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 "ext4_jbd2.h"
32 #include "ext4_extents.h"
35 #include <trace/events/ext4.h>
38 * used by extent splitting.
40 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
42 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
43 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
45 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
46 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
48 static __le32 ext4_extent_block_csum(struct inode *inode,
49 struct ext4_extent_header *eh)
51 struct ext4_inode_info *ei = EXT4_I(inode);
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
55 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
56 EXT4_EXTENT_TAIL_OFFSET(eh));
57 return cpu_to_le32(csum);
60 static int ext4_extent_block_csum_verify(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_extent_tail *et;
65 if (!ext4_has_metadata_csum(inode->i_sb))
68 et = find_ext4_extent_tail(eh);
69 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
74 static void ext4_extent_block_csum_set(struct inode *inode,
75 struct ext4_extent_header *eh)
77 struct ext4_extent_tail *et;
79 if (!ext4_has_metadata_csum(inode->i_sb))
82 et = find_ext4_extent_tail(eh);
83 et->et_checksum = ext4_extent_block_csum(inode, eh);
86 static int ext4_split_extent(handle_t *handle,
88 struct ext4_ext_path **ppath,
89 struct ext4_map_blocks *map,
93 static int ext4_split_extent_at(handle_t *handle,
95 struct ext4_ext_path **ppath,
100 static int ext4_find_delayed_extent(struct inode *inode,
101 struct extent_status *newes);
103 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
106 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
107 * moment, get_block can be called only for blocks inside i_size since
108 * page cache has been already dropped and writes are blocked by
109 * i_mutex. So we can safely drop the i_data_sem here.
111 BUG_ON(EXT4_JOURNAL(inode) == NULL);
112 ext4_discard_preallocations(inode);
113 up_write(&EXT4_I(inode)->i_data_sem);
119 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
120 * transaction with 'restart_cred' credits. The function drops i_data_sem
121 * when restarting transaction and gets it after transaction is restarted.
123 * The function returns 0 on success, 1 if transaction had to be restarted,
124 * and < 0 in case of fatal error.
126 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
127 int check_cred, int restart_cred,
133 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
134 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
136 down_write(&EXT4_I(inode)->i_data_sem);
145 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
146 struct ext4_ext_path *path)
149 /* path points to block */
150 BUFFER_TRACE(path->p_bh, "get_write_access");
151 return ext4_journal_get_write_access(handle, path->p_bh);
153 /* path points to leaf/index in inode body */
154 /* we use in-core data, no need to protect them */
164 static int __ext4_ext_dirty(const char *where, unsigned int line,
165 handle_t *handle, struct inode *inode,
166 struct ext4_ext_path *path)
170 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
172 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
173 /* path points to block */
174 err = __ext4_handle_dirty_metadata(where, line, handle,
177 /* path points to leaf/index in inode body */
178 err = ext4_mark_inode_dirty(handle, inode);
183 #define ext4_ext_dirty(handle, inode, path) \
184 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
186 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
187 struct ext4_ext_path *path,
191 int depth = path->p_depth;
192 struct ext4_extent *ex;
195 * Try to predict block placement assuming that we are
196 * filling in a file which will eventually be
197 * non-sparse --- i.e., in the case of libbfd writing
198 * an ELF object sections out-of-order but in a way
199 * the eventually results in a contiguous object or
200 * executable file, or some database extending a table
201 * space file. However, this is actually somewhat
202 * non-ideal if we are writing a sparse file such as
203 * qemu or KVM writing a raw image file that is going
204 * to stay fairly sparse, since it will end up
205 * fragmenting the file system's free space. Maybe we
206 * should have some hueristics or some way to allow
207 * userspace to pass a hint to file system,
208 * especially if the latter case turns out to be
211 ex = path[depth].p_ext;
213 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
214 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
216 if (block > ext_block)
217 return ext_pblk + (block - ext_block);
219 return ext_pblk - (ext_block - block);
222 /* it looks like index is empty;
223 * try to find starting block from index itself */
224 if (path[depth].p_bh)
225 return path[depth].p_bh->b_blocknr;
228 /* OK. use inode's group */
229 return ext4_inode_to_goal_block(inode);
233 * Allocation for a meta data block
236 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
237 struct ext4_ext_path *path,
238 struct ext4_extent *ex, int *err, unsigned int flags)
240 ext4_fsblk_t goal, newblock;
242 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
243 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
248 static inline int ext4_ext_space_block(struct inode *inode, int check)
252 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
253 / sizeof(struct ext4_extent);
254 #ifdef AGGRESSIVE_TEST
255 if (!check && size > 6)
261 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
265 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
266 / sizeof(struct ext4_extent_idx);
267 #ifdef AGGRESSIVE_TEST
268 if (!check && size > 5)
274 static inline int ext4_ext_space_root(struct inode *inode, int check)
278 size = sizeof(EXT4_I(inode)->i_data);
279 size -= sizeof(struct ext4_extent_header);
280 size /= sizeof(struct ext4_extent);
281 #ifdef AGGRESSIVE_TEST
282 if (!check && size > 3)
288 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
292 size = sizeof(EXT4_I(inode)->i_data);
293 size -= sizeof(struct ext4_extent_header);
294 size /= sizeof(struct ext4_extent_idx);
295 #ifdef AGGRESSIVE_TEST
296 if (!check && size > 4)
303 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
304 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
307 struct ext4_ext_path *path = *ppath;
308 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
310 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
311 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
312 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
313 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
317 ext4_ext_max_entries(struct inode *inode, int depth)
321 if (depth == ext_depth(inode)) {
323 max = ext4_ext_space_root(inode, 1);
325 max = ext4_ext_space_root_idx(inode, 1);
328 max = ext4_ext_space_block(inode, 1);
330 max = ext4_ext_space_block_idx(inode, 1);
336 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
338 ext4_fsblk_t block = ext4_ext_pblock(ext);
339 int len = ext4_ext_get_actual_len(ext);
340 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
345 * - overflow/wrap-around
347 if (lblock + len <= lblock)
349 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
352 static int ext4_valid_extent_idx(struct inode *inode,
353 struct ext4_extent_idx *ext_idx)
355 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
357 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
360 static int ext4_valid_extent_entries(struct inode *inode,
361 struct ext4_extent_header *eh,
364 unsigned short entries;
365 if (eh->eh_entries == 0)
368 entries = le16_to_cpu(eh->eh_entries);
372 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
373 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
374 ext4_fsblk_t pblock = 0;
375 ext4_lblk_t lblock = 0;
376 ext4_lblk_t prev = 0;
379 if (!ext4_valid_extent(inode, ext))
382 /* Check for overlapping extents */
383 lblock = le32_to_cpu(ext->ee_block);
384 len = ext4_ext_get_actual_len(ext);
385 if ((lblock <= prev) && prev) {
386 pblock = ext4_ext_pblock(ext);
387 es->s_last_error_block = cpu_to_le64(pblock);
392 prev = lblock + len - 1;
395 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
397 if (!ext4_valid_extent_idx(inode, ext_idx))
406 static int __ext4_ext_check(const char *function, unsigned int line,
407 struct inode *inode, struct ext4_extent_header *eh,
408 int depth, ext4_fsblk_t pblk)
410 const char *error_msg;
411 int max = 0, err = -EFSCORRUPTED;
413 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
414 error_msg = "invalid magic";
417 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
418 error_msg = "unexpected eh_depth";
421 if (unlikely(eh->eh_max == 0)) {
422 error_msg = "invalid eh_max";
425 max = ext4_ext_max_entries(inode, depth);
426 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
427 error_msg = "too large eh_max";
430 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
431 error_msg = "invalid eh_entries";
434 if (!ext4_valid_extent_entries(inode, eh, depth)) {
435 error_msg = "invalid extent entries";
438 if (unlikely(depth > 32)) {
439 error_msg = "too large eh_depth";
442 /* Verify checksum on non-root extent tree nodes */
443 if (ext_depth(inode) != depth &&
444 !ext4_extent_block_csum_verify(inode, eh)) {
445 error_msg = "extent tree corrupted";
452 ext4_set_errno(inode->i_sb, -err);
453 ext4_error_inode(inode, function, line, 0,
454 "pblk %llu bad header/extent: %s - magic %x, "
455 "entries %u, max %u(%u), depth %u(%u)",
456 (unsigned long long) pblk, error_msg,
457 le16_to_cpu(eh->eh_magic),
458 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
459 max, le16_to_cpu(eh->eh_depth), depth);
463 #define ext4_ext_check(inode, eh, depth, pblk) \
464 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
466 int ext4_ext_check_inode(struct inode *inode)
468 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
471 static struct buffer_head *
472 __read_extent_tree_block(const char *function, unsigned int line,
473 struct inode *inode, ext4_fsblk_t pblk, int depth,
476 struct buffer_head *bh;
479 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
481 return ERR_PTR(-ENOMEM);
483 if (!bh_uptodate_or_lock(bh)) {
484 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
485 err = bh_submit_read(bh);
489 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
491 if (!ext4_has_feature_journal(inode->i_sb) ||
493 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
494 err = __ext4_ext_check(function, line, inode,
495 ext_block_hdr(bh), depth, pblk);
499 set_buffer_verified(bh);
501 * If this is a leaf block, cache all of its entries
503 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
504 struct ext4_extent_header *eh = ext_block_hdr(bh);
505 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
506 ext4_lblk_t prev = 0;
509 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
510 unsigned int status = EXTENT_STATUS_WRITTEN;
511 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
512 int len = ext4_ext_get_actual_len(ex);
514 if (prev && (prev != lblk))
515 ext4_es_cache_extent(inode, prev,
519 if (ext4_ext_is_unwritten(ex))
520 status = EXTENT_STATUS_UNWRITTEN;
521 ext4_es_cache_extent(inode, lblk, len,
522 ext4_ext_pblock(ex), status);
533 #define read_extent_tree_block(inode, pblk, depth, flags) \
534 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
538 * This function is called to cache a file's extent information in the
541 int ext4_ext_precache(struct inode *inode)
543 struct ext4_inode_info *ei = EXT4_I(inode);
544 struct ext4_ext_path *path = NULL;
545 struct buffer_head *bh;
546 int i = 0, depth, ret = 0;
548 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
549 return 0; /* not an extent-mapped inode */
551 down_read(&ei->i_data_sem);
552 depth = ext_depth(inode);
554 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
557 up_read(&ei->i_data_sem);
561 /* Don't cache anything if there are no external extent blocks */
564 path[0].p_hdr = ext_inode_hdr(inode);
565 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
568 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
571 * If this is a leaf block or we've reached the end of
572 * the index block, go up
575 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
576 brelse(path[i].p_bh);
581 bh = read_extent_tree_block(inode,
582 ext4_idx_pblock(path[i].p_idx++),
584 EXT4_EX_FORCE_CACHE);
591 path[i].p_hdr = ext_block_hdr(bh);
592 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
594 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
596 up_read(&ei->i_data_sem);
597 ext4_ext_drop_refs(path);
603 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
605 int k, l = path->p_depth;
608 for (k = 0; k <= l; k++, path++) {
610 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
611 ext4_idx_pblock(path->p_idx));
612 } else if (path->p_ext) {
613 ext_debug(" %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));
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("Displaying leaf extents for inode %lu\n", inode->i_ino);
639 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
640 ext_debug("%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));
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("%d: move %d:%llu in new index %llu\n", level,
658 le32_to_cpu(idx->ei_block),
659 ext4_idx_pblock(idx),
667 ex = path[depth].p_ext;
668 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
669 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
670 le32_to_cpu(ex->ee_block),
672 ext4_ext_is_unwritten(ex),
673 ext4_ext_get_actual_len(ex),
680 #define ext4_ext_show_path(inode, path)
681 #define ext4_ext_show_leaf(inode, path)
682 #define ext4_ext_show_move(inode, path, newblock, level)
685 void ext4_ext_drop_refs(struct ext4_ext_path *path)
691 depth = path->p_depth;
692 for (i = 0; i <= depth; i++, path++)
700 * ext4_ext_binsearch_idx:
701 * binary search for the closest index of the given block
702 * the header must be checked before calling this
705 ext4_ext_binsearch_idx(struct inode *inode,
706 struct ext4_ext_path *path, ext4_lblk_t block)
708 struct ext4_extent_header *eh = path->p_hdr;
709 struct ext4_extent_idx *r, *l, *m;
712 ext_debug("binsearch for %u(idx): ", block);
714 l = EXT_FIRST_INDEX(eh) + 1;
715 r = EXT_LAST_INDEX(eh);
718 if (block < le32_to_cpu(m->ei_block))
722 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
723 m, le32_to_cpu(m->ei_block),
724 r, le32_to_cpu(r->ei_block));
728 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
729 ext4_idx_pblock(path->p_idx));
731 #ifdef CHECK_BINSEARCH
733 struct ext4_extent_idx *chix, *ix;
736 chix = ix = EXT_FIRST_INDEX(eh);
737 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
739 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
740 printk(KERN_DEBUG "k=%d, ix=0x%p, "
742 ix, EXT_FIRST_INDEX(eh));
743 printk(KERN_DEBUG "%u <= %u\n",
744 le32_to_cpu(ix->ei_block),
745 le32_to_cpu(ix[-1].ei_block));
747 BUG_ON(k && le32_to_cpu(ix->ei_block)
748 <= le32_to_cpu(ix[-1].ei_block));
749 if (block < le32_to_cpu(ix->ei_block))
753 BUG_ON(chix != path->p_idx);
760 * ext4_ext_binsearch:
761 * binary search for closest extent of the given block
762 * the header must be checked before calling this
765 ext4_ext_binsearch(struct inode *inode,
766 struct ext4_ext_path *path, ext4_lblk_t block)
768 struct ext4_extent_header *eh = path->p_hdr;
769 struct ext4_extent *r, *l, *m;
771 if (eh->eh_entries == 0) {
773 * this leaf is empty:
774 * we get such a leaf in split/add case
779 ext_debug("binsearch for %u: ", block);
781 l = EXT_FIRST_EXTENT(eh) + 1;
782 r = EXT_LAST_EXTENT(eh);
786 if (block < le32_to_cpu(m->ee_block))
790 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
791 m, le32_to_cpu(m->ee_block),
792 r, le32_to_cpu(r->ee_block));
796 ext_debug(" -> %d:%llu:[%d]%d ",
797 le32_to_cpu(path->p_ext->ee_block),
798 ext4_ext_pblock(path->p_ext),
799 ext4_ext_is_unwritten(path->p_ext),
800 ext4_ext_get_actual_len(path->p_ext));
802 #ifdef CHECK_BINSEARCH
804 struct ext4_extent *chex, *ex;
807 chex = ex = EXT_FIRST_EXTENT(eh);
808 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
809 BUG_ON(k && le32_to_cpu(ex->ee_block)
810 <= le32_to_cpu(ex[-1].ee_block));
811 if (block < le32_to_cpu(ex->ee_block))
815 BUG_ON(chex != path->p_ext);
821 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
823 struct ext4_extent_header *eh;
825 eh = ext_inode_hdr(inode);
828 eh->eh_magic = EXT4_EXT_MAGIC;
829 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
830 ext4_mark_inode_dirty(handle, inode);
834 struct ext4_ext_path *
835 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
836 struct ext4_ext_path **orig_path, int flags)
838 struct ext4_extent_header *eh;
839 struct buffer_head *bh;
840 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
841 short int depth, i, ppos = 0;
844 eh = ext_inode_hdr(inode);
845 depth = ext_depth(inode);
846 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
847 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
854 ext4_ext_drop_refs(path);
855 if (depth > path[0].p_maxdepth) {
857 *orig_path = path = NULL;
861 /* account possible depth increase */
862 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
865 return ERR_PTR(-ENOMEM);
866 path[0].p_maxdepth = depth + 1;
872 /* walk through the tree */
874 ext_debug("depth %d: num %d, max %d\n",
875 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
877 ext4_ext_binsearch_idx(inode, path + ppos, block);
878 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
879 path[ppos].p_depth = i;
880 path[ppos].p_ext = NULL;
882 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
889 eh = ext_block_hdr(bh);
891 path[ppos].p_bh = bh;
892 path[ppos].p_hdr = eh;
895 path[ppos].p_depth = i;
896 path[ppos].p_ext = NULL;
897 path[ppos].p_idx = NULL;
900 ext4_ext_binsearch(inode, path + ppos, block);
901 /* if not an empty leaf */
902 if (path[ppos].p_ext)
903 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
905 ext4_ext_show_path(inode, path);
910 ext4_ext_drop_refs(path);
918 * ext4_ext_insert_index:
919 * insert new index [@logical;@ptr] into the block at @curp;
920 * check where to insert: before @curp or after @curp
922 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
923 struct ext4_ext_path *curp,
924 int logical, ext4_fsblk_t ptr)
926 struct ext4_extent_idx *ix;
929 err = ext4_ext_get_access(handle, inode, curp);
933 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
934 EXT4_ERROR_INODE(inode,
935 "logical %d == ei_block %d!",
936 logical, le32_to_cpu(curp->p_idx->ei_block));
937 return -EFSCORRUPTED;
940 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
941 >= le16_to_cpu(curp->p_hdr->eh_max))) {
942 EXT4_ERROR_INODE(inode,
943 "eh_entries %d >= eh_max %d!",
944 le16_to_cpu(curp->p_hdr->eh_entries),
945 le16_to_cpu(curp->p_hdr->eh_max));
946 return -EFSCORRUPTED;
949 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
951 ext_debug("insert new index %d after: %llu\n", logical, ptr);
952 ix = curp->p_idx + 1;
955 ext_debug("insert new index %d before: %llu\n", logical, ptr);
959 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
962 ext_debug("insert new index %d: "
963 "move %d indices from 0x%p to 0x%p\n",
964 logical, len, ix, ix + 1);
965 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
968 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
969 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
970 return -EFSCORRUPTED;
973 ix->ei_block = cpu_to_le32(logical);
974 ext4_idx_store_pblock(ix, ptr);
975 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
977 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
978 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
979 return -EFSCORRUPTED;
982 err = ext4_ext_dirty(handle, inode, curp);
983 ext4_std_error(inode->i_sb, err);
990 * inserts new subtree into the path, using free index entry
992 * - allocates all needed blocks (new leaf and all intermediate index blocks)
993 * - makes decision where to split
994 * - moves remaining extents and index entries (right to the split point)
995 * into the newly allocated blocks
996 * - initializes subtree
998 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1000 struct ext4_ext_path *path,
1001 struct ext4_extent *newext, int at)
1003 struct buffer_head *bh = NULL;
1004 int depth = ext_depth(inode);
1005 struct ext4_extent_header *neh;
1006 struct ext4_extent_idx *fidx;
1007 int i = at, k, m, a;
1008 ext4_fsblk_t newblock, oldblock;
1010 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1012 size_t ext_size = 0;
1014 /* make decision: where to split? */
1015 /* FIXME: now decision is simplest: at current extent */
1017 /* if current leaf will be split, then we should use
1018 * border from split point */
1019 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1020 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1021 return -EFSCORRUPTED;
1023 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1024 border = path[depth].p_ext[1].ee_block;
1025 ext_debug("leaf will be split."
