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/backing-dev.h>
31 #include <linux/iomap.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_mutex. 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)
140 /* path points to block */
141 BUFFER_TRACE(path->p_bh, "get_write_access");
142 return ext4_journal_get_write_access(handle, path->p_bh);
144 /* path points to leaf/index in inode body */
145 /* we use in-core data, no need to protect them */
155 static int __ext4_ext_dirty(const char *where, unsigned int line,
156 handle_t *handle, struct inode *inode,
157 struct ext4_ext_path *path)
161 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
163 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
164 /* path points to block */
165 err = __ext4_handle_dirty_metadata(where, line, handle,
168 /* path points to leaf/index in inode body */
169 err = ext4_mark_inode_dirty(handle, inode);
174 #define ext4_ext_dirty(handle, inode, path) \
175 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
177 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
178 struct ext4_ext_path *path,
182 int depth = path->p_depth;
183 struct ext4_extent *ex;
186 * Try to predict block placement assuming that we are
187 * filling in a file which will eventually be
188 * non-sparse --- i.e., in the case of libbfd writing
189 * an ELF object sections out-of-order but in a way
190 * the eventually results in a contiguous object or
191 * executable file, or some database extending a table
192 * space file. However, this is actually somewhat
193 * non-ideal if we are writing a sparse file such as
194 * qemu or KVM writing a raw image file that is going
195 * to stay fairly sparse, since it will end up
196 * fragmenting the file system's free space. Maybe we
197 * should have some hueristics or some way to allow
198 * userspace to pass a hint to file system,
199 * especially if the latter case turns out to be
202 ex = path[depth].p_ext;
204 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
205 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
207 if (block > ext_block)
208 return ext_pblk + (block - ext_block);
210 return ext_pblk - (ext_block - block);
213 /* it looks like index is empty;
214 * try to find starting block from index itself */
215 if (path[depth].p_bh)
216 return path[depth].p_bh->b_blocknr;
219 /* OK. use inode's group */
220 return ext4_inode_to_goal_block(inode);
224 * Allocation for a meta data block
227 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
228 struct ext4_ext_path *path,
229 struct ext4_extent *ex, int *err, unsigned int flags)
231 ext4_fsblk_t goal, newblock;
233 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
234 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
239 static inline int ext4_ext_space_block(struct inode *inode, int check)
243 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
244 / sizeof(struct ext4_extent);
245 #ifdef AGGRESSIVE_TEST
246 if (!check && size > 6)
252 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
256 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
257 / sizeof(struct ext4_extent_idx);
258 #ifdef AGGRESSIVE_TEST
259 if (!check && size > 5)
265 static inline int ext4_ext_space_root(struct inode *inode, int check)
269 size = sizeof(EXT4_I(inode)->i_data);
270 size -= sizeof(struct ext4_extent_header);
271 size /= sizeof(struct ext4_extent);
272 #ifdef AGGRESSIVE_TEST
273 if (!check && size > 3)
279 static inline int ext4_ext_space_root_idx(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_idx);
286 #ifdef AGGRESSIVE_TEST
287 if (!check && size > 4)
294 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
295 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
298 struct ext4_ext_path *path = *ppath;
299 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
300 int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
303 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
305 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
306 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
311 ext4_ext_max_entries(struct inode *inode, int depth)
315 if (depth == ext_depth(inode)) {
317 max = ext4_ext_space_root(inode, 1);
319 max = ext4_ext_space_root_idx(inode, 1);
322 max = ext4_ext_space_block(inode, 1);
324 max = ext4_ext_space_block_idx(inode, 1);
330 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
332 ext4_fsblk_t block = ext4_ext_pblock(ext);
333 int len = ext4_ext_get_actual_len(ext);
334 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
339 * - overflow/wrap-around
341 if (lblock + len <= lblock)
343 return ext4_inode_block_valid(inode, block, len);
346 static int ext4_valid_extent_idx(struct inode *inode,
347 struct ext4_extent_idx *ext_idx)
349 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
351 return ext4_inode_block_valid(inode, block, 1);
354 static int ext4_valid_extent_entries(struct inode *inode,
355 struct ext4_extent_header *eh,
356 ext4_fsblk_t *pblk, int depth)
358 unsigned short entries;
359 if (eh->eh_entries == 0)
362 entries = le16_to_cpu(eh->eh_entries);
366 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
367 ext4_lblk_t lblock = 0;
368 ext4_lblk_t prev = 0;
371 if (!ext4_valid_extent(inode, ext))
374 /* Check for overlapping extents */
375 lblock = le32_to_cpu(ext->ee_block);
376 len = ext4_ext_get_actual_len(ext);
377 if ((lblock <= prev) && prev) {
378 *pblk = ext4_ext_pblock(ext);
383 prev = lblock + len - 1;
386 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
388 if (!ext4_valid_extent_idx(inode, ext_idx))
397 static int __ext4_ext_check(const char *function, unsigned int line,
398 struct inode *inode, struct ext4_extent_header *eh,
399 int depth, ext4_fsblk_t pblk)
401 const char *error_msg;
402 int max = 0, err = -EFSCORRUPTED;
404 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
405 error_msg = "invalid magic";
408 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
409 error_msg = "unexpected eh_depth";
412 if (unlikely(eh->eh_max == 0)) {
413 error_msg = "invalid eh_max";
416 max = ext4_ext_max_entries(inode, depth);
417 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
418 error_msg = "too large eh_max";
421 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
422 error_msg = "invalid eh_entries";
425 if (!ext4_valid_extent_entries(inode, eh, &pblk, depth)) {
426 error_msg = "invalid extent entries";
429 if (unlikely(depth > 32)) {
430 error_msg = "too large eh_depth";
433 /* Verify checksum on non-root extent tree nodes */
434 if (ext_depth(inode) != depth &&
435 !ext4_extent_block_csum_verify(inode, eh)) {
436 error_msg = "extent tree corrupted";
443 ext4_error_inode_err(inode, function, line, 0, -err,
444 "pblk %llu bad header/extent: %s - magic %x, "
445 "entries %u, max %u(%u), depth %u(%u)",
446 (unsigned long long) pblk, error_msg,
447 le16_to_cpu(eh->eh_magic),
448 le16_to_cpu(eh->eh_entries),
449 le16_to_cpu(eh->eh_max),
450 max, le16_to_cpu(eh->eh_depth), depth);
454 #define ext4_ext_check(inode, eh, depth, pblk) \
455 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
457 int ext4_ext_check_inode(struct inode *inode)
459 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
462 static void ext4_cache_extents(struct inode *inode,
463 struct ext4_extent_header *eh)
465 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
466 ext4_lblk_t prev = 0;
469 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
470 unsigned int status = EXTENT_STATUS_WRITTEN;
471 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
472 int len = ext4_ext_get_actual_len(ex);
474 if (prev && (prev != lblk))
475 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
478 if (ext4_ext_is_unwritten(ex))
479 status = EXTENT_STATUS_UNWRITTEN;
480 ext4_es_cache_extent(inode, lblk, len,
481 ext4_ext_pblock(ex), status);
486 static struct buffer_head *
487 __read_extent_tree_block(const char *function, unsigned int line,
488 struct inode *inode, ext4_fsblk_t pblk, int depth,
491 struct buffer_head *bh;
493 gfp_t gfp_flags = __GFP_MOVABLE | GFP_NOFS;
495 if (flags & EXT4_EX_NOFAIL)
496 gfp_flags |= __GFP_NOFAIL;
498 bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
500 return ERR_PTR(-ENOMEM);
502 if (!bh_uptodate_or_lock(bh)) {
503 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
504 err = ext4_read_bh(bh, 0, NULL);
508 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
510 err = __ext4_ext_check(function, line, inode,
511 ext_block_hdr(bh), depth, pblk);
514 set_buffer_verified(bh);
516 * If this is a leaf block, cache all of its entries
518 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
519 struct ext4_extent_header *eh = ext_block_hdr(bh);
520 ext4_cache_extents(inode, eh);
529 #define read_extent_tree_block(inode, pblk, depth, flags) \
530 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
534 * This function is called to cache a file's extent information in the
537 int ext4_ext_precache(struct inode *inode)
539 struct ext4_inode_info *ei = EXT4_I(inode);
540 struct ext4_ext_path *path = NULL;
541 struct buffer_head *bh;
542 int i = 0, depth, ret = 0;
544 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
545 return 0; /* not an extent-mapped inode */
547 down_read(&ei->i_data_sem);
548 depth = ext_depth(inode);
550 /* Don't cache anything if there are no external extent blocks */
552 up_read(&ei->i_data_sem);
556 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
559 up_read(&ei->i_data_sem);
563 path[0].p_hdr = ext_inode_hdr(inode);
564 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
567 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
570 * If this is a leaf block or we've reached the end of
571 * the index block, go up
574 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
575 brelse(path[i].p_bh);
580 bh = read_extent_tree_block(inode,
581 ext4_idx_pblock(path[i].p_idx++),
583 EXT4_EX_FORCE_CACHE);
590 path[i].p_hdr = ext_block_hdr(bh);
591 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
593 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
595 up_read(&ei->i_data_sem);
596 ext4_ext_drop_refs(path);
602 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
604 int k, l = path->p_depth;
606 ext_debug(inode, "path:");
607 for (k = 0; k <= l; k++, path++) {
609 ext_debug(inode, " %d->%llu",
610 le32_to_cpu(path->p_idx->ei_block),
611 ext4_idx_pblock(path->p_idx));
612 } else if (path->p_ext) {
613 ext_debug(inode, " %d:[%d]%d:%llu ",
614 le32_to_cpu(path->p_ext->ee_block),
615 ext4_ext_is_unwritten(path->p_ext),
616 ext4_ext_get_actual_len(path->p_ext),
617 ext4_ext_pblock(path->p_ext));
619 ext_debug(inode, " []");
621 ext_debug(inode, "\n");
624 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
626 int depth = ext_depth(inode);
627 struct ext4_extent_header *eh;
628 struct ext4_extent *ex;
634 eh = path[depth].p_hdr;
635 ex = EXT_FIRST_EXTENT(eh);
637 ext_debug(inode, "Displaying leaf extents\n");
639 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
640 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
641 ext4_ext_is_unwritten(ex),
642 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
644 ext_debug(inode, "\n");
647 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
648 ext4_fsblk_t newblock, int level)
650 int depth = ext_depth(inode);
651 struct ext4_extent *ex;
653 if (depth != level) {
654 struct ext4_extent_idx *idx;
655 idx = path[level].p_idx;
656 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
657 ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
658 level, le32_to_cpu(idx->ei_block),
659 ext4_idx_pblock(idx), newblock);
666 ex = path[depth].p_ext;
667 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
668 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
669 le32_to_cpu(ex->ee_block),
671 ext4_ext_is_unwritten(ex),
672 ext4_ext_get_actual_len(ex),
679 #define ext4_ext_show_path(inode, path)
680 #define ext4_ext_show_leaf(inode, path)
681 #define ext4_ext_show_move(inode, path, newblock, level)
684 void ext4_ext_drop_refs(struct ext4_ext_path *path)
690 depth = path->p_depth;
691 for (i = 0; i <= depth; i++, path++) {
698 * ext4_ext_binsearch_idx:
699 * binary search for the closest index of the given block
700 * the header must be checked before calling this
703 ext4_ext_binsearch_idx(struct inode *inode,
704 struct ext4_ext_path *path, ext4_lblk_t block)
706 struct ext4_extent_header *eh = path->p_hdr;
707 struct ext4_extent_idx *r, *l, *m;
710 ext_debug(inode, "binsearch for %u(idx): ", block);
712 l = EXT_FIRST_INDEX(eh) + 1;
713 r = EXT_LAST_INDEX(eh);
716 if (block < le32_to_cpu(m->ei_block))
720 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
721 le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
722 r, le32_to_cpu(r->ei_block));
726 ext_debug(inode, " -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
727 ext4_idx_pblock(path->p_idx));
729 #ifdef CHECK_BINSEARCH
731 struct ext4_extent_idx *chix, *ix;
734 chix = ix = EXT_FIRST_INDEX(eh);
735 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
736 if (k != 0 && le32_to_cpu(ix->ei_block) <=
737 le32_to_cpu(ix[-1].ei_block)) {
738 printk(KERN_DEBUG "k=%d, ix=0x%p, "
740 ix, EXT_FIRST_INDEX(eh));
741 printk(KERN_DEBUG "%u <= %u\n",
742 le32_to_cpu(ix->ei_block),
743 le32_to_cpu(ix[-1].ei_block));
745 BUG_ON(k && le32_to_cpu(ix->ei_block)
746 <= le32_to_cpu(ix[-1].ei_block));
747 if (block < le32_to_cpu(ix->ei_block))
751 BUG_ON(chix != path->p_idx);
758 * ext4_ext_binsearch:
759 * binary search for closest extent of the given block
760 * the header must be checked before calling this
763 ext4_ext_binsearch(struct inode *inode,
764 struct ext4_ext_path *path, ext4_lblk_t block)
766 struct ext4_extent_header *eh = path->p_hdr;
767 struct ext4_extent *r, *l, *m;
769 if (eh->eh_entries == 0) {
771 * this leaf is empty:
772 * we get such a leaf in split/add case
777 ext_debug(inode, "binsearch for %u: ", block);
779 l = EXT_FIRST_EXTENT(eh) + 1;
780 r = EXT_LAST_EXTENT(eh);
784 if (block < le32_to_cpu(m->ee_block))
788 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
789 le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
790 r, le32_to_cpu(r->ee_block));
794 ext_debug(inode, " -> %d:%llu:[%d]%d ",
795 le32_to_cpu(path->p_ext->ee_block),
796 ext4_ext_pblock(path->p_ext),
797 ext4_ext_is_unwritten(path->p_ext),
798 ext4_ext_get_actual_len(path->p_ext));
800 #ifdef CHECK_BINSEARCH
802 struct ext4_extent *chex, *ex;
805 chex = ex = EXT_FIRST_EXTENT(eh);
806 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
807 BUG_ON(k && le32_to_cpu(ex->ee_block)
808 <= le32_to_cpu(ex[-1].ee_block));
809 if (block < le32_to_cpu(ex->ee_block))
813 BUG_ON(chex != path->p_ext);
819 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
821 struct ext4_extent_header *eh;
823 eh = ext_inode_hdr(inode);
826 eh->eh_magic = EXT4_EXT_MAGIC;
827 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
828 ext4_mark_inode_dirty(handle, inode);
831 struct ext4_ext_path *
832 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
833 struct ext4_ext_path **orig_path, int flags)
835 struct ext4_extent_header *eh;
836 struct buffer_head *bh;
837 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
838 short int depth, i, ppos = 0;
840 gfp_t gfp_flags = GFP_NOFS;
842 if (flags & EXT4_EX_NOFAIL)
843 gfp_flags |= __GFP_NOFAIL;
845 eh = ext_inode_hdr(inode);
846 depth = ext_depth(inode);
847 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
848 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
855 ext4_ext_drop_refs(path);
856 if (depth > path[0].p_maxdepth) {
858 *orig_path = path = NULL;
862 /* account possible depth increase */
863 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
866 return ERR_PTR(-ENOMEM);
867 path[0].p_maxdepth = depth + 1;
873 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
874 ext4_cache_extents(inode, eh);
875 /* walk through the tree */
877 ext_debug(inode, "depth %d: num %d, max %d\n",
878 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
880 ext4_ext_binsearch_idx(inode, path + ppos, block);
881 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
882 path[ppos].p_depth = i;
883 path[ppos].p_ext = NULL;
885 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
892 eh = ext_block_hdr(bh);
894 path[ppos].p_bh = bh;
895 path[ppos].p_hdr = eh;
898 path[ppos].p_depth = i;
899 path[ppos].p_ext = NULL;
900 path[ppos].p_idx = NULL;
903 ext4_ext_binsearch(inode, path + ppos, block);
904 /* if not an empty leaf */
905 if (path[ppos].p_ext)
906 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
908 ext4_ext_show_path(inode, path);
913 ext4_ext_drop_refs(path);
921 * ext4_ext_insert_index:
922 * insert new index [@logical;@ptr] into the block at @curp;
923 * check where to insert: before @curp or after @curp
925 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
926 struct ext4_ext_path *curp,
927 int logical, ext4_fsblk_t ptr)
929 struct ext4_extent_idx *ix;
932 err = ext4_ext_get_access(handle, inode, curp);
936 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
937 EXT4_ERROR_INODE(inode,
938 "logical %d == ei_block %d!",
939 logical, le32_to_cpu(curp->p_idx->ei_block));
940 return -EFSCORRUPTED;
943 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
944 >= le16_to_cpu(curp->p_hdr->eh_max))) {
945 EXT4_ERROR_INODE(inode,
946 "eh_entries %d >= eh_max %d!",
947 le16_to_cpu(curp->p_hdr->eh_entries),
948 le16_to_cpu(curp->p_hdr->eh_max));
949 return -EFSCORRUPTED;
952 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
954 ext_debug(inode, "insert new index %d after: %llu\n",
956 ix = curp->p_idx + 1;
959 ext_debug(inode, "insert new index %d before: %llu\n",
964 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
967 ext_debug(inode, "insert new index %d: "
968 "move %d indices from 0x%p to 0x%p\n",
969 logical, len, ix, ix + 1);
970 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
973 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
974 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
975 return -EFSCORRUPTED;
978 ix->ei_block = cpu_to_le32(logical);
979 ext4_idx_store_pblock(ix, ptr);
980 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
982 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
983 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
984 return -EFSCORRUPTED;
987 err = ext4_ext_dirty(handle, inode, curp);
988 ext4_std_error(inode->i_sb, err);
995 * inserts new subtree into the path, using free index entry
997 * - allocates all needed blocks (new leaf and all intermediate index blocks)
998 * - makes decision where to split
999 * - moves remaining extents and index entries (right to the split point)
1000 * into the newly allocated blocks
1001 * - initializes subtree
1003 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1005 struct ext4_ext_path *path,
1006 struct ext4_extent *newext, int at)
1008 struct buffer_head *bh = NULL;
1009 int depth = ext_depth(inode);
1010 struct ext4_extent_header *neh;
1011 struct ext4_extent_idx *fidx;
1012 int i = at, k, m, a;
1013 ext4_fsblk_t newblock, oldblock;
1015 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1016 gfp_t gfp_flags = GFP_NOFS;
1018 size_t ext_size = 0;
1020 if (flags & EXT4_EX_NOFAIL)
1021 gfp_flags |= __GFP_NOFAIL;
1023 /* make decision: where to split? */
1024 /* FIXME: now decision is simplest: at current extent */
1026 /* if current leaf will be split, then we should use
1027 * border from split point */
1028 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1029 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1030 return -EFSCORRUPTED;
1032 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1033 border = path[depth].p_ext[1].ee_block;
1034 ext_debug(inode, "leaf will be split."
