemap_fill_next_extent(fieinfo,
2212 (__u64)es.es_lblk << blksize_bits,
2213 (__u64)es.es_pblk << blksize_bits,
2214 (__u64)es.es_len << blksize_bits,
2229 * ext4_ext_determine_hole - determine hole around given block
2230 * @inode: inode we lookup in
2231 * @path: path in extent tree to @lblk
2232 * @lblk: pointer to logical block around which we want to determine hole
2234 * Determine hole length (and start if easily possible) around given logical
2235 * block. We don't try too hard to find the beginning of the hole but @path
2236 * actually points to extent before @lblk, we provide it.
2238 * The function returns the length of a hole starting at @lblk. We update @lblk
2239 * to the beginning of the hole if we managed to find it.
2241 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2242 struct ext4_ext_path *path,
2245 int depth = ext_depth(inode);
2246 struct ext4_extent *ex;
2249 ex = path[depth].p_ext;
2251 /* there is no extent yet, so gap is [0;-] */
2253 len = EXT_MAX_BLOCKS;
2254 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2255 len = le32_to_cpu(ex->ee_block) - *lblk;
2256 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2257 + ext4_ext_get_actual_len(ex)) {
2260 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2261 next = ext4_ext_next_allocated_block(path);
2262 BUG_ON(next == *lblk);
2271 * ext4_ext_put_gap_in_cache:
2272 * calculate boundaries of the gap that the requested block fits into
2273 * and cache this gap
2276 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2277 ext4_lblk_t hole_len)
2279 struct extent_status es;
2281 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2282 hole_start + hole_len - 1, &es);
2284 /* There's delayed extent containing lblock? */
2285 if (es.es_lblk <= hole_start)
2287 hole_len = min(es.es_lblk - hole_start, hole_len);
2289 ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2290 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2291 EXTENT_STATUS_HOLE);
2296 * removes index from the index block.
2298 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2299 struct ext4_ext_path *path, int depth)
2304 /* free index block */
2306 path = path + depth;
2307 leaf = ext4_idx_pblock(path->p_idx);
2308 if (unlikely(path->p_hdr->eh_entries == 0)) {
2309 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2310 return -EFSCORRUPTED;
2312 err = ext4_ext_get_access(handle, inode, path);
2316 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2317 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2318 len *= sizeof(struct ext4_extent_idx);
2319 memmove(path->p_idx, path->p_idx + 1, len);
2322 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2323 err = ext4_ext_dirty(handle, inode, path);
2326 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2327 trace_ext4_ext_rm_idx(inode, leaf);
2329 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2330 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2332 while (--depth >= 0) {
2333 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2336 err = ext4_ext_get_access(handle, inode, path);
2339 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2340 err = ext4_ext_dirty(handle, inode, path);
2348 * ext4_ext_calc_credits_for_single_extent:
2349 * This routine returns max. credits that needed to insert an extent
2350 * to the extent tree.
2351 * When pass the actual path, the caller should calculate credits
2354 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2355 struct ext4_ext_path *path)
2358 int depth = ext_depth(inode);
2361 /* probably there is space in leaf? */
2362 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2363 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2366 * There are some space in the leaf tree, no
2367 * need to account for leaf block credit
2369 * bitmaps and block group descriptor blocks
2370 * and other metadata blocks still need to be
2373 /* 1 bitmap, 1 block group descriptor */
2374 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2379 return ext4_chunk_trans_blocks(inode, nrblocks);
2383 * How many index/leaf blocks need to change/allocate to add @extents extents?
2385 * If we add a single extent, then in the worse case, each tree level
2386 * index/leaf need to be changed in case of the tree split.
2388 * If more extents are inserted, they could cause the whole tree split more
2389 * than once, but this is really rare.
2391 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2396 /* If we are converting the inline data, only one is needed here. */
2397 if (ext4_has_inline_data(inode))
2400 depth = ext_depth(inode);
2410 static inline int get_default_free_blocks_flags(struct inode *inode)
2412 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2413 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2414 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2415 else if (ext4_should_journal_data(inode))
2416 return EXT4_FREE_BLOCKS_FORGET;
2421 * ext4_rereserve_cluster - increment the reserved cluster count when
2422 * freeing a cluster with a pending reservation
2424 * @inode - file containing the cluster
2425 * @lblk - logical block in cluster to be reserved
2427 * Increments the reserved cluster count and adjusts quota in a bigalloc
2428 * file system when freeing a partial cluster containing at least one
2429 * delayed and unwritten block. A partial cluster meeting that
2430 * requirement will have a pending reservation. If so, the
2431 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2432 * defer reserved and allocated space accounting to a subsequent call
2435 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2437 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2438 struct ext4_inode_info *ei = EXT4_I(inode);
2440 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2442 spin_lock(&ei->i_block_reservation_lock);
2443 ei->i_reserved_data_blocks++;
2444 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2445 spin_unlock(&ei->i_block_reservation_lock);
2447 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2448 ext4_remove_pending(inode, lblk);
2451 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2452 struct ext4_extent *ex,
2453 struct partial_cluster *partial,
2454 ext4_lblk_t from, ext4_lblk_t to)
2456 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2457 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2458 ext4_fsblk_t last_pblk, pblk;
2462 /* only extent tail removal is allowed */
2463 if (from < le32_to_cpu(ex->ee_block) ||
2464 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2465 ext4_error(sbi->s_sb,
2466 "strange request: removal(2) %u-%u from %u:%u",
2467 from, to, le32_to_cpu(ex->ee_block), ee_len);
2471 #ifdef EXTENTS_STATS
2472 spin_lock(&sbi->s_ext_stats_lock);
2473 sbi->s_ext_blocks += ee_len;
2474 sbi->s_ext_extents++;
2475 if (ee_len < sbi->s_ext_min)
2476 sbi->s_ext_min = ee_len;
2477 if (ee_len > sbi->s_ext_max)
2478 sbi->s_ext_max = ee_len;
2479 if (ext_depth(inode) > sbi->s_depth_max)
2480 sbi->s_depth_max = ext_depth(inode);
2481 spin_unlock(&sbi->s_ext_stats_lock);
2484 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2487 * if we have a partial cluster, and it's different from the
2488 * cluster of the last block in the extent, we free it
2490 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2492 if (partial->state != initial &&
2493 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2494 if (partial->state == tofree) {
2495 flags = get_default_free_blocks_flags(inode);
2496 if (ext4_is_pending(inode, partial->lblk))
2497 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2498 ext4_free_blocks(handle, inode, NULL,
2499 EXT4_C2B(sbi, partial->pclu),
2500 sbi->s_cluster_ratio, flags);
2501 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2502 ext4_rereserve_cluster(inode, partial->lblk);
2504 partial->state = initial;
2507 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2508 pblk = ext4_ext_pblock(ex) + ee_len - num;
2511 * We free the partial cluster at the end of the extent (if any),
2512 * unless the cluster is used by another extent (partial_cluster
2513 * state is nofree). If a partial cluster exists here, it must be
2514 * shared with the last block in the extent.
2516 flags = get_default_free_blocks_flags(inode);
2518 /* partial, left end cluster aligned, right end unaligned */
2519 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2520 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2521 (partial->state != nofree)) {
2522 if (ext4_is_pending(inode, to))
2523 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2524 ext4_free_blocks(handle, inode, NULL,
2525 EXT4_PBLK_CMASK(sbi, last_pblk),
2526 sbi->s_cluster_ratio, flags);
2527 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2528 ext4_rereserve_cluster(inode, to);
2529 partial->state = initial;
2530 flags = get_default_free_blocks_flags(inode);
2533 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2536 * For bigalloc file systems, we never free a partial cluster
2537 * at the beginning of the extent. Instead, we check to see if we
2538 * need to free it on a subsequent call to ext4_remove_blocks,
2539 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2541 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2542 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2544 /* reset the partial cluster if we've freed past it */
2545 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2546 partial->state = initial;
2549 * If we've freed the entire extent but the beginning is not left
2550 * cluster aligned and is not marked as ineligible for freeing we
2551 * record the partial cluster at the beginning of the extent. It
2552 * wasn't freed by the preceding ext4_free_blocks() call, and we
2553 * need to look farther to the left to determine if it's to be freed
2554 * (not shared with another extent). Else, reset the partial
2555 * cluster - we're either done freeing or the beginning of the
2556 * extent is left cluster aligned.
2558 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2559 if (partial->state == initial) {
2560 partial->pclu = EXT4_B2C(sbi, pblk);
2561 partial->lblk = from;
2562 partial->state = tofree;
2565 partial->state = initial;
2572 * ext4_ext_rm_leaf() Removes the extents associated with the
2573 * blocks appearing between "start" and "end". Both "start"
2574 * and "end" must appear in the same extent or EIO is returned.
2576 * @handle: The journal handle
2577 * @inode: The files inode
2578 * @path: The path to the leaf
2579 * @partial_cluster: The cluster which we'll have to free if all extents
2580 * has been released from it. However, if this value is
2581 * negative, it's a cluster just to the right of the
2582 * punched region and it must not be freed.
2583 * @start: The first block to remove
2584 * @end: The last block to remove
2587 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2588 struct ext4_ext_path *path,
2589 struct partial_cluster *partial,
2590 ext4_lblk_t start, ext4_lblk_t end)
2592 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2593 int err = 0, correct_index = 0;
2594 int depth = ext_depth(inode), credits, revoke_credits;
2595 struct ext4_extent_header *eh;
2598 ext4_lblk_t ex_ee_block;
2599 unsigned short ex_ee_len;
2600 unsigned unwritten = 0;
2601 struct ext4_extent *ex;
2604 /* the header must be checked already in ext4_ext_remove_space() */
2605 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2606 if (!path[depth].p_hdr)
2607 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2608 eh = path[depth].p_hdr;
2609 if (unlikely(path[depth].p_hdr == NULL)) {
2610 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2611 return -EFSCORRUPTED;
2613 /* find where to start removing */
2614 ex = path[depth].p_ext;
2616 ex = EXT_LAST_EXTENT(eh);
2618 ex_ee_block = le32_to_cpu(ex->ee_block);
2619 ex_ee_len = ext4_ext_get_actual_len(ex);
2621 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2623 while (ex >= EXT_FIRST_EXTENT(eh) &&
2624 ex_ee_block + ex_ee_len > start) {
2626 if (ext4_ext_is_unwritten(ex))
2631 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2632 unwritten, ex_ee_len);
2633 path[depth].p_ext = ex;
2635 a = ex_ee_block > start ? ex_ee_block : start;
2636 b = ex_ee_block+ex_ee_len - 1 < end ?
2637 ex_ee_block+ex_ee_len - 1 : end;
2639 ext_debug(inode, " border %u:%u\n", a, b);
2641 /* If this extent is beyond the end of the hole, skip it */
2642 if (end < ex_ee_block) {
2644 * We're going to skip this extent and move to another,
2645 * so note that its first cluster is in use to avoid
2646 * freeing it when removing blocks. Eventually, the
2647 * right edge of the truncated/punched region will
2648 * be just to the left.
2650 if (sbi->s_cluster_ratio > 1) {
2651 pblk = ext4_ext_pblock(ex);
2652 partial->pclu = EXT4_B2C(sbi, pblk);
2653 partial->state = nofree;
2656 ex_ee_block = le32_to_cpu(ex->ee_block);
2657 ex_ee_len = ext4_ext_get_actual_len(ex);
2659 } else if (b != ex_ee_block + ex_ee_len - 1) {
2660 EXT4_ERROR_INODE(inode,
2661 "can not handle truncate %u:%u "
2663 start, end, ex_ee_block,
2664 ex_ee_block + ex_ee_len - 1);
2665 err = -EFSCORRUPTED;
2667 } else if (a != ex_ee_block) {
2668 /* remove tail of the extent */
2669 num = a - ex_ee_block;
2671 /* remove whole extent: excellent! */
2675 * 3 for leaf, sb, and inode plus 2 (bmap and group
2676 * descriptor) for each block group; assume two block
2677 * groups plus ex_ee_len/blocks_per_block_group for
2680 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2681 if (ex == EXT_FIRST_EXTENT(eh)) {
2683 credits += (ext_depth(inode)) + 1;
2685 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2687 * We may end up freeing some index blocks and data from the
2688 * punched range. Note that partial clusters are accounted for
2689 * by ext4_free_data_revoke_credits().