1026 " next leaf starts at %d\n",
1027 le32_to_cpu(border));
1029 border = newext->ee_block;
1030 ext_debug("leaf will be added."
1031 " next leaf starts at %d\n",
1032 le32_to_cpu(border));
1036 * If error occurs, then we break processing
1037 * and mark filesystem read-only. index won't
1038 * be inserted and tree will be in consistent
1039 * state. Next mount will repair buffers too.
1043 * Get array to track all allocated blocks.
1044 * We need this to handle errors and free blocks
1047 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1051 /* allocate all needed blocks */
1052 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1053 for (a = 0; a < depth - at; a++) {
1054 newblock = ext4_ext_new_meta_block(handle, inode, path,
1055 newext, &err, flags);
1058 ablocks[a] = newblock;
1061 /* initialize new leaf */
1062 newblock = ablocks[--a];
1063 if (unlikely(newblock == 0)) {
1064 EXT4_ERROR_INODE(inode, "newblock == 0!");
1065 err = -EFSCORRUPTED;
1068 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1069 if (unlikely(!bh)) {
1075 err = ext4_journal_get_create_access(handle, bh);
1079 neh = ext_block_hdr(bh);
1080 neh->eh_entries = 0;
1081 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1082 neh->eh_magic = EXT4_EXT_MAGIC;
1085 /* move remainder of path[depth] to the new leaf */
1086 if (unlikely(path[depth].p_hdr->eh_entries !=
1087 path[depth].p_hdr->eh_max)) {
1088 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1089 path[depth].p_hdr->eh_entries,
1090 path[depth].p_hdr->eh_max);
1091 err = -EFSCORRUPTED;
1094 /* start copy from next extent */
1095 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1096 ext4_ext_show_move(inode, path, newblock, depth);
1098 struct ext4_extent *ex;
1099 ex = EXT_FIRST_EXTENT(neh);
1100 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1101 le16_add_cpu(&neh->eh_entries, m);
1104 /* zero out unused area in the extent block */
1105 ext_size = sizeof(struct ext4_extent_header) +
1106 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1107 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1108 ext4_extent_block_csum_set(inode, neh);
1109 set_buffer_uptodate(bh);
1112 err = ext4_handle_dirty_metadata(handle, inode, bh);
1118 /* correct old leaf */
1120 err = ext4_ext_get_access(handle, inode, path + depth);
1123 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1124 err = ext4_ext_dirty(handle, inode, path + depth);
1130 /* create intermediate indexes */
1132 if (unlikely(k < 0)) {
1133 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1134 err = -EFSCORRUPTED;
1138 ext_debug("create %d intermediate indices\n", k);
1139 /* insert new index into current index block */
1140 /* current depth stored in i var */
1143 oldblock = newblock;
1144 newblock = ablocks[--a];
1145 bh = sb_getblk(inode->i_sb, newblock);
1146 if (unlikely(!bh)) {
1152 err = ext4_journal_get_create_access(handle, bh);
1156 neh = ext_block_hdr(bh);
1157 neh->eh_entries = cpu_to_le16(1);
1158 neh->eh_magic = EXT4_EXT_MAGIC;
1159 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1160 neh->eh_depth = cpu_to_le16(depth - i);
1161 fidx = EXT_FIRST_INDEX(neh);
1162 fidx->ei_block = border;
1163 ext4_idx_store_pblock(fidx, oldblock);
1165 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1166 i, newblock, le32_to_cpu(border), oldblock);
1168 /* move remainder of path[i] to the new index block */
1169 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1170 EXT_LAST_INDEX(path[i].p_hdr))) {
1171 EXT4_ERROR_INODE(inode,
1172 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1173 le32_to_cpu(path[i].p_ext->ee_block));
1174 err = -EFSCORRUPTED;
1177 /* start copy indexes */
1178 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1179 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1180 EXT_MAX_INDEX(path[i].p_hdr));
1181 ext4_ext_show_move(inode, path, newblock, i);
1183 memmove(++fidx, path[i].p_idx,
1184 sizeof(struct ext4_extent_idx) * m);
1185 le16_add_cpu(&neh->eh_entries, m);
1187 /* zero out unused area in the extent block */
1188 ext_size = sizeof(struct ext4_extent_header) +
1189 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1190 memset(bh->b_data + ext_size, 0,
1191 inode->i_sb->s_blocksize - ext_size);
1192 ext4_extent_block_csum_set(inode, neh);
1193 set_buffer_uptodate(bh);
1196 err = ext4_handle_dirty_metadata(handle, inode, bh);
1202 /* correct old index */
1204 err = ext4_ext_get_access(handle, inode, path + i);
1207 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1208 err = ext4_ext_dirty(handle, inode, path + i);
1216 /* insert new index */
1217 err = ext4_ext_insert_index(handle, inode, path + at,
1218 le32_to_cpu(border), newblock);
1222 if (buffer_locked(bh))
1228 /* free all allocated blocks in error case */
1229 for (i = 0; i < depth; i++) {
1232 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1233 EXT4_FREE_BLOCKS_METADATA);
1242 * ext4_ext_grow_indepth:
1243 * implements tree growing procedure:
1244 * - allocates new block
1245 * - moves top-level data (index block or leaf) into the new block
1246 * - initializes new top-level, creating index that points to the
1247 * just created block
1249 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1252 struct ext4_extent_header *neh;
1253 struct buffer_head *bh;
1254 ext4_fsblk_t newblock, goal = 0;
1255 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1257 size_t ext_size = 0;
1259 /* Try to prepend new index to old one */
1260 if (ext_depth(inode))
1261 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1262 if (goal > le32_to_cpu(es->s_first_data_block)) {
1263 flags |= EXT4_MB_HINT_TRY_GOAL;
1266 goal = ext4_inode_to_goal_block(inode);
1267 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1272 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1277 err = ext4_journal_get_create_access(handle, bh);
1283 ext_size = sizeof(EXT4_I(inode)->i_data);
1284 /* move top-level index/leaf into new block */
1285 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1286 /* zero out unused area in the extent block */
1287 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1289 /* set size of new block */
1290 neh = ext_block_hdr(bh);
1291 /* old root could have indexes or leaves
1292 * so calculate e_max right way */
1293 if (ext_depth(inode))
1294 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1296 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1297 neh->eh_magic = EXT4_EXT_MAGIC;
1298 ext4_extent_block_csum_set(inode, neh);
1299 set_buffer_uptodate(bh);
1302 err = ext4_handle_dirty_metadata(handle, inode, bh);
1306 /* Update top-level index: num,max,pointer */
1307 neh = ext_inode_hdr(inode);
1308 neh->eh_entries = cpu_to_le16(1);
1309 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1310 if (neh->eh_depth == 0) {
1311 /* Root extent block becomes index block */
1312 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1313 EXT_FIRST_INDEX(neh)->ei_block =
1314 EXT_FIRST_EXTENT(neh)->ee_block;
1316 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1317 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1318 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1319 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1321 le16_add_cpu(&neh->eh_depth, 1);
1322 ext4_mark_inode_dirty(handle, inode);
1330 * ext4_ext_create_new_leaf:
1331 * finds empty index and adds new leaf.
1332 * if no free index is found, then it requests in-depth growing.
1334 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1335 unsigned int mb_flags,
1336 unsigned int gb_flags,
1337 struct ext4_ext_path **ppath,
1338 struct ext4_extent *newext)
1340 struct ext4_ext_path *path = *ppath;
1341 struct ext4_ext_path *curp;
1342 int depth, i, err = 0;
1345 i = depth = ext_depth(inode);
1347 /* walk up to the tree and look for free index entry */
1348 curp = path + depth;
1349 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1354 /* we use already allocated block for index block,
1355 * so subsequent data blocks should be contiguous */
1356 if (EXT_HAS_FREE_INDEX(curp)) {
1357 /* if we found index with free entry, then use that
1358 * entry: create all needed subtree and add new leaf */
1359 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1364 path = ext4_find_extent(inode,
1365 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1368 err = PTR_ERR(path);
1370 /* tree is full, time to grow in depth */
1371 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1376 path = ext4_find_extent(inode,
1377 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1380 err = PTR_ERR(path);
1385 * only first (depth 0 -> 1) produces free space;
1386 * in all other cases we have to split the grown tree
1388 depth = ext_depth(inode);
1389 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1390 /* now we need to split */
1400 * search the closest allocated block to the left for *logical
1401 * and returns it at @logical + it's physical address at @phys
1402 * if *logical is the smallest allocated block, the function
1403 * returns 0 at @phys
1404 * return value contains 0 (success) or error code
1406 static int ext4_ext_search_left(struct inode *inode,
1407 struct ext4_ext_path *path,
1408 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1410 struct ext4_extent_idx *ix;
1411 struct ext4_extent *ex;
1414 if (unlikely(path == NULL)) {
1415 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1416 return -EFSCORRUPTED;
1418 depth = path->p_depth;
1421 if (depth == 0 && path->p_ext == NULL)
1424 /* usually extent in the path covers blocks smaller
1425 * then *logical, but it can be that extent is the
1426 * first one in the file */
1428 ex = path[depth].p_ext;
1429 ee_len = ext4_ext_get_actual_len(ex);
1430 if (*logical < le32_to_cpu(ex->ee_block)) {
1431 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1432 EXT4_ERROR_INODE(inode,
1433 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1434 *logical, le32_to_cpu(ex->ee_block));
1435 return -EFSCORRUPTED;
1437 while (--depth >= 0) {
1438 ix = path[depth].p_idx;
1439 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1440 EXT4_ERROR_INODE(inode,
1441 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1442 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1443 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1444 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1446 return -EFSCORRUPTED;
1452 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1453 EXT4_ERROR_INODE(inode,
1454 "logical %d < ee_block %d + ee_len %d!",
1455 *logical, le32_to_cpu(ex->ee_block), ee_len);
1456 return -EFSCORRUPTED;
1459 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1460 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1465 * search the closest allocated block to the right for *logical
1466 * and returns it at @logical + it's physical address at @phys
1467 * if *logical is the largest allocated block, the function
1468 * returns 0 at @phys
1469 * return value contains 0 (success) or error code
1471 static int ext4_ext_search_right(struct inode *inode,
1472 struct ext4_ext_path *path,
1473 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1474 struct ext4_extent **ret_ex)
1476 struct buffer_head *bh = NULL;
1477 struct ext4_extent_header *eh;
1478 struct ext4_extent_idx *ix;
1479 struct ext4_extent *ex;
1481 int depth; /* Note, NOT eh_depth; depth from top of tree */
1484 if (unlikely(path == NULL)) {
1485 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1486 return -EFSCORRUPTED;
1488 depth = path->p_depth;
1491 if (depth == 0 && path->p_ext == NULL)
1494 /* usually extent in the path covers blocks smaller
1495 * then *logical, but it can be that extent is the
1496 * first one in the file */
1498 ex = path[depth].p_ext;
1499 ee_len = ext4_ext_get_actual_len(ex);
1500 if (*logical < le32_to_cpu(ex->ee_block)) {
1501 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1502 EXT4_ERROR_INODE(inode,
1503 "first_extent(path[%d].p_hdr) != ex",
1505 return -EFSCORRUPTED;
1507 while (--depth >= 0) {
1508 ix = path[depth].p_idx;
1509 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1510 EXT4_ERROR_INODE(inode,
1511 "ix != EXT_FIRST_INDEX *logical %d!",
1513 return -EFSCORRUPTED;
1519 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1520 EXT4_ERROR_INODE(inode,
1521 "logical %d < ee_block %d + ee_len %d!",
1522 *logical, le32_to_cpu(ex->ee_block), ee_len);
1523 return -EFSCORRUPTED;
1526 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1527 /* next allocated block in this leaf */
1532 /* go up and search for index to the right */
1533 while (--depth >= 0) {
1534 ix = path[depth].p_idx;
1535 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1539 /* we've gone up to the root and found no index to the right */
1543 /* we've found index to the right, let's
1544 * follow it and find the closest allocated
1545 * block to the right */
1547 block = ext4_idx_pblock(ix);
1548 while (++depth < path->p_depth) {
1549 /* subtract from p_depth to get proper eh_depth */
1550 bh = read_extent_tree_block(inode, block,
1551 path->p_depth - depth, 0);
1554 eh = ext_block_hdr(bh);
1555 ix = EXT_FIRST_INDEX(eh);
1556 block = ext4_idx_pblock(ix);
1560 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1563 eh = ext_block_hdr(bh);
1564 ex = EXT_FIRST_EXTENT(eh);
1566 *logical = le32_to_cpu(ex->ee_block);
1567 *phys = ext4_ext_pblock(ex);
1575 * ext4_ext_next_allocated_block:
1576 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1577 * NOTE: it considers block number from index entry as
1578 * allocated block. Thus, index entries have to be consistent
1582 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1586 BUG_ON(path == NULL);
1587 depth = path->p_depth;
1589 if (depth == 0 && path->p_ext == NULL)
1590 return EXT_MAX_BLOCKS;
1592 while (depth >= 0) {
1593 if (depth == path->p_depth) {
1595 if (path[depth].p_ext &&
1596 path[depth].p_ext !=
1597 EXT_LAST_EXTENT(path[depth].p_hdr))
1598 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1601 if (path[depth].p_idx !=
1602 EXT_LAST_INDEX(path[depth].p_hdr))
1603 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1608 return EXT_MAX_BLOCKS;
1612 * ext4_ext_next_leaf_block:
1613 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1615 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1619 BUG_ON(path == NULL);
1620 depth = path->p_depth;
1622 /* zero-tree has no leaf blocks at all */
1624 return EXT_MAX_BLOCKS;
1626 /* go to index block */
1629 while (depth >= 0) {
1630 if (path[depth].p_idx !=
1631 EXT_LAST_INDEX(path[depth].p_hdr))
1632 return (ext4_lblk_t)
1633 le32_to_cpu(path[depth].p_idx[1].ei_block);
1637 return EXT_MAX_BLOCKS;
1641 * ext4_ext_correct_indexes:
1642 * if leaf gets modified and modified extent is first in the leaf,
1643 * then we have to correct all indexes above.