1035 " next leaf starts at %d\n",
1036 le32_to_cpu(border));
1038 border = newext->ee_block;
1039 ext_debug(inode, "leaf will be added."
1040 " next leaf starts at %d\n",
1041 le32_to_cpu(border));
1045 * If error occurs, then we break processing
1046 * and mark filesystem read-only. index won't
1047 * be inserted and tree will be in consistent
1048 * state. Next mount will repair buffers too.
1052 * Get array to track all allocated blocks.
1053 * We need this to handle errors and free blocks
1056 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1060 /* allocate all needed blocks */
1061 ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1062 for (a = 0; a < depth - at; a++) {
1063 newblock = ext4_ext_new_meta_block(handle, inode, path,
1064 newext, &err, flags);
1067 ablocks[a] = newblock;
1070 /* initialize new leaf */
1071 newblock = ablocks[--a];
1072 if (unlikely(newblock == 0)) {
1073 EXT4_ERROR_INODE(inode, "newblock == 0!");
1074 err = -EFSCORRUPTED;
1077 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1078 if (unlikely(!bh)) {
1084 err = ext4_journal_get_create_access(handle, bh);
1088 neh = ext_block_hdr(bh);
1089 neh->eh_entries = 0;
1090 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1091 neh->eh_magic = EXT4_EXT_MAGIC;
1094 /* move remainder of path[depth] to the new leaf */
1095 if (unlikely(path[depth].p_hdr->eh_entries !=
1096 path[depth].p_hdr->eh_max)) {
1097 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1098 path[depth].p_hdr->eh_entries,
1099 path[depth].p_hdr->eh_max);
1100 err = -EFSCORRUPTED;
1103 /* start copy from next extent */
1104 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1105 ext4_ext_show_move(inode, path, newblock, depth);
1107 struct ext4_extent *ex;
1108 ex = EXT_FIRST_EXTENT(neh);
1109 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1110 le16_add_cpu(&neh->eh_entries, m);
1113 /* zero out unused area in the extent block */
1114 ext_size = sizeof(struct ext4_extent_header) +
1115 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1116 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1117 ext4_extent_block_csum_set(inode, neh);
1118 set_buffer_uptodate(bh);
1121 err = ext4_handle_dirty_metadata(handle, inode, bh);
1127 /* correct old leaf */
1129 err = ext4_ext_get_access(handle, inode, path + depth);
1132 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1133 err = ext4_ext_dirty(handle, inode, path + depth);
1139 /* create intermediate indexes */
1141 if (unlikely(k < 0)) {
1142 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1143 err = -EFSCORRUPTED;
1147 ext_debug(inode, "create %d intermediate indices\n", k);
1148 /* insert new index into current index block */
1149 /* current depth stored in i var */
1152 oldblock = newblock;
1153 newblock = ablocks[--a];
1154 bh = sb_getblk(inode->i_sb, newblock);
1155 if (unlikely(!bh)) {
1161 err = ext4_journal_get_create_access(handle, bh);
1165 neh = ext_block_hdr(bh);
1166 neh->eh_entries = cpu_to_le16(1);
1167 neh->eh_magic = EXT4_EXT_MAGIC;
1168 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1169 neh->eh_depth = cpu_to_le16(depth - i);
1170 fidx = EXT_FIRST_INDEX(neh);
1171 fidx->ei_block = border;
1172 ext4_idx_store_pblock(fidx, oldblock);
1174 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1175 i, newblock, le32_to_cpu(border), oldblock);
1177 /* move remainder of path[i] to the new index block */
1178 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1179 EXT_LAST_INDEX(path[i].p_hdr))) {
1180 EXT4_ERROR_INODE(inode,
1181 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1182 le32_to_cpu(path[i].p_ext->ee_block));
1183 err = -EFSCORRUPTED;
1186 /* start copy indexes */
1187 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1188 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1189 EXT_MAX_INDEX(path[i].p_hdr));
1190 ext4_ext_show_move(inode, path, newblock, i);
1192 memmove(++fidx, path[i].p_idx,
1193 sizeof(struct ext4_extent_idx) * m);
1194 le16_add_cpu(&neh->eh_entries, m);
1196 /* zero out unused area in the extent block */
1197 ext_size = sizeof(struct ext4_extent_header) +
1198 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1199 memset(bh->b_data + ext_size, 0,
1200 inode->i_sb->s_blocksize - ext_size);
1201 ext4_extent_block_csum_set(inode, neh);
1202 set_buffer_uptodate(bh);
1205 err = ext4_handle_dirty_metadata(handle, inode, bh);
1211 /* correct old index */
1213 err = ext4_ext_get_access(handle, inode, path + i);
1216 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1217 err = ext4_ext_dirty(handle, inode, path + i);
1225 /* insert new index */
1226 err = ext4_ext_insert_index(handle, inode, path + at,
1227 le32_to_cpu(border), newblock);
1231 if (buffer_locked(bh))
1237 /* free all allocated blocks in error case */
1238 for (i = 0; i < depth; i++) {
1241 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1242 EXT4_FREE_BLOCKS_METADATA);
1251 * ext4_ext_grow_indepth:
1252 * implements tree growing procedure:
1253 * - allocates new block
1254 * - moves top-level data (index block or leaf) into the new block
1255 * - initializes new top-level, creating index that points to the
1256 * just created block
1258 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1261 struct ext4_extent_header *neh;
1262 struct buffer_head *bh;
1263 ext4_fsblk_t newblock, goal = 0;
1264 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1266 size_t ext_size = 0;
1268 /* Try to prepend new index to old one */
1269 if (ext_depth(inode))
1270 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1271 if (goal > le32_to_cpu(es->s_first_data_block)) {
1272 flags |= EXT4_MB_HINT_TRY_GOAL;
1275 goal = ext4_inode_to_goal_block(inode);
1276 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1281 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1286 err = ext4_journal_get_create_access(handle, bh);
1292 ext_size = sizeof(EXT4_I(inode)->i_data);
1293 /* move top-level index/leaf into new block */
1294 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1295 /* zero out unused area in the extent block */
1296 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1298 /* set size of new block */
1299 neh = ext_block_hdr(bh);
1300 /* old root could have indexes or leaves
1301 * so calculate e_max right way */
1302 if (ext_depth(inode))
1303 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1305 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1306 neh->eh_magic = EXT4_EXT_MAGIC;
1307 ext4_extent_block_csum_set(inode, neh);
1308 set_buffer_uptodate(bh);
1311 err = ext4_handle_dirty_metadata(handle, inode, bh);
1315 /* Update top-level index: num,max,pointer */
1316 neh = ext_inode_hdr(inode);
1317 neh->eh_entries = cpu_to_le16(1);
1318 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1319 if (neh->eh_depth == 0) {
1320 /* Root extent block becomes index block */
1321 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1322 EXT_FIRST_INDEX(neh)->ei_block =
1323 EXT_FIRST_EXTENT(neh)->ee_block;
1325 ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1326 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1327 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1328 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1330 le16_add_cpu(&neh->eh_depth, 1);
1331 err = ext4_mark_inode_dirty(handle, inode);
1339 * ext4_ext_create_new_leaf:
1340 * finds empty index and adds new leaf.
1341 * if no free index is found, then it requests in-depth growing.
1343 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1344 unsigned int mb_flags,
1345 unsigned int gb_flags,
1346 struct ext4_ext_path **ppath,
1347 struct ext4_extent *newext)
1349 struct ext4_ext_path *path = *ppath;
1350 struct ext4_ext_path *curp;
1351 int depth, i, err = 0;
1354 i = depth = ext_depth(inode);
1356 /* walk up to the tree and look for free index entry */
1357 curp = path + depth;
1358 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1363 /* we use already allocated block for index block,
1364 * so subsequent data blocks should be contiguous */
1365 if (EXT_HAS_FREE_INDEX(curp)) {
1366 /* if we found index with free entry, then use that
1367 * entry: create all needed subtree and add new leaf */
1368 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1373 path = ext4_find_extent(inode,
1374 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1377 err = PTR_ERR(path);
1379 /* tree is full, time to grow in depth */
1380 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1385 path = ext4_find_extent(inode,
1386 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1389 err = PTR_ERR(path);
1394 * only first (depth 0 -> 1) produces free space;
1395 * in all other cases we have to split the grown tree
1397 depth = ext_depth(inode);
1398 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1399 /* now we need to split */
1409 * search the closest allocated block to the left for *logical
1410 * and returns it at @logical + it's physical address at @phys
1411 * if *logical is the smallest allocated block, the function
1412 * returns 0 at @phys
1413 * return value contains 0 (success) or error code
1415 static int ext4_ext_search_left(struct inode *inode,
1416 struct ext4_ext_path *path,
1417 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1419 struct ext4_extent_idx *ix;
1420 struct ext4_extent *ex;
1423 if (unlikely(path == NULL)) {
1424 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1425 return -EFSCORRUPTED;
1427 depth = path->p_depth;
1430 if (depth == 0 && path->p_ext == NULL)
1433 /* usually extent in the path covers blocks smaller
1434 * then *logical, but it can be that extent is the
1435 * first one in the file */
1437 ex = path[depth].p_ext;
1438 ee_len = ext4_ext_get_actual_len(ex);
1439 if (*logical < le32_to_cpu(ex->ee_block)) {
1440 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1441 EXT4_ERROR_INODE(inode,
1442 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1443 *logical, le32_to_cpu(ex->ee_block));
1444 return -EFSCORRUPTED;
1446 while (--depth >= 0) {
1447 ix = path[depth].p_idx;
1448 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1449 EXT4_ERROR_INODE(inode,
1450 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1451 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1452 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1453 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1455 return -EFSCORRUPTED;
1461 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1462 EXT4_ERROR_INODE(inode,
1463 "logical %d < ee_block %d + ee_len %d!",
1464 *logical, le32_to_cpu(ex->ee_block), ee_len);
1465 return -EFSCORRUPTED;
1468 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1469 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1474 * Search the closest allocated block to the right for *logical
1475 * and returns it at @logical + it's physical address at @phys.
1476 * If not exists, return 0 and @phys is set to 0. We will return
1477 * 1 which means we found an allocated block and ret_ex is valid.
1478 * Or return a (< 0) error code.
1480 static int ext4_ext_search_right(struct inode *inode,
1481 struct ext4_ext_path *path,
1482 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1483 struct ext4_extent *ret_ex)
1485 struct buffer_head *bh = NULL;
1486 struct ext4_extent_header *eh;
1487 struct ext4_extent_idx *ix;
1488 struct ext4_extent *ex;
1490 int depth; /* Note, NOT eh_depth; depth from top of tree */
1493 if (unlikely(path == NULL)) {
1494 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1495 return -EFSCORRUPTED;
1497 depth = path->p_depth;
1500 if (depth == 0 && path->p_ext == NULL)
1503 /* usually extent in the path covers blocks smaller
1504 * then *logical, but it can be that extent is the
1505 * first one in the file */
1507 ex = path[depth].p_ext;
1508 ee_len = ext4_ext_get_actual_len(ex);
1509 if (*logical < le32_to_cpu(ex->ee_block)) {
1510 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1511 EXT4_ERROR_INODE(inode,
1512 "first_extent(path[%d].p_hdr) != ex",
1514 return -EFSCORRUPTED;
1516 while (--depth >= 0) {
1517 ix = path[depth].p_idx;
1518 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1519 EXT4_ERROR_INODE(inode,
1520 "ix != EXT_FIRST_INDEX *logical %d!",
1522 return -EFSCORRUPTED;
1528 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1529 EXT4_ERROR_INODE(inode,
1530 "logical %d < ee_block %d + ee_len %d!",
1531 *logical, le32_to_cpu(ex->ee_block), ee_len);
1532 return -EFSCORRUPTED;
1535 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1536 /* next allocated block in this leaf */
1541 /* go up and search for index to the right */
1542 while (--depth >= 0) {
1543 ix = path[depth].p_idx;
1544 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1548 /* we've gone up to the root and found no index to the right */
1552 /* we've found index to the right, let's
1553 * follow it and find the closest allocated
1554 * block to the right */
1556 block = ext4_idx_pblock(ix);
1557 while (++depth < path->p_depth) {
1558 /* subtract from p_depth to get proper eh_depth */
1559 bh = read_extent_tree_block(inode, block,
1560 path->p_depth - depth, 0);
1563 eh = ext_block_hdr(bh);
1564 ix = EXT_FIRST_INDEX(eh);
1565 block = ext4_idx_pblock(ix);
1569 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1572 eh = ext_block_hdr(bh);
1573 ex = EXT_FIRST_EXTENT(eh);
1575 *logical = le32_to_cpu(ex->ee_block);
1576 *phys = ext4_ext_pblock(ex);
1585 * ext4_ext_next_allocated_block:
1586 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1587 * NOTE: it considers block number from index entry as
1588 * allocated block. Thus, index entries have to be consistent
1592 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1596 BUG_ON(path == NULL);
1597 depth = path->p_depth;
1599 if (depth == 0 && path->p_ext == NULL)
1600 return EXT_MAX_BLOCKS;
1602 while (depth >= 0) {
1603 struct ext4_ext_path *p = &path[depth];
1605 if (depth == path->p_depth) {
1607 if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1608 return le32_to_cpu(p->p_ext[1].ee_block);
1611 if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1612 return le32_to_cpu(p->p_idx[1].ei_block);
1617 return EXT_MAX_BLOCKS;
1621 * ext4_ext_next_leaf_block:
1622 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1624 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1628 BUG_ON(path == NULL);
1629 depth = path->p_depth;
1631 /* zero-tree has no leaf blocks at all */
1633 return EXT_MAX_BLOCKS;
1635 /* go to index block */
1638 while (depth >= 0) {
1639 if (path[depth].p_idx !=
1640 EXT_LAST_INDEX(path[depth].p_hdr))
1641 return (ext4_lblk_t)
1642 le32_to_cpu(path[depth].p_idx[1].ei_block);
1646 return EXT_MAX_BLOCKS;
1650 * ext4_ext_correct_indexes:
1651 * if leaf gets modified and modified extent is first in the leaf,
1652 * then we have to correct all indexes above.
1653 * TODO: do we need to correct tree in all cases?
1655 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1656 struct ext4_ext_path *path)
1658 struct ext4_extent_header *eh;
1659 int depth = ext_depth(inode);
1660 struct ext4_extent *ex;
1664 eh = path[depth].p_hdr;
1665 ex = path[depth].p_ext;
1667 if (unlikely(ex == NULL || eh == NULL)) {
1668 EXT4_ERROR_INODE(inode,
1669 "ex %p == NULL or eh %p == NULL", ex, eh);
1670 return -EFSCORRUPTED;
1674 /* there is no tree at all */
1678 if (ex != EXT_FIRST_EXTENT(eh)) {
1679 /* we correct tree if first leaf got modified only */
1684 * TODO: we need correction if border is smaller than current one
1687 border = path[depth].p_ext->ee_block;
1688 err = ext4_ext_get_access(handle, inode, path + k);
1691 path[k].p_idx->ei_block = border;
1692 err = ext4_ext_dirty(handle, inode, path + k);
1697 /* change all left-side indexes */
1698 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1700 err = ext4_ext_get_access(handle, inode, path + k);
1703 path[k].p_idx->ei_block = border;
1704 err = ext4_ext_dirty(handle, inode, path + k);
1712 static int ext4_can_extents_be_merged(struct inode *inode,
1713 struct ext4_extent *ex1,
1714 struct ext4_extent *ex2)
1716 unsigned short ext1_ee_len, ext2_ee_len;
1718 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1721 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1722 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1724 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1725 le32_to_cpu(ex2->ee_block))
1728 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1731 if (ext4_ext_is_unwritten(ex1) &&
1732 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1734 #ifdef AGGRESSIVE_TEST
1735 if (ext1_ee_len >= 4)
1739 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1745 * This function tries to merge the "ex" extent to the next extent in the tree.
1746 * It always tries to merge towards right. If you want to merge towards
1747 * left, pass "ex - 1" as argument instead of "ex".
1748 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1749 * 1 if they got merged.