2692 ext4_free_metadata_revoke_credits(inode->i_sb,
2694 ext4_free_data_revoke_credits(inode, b - a + 1);
2696 err = ext4_datasem_ensure_credits(handle, inode, credits,
2697 credits, revoke_credits);
2704 err = ext4_ext_get_access(handle, inode, path + depth);
2708 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2713 /* this extent is removed; mark slot entirely unused */
2714 ext4_ext_store_pblock(ex, 0);
2716 ex->ee_len = cpu_to_le16(num);
2718 * Do not mark unwritten if all the blocks in the
2719 * extent have been removed.
2721 if (unwritten && num)
2722 ext4_ext_mark_unwritten(ex);
2724 * If the extent was completely released,
2725 * we need to remove it from the leaf
2728 if (end != EXT_MAX_BLOCKS - 1) {
2730 * For hole punching, we need to scoot all the
2731 * extents up when an extent is removed so that
2732 * we dont have blank extents in the middle
2734 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2735 sizeof(struct ext4_extent));
2737 /* Now get rid of the one at the end */
2738 memset(EXT_LAST_EXTENT(eh), 0,
2739 sizeof(struct ext4_extent));
2741 le16_add_cpu(&eh->eh_entries, -1);
2744 err = ext4_ext_dirty(handle, inode, path + depth);
2748 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2749 ext4_ext_pblock(ex));
2751 ex_ee_block = le32_to_cpu(ex->ee_block);
2752 ex_ee_len = ext4_ext_get_actual_len(ex);
2755 if (correct_index && eh->eh_entries)
2756 err = ext4_ext_correct_indexes(handle, inode, path);
2759 * If there's a partial cluster and at least one extent remains in
2760 * the leaf, free the partial cluster if it isn't shared with the
2761 * current extent. If it is shared with the current extent
2762 * we reset the partial cluster because we've reached the start of the
2763 * truncated/punched region and we're done removing blocks.
2765 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2766 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2767 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2768 int flags = get_default_free_blocks_flags(inode);
2770 if (ext4_is_pending(inode, partial->lblk))
2771 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2772 ext4_free_blocks(handle, inode, NULL,
2773 EXT4_C2B(sbi, partial->pclu),
2774 sbi->s_cluster_ratio, flags);
2775 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2776 ext4_rereserve_cluster(inode, partial->lblk);
2778 partial->state = initial;
2781 /* if this leaf is free, then we should
2782 * remove it from index block above */
2783 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2784 err = ext4_ext_rm_idx(handle, inode, path, depth);
2791 * ext4_ext_more_to_rm:
2792 * returns 1 if current index has to be freed (even partial)
2795 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2797 BUG_ON(path->p_idx == NULL);
2799 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2803 * if truncate on deeper level happened, it wasn't partial,
2804 * so we have to consider current index for truncation
2806 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2811 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2814 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2815 int depth = ext_depth(inode);
2816 struct ext4_ext_path *path = NULL;
2817 struct partial_cluster partial;
2823 partial.state = initial;
2825 ext_debug(inode, "truncate since %u to %u\n", start, end);
2827 /* probably first extent we're gonna free will be last in block */
2828 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2830 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2832 return PTR_ERR(handle);
2835 trace_ext4_ext_remove_space(inode, start, end, depth);
2838 * Check if we are removing extents inside the extent tree. If that
2839 * is the case, we are going to punch a hole inside the extent tree
2840 * so we have to check whether we need to split the extent covering
2841 * the last block to remove so we can easily remove the part of it
2842 * in ext4_ext_rm_leaf().
2844 if (end < EXT_MAX_BLOCKS - 1) {
2845 struct ext4_extent *ex;
2846 ext4_lblk_t ee_block, ex_end, lblk;
2849 /* find extent for or closest extent to this block */
2850 path = ext4_find_extent(inode, end, NULL,
2851 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2853 ext4_journal_stop(handle);
2854 return PTR_ERR(path);
2856 depth = ext_depth(inode);
2857 /* Leaf not may not exist only if inode has no blocks at all */
2858 ex = path[depth].p_ext;
2861 EXT4_ERROR_INODE(inode,
2862 "path[%d].p_hdr == NULL",
2864 err = -EFSCORRUPTED;
2869 ee_block = le32_to_cpu(ex->ee_block);
2870 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2873 * See if the last block is inside the extent, if so split
2874 * the extent at 'end' block so we can easily remove the
2875 * tail of the first part of the split extent in
2876 * ext4_ext_rm_leaf().
2878 if (end >= ee_block && end < ex_end) {
2881 * If we're going to split the extent, note that
2882 * the cluster containing the block after 'end' is
2883 * in use to avoid freeing it when removing blocks.
2885 if (sbi->s_cluster_ratio > 1) {
2886 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2887 partial.pclu = EXT4_B2C(sbi, pblk);
2888 partial.state = nofree;
2892 * Split the extent in two so that 'end' is the last
2893 * block in the first new extent. Also we should not
2894 * fail removing space due to ENOSPC so try to use
2895 * reserved block if that happens.
2897 err = ext4_force_split_extent_at(handle, inode, &path,
2902 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2903 partial.state == initial) {
2905 * If we're punching, there's an extent to the right.
2906 * If the partial cluster hasn't been set, set it to
2907 * that extent's first cluster and its state to nofree
2908 * so it won't be freed should it contain blocks to be
2909 * removed. If it's already set (tofree/nofree), we're
2910 * retrying and keep the original partial cluster info
2911 * so a cluster marked tofree as a result of earlier
2912 * extent removal is not lost.
2915 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2920 partial.pclu = EXT4_B2C(sbi, pblk);
2921 partial.state = nofree;
2926 * We start scanning from right side, freeing all the blocks
2927 * after i_size and walking into the tree depth-wise.
2929 depth = ext_depth(inode);
2934 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2936 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2937 GFP_NOFS | __GFP_NOFAIL);
2939 ext4_journal_stop(handle);
2942 path[0].p_maxdepth = path[0].p_depth = depth;
2943 path[0].p_hdr = ext_inode_hdr(inode);
2946 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2947 err = -EFSCORRUPTED;
2953 while (i >= 0 && err == 0) {
2955 /* this is leaf block */
2956 err = ext4_ext_rm_leaf(handle, inode, path,
2957 &partial, start, end);
2958 /* root level has p_bh == NULL, brelse() eats this */
2959 brelse(path[i].p_bh);
2960 path[i].p_bh = NULL;
2965 /* this is index block */
2966 if (!path[i].p_hdr) {
2967 ext_debug(inode, "initialize header\n");
2968 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2971 if (!path[i].p_idx) {
2972 /* this level hasn't been touched yet */
2973 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2974 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2975 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2977 le16_to_cpu(path[i].p_hdr->eh_entries));
2979 /* we were already here, see at next index */
2983 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2984 i, EXT_FIRST_INDEX(path[i].p_hdr),
2986 if (ext4_ext_more_to_rm(path + i)) {
2987 struct buffer_head *bh;
2988 /* go to the next level */
2989 ext_debug(inode, "move to level %d (block %llu)\n",
2990 i + 1, ext4_idx_pblock(path[i].p_idx));
2991 memset(path + i + 1, 0, sizeof(*path));
2992 bh = read_extent_tree_block(inode, path[i].p_idx,
2996 /* should we reset i_size? */
3000 /* Yield here to deal with large extent trees.
3001 * Should be a no-op if we did IO above. */
3003 if (WARN_ON(i + 1 > depth)) {
3004 err = -EFSCORRUPTED;
3007 path[i + 1].p_bh = bh;
3009 /* save actual number of indexes since this
3010 * number is changed at the next iteration */
3011 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3014 /* we finished processing this index, go up */
3015 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3016 /* index is empty, remove it;
3017 * handle must be already prepared by the
3018 * truncatei_leaf() */
3019 err = ext4_ext_rm_idx(handle, inode, path, i);
3021 /* root level has p_bh == NULL, brelse() eats this */
3022 brelse(path[i].p_bh);
3023 path[i].p_bh = NULL;
3025 ext_debug(inode, "return to level %d\n", i);
3029 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3030 path->p_hdr->eh_entries);
3033 * if there's a partial cluster and we have removed the first extent
3034 * in the file, then we also free the partial cluster, if any
3036 if (partial.state == tofree && err == 0) {
3037 int flags = get_default_free_blocks_flags(inode);
3039 if (ext4_is_pending(inode, partial.lblk))
3040 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3041 ext4_free_blocks(handle, inode, NULL,
3042 EXT4_C2B(sbi, partial.pclu),
3043 sbi->s_cluster_ratio, flags);
3044 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3045 ext4_rereserve_cluster(inode, partial.lblk);
3046 partial.state = initial;
3049 /* TODO: flexible tree reduction should be here */
3050 if (path->p_hdr->eh_entries == 0) {
3052 * truncate to zero freed all the tree,
3053 * so we need to correct eh_depth
3055 err = ext4_ext_get_access(handle, inode, path);
3057 ext_inode_hdr(inode)->eh_depth = 0;
3058 ext_inode_hdr(inode)->eh_max =
3059 cpu_to_le16(ext4_ext_space_root(inode, 0));
3060 err = ext4_ext_dirty(handle, inode, path);
3064 ext4_ext_drop_refs(path);
3069 ext4_journal_stop(handle);
3075 * called at mount time
3077 void ext4_ext_init(struct super_block *sb)
3080 * possible initialization would be here
3083 if (ext4_has_feature_extents(sb)) {
3084 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3085 printk(KERN_INFO "EXT4-fs: file extents enabled"
3086 #ifdef AGGRESSIVE_TEST
3087 ", aggressive tests"
3089 #ifdef CHECK_BINSEARCH
3092 #ifdef EXTENTS_STATS
3097 #ifdef EXTENTS_STATS
3098 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3099 EXT4_SB(sb)->s_ext_min = 1 << 30;
3100 EXT4_SB(sb)->s_ext_max = 0;
3106 * called at umount time
3108 void ext4_ext_release(struct super_block *sb)
3110 if (!ext4_has_feature_extents(sb))
3113 #ifdef EXTENTS_STATS
3114 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3115 struct ext4_sb_info *sbi = EXT4_SB(sb);
3116 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3117 sbi->s_ext_blocks, sbi->s_ext_extents,
3118 sbi->s_ext_blocks / sbi->s_ext_extents);
3119 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3120 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3125 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3127 ext4_lblk_t ee_block;
3128 ext4_fsblk_t ee_pblock;
3129 unsigned int ee_len;
3131 ee_block = le32_to_cpu(ex->ee_block);
3132 ee_len = ext4_ext_get_actual_len(ex);
3133 ee_pblock = ext4_ext_pblock(ex);
3138 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3139 EXTENT_STATUS_WRITTEN);
3142 /* FIXME!! we need to try to merge to left or right after zero-out */
3143 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3145 ext4_fsblk_t ee_pblock;
3146 unsigned int ee_len;
3148 ee_len = ext4_ext_get_actual_len(ex);
3149 ee_pblock = ext4_ext_pblock(ex);
3150 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3155 * ext4_split_extent_at() splits an extent at given block.
3157 * @handle: the journal handle
3158 * @inode: the file inode
3159 * @path: the path to the extent
3160 * @split: the logical block where the extent is splitted.
3161 * @split_flags: indicates if the extent could be zeroout if split fails, and
3162 * the states(init or unwritten) of new extents.
3163 * @flags: flags used to insert new extent to extent tree.
3166 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3167 * of which are determined by split_flag.
3169 * There are two cases:
3170 * a> the extent are splitted into two extent.
3171 * b> split is not needed, and just mark the extent.
3173 * return 0 on success.