1644 * TODO: do we need to correct tree in all cases?
1646 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1647 struct ext4_ext_path *path)
1649 struct ext4_extent_header *eh;
1650 int depth = ext_depth(inode);
1651 struct ext4_extent *ex;
1655 eh = path[depth].p_hdr;
1656 ex = path[depth].p_ext;
1658 if (unlikely(ex == NULL || eh == NULL)) {
1659 EXT4_ERROR_INODE(inode,
1660 "ex %p == NULL or eh %p == NULL", ex, eh);
1661 return -EFSCORRUPTED;
1665 /* there is no tree at all */
1669 if (ex != EXT_FIRST_EXTENT(eh)) {
1670 /* we correct tree if first leaf got modified only */
1675 * TODO: we need correction if border is smaller than current one
1678 border = path[depth].p_ext->ee_block;
1679 err = ext4_ext_get_access(handle, inode, path + k);
1682 path[k].p_idx->ei_block = border;
1683 err = ext4_ext_dirty(handle, inode, path + k);
1688 /* change all left-side indexes */
1689 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1691 err = ext4_ext_get_access(handle, inode, path + k);
1694 path[k].p_idx->ei_block = border;
1695 err = ext4_ext_dirty(handle, inode, path + k);
1703 static int ext4_can_extents_be_merged(struct inode *inode,
1704 struct ext4_extent *ex1,
1705 struct ext4_extent *ex2)
1707 unsigned short ext1_ee_len, ext2_ee_len;
1709 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1712 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1713 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1715 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1716 le32_to_cpu(ex2->ee_block))
1720 * To allow future support for preallocated extents to be added
1721 * as an RO_COMPAT feature, refuse to merge to extents if
1722 * this can result in the top bit of ee_len being set.
1724 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1727 if (ext4_ext_is_unwritten(ex1) &&
1728 ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1730 #ifdef AGGRESSIVE_TEST
1731 if (ext1_ee_len >= 4)
1735 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1741 * This function tries to merge the "ex" extent to the next extent in the tree.
1742 * It always tries to merge towards right. If you want to merge towards
1743 * left, pass "ex - 1" as argument instead of "ex".
1744 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1745 * 1 if they got merged.
1747 static int ext4_ext_try_to_merge_right(struct inode *inode,
1748 struct ext4_ext_path *path,
1749 struct ext4_extent *ex)
1751 struct ext4_extent_header *eh;
1752 unsigned int depth, len;
1753 int merge_done = 0, unwritten;
1755 depth = ext_depth(inode);
1756 BUG_ON(path[depth].p_hdr == NULL);
1757 eh = path[depth].p_hdr;
1759 while (ex < EXT_LAST_EXTENT(eh)) {
1760 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1762 /* merge with next extent! */
1763 unwritten = ext4_ext_is_unwritten(ex);
1764 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1765 + ext4_ext_get_actual_len(ex + 1));
1767 ext4_ext_mark_unwritten(ex);
1769 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1770 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1771 * sizeof(struct ext4_extent);
1772 memmove(ex + 1, ex + 2, len);
1774 le16_add_cpu(&eh->eh_entries, -1);
1776 WARN_ON(eh->eh_entries == 0);
1777 if (!eh->eh_entries)
1778 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1785 * This function does a very simple check to see if we can collapse
1786 * an extent tree with a single extent tree leaf block into the inode.
1788 static void ext4_ext_try_to_merge_up(handle_t *handle,
1789 struct inode *inode,
1790 struct ext4_ext_path *path)
1793 unsigned max_root = ext4_ext_space_root(inode, 0);
1796 if ((path[0].p_depth != 1) ||
1797 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1798 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1802 * We need to modify the block allocation bitmap and the block
1803 * group descriptor to release the extent tree block. If we
1804 * can't get the journal credits, give up.
1806 if (ext4_journal_extend(handle, 2,
1807 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1811 * Copy the extent data up to the inode
1813 blk = ext4_idx_pblock(path[0].p_idx);
1814 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1815 sizeof(struct ext4_extent_idx);
1816 s += sizeof(struct ext4_extent_header);
1818 path[1].p_maxdepth = path[0].p_maxdepth;
1819 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1820 path[0].p_depth = 0;
1821 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1822 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1823 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1825 brelse(path[1].p_bh);
1826 ext4_free_blocks(handle, inode, NULL, blk, 1,
1827 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1831 * This function tries to merge the @ex extent to neighbours in the tree.
1832 * return 1 if merge left else 0.
1834 static void ext4_ext_try_to_merge(handle_t *handle,
1835 struct inode *inode,
1836 struct ext4_ext_path *path,
1837 struct ext4_extent *ex) {
1838 struct ext4_extent_header *eh;
1842 depth = ext_depth(inode);
1843 BUG_ON(path[depth].p_hdr == NULL);
1844 eh = path[depth].p_hdr;
1846 if (ex > EXT_FIRST_EXTENT(eh))
1847 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1850 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1852 ext4_ext_try_to_merge_up(handle, inode, path);
1856 * check if a portion of the "newext" extent overlaps with an
1859 * If there is an overlap discovered, it updates the length of the newext
1860 * such that there will be no overlap, and then returns 1.
1861 * If there is no overlap found, it returns 0.
1863 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1864 struct inode *inode,
1865 struct ext4_extent *newext,
1866 struct ext4_ext_path *path)
1869 unsigned int depth, len1;
1870 unsigned int ret = 0;
1872 b1 = le32_to_cpu(newext->ee_block);
1873 len1 = ext4_ext_get_actual_len(newext);
1874 depth = ext_depth(inode);
1875 if (!path[depth].p_ext)
1877 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1880 * get the next allocated block if the extent in the path
1881 * is before the requested block(s)
1884 b2 = ext4_ext_next_allocated_block(path);
1885 if (b2 == EXT_MAX_BLOCKS)
1887 b2 = EXT4_LBLK_CMASK(sbi, b2);
1890 /* check for wrap through zero on extent logical start block*/
1891 if (b1 + len1 < b1) {
1892 len1 = EXT_MAX_BLOCKS - b1;
1893 newext->ee_len = cpu_to_le16(len1);
1897 /* check for overlap */
1898 if (b1 + len1 > b2) {
1899 newext->ee_len = cpu_to_le16(b2 - b1);
1907 * ext4_ext_insert_extent:
1908 * tries to merge requsted extent into the existing extent or
1909 * inserts requested extent as new one into the tree,
1910 * creating new leaf in the no-space case.
1912 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1913 struct ext4_ext_path **ppath,
1914 struct ext4_extent *newext, int gb_flags)
1916 struct ext4_ext_path *path = *ppath;
1917 struct ext4_extent_header *eh;
1918 struct ext4_extent *ex, *fex;
1919 struct ext4_extent *nearex; /* nearest extent */
1920 struct ext4_ext_path *npath = NULL;
1921 int depth, len, err;
1923 int mb_flags = 0, unwritten;
1925 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1926 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1927 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1928 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1929 return -EFSCORRUPTED;
1931 depth = ext_depth(inode);
1932 ex = path[depth].p_ext;
1933 eh = path[depth].p_hdr;
1934 if (unlikely(path[depth].p_hdr == NULL)) {
1935 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1936 return -EFSCORRUPTED;
1939 /* try to insert block into found extent and return */
1940 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1943 * Try to see whether we should rather test the extent on
1944 * right from ex, or from the left of ex. This is because
1945 * ext4_find_extent() can return either extent on the
1946 * left, or on the right from the searched position. This
1947 * will make merging more effective.
1949 if (ex < EXT_LAST_EXTENT(eh) &&
1950 (le32_to_cpu(ex->ee_block) +
1951 ext4_ext_get_actual_len(ex) <
1952 le32_to_cpu(newext->ee_block))) {
1955 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1956 (le32_to_cpu(newext->ee_block) +
1957 ext4_ext_get_actual_len(newext) <
1958 le32_to_cpu(ex->ee_block)))
1961 /* Try to append newex to the ex */
1962 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1963 ext_debug("append [%d]%d block to %u:[%d]%d"
1965 ext4_ext_is_unwritten(newext),
1966 ext4_ext_get_actual_len(newext),
1967 le32_to_cpu(ex->ee_block),
1968 ext4_ext_is_unwritten(ex),
1969 ext4_ext_get_actual_len(ex),
1970 ext4_ext_pblock(ex));
1971 err = ext4_ext_get_access(handle, inode,
1975 unwritten = ext4_ext_is_unwritten(ex);
1976 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1977 + ext4_ext_get_actual_len(newext));
1979 ext4_ext_mark_unwritten(ex);
1980 eh = path[depth].p_hdr;
1986 /* Try to prepend newex to the ex */
1987 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1988 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1990 le32_to_cpu(newext->ee_block),
1991 ext4_ext_is_unwritten(newext),
1992 ext4_ext_get_actual_len(newext),
1993 le32_to_cpu(ex->ee_block),
1994 ext4_ext_is_unwritten(ex),
1995 ext4_ext_get_actual_len(ex),
1996 ext4_ext_pblock(ex));
1997 err = ext4_ext_get_access(handle, inode,
2002 unwritten = ext4_ext_is_unwritten(ex);
2003 ex->ee_block = newext->ee_block;
2004 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2005 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2006 + ext4_ext_get_actual_len(newext));
2008 ext4_ext_mark_unwritten(ex);
2009 eh = path[depth].p_hdr;
2015 depth = ext_depth(inode);
2016 eh = path[depth].p_hdr;
2017 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2020 /* probably next leaf has space for us? */
2021 fex = EXT_LAST_EXTENT(eh);
2022 next = EXT_MAX_BLOCKS;
2023 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2024 next = ext4_ext_next_leaf_block(path);
2025 if (next != EXT_MAX_BLOCKS) {
2026 ext_debug("next leaf block - %u\n", next);
2027 BUG_ON(npath != NULL);
2028 npath = ext4_find_extent(inode, next, NULL, 0);
2030 return PTR_ERR(npath);
2031 BUG_ON(npath->p_depth != path->p_depth);
2032 eh = npath[depth].p_hdr;
2033 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2034 ext_debug("next leaf isn't full(%d)\n",
2035 le16_to_cpu(eh->eh_entries));
2039 ext_debug("next leaf has no free space(%d,%d)\n",
2040 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2044 * There is no free space in the found leaf.
2045 * We're gonna add a new leaf in the tree.
2047 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2048 mb_flags |= EXT4_MB_USE_RESERVED;
2049 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2053 depth = ext_depth(inode);
2054 eh = path[depth].p_hdr;
2057 nearex = path[depth].p_ext;
2059 err = ext4_ext_get_access(handle, inode, path + depth);
2064 /* there is no extent in this leaf, create first one */
2065 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2066 le32_to_cpu(newext->ee_block),
2067 ext4_ext_pblock(newext),
2068 ext4_ext_is_unwritten(newext),
2069 ext4_ext_get_actual_len(newext));
2070 nearex = EXT_FIRST_EXTENT(eh);
2072 if (le32_to_cpu(newext->ee_block)
2073 > le32_to_cpu(nearex->ee_block)) {
2075 ext_debug("insert %u:%llu:[%d]%d before: "
2077 le32_to_cpu(newext->ee_block),
2078 ext4_ext_pblock(newext),
2079 ext4_ext_is_unwritten(newext),
2080 ext4_ext_get_actual_len(newext),
2085 BUG_ON(newext->ee_block == nearex->ee_block);
2086 ext_debug("insert %u:%llu:[%d]%d after: "
2088 le32_to_cpu(newext->ee_block),
2089 ext4_ext_pblock(newext),
2090 ext4_ext_is_unwritten(newext),
2091 ext4_ext_get_actual_len(newext),
2094 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2096 ext_debug("insert %u:%llu:[%d]%d: "
2097 "move %d extents from 0x%p to 0x%p\n",
2098 le32_to_cpu(newext->ee_block),
2099 ext4_ext_pblock(newext),
2100 ext4_ext_is_unwritten(newext),
2101 ext4_ext_get_actual_len(newext),
2102 len, nearex, nearex + 1);
2103 memmove(nearex + 1, nearex,
2104 len * sizeof(struct ext4_extent));
2108 le16_add_cpu(&eh->eh_entries, 1);
2109 path[depth].p_ext = nearex;
2110 nearex->ee_block = newext->ee_block;
2111 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2112 nearex->ee_len = newext->ee_len;
2115 /* try to merge extents */
2116 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2117 ext4_ext_try_to_merge(handle, inode, path, nearex);
2120 /* time to correct all indexes above */
2121 err = ext4_ext_correct_indexes(handle, inode, path);
2125 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2128 ext4_ext_drop_refs(npath);
2133 static int ext4_fill_fiemap_extents(struct inode *inode,
2134 ext4_lblk_t block, ext4_lblk_t num,
2135 struct fiemap_extent_info *fieinfo)
2137 struct ext4_ext_path *path = NULL;
2138 struct ext4_extent *ex;
2139 struct extent_status es;
2140 ext4_lblk_t next, next_del, start = 0, end = 0;
2141 ext4_lblk_t last = block + num;
2142 int exists, depth = 0, err = 0;
2143 unsigned int flags = 0;
2144 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2146 while (block < last && block != EXT_MAX_BLOCKS) {
2148 /* find extent for this block */
2149 down_read(&EXT4_I(inode)->i_data_sem);
2151 path = ext4_find_extent(inode, block, &path, 0);
2153 up_read(&EXT4_I(inode)->i_data_sem);
2154 err = PTR_ERR(path);
2159 depth = ext_depth(inode);
2160 if (unlikely(path[depth].p_hdr == NULL)) {
2161 up_read(&EXT4_I(inode)->i_data_sem);
2162 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2163 err = -EFSCORRUPTED;
2166 ex = path[depth].p_ext;
2167 next = ext4_ext_next_allocated_block(path);
2172 /* there is no extent yet, so try to allocate
2173 * all requested space */
2176 } else if (le32_to_cpu(ex->ee_block) > block) {
2177 /* need to allocate space before found extent */
2179 end = le32_to_cpu(ex->ee_block);
2180 if (block + num < end)
2182 } else if (block >= le32_to_cpu(ex->ee_block)
2183 + ext4_ext_get_actual_len(ex)) {
2184 /* need to allocate space after found extent */
2189 } else if (block >= le32_to_cpu(ex->ee_block)) {
2191 * some part of requested space is covered
2195 end = le32_to_cpu(ex->ee_block)
2196 + ext4_ext_get_actual_len(ex);
2197 if (block + num < end)
2203 BUG_ON(end <= start);
2207 es.es_len = end - start;
2210 es.es_lblk = le32_to_cpu(ex->ee_block);
2211 es.es_len = ext4_ext_get_actual_len(ex);
2212 es.es_pblk = ext4_ext_pblock(ex);
2213 if (ext4_ext_is_unwritten(ex))
2214 flags |= FIEMAP_EXTENT_UNWRITTEN;
2218 * Find delayed extent and update es accordingly. We call
2219 * it even in !exists case to find out whether es is the
2220 * last existing extent or not.
2222 next_del = ext4_find_delayed_extent(inode, &es);
2223 if (!exists && next_del) {
2225 flags |= (FIEMAP_EXTENT_DELALLOC |
2226 FIEMAP_EXTENT_UNKNOWN);
2228 up_read(&EXT4_I(inode)->i_data_sem);
2230 if (unlikely(es.es_len == 0)) {
2231 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2232 err = -EFSCORRUPTED;
2237 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2238 * we need to check next == EXT_MAX_BLOCKS because it is
2239 * possible that an extent is with unwritten and delayed
2240 * status due to when an extent is delayed allocated and
2241 * is allocated by fallocate status tree will track both of
2244 * So we could return a unwritten and delayed extent, and
2245 * its block is equal to 'next'.