1751 static int ext4_ext_try_to_merge_right(struct inode *inode,
1752 struct ext4_ext_path *path,
1753 struct ext4_extent *ex)
1755 struct ext4_extent_header *eh;
1756 unsigned int depth, len;
1757 int merge_done = 0, unwritten;
1759 depth = ext_depth(inode);
1760 BUG_ON(path[depth].p_hdr == NULL);
1761 eh = path[depth].p_hdr;
1763 while (ex < EXT_LAST_EXTENT(eh)) {
1764 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1766 /* merge with next extent! */
1767 unwritten = ext4_ext_is_unwritten(ex);
1768 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1769 + ext4_ext_get_actual_len(ex + 1));
1771 ext4_ext_mark_unwritten(ex);
1773 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1774 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1775 * sizeof(struct ext4_extent);
1776 memmove(ex + 1, ex + 2, len);
1778 le16_add_cpu(&eh->eh_entries, -1);
1780 WARN_ON(eh->eh_entries == 0);
1781 if (!eh->eh_entries)
1782 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1789 * This function does a very simple check to see if we can collapse
1790 * an extent tree with a single extent tree leaf block into the inode.
1792 static void ext4_ext_try_to_merge_up(handle_t *handle,
1793 struct inode *inode,
1794 struct ext4_ext_path *path)
1797 unsigned max_root = ext4_ext_space_root(inode, 0);
1800 if ((path[0].p_depth != 1) ||
1801 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1802 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1806 * We need to modify the block allocation bitmap and the block
1807 * group descriptor to release the extent tree block. If we
1808 * can't get the journal credits, give up.
1810 if (ext4_journal_extend(handle, 2,
1811 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1815 * Copy the extent data up to the inode
1817 blk = ext4_idx_pblock(path[0].p_idx);
1818 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1819 sizeof(struct ext4_extent_idx);
1820 s += sizeof(struct ext4_extent_header);
1822 path[1].p_maxdepth = path[0].p_maxdepth;
1823 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1824 path[0].p_depth = 0;
1825 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1826 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1827 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1829 brelse(path[1].p_bh);
1830 ext4_free_blocks(handle, inode, NULL, blk, 1,
1831 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1835 * This function tries to merge the @ex extent to neighbours in the tree, then
1836 * tries to collapse the extent tree into the inode.
1838 static void ext4_ext_try_to_merge(handle_t *handle,
1839 struct inode *inode,
1840 struct ext4_ext_path *path,
1841 struct ext4_extent *ex)
1843 struct ext4_extent_header *eh;
1847 depth = ext_depth(inode);
1848 BUG_ON(path[depth].p_hdr == NULL);
1849 eh = path[depth].p_hdr;
1851 if (ex > EXT_FIRST_EXTENT(eh))
1852 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1855 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1857 ext4_ext_try_to_merge_up(handle, inode, path);
1861 * check if a portion of the "newext" extent overlaps with an
1864 * If there is an overlap discovered, it updates the length of the newext
1865 * such that there will be no overlap, and then returns 1.
1866 * If there is no overlap found, it returns 0.
1868 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1869 struct inode *inode,
1870 struct ext4_extent *newext,
1871 struct ext4_ext_path *path)
1874 unsigned int depth, len1;
1875 unsigned int ret = 0;
1877 b1 = le32_to_cpu(newext->ee_block);
1878 len1 = ext4_ext_get_actual_len(newext);
1879 depth = ext_depth(inode);
1880 if (!path[depth].p_ext)
1882 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1885 * get the next allocated block if the extent in the path
1886 * is before the requested block(s)
1889 b2 = ext4_ext_next_allocated_block(path);
1890 if (b2 == EXT_MAX_BLOCKS)
1892 b2 = EXT4_LBLK_CMASK(sbi, b2);
1895 /* check for wrap through zero on extent logical start block*/
1896 if (b1 + len1 < b1) {
1897 len1 = EXT_MAX_BLOCKS - b1;
1898 newext->ee_len = cpu_to_le16(len1);
1902 /* check for overlap */
1903 if (b1 + len1 > b2) {
1904 newext->ee_len = cpu_to_le16(b2 - b1);
1912 * ext4_ext_insert_extent:
1913 * tries to merge requested extent into the existing extent or
1914 * inserts requested extent as new one into the tree,
1915 * creating new leaf in the no-space case.
1917 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1918 struct ext4_ext_path **ppath,
1919 struct ext4_extent *newext, int gb_flags)
1921 struct ext4_ext_path *path = *ppath;
1922 struct ext4_extent_header *eh;
1923 struct ext4_extent *ex, *fex;
1924 struct ext4_extent *nearex; /* nearest extent */
1925 struct ext4_ext_path *npath = NULL;
1926 int depth, len, err;
1928 int mb_flags = 0, unwritten;
1930 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1931 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1932 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1933 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1934 return -EFSCORRUPTED;
1936 depth = ext_depth(inode);
1937 ex = path[depth].p_ext;
1938 eh = path[depth].p_hdr;
1939 if (unlikely(path[depth].p_hdr == NULL)) {
1940 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1941 return -EFSCORRUPTED;
1944 /* try to insert block into found extent and return */
1945 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1948 * Try to see whether we should rather test the extent on
1949 * right from ex, or from the left of ex. This is because
1950 * ext4_find_extent() can return either extent on the
1951 * left, or on the right from the searched position. This
1952 * will make merging more effective.
1954 if (ex < EXT_LAST_EXTENT(eh) &&
1955 (le32_to_cpu(ex->ee_block) +
1956 ext4_ext_get_actual_len(ex) <
1957 le32_to_cpu(newext->ee_block))) {
1960 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1961 (le32_to_cpu(newext->ee_block) +
1962 ext4_ext_get_actual_len(newext) <
1963 le32_to_cpu(ex->ee_block)))
1966 /* Try to append newex to the ex */
1967 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1968 ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
1970 ext4_ext_is_unwritten(newext),
1971 ext4_ext_get_actual_len(newext),
1972 le32_to_cpu(ex->ee_block),
1973 ext4_ext_is_unwritten(ex),
1974 ext4_ext_get_actual_len(ex),
1975 ext4_ext_pblock(ex));
1976 err = ext4_ext_get_access(handle, inode,
1980 unwritten = ext4_ext_is_unwritten(ex);
1981 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1982 + ext4_ext_get_actual_len(newext));
1984 ext4_ext_mark_unwritten(ex);
1985 eh = path[depth].p_hdr;
1991 /* Try to prepend newex to the ex */
1992 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1993 ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
1995 le32_to_cpu(newext->ee_block),
1996 ext4_ext_is_unwritten(newext),
1997 ext4_ext_get_actual_len(newext),
1998 le32_to_cpu(ex->ee_block),
1999 ext4_ext_is_unwritten(ex),
2000 ext4_ext_get_actual_len(ex),
2001 ext4_ext_pblock(ex));
2002 err = ext4_ext_get_access(handle, inode,
2007 unwritten = ext4_ext_is_unwritten(ex);
2008 ex->ee_block = newext->ee_block;
2009 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2010 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2011 + ext4_ext_get_actual_len(newext));
2013 ext4_ext_mark_unwritten(ex);
2014 eh = path[depth].p_hdr;
2020 depth = ext_depth(inode);
2021 eh = path[depth].p_hdr;
2022 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2025 /* probably next leaf has space for us? */
2026 fex = EXT_LAST_EXTENT(eh);
2027 next = EXT_MAX_BLOCKS;
2028 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2029 next = ext4_ext_next_leaf_block(path);
2030 if (next != EXT_MAX_BLOCKS) {
2031 ext_debug(inode, "next leaf block - %u\n", next);
2032 BUG_ON(npath != NULL);
2033 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2035 return PTR_ERR(npath);
2036 BUG_ON(npath->p_depth != path->p_depth);
2037 eh = npath[depth].p_hdr;
2038 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2039 ext_debug(inode, "next leaf isn't full(%d)\n",
2040 le16_to_cpu(eh->eh_entries));
2044 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2045 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2049 * There is no free space in the found leaf.
2050 * We're gonna add a new leaf in the tree.
2052 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2053 mb_flags |= EXT4_MB_USE_RESERVED;
2054 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2058 depth = ext_depth(inode);
2059 eh = path[depth].p_hdr;
2062 nearex = path[depth].p_ext;
2064 err = ext4_ext_get_access(handle, inode, path + depth);
2069 /* there is no extent in this leaf, create first one */
2070 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2071 le32_to_cpu(newext->ee_block),
2072 ext4_ext_pblock(newext),
2073 ext4_ext_is_unwritten(newext),
2074 ext4_ext_get_actual_len(newext));
2075 nearex = EXT_FIRST_EXTENT(eh);
2077 if (le32_to_cpu(newext->ee_block)
2078 > le32_to_cpu(nearex->ee_block)) {
2080 ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2082 le32_to_cpu(newext->ee_block),
2083 ext4_ext_pblock(newext),
2084 ext4_ext_is_unwritten(newext),
2085 ext4_ext_get_actual_len(newext),
2090 BUG_ON(newext->ee_block == nearex->ee_block);
2091 ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2093 le32_to_cpu(newext->ee_block),
2094 ext4_ext_pblock(newext),
2095 ext4_ext_is_unwritten(newext),
2096 ext4_ext_get_actual_len(newext),
2099 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2101 ext_debug(inode, "insert %u:%llu:[%d]%d: "
2102 "move %d extents from 0x%p to 0x%p\n",
2103 le32_to_cpu(newext->ee_block),
2104 ext4_ext_pblock(newext),
2105 ext4_ext_is_unwritten(newext),
2106 ext4_ext_get_actual_len(newext),
2107 len, nearex, nearex + 1);
2108 memmove(nearex + 1, nearex,
2109 len * sizeof(struct ext4_extent));
2113 le16_add_cpu(&eh->eh_entries, 1);
2114 path[depth].p_ext = nearex;
2115 nearex->ee_block = newext->ee_block;
2116 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2117 nearex->ee_len = newext->ee_len;
2120 /* try to merge extents */
2121 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2122 ext4_ext_try_to_merge(handle, inode, path, nearex);
2125 /* time to correct all indexes above */
2126 err = ext4_ext_correct_indexes(handle, inode, path);
2130 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2133 ext4_ext_drop_refs(npath);
2138 static int ext4_fill_es_cache_info(struct inode *inode,
2139 ext4_lblk_t block, ext4_lblk_t num,
2140 struct fiemap_extent_info *fieinfo)
2142 ext4_lblk_t next, end = block + num - 1;
2143 struct extent_status es;
2144 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2148 while (block <= end) {
2151 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2153 if (ext4_es_is_unwritten(&es))
2154 flags |= FIEMAP_EXTENT_UNWRITTEN;
2155 if (ext4_es_is_delayed(&es))
2156 flags |= (FIEMAP_EXTENT_DELALLOC |
2157 FIEMAP_EXTENT_UNKNOWN);
2158 if (ext4_es_is_hole(&es))
2159 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2161 flags |= FIEMAP_EXTENT_LAST;
2162 if (flags & (FIEMAP_EXTENT_DELALLOC|
2163 EXT4_FIEMAP_EXTENT_HOLE))
2166 es.es_pblk = ext4_es_pblock(&es);
2167 err = fiemap_fill_next_extent(fieinfo,
2168 (__u64)es.es_lblk << blksize_bits,
2169 (__u64)es.es_pblk << blksize_bits,
2170 (__u64)es.es_len << blksize_bits,
2185 * ext4_ext_determine_hole - determine hole around given block
2186 * @inode: inode we lookup in
2187 * @path: path in extent tree to @lblk
2188 * @lblk: pointer to logical block around which we want to determine hole
2190 * Determine hole length (and start if easily possible) around given logical
2191 * block. We don't try too hard to find the beginning of the hole but @path
2192 * actually points to extent before @lblk, we provide it.
2194 * The function returns the length of a hole starting at @lblk. We update @lblk
2195 * to the beginning of the hole if we managed to find it.
2197 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2198 struct ext4_ext_path *path,
2201 int depth = ext_depth(inode);
2202 struct ext4_extent *ex;
2205 ex = path[depth].p_ext;
2207 /* there is no extent yet, so gap is [0;-] */
2209 len = EXT_MAX_BLOCKS;
2210 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2211 len = le32_to_cpu(ex->ee_block) - *lblk;
2212 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2213 + ext4_ext_get_actual_len(ex)) {
2216 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2217 next = ext4_ext_next_allocated_block(path);
2218 BUG_ON(next == *lblk);
2227 * ext4_ext_put_gap_in_cache:
2228 * calculate boundaries of the gap that the requested block fits into
2229 * and cache this gap
2232 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2233 ext4_lblk_t hole_len)
2235 struct extent_status es;
2237 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2238 hole_start + hole_len - 1, &es);
2240 /* There's delayed extent containing lblock? */
2241 if (es.es_lblk <= hole_start)
2243 hole_len = min(es.es_lblk - hole_start, hole_len);
2245 ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2246 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2247 EXTENT_STATUS_HOLE);
2252 * removes index from the index block.
2254 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2255 struct ext4_ext_path *path, int depth)
2260 /* free index block */
2262 path = path + depth;
2263 leaf = ext4_idx_pblock(path->p_idx);
2264 if (unlikely(path->p_hdr->eh_entries == 0)) {
2265 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2266 return -EFSCORRUPTED;
2268 err = ext4_ext_get_access(handle, inode, path);
2272 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2273 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2274 len *= sizeof(struct ext4_extent_idx);
2275 memmove(path->p_idx, path->p_idx + 1, len);
2278 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2279 err = ext4_ext_dirty(handle, inode, path);
2282 ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2283 trace_ext4_ext_rm_idx(inode, leaf);
2285 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2286 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2288 while (--depth >= 0) {
2289 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2292 err = ext4_ext_get_access(handle, inode, path);
2295 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2296 err = ext4_ext_dirty(handle, inode, path);
2304 * ext4_ext_calc_credits_for_single_extent:
2305 * This routine returns max. credits that needed to insert an extent
2306 * to the extent tree.
2307 * When pass the actual path, the caller should calculate credits
2310 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2311 struct ext4_ext_path *path)
2314 int depth = ext_depth(inode);
2317 /* probably there is space in leaf? */
2318 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2319 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2322 * There are some space in the leaf tree, no
2323 * need to account for leaf block credit
2325 * bitmaps and block group descriptor blocks
2326 * and other metadata blocks still need to be
2329 /* 1 bitmap, 1 block group descriptor */
2330 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2335 return ext4_chunk_trans_blocks(inode, nrblocks);
2339 * How many index/leaf blocks need to change/allocate to add @extents extents?
2341 * If we add a single extent, then in the worse case, each tree level
2342 * index/leaf need to be changed in case of the tree split.
2344 * If more extents are inserted, they could cause the whole tree split more
2345 * than once, but this is really rare.
2347 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2352 /* If we are converting the inline data, only one is needed here. */
2353 if (ext4_has_inline_data(inode))
2356 depth = ext_depth(inode);
2366 static inline int get_default_free_blocks_flags(struct inode *inode)
2368 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2369 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2370 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2371 else if (ext4_should_journal_data(inode))
2372 return EXT4_FREE_BLOCKS_FORGET;
2377 * ext4_rereserve_cluster - increment the reserved cluster count when
2378 * freeing a cluster with a pending reservation
2380 * @inode - file containing the cluster
2381 * @lblk - logical block in cluster to be reserved
2383 * Increments the reserved cluster count and adjusts quota in a bigalloc
2384 * file system when freeing a partial cluster containing at least one
2385 * delayed and unwritten block. A partial cluster meeting that
2386 * requirement will have a pending reservation. If so, the
2387 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2388 * defer reserved and allocated space accounting to a subsequent call
2391 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2393 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2394 struct ext4_inode_info *ei = EXT4_I(inode);
2396 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2398 spin_lock(&ei->i_block_reservation_lock);
2399 ei->i_reserved_data_blocks++;
2400 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2401 spin_unlock(&ei->i_block_reservation_lock);
2403 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2404 ext4_remove_pending(inode, lblk);
2407 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2408 struct ext4_extent *ex,
2409 struct partial_cluster *partial,
2410 ext4_lblk_t from, ext4_lblk_t to)
2412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2413 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2414 ext4_fsblk_t last_pblk, pblk;
2418 /* only extent tail removal is allowed */
2419 if (from < le32_to_cpu(ex->ee_block) ||
2420 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2421 ext4_error(sbi->s_sb,
2422 "strange request: removal(2) %u-%u from %u:%u",
2423 from, to, le32_to_cpu(ex->ee_block), ee_len);
2427 #ifdef EXTENTS_STATS
2428 spin_lock(&sbi->s_ext_stats_lock);
2429 sbi->s_ext_blocks += ee_len;
2430 sbi->s_ext_extents++;
2431 if (ee_len < sbi->s_ext_min)
2432 sbi->s_ext_min = ee_len;
2433 if (ee_len > sbi->s_ext_max)
2434 sbi->s_ext_max = ee_len;
2435 if (ext_depth(inode) > sbi->s_depth_max)
2436 sbi->s_depth_max = ext_depth(inode);
2437 spin_unlock(&sbi->s_ext_stats_lock);
2440 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2443 * if we have a partial cluster, and it's different from the
2444 * cluster of the last block in the extent, we free it
2446 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2448 if (partial->state != initial &&
2449 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2450 if (partial->state == tofree) {
2451 flags = get_default_free_blocks_flags(inode);
2452 if (ext4_is_pending(inode, partial->lblk))
2453 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2454 ext4_free_blocks(handle, inode, NULL,
2455 EXT4_C2B(sbi, partial->pclu),
2456 sbi->s_cluster_ratio, flags);
2457 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2458 ext4_rereserve_cluster(inode, partial->lblk);
2460 partial->state = initial;
2463 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2464 pblk = ext4_ext_pblock(ex) + ee_len - num;
2467 * We free the partial cluster at the end of the extent (if any),
2468 * unless the cluster is used by another extent (partial_cluster
2469 * state is nofree). If a partial cluster exists here, it must be
2470 * shared with the last block in the extent.