3175 static int ext4_split_extent_at(handle_t *handle,
3176 struct inode *inode,
3177 struct ext4_ext_path **ppath,
3182 struct ext4_ext_path *path = *ppath;
3183 ext4_fsblk_t newblock;
3184 ext4_lblk_t ee_block;
3185 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3186 struct ext4_extent *ex2 = NULL;
3187 unsigned int ee_len, depth;
3190 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3191 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3193 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3195 ext4_ext_show_leaf(inode, path);
3197 depth = ext_depth(inode);
3198 ex = path[depth].p_ext;
3199 ee_block = le32_to_cpu(ex->ee_block);
3200 ee_len = ext4_ext_get_actual_len(ex);
3201 newblock = split - ee_block + ext4_ext_pblock(ex);
3203 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3204 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3205 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3206 EXT4_EXT_MARK_UNWRIT1 |
3207 EXT4_EXT_MARK_UNWRIT2));
3209 err = ext4_ext_get_access(handle, inode, path + depth);
3213 if (split == ee_block) {
3215 * case b: block @split is the block that the extent begins with
3216 * then we just change the state of the extent, and splitting
3219 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3220 ext4_ext_mark_unwritten(ex);
3222 ext4_ext_mark_initialized(ex);
3224 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3225 ext4_ext_try_to_merge(handle, inode, path, ex);
3227 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3232 memcpy(&orig_ex, ex, sizeof(orig_ex));
3233 ex->ee_len = cpu_to_le16(split - ee_block);
3234 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3235 ext4_ext_mark_unwritten(ex);
3238 * path may lead to new leaf, not to original leaf any more
3239 * after ext4_ext_insert_extent() returns,
3241 err = ext4_ext_dirty(handle, inode, path + depth);
3243 goto fix_extent_len;
3246 ex2->ee_block = cpu_to_le32(split);
3247 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3248 ext4_ext_store_pblock(ex2, newblock);
3249 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3250 ext4_ext_mark_unwritten(ex2);
3252 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3253 if (err != -ENOSPC && err != -EDQUOT)
3256 if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3257 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3258 if (split_flag & EXT4_EXT_DATA_VALID1) {
3259 err = ext4_ext_zeroout(inode, ex2);
3260 zero_ex.ee_block = ex2->ee_block;
3261 zero_ex.ee_len = cpu_to_le16(
3262 ext4_ext_get_actual_len(ex2));
3263 ext4_ext_store_pblock(&zero_ex,
3264 ext4_ext_pblock(ex2));
3266 err = ext4_ext_zeroout(inode, ex);
3267 zero_ex.ee_block = ex->ee_block;
3268 zero_ex.ee_len = cpu_to_le16(
3269 ext4_ext_get_actual_len(ex));
3270 ext4_ext_store_pblock(&zero_ex,
3271 ext4_ext_pblock(ex));
3274 err = ext4_ext_zeroout(inode, &orig_ex);
3275 zero_ex.ee_block = orig_ex.ee_block;
3276 zero_ex.ee_len = cpu_to_le16(
3277 ext4_ext_get_actual_len(&orig_ex));
3278 ext4_ext_store_pblock(&zero_ex,
3279 ext4_ext_pblock(&orig_ex));
3283 /* update the extent length and mark as initialized */
3284 ex->ee_len = cpu_to_le16(ee_len);
3285 ext4_ext_try_to_merge(handle, inode, path, ex);
3286 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3288 /* update extent status tree */
3289 err = ext4_zeroout_es(inode, &zero_ex);
3290 /* If we failed at this point, we don't know in which
3291 * state the extent tree exactly is so don't try to fix
3292 * length of the original extent as it may do even more
3300 ex->ee_len = orig_ex.ee_len;
3302 * Ignore ext4_ext_dirty return value since we are already in error path
3303 * and err is a non-zero error code.
3305 ext4_ext_dirty(handle, inode, path + path->p_depth);
3308 ext4_ext_show_leaf(inode, path);
3313 * ext4_split_extents() splits an extent and mark extent which is covered
3314 * by @map as split_flags indicates
3316 * It may result in splitting the extent into multiple extents (up to three)
3317 * There are three possibilities:
3318 * a> There is no split required
3319 * b> Splits in two extents: Split is happening at either end of the extent
3320 * c> Splits in three extents: Somone is splitting in middle of the extent
3323 static int ext4_split_extent(handle_t *handle,
3324 struct inode *inode,
3325 struct ext4_ext_path **ppath,
3326 struct ext4_map_blocks *map,
3330 struct ext4_ext_path *path = *ppath;
3331 ext4_lblk_t ee_block;
3332 struct ext4_extent *ex;
3333 unsigned int ee_len, depth;
3336 int split_flag1, flags1;
3337 int allocated = map->m_len;
3339 depth = ext_depth(inode);
3340 ex = path[depth].p_ext;
3341 ee_block = le32_to_cpu(ex->ee_block);
3342 ee_len = ext4_ext_get_actual_len(ex);
3343 unwritten = ext4_ext_is_unwritten(ex);
3345 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3346 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3347 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3349 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3350 EXT4_EXT_MARK_UNWRIT2;
3351 if (split_flag & EXT4_EXT_DATA_VALID2)
3352 split_flag1 |= EXT4_EXT_DATA_VALID1;
3353 err = ext4_split_extent_at(handle, inode, ppath,
3354 map->m_lblk + map->m_len, split_flag1, flags1);
3358 allocated = ee_len - (map->m_lblk - ee_block);
3361 * Update path is required because previous ext4_split_extent_at() may
3362 * result in split of original leaf or extent zeroout.
3364 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3366 return PTR_ERR(path);
3367 depth = ext_depth(inode);
3368 ex = path[depth].p_ext;
3370 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3371 (unsigned long) map->m_lblk);
3372 return -EFSCORRUPTED;
3374 unwritten = ext4_ext_is_unwritten(ex);
3376 if (map->m_lblk >= ee_block) {
3377 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3379 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3380 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3381 EXT4_EXT_MARK_UNWRIT2);
3383 err = ext4_split_extent_at(handle, inode, ppath,
3384 map->m_lblk, split_flag1, flags);
3389 ext4_ext_show_leaf(inode, path);
3391 return err ? err : allocated;
3395 * This function is called by ext4_ext_map_blocks() if someone tries to write
3396 * to an unwritten extent. It may result in splitting the unwritten
3397 * extent into multiple extents (up to three - one initialized and two
3399 * There are three possibilities:
3400 * a> There is no split required: Entire extent should be initialized
3401 * b> Splits in two extents: Write is happening at either end of the extent
3402 * c> Splits in three extents: Somone is writing in middle of the extent
3405 * - The extent pointed to by 'path' is unwritten.
3406 * - The extent pointed to by 'path' contains a superset
3407 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3409 * Post-conditions on success:
3410 * - the returned value is the number of blocks beyond map->l_lblk
3411 * that are allocated and initialized.
3412 * It is guaranteed to be >= map->m_len.
3414 static int ext4_ext_convert_to_initialized(handle_t *handle,
3415 struct inode *inode,
3416 struct ext4_map_blocks *map,
3417 struct ext4_ext_path **ppath,
3420 struct ext4_ext_path *path = *ppath;
3421 struct ext4_sb_info *sbi;
3422 struct ext4_extent_header *eh;
3423 struct ext4_map_blocks split_map;
3424 struct ext4_extent zero_ex1, zero_ex2;
3425 struct ext4_extent *ex, *abut_ex;
3426 ext4_lblk_t ee_block, eof_block;
3427 unsigned int ee_len, depth, map_len = map->m_len;
3428 int allocated = 0, max_zeroout = 0;
3430 int split_flag = EXT4_EXT_DATA_VALID2;
3432 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3433 (unsigned long long)map->m_lblk, map_len);
3435 sbi = EXT4_SB(inode->i_sb);
3436 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3437 >> inode->i_sb->s_blocksize_bits;
3438 if (eof_block < map->m_lblk + map_len)
3439 eof_block = map->m_lblk + map_len;
3441 depth = ext_depth(inode);
3442 eh = path[depth].p_hdr;
3443 ex = path[depth].p_ext;
3444 ee_block = le32_to_cpu(ex->ee_block);
3445 ee_len = ext4_ext_get_actual_len(ex);
3446 zero_ex1.ee_len = 0;
3447 zero_ex2.ee_len = 0;
3449 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3451 /* Pre-conditions */
3452 BUG_ON(!ext4_ext_is_unwritten(ex));
3453 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3456 * Attempt to transfer newly initialized blocks from the currently
3457 * unwritten extent to its neighbor. This is much cheaper
3458 * than an insertion followed by a merge as those involve costly
3459 * memmove() calls. Transferring to the left is the common case in
3460 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3461 * followed by append writes.
3463 * Limitations of the current logic:
3464 * - L1: we do not deal with writes covering the whole extent.
3465 * This would require removing the extent if the transfer
3467 * - L2: we only attempt to merge with an extent stored in the
3468 * same extent tree node.
3470 if ((map->m_lblk == ee_block) &&
3471 /* See if we can merge left */
3472 (map_len < ee_len) && /*L1*/
3473 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3474 ext4_lblk_t prev_lblk;
3475 ext4_fsblk_t prev_pblk, ee_pblk;
3476 unsigned int prev_len;
3479 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3480 prev_len = ext4_ext_get_actual_len(abut_ex);
3481 prev_pblk = ext4_ext_pblock(abut_ex);
3482 ee_pblk = ext4_ext_pblock(ex);
3485 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3486 * upon those conditions:
3487 * - C1: abut_ex is initialized,
3488 * - C2: abut_ex is logically abutting ex,
3489 * - C3: abut_ex is physically abutting ex,
3490 * - C4: abut_ex can receive the additional blocks without
3491 * overflowing the (initialized) length limit.
3493 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3494 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3495 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3496 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3497 err = ext4_ext_get_access(handle, inode, path + depth);
3501 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3504 /* Shift the start of ex by 'map_len' blocks */
3505 ex->ee_block = cpu_to_le32(ee_block + map_len);
3506 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3507 ex->ee_len = cpu_to_le16(ee_len - map_len);
3508 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3510 /* Extend abut_ex by 'map_len' blocks */
3511 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3513 /* Result: number of initialized blocks past m_lblk */
3514 allocated = map_len;
3516 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3517 (map_len < ee_len) && /*L1*/
3518 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3519 /* See if we can merge right */
3520 ext4_lblk_t next_lblk;
3521 ext4_fsblk_t next_pblk, ee_pblk;
3522 unsigned int next_len;
3525 next_lblk = le32_to_cpu(abut_ex->ee_block);
3526 next_len = ext4_ext_get_actual_len(abut_ex);
3527 next_pblk = ext4_ext_pblock(abut_ex);
3528 ee_pblk = ext4_ext_pblock(ex);
3531 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3532 * upon those conditions:
3533 * - C1: abut_ex is initialized,
3534 * - C2: abut_ex is logically abutting ex,
3535 * - C3: abut_ex is physically abutting ex,
3536 * - C4: abut_ex can receive the additional blocks without
3537 * overflowing the (initialized) length limit.
3539 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3540 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3541 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3542 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3543 err = ext4_ext_get_access(handle, inode, path + depth);
3547 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3550 /* Shift the start of abut_ex by 'map_len' blocks */
3551 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3552 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3553 ex->ee_len = cpu_to_le16(ee_len - map_len);
3554 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3556 /* Extend abut_ex by 'map_len' blocks */
3557 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3559 /* Result: number of initialized blocks past m_lblk */
3560 allocated = map_len;
3564 /* Mark the block containing both extents as dirty */
3565 err = ext4_ext_dirty(handle, inode, path + depth);
3567 /* Update path to point to the right extent */
3568 path[depth].p_ext = abut_ex;
3571 allocated = ee_len - (map->m_lblk - ee_block);
3573 WARN_ON(map->m_lblk < ee_block);
3575 * It is safe to convert extent to initialized via explicit
3576 * zeroout only if extent is fully inside i_size or new_size.
3578 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3580 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3581 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3582 (inode->i_sb->s_blocksize_bits - 10);
3586 * 1. split the extent into three extents.