2247 if (next == next_del && next == EXT_MAX_BLOCKS) {
2248 flags |= FIEMAP_EXTENT_LAST;
2249 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2250 next != EXT_MAX_BLOCKS)) {
2251 EXT4_ERROR_INODE(inode,
2252 "next extent == %u, next "
2253 "delalloc extent = %u",
2255 err = -EFSCORRUPTED;
2261 err = fiemap_fill_next_extent(fieinfo,
2262 (__u64)es.es_lblk << blksize_bits,
2263 (__u64)es.es_pblk << blksize_bits,
2264 (__u64)es.es_len << blksize_bits,
2274 block = es.es_lblk + es.es_len;
2277 ext4_ext_drop_refs(path);
2282 static int ext4_fill_es_cache_info(struct inode *inode,
2283 ext4_lblk_t block, ext4_lblk_t num,
2284 struct fiemap_extent_info *fieinfo)
2286 ext4_lblk_t next, end = block + num - 1;
2287 struct extent_status es;
2288 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2292 while (block <= end) {
2295 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2297 if (ext4_es_is_unwritten(&es))
2298 flags |= FIEMAP_EXTENT_UNWRITTEN;
2299 if (ext4_es_is_delayed(&es))
2300 flags |= (FIEMAP_EXTENT_DELALLOC |
2301 FIEMAP_EXTENT_UNKNOWN);
2302 if (ext4_es_is_hole(&es))
2303 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2305 flags |= FIEMAP_EXTENT_LAST;
2306 if (flags & (FIEMAP_EXTENT_DELALLOC|
2307 EXT4_FIEMAP_EXTENT_HOLE))
2310 es.es_pblk = ext4_es_pblock(&es);
2311 err = fiemap_fill_next_extent(fieinfo,
2312 (__u64)es.es_lblk << blksize_bits,
2313 (__u64)es.es_pblk << blksize_bits,
2314 (__u64)es.es_len << blksize_bits,
2329 * ext4_ext_determine_hole - determine hole around given block
2330 * @inode: inode we lookup in
2331 * @path: path in extent tree to @lblk
2332 * @lblk: pointer to logical block around which we want to determine hole
2334 * Determine hole length (and start if easily possible) around given logical
2335 * block. We don't try too hard to find the beginning of the hole but @path
2336 * actually points to extent before @lblk, we provide it.
2338 * The function returns the length of a hole starting at @lblk. We update @lblk
2339 * to the beginning of the hole if we managed to find it.
2341 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2342 struct ext4_ext_path *path,
2345 int depth = ext_depth(inode);
2346 struct ext4_extent *ex;
2349 ex = path[depth].p_ext;
2351 /* there is no extent yet, so gap is [0;-] */
2353 len = EXT_MAX_BLOCKS;
2354 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2355 len = le32_to_cpu(ex->ee_block) - *lblk;
2356 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2357 + ext4_ext_get_actual_len(ex)) {
2360 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2361 next = ext4_ext_next_allocated_block(path);
2362 BUG_ON(next == *lblk);
2371 * ext4_ext_put_gap_in_cache:
2372 * calculate boundaries of the gap that the requested block fits into
2373 * and cache this gap
2376 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2377 ext4_lblk_t hole_len)
2379 struct extent_status es;
2381 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2382 hole_start + hole_len - 1, &es);
2384 /* There's delayed extent containing lblock? */
2385 if (es.es_lblk <= hole_start)
2387 hole_len = min(es.es_lblk - hole_start, hole_len);
2389 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2390 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2391 EXTENT_STATUS_HOLE);
2396 * removes index from the index block.
2398 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2399 struct ext4_ext_path *path, int depth)
2404 /* free index block */
2406 path = path + depth;
2407 leaf = ext4_idx_pblock(path->p_idx);
2408 if (unlikely(path->p_hdr->eh_entries == 0)) {
2409 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2410 return -EFSCORRUPTED;
2412 err = ext4_ext_get_access(handle, inode, path);
2416 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2417 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2418 len *= sizeof(struct ext4_extent_idx);
2419 memmove(path->p_idx, path->p_idx + 1, len);
2422 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2423 err = ext4_ext_dirty(handle, inode, path);
2426 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2427 trace_ext4_ext_rm_idx(inode, leaf);
2429 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2430 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2432 while (--depth >= 0) {
2433 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2436 err = ext4_ext_get_access(handle, inode, path);
2439 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2440 err = ext4_ext_dirty(handle, inode, path);
2448 * ext4_ext_calc_credits_for_single_extent:
2449 * This routine returns max. credits that needed to insert an extent
2450 * to the extent tree.
2451 * When pass the actual path, the caller should calculate credits
2454 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2455 struct ext4_ext_path *path)
2458 int depth = ext_depth(inode);
2461 /* probably there is space in leaf? */
2462 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2463 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2466 * There are some space in the leaf tree, no
2467 * need to account for leaf block credit
2469 * bitmaps and block group descriptor blocks
2470 * and other metadata blocks still need to be
2473 /* 1 bitmap, 1 block group descriptor */
2474 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2479 return ext4_chunk_trans_blocks(inode, nrblocks);
2483 * How many index/leaf blocks need to change/allocate to add @extents extents?
2485 * If we add a single extent, then in the worse case, each tree level
2486 * index/leaf need to be changed in case of the tree split.
2488 * If more extents are inserted, they could cause the whole tree split more
2489 * than once, but this is really rare.
2491 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2496 /* If we are converting the inline data, only one is needed here. */
2497 if (ext4_has_inline_data(inode))
2500 depth = ext_depth(inode);
2510 static inline int get_default_free_blocks_flags(struct inode *inode)
2512 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2513 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2514 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2515 else if (ext4_should_journal_data(inode))
2516 return EXT4_FREE_BLOCKS_FORGET;
2521 * ext4_rereserve_cluster - increment the reserved cluster count when
2522 * freeing a cluster with a pending reservation
2524 * @inode - file containing the cluster
2525 * @lblk - logical block in cluster to be reserved
2527 * Increments the reserved cluster count and adjusts quota in a bigalloc
2528 * file system when freeing a partial cluster containing at least one
2529 * delayed and unwritten block. A partial cluster meeting that
2530 * requirement will have a pending reservation. If so, the
2531 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2532 * defer reserved and allocated space accounting to a subsequent call
2535 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2537 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2538 struct ext4_inode_info *ei = EXT4_I(inode);
2540 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2542 spin_lock(&ei->i_block_reservation_lock);
2543 ei->i_reserved_data_blocks++;
2544 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2545 spin_unlock(&ei->i_block_reservation_lock);
2547 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2548 ext4_remove_pending(inode, lblk);
2551 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2552 struct ext4_extent *ex,
2553 struct partial_cluster *partial,
2554 ext4_lblk_t from, ext4_lblk_t to)
2556 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2557 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2558 ext4_fsblk_t last_pblk, pblk;
2562 /* only extent tail removal is allowed */
2563 if (from < le32_to_cpu(ex->ee_block) ||
2564 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2565 ext4_error(sbi->s_sb,
2566 "strange request: removal(2) %u-%u from %u:%u",
2567 from, to, le32_to_cpu(ex->ee_block), ee_len);
2571 #ifdef EXTENTS_STATS
2572 spin_lock(&sbi->s_ext_stats_lock);
2573 sbi->s_ext_blocks += ee_len;
2574 sbi->s_ext_extents++;
2575 if (ee_len < sbi->s_ext_min)
2576 sbi->s_ext_min = ee_len;
2577 if (ee_len > sbi->s_ext_max)
2578 sbi->s_ext_max = ee_len;
2579 if (ext_depth(inode) > sbi->s_depth_max)
2580 sbi->s_depth_max = ext_depth(inode);
2581 spin_unlock(&sbi->s_ext_stats_lock);
2584 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2587 * if we have a partial cluster, and it's different from the
2588 * cluster of the last block in the extent, we free it
2590 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2592 if (partial->state != initial &&
2593 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2594 if (partial->state == tofree) {
2595 flags = get_default_free_blocks_flags(inode);
2596 if (ext4_is_pending(inode, partial->lblk))
2597 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2598 ext4_free_blocks(handle, inode, NULL,
2599 EXT4_C2B(sbi, partial->pclu),
2600 sbi->s_cluster_ratio, flags);
2601 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2602 ext4_rereserve_cluster(inode, partial->lblk);
2604 partial->state = initial;
2607 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2608 pblk = ext4_ext_pblock(ex) + ee_len - num;
2611 * We free the partial cluster at the end of the extent (if any),
2612 * unless the cluster is used by another extent (partial_cluster
2613 * state is nofree). If a partial cluster exists here, it must be
2614 * shared with the last block in the extent.
2616 flags = get_default_free_blocks_flags(inode);
2618 /* partial, left end cluster aligned, right end unaligned */
2619 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2620 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2621 (partial->state != nofree)) {
2622 if (ext4_is_pending(inode, to))
2623 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2624 ext4_free_blocks(handle, inode, NULL,
2625 EXT4_PBLK_CMASK(sbi, last_pblk),
2626 sbi->s_cluster_ratio, flags);
2627 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2628 ext4_rereserve_cluster(inode, to);
2629 partial->state = initial;
2630 flags = get_default_free_blocks_flags(inode);
2633 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2636 * For bigalloc file systems, we never free a partial cluster
2637 * at the beginning of the extent. Instead, we check to see if we
2638 * need to free it on a subsequent call to ext4_remove_blocks,
2639 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2641 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2642 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2644 /* reset the partial cluster if we've freed past it */
2645 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2646 partial->state = initial;
2649 * If we've freed the entire extent but the beginning is not left
2650 * cluster aligned and is not marked as ineligible for freeing we
2651 * record the partial cluster at the beginning of the extent. It
2652 * wasn't freed by the preceding ext4_free_blocks() call, and we
2653 * need to look farther to the left to determine if it's to be freed
2654 * (not shared with another extent). Else, reset the partial
2655 * cluster - we're either done freeing or the beginning of the
2656 * extent is left cluster aligned.
2658 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2659 if (partial->state == initial) {
2660 partial->pclu = EXT4_B2C(sbi, pblk);
2661 partial->lblk = from;
2662 partial->state = tofree;
2665 partial->state = initial;
2672 * ext4_ext_rm_leaf() Removes the extents associated with the
2673 * blocks appearing between "start" and "end". Both "start"
2674 * and "end" must appear in the same extent or EIO is returned.
2676 * @handle: The journal handle
2677 * @inode: The files inode
2678 * @path: The path to the leaf
2679 * @partial_cluster: The cluster which we'll have to free if all extents
2680 * has been released from it. However, if this value is
2681 * negative, it's a cluster just to the right of the
2682 * punched region and it must not be freed.
2683 * @start: The first block to remove
2684 * @end: The last block to remove
2687 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2688 struct ext4_ext_path *path,
2689 struct partial_cluster *partial,
2690 ext4_lblk_t start, ext4_lblk_t end)
2692 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2693 int err = 0, correct_index = 0;
2694 int depth = ext_depth(inode), credits, revoke_credits;
2695 struct ext4_extent_header *eh;
2698 ext4_lblk_t ex_ee_block;
2699 unsigned short ex_ee_len;
2700 unsigned unwritten = 0;
2701 struct ext4_extent *ex;
2704 /* the header must be checked already in ext4_ext_remove_space() */
2705 ext_debug("truncate since %u in leaf to %u\n", start, end);
2706 if (!path[depth].p_hdr)
2707 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2708 eh = path[depth].p_hdr;
2709 if (unlikely(path[depth].p_hdr == NULL)) {
2710 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2711 return -EFSCORRUPTED;
2713 /* find where to start removing */
2714 ex = path[depth].p_ext;
2716 ex = EXT_LAST_EXTENT(eh);
2718 ex_ee_block = le32_to_cpu(ex->ee_block);
2719 ex_ee_len = ext4_ext_get_actual_len(ex);
2721 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2723 while (ex >= EXT_FIRST_EXTENT(eh) &&
2724 ex_ee_block + ex_ee_len > start) {
2726 if (ext4_ext_is_unwritten(ex))
2731 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2732 unwritten, ex_ee_len);
2733 path[depth].p_ext = ex;
2735 a = ex_ee_block > start ? ex_ee_block : start;
2736 b = ex_ee_block+ex_ee_len - 1 < end ?
2737 ex_ee_block+ex_ee_len - 1 : end;
2739 ext_debug(" border %u:%u\n", a, b);
2741 /* If this extent is beyond the end of the hole, skip it */
2742 if (end < ex_ee_block) {
2744 * We're going to skip this extent and move to another,
2745 * so note that its first cluster is in use to avoid
2746 * freeing it when removing blocks. Eventually, the
2747 * right edge of the truncated/punched region will
2748 * be just to the left.
2750 if (sbi->s_cluster_ratio > 1) {
2751 pblk = ext4_ext_pblock(ex);
2752 partial->pclu = EXT4_B2C(sbi, pblk);
2753 partial->state = nofree;
2756 ex_ee_block = le32_to_cpu(ex->ee_block);
2757 ex_ee_len = ext4_ext_get_actual_len(ex);
2759 } else if (b != ex_ee_block + ex_ee_len - 1) {
2760 EXT4_ERROR_INODE(inode,
2761 "can not handle truncate %u:%u "
2763 start, end, ex_ee_block,
2764 ex_ee_block + ex_ee_len - 1);
2765 err = -EFSCORRUPTED;
2767 } else if (a != ex_ee_block) {
2768 /* remove tail of the extent */
2769 num = a - ex_ee_block;
2771 /* remove whole extent: excellent! */
2775 * 3 for leaf, sb, and inode plus 2 (bmap and group
2776 * descriptor) for each block group; assume two block
2777 * groups plus ex_ee_len/blocks_per_block_group for
2780 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2781 if (ex == EXT_FIRST_EXTENT(eh)) {
2783 credits += (ext_depth(inode)) + 1;
2785 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2787 * We may end up freeing some index blocks and data from the
2788 * punched range. Note that partial clusters are accounted for
2789 * by ext4_free_data_revoke_credits().
2792 ext4_free_metadata_revoke_credits(inode->i_sb,
2794 ext4_free_data_revoke_credits(inode, b - a + 1);
2796 err = ext4_datasem_ensure_credits(handle, inode, credits,
2797 credits, revoke_credits);
2804 err = ext4_ext_get_access(handle, inode, path + depth);
2808 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2813 /* this extent is removed; mark slot entirely unused */
2814 ext4_ext_store_pblock(ex, 0);
2816 ex->ee_len = cpu_to_le16(num);
2818 * Do not mark unwritten if all the blocks in the
2819 * extent have been removed.
2821 if (unwritten && num)
2822 ext4_ext_mark_unwritten(ex);
2824 * If the extent was completely released,
2825 * we need to remove it from the leaf
2828 if (end != EXT_MAX_BLOCKS - 1) {
2830 * For hole punching, we need to scoot all the
2831 * extents up when an extent is removed so that
2832 * we dont have blank extents in the middle
2834 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2835 sizeof(struct ext4_extent));
2837 /* Now get rid of the one at the end */
2838 memset(EXT_LAST_EXTENT(eh), 0,
2839 sizeof(struct ext4_extent));
2841 le16_add_cpu(&eh->eh_entries, -1);
2844 err = ext4_ext_dirty(handle, inode, path + depth);
2848 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2849 ext4_ext_pblock(ex));
2851 ex_ee_block = le32_to_cpu(ex->ee_block);
2852 ex_ee_len = ext4_ext_get_actual_len(ex);
2855 if (correct_index && eh->eh_entries)
2856 err = ext4_ext_correct_indexes(handle, inode, path);
2859 * If there's a partial cluster and at least one extent remains in
2860 * the leaf, free the partial cluster if it isn't shared with the
2861 * current extent. If it is shared with the current extent
2862 * we reset the partial cluster because we've reached the start of the
2863 * truncated/punched region and we're done removing blocks.