2472 flags = get_default_free_blocks_flags(inode);
2474 /* partial, left end cluster aligned, right end unaligned */
2475 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2476 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2477 (partial->state != nofree)) {
2478 if (ext4_is_pending(inode, to))
2479 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2480 ext4_free_blocks(handle, inode, NULL,
2481 EXT4_PBLK_CMASK(sbi, last_pblk),
2482 sbi->s_cluster_ratio, flags);
2483 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2484 ext4_rereserve_cluster(inode, to);
2485 partial->state = initial;
2486 flags = get_default_free_blocks_flags(inode);
2489 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2492 * For bigalloc file systems, we never free a partial cluster
2493 * at the beginning of the extent. Instead, we check to see if we
2494 * need to free it on a subsequent call to ext4_remove_blocks,
2495 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2497 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2498 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2500 /* reset the partial cluster if we've freed past it */
2501 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2502 partial->state = initial;
2505 * If we've freed the entire extent but the beginning is not left
2506 * cluster aligned and is not marked as ineligible for freeing we
2507 * record the partial cluster at the beginning of the extent. It
2508 * wasn't freed by the preceding ext4_free_blocks() call, and we
2509 * need to look farther to the left to determine if it's to be freed
2510 * (not shared with another extent). Else, reset the partial
2511 * cluster - we're either done freeing or the beginning of the
2512 * extent is left cluster aligned.
2514 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2515 if (partial->state == initial) {
2516 partial->pclu = EXT4_B2C(sbi, pblk);
2517 partial->lblk = from;
2518 partial->state = tofree;
2521 partial->state = initial;
2528 * ext4_ext_rm_leaf() Removes the extents associated with the
2529 * blocks appearing between "start" and "end". Both "start"
2530 * and "end" must appear in the same extent or EIO is returned.
2532 * @handle: The journal handle
2533 * @inode: The files inode
2534 * @path: The path to the leaf
2535 * @partial_cluster: The cluster which we'll have to free if all extents
2536 * has been released from it. However, if this value is
2537 * negative, it's a cluster just to the right of the
2538 * punched region and it must not be freed.
2539 * @start: The first block to remove
2540 * @end: The last block to remove
2543 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2544 struct ext4_ext_path *path,
2545 struct partial_cluster *partial,
2546 ext4_lblk_t start, ext4_lblk_t end)
2548 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2549 int err = 0, correct_index = 0;
2550 int depth = ext_depth(inode), credits, revoke_credits;
2551 struct ext4_extent_header *eh;
2554 ext4_lblk_t ex_ee_block;
2555 unsigned short ex_ee_len;
2556 unsigned unwritten = 0;
2557 struct ext4_extent *ex;
2560 /* the header must be checked already in ext4_ext_remove_space() */
2561 ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2562 if (!path[depth].p_hdr)
2563 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2564 eh = path[depth].p_hdr;
2565 if (unlikely(path[depth].p_hdr == NULL)) {
2566 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2567 return -EFSCORRUPTED;
2569 /* find where to start removing */
2570 ex = path[depth].p_ext;
2572 ex = EXT_LAST_EXTENT(eh);
2574 ex_ee_block = le32_to_cpu(ex->ee_block);
2575 ex_ee_len = ext4_ext_get_actual_len(ex);
2577 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2579 while (ex >= EXT_FIRST_EXTENT(eh) &&
2580 ex_ee_block + ex_ee_len > start) {
2582 if (ext4_ext_is_unwritten(ex))
2587 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2588 unwritten, ex_ee_len);
2589 path[depth].p_ext = ex;
2591 a = ex_ee_block > start ? ex_ee_block : start;
2592 b = ex_ee_block+ex_ee_len - 1 < end ?
2593 ex_ee_block+ex_ee_len - 1 : end;
2595 ext_debug(inode, " border %u:%u\n", a, b);
2597 /* If this extent is beyond the end of the hole, skip it */
2598 if (end < ex_ee_block) {
2600 * We're going to skip this extent and move to another,
2601 * so note that its first cluster is in use to avoid
2602 * freeing it when removing blocks. Eventually, the
2603 * right edge of the truncated/punched region will
2604 * be just to the left.
2606 if (sbi->s_cluster_ratio > 1) {
2607 pblk = ext4_ext_pblock(ex);
2608 partial->pclu = EXT4_B2C(sbi, pblk);
2609 partial->state = nofree;
2612 ex_ee_block = le32_to_cpu(ex->ee_block);
2613 ex_ee_len = ext4_ext_get_actual_len(ex);
2615 } else if (b != ex_ee_block + ex_ee_len - 1) {
2616 EXT4_ERROR_INODE(inode,
2617 "can not handle truncate %u:%u "
2619 start, end, ex_ee_block,
2620 ex_ee_block + ex_ee_len - 1);
2621 err = -EFSCORRUPTED;
2623 } else if (a != ex_ee_block) {
2624 /* remove tail of the extent */
2625 num = a - ex_ee_block;
2627 /* remove whole extent: excellent! */
2631 * 3 for leaf, sb, and inode plus 2 (bmap and group
2632 * descriptor) for each block group; assume two block
2633 * groups plus ex_ee_len/blocks_per_block_group for
2636 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2637 if (ex == EXT_FIRST_EXTENT(eh)) {
2639 credits += (ext_depth(inode)) + 1;
2641 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2643 * We may end up freeing some index blocks and data from the
2644 * punched range. Note that partial clusters are accounted for
2645 * by ext4_free_data_revoke_credits().
2648 ext4_free_metadata_revoke_credits(inode->i_sb,
2650 ext4_free_data_revoke_credits(inode, b - a + 1);
2652 err = ext4_datasem_ensure_credits(handle, inode, credits,
2653 credits, revoke_credits);
2660 err = ext4_ext_get_access(handle, inode, path + depth);
2664 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2669 /* this extent is removed; mark slot entirely unused */
2670 ext4_ext_store_pblock(ex, 0);
2672 ex->ee_len = cpu_to_le16(num);
2674 * Do not mark unwritten if all the blocks in the
2675 * extent have been removed.
2677 if (unwritten && num)
2678 ext4_ext_mark_unwritten(ex);
2680 * If the extent was completely released,
2681 * we need to remove it from the leaf
2684 if (end != EXT_MAX_BLOCKS - 1) {
2686 * For hole punching, we need to scoot all the
2687 * extents up when an extent is removed so that
2688 * we dont have blank extents in the middle
2690 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2691 sizeof(struct ext4_extent));
2693 /* Now get rid of the one at the end */
2694 memset(EXT_LAST_EXTENT(eh), 0,
2695 sizeof(struct ext4_extent));
2697 le16_add_cpu(&eh->eh_entries, -1);
2700 err = ext4_ext_dirty(handle, inode, path + depth);
2704 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2705 ext4_ext_pblock(ex));
2707 ex_ee_block = le32_to_cpu(ex->ee_block);
2708 ex_ee_len = ext4_ext_get_actual_len(ex);
2711 if (correct_index && eh->eh_entries)
2712 err = ext4_ext_correct_indexes(handle, inode, path);
2715 * If there's a partial cluster and at least one extent remains in
2716 * the leaf, free the partial cluster if it isn't shared with the
2717 * current extent. If it is shared with the current extent
2718 * we reset the partial cluster because we've reached the start of the
2719 * truncated/punched region and we're done removing blocks.
2721 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2722 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2723 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2724 int flags = get_default_free_blocks_flags(inode);
2726 if (ext4_is_pending(inode, partial->lblk))
2727 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2728 ext4_free_blocks(handle, inode, NULL,
2729 EXT4_C2B(sbi, partial->pclu),
2730 sbi->s_cluster_ratio, flags);
2731 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2732 ext4_rereserve_cluster(inode, partial->lblk);
2734 partial->state = initial;
2737 /* if this leaf is free, then we should
2738 * remove it from index block above */
2739 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2740 err = ext4_ext_rm_idx(handle, inode, path, depth);
2747 * ext4_ext_more_to_rm:
2748 * returns 1 if current index has to be freed (even partial)
2751 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2753 BUG_ON(path->p_idx == NULL);
2755 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2759 * if truncate on deeper level happened, it wasn't partial,
2760 * so we have to consider current index for truncation
2762 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2767 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2770 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2771 int depth = ext_depth(inode);
2772 struct ext4_ext_path *path = NULL;
2773 struct partial_cluster partial;
2779 partial.state = initial;
2781 ext_debug(inode, "truncate since %u to %u\n", start, end);
2783 /* probably first extent we're gonna free will be last in block */
2784 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2786 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2788 return PTR_ERR(handle);
2791 trace_ext4_ext_remove_space(inode, start, end, depth);
2794 * Check if we are removing extents inside the extent tree. If that
2795 * is the case, we are going to punch a hole inside the extent tree
2796 * so we have to check whether we need to split the extent covering
2797 * the last block to remove so we can easily remove the part of it
2798 * in ext4_ext_rm_leaf().
2800 if (end < EXT_MAX_BLOCKS - 1) {
2801 struct ext4_extent *ex;
2802 ext4_lblk_t ee_block, ex_end, lblk;
2805 /* find extent for or closest extent to this block */
2806 path = ext4_find_extent(inode, end, NULL,
2807 EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2809 ext4_journal_stop(handle);
2810 return PTR_ERR(path);
2812 depth = ext_depth(inode);
2813 /* Leaf not may not exist only if inode has no blocks at all */
2814 ex = path[depth].p_ext;
2817 EXT4_ERROR_INODE(inode,
2818 "path[%d].p_hdr == NULL",
2820 err = -EFSCORRUPTED;
2825 ee_block = le32_to_cpu(ex->ee_block);
2826 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2829 * See if the last block is inside the extent, if so split
2830 * the extent at 'end' block so we can easily remove the
2831 * tail of the first part of the split extent in
2832 * ext4_ext_rm_leaf().
2834 if (end >= ee_block && end < ex_end) {
2837 * If we're going to split the extent, note that
2838 * the cluster containing the block after 'end' is
2839 * in use to avoid freeing it when removing blocks.
2841 if (sbi->s_cluster_ratio > 1) {
2842 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2843 partial.pclu = EXT4_B2C(sbi, pblk);
2844 partial.state = nofree;
2848 * Split the extent in two so that 'end' is the last
2849 * block in the first new extent. Also we should not
2850 * fail removing space due to ENOSPC so try to use
2851 * reserved block if that happens.
2853 err = ext4_force_split_extent_at(handle, inode, &path,
2858 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2859 partial.state == initial) {
2861 * If we're punching, there's an extent to the right.
2862 * If the partial cluster hasn't been set, set it to
2863 * that extent's first cluster and its state to nofree
2864 * so it won't be freed should it contain blocks to be
2865 * removed. If it's already set (tofree/nofree), we're
2866 * retrying and keep the original partial cluster info
2867 * so a cluster marked tofree as a result of earlier
2868 * extent removal is not lost.
2871 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2876 partial.pclu = EXT4_B2C(sbi, pblk);
2877 partial.state = nofree;
2882 * We start scanning from right side, freeing all the blocks
2883 * after i_size and walking into the tree depth-wise.
2885 depth = ext_depth(inode);
2890 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2892 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2893 GFP_NOFS | __GFP_NOFAIL);
2895 ext4_journal_stop(handle);
2898 path[0].p_maxdepth = path[0].p_depth = depth;
2899 path[0].p_hdr = ext_inode_hdr(inode);
2902 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2903 err = -EFSCORRUPTED;
2909 while (i >= 0 && err == 0) {
2911 /* this is leaf block */
2912 err = ext4_ext_rm_leaf(handle, inode, path,
2913 &partial, start, end);
2914 /* root level has p_bh == NULL, brelse() eats this */
2915 brelse(path[i].p_bh);
2916 path[i].p_bh = NULL;
2921 /* this is index block */
2922 if (!path[i].p_hdr) {
2923 ext_debug(inode, "initialize header\n");
2924 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2927 if (!path[i].p_idx) {
2928 /* this level hasn't been touched yet */
2929 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2930 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2931 ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2933 le16_to_cpu(path[i].p_hdr->eh_entries));
2935 /* we were already here, see at next index */
2939 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2940 i, EXT_FIRST_INDEX(path[i].p_hdr),
2942 if (ext4_ext_more_to_rm(path + i)) {
2943 struct buffer_head *bh;
2944 /* go to the next level */
2945 ext_debug(inode, "move to level %d (block %llu)\n",
2946 i + 1, ext4_idx_pblock(path[i].p_idx));
2947 memset(path + i + 1, 0, sizeof(*path));
2948 bh = read_extent_tree_block(inode,
2949 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2952 /* should we reset i_size? */
2956 /* Yield here to deal with large extent trees.
2957 * Should be a no-op if we did IO above. */
2959 if (WARN_ON(i + 1 > depth)) {
2960 err = -EFSCORRUPTED;
2963 path[i + 1].p_bh = bh;
2965 /* save actual number of indexes since this
2966 * number is changed at the next iteration */
2967 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2970 /* we finished processing this index, go up */
2971 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2972 /* index is empty, remove it;
2973 * handle must be already prepared by the
2974 * truncatei_leaf() */
2975 err = ext4_ext_rm_idx(handle, inode, path, i);
2977 /* root level has p_bh == NULL, brelse() eats this */
2978 brelse(path[i].p_bh);
2979 path[i].p_bh = NULL;
2981 ext_debug(inode, "return to level %d\n", i);
2985 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
2986 path->p_hdr->eh_entries);
2989 * if there's a partial cluster and we have removed the first extent
2990 * in the file, then we also free the partial cluster, if any
2992 if (partial.state == tofree && err == 0) {
2993 int flags = get_default_free_blocks_flags(inode);
2995 if (ext4_is_pending(inode, partial.lblk))
2996 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2997 ext4_free_blocks(handle, inode, NULL,
2998 EXT4_C2B(sbi, partial.pclu),
2999 sbi->s_cluster_ratio, flags);
3000 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3001 ext4_rereserve_cluster(inode, partial.lblk);
3002 partial.state = initial;
3005 /* TODO: flexible tree reduction should be here */
3006 if (path->p_hdr->eh_entries == 0) {
3008 * truncate to zero freed all the tree,
3009 * so we need to correct eh_depth
3011 err = ext4_ext_get_access(handle, inode, path);
3013 ext_inode_hdr(inode)->eh_depth = 0;
3014 ext_inode_hdr(inode)->eh_max =
3015 cpu_to_le16(ext4_ext_space_root(inode, 0));
3016 err = ext4_ext_dirty(handle, inode, path);
3020 ext4_ext_drop_refs(path);
3025 ext4_journal_stop(handle);
3031 * called at mount time
3033 void ext4_ext_init(struct super_block *sb)
3036 * possible initialization would be here
3039 if (ext4_has_feature_extents(sb)) {
3040 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3041 printk(KERN_INFO "EXT4-fs: file extents enabled"
3042 #ifdef AGGRESSIVE_TEST
3043 ", aggressive tests"
3045 #ifdef CHECK_BINSEARCH
3048 #ifdef EXTENTS_STATS
3053 #ifdef EXTENTS_STATS
3054 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3055 EXT4_SB(sb)->s_ext_min = 1 << 30;
3056 EXT4_SB(sb)->s_ext_max = 0;
3062 * called at umount time
3064 void ext4_ext_release(struct super_block *sb)
3066 if (!ext4_has_feature_extents(sb))
3069 #ifdef EXTENTS_STATS
3070 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3071 struct ext4_sb_info *sbi = EXT4_SB(sb);
3072 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3073 sbi->s_ext_blocks, sbi->s_ext_extents,
3074 sbi->s_ext_blocks / sbi->s_ext_extents);
3075 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3076 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3081 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3083 ext4_lblk_t ee_block;
3084 ext4_fsblk_t ee_pblock;
3085 unsigned int ee_len;
3087 ee_block = le32_to_cpu(ex->ee_block);
3088 ee_len = ext4_ext_get_actual_len(ex);
3089 ee_pblock = ext4_ext_pblock(ex);
3094 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3095 EXTENT_STATUS_WRITTEN);
3098 /* FIXME!! we need to try to merge to left or right after zero-out */
3099 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3101 ext4_fsblk_t ee_pblock;
3102 unsigned int ee_len;
3104 ee_len = ext4_ext_get_actual_len(ex);
3105 ee_pblock = ext4_ext_pblock(ex);
3106 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3111 * ext4_split_extent_at() splits an extent at given block.
3113 * @handle: the journal handle
3114 * @inode: the file inode
3115 * @path: the path to the extent
3116 * @split: the logical block where the extent is splitted.
3117 * @split_flags: indicates if the extent could be zeroout if split fails, and
3118 * the states(init or unwritten) of new extents.
3119 * @flags: flags used to insert new extent to extent tree.
3122 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3123 * of which are determined by split_flag.
3125 * There are two cases:
3126 * a> the extent are splitted into two extent.
3127 * b> split is not needed, and just mark the extent.
3129 * return 0 on success.