3587 * 2. split the extent into two extents, zeroout the head of the first
3589 * 3. split the extent into two extents, zeroout the tail of the second
3591 * 4. split the extent into two extents with out zeroout.
3592 * 5. no splitting needed, just possibly zeroout the head and / or the
3593 * tail of the extent.
3595 split_map.m_lblk = map->m_lblk;
3596 split_map.m_len = map->m_len;
3598 if (max_zeroout && (allocated > split_map.m_len)) {
3599 if (allocated <= max_zeroout) {
3602 cpu_to_le32(split_map.m_lblk +
3605 cpu_to_le16(allocated - split_map.m_len);
3606 ext4_ext_store_pblock(&zero_ex1,
3607 ext4_ext_pblock(ex) + split_map.m_lblk +
3608 split_map.m_len - ee_block);
3609 err = ext4_ext_zeroout(inode, &zero_ex1);
3612 split_map.m_len = allocated;
3614 if (split_map.m_lblk - ee_block + split_map.m_len <
3617 if (split_map.m_lblk != ee_block) {
3618 zero_ex2.ee_block = ex->ee_block;
3619 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3621 ext4_ext_store_pblock(&zero_ex2,
3622 ext4_ext_pblock(ex));
3623 err = ext4_ext_zeroout(inode, &zero_ex2);
3628 split_map.m_len += split_map.m_lblk - ee_block;
3629 split_map.m_lblk = ee_block;
3630 allocated = map->m_len;
3635 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3640 /* If we have gotten a failure, don't zero out status tree */
3642 err = ext4_zeroout_es(inode, &zero_ex1);
3644 err = ext4_zeroout_es(inode, &zero_ex2);
3646 return err ? err : allocated;
3650 * This function is called by ext4_ext_map_blocks() from
3651 * ext4_get_blocks_dio_write() when DIO to write
3652 * to an unwritten extent.
3654 * Writing to an unwritten extent may result in splitting the unwritten
3655 * extent into multiple initialized/unwritten extents (up to three)
3656 * There are three possibilities:
3657 * a> There is no split required: Entire extent should be unwritten
3658 * b> Splits in two extents: Write is happening at either end of the extent
3659 * c> Splits in three extents: Somone is writing in middle of the extent
3661 * This works the same way in the case of initialized -> unwritten conversion.
3663 * One of more index blocks maybe needed if the extent tree grow after
3664 * the unwritten extent split. To prevent ENOSPC occur at the IO
3665 * complete, we need to split the unwritten extent before DIO submit
3666 * the IO. The unwritten extent called at this time will be split
3667 * into three unwritten extent(at most). After IO complete, the part
3668 * being filled will be convert to initialized by the end_io callback function
3669 * via ext4_convert_unwritten_extents().
3671 * Returns the size of unwritten extent to be written on success.
3673 static int ext4_split_convert_extents(handle_t *handle,
3674 struct inode *inode,
3675 struct ext4_map_blocks *map,
3676 struct ext4_ext_path **ppath,
3679 struct ext4_ext_path *path = *ppath;
3680 ext4_lblk_t eof_block;
3681 ext4_lblk_t ee_block;
3682 struct ext4_extent *ex;
3683 unsigned int ee_len;
3684 int split_flag = 0, depth;
3686 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3687 (unsigned long long)map->m_lblk, map->m_len);
3689 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3690 >> inode->i_sb->s_blocksize_bits;
3691 if (eof_block < map->m_lblk + map->m_len)
3692 eof_block = map->m_lblk + map->m_len;
3694 * It is safe to convert extent to initialized via explicit
3695 * zeroout only if extent is fully inside i_size or new_size.
3697 depth = ext_depth(inode);
3698 ex = path[depth].p_ext;
3699 ee_block = le32_to_cpu(ex->ee_block);
3700 ee_len = ext4_ext_get_actual_len(ex);
3702 /* Convert to unwritten */
3703 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3704 split_flag |= EXT4_EXT_DATA_VALID1;
3705 /* Convert to initialized */
3706 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3707 split_flag |= ee_block + ee_len <= eof_block ?
3708 EXT4_EXT_MAY_ZEROOUT : 0;
3709 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3711 flags |= EXT4_GET_BLOCKS_PRE_IO;
3712 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3715 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3716 struct inode *inode,
3717 struct ext4_map_blocks *map,
3718 struct ext4_ext_path **ppath)
3720 struct ext4_ext_path *path = *ppath;
3721 struct ext4_extent *ex;
3722 ext4_lblk_t ee_block;
3723 unsigned int ee_len;
3727 depth = ext_depth(inode);
3728 ex = path[depth].p_ext;
3729 ee_block = le32_to_cpu(ex->ee_block);
3730 ee_len = ext4_ext_get_actual_len(ex);
3732 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3733 (unsigned long long)ee_block, ee_len);
3735 /* If extent is larger than requested it is a clear sign that we still
3736 * have some extent state machine issues left. So extent_split is still
3738 * TODO: Once all related issues will be fixed this situation should be
3741 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3742 #ifdef CONFIG_EXT4_DEBUG
3743 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3744 " len %u; IO logical block %llu, len %u",
3745 inode->i_ino, (unsigned long long)ee_block, ee_len,
3746 (unsigned long long)map->m_lblk, map->m_len);
3748 err = ext4_split_convert_extents(handle, inode, map, ppath,
3749 EXT4_GET_BLOCKS_CONVERT);
3752 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3754 return PTR_ERR(path);
3755 depth = ext_depth(inode);
3756 ex = path[depth].p_ext;
3759 err = ext4_ext_get_access(handle, inode, path + depth);
3762 /* first mark the extent as initialized */
3763 ext4_ext_mark_initialized(ex);
3765 /* note: ext4_ext_correct_indexes() isn't needed here because
3766 * borders are not changed
3768 ext4_ext_try_to_merge(handle, inode, path, ex);
3770 /* Mark modified extent as dirty */
3771 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3773 ext4_ext_show_leaf(inode, path);
3778 convert_initialized_extent(handle_t *handle, struct inode *inode,
3779 struct ext4_map_blocks *map,
3780 struct ext4_ext_path **ppath,
3781 unsigned int *allocated)
3783 struct ext4_ext_path *path = *ppath;
3784 struct ext4_extent *ex;
3785 ext4_lblk_t ee_block;
3786 unsigned int ee_len;
3791 * Make sure that the extent is no bigger than we support with
3794 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3795 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3797 depth = ext_depth(inode);
3798 ex = path[depth].p_ext;
3799 ee_block = le32_to_cpu(ex->ee_block);
3800 ee_len = ext4_ext_get_actual_len(ex);
3802 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3803 (unsigned long long)ee_block, ee_len);
3805 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3806 err = ext4_split_convert_extents(handle, inode, map, ppath,
3807 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3810 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3812 return PTR_ERR(path);
3813 depth = ext_depth(inode);
3814 ex = path[depth].p_ext;
3816 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3817 (unsigned long) map->m_lblk);
3818 return -EFSCORRUPTED;
3822 err = ext4_ext_get_access(handle, inode, path + depth);
3825 /* first mark the extent as unwritten */
3826 ext4_ext_mark_unwritten(ex);
3828 /* note: ext4_ext_correct_indexes() isn't needed here because
3829 * borders are not changed
3831 ext4_ext_try_to_merge(handle, inode, path, ex);
3833 /* Mark modified extent as dirty */
3834 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3837 ext4_ext_show_leaf(inode, path);
3839 ext4_update_inode_fsync_trans(handle, inode, 1);
3841 map->m_flags |= EXT4_MAP_UNWRITTEN;
3842 if (*allocated > map->m_len)
3843 *allocated = map->m_len;
3844 map->m_len = *allocated;
3849 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3850 struct ext4_map_blocks *map,
3851 struct ext4_ext_path **ppath, int flags,
3852 unsigned int allocated, ext4_fsblk_t newblock)
3854 struct ext4_ext_path __maybe_unused *path = *ppath;
3858 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3859 (unsigned long long)map->m_lblk, map->m_len, flags,
3861 ext4_ext_show_leaf(inode, path);
3864 * When writing into unwritten space, we should not fail to
3865 * allocate metadata blocks for the new extent block if needed.
3867 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3869 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3870 allocated, newblock);
3872 /* get_block() before submitting IO, split the extent */
3873 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3874 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3875 flags | EXT4_GET_BLOCKS_CONVERT);
3881 * shouldn't get a 0 return when splitting an extent unless
3882 * m_len is 0 (bug) or extent has been corrupted
3884 if (unlikely(ret == 0)) {
3885 EXT4_ERROR_INODE(inode,
3886 "unexpected ret == 0, m_len = %u",
3888 err = -EFSCORRUPTED;
3891 map->m_flags |= EXT4_MAP_UNWRITTEN;
3894 /* IO end_io complete, convert the filled extent to written */
3895 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3896 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3900 ext4_update_inode_fsync_trans(handle, inode, 1);
3903 /* buffered IO cases */
3905 * repeat fallocate creation request
3906 * we already have an unwritten extent
3908 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3909 map->m_flags |= EXT4_MAP_UNWRITTEN;
3913 /* buffered READ or buffered write_begin() lookup */
3914 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3916 * We have blocks reserved already. We
3917 * return allocated blocks so that delalloc
3918 * won't do block reservation for us. But
3919 * the buffer head will be unmapped so that
3920 * a read from the block returns 0s.
3922 map->m_flags |= EXT4_MAP_UNWRITTEN;
3927 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3928 * For buffered writes, at writepage time, etc. Convert a
3929 * discovered unwritten extent to written.
3931 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3936 ext4_update_inode_fsync_trans(handle, inode, 1);
3938 * shouldn't get a 0 return when converting an unwritten extent
3939 * unless m_len is 0 (bug) or extent has been corrupted
3941 if (unlikely(ret == 0)) {
3942 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3944 err = -EFSCORRUPTED;
3950 map->m_flags |= EXT4_MAP_NEW;
3952 map->m_flags |= EXT4_MAP_MAPPED;
3954 map->m_pblk = newblock;
3955 if (allocated > map->m_len)
3956 allocated = map->m_len;
3957 map->m_len = allocated;
3958 ext4_ext_show_leaf(inode, path);
3960 return err ? err : allocated;
3964 * get_implied_cluster_alloc - check to see if the requested
3965 * allocation (in the map structure) overlaps with a cluster already
3966 * allocated in an extent.
3967 * @sb The filesystem superblock structure
3968 * @map The requested lblk->pblk mapping
3969 * @ex The extent structure which might contain an implied
3970 * cluster allocation
3972 * This function is called by ext4_ext_map_blocks() after we failed to
3973 * find blocks that were already in the inode's extent tree. Hence,
3974 * we know that the beginning of the requested region cannot overlap
3975 * the extent from the inode's extent tree. There are three cases we
3976 * want to catch. The first is this case:
3978 * |--- cluster # N--|
3979 * |--- extent ---| |---- requested region ---|
3982 * The second case that we need to test for is this one:
3984 * |--------- cluster # N ----------------|
3985 * |--- requested region --| |------- extent ----|
3986 * |=======================|
3988 * The third case is when the requested region lies between two extents
3989 * within the same cluster:
3990 * |------------- cluster # N-------------|
3991 * |----- ex -----| |---- ex_right ----|
3992 * |------ requested region ------|
3993 * |================|
3995 * In each of the above cases, we need to set the map->m_pblk and
3996 * map->m_len so it corresponds to the return the extent labelled as
3997 * "|====|" from cluster #N, since it is already in use for data in
3998 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3999 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4000 * as a new "allocated" block region. Otherwise, we will return 0 and
4001 * ext4_ext_map_blocks() will then allocate one or more new clusters
4002 * by calling ext4_mb_new_blocks().