2865 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2866 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2867 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2868 int flags = get_default_free_blocks_flags(inode);
2870 if (ext4_is_pending(inode, partial->lblk))
2871 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2872 ext4_free_blocks(handle, inode, NULL,
2873 EXT4_C2B(sbi, partial->pclu),
2874 sbi->s_cluster_ratio, flags);
2875 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2876 ext4_rereserve_cluster(inode, partial->lblk);
2878 partial->state = initial;
2881 /* if this leaf is free, then we should
2882 * remove it from index block above */
2883 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2884 err = ext4_ext_rm_idx(handle, inode, path, depth);
2891 * ext4_ext_more_to_rm:
2892 * returns 1 if current index has to be freed (even partial)
2895 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2897 BUG_ON(path->p_idx == NULL);
2899 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2903 * if truncate on deeper level happened, it wasn't partial,
2904 * so we have to consider current index for truncation
2906 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2911 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2914 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2915 int depth = ext_depth(inode);
2916 struct ext4_ext_path *path = NULL;
2917 struct partial_cluster partial;
2923 partial.state = initial;
2925 ext_debug("truncate since %u to %u\n", start, end);
2927 /* probably first extent we're gonna free will be last in block */
2928 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2930 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2932 return PTR_ERR(handle);
2935 trace_ext4_ext_remove_space(inode, start, end, depth);
2938 * Check if we are removing extents inside the extent tree. If that
2939 * is the case, we are going to punch a hole inside the extent tree
2940 * so we have to check whether we need to split the extent covering
2941 * the last block to remove so we can easily remove the part of it
2942 * in ext4_ext_rm_leaf().
2944 if (end < EXT_MAX_BLOCKS - 1) {
2945 struct ext4_extent *ex;
2946 ext4_lblk_t ee_block, ex_end, lblk;
2949 /* find extent for or closest extent to this block */
2950 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2952 ext4_journal_stop(handle);
2953 return PTR_ERR(path);
2955 depth = ext_depth(inode);
2956 /* Leaf not may not exist only if inode has no blocks at all */
2957 ex = path[depth].p_ext;
2960 EXT4_ERROR_INODE(inode,
2961 "path[%d].p_hdr == NULL",
2963 err = -EFSCORRUPTED;
2968 ee_block = le32_to_cpu(ex->ee_block);
2969 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2972 * See if the last block is inside the extent, if so split
2973 * the extent at 'end' block so we can easily remove the
2974 * tail of the first part of the split extent in
2975 * ext4_ext_rm_leaf().
2977 if (end >= ee_block && end < ex_end) {
2980 * If we're going to split the extent, note that
2981 * the cluster containing the block after 'end' is
2982 * in use to avoid freeing it when removing blocks.
2984 if (sbi->s_cluster_ratio > 1) {
2985 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2986 partial.pclu = EXT4_B2C(sbi, pblk);
2987 partial.state = nofree;
2991 * Split the extent in two so that 'end' is the last
2992 * block in the first new extent. Also we should not
2993 * fail removing space due to ENOSPC so try to use
2994 * reserved block if that happens.
2996 err = ext4_force_split_extent_at(handle, inode, &path,
3001 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
3002 partial.state == initial) {
3004 * If we're punching, there's an extent to the right.
3005 * If the partial cluster hasn't been set, set it to
3006 * that extent's first cluster and its state to nofree
3007 * so it won't be freed should it contain blocks to be
3008 * removed. If it's already set (tofree/nofree), we're
3009 * retrying and keep the original partial cluster info
3010 * so a cluster marked tofree as a result of earlier
3011 * extent removal is not lost.
3014 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3019 partial.pclu = EXT4_B2C(sbi, pblk);
3020 partial.state = nofree;
3025 * We start scanning from right side, freeing all the blocks
3026 * after i_size and walking into the tree depth-wise.
3028 depth = ext_depth(inode);
3033 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3035 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3038 ext4_journal_stop(handle);
3041 path[0].p_maxdepth = path[0].p_depth = depth;
3042 path[0].p_hdr = ext_inode_hdr(inode);
3045 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3046 err = -EFSCORRUPTED;
3052 while (i >= 0 && err == 0) {
3054 /* this is leaf block */
3055 err = ext4_ext_rm_leaf(handle, inode, path,
3056 &partial, start, end);
3057 /* root level has p_bh == NULL, brelse() eats this */
3058 brelse(path[i].p_bh);
3059 path[i].p_bh = NULL;
3064 /* this is index block */
3065 if (!path[i].p_hdr) {
3066 ext_debug("initialize header\n");
3067 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3070 if (!path[i].p_idx) {
3071 /* this level hasn't been touched yet */
3072 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3073 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3074 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3076 le16_to_cpu(path[i].p_hdr->eh_entries));
3078 /* we were already here, see at next index */
3082 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3083 i, EXT_FIRST_INDEX(path[i].p_hdr),
3085 if (ext4_ext_more_to_rm(path + i)) {
3086 struct buffer_head *bh;
3087 /* go to the next level */
3088 ext_debug("move to level %d (block %llu)\n",
3089 i + 1, ext4_idx_pblock(path[i].p_idx));
3090 memset(path + i + 1, 0, sizeof(*path));
3091 bh = read_extent_tree_block(inode,
3092 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3095 /* should we reset i_size? */
3099 /* Yield here to deal with large extent trees.
3100 * Should be a no-op if we did IO above. */
3102 if (WARN_ON(i + 1 > depth)) {
3103 err = -EFSCORRUPTED;
3106 path[i + 1].p_bh = bh;
3108 /* save actual number of indexes since this
3109 * number is changed at the next iteration */
3110 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3113 /* we finished processing this index, go up */
3114 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3115 /* index is empty, remove it;
3116 * handle must be already prepared by the
3117 * truncatei_leaf() */
3118 err = ext4_ext_rm_idx(handle, inode, path, i);
3120 /* root level has p_bh == NULL, brelse() eats this */
3121 brelse(path[i].p_bh);
3122 path[i].p_bh = NULL;
3124 ext_debug("return to level %d\n", i);
3128 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3129 path->p_hdr->eh_entries);
3132 * if there's a partial cluster and we have removed the first extent
3133 * in the file, then we also free the partial cluster, if any
3135 if (partial.state == tofree && err == 0) {
3136 int flags = get_default_free_blocks_flags(inode);
3138 if (ext4_is_pending(inode, partial.lblk))
3139 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3140 ext4_free_blocks(handle, inode, NULL,
3141 EXT4_C2B(sbi, partial.pclu),
3142 sbi->s_cluster_ratio, flags);
3143 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3144 ext4_rereserve_cluster(inode, partial.lblk);
3145 partial.state = initial;
3148 /* TODO: flexible tree reduction should be here */
3149 if (path->p_hdr->eh_entries == 0) {
3151 * truncate to zero freed all the tree,
3152 * so we need to correct eh_depth
3154 err = ext4_ext_get_access(handle, inode, path);
3156 ext_inode_hdr(inode)->eh_depth = 0;
3157 ext_inode_hdr(inode)->eh_max =
3158 cpu_to_le16(ext4_ext_space_root(inode, 0));
3159 err = ext4_ext_dirty(handle, inode, path);
3163 ext4_ext_drop_refs(path);
3168 ext4_journal_stop(handle);
3174 * called at mount time
3176 void ext4_ext_init(struct super_block *sb)
3179 * possible initialization would be here
3182 if (ext4_has_feature_extents(sb)) {
3183 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3184 printk(KERN_INFO "EXT4-fs: file extents enabled"
3185 #ifdef AGGRESSIVE_TEST
3186 ", aggressive tests"
3188 #ifdef CHECK_BINSEARCH
3191 #ifdef EXTENTS_STATS
3196 #ifdef EXTENTS_STATS
3197 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3198 EXT4_SB(sb)->s_ext_min = 1 << 30;
3199 EXT4_SB(sb)->s_ext_max = 0;
3205 * called at umount time
3207 void ext4_ext_release(struct super_block *sb)
3209 if (!ext4_has_feature_extents(sb))
3212 #ifdef EXTENTS_STATS
3213 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3214 struct ext4_sb_info *sbi = EXT4_SB(sb);
3215 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3216 sbi->s_ext_blocks, sbi->s_ext_extents,
3217 sbi->s_ext_blocks / sbi->s_ext_extents);
3218 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3219 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3224 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3226 ext4_lblk_t ee_block;
3227 ext4_fsblk_t ee_pblock;
3228 unsigned int ee_len;
3230 ee_block = le32_to_cpu(ex->ee_block);
3231 ee_len = ext4_ext_get_actual_len(ex);
3232 ee_pblock = ext4_ext_pblock(ex);
3237 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3238 EXTENT_STATUS_WRITTEN);
3241 /* FIXME!! we need to try to merge to left or right after zero-out */
3242 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3244 ext4_fsblk_t ee_pblock;
3245 unsigned int ee_len;
3247 ee_len = ext4_ext_get_actual_len(ex);
3248 ee_pblock = ext4_ext_pblock(ex);
3249 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3254 * ext4_split_extent_at() splits an extent at given block.
3256 * @handle: the journal handle
3257 * @inode: the file inode
3258 * @path: the path to the extent
3259 * @split: the logical block where the extent is splitted.
3260 * @split_flags: indicates if the extent could be zeroout if split fails, and
3261 * the states(init or unwritten) of new extents.
3262 * @flags: flags used to insert new extent to extent tree.
3265 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3266 * of which are deterimined by split_flag.
3268 * There are two cases:
3269 * a> the extent are splitted into two extent.
3270 * b> split is not needed, and just mark the extent.
3272 * return 0 on success.
3274 static int ext4_split_extent_at(handle_t *handle,
3275 struct inode *inode,
3276 struct ext4_ext_path **ppath,
3281 struct ext4_ext_path *path = *ppath;
3282 ext4_fsblk_t newblock;
3283 ext4_lblk_t ee_block;
3284 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3285 struct ext4_extent *ex2 = NULL;
3286 unsigned int ee_len, depth;
3289 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3290 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3292 ext_debug("ext4_split_extents_at: inode %lu, logical"
3293 "block %llu\n", inode->i_ino, (unsigned long long)split);
3295 ext4_ext_show_leaf(inode, path);
3297 depth = ext_depth(inode);
3298 ex = path[depth].p_ext;
3299 ee_block = le32_to_cpu(ex->ee_block);
3300 ee_len = ext4_ext_get_actual_len(ex);
3301 newblock = split - ee_block + ext4_ext_pblock(ex);
3303 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3304 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3305 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3306 EXT4_EXT_MARK_UNWRIT1 |
3307 EXT4_EXT_MARK_UNWRIT2));
3309 err = ext4_ext_get_access(handle, inode, path + depth);
3313 if (split == ee_block) {
3315 * case b: block @split is the block that the extent begins with
3316 * then we just change the state of the extent, and splitting
3319 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3320 ext4_ext_mark_unwritten(ex);
3322 ext4_ext_mark_initialized(ex);
3324 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3325 ext4_ext_try_to_merge(handle, inode, path, ex);
3327 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3332 memcpy(&orig_ex, ex, sizeof(orig_ex));
3333 ex->ee_len = cpu_to_le16(split - ee_block);
3334 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3335 ext4_ext_mark_unwritten(ex);
3338 * path may lead to new leaf, not to original leaf any more
3339 * after ext4_ext_insert_extent() returns,
3341 err = ext4_ext_dirty(handle, inode, path + depth);
3343 goto fix_extent_len;
3346 ex2->ee_block = cpu_to_le32(split);
3347 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3348 ext4_ext_store_pblock(ex2, newblock);
3349 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3350 ext4_ext_mark_unwritten(ex2);
3352 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3353 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3354 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3355 if (split_flag & EXT4_EXT_DATA_VALID1) {
3356 err = ext4_ext_zeroout(inode, ex2);
3357 zero_ex.ee_block = ex2->ee_block;
3358 zero_ex.ee_len = cpu_to_le16(
3359 ext4_ext_get_actual_len(ex2));
3360 ext4_ext_store_pblock(&zero_ex,
3361 ext4_ext_pblock(ex2));
3363 err = ext4_ext_zeroout(inode, ex);
3364 zero_ex.ee_block = ex->ee_block;
3365 zero_ex.ee_len = cpu_to_le16(
3366 ext4_ext_get_actual_len(ex));
3367 ext4_ext_store_pblock(&zero_ex,
3368 ext4_ext_pblock(ex));
3371 err = ext4_ext_zeroout(inode, &orig_ex);
3372 zero_ex.ee_block = orig_ex.ee_block;
3373 zero_ex.ee_len = cpu_to_le16(
3374 ext4_ext_get_actual_len(&orig_ex));
3375 ext4_ext_store_pblock(&zero_ex,
3376 ext4_ext_pblock(&orig_ex));
3380 goto fix_extent_len;
3381 /* update the extent length and mark as initialized */
3382 ex->ee_len = cpu_to_le16(ee_len);
3383 ext4_ext_try_to_merge(handle, inode, path, ex);
3384 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3386 goto fix_extent_len;
3388 /* update extent status tree */
3389 err = ext4_zeroout_es(inode, &zero_ex);
3393 goto fix_extent_len;
3396 ext4_ext_show_leaf(inode, path);
3400 ex->ee_len = orig_ex.ee_len;
3401 ext4_ext_dirty(handle, inode, path + path->p_depth);
3406 * ext4_split_extents() splits an extent and mark extent which is covered
3407 * by @map as split_flags indicates
3409 * It may result in splitting the extent into multiple extents (up to three)
3410 * There are three possibilities:
3411 * a> There is no split required
3412 * b> Splits in two extents: Split is happening at either end of the extent
3413 * c> Splits in three extents: Somone is splitting in middle of the extent
3416 static int ext4_split_extent(handle_t *handle,
3417 struct inode *inode,
3418 struct ext4_ext_path **ppath,
3419 struct ext4_map_blocks *map,
3423 struct ext4_ext_path *path = *ppath;
3424 ext4_lblk_t ee_block;
3425 struct ext4_extent *ex;
3426 unsigned int ee_len, depth;
3429 int split_flag1, flags1;
3430 int allocated = map->m_len;
3432 depth = ext_depth(inode);
3433 ex = path[depth].p_ext;
3434 ee_block = le32_to_cpu(ex->ee_block);
3435 ee_len = ext4_ext_get_actual_len(ex);
3436 unwritten = ext4_ext_is_unwritten(ex);
3438 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3439 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3440 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3442 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3443 EXT4_EXT_MARK_UNWRIT2;
3444 if (split_flag & EXT4_EXT_DATA_VALID2)
3445 split_flag1 |= EXT4_EXT_DATA_VALID1;
3446 err = ext4_split_extent_at(handle, inode, ppath,
3447 map->m_lblk + map->m_len, split_flag1, flags1);
3451 allocated = ee_len - (map->m_lblk - ee_block);
3454 * Update path is required because previous ext4_split_extent_at() may
3455 * result in split of original leaf or extent zeroout.
3457 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3459 return PTR_ERR(path);
3460 depth = ext_depth(inode);
3461 ex = path[depth].p_ext;
3463 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3464 (unsigned long) map->m_lblk);
3465 return -EFSCORRUPTED;
3467 unwritten = ext4_ext_is_unwritten(ex);
3470 if (map->m_lblk >= ee_block) {
3471 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3473 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3474 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3475 EXT4_EXT_MARK_UNWRIT2);
3477 err = ext4_split_extent_at(handle, inode, ppath,
3478 map->m_lblk, split_flag1, flags);
3483 ext4_ext_show_leaf(inode, path);
3485 return err ? err : allocated;
3489 * This function is called by ext4_ext_map_blocks() if someone tries to write
3490 * to an unwritten extent. It may result in splitting the unwritten
3491 * extent into multiple extents (up to three - one initialized and two
3493 * There are three possibilities:
3494 * a> There is no split required: Entire extent should be initialized
3495 * b> Splits in two extents: Write is happening at either end of the extent
3496 * c> Splits in three extents: Somone is writing in middle of the extent
3499 * - The extent pointed to by 'path' is unwritten.
3500 * - The extent pointed to by 'path' contains a superset
3501 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3503 * Post-conditions on success:
3504 * - the returned value is the number of blocks beyond map->l_lblk
3505 * that are allocated and initialized.
3506 * It is guaranteed to be >= map->m_len.