3131 static int ext4_split_extent_at(handle_t *handle,
3132 struct inode *inode,
3133 struct ext4_ext_path **ppath,
3138 struct ext4_ext_path *path = *ppath;
3139 ext4_fsblk_t newblock;
3140 ext4_lblk_t ee_block;
3141 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3142 struct ext4_extent *ex2 = NULL;
3143 unsigned int ee_len, depth;
3146 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3147 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3149 ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3151 ext4_ext_show_leaf(inode, path);
3153 depth = ext_depth(inode);
3154 ex = path[depth].p_ext;
3155 ee_block = le32_to_cpu(ex->ee_block);
3156 ee_len = ext4_ext_get_actual_len(ex);
3157 newblock = split - ee_block + ext4_ext_pblock(ex);
3159 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3160 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3161 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3162 EXT4_EXT_MARK_UNWRIT1 |
3163 EXT4_EXT_MARK_UNWRIT2));
3165 err = ext4_ext_get_access(handle, inode, path + depth);
3169 if (split == ee_block) {
3171 * case b: block @split is the block that the extent begins with
3172 * then we just change the state of the extent, and splitting
3175 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3176 ext4_ext_mark_unwritten(ex);
3178 ext4_ext_mark_initialized(ex);
3180 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3181 ext4_ext_try_to_merge(handle, inode, path, ex);
3183 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3188 memcpy(&orig_ex, ex, sizeof(orig_ex));
3189 ex->ee_len = cpu_to_le16(split - ee_block);
3190 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3191 ext4_ext_mark_unwritten(ex);
3194 * path may lead to new leaf, not to original leaf any more
3195 * after ext4_ext_insert_extent() returns,
3197 err = ext4_ext_dirty(handle, inode, path + depth);
3199 goto fix_extent_len;
3202 ex2->ee_block = cpu_to_le32(split);
3203 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3204 ext4_ext_store_pblock(ex2, newblock);
3205 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3206 ext4_ext_mark_unwritten(ex2);
3208 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3209 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3210 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3211 if (split_flag & EXT4_EXT_DATA_VALID1) {
3212 err = ext4_ext_zeroout(inode, ex2);
3213 zero_ex.ee_block = ex2->ee_block;
3214 zero_ex.ee_len = cpu_to_le16(
3215 ext4_ext_get_actual_len(ex2));
3216 ext4_ext_store_pblock(&zero_ex,
3217 ext4_ext_pblock(ex2));
3219 err = ext4_ext_zeroout(inode, ex);
3220 zero_ex.ee_block = ex->ee_block;
3221 zero_ex.ee_len = cpu_to_le16(
3222 ext4_ext_get_actual_len(ex));
3223 ext4_ext_store_pblock(&zero_ex,
3224 ext4_ext_pblock(ex));
3227 err = ext4_ext_zeroout(inode, &orig_ex);
3228 zero_ex.ee_block = orig_ex.ee_block;
3229 zero_ex.ee_len = cpu_to_le16(
3230 ext4_ext_get_actual_len(&orig_ex));
3231 ext4_ext_store_pblock(&zero_ex,
3232 ext4_ext_pblock(&orig_ex));
3236 goto fix_extent_len;
3237 /* update the extent length and mark as initialized */
3238 ex->ee_len = cpu_to_le16(ee_len);
3239 ext4_ext_try_to_merge(handle, inode, path, ex);
3240 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3242 goto fix_extent_len;
3244 /* update extent status tree */
3245 err = ext4_zeroout_es(inode, &zero_ex);
3249 goto fix_extent_len;
3252 ext4_ext_show_leaf(inode, path);
3256 ex->ee_len = orig_ex.ee_len;
3258 * Ignore ext4_ext_dirty return value since we are already in error path
3259 * and err is a non-zero error code.
3261 ext4_ext_dirty(handle, inode, path + path->p_depth);
3266 * ext4_split_extents() splits an extent and mark extent which is covered
3267 * by @map as split_flags indicates
3269 * It may result in splitting the extent into multiple extents (up to three)
3270 * There are three possibilities:
3271 * a> There is no split required
3272 * b> Splits in two extents: Split is happening at either end of the extent
3273 * c> Splits in three extents: Somone is splitting in middle of the extent
3276 static int ext4_split_extent(handle_t *handle,
3277 struct inode *inode,
3278 struct ext4_ext_path **ppath,
3279 struct ext4_map_blocks *map,
3283 struct ext4_ext_path *path = *ppath;
3284 ext4_lblk_t ee_block;
3285 struct ext4_extent *ex;
3286 unsigned int ee_len, depth;
3289 int split_flag1, flags1;
3290 int allocated = map->m_len;
3292 depth = ext_depth(inode);
3293 ex = path[depth].p_ext;
3294 ee_block = le32_to_cpu(ex->ee_block);
3295 ee_len = ext4_ext_get_actual_len(ex);
3296 unwritten = ext4_ext_is_unwritten(ex);
3298 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3299 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3300 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3302 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3303 EXT4_EXT_MARK_UNWRIT2;
3304 if (split_flag & EXT4_EXT_DATA_VALID2)
3305 split_flag1 |= EXT4_EXT_DATA_VALID1;
3306 err = ext4_split_extent_at(handle, inode, ppath,
3307 map->m_lblk + map->m_len, split_flag1, flags1);
3311 allocated = ee_len - (map->m_lblk - ee_block);
3314 * Update path is required because previous ext4_split_extent_at() may
3315 * result in split of original leaf or extent zeroout.
3317 path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3319 return PTR_ERR(path);
3320 depth = ext_depth(inode);
3321 ex = path[depth].p_ext;
3323 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3324 (unsigned long) map->m_lblk);
3325 return -EFSCORRUPTED;
3327 unwritten = ext4_ext_is_unwritten(ex);
3330 if (map->m_lblk >= ee_block) {
3331 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3333 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3334 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3335 EXT4_EXT_MARK_UNWRIT2);
3337 err = ext4_split_extent_at(handle, inode, ppath,
3338 map->m_lblk, split_flag1, flags);
3343 ext4_ext_show_leaf(inode, path);
3345 return err ? err : allocated;
3349 * This function is called by ext4_ext_map_blocks() if someone tries to write
3350 * to an unwritten extent. It may result in splitting the unwritten
3351 * extent into multiple extents (up to three - one initialized and two
3353 * There are three possibilities:
3354 * a> There is no split required: Entire extent should be initialized
3355 * b> Splits in two extents: Write is happening at either end of the extent
3356 * c> Splits in three extents: Somone is writing in middle of the extent
3359 * - The extent pointed to by 'path' is unwritten.
3360 * - The extent pointed to by 'path' contains a superset
3361 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3363 * Post-conditions on success:
3364 * - the returned value is the number of blocks beyond map->l_lblk
3365 * that are allocated and initialized.
3366 * It is guaranteed to be >= map->m_len.
3368 static int ext4_ext_convert_to_initialized(handle_t *handle,
3369 struct inode *inode,
3370 struct ext4_map_blocks *map,
3371 struct ext4_ext_path **ppath,
3374 struct ext4_ext_path *path = *ppath;
3375 struct ext4_sb_info *sbi;
3376 struct ext4_extent_header *eh;
3377 struct ext4_map_blocks split_map;
3378 struct ext4_extent zero_ex1, zero_ex2;
3379 struct ext4_extent *ex, *abut_ex;
3380 ext4_lblk_t ee_block, eof_block;
3381 unsigned int ee_len, depth, map_len = map->m_len;
3382 int allocated = 0, max_zeroout = 0;
3384 int split_flag = EXT4_EXT_DATA_VALID2;
3386 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3387 (unsigned long long)map->m_lblk, map_len);
3389 sbi = EXT4_SB(inode->i_sb);
3390 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3391 >> inode->i_sb->s_blocksize_bits;
3392 if (eof_block < map->m_lblk + map_len)
3393 eof_block = map->m_lblk + map_len;
3395 depth = ext_depth(inode);
3396 eh = path[depth].p_hdr;
3397 ex = path[depth].p_ext;
3398 ee_block = le32_to_cpu(ex->ee_block);
3399 ee_len = ext4_ext_get_actual_len(ex);
3400 zero_ex1.ee_len = 0;
3401 zero_ex2.ee_len = 0;
3403 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3405 /* Pre-conditions */
3406 BUG_ON(!ext4_ext_is_unwritten(ex));
3407 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3410 * Attempt to transfer newly initialized blocks from the currently
3411 * unwritten extent to its neighbor. This is much cheaper
3412 * than an insertion followed by a merge as those involve costly
3413 * memmove() calls. Transferring to the left is the common case in
3414 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3415 * followed by append writes.
3417 * Limitations of the current logic:
3418 * - L1: we do not deal with writes covering the whole extent.
3419 * This would require removing the extent if the transfer
3421 * - L2: we only attempt to merge with an extent stored in the
3422 * same extent tree node.
3424 if ((map->m_lblk == ee_block) &&
3425 /* See if we can merge left */
3426 (map_len < ee_len) && /*L1*/
3427 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3428 ext4_lblk_t prev_lblk;
3429 ext4_fsblk_t prev_pblk, ee_pblk;
3430 unsigned int prev_len;
3433 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3434 prev_len = ext4_ext_get_actual_len(abut_ex);
3435 prev_pblk = ext4_ext_pblock(abut_ex);
3436 ee_pblk = ext4_ext_pblock(ex);
3439 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3440 * upon those conditions:
3441 * - C1: abut_ex is initialized,
3442 * - C2: abut_ex is logically abutting ex,
3443 * - C3: abut_ex is physically abutting ex,
3444 * - C4: abut_ex can receive the additional blocks without
3445 * overflowing the (initialized) length limit.
3447 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3448 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3449 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3450 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3451 err = ext4_ext_get_access(handle, inode, path + depth);
3455 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3458 /* Shift the start of ex by 'map_len' blocks */
3459 ex->ee_block = cpu_to_le32(ee_block + map_len);
3460 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3461 ex->ee_len = cpu_to_le16(ee_len - map_len);
3462 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3464 /* Extend abut_ex by 'map_len' blocks */
3465 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3467 /* Result: number of initialized blocks past m_lblk */
3468 allocated = map_len;
3470 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3471 (map_len < ee_len) && /*L1*/
3472 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3473 /* See if we can merge right */
3474 ext4_lblk_t next_lblk;
3475 ext4_fsblk_t next_pblk, ee_pblk;
3476 unsigned int next_len;
3479 next_lblk = le32_to_cpu(abut_ex->ee_block);
3480 next_len = ext4_ext_get_actual_len(abut_ex);
3481 next_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 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3495 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3496 (next_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 abut_ex by 'map_len' blocks */
3505 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3506 ext4_ext_store_pblock(abut_ex, next_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(next_len + map_len);
3513 /* Result: number of initialized blocks past m_lblk */
3514 allocated = map_len;
3518 /* Mark the block containing both extents as dirty */
3519 err = ext4_ext_dirty(handle, inode, path + depth);
3521 /* Update path to point to the right extent */
3522 path[depth].p_ext = abut_ex;
3525 allocated = ee_len - (map->m_lblk - ee_block);
3527 WARN_ON(map->m_lblk < ee_block);
3529 * It is safe to convert extent to initialized via explicit
3530 * zeroout only if extent is fully inside i_size or new_size.
3532 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3534 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3535 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3536 (inode->i_sb->s_blocksize_bits - 10);
3540 * 1. split the extent into three extents.
3541 * 2. split the extent into two extents, zeroout the head of the first
3543 * 3. split the extent into two extents, zeroout the tail of the second
3545 * 4. split the extent into two extents with out zeroout.
3546 * 5. no splitting needed, just possibly zeroout the head and / or the
3547 * tail of the extent.
3549 split_map.m_lblk = map->m_lblk;
3550 split_map.m_len = map->m_len;
3552 if (max_zeroout && (allocated > split_map.m_len)) {
3553 if (allocated <= max_zeroout) {
3556 cpu_to_le32(split_map.m_lblk +
3559 cpu_to_le16(allocated - split_map.m_len);
3560 ext4_ext_store_pblock(&zero_ex1,
3561 ext4_ext_pblock(ex) + split_map.m_lblk +
3562 split_map.m_len - ee_block);
3563 err = ext4_ext_zeroout(inode, &zero_ex1);
3566 split_map.m_len = allocated;
3568 if (split_map.m_lblk - ee_block + split_map.m_len <
3571 if (split_map.m_lblk != ee_block) {
3572 zero_ex2.ee_block = ex->ee_block;
3573 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3575 ext4_ext_store_pblock(&zero_ex2,
3576 ext4_ext_pblock(ex));
3577 err = ext4_ext_zeroout(inode, &zero_ex2);
3582 split_map.m_len += split_map.m_lblk - ee_block;
3583 split_map.m_lblk = ee_block;
3584 allocated = map->m_len;
3588 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3593 /* If we have gotten a failure, don't zero out status tree */
3595 err = ext4_zeroout_es(inode, &zero_ex1);
3597 err = ext4_zeroout_es(inode, &zero_ex2);
3599 return err ? err : allocated;
3603 * This function is called by ext4_ext_map_blocks() from
3604 * ext4_get_blocks_dio_write() when DIO to write
3605 * to an unwritten extent.
3607 * Writing to an unwritten extent may result in splitting the unwritten
3608 * extent into multiple initialized/unwritten extents (up to three)
3609 * There are three possibilities:
3610 * a> There is no split required: Entire extent should be unwritten
3611 * b> Splits in two extents: Write is happening at either end of the extent
3612 * c> Splits in three extents: Somone is writing in middle of the extent
3614 * This works the same way in the case of initialized -> unwritten conversion.
3616 * One of more index blocks maybe needed if the extent tree grow after
3617 * the unwritten extent split. To prevent ENOSPC occur at the IO
3618 * complete, we need to split the unwritten extent before DIO submit
3619 * the IO. The unwritten extent called at this time will be split
3620 * into three unwritten extent(at most). After IO complete, the part
3621 * being filled will be convert to initialized by the end_io callback function
3622 * via ext4_convert_unwritten_extents().
3624 * Returns the size of unwritten extent to be written on success.
3626 static int ext4_split_convert_extents(handle_t *handle,
3627 struct inode *inode,
3628 struct ext4_map_blocks *map,
3629 struct ext4_ext_path **ppath,
3632 struct ext4_ext_path *path = *ppath;
3633 ext4_lblk_t eof_block;
3634 ext4_lblk_t ee_block;
3635 struct ext4_extent *ex;
3636 unsigned int ee_len;
3637 int split_flag = 0, depth;
3639 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3640 (unsigned long long)map->m_lblk, map->m_len);
3642 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3643 >> inode->i_sb->s_blocksize_bits;
3644 if (eof_block < map->m_lblk + map->m_len)
3645 eof_block = map->m_lblk + map->m_len;
3647 * It is safe to convert extent to initialized via explicit
3648 * zeroout only if extent is fully inside i_size or new_size.
3650 depth = ext_depth(inode);
3651 ex = path[depth].p_ext;
3652 ee_block = le32_to_cpu(ex->ee_block);
3653 ee_len = ext4_ext_get_actual_len(ex);
3655 /* Convert to unwritten */
3656 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3657 split_flag |= EXT4_EXT_DATA_VALID1;
3658 /* Convert to initialized */
3659 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3660 split_flag |= ee_block + ee_len <= eof_block ?
3661 EXT4_EXT_MAY_ZEROOUT : 0;
3662 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3664 flags |= EXT4_GET_BLOCKS_PRE_IO;
3665 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3668 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3669 struct inode *inode,
3670 struct ext4_map_blocks *map,
3671 struct ext4_ext_path **ppath)
3673 struct ext4_ext_path *path = *ppath;
3674 struct ext4_extent *ex;
3675 ext4_lblk_t ee_block;
3676 unsigned int ee_len;
3680 depth = ext_depth(inode);
3681 ex = path[depth].p_ext;
3682 ee_block = le32_to_cpu(ex->ee_block);
3683 ee_len = ext4_ext_get_actual_len(ex);
3685 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3686 (unsigned long long)ee_block, ee_len);
3688 /* If extent is larger than requested it is a clear sign that we still
3689 * have some extent state machine issues left. So extent_split is still
3691 * TODO: Once all related issues will be fixed this situation should be
3694 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3695 #ifdef CONFIG_EXT4_DEBUG
3696 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3697 " len %u; IO logical block %llu, len %u",
3698 inode->i_ino, (unsigned long long)ee_block, ee_len,
3699 (unsigned long long)map->m_lblk, map->m_len);
3701 err = ext4_split_convert_extents(handle, inode, map, ppath,
3702 EXT4_GET_BLOCKS_CONVERT);
3705 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3707 return PTR_ERR(path);
3708 depth = ext_depth(inode);
3709 ex = path[depth].p_ext;
3712 err = ext4_ext_get_access(handle, inode, path + depth);
3715 /* first mark the extent as initialized */
3716 ext4_ext_mark_initialized(ex);
3718 /* note: ext4_ext_correct_indexes() isn't needed here because
3719 * borders are not changed
3721 ext4_ext_try_to_merge(handle, inode, path, ex);
3723 /* Mark modified extent as dirty */
3724 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3726 ext4_ext_show_leaf(inode, path);
3727 ext4_fc_track_range(inode, ee_block, ee_block + ee_len - 1);
3732 convert_initialized_extent(handle_t *handle, struct inode *inode,
3733 struct ext4_map_blocks *map,
3734 struct ext4_ext_path **ppath,
3735 unsigned int *allocated)
3737 struct ext4_ext_path *path = *ppath;
3738 struct ext4_extent *ex;
3739 ext4_lblk_t ee_block;
3740 unsigned int ee_len;
3745 * Make sure that the extent is no bigger than we support with
3748 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3749 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3751 depth = ext_depth(inode);
3752 ex = path[depth].p_ext;
3753 ee_block = le32_to_cpu(ex->ee_block);
3754 ee_len = ext4_ext_get_actual_len(ex);
3756 ext_debug(inode, "logical block %llu, max_blocks %u\n",
3757 (unsigned long long)ee_block, ee_len);
3759 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3760 err = ext4_split_convert_extents(handle, inode, map, ppath,
3761 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3764 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3766 return PTR_ERR(path);
3767 depth = ext_depth(inode);
3768 ex = path[depth].p_ext;
3770 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3771 (unsigned long) map->m_lblk);
3772 return -EFSCORRUPTED;
3776 err = ext4_ext_get_access(handle, inode, path + depth);
3779 /* first mark the extent as unwritten */
3780 ext4_ext_mark_unwritten(ex);
3782 /* note: ext4_ext_correct_indexes() isn't needed here because
3783 * borders are not changed
3785 ext4_ext_try_to_merge(handle, inode, path, ex);
3787 /* Mark modified extent as dirty */
3788 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3791 ext4_ext_show_leaf(inode, path);
3793 ext4_update_inode_fsync_trans(handle, inode, 1);
3795 map->m_flags |= EXT4_MAP_UNWRITTEN;
3796 if (*allocated > map->m_len)
3797 *allocated = map->m_len;
3798 map->m_len = *allocated;
3799 ext4_fc_track_range(inode, ee_block, ee_block + ee_len - 1);
3804 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3805 struct ext4_map_blocks *map,
3806 struct ext4_ext_path **ppath, int flags,
3807 unsigned int allocated, ext4_fsblk_t newblock)
3809 struct ext4_ext_path __maybe_unused *path = *ppath;
3813 ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3814 (unsigned long long)map->m_lblk, map->m_len, flags,
3816 ext4_ext_show_leaf(inode, path);
3819 * When writing into unwritten space, we should not fail to
3820 * allocate metadata blocks for the new extent block if needed.