4004 static int get_implied_cluster_alloc(struct super_block *sb,
4005 struct ext4_map_blocks *map,
4006 struct ext4_extent *ex,
4007 struct ext4_ext_path *path)
4009 struct ext4_sb_info *sbi = EXT4_SB(sb);
4010 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4011 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4012 ext4_lblk_t rr_cluster_start;
4013 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4014 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4015 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4017 /* The extent passed in that we are trying to match */
4018 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4019 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4021 /* The requested region passed into ext4_map_blocks() */
4022 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4024 if ((rr_cluster_start == ex_cluster_end) ||
4025 (rr_cluster_start == ex_cluster_start)) {
4026 if (rr_cluster_start == ex_cluster_end)
4027 ee_start += ee_len - 1;
4028 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4029 map->m_len = min(map->m_len,
4030 (unsigned) sbi->s_cluster_ratio - c_offset);
4032 * Check for and handle this case:
4034 * |--------- cluster # N-------------|
4035 * |------- extent ----|
4036 * |--- requested region ---|
4040 if (map->m_lblk < ee_block)
4041 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4044 * Check for the case where there is already another allocated
4045 * block to the right of 'ex' but before the end of the cluster.
4047 * |------------- cluster # N-------------|
4048 * |----- ex -----| |---- ex_right ----|
4049 * |------ requested region ------|
4050 * |================|
4052 if (map->m_lblk > ee_block) {
4053 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4054 map->m_len = min(map->m_len, next - map->m_lblk);
4057 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4061 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4067 * Block allocation/map/preallocation routine for extents based files
4070 * Need to be called with
4071 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4072 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4074 * return > 0, number of blocks already mapped/allocated
4075 * if create == 0 and these are pre-allocated blocks
4076 * buffer head is unmapped
4077 * otherwise blocks are mapped
4079 * return = 0, if plain look up failed (blocks have not been allocated)
4080 * buffer head is unmapped
4082 * return < 0, error case.
4084 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4085 struct ext4_map_blocks *map, int flags)
4087 struct ext4_ext_path *path = NULL;
4088 struct ext4_extent newex, *ex, ex2;
4089 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4090 ext4_fsblk_t newblock = 0, pblk;
4091 int err = 0, depth, ret;
4092 unsigned int allocated = 0, offset = 0;
4093 unsigned int allocated_clusters = 0;
4094 struct ext4_allocation_request ar;
4095 ext4_lblk_t cluster_offset;
4097 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4098 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4100 /* find extent for this block */
4101 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4103 err = PTR_ERR(path);
4108 depth = ext_depth(inode);
4111 * consistent leaf must not be empty;
4112 * this situation is possible, though, _during_ tree modification;
4113 * this is why assert can't be put in ext4_find_extent()
4115 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4116 EXT4_ERROR_INODE(inode, "bad extent address "
4117 "lblock: %lu, depth: %d pblock %lld",
4118 (unsigned long) map->m_lblk, depth,
4119 path[depth].p_block);
4120 err = -EFSCORRUPTED;
4124 ex = path[depth].p_ext;
4126 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4127 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4128 unsigned short ee_len;
4132 * unwritten extents are treated as holes, except that
4133 * we split out initialized portions during a write.
4135 ee_len = ext4_ext_get_actual_len(ex);
4137 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4139 /* if found extent covers block, simply return it */
4140 if (in_range(map->m_lblk, ee_block, ee_len)) {
4141 newblock = map->m_lblk - ee_block + ee_start;
4142 /* number of remaining blocks in the extent */
4143 allocated = ee_len - (map->m_lblk - ee_block);
4144 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4145 map->m_lblk, ee_block, ee_len, newblock);
4148 * If the extent is initialized check whether the
4149 * caller wants to convert it to unwritten.
4151 if ((!ext4_ext_is_unwritten(ex)) &&
4152 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4153 err = convert_initialized_extent(handle,
4154 inode, map, &path, &allocated);
4156 } else if (!ext4_ext_is_unwritten(ex)) {
4157 map->m_flags |= EXT4_MAP_MAPPED;
4158 map->m_pblk = newblock;
4159 if (allocated > map->m_len)
4160 allocated = map->m_len;
4161 map->m_len = allocated;
4162 ext4_ext_show_leaf(inode, path);
4166 ret = ext4_ext_handle_unwritten_extents(
4167 handle, inode, map, &path, flags,
4168 allocated, newblock);
4178 * requested block isn't allocated yet;
4179 * we couldn't try to create block if create flag is zero
4181 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4182 ext4_lblk_t hole_start, hole_len;
4184 hole_start = map->m_lblk;
4185 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4187 * put just found gap into cache to speed up
4188 * subsequent requests
4190 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4192 /* Update hole_len to reflect hole size after map->m_lblk */
4193 if (hole_start != map->m_lblk)
4194 hole_len -= map->m_lblk - hole_start;
4196 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4202 * Okay, we need to do block allocation.
4204 newex.ee_block = cpu_to_le32(map->m_lblk);
4205 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4208 * If we are doing bigalloc, check to see if the extent returned
4209 * by ext4_find_extent() implies a cluster we can use.
4211 if (cluster_offset && ex &&
4212 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4213 ar.len = allocated = map->m_len;
4214 newblock = map->m_pblk;
4215 goto got_allocated_blocks;
4218 /* find neighbour allocated blocks */
4219 ar.lleft = map->m_lblk;
4220 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4223 ar.lright = map->m_lblk;
4224 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4228 /* Check if the extent after searching to the right implies a
4229 * cluster we can use. */
4230 if ((sbi->s_cluster_ratio > 1) && err &&
4231 get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4232 ar.len = allocated = map->m_len;
4233 newblock = map->m_pblk;
4234 goto got_allocated_blocks;
4238 * See if request is beyond maximum number of blocks we can have in
4239 * a single extent. For an initialized extent this limit is
4240 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4241 * EXT_UNWRITTEN_MAX_LEN.
4243 if (map->m_len > EXT_INIT_MAX_LEN &&
4244 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4245 map->m_len = EXT_INIT_MAX_LEN;
4246 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4247 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4248 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4250 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4251 newex.ee_len = cpu_to_le16(map->m_len);
4252 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4254 allocated = ext4_ext_get_actual_len(&newex);
4256 allocated = map->m_len;
4258 /* allocate new block */
4260 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4261 ar.logical = map->m_lblk;
4263 * We calculate the offset from the beginning of the cluster
4264 * for the logical block number, since when we allocate a
4265 * physical cluster, the physical block should start at the
4266 * same offset from the beginning of the cluster. This is
4267 * needed so that future calls to get_implied_cluster_alloc()
4270 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4271 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4273 ar.logical -= offset;
4274 if (S_ISREG(inode->i_mode))
4275 ar.flags = EXT4_MB_HINT_DATA;
4277 /* disable in-core preallocation for non-regular files */
4279 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4280 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4281 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4282 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4283 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4284 ar.flags |= EXT4_MB_USE_RESERVED;
4285 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4288 allocated_clusters = ar.len;
4289 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4290 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4291 ar.goal, newblock, ar.len, allocated);
4292 if (ar.len > allocated)
4295 got_allocated_blocks:
4296 /* try to insert new extent into found leaf and return */
4297 pblk = newblock + offset;
4298 ext4_ext_store_pblock(&newex, pblk);
4299 newex.ee_len = cpu_to_le16(ar.len);
4300 /* Mark unwritten */
4301 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4302 ext4_ext_mark_unwritten(&newex);
4303 map->m_flags |= EXT4_MAP_UNWRITTEN;
4306 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4308 if (allocated_clusters) {
4312 * free data blocks we just allocated.
4313 * not a good idea to call discard here directly,
4314 * but otherwise we'd need to call it every free().
4316 ext4_discard_preallocations(inode, 0);
4317 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4318 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4319 ext4_free_blocks(handle, inode, NULL, newblock,
4320 EXT4_C2B(sbi, allocated_clusters),
4327 * Reduce the reserved cluster count to reflect successful deferred
4328 * allocation of delayed allocated clusters or direct allocation of
4329 * clusters discovered to be delayed allocated. Once allocated, a
4330 * cluster is not included in the reserved count.
4332 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4333 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4335 * When allocating delayed allocated clusters, simply
4336 * reduce the reserved cluster count and claim quota
4338 ext4_da_update_reserve_space(inode, allocated_clusters,
4341 ext4_lblk_t lblk, len;
4345 * When allocating non-delayed allocated clusters
4346 * (from fallocate, filemap, DIO, or clusters
4347 * allocated when delalloc has been disabled by
4348 * ext4_nonda_switch), reduce the reserved cluster
4349 * count by the number of allocated clusters that
4350 * have previously been delayed allocated. Quota
4351 * has been claimed by ext4_mb_new_blocks() above,
4352 * so release the quota reservations made for any
4353 * previously delayed allocated clusters.
4355 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4356 len = allocated_clusters << sbi->s_cluster_bits;
4357 n = ext4_es_delayed_clu(inode, lblk, len);
4359 ext4_da_update_reserve_space(inode, (int) n, 0);
4364 * Cache the extent and update transaction to commit on fdatasync only
4365 * when it is _not_ an unwritten extent.
4367 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4368 ext4_update_inode_fsync_trans(handle, inode, 1);
4370 ext4_update_inode_fsync_trans(handle, inode, 0);
4372 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4374 map->m_len = ar.len;
4375 allocated = map->m_len;
4376 ext4_ext_show_leaf(inode, path);
4378 ext4_ext_drop_refs(path);
4381 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4382 err ? err : allocated);
4383 return err ? err : allocated;
4386 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4388 struct super_block *sb = inode->i_sb;
4389 ext4_lblk_t last_block;
4393 * TODO: optimization is possible here.
4394 * Probably we need not scan at all,
4395 * because page truncation is enough.
4398 /* we have to know where to truncate from in crash case */
4399 EXT4_I(inode)->i_disksize = inode->i_size;
4400 err = ext4_mark_inode_dirty(handle, inode);
4404 last_block = (inode->i_size + sb->s_blocksize - 1)
4405 >> EXT4_BLOCK_SIZE_BITS(sb);
4407 err = ext4_es_remove_extent(inode, last_block,
4408 EXT_MAX_BLOCKS - last_block);
4409 if (err == -ENOMEM) {
4410 memalloc_retry_wait(GFP_ATOMIC);
4416 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4417 if (err == -ENOMEM) {
4418 memalloc_retry_wait(GFP_ATOMIC);
4419 goto retry_remove_space;
4424 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4425 ext4_lblk_t len, loff_t new_size,
4428 struct inode *inode = file_inode(file);
4430 int ret = 0, ret2 = 0, ret3 = 0;
4433 struct ext4_map_blocks map;
4434 unsigned int credits;
4437 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4438 map.m_lblk = offset;
4441 * Don't normalize the request if it can fit in one extent so
4442 * that it doesn't get unnecessarily split into multiple
4445 if (len <= EXT_UNWRITTEN_MAX_LEN)
4446 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4449 * credits to insert 1 extent into extent tree
4451 credits = ext4_chunk_trans_blocks(inode, len);
4452 depth = ext_depth(inode);
4457 * Recalculate credits when extent tree depth changes.