3508 static int ext4_ext_convert_to_initialized(handle_t *handle,
3509 struct inode *inode,
3510 struct ext4_map_blocks *map,
3511 struct ext4_ext_path **ppath,
3514 struct ext4_ext_path *path = *ppath;
3515 struct ext4_sb_info *sbi;
3516 struct ext4_extent_header *eh;
3517 struct ext4_map_blocks split_map;
3518 struct ext4_extent zero_ex1, zero_ex2;
3519 struct ext4_extent *ex, *abut_ex;
3520 ext4_lblk_t ee_block, eof_block;
3521 unsigned int ee_len, depth, map_len = map->m_len;
3522 int allocated = 0, max_zeroout = 0;
3524 int split_flag = EXT4_EXT_DATA_VALID2;
3526 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3527 "block %llu, max_blocks %u\n", inode->i_ino,
3528 (unsigned long long)map->m_lblk, map_len);
3530 sbi = EXT4_SB(inode->i_sb);
3531 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3532 inode->i_sb->s_blocksize_bits;
3533 if (eof_block < map->m_lblk + map_len)
3534 eof_block = map->m_lblk + map_len;
3536 depth = ext_depth(inode);
3537 eh = path[depth].p_hdr;
3538 ex = path[depth].p_ext;
3539 ee_block = le32_to_cpu(ex->ee_block);
3540 ee_len = ext4_ext_get_actual_len(ex);
3541 zero_ex1.ee_len = 0;
3542 zero_ex2.ee_len = 0;
3544 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3546 /* Pre-conditions */
3547 BUG_ON(!ext4_ext_is_unwritten(ex));
3548 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3551 * Attempt to transfer newly initialized blocks from the currently
3552 * unwritten extent to its neighbor. This is much cheaper
3553 * than an insertion followed by a merge as those involve costly
3554 * memmove() calls. Transferring to the left is the common case in
3555 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3556 * followed by append writes.
3558 * Limitations of the current logic:
3559 * - L1: we do not deal with writes covering the whole extent.
3560 * This would require removing the extent if the transfer
3562 * - L2: we only attempt to merge with an extent stored in the
3563 * same extent tree node.
3565 if ((map->m_lblk == ee_block) &&
3566 /* See if we can merge left */
3567 (map_len < ee_len) && /*L1*/
3568 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3569 ext4_lblk_t prev_lblk;
3570 ext4_fsblk_t prev_pblk, ee_pblk;
3571 unsigned int prev_len;
3574 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3575 prev_len = ext4_ext_get_actual_len(abut_ex);
3576 prev_pblk = ext4_ext_pblock(abut_ex);
3577 ee_pblk = ext4_ext_pblock(ex);
3580 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3581 * upon those conditions:
3582 * - C1: abut_ex is initialized,
3583 * - C2: abut_ex is logically abutting ex,
3584 * - C3: abut_ex is physically abutting ex,
3585 * - C4: abut_ex can receive the additional blocks without
3586 * overflowing the (initialized) length limit.
3588 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3589 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3590 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3591 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3592 err = ext4_ext_get_access(handle, inode, path + depth);
3596 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3599 /* Shift the start of ex by 'map_len' blocks */
3600 ex->ee_block = cpu_to_le32(ee_block + map_len);
3601 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3602 ex->ee_len = cpu_to_le16(ee_len - map_len);
3603 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3605 /* Extend abut_ex by 'map_len' blocks */
3606 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3608 /* Result: number of initialized blocks past m_lblk */
3609 allocated = map_len;
3611 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3612 (map_len < ee_len) && /*L1*/
3613 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3614 /* See if we can merge right */
3615 ext4_lblk_t next_lblk;
3616 ext4_fsblk_t next_pblk, ee_pblk;
3617 unsigned int next_len;
3620 next_lblk = le32_to_cpu(abut_ex->ee_block);
3621 next_len = ext4_ext_get_actual_len(abut_ex);
3622 next_pblk = ext4_ext_pblock(abut_ex);
3623 ee_pblk = ext4_ext_pblock(ex);
3626 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3627 * upon those conditions:
3628 * - C1: abut_ex is initialized,
3629 * - C2: abut_ex is logically abutting ex,
3630 * - C3: abut_ex is physically abutting ex,
3631 * - C4: abut_ex can receive the additional blocks without
3632 * overflowing the (initialized) length limit.
3634 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3635 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3636 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3637 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3638 err = ext4_ext_get_access(handle, inode, path + depth);
3642 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3645 /* Shift the start of abut_ex by 'map_len' blocks */
3646 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3647 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3648 ex->ee_len = cpu_to_le16(ee_len - map_len);
3649 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3651 /* Extend abut_ex by 'map_len' blocks */
3652 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3654 /* Result: number of initialized blocks past m_lblk */
3655 allocated = map_len;
3659 /* Mark the block containing both extents as dirty */
3660 ext4_ext_dirty(handle, inode, path + depth);
3662 /* Update path to point to the right extent */
3663 path[depth].p_ext = abut_ex;
3666 allocated = ee_len - (map->m_lblk - ee_block);
3668 WARN_ON(map->m_lblk < ee_block);
3670 * It is safe to convert extent to initialized via explicit
3671 * zeroout only if extent is fully inside i_size or new_size.
3673 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3675 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3676 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3677 (inode->i_sb->s_blocksize_bits - 10);
3681 * 1. split the extent into three extents.
3682 * 2. split the extent into two extents, zeroout the head of the first
3684 * 3. split the extent into two extents, zeroout the tail of the second
3686 * 4. split the extent into two extents with out zeroout.
3687 * 5. no splitting needed, just possibly zeroout the head and / or the
3688 * tail of the extent.
3690 split_map.m_lblk = map->m_lblk;
3691 split_map.m_len = map->m_len;
3693 if (max_zeroout && (allocated > split_map.m_len)) {
3694 if (allocated <= max_zeroout) {
3697 cpu_to_le32(split_map.m_lblk +
3700 cpu_to_le16(allocated - split_map.m_len);
3701 ext4_ext_store_pblock(&zero_ex1,
3702 ext4_ext_pblock(ex) + split_map.m_lblk +
3703 split_map.m_len - ee_block);
3704 err = ext4_ext_zeroout(inode, &zero_ex1);
3707 split_map.m_len = allocated;
3709 if (split_map.m_lblk - ee_block + split_map.m_len <
3712 if (split_map.m_lblk != ee_block) {
3713 zero_ex2.ee_block = ex->ee_block;
3714 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3716 ext4_ext_store_pblock(&zero_ex2,
3717 ext4_ext_pblock(ex));
3718 err = ext4_ext_zeroout(inode, &zero_ex2);
3723 split_map.m_len += split_map.m_lblk - ee_block;
3724 split_map.m_lblk = ee_block;
3725 allocated = map->m_len;
3729 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3734 /* If we have gotten a failure, don't zero out status tree */
3736 err = ext4_zeroout_es(inode, &zero_ex1);
3738 err = ext4_zeroout_es(inode, &zero_ex2);
3740 return err ? err : allocated;
3744 * This function is called by ext4_ext_map_blocks() from
3745 * ext4_get_blocks_dio_write() when DIO to write
3746 * to an unwritten extent.
3748 * Writing to an unwritten extent may result in splitting the unwritten
3749 * extent into multiple initialized/unwritten extents (up to three)
3750 * There are three possibilities:
3751 * a> There is no split required: Entire extent should be unwritten
3752 * b> Splits in two extents: Write is happening at either end of the extent
3753 * c> Splits in three extents: Somone is writing in middle of the extent
3755 * This works the same way in the case of initialized -> unwritten conversion.
3757 * One of more index blocks maybe needed if the extent tree grow after
3758 * the unwritten extent split. To prevent ENOSPC occur at the IO
3759 * complete, we need to split the unwritten extent before DIO submit
3760 * the IO. The unwritten extent called at this time will be split
3761 * into three unwritten extent(at most). After IO complete, the part
3762 * being filled will be convert to initialized by the end_io callback function
3763 * via ext4_convert_unwritten_extents().
3765 * Returns the size of unwritten extent to be written on success.
3767 static int ext4_split_convert_extents(handle_t *handle,
3768 struct inode *inode,
3769 struct ext4_map_blocks *map,
3770 struct ext4_ext_path **ppath,
3773 struct ext4_ext_path *path = *ppath;
3774 ext4_lblk_t eof_block;
3775 ext4_lblk_t ee_block;
3776 struct ext4_extent *ex;
3777 unsigned int ee_len;
3778 int split_flag = 0, depth;
3780 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3781 __func__, inode->i_ino,
3782 (unsigned long long)map->m_lblk, map->m_len);
3784 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3785 inode->i_sb->s_blocksize_bits;
3786 if (eof_block < map->m_lblk + map->m_len)
3787 eof_block = map->m_lblk + map->m_len;
3789 * It is safe to convert extent to initialized via explicit
3790 * zeroout only if extent is fully insde i_size or new_size.
3792 depth = ext_depth(inode);
3793 ex = path[depth].p_ext;
3794 ee_block = le32_to_cpu(ex->ee_block);
3795 ee_len = ext4_ext_get_actual_len(ex);
3797 /* Convert to unwritten */
3798 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3799 split_flag |= EXT4_EXT_DATA_VALID1;
3800 /* Convert to initialized */
3801 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3802 split_flag |= ee_block + ee_len <= eof_block ?
3803 EXT4_EXT_MAY_ZEROOUT : 0;
3804 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3806 flags |= EXT4_GET_BLOCKS_PRE_IO;
3807 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3810 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3811 struct inode *inode,
3812 struct ext4_map_blocks *map,
3813 struct ext4_ext_path **ppath)
3815 struct ext4_ext_path *path = *ppath;
3816 struct ext4_extent *ex;
3817 ext4_lblk_t ee_block;
3818 unsigned int ee_len;
3822 depth = ext_depth(inode);
3823 ex = path[depth].p_ext;
3824 ee_block = le32_to_cpu(ex->ee_block);
3825 ee_len = ext4_ext_get_actual_len(ex);
3827 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3828 "block %llu, max_blocks %u\n", inode->i_ino,
3829 (unsigned long long)ee_block, ee_len);
3831 /* If extent is larger than requested it is a clear sign that we still
3832 * have some extent state machine issues left. So extent_split is still
3834 * TODO: Once all related issues will be fixed this situation should be
3837 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3838 #ifdef CONFIG_EXT4_DEBUG
3839 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3840 " len %u; IO logical block %llu, len %u",
3841 inode->i_ino, (unsigned long long)ee_block, ee_len,
3842 (unsigned long long)map->m_lblk, map->m_len);
3844 err = ext4_split_convert_extents(handle, inode, map, ppath,
3845 EXT4_GET_BLOCKS_CONVERT);
3848 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3850 return PTR_ERR(path);
3851 depth = ext_depth(inode);
3852 ex = path[depth].p_ext;
3855 err = ext4_ext_get_access(handle, inode, path + depth);
3858 /* first mark the extent as initialized */
3859 ext4_ext_mark_initialized(ex);
3861 /* note: ext4_ext_correct_indexes() isn't needed here because
3862 * borders are not changed
3864 ext4_ext_try_to_merge(handle, inode, path, ex);
3866 /* Mark modified extent as dirty */
3867 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3869 ext4_ext_show_leaf(inode, path);
3874 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3876 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3878 struct ext4_ext_path *path,
3882 struct ext4_extent_header *eh;
3883 struct ext4_extent *last_ex;
3885 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3888 depth = ext_depth(inode);
3889 eh = path[depth].p_hdr;
3892 * We're going to remove EOFBLOCKS_FL entirely in future so we
3893 * do not care for this case anymore. Simply remove the flag
3894 * if there are no extents.
3896 if (unlikely(!eh->eh_entries))
3898 last_ex = EXT_LAST_EXTENT(eh);
3900 * We should clear the EOFBLOCKS_FL flag if we are writing the
3901 * last block in the last extent in the file. We test this by
3902 * first checking to see if the caller to
3903 * ext4_ext_get_blocks() was interested in the last block (or
3904 * a block beyond the last block) in the current extent. If
3905 * this turns out to be false, we can bail out from this
3906 * function immediately.
3908 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3909 ext4_ext_get_actual_len(last_ex))
3912 * If the caller does appear to be planning to write at or
3913 * beyond the end of the current extent, we then test to see
3914 * if the current extent is the last extent in the file, by
3915 * checking to make sure it was reached via the rightmost node
3916 * at each level of the tree.
3918 for (i = depth-1; i >= 0; i--)
3919 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3922 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3923 return ext4_mark_inode_dirty(handle, inode);
3927 convert_initialized_extent(handle_t *handle, struct inode *inode,
3928 struct ext4_map_blocks *map,
3929 struct ext4_ext_path **ppath,
3930 unsigned int allocated)
3932 struct ext4_ext_path *path = *ppath;
3933 struct ext4_extent *ex;
3934 ext4_lblk_t ee_block;
3935 unsigned int ee_len;
3940 * Make sure that the extent is no bigger than we support with
3943 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3944 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3946 depth = ext_depth(inode);
3947 ex = path[depth].p_ext;
3948 ee_block = le32_to_cpu(ex->ee_block);
3949 ee_len = ext4_ext_get_actual_len(ex);
3951 ext_debug("%s: inode %lu, logical"
3952 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3953 (unsigned long long)ee_block, ee_len);
3955 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3956 err = ext4_split_convert_extents(handle, inode, map, ppath,
3957 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3960 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3962 return PTR_ERR(path);
3963 depth = ext_depth(inode);
3964 ex = path[depth].p_ext;
3966 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3967 (unsigned long) map->m_lblk);
3968 return -EFSCORRUPTED;
3972 err = ext4_ext_get_access(handle, inode, path + depth);
3975 /* first mark the extent as unwritten */
3976 ext4_ext_mark_unwritten(ex);
3978 /* note: ext4_ext_correct_indexes() isn't needed here because
3979 * borders are not changed
3981 ext4_ext_try_to_merge(handle, inode, path, ex);
3983 /* Mark modified extent as dirty */
3984 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3987 ext4_ext_show_leaf(inode, path);
3989 ext4_update_inode_fsync_trans(handle, inode, 1);
3990 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3993 map->m_flags |= EXT4_MAP_UNWRITTEN;
3994 if (allocated > map->m_len)
3995 allocated = map->m_len;
3996 map->m_len = allocated;
4001 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4002 struct ext4_map_blocks *map,
4003 struct ext4_ext_path **ppath, int flags,
4004 unsigned int allocated, ext4_fsblk_t newblock)
4006 struct ext4_ext_path *path = *ppath;
4010 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4011 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4012 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4014 ext4_ext_show_leaf(inode, path);
4017 * When writing into unwritten space, we should not fail to
4018 * allocate metadata blocks for the new extent block if needed.
4020 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4022 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4023 allocated, newblock);
4025 /* get_block() before submit the IO, split the extent */
4026 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4027 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4028 flags | EXT4_GET_BLOCKS_CONVERT);
4031 map->m_flags |= EXT4_MAP_UNWRITTEN;
4034 /* IO end_io complete, convert the filled extent to written */
4035 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4036 if (flags & EXT4_GET_BLOCKS_ZERO) {
4037 if (allocated > map->m_len)
4038 allocated = map->m_len;
4039 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4044 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4047 ext4_update_inode_fsync_trans(handle, inode, 1);
4048 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4052 map->m_flags |= EXT4_MAP_MAPPED;
4053 map->m_pblk = newblock;
4054 if (allocated > map->m_len)
4055 allocated = map->m_len;
4056 map->m_len = allocated;
4059 /* buffered IO case */
4061 * repeat fallocate creation request
4062 * we already have an unwritten extent
4064 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4065 map->m_flags |= EXT4_MAP_UNWRITTEN;
4069 /* buffered READ or buffered write_begin() lookup */
4070 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4072 * We have blocks reserved already. We
4073 * return allocated blocks so that delalloc
4074 * won't do block reservation for us. But
4075 * the buffer head will be unmapped so that
4076 * a read from the block returns 0s.
4078 map->m_flags |= EXT4_MAP_UNWRITTEN;
4082 /* buffered write, writepage time, convert*/
4083 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4085 ext4_update_inode_fsync_trans(handle, inode, 1);
4092 map->m_flags |= EXT4_MAP_NEW;
4093 if (allocated > map->m_len)
4094 allocated = map->m_len;
4095 map->m_len = allocated;
4098 map->m_flags |= EXT4_MAP_MAPPED;
4099 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4100 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4106 if (allocated > map->m_len)
4107 allocated = map->m_len;
4108 ext4_ext_show_leaf(inode, path);
4109 map->m_pblk = newblock;
4110 map->m_len = allocated;
4112 return err ? err : allocated;
4116 * get_implied_cluster_alloc - check to see if the requested
4117 * allocation (in the map structure) overlaps with a cluster already
4118 * allocated in an extent.