3822 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3824 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3825 allocated, newblock);
3827 /* get_block() before submitting IO, split the extent */
3828 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3829 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3830 flags | EXT4_GET_BLOCKS_CONVERT);
3836 * shouldn't get a 0 return when splitting an extent unless
3837 * m_len is 0 (bug) or extent has been corrupted
3839 if (unlikely(ret == 0)) {
3840 EXT4_ERROR_INODE(inode,
3841 "unexpected ret == 0, m_len = %u",
3843 err = -EFSCORRUPTED;
3846 map->m_flags |= EXT4_MAP_UNWRITTEN;
3849 /* IO end_io complete, convert the filled extent to written */
3850 if (flags & EXT4_GET_BLOCKS_CONVERT) {
3851 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3855 ext4_update_inode_fsync_trans(handle, inode, 1);
3858 /* buffered IO cases */
3860 * repeat fallocate creation request
3861 * we already have an unwritten extent
3863 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3864 map->m_flags |= EXT4_MAP_UNWRITTEN;
3868 /* buffered READ or buffered write_begin() lookup */
3869 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3871 * We have blocks reserved already. We
3872 * return allocated blocks so that delalloc
3873 * won't do block reservation for us. But
3874 * the buffer head will be unmapped so that
3875 * a read from the block returns 0s.
3877 map->m_flags |= EXT4_MAP_UNWRITTEN;
3882 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3883 * For buffered writes, at writepage time, etc. Convert a
3884 * discovered unwritten extent to written.
3886 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3891 ext4_update_inode_fsync_trans(handle, inode, 1);
3893 * shouldn't get a 0 return when converting an unwritten extent
3894 * unless m_len is 0 (bug) or extent has been corrupted
3896 if (unlikely(ret == 0)) {
3897 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3899 err = -EFSCORRUPTED;
3905 map->m_flags |= EXT4_MAP_NEW;
3907 map->m_flags |= EXT4_MAP_MAPPED;
3909 map->m_pblk = newblock;
3910 if (allocated > map->m_len)
3911 allocated = map->m_len;
3912 map->m_len = allocated;
3913 ext4_ext_show_leaf(inode, path);
3915 return err ? err : allocated;
3919 * get_implied_cluster_alloc - check to see if the requested
3920 * allocation (in the map structure) overlaps with a cluster already
3921 * allocated in an extent.
3922 * @sb The filesystem superblock structure
3923 * @map The requested lblk->pblk mapping
3924 * @ex The extent structure which might contain an implied
3925 * cluster allocation
3927 * This function is called by ext4_ext_map_blocks() after we failed to
3928 * find blocks that were already in the inode's extent tree. Hence,
3929 * we know that the beginning of the requested region cannot overlap
3930 * the extent from the inode's extent tree. There are three cases we
3931 * want to catch. The first is this case:
3933 * |--- cluster # N--|
3934 * |--- extent ---| |---- requested region ---|
3937 * The second case that we need to test for is this one:
3939 * |--------- cluster # N ----------------|
3940 * |--- requested region --| |------- extent ----|
3941 * |=======================|
3943 * The third case is when the requested region lies between two extents
3944 * within the same cluster:
3945 * |------------- cluster # N-------------|
3946 * |----- ex -----| |---- ex_right ----|
3947 * |------ requested region ------|
3948 * |================|
3950 * In each of the above cases, we need to set the map->m_pblk and
3951 * map->m_len so it corresponds to the return the extent labelled as
3952 * "|====|" from cluster #N, since it is already in use for data in
3953 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3954 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3955 * as a new "allocated" block region. Otherwise, we will return 0 and
3956 * ext4_ext_map_blocks() will then allocate one or more new clusters
3957 * by calling ext4_mb_new_blocks().
3959 static int get_implied_cluster_alloc(struct super_block *sb,
3960 struct ext4_map_blocks *map,
3961 struct ext4_extent *ex,
3962 struct ext4_ext_path *path)
3964 struct ext4_sb_info *sbi = EXT4_SB(sb);
3965 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3966 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3967 ext4_lblk_t rr_cluster_start;
3968 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3969 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3970 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3972 /* The extent passed in that we are trying to match */
3973 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3974 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3976 /* The requested region passed into ext4_map_blocks() */
3977 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3979 if ((rr_cluster_start == ex_cluster_end) ||
3980 (rr_cluster_start == ex_cluster_start)) {
3981 if (rr_cluster_start == ex_cluster_end)
3982 ee_start += ee_len - 1;
3983 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
3984 map->m_len = min(map->m_len,
3985 (unsigned) sbi->s_cluster_ratio - c_offset);
3987 * Check for and handle this case:
3989 * |--------- cluster # N-------------|
3990 * |------- extent ----|
3991 * |--- requested region ---|
3995 if (map->m_lblk < ee_block)
3996 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3999 * Check for the case where there is already another allocated
4000 * block to the right of 'ex' but before the end of the cluster.
4002 * |------------- cluster # N-------------|
4003 * |----- ex -----| |---- ex_right ----|
4004 * |------ requested region ------|
4005 * |================|
4007 if (map->m_lblk > ee_block) {
4008 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4009 map->m_len = min(map->m_len, next - map->m_lblk);
4012 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4016 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4022 * Block allocation/map/preallocation routine for extents based files
4025 * Need to be called with
4026 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4027 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4029 * return > 0, number of blocks already mapped/allocated
4030 * if create == 0 and these are pre-allocated blocks
4031 * buffer head is unmapped
4032 * otherwise blocks are mapped
4034 * return = 0, if plain look up failed (blocks have not been allocated)
4035 * buffer head is unmapped
4037 * return < 0, error case.
4039 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4040 struct ext4_map_blocks *map, int flags)
4042 struct ext4_ext_path *path = NULL;
4043 struct ext4_extent newex, *ex, ex2;
4044 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4045 ext4_fsblk_t newblock = 0, pblk;
4046 int err = 0, depth, ret;
4047 unsigned int allocated = 0, offset = 0;
4048 unsigned int allocated_clusters = 0;
4049 struct ext4_allocation_request ar;
4050 ext4_lblk_t cluster_offset;
4052 ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4053 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4055 /* find extent for this block */
4056 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4058 err = PTR_ERR(path);
4063 depth = ext_depth(inode);
4066 * consistent leaf must not be empty;
4067 * this situation is possible, though, _during_ tree modification;
4068 * this is why assert can't be put in ext4_find_extent()
4070 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4071 EXT4_ERROR_INODE(inode, "bad extent address "
4072 "lblock: %lu, depth: %d pblock %lld",
4073 (unsigned long) map->m_lblk, depth,
4074 path[depth].p_block);
4075 err = -EFSCORRUPTED;
4079 ex = path[depth].p_ext;
4081 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4082 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4083 unsigned short ee_len;
4087 * unwritten extents are treated as holes, except that
4088 * we split out initialized portions during a write.
4090 ee_len = ext4_ext_get_actual_len(ex);
4092 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4094 /* if found extent covers block, simply return it */
4095 if (in_range(map->m_lblk, ee_block, ee_len)) {
4096 newblock = map->m_lblk - ee_block + ee_start;
4097 /* number of remaining blocks in the extent */
4098 allocated = ee_len - (map->m_lblk - ee_block);
4099 ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4100 map->m_lblk, ee_block, ee_len, newblock);
4103 * If the extent is initialized check whether the
4104 * caller wants to convert it to unwritten.
4106 if ((!ext4_ext_is_unwritten(ex)) &&
4107 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4108 err = convert_initialized_extent(handle,
4109 inode, map, &path, &allocated);
4111 } else if (!ext4_ext_is_unwritten(ex)) {
4112 map->m_flags |= EXT4_MAP_MAPPED;
4113 map->m_pblk = newblock;
4114 if (allocated > map->m_len)
4115 allocated = map->m_len;
4116 map->m_len = allocated;
4117 ext4_ext_show_leaf(inode, path);
4121 ret = ext4_ext_handle_unwritten_extents(
4122 handle, inode, map, &path, flags,
4123 allocated, newblock);
4133 * requested block isn't allocated yet;
4134 * we couldn't try to create block if create flag is zero
4136 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4137 ext4_lblk_t hole_start, hole_len;
4139 hole_start = map->m_lblk;
4140 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4142 * put just found gap into cache to speed up
4143 * subsequent requests
4145 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4147 /* Update hole_len to reflect hole size after map->m_lblk */
4148 if (hole_start != map->m_lblk)
4149 hole_len -= map->m_lblk - hole_start;
4151 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4157 * Okay, we need to do block allocation.
4159 newex.ee_block = cpu_to_le32(map->m_lblk);
4160 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4163 * If we are doing bigalloc, check to see if the extent returned
4164 * by ext4_find_extent() implies a cluster we can use.
4166 if (cluster_offset && ex &&
4167 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4168 ar.len = allocated = map->m_len;
4169 newblock = map->m_pblk;
4170 goto got_allocated_blocks;
4173 /* find neighbour allocated blocks */
4174 ar.lleft = map->m_lblk;
4175 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4178 ar.lright = map->m_lblk;
4179 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4183 /* Check if the extent after searching to the right implies a
4184 * cluster we can use. */
4185 if ((sbi->s_cluster_ratio > 1) && err &&
4186 get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4187 ar.len = allocated = map->m_len;
4188 newblock = map->m_pblk;
4189 goto got_allocated_blocks;
4193 * See if request is beyond maximum number of blocks we can have in
4194 * a single extent. For an initialized extent this limit is
4195 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4196 * EXT_UNWRITTEN_MAX_LEN.
4198 if (map->m_len > EXT_INIT_MAX_LEN &&
4199 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4200 map->m_len = EXT_INIT_MAX_LEN;
4201 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4202 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4203 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4205 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4206 newex.ee_len = cpu_to_le16(map->m_len);
4207 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4209 allocated = ext4_ext_get_actual_len(&newex);
4211 allocated = map->m_len;
4213 /* allocate new block */
4215 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4216 ar.logical = map->m_lblk;
4218 * We calculate the offset from the beginning of the cluster
4219 * for the logical block number, since when we allocate a
4220 * physical cluster, the physical block should start at the
4221 * same offset from the beginning of the cluster. This is
4222 * needed so that future calls to get_implied_cluster_alloc()
4225 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4226 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4228 ar.logical -= offset;
4229 if (S_ISREG(inode->i_mode))
4230 ar.flags = EXT4_MB_HINT_DATA;
4232 /* disable in-core preallocation for non-regular files */
4234 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4235 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4236 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4237 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4238 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4239 ar.flags |= EXT4_MB_USE_RESERVED;
4240 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4243 allocated_clusters = ar.len;
4244 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4245 ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4246 ar.goal, newblock, ar.len, allocated);
4247 if (ar.len > allocated)
4250 got_allocated_blocks:
4251 /* try to insert new extent into found leaf and return */
4252 pblk = newblock + offset;
4253 ext4_ext_store_pblock(&newex, pblk);
4254 newex.ee_len = cpu_to_le16(ar.len);
4255 /* Mark unwritten */
4256 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4257 ext4_ext_mark_unwritten(&newex);
4258 map->m_flags |= EXT4_MAP_UNWRITTEN;
4261 err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4263 if (allocated_clusters) {
4267 * free data blocks we just allocated.
4268 * not a good idea to call discard here directly,
4269 * but otherwise we'd need to call it every free().
4271 ext4_discard_preallocations(inode, 0);
4272 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4273 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4274 ext4_free_blocks(handle, inode, NULL, newblock,
4275 EXT4_C2B(sbi, allocated_clusters),
4282 * Reduce the reserved cluster count to reflect successful deferred
4283 * allocation of delayed allocated clusters or direct allocation of
4284 * clusters discovered to be delayed allocated. Once allocated, a
4285 * cluster is not included in the reserved count.
4287 if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4288 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4290 * When allocating delayed allocated clusters, simply
4291 * reduce the reserved cluster count and claim quota
4293 ext4_da_update_reserve_space(inode, allocated_clusters,
4296 ext4_lblk_t lblk, len;
4300 * When allocating non-delayed allocated clusters
4301 * (from fallocate, filemap, DIO, or clusters
4302 * allocated when delalloc has been disabled by
4303 * ext4_nonda_switch), reduce the reserved cluster
4304 * count by the number of allocated clusters that
4305 * have previously been delayed allocated. Quota
4306 * has been claimed by ext4_mb_new_blocks() above,
4307 * so release the quota reservations made for any
4308 * previously delayed allocated clusters.
4310 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4311 len = allocated_clusters << sbi->s_cluster_bits;
4312 n = ext4_es_delayed_clu(inode, lblk, len);
4314 ext4_da_update_reserve_space(inode, (int) n, 0);
4319 * Cache the extent and update transaction to commit on fdatasync only
4320 * when it is _not_ an unwritten extent.
4322 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4323 ext4_update_inode_fsync_trans(handle, inode, 1);
4325 ext4_update_inode_fsync_trans(handle, inode, 0);
4327 map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4329 map->m_len = ar.len;
4330 allocated = map->m_len;
4331 ext4_ext_show_leaf(inode, path);
4332 ext4_fc_track_range(inode, map->m_lblk, map->m_lblk + map->m_len - 1);
4334 ext4_ext_drop_refs(path);
4337 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4338 err ? err : allocated);
4339 return err ? err : allocated;
4342 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4344 struct super_block *sb = inode->i_sb;
4345 ext4_lblk_t last_block;
4349 * TODO: optimization is possible here.
4350 * Probably we need not scan at all,
4351 * because page truncation is enough.
4354 /* we have to know where to truncate from in crash case */
4355 EXT4_I(inode)->i_disksize = inode->i_size;
4356 err = ext4_mark_inode_dirty(handle, inode);
4360 last_block = (inode->i_size + sb->s_blocksize - 1)
4361 >> EXT4_BLOCK_SIZE_BITS(sb);
4363 err = ext4_es_remove_extent(inode, last_block,
4364 EXT_MAX_BLOCKS - last_block);
4365 if (err == -ENOMEM) {
4367 congestion_wait(BLK_RW_ASYNC, HZ/50);
4373 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4374 if (err == -ENOMEM) {
4376 congestion_wait(BLK_RW_ASYNC, HZ/50);
4377 goto retry_remove_space;
4382 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4383 ext4_lblk_t len, loff_t new_size,
4386 struct inode *inode = file_inode(file);
4389 int ret2 = 0, ret3 = 0;
4392 struct ext4_map_blocks map;
4393 unsigned int credits;
4396 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4397 map.m_lblk = offset;
4400 * Don't normalize the request if it can fit in one extent so
4401 * that it doesn't get unnecessarily split into multiple
4404 if (len <= EXT_UNWRITTEN_MAX_LEN)
4405 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4408 * credits to insert 1 extent into extent tree
4410 credits = ext4_chunk_trans_blocks(inode, len);
4411 depth = ext_depth(inode);
4414 while (ret >= 0 && len) {
4416 * Recalculate credits when extent tree depth changes.
4418 if (depth != ext_depth(inode)) {
4419 credits = ext4_chunk_trans_blocks(inode, len);
4420 depth = ext_depth(inode);
4423 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4425 if (IS_ERR(handle)) {
4426 ret = PTR_ERR(handle);
4429 ret = ext4_map_blocks(handle, inode, &map, flags);
4431 ext4_debug("inode #%lu: block %u: len %u: "
4432 "ext4_ext_map_blocks returned %d",
4433 inode->i_ino, map.m_lblk,
4435 ext4_mark_inode_dirty(handle, inode);
4436 ret2 = ext4_journal_stop(handle);
4440 map.m_len = len = len - ret;
4441 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4442 inode->i_ctime = current_time(inode);
4444 if (epos > new_size)
4446 if (ext4_update_inode_size(inode, epos) & 0x1)
4447 inode->i_mtime = inode->i_ctime;
4449 ret2 = ext4_mark_inode_dirty(handle, inode);
4450 ext4_update_inode_fsync_trans(handle, inode, 1);
4451 ret3 = ext4_journal_stop(handle);
4452 ret2 = ret3 ? ret3 : ret2;
4456 if (ret == -ENOSPC &&
4457 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4462 return ret > 0 ? ret2 : ret;
4465 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4467 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4469 static long ext4_zero_range(struct file *file, loff_t offset,
4470 loff_t len, int mode)
4472 struct inode *inode = file_inode(file);
4473 handle_t *handle = NULL;
4474 unsigned int max_blocks;
4475 loff_t new_size = 0;
4479 int partial_begin, partial_end;
4482 unsigned int blkbits = inode->i_blkbits;
4484 trace_ext4_zero_range(inode, offset, len, mode);
4486 /* Call ext4_force_commit to flush all data in case of data=journal. */
4487 if (ext4_should_journal_data(inode)) {
4488 ret = ext4_force_commit(inode->i_sb);
4494 * Round up offset. This is not fallocate, we need to zero out
4495 * blocks, so convert interior block aligned part of the range to
4496 * unwritten and possibly manually zero out unaligned parts of the
4499 start = round_up(offset, 1 << blkbits);
4500 end = round_down((offset + len), 1 << blkbits);
4502 if (start < offset || end > offset + len)
4504 partial_begin = offset & ((1 << blkbits) - 1);
4505 partial_end = (offset + len) & ((1 << blkbits) - 1);
4507 lblk = start >> blkbits;
4508 max_blocks = (end >> blkbits);
4509 if (max_blocks < lblk)
4517 * Indirect files do not support unwritten extents
4519 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4524 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4525 (offset + len > inode->i_size ||
4526 offset + len > EXT4_I(inode)->i_disksize)) {
4527 new_size = offset + len;
4528 ret = inode_newsize_ok(inode, new_size);
4533 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4535 /* Wait all existing dio workers, newcomers will block on i_mutex */
4536 inode_dio_wait(inode);
4538 /* Preallocate the range including the unaligned edges */
4539 if (partial_begin || partial_end) {
4540 ret = ext4_alloc_file_blocks(file,
4541 round_down(offset, 1 << blkbits) >> blkbits,
4542 (round_up((offset + len), 1 << blkbits) -
4543 round_down(offset, 1 << blkbits)) >> blkbits,
4550 /* Zero range excluding the unaligned edges */
4551 if (max_blocks > 0) {
4552 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4556 * Prevent page faults from reinstantiating pages we have
4557 * released from page cache.