4459 if (depth != ext_depth(inode)) {
4460 credits = ext4_chunk_trans_blocks(inode, len);
4461 depth = ext_depth(inode);
4464 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4466 if (IS_ERR(handle)) {
4467 ret = PTR_ERR(handle);
4470 ret = ext4_map_blocks(handle, inode, &map, flags);
4472 ext4_debug("inode #%lu: block %u: len %u: "
4473 "ext4_ext_map_blocks returned %d",
4474 inode->i_ino, map.m_lblk,
4476 ext4_mark_inode_dirty(handle, inode);
4477 ext4_journal_stop(handle);
4481 * allow a full retry cycle for any remaining allocations
4485 map.m_len = len = len - ret;
4486 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4487 inode->i_ctime = current_time(inode);
4489 if (epos > new_size)
4491 if (ext4_update_inode_size(inode, epos) & 0x1)
4492 inode->i_mtime = inode->i_ctime;
4494 ret2 = ext4_mark_inode_dirty(handle, inode);
4495 ext4_update_inode_fsync_trans(handle, inode, 1);
4496 ret3 = ext4_journal_stop(handle);
4497 ret2 = ret3 ? ret3 : ret2;
4501 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4504 return ret > 0 ? ret2 : ret;
4507 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4509 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4511 static long ext4_zero_range(struct file *file, loff_t offset,
4512 loff_t len, int mode)
4514 struct inode *inode = file_inode(file);
4515 struct address_space *mapping = file->f_mapping;
4516 handle_t *handle = NULL;
4517 unsigned int max_blocks;
4518 loff_t new_size = 0;
4522 int partial_begin, partial_end;
4525 unsigned int blkbits = inode->i_blkbits;
4527 trace_ext4_zero_range(inode, offset, len, mode);
4529 /* Call ext4_force_commit to flush all data in case of data=journal. */
4530 if (ext4_should_journal_data(inode)) {
4531 ret = ext4_force_commit(inode->i_sb);
4537 * Round up offset. This is not fallocate, we need to zero out
4538 * blocks, so convert interior block aligned part of the range to
4539 * unwritten and possibly manually zero out unaligned parts of the
4542 start = round_up(offset, 1 << blkbits);
4543 end = round_down((offset + len), 1 << blkbits);
4545 if (start < offset || end > offset + len)
4547 partial_begin = offset & ((1 << blkbits) - 1);
4548 partial_end = (offset + len) & ((1 << blkbits) - 1);
4550 lblk = start >> blkbits;
4551 max_blocks = (end >> blkbits);
4552 if (max_blocks < lblk)
4560 * Indirect files do not support unwritten extents
4562 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4567 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4568 (offset + len > inode->i_size ||
4569 offset + len > EXT4_I(inode)->i_disksize)) {
4570 new_size = offset + len;
4571 ret = inode_newsize_ok(inode, new_size);
4576 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4578 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4579 inode_dio_wait(inode);
4581 ret = file_modified(file);
4585 /* Preallocate the range including the unaligned edges */
4586 if (partial_begin || partial_end) {
4587 ret = ext4_alloc_file_blocks(file,
4588 round_down(offset, 1 << blkbits) >> blkbits,
4589 (round_up((offset + len), 1 << blkbits) -
4590 round_down(offset, 1 << blkbits)) >> blkbits,
4597 /* Zero range excluding the unaligned edges */
4598 if (max_blocks > 0) {
4599 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4603 * Prevent page faults from reinstantiating pages we have
4604 * released from page cache.
4606 filemap_invalidate_lock(mapping);
4608 ret = ext4_break_layouts(inode);
4610 filemap_invalidate_unlock(mapping);
4614 ret = ext4_update_disksize_before_punch(inode, offset, len);
4616 filemap_invalidate_unlock(mapping);
4619 /* Now release the pages and zero block aligned part of pages */
4620 truncate_pagecache_range(inode, start, end - 1);
4621 inode->i_mtime = inode->i_ctime = current_time(inode);
4623 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4625 filemap_invalidate_unlock(mapping);
4629 if (!partial_begin && !partial_end)
4633 * In worst case we have to writeout two nonadjacent unwritten
4634 * blocks and update the inode
4636 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4637 if (ext4_should_journal_data(inode))
4639 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4640 if (IS_ERR(handle)) {
4641 ret = PTR_ERR(handle);
4642 ext4_std_error(inode->i_sb, ret);
4646 inode->i_mtime = inode->i_ctime = current_time(inode);
4648 ext4_update_inode_size(inode, new_size);
4649 ret = ext4_mark_inode_dirty(handle, inode);
4652 /* Zero out partial block at the edges of the range */
4653 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4655 ext4_update_inode_fsync_trans(handle, inode, 1);
4657 if (file->f_flags & O_SYNC)
4658 ext4_handle_sync(handle);
4661 ext4_journal_stop(handle);
4663 inode_unlock(inode);
4668 * preallocate space for a file. This implements ext4's fallocate file
4669 * operation, which gets called from sys_fallocate system call.
4670 * For block-mapped files, posix_fallocate should fall back to the method
4671 * of writing zeroes to the required new blocks (the same behavior which is
4672 * expected for file systems which do not support fallocate() system call).
4674 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4676 struct inode *inode = file_inode(file);
4677 loff_t new_size = 0;
4678 unsigned int max_blocks;
4682 unsigned int blkbits = inode->i_blkbits;
4685 * Encrypted inodes can't handle collapse range or insert
4686 * range since we would need to re-encrypt blocks with a
4687 * different IV or XTS tweak (which are based on the logical
4690 if (IS_ENCRYPTED(inode) &&
4691 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4694 /* Return error if mode is not supported */
4695 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4696 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4697 FALLOC_FL_INSERT_RANGE))
4701 ret = ext4_convert_inline_data(inode);
4702 inode_unlock(inode);
4706 if (mode & FALLOC_FL_PUNCH_HOLE) {
4707 ret = ext4_punch_hole(file, offset, len);
4711 if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4712 ret = ext4_collapse_range(file, offset, len);
4716 if (mode & FALLOC_FL_INSERT_RANGE) {
4717 ret = ext4_insert_range(file, offset, len);
4721 if (mode & FALLOC_FL_ZERO_RANGE) {
4722 ret = ext4_zero_range(file, offset, len, mode);
4725 trace_ext4_fallocate_enter(inode, offset, len, mode);
4726 lblk = offset >> blkbits;
4728 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4729 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4734 * We only support preallocation for extent-based files only
4736 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4741 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4742 (offset + len > inode->i_size ||
4743 offset + len > EXT4_I(inode)->i_disksize)) {
4744 new_size = offset + len;
4745 ret = inode_newsize_ok(inode, new_size);
4750 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4751 inode_dio_wait(inode);
4753 ret = file_modified(file);
4757 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4761 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4762 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4763 EXT4_I(inode)->i_sync_tid);
4766 inode_unlock(inode);
4767 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4773 * This function convert a range of blocks to written extents
4774 * The caller of this function will pass the start offset and the size.
4775 * all unwritten extents within this range will be converted to
4778 * This function is called from the direct IO end io call back
4779 * function, to convert the fallocated extents after IO is completed.
4780 * Returns 0 on success.
4782 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4783 loff_t offset, ssize_t len)
4785 unsigned int max_blocks;
4786 int ret = 0, ret2 = 0, ret3 = 0;
4787 struct ext4_map_blocks map;
4788 unsigned int blkbits = inode->i_blkbits;
4789 unsigned int credits = 0;
4791 map.m_lblk = offset >> blkbits;
4792 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4796 * credits to insert 1 extent into extent tree
4798 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4800 while (ret >= 0 && ret < max_blocks) {
4802 map.m_len = (max_blocks -= ret);
4804 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4806 if (IS_ERR(handle)) {
4807 ret = PTR_ERR(handle);
4811 ret = ext4_map_blocks(handle, inode, &map,
4812 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4814 ext4_warning(inode->i_sb,
4815 "inode #%lu: block %u: len %u: "
4816 "ext4_ext_map_blocks returned %d",
4817 inode->i_ino, map.m_lblk,
4819 ret2 = ext4_mark_inode_dirty(handle, inode);
4821 ret3 = ext4_journal_stop(handle);
4826 if (ret <= 0 || ret2)
4829 return ret > 0 ? ret2 : ret;
4832 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4834 int ret = 0, err = 0;
4835 struct ext4_io_end_vec *io_end_vec;
4838 * This is somewhat ugly but the idea is clear: When transaction is
4839 * reserved, everything goes into it. Otherwise we rather start several
4840 * smaller transactions for conversion of each extent separately.
4843 handle = ext4_journal_start_reserved(handle,
4844 EXT4_HT_EXT_CONVERT);
4846 return PTR_ERR(handle);
4849 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4850 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4858 err = ext4_journal_stop(handle);
4860 return ret < 0 ? ret : err;
4863 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4867 int blockbits = inode->i_sb->s_blocksize_bits;
4872 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4873 struct ext4_iloc iloc;
4874 int offset; /* offset of xattr in inode */
4876 error = ext4_get_inode_loc(inode, &iloc);
4879 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4880 offset = EXT4_GOOD_OLD_INODE_SIZE +
4881 EXT4_I(inode)->i_extra_isize;
4883 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4885 iomap_type = IOMAP_INLINE;
4886 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4887 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4888 length = inode->i_sb->s_blocksize;
4889 iomap_type = IOMAP_MAPPED;
4891 /* no in-inode or external block for xattr, so return -ENOENT */
4896 iomap->addr = physical;
4898 iomap->length = length;
4899 iomap->type = iomap_type;
4905 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4906 loff_t length, unsigned flags,
4907 struct iomap *iomap, struct iomap *srcmap)
4911 error = ext4_iomap_xattr_fiemap(inode, iomap);
4912 if (error == 0 && (offset >= iomap->length))
4917 static const struct iomap_ops ext4_iomap_xattr_ops = {
4918 .iomap_begin = ext4_iomap_xattr_begin,
4921 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4925 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4926 maxbytes = inode->i_sb->s_maxbytes;
4928 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4932 if (start > maxbytes)
4936 * Shrink request scope to what the fs can actually handle.
4938 if (*len > maxbytes || (maxbytes - *len) < start)
4939 *len = maxbytes - start;
4943 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4948 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4949 error = ext4_ext_precache(inode);
4952 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4956 * For bitmap files the maximum size limit could be smaller than
4957 * s_maxbytes, so check len here manually instead of just relying on the
4960 error = ext4_fiemap_check_ranges(inode, start, &len);
4964 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4965 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4966 return iomap_fiemap(inode, fieinfo, start, len,
4967 &ext4_iomap_xattr_ops);
4970 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4973 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4974 __u64 start, __u64 len)
4976 ext4_lblk_t start_blk, len_blks;
4980 if (ext4_has_inline_data(inode)) {
4983 down_read(&EXT4_I(inode)->xattr_sem);
4984 has_inline = ext4_has_inline_data(inode);
4985 up_read(&EXT4_I(inode)->xattr_sem);
4990 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4991 error = ext4_ext_precache(inode);
4994 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4997 error = fiemap_prep(inode, fieinfo, start, &len, 0);
5001 error = ext4_fiemap_check_ranges(inode, start, &len);
5005 start_blk = start >> inode->i_sb->s_blocksize_bits;
5006 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5007 if (last_blk >= EXT_MAX_BLOCKS)
5008 last_blk = EXT_MAX_BLOCKS-1;
5009 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5012 * Walk the extent tree gathering extent information
5013 * and pushing extents back to the user.
5015 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5019 * ext4_ext_shift_path_extents:
5020 * Shift the extents of a path structure lying between path[depth].p_ext
5021 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5022 * if it is right shift or left shift operation.
5025 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5026 struct inode *inode, handle_t *handle,
5027 enum SHIFT_DIRECTION SHIFT)
5030 struct ext4_extent *ex_start, *ex_last;
5031 bool update = false;
5032 int credits, restart_credits;
5033 depth = path->p_depth;
5035 while (depth >= 0) {
5036 if (depth == path->p_depth) {
5037 ex_start = path[depth].p_ext;
5039 return -EFSCORRUPTED;
5041 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5042 /* leaf + sb + inode */
5044 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5046 /* extent tree + sb + inode */
5047 credits = depth + 2;
5050 restart_credits = ext4_writepage_trans_blocks(inode);
5051 err = ext4_datasem_ensure_credits(handle, inode, credits,
5052 restart_credits, 0);
5059 err = ext4_ext_get_access(handle, inode, path + depth);
5063 while (ex_start <= ex_last) {
5064 if (SHIFT == SHIFT_LEFT) {
5065 le32_add_cpu(&ex_start->ee_block,
5067 /* Try to merge to the left. */
5069 EXT_FIRST_EXTENT(path[depth].p_hdr))
5071 ext4_ext_try_to_merge_right(inode,
5072 path, ex_start - 1))
5077 le32_add_cpu(&ex_last->ee_block, shift);
5078 ext4_ext_try_to_merge_right(inode, path,
5083 err = ext4_ext_dirty(handle, inode, path + depth);
5087 if (--depth < 0 || !update)
5091 /* Update index too */
5092 err = ext4_ext_get_access(handle, inode, path + depth);
5096 if (SHIFT == SHIFT_LEFT)
5097 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5099 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5100 err = ext4_ext_dirty(handle, inode, path + depth);
5104 /* we are done if current index is not a starting index */
5105 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5116 * ext4_ext_shift_extents:
5117 * All the extents which lies in the range from @start to the last allocated
5118 * block for the @inode are shifted either towards left or right (depending
5119 * upon @SHIFT) by @shift blocks.