4119 * @sb The filesystem superblock structure
4120 * @map The requested lblk->pblk mapping
4121 * @ex The extent structure which might contain an implied
4122 * cluster allocation
4124 * This function is called by ext4_ext_map_blocks() after we failed to
4125 * find blocks that were already in the inode's extent tree. Hence,
4126 * we know that the beginning of the requested region cannot overlap
4127 * the extent from the inode's extent tree. There are three cases we
4128 * want to catch. The first is this case:
4130 * |--- cluster # N--|
4131 * |--- extent ---| |---- requested region ---|
4134 * The second case that we need to test for is this one:
4136 * |--------- cluster # N ----------------|
4137 * |--- requested region --| |------- extent ----|
4138 * |=======================|
4140 * The third case is when the requested region lies between two extents
4141 * within the same cluster:
4142 * |------------- cluster # N-------------|
4143 * |----- ex -----| |---- ex_right ----|
4144 * |------ requested region ------|
4145 * |================|
4147 * In each of the above cases, we need to set the map->m_pblk and
4148 * map->m_len so it corresponds to the return the extent labelled as
4149 * "|====|" from cluster #N, since it is already in use for data in
4150 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4151 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4152 * as a new "allocated" block region. Otherwise, we will return 0 and
4153 * ext4_ext_map_blocks() will then allocate one or more new clusters
4154 * by calling ext4_mb_new_blocks().
4156 static int get_implied_cluster_alloc(struct super_block *sb,
4157 struct ext4_map_blocks *map,
4158 struct ext4_extent *ex,
4159 struct ext4_ext_path *path)
4161 struct ext4_sb_info *sbi = EXT4_SB(sb);
4162 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4163 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4164 ext4_lblk_t rr_cluster_start;
4165 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4166 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4167 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4169 /* The extent passed in that we are trying to match */
4170 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4171 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4173 /* The requested region passed into ext4_map_blocks() */
4174 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4176 if ((rr_cluster_start == ex_cluster_end) ||
4177 (rr_cluster_start == ex_cluster_start)) {
4178 if (rr_cluster_start == ex_cluster_end)
4179 ee_start += ee_len - 1;
4180 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4181 map->m_len = min(map->m_len,
4182 (unsigned) sbi->s_cluster_ratio - c_offset);
4184 * Check for and handle this case:
4186 * |--------- cluster # N-------------|
4187 * |------- extent ----|
4188 * |--- requested region ---|
4192 if (map->m_lblk < ee_block)
4193 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4196 * Check for the case where there is already another allocated
4197 * block to the right of 'ex' but before the end of the cluster.
4199 * |------------- cluster # N-------------|
4200 * |----- ex -----| |---- ex_right ----|
4201 * |------ requested region ------|
4202 * |================|
4204 if (map->m_lblk > ee_block) {
4205 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4206 map->m_len = min(map->m_len, next - map->m_lblk);
4209 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4213 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4219 * Block allocation/map/preallocation routine for extents based files
4222 * Need to be called with
4223 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4224 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4226 * return > 0, number of of blocks already mapped/allocated
4227 * if create == 0 and these are pre-allocated blocks
4228 * buffer head is unmapped
4229 * otherwise blocks are mapped
4231 * return = 0, if plain look up failed (blocks have not been allocated)
4232 * buffer head is unmapped
4234 * return < 0, error case.
4236 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4237 struct ext4_map_blocks *map, int flags)
4239 struct ext4_ext_path *path = NULL;
4240 struct ext4_extent newex, *ex, *ex2;
4241 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4242 ext4_fsblk_t newblock = 0;
4243 int free_on_err = 0, err = 0, depth, ret;
4244 unsigned int allocated = 0, offset = 0;
4245 unsigned int allocated_clusters = 0;
4246 struct ext4_allocation_request ar;
4247 ext4_lblk_t cluster_offset;
4248 bool map_from_cluster = false;
4250 ext_debug("blocks %u/%u requested for inode %lu\n",
4251 map->m_lblk, map->m_len, inode->i_ino);
4252 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4254 /* find extent for this block */
4255 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4257 err = PTR_ERR(path);
4262 depth = ext_depth(inode);
4265 * consistent leaf must not be empty;
4266 * this situation is possible, though, _during_ tree modification;
4267 * this is why assert can't be put in ext4_find_extent()
4269 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4270 EXT4_ERROR_INODE(inode, "bad extent address "
4271 "lblock: %lu, depth: %d pblock %lld",
4272 (unsigned long) map->m_lblk, depth,
4273 path[depth].p_block);
4274 err = -EFSCORRUPTED;
4278 ex = path[depth].p_ext;
4280 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4281 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4282 unsigned short ee_len;
4286 * unwritten extents are treated as holes, except that
4287 * we split out initialized portions during a write.
4289 ee_len = ext4_ext_get_actual_len(ex);
4291 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4293 /* if found extent covers block, simply return it */
4294 if (in_range(map->m_lblk, ee_block, ee_len)) {
4295 newblock = map->m_lblk - ee_block + ee_start;
4296 /* number of remaining blocks in the extent */
4297 allocated = ee_len - (map->m_lblk - ee_block);
4298 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4299 ee_block, ee_len, newblock);
4302 * If the extent is initialized check whether the
4303 * caller wants to convert it to unwritten.
4305 if ((!ext4_ext_is_unwritten(ex)) &&
4306 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4307 allocated = convert_initialized_extent(
4308 handle, inode, map, &path,
4311 } else if (!ext4_ext_is_unwritten(ex))
4314 ret = ext4_ext_handle_unwritten_extents(
4315 handle, inode, map, &path, flags,
4316 allocated, newblock);
4326 * requested block isn't allocated yet;
4327 * we couldn't try to create block if create flag is zero
4329 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4330 ext4_lblk_t hole_start, hole_len;
4332 hole_start = map->m_lblk;
4333 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4335 * put just found gap into cache to speed up
4336 * subsequent requests
4338 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4340 /* Update hole_len to reflect hole size after map->m_lblk */
4341 if (hole_start != map->m_lblk)
4342 hole_len -= map->m_lblk - hole_start;
4344 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4350 * Okay, we need to do block allocation.
4352 newex.ee_block = cpu_to_le32(map->m_lblk);
4353 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4356 * If we are doing bigalloc, check to see if the extent returned
4357 * by ext4_find_extent() implies a cluster we can use.
4359 if (cluster_offset && ex &&
4360 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4361 ar.len = allocated = map->m_len;
4362 newblock = map->m_pblk;
4363 map_from_cluster = true;
4364 goto got_allocated_blocks;
4367 /* find neighbour allocated blocks */
4368 ar.lleft = map->m_lblk;
4369 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4372 ar.lright = map->m_lblk;
4374 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4378 /* Check if the extent after searching to the right implies a
4379 * cluster we can use. */
4380 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4381 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4382 ar.len = allocated = map->m_len;
4383 newblock = map->m_pblk;
4384 map_from_cluster = true;
4385 goto got_allocated_blocks;
4389 * See if request is beyond maximum number of blocks we can have in
4390 * a single extent. For an initialized extent this limit is
4391 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4392 * EXT_UNWRITTEN_MAX_LEN.
4394 if (map->m_len > EXT_INIT_MAX_LEN &&
4395 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4396 map->m_len = EXT_INIT_MAX_LEN;
4397 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4398 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4399 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4401 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4402 newex.ee_len = cpu_to_le16(map->m_len);
4403 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4405 allocated = ext4_ext_get_actual_len(&newex);
4407 allocated = map->m_len;
4409 /* allocate new block */
4411 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4412 ar.logical = map->m_lblk;
4414 * We calculate the offset from the beginning of the cluster
4415 * for the logical block number, since when we allocate a
4416 * physical cluster, the physical block should start at the
4417 * same offset from the beginning of the cluster. This is
4418 * needed so that future calls to get_implied_cluster_alloc()
4421 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4422 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4424 ar.logical -= offset;
4425 if (S_ISREG(inode->i_mode))
4426 ar.flags = EXT4_MB_HINT_DATA;
4428 /* disable in-core preallocation for non-regular files */
4430 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4431 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4432 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4433 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4434 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4435 ar.flags |= EXT4_MB_USE_RESERVED;
4436 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4439 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4440 ar.goal, newblock, allocated);
4442 allocated_clusters = ar.len;
4443 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4444 if (ar.len > allocated)
4447 got_allocated_blocks:
4448 /* try to insert new extent into found leaf and return */
4449 ext4_ext_store_pblock(&newex, newblock + offset);
4450 newex.ee_len = cpu_to_le16(ar.len);
4451 /* Mark unwritten */
4452 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4453 ext4_ext_mark_unwritten(&newex);
4454 map->m_flags |= EXT4_MAP_UNWRITTEN;
4458 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4459 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4462 err = ext4_ext_insert_extent(handle, inode, &path,
4465 if (err && free_on_err) {
4466 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4467 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4468 /* free data blocks we just allocated */
4469 /* not a good idea to call discard here directly,
4470 * but otherwise we'd need to call it every free() */
4471 ext4_discard_preallocations(inode);
4472 ext4_free_blocks(handle, inode, NULL, newblock,
4473 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4477 /* previous routine could use block we allocated */
4478 newblock = ext4_ext_pblock(&newex);
4479 allocated = ext4_ext_get_actual_len(&newex);
4480 if (allocated > map->m_len)
4481 allocated = map->m_len;
4482 map->m_flags |= EXT4_MAP_NEW;
4485 * Reduce the reserved cluster count to reflect successful deferred
4486 * allocation of delayed allocated clusters or direct allocation of
4487 * clusters discovered to be delayed allocated. Once allocated, a
4488 * cluster is not included in the reserved count.
4490 if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4491 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4493 * When allocating delayed allocated clusters, simply
4494 * reduce the reserved cluster count and claim quota
4496 ext4_da_update_reserve_space(inode, allocated_clusters,
4499 ext4_lblk_t lblk, len;
4503 * When allocating non-delayed allocated clusters
4504 * (from fallocate, filemap, DIO, or clusters
4505 * allocated when delalloc has been disabled by
4506 * ext4_nonda_switch), reduce the reserved cluster
4507 * count by the number of allocated clusters that
4508 * have previously been delayed allocated. Quota
4509 * has been claimed by ext4_mb_new_blocks() above,
4510 * so release the quota reservations made for any
4511 * previously delayed allocated clusters.
4513 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4514 len = allocated_clusters << sbi->s_cluster_bits;
4515 n = ext4_es_delayed_clu(inode, lblk, len);
4517 ext4_da_update_reserve_space(inode, (int) n, 0);
4522 * Cache the extent and update transaction to commit on fdatasync only
4523 * when it is _not_ an unwritten extent.
4525 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4526 ext4_update_inode_fsync_trans(handle, inode, 1);
4528 ext4_update_inode_fsync_trans(handle, inode, 0);
4530 if (allocated > map->m_len)
4531 allocated = map->m_len;
4532 ext4_ext_show_leaf(inode, path);
4533 map->m_flags |= EXT4_MAP_MAPPED;
4534 map->m_pblk = newblock;
4535 map->m_len = allocated;
4537 ext4_ext_drop_refs(path);
4540 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4541 err ? err : allocated);
4542 return err ? err : allocated;
4545 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4547 struct super_block *sb = inode->i_sb;
4548 ext4_lblk_t last_block;
4552 * TODO: optimization is possible here.
4553 * Probably we need not scan at all,
4554 * because page truncation is enough.
4557 /* we have to know where to truncate from in crash case */
4558 EXT4_I(inode)->i_disksize = inode->i_size;
4559 err = ext4_mark_inode_dirty(handle, inode);
4563 last_block = (inode->i_size + sb->s_blocksize - 1)
4564 >> EXT4_BLOCK_SIZE_BITS(sb);
4566 err = ext4_es_remove_extent(inode, last_block,
4567 EXT_MAX_BLOCKS - last_block);
4568 if (err == -ENOMEM) {
4570 congestion_wait(BLK_RW_ASYNC, HZ/50);
4575 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4578 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4579 ext4_lblk_t len, loff_t new_size,
4582 struct inode *inode = file_inode(file);
4588 struct ext4_map_blocks map;
4589 unsigned int credits;
4592 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4593 map.m_lblk = offset;
4596 * Don't normalize the request if it can fit in one extent so
4597 * that it doesn't get unnecessarily split into multiple
4600 if (len <= EXT_UNWRITTEN_MAX_LEN)
4601 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4604 * credits to insert 1 extent into extent tree
4606 credits = ext4_chunk_trans_blocks(inode, len);
4607 depth = ext_depth(inode);
4610 while (ret >= 0 && len) {
4612 * Recalculate credits when extent tree depth changes.
4614 if (depth != ext_depth(inode)) {
4615 credits = ext4_chunk_trans_blocks(inode, len);
4616 depth = ext_depth(inode);
4619 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4621 if (IS_ERR(handle)) {
4622 ret = PTR_ERR(handle);
4625 ret = ext4_map_blocks(handle, inode, &map, flags);
4627 ext4_debug("inode #%lu: block %u: len %u: "
4628 "ext4_ext_map_blocks returned %d",
4629 inode->i_ino, map.m_lblk,
4631 ext4_mark_inode_dirty(handle, inode);
4632 ret2 = ext4_journal_stop(handle);
4636 map.m_len = len = len - ret;
4637 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4638 inode->i_ctime = current_time(inode);
4640 if (epos > new_size)
4642 if (ext4_update_inode_size(inode, epos) & 0x1)
4643 inode->i_mtime = inode->i_ctime;
4645 if (epos > inode->i_size)
4646 ext4_set_inode_flag(inode,
4647 EXT4_INODE_EOFBLOCKS);
4649 ext4_mark_inode_dirty(handle, inode);
4650 ext4_update_inode_fsync_trans(handle, inode, 1);
4651 ret2 = ext4_journal_stop(handle);
4655 if (ret == -ENOSPC &&
4656 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4661 return ret > 0 ? ret2 : ret;
4664 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4666 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4668 static long ext4_zero_range(struct file *file, loff_t offset,
4669 loff_t len, int mode)
4671 struct inode *inode = file_inode(file);
4672 handle_t *handle = NULL;
4673 unsigned int max_blocks;
4674 loff_t new_size = 0;
4678 int partial_begin, partial_end;
4681 unsigned int blkbits = inode->i_blkbits;
4683 trace_ext4_zero_range(inode, offset, len, mode);
4685 /* Call ext4_force_commit to flush all data in case of data=journal. */
4686 if (ext4_should_journal_data(inode)) {
4687 ret = ext4_force_commit(inode->i_sb);
4693 * Round up offset. This is not fallocate, we neet to zero out
4694 * blocks, so convert interior block aligned part of the range to
4695 * unwritten and possibly manually zero out unaligned parts of the
4698 start = round_up(offset, 1 << blkbits);
4699 end = round_down((offset + len), 1 << blkbits);
4701 if (start < offset || end > offset + len)
4703 partial_begin = offset & ((1 << blkbits) - 1);
4704 partial_end = (offset + len) & ((1 << blkbits) - 1);
4706 lblk = start >> blkbits;
4707 max_blocks = (end >> blkbits);
4708 if (max_blocks < lblk)
4716 * Indirect files do not support unwritten extnets
4718 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4723 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4724 (offset + len > inode->i_size ||
4725 offset + len > EXT4_I(inode)->i_disksize)) {
4726 new_size = offset + len;
4727 ret = inode_newsize_ok(inode, new_size);
4732 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4733 if (mode & FALLOC_FL_KEEP_SIZE)
4734 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4736 /* Wait all existing dio workers, newcomers will block on i_mutex */
4737 inode_dio_wait(inode);
4739 /* Preallocate the range including the unaligned edges */
4740 if (partial_begin || partial_end) {
4741 ret = ext4_alloc_file_blocks(file,
4742 round_down(offset, 1 << blkbits) >> blkbits,
4743 (round_up((offset + len), 1 << blkbits) -
4744 round_down(offset, 1 << blkbits)) >> blkbits,
4751 /* Zero range excluding the unaligned edges */
4752 if (max_blocks > 0) {
4753 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4757 * Prevent page faults from reinstantiating pages we have
4758 * released from page cache.