4559 down_write(&EXT4_I(inode)->i_mmap_sem);
4561 ret = ext4_break_layouts(inode);
4563 up_write(&EXT4_I(inode)->i_mmap_sem);
4567 ret = ext4_update_disksize_before_punch(inode, offset, len);
4569 up_write(&EXT4_I(inode)->i_mmap_sem);
4572 /* Now release the pages and zero block aligned part of pages */
4573 truncate_pagecache_range(inode, start, end - 1);
4574 inode->i_mtime = inode->i_ctime = current_time(inode);
4576 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4578 up_write(&EXT4_I(inode)->i_mmap_sem);
4582 if (!partial_begin && !partial_end)
4586 * In worst case we have to writeout two nonadjacent unwritten
4587 * blocks and update the inode
4589 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4590 if (ext4_should_journal_data(inode))
4592 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4593 if (IS_ERR(handle)) {
4594 ret = PTR_ERR(handle);
4595 ext4_std_error(inode->i_sb, ret);
4599 inode->i_mtime = inode->i_ctime = current_time(inode);
4601 ext4_update_inode_size(inode, new_size);
4602 ret = ext4_mark_inode_dirty(handle, inode);
4605 ext4_fc_track_range(inode, offset >> inode->i_sb->s_blocksize_bits,
4606 (offset + len - 1) >> inode->i_sb->s_blocksize_bits);
4607 /* Zero out partial block at the edges of the range */
4608 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4610 ext4_update_inode_fsync_trans(handle, inode, 1);
4612 if (file->f_flags & O_SYNC)
4613 ext4_handle_sync(handle);
4616 ext4_journal_stop(handle);
4618 inode_unlock(inode);
4623 * preallocate space for a file. This implements ext4's fallocate file
4624 * operation, which gets called from sys_fallocate system call.
4625 * For block-mapped files, posix_fallocate should fall back to the method
4626 * of writing zeroes to the required new blocks (the same behavior which is
4627 * expected for file systems which do not support fallocate() system call).
4629 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4631 struct inode *inode = file_inode(file);
4632 loff_t new_size = 0;
4633 unsigned int max_blocks;
4637 unsigned int blkbits = inode->i_blkbits;
4640 * Encrypted inodes can't handle collapse range or insert
4641 * range since we would need to re-encrypt blocks with a
4642 * different IV or XTS tweak (which are based on the logical
4645 if (IS_ENCRYPTED(inode) &&
4646 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4649 /* Return error if mode is not supported */
4650 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4651 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4652 FALLOC_FL_INSERT_RANGE))
4654 ext4_fc_track_range(inode, offset >> blkbits,
4655 (offset + len - 1) >> blkbits);
4657 ext4_fc_start_update(inode);
4659 if (mode & FALLOC_FL_PUNCH_HOLE) {
4660 ret = ext4_punch_hole(inode, offset, len);
4664 ret = ext4_convert_inline_data(inode);
4668 if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4669 ret = ext4_collapse_range(inode, offset, len);
4673 if (mode & FALLOC_FL_INSERT_RANGE) {
4674 ret = ext4_insert_range(inode, offset, len);
4678 if (mode & FALLOC_FL_ZERO_RANGE) {
4679 ret = ext4_zero_range(file, offset, len, mode);
4682 trace_ext4_fallocate_enter(inode, offset, len, mode);
4683 lblk = offset >> blkbits;
4685 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4686 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4691 * We only support preallocation for extent-based files only
4693 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4698 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4699 (offset + len > inode->i_size ||
4700 offset + len > EXT4_I(inode)->i_disksize)) {
4701 new_size = offset + len;
4702 ret = inode_newsize_ok(inode, new_size);
4707 /* Wait all existing dio workers, newcomers will block on i_mutex */
4708 inode_dio_wait(inode);
4710 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4714 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4715 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4716 EXT4_I(inode)->i_sync_tid);
4719 inode_unlock(inode);
4720 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4722 ext4_fc_stop_update(inode);
4727 * This function convert a range of blocks to written extents
4728 * The caller of this function will pass the start offset and the size.
4729 * all unwritten extents within this range will be converted to
4732 * This function is called from the direct IO end io call back
4733 * function, to convert the fallocated extents after IO is completed.
4734 * Returns 0 on success.
4736 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4737 loff_t offset, ssize_t len)
4739 unsigned int max_blocks;
4740 int ret = 0, ret2 = 0, ret3 = 0;
4741 struct ext4_map_blocks map;
4742 unsigned int blkbits = inode->i_blkbits;
4743 unsigned int credits = 0;
4745 map.m_lblk = offset >> blkbits;
4746 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4750 * credits to insert 1 extent into extent tree
4752 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4754 while (ret >= 0 && ret < max_blocks) {
4756 map.m_len = (max_blocks -= ret);
4758 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4760 if (IS_ERR(handle)) {
4761 ret = PTR_ERR(handle);
4765 ret = ext4_map_blocks(handle, inode, &map,
4766 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4768 ext4_warning(inode->i_sb,
4769 "inode #%lu: block %u: len %u: "
4770 "ext4_ext_map_blocks returned %d",
4771 inode->i_ino, map.m_lblk,
4773 ret2 = ext4_mark_inode_dirty(handle, inode);
4775 ret3 = ext4_journal_stop(handle);
4780 if (ret <= 0 || ret2)
4783 return ret > 0 ? ret2 : ret;
4786 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4788 int ret = 0, err = 0;
4789 struct ext4_io_end_vec *io_end_vec;
4792 * This is somewhat ugly but the idea is clear: When transaction is
4793 * reserved, everything goes into it. Otherwise we rather start several
4794 * smaller transactions for conversion of each extent separately.
4797 handle = ext4_journal_start_reserved(handle,
4798 EXT4_HT_EXT_CONVERT);
4800 return PTR_ERR(handle);
4803 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4804 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4812 err = ext4_journal_stop(handle);
4814 return ret < 0 ? ret : err;
4817 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4821 int blockbits = inode->i_sb->s_blocksize_bits;
4826 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4827 struct ext4_iloc iloc;
4828 int offset; /* offset of xattr in inode */
4830 error = ext4_get_inode_loc(inode, &iloc);
4833 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4834 offset = EXT4_GOOD_OLD_INODE_SIZE +
4835 EXT4_I(inode)->i_extra_isize;
4837 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4839 iomap_type = IOMAP_INLINE;
4840 } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4841 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4842 length = inode->i_sb->s_blocksize;
4843 iomap_type = IOMAP_MAPPED;
4845 /* no in-inode or external block for xattr, so return -ENOENT */
4850 iomap->addr = physical;
4852 iomap->length = length;
4853 iomap->type = iomap_type;
4859 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4860 loff_t length, unsigned flags,
4861 struct iomap *iomap, struct iomap *srcmap)
4865 error = ext4_iomap_xattr_fiemap(inode, iomap);
4866 if (error == 0 && (offset >= iomap->length))
4871 static const struct iomap_ops ext4_iomap_xattr_ops = {
4872 .iomap_begin = ext4_iomap_xattr_begin,
4875 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4879 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4880 maxbytes = inode->i_sb->s_maxbytes;
4882 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4886 if (start > maxbytes)
4890 * Shrink request scope to what the fs can actually handle.
4892 if (*len > maxbytes || (maxbytes - *len) < start)
4893 *len = maxbytes - start;
4897 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4902 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4903 error = ext4_ext_precache(inode);
4906 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4910 * For bitmap files the maximum size limit could be smaller than
4911 * s_maxbytes, so check len here manually instead of just relying on the
4914 error = ext4_fiemap_check_ranges(inode, start, &len);
4918 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4919 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4920 return iomap_fiemap(inode, fieinfo, start, len,
4921 &ext4_iomap_xattr_ops);
4924 return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4927 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4928 __u64 start, __u64 len)
4930 ext4_lblk_t start_blk, len_blks;
4934 if (ext4_has_inline_data(inode)) {
4937 down_read(&EXT4_I(inode)->xattr_sem);
4938 has_inline = ext4_has_inline_data(inode);
4939 up_read(&EXT4_I(inode)->xattr_sem);
4944 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4945 error = ext4_ext_precache(inode);
4948 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4951 error = fiemap_prep(inode, fieinfo, start, &len, 0);
4955 error = ext4_fiemap_check_ranges(inode, start, &len);
4959 start_blk = start >> inode->i_sb->s_blocksize_bits;
4960 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4961 if (last_blk >= EXT_MAX_BLOCKS)
4962 last_blk = EXT_MAX_BLOCKS-1;
4963 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4966 * Walk the extent tree gathering extent information
4967 * and pushing extents back to the user.
4969 return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
4974 * Function to access the path buffer for marking it dirty.
4975 * It also checks if there are sufficient credits left in the journal handle
4979 ext4_access_path(handle_t *handle, struct inode *inode,
4980 struct ext4_ext_path *path)
4984 if (!ext4_handle_valid(handle))
4988 * Check if need to extend journal credits
4989 * 3 for leaf, sb, and inode plus 2 (bmap and group
4990 * descriptor) for each block group; assume two block
4993 credits = ext4_writepage_trans_blocks(inode);
4994 err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
4998 err = ext4_ext_get_access(handle, inode, path);
5003 * ext4_ext_shift_path_extents:
5004 * Shift the extents of a path structure lying between path[depth].p_ext
5005 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5006 * if it is right shift or left shift operation.
5009 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5010 struct inode *inode, handle_t *handle,
5011 enum SHIFT_DIRECTION SHIFT)
5014 struct ext4_extent *ex_start, *ex_last;
5015 bool update = false;
5016 depth = path->p_depth;
5018 while (depth >= 0) {
5019 if (depth == path->p_depth) {
5020 ex_start = path[depth].p_ext;
5022 return -EFSCORRUPTED;
5024 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5026 err = ext4_access_path(handle, inode, path + depth);
5030 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5033 while (ex_start <= ex_last) {
5034 if (SHIFT == SHIFT_LEFT) {
5035 le32_add_cpu(&ex_start->ee_block,
5037 /* Try to merge to the left. */
5039 EXT_FIRST_EXTENT(path[depth].p_hdr))
5041 ext4_ext_try_to_merge_right(inode,
5042 path, ex_start - 1))
5047 le32_add_cpu(&ex_last->ee_block, shift);
5048 ext4_ext_try_to_merge_right(inode, path,
5053 err = ext4_ext_dirty(handle, inode, path + depth);
5057 if (--depth < 0 || !update)
5061 /* Update index too */
5062 err = ext4_access_path(handle, inode, path + depth);
5066 if (SHIFT == SHIFT_LEFT)
5067 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5069 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5070 err = ext4_ext_dirty(handle, inode, path + depth);
5074 /* we are done if current index is not a starting index */
5075 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5086 * ext4_ext_shift_extents:
5087 * All the extents which lies in the range from @start to the last allocated
5088 * block for the @inode are shifted either towards left or right (depending
5089 * upon @SHIFT) by @shift blocks.
5090 * On success, 0 is returned, error otherwise.
5093 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5094 ext4_lblk_t start, ext4_lblk_t shift,
5095 enum SHIFT_DIRECTION SHIFT)
5097 struct ext4_ext_path *path;
5099 struct ext4_extent *extent;
5100 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5102 /* Let path point to the last extent */
5103 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5106 return PTR_ERR(path);
5108 depth = path->p_depth;
5109 extent = path[depth].p_ext;
5113 stop = le32_to_cpu(extent->ee_block);
5116 * For left shifts, make sure the hole on the left is big enough to
5117 * accommodate the shift. For right shifts, make sure the last extent
5118 * won't be shifted beyond EXT_MAX_BLOCKS.
5120 if (SHIFT == SHIFT_LEFT) {
5121 path = ext4_find_extent(inode, start - 1, &path,
5124 return PTR_ERR(path);
5125 depth = path->p_depth;
5126 extent = path[depth].p_ext;
5128 ex_start = le32_to_cpu(extent->ee_block);
5129 ex_end = le32_to_cpu(extent->ee_block) +
5130 ext4_ext_get_actual_len(extent);
5136 if ((start == ex_start && shift > ex_start) ||
5137 (shift > start - ex_end)) {
5142 if (shift > EXT_MAX_BLOCKS -
5143 (stop + ext4_ext_get_actual_len(extent))) {
5150 * In case of left shift, iterator points to start and it is increased
5151 * till we reach stop. In case of right shift, iterator points to stop
5152 * and it is decreased till we reach start.
5154 if (SHIFT == SHIFT_LEFT)
5160 * Its safe to start updating extents. Start and stop are unsigned, so
5161 * in case of right shift if extent with 0 block is reached, iterator
5162 * becomes NULL to indicate the end of the loop.
5164 while (iterator && start <= stop) {
5165 path = ext4_find_extent(inode, *iterator, &path,
5168 return PTR_ERR(path);
5169 depth = path->p_depth;
5170 extent = path[depth].p_ext;
5172 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5173 (unsigned long) *iterator);
5174 return -EFSCORRUPTED;
5176 if (SHIFT == SHIFT_LEFT && *iterator >
5177 le32_to_cpu(extent->ee_block)) {
5178 /* Hole, move to the next extent */
5179 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5180 path[depth].p_ext++;
5182 *iterator = ext4_ext_next_allocated_block(path);
5187 if (SHIFT == SHIFT_LEFT) {
5188 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5189 *iterator = le32_to_cpu(extent->ee_block) +
5190 ext4_ext_get_actual_len(extent);
5192 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5193 if (le32_to_cpu(extent->ee_block) > 0)
5194 *iterator = le32_to_cpu(extent->ee_block) - 1;
5196 /* Beginning is reached, end of the loop */
5198 /* Update path extent in case we need to stop */
5199 while (le32_to_cpu(extent->ee_block) < start)
5201 path[depth].p_ext = extent;
5203 ret = ext4_ext_shift_path_extents(path, shift, inode,
5209 ext4_ext_drop_refs(path);
5215 * ext4_collapse_range:
5216 * This implements the fallocate's collapse range functionality for ext4
5217 * Returns: 0 and non-zero on error.
5219 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5221 struct super_block *sb = inode->i_sb;
5222 ext4_lblk_t punch_start, punch_stop;
5224 unsigned int credits;
5225 loff_t new_size, ioffset;
5229 * We need to test this early because xfstests assumes that a
5230 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5231 * system does not support collapse range.
5233 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5236 /* Collapse range works only on fs cluster size aligned regions. */
5237 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5240 trace_ext4_collapse_range(inode, offset, len);
5242 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5243 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5245 /* Call ext4_force_commit to flush all data in case of data=journal. */
5246 if (ext4_should_journal_data(inode)) {
5247 ret = ext4_force_commit(inode->i_sb);
5254 * There is no need to overlap collapse range with EOF, in which case
5255 * it is effectively a truncate operation
5257 if (offset + len >= inode->i_size) {
5262 /* Currently just for extent based files */
5263 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5268 /* Wait for existing dio to complete */
5269 inode_dio_wait(inode);
5272 * Prevent page faults from reinstantiating pages we have released from
5275 down_write(&EXT4_I(inode)->i_mmap_sem);
5277 ret = ext4_break_layouts(inode);
5282 * Need to round down offset to be aligned with page size boundary
5283 * for page size > block size.
5285 ioffset = round_down(offset, PAGE_SIZE);
5287 * Write tail of the last page before removed range since it will get
5288 * removed from the page cache below.