5120 * On success, 0 is returned, error otherwise.
5123 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5124 ext4_lblk_t start, ext4_lblk_t shift,
5125 enum SHIFT_DIRECTION SHIFT)
5127 struct ext4_ext_path *path;
5129 struct ext4_extent *extent;
5130 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5131 ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5133 /* Let path point to the last extent */
5134 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5137 return PTR_ERR(path);
5139 depth = path->p_depth;
5140 extent = path[depth].p_ext;
5144 stop = le32_to_cpu(extent->ee_block);
5147 * For left shifts, make sure the hole on the left is big enough to
5148 * accommodate the shift. For right shifts, make sure the last extent
5149 * won't be shifted beyond EXT_MAX_BLOCKS.
5151 if (SHIFT == SHIFT_LEFT) {
5152 path = ext4_find_extent(inode, start - 1, &path,
5155 return PTR_ERR(path);
5156 depth = path->p_depth;
5157 extent = path[depth].p_ext;
5159 ex_start = le32_to_cpu(extent->ee_block);
5160 ex_end = le32_to_cpu(extent->ee_block) +
5161 ext4_ext_get_actual_len(extent);
5167 if ((start == ex_start && shift > ex_start) ||
5168 (shift > start - ex_end)) {
5173 if (shift > EXT_MAX_BLOCKS -
5174 (stop + ext4_ext_get_actual_len(extent))) {
5181 * In case of left shift, iterator points to start and it is increased
5182 * till we reach stop. In case of right shift, iterator points to stop
5183 * and it is decreased till we reach start.
5186 if (SHIFT == SHIFT_LEFT)
5191 if (tmp != EXT_MAX_BLOCKS)
5195 * Its safe to start updating extents. Start and stop are unsigned, so
5196 * in case of right shift if extent with 0 block is reached, iterator
5197 * becomes NULL to indicate the end of the loop.
5199 while (iterator && start <= stop) {
5200 path = ext4_find_extent(inode, *iterator, &path,
5203 return PTR_ERR(path);
5204 depth = path->p_depth;
5205 extent = path[depth].p_ext;
5207 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5208 (unsigned long) *iterator);
5209 return -EFSCORRUPTED;
5211 if (SHIFT == SHIFT_LEFT && *iterator >
5212 le32_to_cpu(extent->ee_block)) {
5213 /* Hole, move to the next extent */
5214 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5215 path[depth].p_ext++;
5217 *iterator = ext4_ext_next_allocated_block(path);
5223 if (SHIFT == SHIFT_LEFT) {
5224 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5225 *iterator = le32_to_cpu(extent->ee_block) +
5226 ext4_ext_get_actual_len(extent);
5228 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5229 if (le32_to_cpu(extent->ee_block) > 0)
5230 *iterator = le32_to_cpu(extent->ee_block) - 1;
5232 /* Beginning is reached, end of the loop */
5234 /* Update path extent in case we need to stop */
5235 while (le32_to_cpu(extent->ee_block) < start)
5237 path[depth].p_ext = extent;
5239 ret = ext4_ext_shift_path_extents(path, shift, inode,
5241 /* iterator can be NULL which means we should break */
5248 ext4_ext_drop_refs(path);
5254 * ext4_collapse_range:
5255 * This implements the fallocate's collapse range functionality for ext4
5256 * Returns: 0 and non-zero on error.
5258 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5260 struct inode *inode = file_inode(file);
5261 struct super_block *sb = inode->i_sb;
5262 struct address_space *mapping = inode->i_mapping;
5263 ext4_lblk_t punch_start, punch_stop;
5265 unsigned int credits;
5266 loff_t new_size, ioffset;
5270 * We need to test this early because xfstests assumes that a
5271 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5272 * system does not support collapse range.
5274 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5277 /* Collapse range works only on fs cluster size aligned regions. */
5278 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5281 trace_ext4_collapse_range(inode, offset, len);
5283 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5284 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5286 /* Call ext4_force_commit to flush all data in case of data=journal. */
5287 if (ext4_should_journal_data(inode)) {
5288 ret = ext4_force_commit(inode->i_sb);
5295 * There is no need to overlap collapse range with EOF, in which case
5296 * it is effectively a truncate operation
5298 if (offset + len >= inode->i_size) {
5303 /* Currently just for extent based files */
5304 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5309 /* Wait for existing dio to complete */
5310 inode_dio_wait(inode);
5312 ret = file_modified(file);
5317 * Prevent page faults from reinstantiating pages we have released from
5320 filemap_invalidate_lock(mapping);
5322 ret = ext4_break_layouts(inode);
5327 * Need to round down offset to be aligned with page size boundary
5328 * for page size > block size.
5330 ioffset = round_down(offset, PAGE_SIZE);
5332 * Write tail of the last page before removed range since it will get
5333 * removed from the page cache below.
5335 ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5339 * Write data that will be shifted to preserve them when discarding
5340 * page cache below. We are also protected from pages becoming dirty
5341 * by i_rwsem and invalidate_lock.
5343 ret = filemap_write_and_wait_range(mapping, offset + len,
5347 truncate_pagecache(inode, ioffset);
5349 credits = ext4_writepage_trans_blocks(inode);
5350 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5351 if (IS_ERR(handle)) {
5352 ret = PTR_ERR(handle);
5355 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5357 down_write(&EXT4_I(inode)->i_data_sem);
5358 ext4_discard_preallocations(inode, 0);
5360 ret = ext4_es_remove_extent(inode, punch_start,
5361 EXT_MAX_BLOCKS - punch_start);
5363 up_write(&EXT4_I(inode)->i_data_sem);
5367 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5369 up_write(&EXT4_I(inode)->i_data_sem);
5372 ext4_discard_preallocations(inode, 0);
5374 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5375 punch_stop - punch_start, SHIFT_LEFT);
5377 up_write(&EXT4_I(inode)->i_data_sem);
5381 new_size = inode->i_size - len;
5382 i_size_write(inode, new_size);
5383 EXT4_I(inode)->i_disksize = new_size;
5385 up_write(&EXT4_I(inode)->i_data_sem);
5387 ext4_handle_sync(handle);
5388 inode->i_mtime = inode->i_ctime = current_time(inode);
5389 ret = ext4_mark_inode_dirty(handle, inode);
5390 ext4_update_inode_fsync_trans(handle, inode, 1);
5393 ext4_journal_stop(handle);
5395 filemap_invalidate_unlock(mapping);
5397 inode_unlock(inode);
5402 * ext4_insert_range:
5403 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5404 * The data blocks starting from @offset to the EOF are shifted by @len
5405 * towards right to create a hole in the @inode. Inode size is increased
5407 * Returns 0 on success, error otherwise.
5409 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5411 struct inode *inode = file_inode(file);
5412 struct super_block *sb = inode->i_sb;
5413 struct address_space *mapping = inode->i_mapping;
5415 struct ext4_ext_path *path;
5416 struct ext4_extent *extent;
5417 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5418 unsigned int credits, ee_len;
5419 int ret = 0, depth, split_flag = 0;
5423 * We need to test this early because xfstests assumes that an
5424 * insert range of (0, 1) will return EOPNOTSUPP if the file
5425 * system does not support insert range.
5427 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5430 /* Insert range works only on fs cluster size aligned regions. */
5431 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5434 trace_ext4_insert_range(inode, offset, len);
5436 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5437 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5439 /* Call ext4_force_commit to flush all data in case of data=journal */
5440 if (ext4_should_journal_data(inode)) {
5441 ret = ext4_force_commit(inode->i_sb);
5447 /* Currently just for extent based files */
5448 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5453 /* Check whether the maximum file size would be exceeded */
5454 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5459 /* Offset must be less than i_size */
5460 if (offset >= inode->i_size) {
5465 /* Wait for existing dio to complete */
5466 inode_dio_wait(inode);
5468 ret = file_modified(file);
5473 * Prevent page faults from reinstantiating pages we have released from
5476 filemap_invalidate_lock(mapping);
5478 ret = ext4_break_layouts(inode);
5483 * Need to round down to align start offset to page size boundary
5484 * for page size > block size.
5486 ioffset = round_down(offset, PAGE_SIZE);
5487 /* Write out all dirty pages */
5488 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5492 truncate_pagecache(inode, ioffset);
5494 credits = ext4_writepage_trans_blocks(inode);
5495 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5496 if (IS_ERR(handle)) {
5497 ret = PTR_ERR(handle);
5500 ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5502 /* Expand file to avoid data loss if there is error while shifting */
5503 inode->i_size += len;
5504 EXT4_I(inode)->i_disksize += len;
5505 inode->i_mtime = inode->i_ctime = current_time(inode);
5506 ret = ext4_mark_inode_dirty(handle, inode);
5510 down_write(&EXT4_I(inode)->i_data_sem);
5511 ext4_discard_preallocations(inode, 0);
5513 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5515 up_write(&EXT4_I(inode)->i_data_sem);
5519 depth = ext_depth(inode);
5520 extent = path[depth].p_ext;
5522 ee_start_lblk = le32_to_cpu(extent->ee_block);
5523 ee_len = ext4_ext_get_actual_len(extent);
5526 * If offset_lblk is not the starting block of extent, split
5527 * the extent @offset_lblk
5529 if ((offset_lblk > ee_start_lblk) &&
5530 (offset_lblk < (ee_start_lblk + ee_len))) {
5531 if (ext4_ext_is_unwritten(extent))
5532 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5533 EXT4_EXT_MARK_UNWRIT2;
5534 ret = ext4_split_extent_at(handle, inode, &path,
5535 offset_lblk, split_flag,
5537 EXT4_GET_BLOCKS_PRE_IO |
5538 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5541 ext4_ext_drop_refs(path);
5544 up_write(&EXT4_I(inode)->i_data_sem);
5548 ext4_ext_drop_refs(path);
5552 ret = ext4_es_remove_extent(inode, offset_lblk,
5553 EXT_MAX_BLOCKS - offset_lblk);
5555 up_write(&EXT4_I(inode)->i_data_sem);
5560 * if offset_lblk lies in a hole which is at start of file, use
5561 * ee_start_lblk to shift extents
5563 ret = ext4_ext_shift_extents(inode, handle,
5564 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5565 len_lblk, SHIFT_RIGHT);
5567 up_write(&EXT4_I(inode)->i_data_sem);
5569 ext4_handle_sync(handle);
5571 ext4_update_inode_fsync_trans(handle, inode, 1);
5574 ext4_journal_stop(handle);
5576 filemap_invalidate_unlock(mapping);
5578 inode_unlock(inode);
5583 * ext4_swap_extents() - Swap extents between two inodes
5584 * @handle: handle for this transaction
5585 * @inode1: First inode
5586 * @inode2: Second inode
5587 * @lblk1: Start block for first inode
5588 * @lblk2: Start block for second inode
5589 * @count: Number of blocks to swap
5590 * @unwritten: Mark second inode's extents as unwritten after swap
5591 * @erp: Pointer to save error value
5593 * This helper routine does exactly what is promise "swap extents". All other
5594 * stuff such as page-cache locking consistency, bh mapping consistency or
5595 * extent's data copying must be performed by caller.