5290 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5294 * Write data that will be shifted to preserve them when discarding
5295 * page cache below. We are also protected from pages becoming dirty
5298 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5302 truncate_pagecache(inode, ioffset);
5304 credits = ext4_writepage_trans_blocks(inode);
5305 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5306 if (IS_ERR(handle)) {
5307 ret = PTR_ERR(handle);
5310 ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5312 down_write(&EXT4_I(inode)->i_data_sem);
5313 ext4_discard_preallocations(inode, 0);
5315 ret = ext4_es_remove_extent(inode, punch_start,
5316 EXT_MAX_BLOCKS - punch_start);
5318 up_write(&EXT4_I(inode)->i_data_sem);
5322 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5324 up_write(&EXT4_I(inode)->i_data_sem);
5327 ext4_discard_preallocations(inode, 0);
5329 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5330 punch_stop - punch_start, SHIFT_LEFT);
5332 up_write(&EXT4_I(inode)->i_data_sem);
5336 new_size = inode->i_size - len;
5337 i_size_write(inode, new_size);
5338 EXT4_I(inode)->i_disksize = new_size;
5340 up_write(&EXT4_I(inode)->i_data_sem);
5342 ext4_handle_sync(handle);
5343 inode->i_mtime = inode->i_ctime = current_time(inode);
5344 ret = ext4_mark_inode_dirty(handle, inode);
5345 ext4_update_inode_fsync_trans(handle, inode, 1);
5348 ext4_journal_stop(handle);
5349 ext4_fc_stop_ineligible(sb);
5351 up_write(&EXT4_I(inode)->i_mmap_sem);
5353 inode_unlock(inode);
5358 * ext4_insert_range:
5359 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5360 * The data blocks starting from @offset to the EOF are shifted by @len
5361 * towards right to create a hole in the @inode. Inode size is increased
5363 * Returns 0 on success, error otherwise.
5365 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5367 struct super_block *sb = inode->i_sb;
5369 struct ext4_ext_path *path;
5370 struct ext4_extent *extent;
5371 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5372 unsigned int credits, ee_len;
5373 int ret = 0, depth, split_flag = 0;
5377 * We need to test this early because xfstests assumes that an
5378 * insert range of (0, 1) will return EOPNOTSUPP if the file
5379 * system does not support insert range.
5381 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5384 /* Insert range works only on fs cluster size aligned regions. */
5385 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5388 trace_ext4_insert_range(inode, offset, len);
5390 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5391 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5393 /* Call ext4_force_commit to flush all data in case of data=journal */
5394 if (ext4_should_journal_data(inode)) {
5395 ret = ext4_force_commit(inode->i_sb);
5401 /* Currently just for extent based files */
5402 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5407 /* Check whether the maximum file size would be exceeded */
5408 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5413 /* Offset must be less than i_size */
5414 if (offset >= inode->i_size) {
5419 /* Wait for existing dio to complete */
5420 inode_dio_wait(inode);
5423 * Prevent page faults from reinstantiating pages we have released from
5426 down_write(&EXT4_I(inode)->i_mmap_sem);
5428 ret = ext4_break_layouts(inode);
5433 * Need to round down to align start offset to page size boundary
5434 * for page size > block size.
5436 ioffset = round_down(offset, PAGE_SIZE);
5437 /* Write out all dirty pages */
5438 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5442 truncate_pagecache(inode, ioffset);
5444 credits = ext4_writepage_trans_blocks(inode);
5445 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5446 if (IS_ERR(handle)) {
5447 ret = PTR_ERR(handle);
5450 ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5452 /* Expand file to avoid data loss if there is error while shifting */
5453 inode->i_size += len;
5454 EXT4_I(inode)->i_disksize += len;
5455 inode->i_mtime = inode->i_ctime = current_time(inode);
5456 ret = ext4_mark_inode_dirty(handle, inode);
5460 down_write(&EXT4_I(inode)->i_data_sem);
5461 ext4_discard_preallocations(inode, 0);
5463 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5465 up_write(&EXT4_I(inode)->i_data_sem);
5469 depth = ext_depth(inode);
5470 extent = path[depth].p_ext;
5472 ee_start_lblk = le32_to_cpu(extent->ee_block);
5473 ee_len = ext4_ext_get_actual_len(extent);
5476 * If offset_lblk is not the starting block of extent, split
5477 * the extent @offset_lblk
5479 if ((offset_lblk > ee_start_lblk) &&
5480 (offset_lblk < (ee_start_lblk + ee_len))) {
5481 if (ext4_ext_is_unwritten(extent))
5482 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5483 EXT4_EXT_MARK_UNWRIT2;
5484 ret = ext4_split_extent_at(handle, inode, &path,
5485 offset_lblk, split_flag,
5487 EXT4_GET_BLOCKS_PRE_IO |
5488 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5491 ext4_ext_drop_refs(path);
5494 up_write(&EXT4_I(inode)->i_data_sem);
5498 ext4_ext_drop_refs(path);
5502 ret = ext4_es_remove_extent(inode, offset_lblk,
5503 EXT_MAX_BLOCKS - offset_lblk);
5505 up_write(&EXT4_I(inode)->i_data_sem);
5510 * if offset_lblk lies in a hole which is at start of file, use
5511 * ee_start_lblk to shift extents
5513 ret = ext4_ext_shift_extents(inode, handle,
5514 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5515 len_lblk, SHIFT_RIGHT);
5517 up_write(&EXT4_I(inode)->i_data_sem);
5519 ext4_handle_sync(handle);
5521 ext4_update_inode_fsync_trans(handle, inode, 1);
5524 ext4_journal_stop(handle);
5525 ext4_fc_stop_ineligible(sb);
5527 up_write(&EXT4_I(inode)->i_mmap_sem);
5529 inode_unlock(inode);
5534 * ext4_swap_extents() - Swap extents between two inodes
5535 * @handle: handle for this transaction
5536 * @inode1: First inode
5537 * @inode2: Second inode
5538 * @lblk1: Start block for first inode
5539 * @lblk2: Start block for second inode
5540 * @count: Number of blocks to swap
5541 * @unwritten: Mark second inode's extents as unwritten after swap
5542 * @erp: Pointer to save error value
5544 * This helper routine does exactly what is promise "swap extents". All other
5545 * stuff such as page-cache locking consistency, bh mapping consistency or
5546 * extent's data copying must be performed by caller.
5548 * i_mutex is held for both inodes
5549 * i_data_sem is locked for write for both inodes
5551 * All pages from requested range are locked for both inodes
5554 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5555 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5556 ext4_lblk_t count, int unwritten, int *erp)
5558 struct ext4_ext_path *path1 = NULL;
5559 struct ext4_ext_path *path2 = NULL;
5560 int replaced_count = 0;
5562 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5563 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5564 BUG_ON(!inode_is_locked(inode1));
5565 BUG_ON(!inode_is_locked(inode2));
5567 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5570 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5575 struct ext4_extent *ex1, *ex2, tmp_ex;
5576 ext4_lblk_t e1_blk, e2_blk;
5577 int e1_len, e2_len, len;
5580 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5581 if (IS_ERR(path1)) {
5582 *erp = PTR_ERR(path1);
5588 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5589 if (IS_ERR(path2)) {
5590 *erp = PTR_ERR(path2);
5594 ex1 = path1[path1->p_depth].p_ext;
5595 ex2 = path2[path2->p_depth].p_ext;
5596 /* Do we have something to swap ? */
5597 if (unlikely(!ex2 || !ex1))
5600 e1_blk = le32_to_cpu(ex1->ee_block);
5601 e2_blk = le32_to_cpu(ex2->ee_block);
5602 e1_len = ext4_ext_get_actual_len(ex1);
5603 e2_len = ext4_ext_get_actual_len(ex2);
5606 if (!in_range(lblk1, e1_blk, e1_len) ||
5607 !in_range(lblk2, e2_blk, e2_len)) {
5608 ext4_lblk_t next1, next2;
5610 /* if hole after extent, then go to next extent */
5611 next1 = ext4_ext_next_allocated_block(path1);
5612 next2 = ext4_ext_next_allocated_block(path2);
5613 /* If hole before extent, then shift to that extent */
5618 /* Do we have something to swap */
5619 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5621 /* Move to the rightest boundary */
5622 len = next1 - lblk1;
5623 if (len < next2 - lblk2)
5624 len = next2 - lblk2;
5633 /* Prepare left boundary */
5634 if (e1_blk < lblk1) {
5636 *erp = ext4_force_split_extent_at(handle, inode1,
5641 if (e2_blk < lblk2) {
5643 *erp = ext4_force_split_extent_at(handle, inode2,
5648 /* ext4_split_extent_at() may result in leaf extent split,
5649 * path must to be revalidated. */
5653 /* Prepare right boundary */
5655 if (len > e1_blk + e1_len - lblk1)
5656 len = e1_blk + e1_len - lblk1;
5657 if (len > e2_blk + e2_len - lblk2)
5658 len = e2_blk + e2_len - lblk2;
5660 if (len != e1_len) {
5662 *erp = ext4_force_split_extent_at(handle, inode1,
5663 &path1, lblk1 + len, 0);
5667 if (len != e2_len) {
5669 *erp = ext4_force_split_extent_at(handle, inode2,
5670 &path2, lblk2 + len, 0);
5674 /* ext4_split_extent_at() may result in leaf extent split,
5675 * path must to be revalidated. */
5679 BUG_ON(e2_len != e1_len);
5680 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5683 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5687 /* Both extents are fully inside boundaries. Swap it now */
5689 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5690 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5691 ex1->ee_len = cpu_to_le16(e2_len);
5692 ex2->ee_len = cpu_to_le16(e1_len);
5694 ext4_ext_mark_unwritten(ex2);
5695 if (ext4_ext_is_unwritten(&tmp_ex))
5696 ext4_ext_mark_unwritten(ex1);
5698 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5699 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5700 *erp = ext4_ext_dirty(handle, inode2, path2 +
5704 *erp = ext4_ext_dirty(handle, inode1, path1 +
5707 * Looks scarry ah..? second inode already points to new blocks,
5708 * and it was successfully dirtied. But luckily error may happen
5709 * only due to journal error, so full transaction will be
5716 replaced_count += len;
5720 ext4_ext_drop_refs(path1);
5722 ext4_ext_drop_refs(path2);
5724 path1 = path2 = NULL;
5726 return replaced_count;
5730 * ext4_clu_mapped - determine whether any block in a logical cluster has
5731 * been mapped to a physical cluster
5733 * @inode - file containing the logical cluster
5734 * @lclu - logical cluster of interest
5736 * Returns 1 if any block in the logical cluster is mapped, signifying
5737 * that a physical cluster has been allocated for it. Otherwise,
5738 * returns 0. Can also return negative error codes. Derived from
5739 * ext4_ext_map_blocks().
5741 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5743 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5744 struct ext4_ext_path *path;
5745 int depth, mapped = 0, err = 0;
5746 struct ext4_extent *extent;
5747 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5749 /* search for the extent closest to the first block in the cluster */
5750 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5752 err = PTR_ERR(path);
5757 depth = ext_depth(inode);
5760 * A consistent leaf must not be empty. This situation is possible,
5761 * though, _during_ tree modification, and it's why an assert can't
5762 * be put in ext4_find_extent().
5764 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5765 EXT4_ERROR_INODE(inode,
5766 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5767 (unsigned long) EXT4_C2B(sbi, lclu),
5768 depth, path[depth].p_block);
5769 err = -EFSCORRUPTED;
5773 extent = path[depth].p_ext;
5775 /* can't be mapped if the extent tree is empty */
5779 first_lblk = le32_to_cpu(extent->ee_block);
5780 first_lclu = EXT4_B2C(sbi, first_lblk);
5783 * Three possible outcomes at this point - found extent spanning
5784 * the target cluster, to the left of the target cluster, or to the
5785 * right of the target cluster. The first two cases are handled here.
5786 * The last case indicates the target cluster is not mapped.
5788 if (lclu >= first_lclu) {
5789 last_lclu = EXT4_B2C(sbi, first_lblk +
5790 ext4_ext_get_actual_len(extent) - 1);
5791 if (lclu <= last_lclu) {
5794 first_lblk = ext4_ext_next_allocated_block(path);
5795 first_lclu = EXT4_B2C(sbi, first_lblk);
5796 if (lclu == first_lclu)
5802 ext4_ext_drop_refs(path);
5805 return err ? err : mapped;
5809 * Updates physical block address and unwritten status of extent
5810 * starting at lblk start and of len. If such an extent doesn't exist,
5811 * this function splits the extent tree appropriately to create an
5812 * extent like this. This function is called in the fast commit
5813 * replay path. Returns 0 on success and error on failure.
5815 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5816 int len, int unwritten, ext4_fsblk_t pblk)
5818 struct ext4_ext_path *path = NULL, *ppath;
5819 struct ext4_extent *ex;
5822 path = ext4_find_extent(inode, start, NULL, 0);
5825 ex = path[path->p_depth].p_ext;
5827 ret = -EFSCORRUPTED;
5831 if (le32_to_cpu(ex->ee_block) != start ||
5832 ext4_ext_get_actual_len(ex) != len) {
5833 /* We need to split this extent to match our extent first */
5835 down_write(&EXT4_I(inode)->i_data_sem);
5836 ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5837 up_write(&EXT4_I(inode)->i_data_sem);
5841 path = ext4_find_extent(inode, start, NULL, 0);
5845 ex = path[path->p_depth].p_ext;
5846 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5847 if (ext4_ext_get_actual_len(ex) != len) {
5848 down_write(&EXT4_I(inode)->i_data_sem);
5849 ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5851 up_write(&EXT4_I(inode)->i_data_sem);
5855 path = ext4_find_extent(inode, start, NULL, 0);
5858 ex = path[path->p_depth].p_ext;
5862 ext4_ext_mark_unwritten(ex);
5864 ext4_ext_mark_initialized(ex);
5865 ext4_ext_store_pblock(ex, pblk);
5866 down_write(&EXT4_I(inode)->i_data_sem);
5867 ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5868 up_write(&EXT4_I(inode)->i_data_sem);
5870 ext4_ext_drop_refs(path);
5872 ext4_mark_inode_dirty(NULL, inode);
5876 /* Try to shrink the extent tree */
5877 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5879 struct ext4_ext_path *path = NULL;
5880 struct ext4_extent *ex;
5881 ext4_lblk_t old_cur, cur = 0;
5884 path = ext4_find_extent(inode, cur, NULL, 0);
5887 ex = path[path->p_depth].p_ext;
5889 ext4_ext_drop_refs(path);
5891 ext4_mark_inode_dirty(NULL, inode);
5895 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5898 ext4_ext_try_to_merge(NULL, inode, path, ex);
5899 down_write(&EXT4_I(inode)->i_data_sem);
5900 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5901 up_write(&EXT4_I(inode)->i_data_sem);
5902 ext4_mark_inode_dirty(NULL, inode);
5903 ext4_ext_drop_refs(path);
5908 /* Check if *cur is a hole and if it is, skip it */
5909 static void skip_hole(struct inode *inode, ext4_lblk_t *cur)
5912 struct ext4_map_blocks map;
5915 map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5917 ret = ext4_map_blocks(NULL, inode, &map, 0);
5920 *cur = *cur + map.m_len;
5923 /* Count number of blocks used by this inode and update i_blocks */
5924 int ext4_ext_replay_set_iblocks(struct inode *inode)
5926 struct ext4_ext_path *path = NULL, *path2 = NULL;
5927 struct ext4_extent *ex;
5928 ext4_lblk_t cur = 0, end;
5929 int numblks = 0, i, ret = 0;
5930 ext4_fsblk_t cmp1, cmp2;
5931 struct ext4_map_blocks map;
5933 /* Determin the size of the file first */
5934 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5937 return PTR_ERR(path);
5938 ex = path[path->p_depth].p_ext;
5940 ext4_ext_drop_refs(path);
5944 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5945 ext4_ext_drop_refs(path);
5948 /* Count the number of data blocks */
5952 map.m_len = end - cur;
5953 ret = ext4_map_blocks(NULL, inode, &map, 0);
5958 cur = cur + map.m_len;
5962 * Count the number of extent tree blocks. We do it by looking up
5963 * two successive extents and determining the difference between
5964 * their paths. When path is different for 2 successive extents
5965 * we compare the blocks in the path at each level and increment
5966 * iblocks by total number of differences found.
5969 skip_hole(inode, &cur);
5970 path = ext4_find_extent(inode, cur, NULL, 0);
5973 numblks += path->p_depth;
5974 ext4_ext_drop_refs(path);
5977 path = ext4_find_extent(inode, cur, NULL, 0);
5980 ex = path[path->p_depth].p_ext;
5982 ext4_ext_drop_refs(path);
5986 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
5987 ext4_ext_get_actual_len(ex));
5988 skip_hole(inode, &cur);
5990 path2 = ext4_find_extent(inode, cur, NULL, 0);
5991 if (IS_ERR(path2)) {
5992 ext4_ext_drop_refs(path);
5996 ex = path2[path2->p_depth].p_ext;
5997 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
5999 if (i <= path->p_depth)
6000 cmp1 = path[i].p_bh ?
6001 path[i].p_bh->b_blocknr : 0;
6002 if (i <= path2->p_depth)
6003 cmp2 = path2[i].p_bh ?
6004 path2[i].p_bh->b_blocknr : 0;
6005 if (cmp1 != cmp2 && cmp2 != 0)
6008 ext4_ext_drop_refs(path);
6009 ext4_ext_drop_refs(path2);
6015 inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6016 ext4_mark_inode_dirty(NULL, inode);
6020 int ext4_ext_clear_bb(struct inode *inode)
6022 struct ext4_ext_path *path = NULL;
6023 struct ext4_extent *ex;
6024 ext4_lblk_t cur = 0, end;
6026 struct ext4_map_blocks map;
6028 /* Determin the size of the file first */
6029 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6032 return PTR_ERR(path);
6033 ex = path[path->p_depth].p_ext;
6035 ext4_ext_drop_refs(path);
6039 end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6040 ext4_ext_drop_refs(path);
6046 map.m_len = end - cur;
6047 ret = ext4_map_blocks(NULL, inode, &map, 0);
6051 path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6052 if (!IS_ERR_OR_NULL(path)) {
6053 for (j = 0; j < path->p_depth; j++) {
6055 ext4_mb_mark_bb(inode->i_sb,
6056 path[j].p_block, 1, 0);
6058 ext4_ext_drop_refs(path);
6061 ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6063 cur = cur + map.m_len;