5597 * i_rwsem is held for both inodes
5598 * i_data_sem is locked for write for both inodes
5600 * All pages from requested range are locked for both inodes
5603 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5604 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5605 ext4_lblk_t count, int unwritten, int *erp)
5607 struct ext4_ext_path *path1 = NULL;
5608 struct ext4_ext_path *path2 = NULL;
5609 int replaced_count = 0;
5611 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5612 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5613 BUG_ON(!inode_is_locked(inode1));
5614 BUG_ON(!inode_is_locked(inode2));
5616 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5619 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5624 struct ext4_extent *ex1, *ex2, tmp_ex;
5625 ext4_lblk_t e1_blk, e2_blk;
5626 int e1_len, e2_len, len;
5629 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5630 if (IS_ERR(path1)) {
5631 *erp = PTR_ERR(path1);
5637 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5638 if (IS_ERR(path2)) {
5639 *erp = PTR_ERR(path2);
5643 ex1 = path1[path1->p_depth].p_ext;
5644 ex2 = path2[path2->p_depth].p_ext;
5645 /* Do we have something to swap ? */
5646 if (unlikely(!ex2 || !ex1))
5649 e1_blk = le32_to_cpu(ex1->ee_block);
5650 e2_blk = le32_to_cpu(ex2->ee_block);
5651 e1_len = ext4_ext_get_actual_len(ex1);
5652 e2_len = ext4_ext_get_actual_len(ex2);
5655 if (!in_range(lblk1, e1_blk, e1_len) ||
5656 !in_range(lblk2, e2_blk, e2_len)) {
5657 ext4_lblk_t next1, next2;
5659 /* if hole after extent, then go to next extent */
5660 next1 = ext4_ext_next_allocated_block(path1);
5661 next2 = ext4_ext_next_allocated_block(path2);
5662 /* If hole before extent, then shift to that extent */
5667 /* Do we have something to swap */
5668 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5670 /* Move to the rightest boundary */
5671 len = next1 - lblk1;
5672 if (len < next2 - lblk2)
5673 len = next2 - lblk2;
5682 /* Prepare left boundary */
5683 if (e1_blk < lblk1) {
5685 *erp = ext4_force_split_extent_at(handle, inode1,
5690 if (e2_blk < lblk2) {
5692 *erp = ext4_force_split_extent_at(handle, inode2,
5697 /* ext4_split_extent_at() may result in leaf extent split,
5698 * path must to be revalidated. */
5702 /* Prepare right boundary */
5704 if (len > e1_blk + e1_len - lblk1)
5705 len = e1_blk + e1_len - lblk1;
5706 if (len > e2_blk + e2_len - lblk2)
5707 len = e2_blk + e2_len - lblk2;
5709 if (len != e1_len) {
5711 *erp = ext4_force_split_extent_at(handle, inode1,
5712 &path1, lblk1 + len, 0);
5716 if (len != e2_len) {
5718 *erp = ext4_force_split_extent_at(handle, inode2,
5719 &path2, lblk2 + len, 0);
5723 /* ext4_split_extent_at() may result in leaf extent split,
5724 * path must to be revalidated. */
5728 BUG_ON(e2_len != e1_len);
5729 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5732 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5736 /* Both extents are fully inside boundaries. Swap it now */
5738 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5739 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5740 ex1->ee_len = cpu_to_le16(e2_len);
5741 ex2->ee_len = cpu_to_le16(e1_len);
5743 ext4_ext_mark_unwritten(ex2);
5744 if (ext4_ext_is_unwritten(&tmp_ex))
5745 ext4_ext_mark_unwritten(ex1);
5747 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5748 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5749 *erp = ext4_ext_dirty(handle, inode2, path2 +
5753 *erp = ext4_ext_dirty(handle, inode1, path1 +
5756 * Looks scarry ah..? second inode already points to new blocks,
5757 * and it was successfully dirtied. But luckily error may happen
5758 * only due to journal error, so full transaction will be
5765 replaced_count += len;
5769 ext4_ext_drop_refs(path1);
5771 ext4_ext_drop_refs(path2);
5773 path1 = path2 = NULL;
5775 return replaced_count;
5779 * ext4_clu_mapped - determine whether any block in a logical cluster has
5780 * been mapped to a physical cluster
5782 * @inode - file containing the logical cluster
5783 * @lclu - logical cluster of interest
5785 * Returns 1 if any block in the logical cluster is mapped, signifying
5786 * that a physical cluster has been allocated for it. Otherwise,
5787 * returns 0. Can also return negative error codes. Derived from
5788 * ext4_ext_map_blocks().
5790 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5792 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5793 struct ext4_ext_path *path;
5794 int depth, mapped = 0, err = 0;
5795 struct ext4_extent *extent;
5796 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5798 /* search for the extent closest to the first block in the cluster */
5799 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5801 err = PTR_ERR(path);
5806 depth = ext_depth(inode);
5809 * A consistent leaf must not be empty. This situation is possible,
5810 * though, _during_ tree modification, and it's why an assert can't
5811 * be put in ext4_find_extent().
5813 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5814 EXT4_ERROR_INODE(inode,
5815 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5816 (unsigned long) EXT4_C2B(sbi, lclu),
5817 depth, path[depth].p_block);
5818 err = -EFSCORRUPTED;
5822 extent = path[depth].p_ext;
5824 /* can't be mapped if the extent tree is empty */
5828 first_lblk = le32_to_cpu(extent->ee_block);
5829 first_lclu = EXT4_B2C(sbi, first_lblk);
5832 * Three possible outcomes at this point - found extent spanning
5833 * the target cluster, to the left of the target cluster, or to the
5834 * right of the target cluster. The first two cases are handled here.
5835 * The last case indicates the target cluster is not mapped.
5837 if (lclu >= first_lclu) {
5838 last_lclu = EXT4_B2C(sbi, first_lblk +
5839 ext4_ext_get_actual_len(extent) - 1);
5840 if (lclu <= last_lclu) {
5843 first_lblk = ext4_ext_next_allocated_block(path);
5844 first_lclu = EXT4_B2C(sbi, first_lblk);
5845 if (lclu == first_lclu)
5851 ext4_ext_drop_refs(path);
5854 return err ? err : mapped;
5858 * Updates physical block address and unwritten status of extent
5859 * starting at lblk start and of len. If such an extent doesn't exist,
5860 * this function splits the extent tree appropriately to create an
5861 * extent like this. This function is called in the fast commit
5862 * replay path. Returns 0 on success and error on failure.
5864 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5865 int len, int unwritten, ext4_fsblk_t pblk)
5867 struct ext4_ext_path *path = NULL, *ppath;
5868 struct ext4_extent *ex;
5871 path = ext4_find_extent(inode, start, NULL, 0);
5873 return PTR_ERR(path);
5874 ex = path[path->p_depth].p_ext;
5876 ret = -EFSCORRUPTED;
5880 if (le32_to_cpu(ex->ee_block) != start ||
5881 ext4_ext_get_actual_len(ex) != len) {
5882 /* We need to split this extent to match our extent first */
5884 down_write(&EXT4_I(inode)->i_data_sem);
5885 ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5886 up_write(&EXT4_I(inode)->i_data_sem);
5890 path = ext4_find_extent(inode, start, NULL, 0);
5894 ex = path[path->p_depth].p_ext;
5895 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5896 if (ext4_ext_get_actual_len(ex) != len) {
5897 down_write(&EXT4_I(inode)->i_data_sem);
5898 ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5900 up_write(&EXT4_I(inode)->i_data_sem);
5904 path = ext4_find_extent(inode, start, NULL, 0);
5907 ex = path[path->p_depth].p_ext;
5911 ext4_ext_mark_unwritten(ex);
5913 ext4_ext_mark_initialized(ex);
5914 ext4_ext_store_pblock(ex, pblk);
5915 down_write(&EXT4_I(inode)->i_data_sem);
5916 ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5917 up_write(&EXT4_I(inode)->i_data_sem);
5919 ext4_ext_drop_refs(path);
5921 ext4_mark_inode_dirty(NULL, inode);
5925 /* Try to shrink the extent tree */
5926 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5928 struct ext4_ext_path *path = NULL;
5929 struct ext4_extent *ex;
5930 ext4_lblk_t old_cur, cur = 0;
5933 path = ext4_find_extent(inode, cur, NULL, 0);
5936 ex = path[path->p_depth].p_ext;
5938 ext4_ext_drop_refs(path);
5940 ext4_mark_inode_dirty(NULL, inode);
5944 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5947 ext4_ext_try_to_merge(NULL, inode, path, ex);
5948 down_write(&EXT4_I(inode)->i_data_sem);
5949 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5950 up_write(&EXT4_I(inode)->i_data_sem);
5951 ext4_mark_inode_dirty(NULL, inode);
5952 ext4_ext_drop_refs(path);
5957 /* Check if *cur is a hole and if it is, skip it */
5958 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5961 struct ext4_map_blocks map;
5964 map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5966 ret = ext4_map_blocks(NULL, inode, &map, 0);
5971 *cur = *cur + map.m_len;
5975 /* Count number of blocks used by this inode and update i_blocks */
5976 int ext4_ext_replay_set_iblocks(struct inode *inode)
5978 struct ext4_ext_path *path = NULL, *path2 = NULL;
5979 struct ext4_extent *ex;
5980 ext4_lblk_t cur = 0, end;
5981 int numblks = 0, i, ret = 0;
5982 ext4_fsblk_t cmp1, cmp2;
5983 struct ext4_map_blocks map;
5985 /* Determin the size of the file first */
5986 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5989 return PTR_ERR(path);
5990 ex = path[path->p_depth].p_ext;
5992 ext4_ext_drop_refs(path);
5996 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5997 ext4_ext_drop_refs(path);
6000 /* Count the number of data blocks */
6004 map.m_len = end - cur;
6005 ret = ext4_map_blocks(NULL, inode, &map, 0);
6010 cur = cur + map.m_len;
6014 * Count the number of extent tree blocks. We do it by looking up
6015 * two successive extents and determining the difference between
6016 * their paths. When path is different for 2 successive extents
6017 * we compare the blocks in the path at each level and increment
6018 * iblocks by total number of differences found.
6021 ret = skip_hole(inode, &cur);
6024 path = ext4_find_extent(inode, cur, NULL, 0);
6027 numblks += path->p_depth;
6028 ext4_ext_drop_refs(path);
6031 path = ext4_find_extent(inode, cur, NULL, 0);
6034 ex = path[path->p_depth].p_ext;
6036 ext4_ext_drop_refs(path);
6040 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6041 ext4_ext_get_actual_len(ex));
6042 ret = skip_hole(inode, &cur);
6044 ext4_ext_drop_refs(path);
6048 path2 = ext4_find_extent(inode, cur, NULL, 0);
6049 if (IS_ERR(path2)) {
6050 ext4_ext_drop_refs(path);
6054 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6056 if (i <= path->p_depth)
6057 cmp1 = path[i].p_bh ?
6058 path[i].p_bh->b_blocknr : 0;
6059 if (i <= path2->p_depth)
6060 cmp2 = path2[i].p_bh ?
6061 path2[i].p_bh->b_blocknr : 0;
6062 if (cmp1 != cmp2 && cmp2 != 0)
6065 ext4_ext_drop_refs(path);
6066 ext4_ext_drop_refs(path2);
6072 inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6073 ext4_mark_inode_dirty(NULL, inode);
6077 int ext4_ext_clear_bb(struct inode *inode)
6079 struct ext4_ext_path *path = NULL;
6080 struct ext4_extent *ex;
6081 ext4_lblk_t cur = 0, end;
6083 struct ext4_map_blocks map;
6085 if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6088 /* Determin the size of the file first */
6089 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6092 return PTR_ERR(path);
6093 ex = path[path->p_depth].p_ext;
6095 ext4_ext_drop_refs(path);
6099 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6100 ext4_ext_drop_refs(path);
6106 map.m_len = end - cur;
6107 ret = ext4_map_blocks(NULL, inode, &map, 0);
6111 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6112 if (!IS_ERR_OR_NULL(path)) {
6113 for (j = 0; j < path->p_depth; j++) {
6115 ext4_mb_mark_bb(inode->i_sb,
6116 path[j].p_block, 1, 0);
6117 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6118 0, path[j].p_block, 1, 1);
6120 ext4_ext_drop_refs(path);
6123 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6124 ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6125 map.m_lblk, map.m_pblk, map.m_len, 1);
6127 cur = cur + map.m_len;