4760 down_write(&EXT4_I(inode)->i_mmap_sem);
4762 ret = ext4_break_layouts(inode);
4764 up_write(&EXT4_I(inode)->i_mmap_sem);
4768 ret = ext4_update_disksize_before_punch(inode, offset, len);
4770 up_write(&EXT4_I(inode)->i_mmap_sem);
4773 /* Now release the pages and zero block aligned part of pages */
4774 truncate_pagecache_range(inode, start, end - 1);
4775 inode->i_mtime = inode->i_ctime = current_time(inode);
4777 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4779 up_write(&EXT4_I(inode)->i_mmap_sem);
4783 if (!partial_begin && !partial_end)
4787 * In worst case we have to writeout two nonadjacent unwritten
4788 * blocks and update the inode
4790 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4791 if (ext4_should_journal_data(inode))
4793 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4794 if (IS_ERR(handle)) {
4795 ret = PTR_ERR(handle);
4796 ext4_std_error(inode->i_sb, ret);
4800 inode->i_mtime = inode->i_ctime = current_time(inode);
4802 ext4_update_inode_size(inode, new_size);
4805 * Mark that we allocate beyond EOF so the subsequent truncate
4806 * can proceed even if the new size is the same as i_size.
4808 if (offset + len > inode->i_size)
4809 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4811 ext4_mark_inode_dirty(handle, inode);
4813 /* Zero out partial block at the edges of the range */
4814 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4816 ext4_update_inode_fsync_trans(handle, inode, 1);
4818 if (file->f_flags & O_SYNC)
4819 ext4_handle_sync(handle);
4821 ext4_journal_stop(handle);
4823 inode_unlock(inode);
4828 * preallocate space for a file. This implements ext4's fallocate file
4829 * operation, which gets called from sys_fallocate system call.
4830 * For block-mapped files, posix_fallocate should fall back to the method
4831 * of writing zeroes to the required new blocks (the same behavior which is
4832 * expected for file systems which do not support fallocate() system call).
4834 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4836 struct inode *inode = file_inode(file);
4837 loff_t new_size = 0;
4838 unsigned int max_blocks;
4842 unsigned int blkbits = inode->i_blkbits;
4845 * Encrypted inodes can't handle collapse range or insert
4846 * range since we would need to re-encrypt blocks with a
4847 * different IV or XTS tweak (which are based on the logical
4850 if (IS_ENCRYPTED(inode) &&
4851 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4854 /* Return error if mode is not supported */
4855 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4856 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4857 FALLOC_FL_INSERT_RANGE))
4860 if (mode & FALLOC_FL_PUNCH_HOLE)
4861 return ext4_punch_hole(inode, offset, len);
4863 ret = ext4_convert_inline_data(inode);
4867 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4868 return ext4_collapse_range(inode, offset, len);
4870 if (mode & FALLOC_FL_INSERT_RANGE)
4871 return ext4_insert_range(inode, offset, len);
4873 if (mode & FALLOC_FL_ZERO_RANGE)
4874 return ext4_zero_range(file, offset, len, mode);
4876 trace_ext4_fallocate_enter(inode, offset, len, mode);
4877 lblk = offset >> blkbits;
4879 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4880 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4881 if (mode & FALLOC_FL_KEEP_SIZE)
4882 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4887 * We only support preallocation for extent-based files only
4889 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4894 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4895 (offset + len > inode->i_size ||
4896 offset + len > EXT4_I(inode)->i_disksize)) {
4897 new_size = offset + len;
4898 ret = inode_newsize_ok(inode, new_size);
4903 /* Wait all existing dio workers, newcomers will block on i_mutex */
4904 inode_dio_wait(inode);
4906 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4910 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4911 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4912 EXT4_I(inode)->i_sync_tid);
4915 inode_unlock(inode);
4916 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4921 * This function convert a range of blocks to written extents
4922 * The caller of this function will pass the start offset and the size.
4923 * all unwritten extents within this range will be converted to
4926 * This function is called from the direct IO end io call back
4927 * function, to convert the fallocated extents after IO is completed.
4928 * Returns 0 on success.
4930 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4931 loff_t offset, ssize_t len)
4933 unsigned int max_blocks;
4936 struct ext4_map_blocks map;
4937 unsigned int blkbits = inode->i_blkbits;
4938 unsigned int credits = 0;
4940 map.m_lblk = offset >> blkbits;
4941 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4945 * credits to insert 1 extent into extent tree
4947 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4949 while (ret >= 0 && ret < max_blocks) {
4951 map.m_len = (max_blocks -= ret);
4953 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4955 if (IS_ERR(handle)) {
4956 ret = PTR_ERR(handle);
4960 ret = ext4_map_blocks(handle, inode, &map,
4961 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4963 ext4_warning(inode->i_sb,
4964 "inode #%lu: block %u: len %u: "
4965 "ext4_ext_map_blocks returned %d",
4966 inode->i_ino, map.m_lblk,
4968 ext4_mark_inode_dirty(handle, inode);
4970 ret2 = ext4_journal_stop(handle);
4971 if (ret <= 0 || ret2)
4974 return ret > 0 ? ret2 : ret;
4977 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4980 struct ext4_io_end_vec *io_end_vec;
4983 * This is somewhat ugly but the idea is clear: When transaction is
4984 * reserved, everything goes into it. Otherwise we rather start several
4985 * smaller transactions for conversion of each extent separately.
4988 handle = ext4_journal_start_reserved(handle,
4989 EXT4_HT_EXT_CONVERT);
4991 return PTR_ERR(handle);
4994 list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4995 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
5003 err = ext4_journal_stop(handle);
5005 return ret < 0 ? ret : err;
5009 * If newes is not existing extent (newes->ec_pblk equals zero) find
5010 * delayed extent at start of newes and update newes accordingly and
5011 * return start of the next delayed extent.
5013 * If newes is existing extent (newes->ec_pblk is not equal zero)
5014 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5015 * extent found. Leave newes unmodified.
5017 static int ext4_find_delayed_extent(struct inode *inode,
5018 struct extent_status *newes)
5020 struct extent_status es;
5021 ext4_lblk_t block, next_del;
5023 if (newes->es_pblk == 0) {
5024 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5026 newes->es_lblk + newes->es_len - 1,
5030 * No extent in extent-tree contains block @newes->es_pblk,
5031 * then the block may stay in 1)a hole or 2)delayed-extent.
5037 if (es.es_lblk > newes->es_lblk) {
5039 newes->es_len = min(es.es_lblk - newes->es_lblk,
5044 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5047 block = newes->es_lblk + newes->es_len;
5048 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5049 EXT_MAX_BLOCKS, &es);
5051 next_del = EXT_MAX_BLOCKS;
5053 next_del = es.es_lblk;
5058 static int ext4_xattr_fiemap(struct inode *inode,
5059 struct fiemap_extent_info *fieinfo)
5063 __u32 flags = FIEMAP_EXTENT_LAST;
5064 int blockbits = inode->i_sb->s_blocksize_bits;
5068 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5069 struct ext4_iloc iloc;
5070 int offset; /* offset of xattr in inode */
5072 error = ext4_get_inode_loc(inode, &iloc);
5075 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5076 offset = EXT4_GOOD_OLD_INODE_SIZE +
5077 EXT4_I(inode)->i_extra_isize;
5079 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5080 flags |= FIEMAP_EXTENT_DATA_INLINE;
5082 } else { /* external block */
5083 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5084 length = inode->i_sb->s_blocksize;
5088 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5090 return (error < 0 ? error : 0);
5093 static int _ext4_fiemap(struct inode *inode,
5094 struct fiemap_extent_info *fieinfo,
5095 __u64 start, __u64 len,
5096 int (*fill)(struct inode *, ext4_lblk_t,
5098 struct fiemap_extent_info *))
5100 ext4_lblk_t start_blk;
5101 u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5105 if (ext4_has_inline_data(inode)) {
5108 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5115 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5116 error = ext4_ext_precache(inode);
5119 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5122 /* fallback to generic here if not in extents fmt */
5123 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5124 fill == ext4_fill_fiemap_extents)
5125 return generic_block_fiemap(inode, fieinfo, start, len,
5128 if (fill == ext4_fill_es_cache_info)
5129 ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5130 if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5133 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5134 error = ext4_xattr_fiemap(inode, fieinfo);
5136 ext4_lblk_t len_blks;
5139 start_blk = start >> inode->i_sb->s_blocksize_bits;
5140 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5141 if (last_blk >= EXT_MAX_BLOCKS)
5142 last_blk = EXT_MAX_BLOCKS-1;
5143 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5146 * Walk the extent tree gathering extent information
5147 * and pushing extents back to the user.
5149 error = fill(inode, start_blk, len_blks, fieinfo);
5154 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5155 __u64 start, __u64 len)
5157 return _ext4_fiemap(inode, fieinfo, start, len,
5158 ext4_fill_fiemap_extents);
5161 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5162 __u64 start, __u64 len)
5164 if (ext4_has_inline_data(inode)) {
5167 down_read(&EXT4_I(inode)->xattr_sem);
5168 has_inline = ext4_has_inline_data(inode);
5169 up_read(&EXT4_I(inode)->xattr_sem);
5174 return _ext4_fiemap(inode, fieinfo, start, len,
5175 ext4_fill_es_cache_info);
5181 * Function to access the path buffer for marking it dirty.
5182 * It also checks if there are sufficient credits left in the journal handle
5186 ext4_access_path(handle_t *handle, struct inode *inode,
5187 struct ext4_ext_path *path)
5191 if (!ext4_handle_valid(handle))
5195 * Check if need to extend journal credits
5196 * 3 for leaf, sb, and inode plus 2 (bmap and group
5197 * descriptor) for each block group; assume two block
5200 credits = ext4_writepage_trans_blocks(inode);
5201 err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
5205 err = ext4_ext_get_access(handle, inode, path);
5210 * ext4_ext_shift_path_extents:
5211 * Shift the extents of a path structure lying between path[depth].p_ext
5212 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5213 * if it is right shift or left shift operation.
5216 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5217 struct inode *inode, handle_t *handle,
5218 enum SHIFT_DIRECTION SHIFT)
5221 struct ext4_extent *ex_start, *ex_last;
5222 bool update = false;
5223 depth = path->p_depth;
5225 while (depth >= 0) {
5226 if (depth == path->p_depth) {
5227 ex_start = path[depth].p_ext;
5229 return -EFSCORRUPTED;
5231 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5233 err = ext4_access_path(handle, inode, path + depth);
5237 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5240 while (ex_start <= ex_last) {
5241 if (SHIFT == SHIFT_LEFT) {
5242 le32_add_cpu(&ex_start->ee_block,
5244 /* Try to merge to the left. */
5246 EXT_FIRST_EXTENT(path[depth].p_hdr))
5248 ext4_ext_try_to_merge_right(inode,
5249 path, ex_start - 1))
5254 le32_add_cpu(&ex_last->ee_block, shift);
5255 ext4_ext_try_to_merge_right(inode, path,
5260 err = ext4_ext_dirty(handle, inode, path + depth);
5264 if (--depth < 0 || !update)
5268 /* Update index too */
5269 err = ext4_access_path(handle, inode, path + depth);
5273 if (SHIFT == SHIFT_LEFT)
5274 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5276 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5277 err = ext4_ext_dirty(handle, inode, path + depth);
5281 /* we are done if current index is not a starting index */
5282 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5293 * ext4_ext_shift_extents:
5294 * All the extents which lies in the range from @start to the last allocated
5295 * block for the @inode are shifted either towards left or right (depending
5296 * upon @SHIFT) by @shift blocks.
5297 * On success, 0 is returned, error otherwise.
5300 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5301 ext4_lblk_t start, ext4_lblk_t shift,
5302 enum SHIFT_DIRECTION SHIFT)
5304 struct ext4_ext_path *path;
5306 struct ext4_extent *extent;
5307 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5309 /* Let path point to the last extent */
5310 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5313 return PTR_ERR(path);
5315 depth = path->p_depth;
5316 extent = path[depth].p_ext;
5320 stop = le32_to_cpu(extent->ee_block);
5323 * For left shifts, make sure the hole on the left is big enough to
5324 * accommodate the shift. For right shifts, make sure the last extent
5325 * won't be shifted beyond EXT_MAX_BLOCKS.
5327 if (SHIFT == SHIFT_LEFT) {
5328 path = ext4_find_extent(inode, start - 1, &path,
5331 return PTR_ERR(path);
5332 depth = path->p_depth;
5333 extent = path[depth].p_ext;
5335 ex_start = le32_to_cpu(extent->ee_block);
5336 ex_end = le32_to_cpu(extent->ee_block) +
5337 ext4_ext_get_actual_len(extent);
5343 if ((start == ex_start && shift > ex_start) ||
5344 (shift > start - ex_end)) {
5349 if (shift > EXT_MAX_BLOCKS -
5350 (stop + ext4_ext_get_actual_len(extent))) {
5357 * In case of left shift, iterator points to start and it is increased
5358 * till we reach stop. In case of right shift, iterator points to stop
5359 * and it is decreased till we reach start.
5361 if (SHIFT == SHIFT_LEFT)
5367 * Its safe to start updating extents. Start and stop are unsigned, so
5368 * in case of right shift if extent with 0 block is reached, iterator
5369 * becomes NULL to indicate the end of the loop.
5371 while (iterator && start <= stop) {
5372 path = ext4_find_extent(inode, *iterator, &path,
5375 return PTR_ERR(path);
5376 depth = path->p_depth;
5377 extent = path[depth].p_ext;
5379 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5380 (unsigned long) *iterator);
5381 return -EFSCORRUPTED;
5383 if (SHIFT == SHIFT_LEFT && *iterator >
5384 le32_to_cpu(extent->ee_block)) {
5385 /* Hole, move to the next extent */
5386 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5387 path[depth].p_ext++;
5389 *iterator = ext4_ext_next_allocated_block(path);
5394 if (SHIFT == SHIFT_LEFT) {
5395 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5396 *iterator = le32_to_cpu(extent->ee_block) +
5397 ext4_ext_get_actual_len(extent);
5399 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5400 if (le32_to_cpu(extent->ee_block) > 0)
5401 *iterator = le32_to_cpu(extent->ee_block) - 1;
5403 /* Beginning is reached, end of the loop */
5405 /* Update path extent in case we need to stop */
5406 while (le32_to_cpu(extent->ee_block) < start)
5408 path[depth].p_ext = extent;
5410 ret = ext4_ext_shift_path_extents(path, shift, inode,
5416 ext4_ext_drop_refs(path);
5422 * ext4_collapse_range:
5423 * This implements the fallocate's collapse range functionality for ext4
5424 * Returns: 0 and non-zero on error.
5426 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5428 struct super_block *sb = inode->i_sb;
5429 ext4_lblk_t punch_start, punch_stop;
5431 unsigned int credits;
5432 loff_t new_size, ioffset;
5436 * We need to test this early because xfstests assumes that a
5437 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5438 * system does not support collapse range.
5440 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5443 /* Collapse range works only on fs cluster size aligned regions. */
5444 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5447 trace_ext4_collapse_range(inode, offset, len);
5449 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5450 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5452 /* Call ext4_force_commit to flush all data in case of data=journal. */
5453 if (ext4_should_journal_data(inode)) {
5454 ret = ext4_force_commit(inode->i_sb);
5461 * There is no need to overlap collapse range with EOF, in which case
5462 * it is effectively a truncate operation
5464 if (offset + len >= inode->i_size) {
5469 /* Currently just for extent based files */
5470 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5475 /* Wait for existing dio to complete */
5476 inode_dio_wait(inode);
5479 * Prevent page faults from reinstantiating pages we have released from
5482 down_write(&EXT4_I(inode)->i_mmap_sem);
5484 ret = ext4_break_layouts(inode);
5489 * Need to round down offset to be aligned with page size boundary
5490 * for page size > block size.
5492 ioffset = round_down(offset, PAGE_SIZE);
5494 * Write tail of the last page before removed range since it will get
5495 * removed from the page cache below.
5497 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5501 * Write data that will be shifted to preserve them when discarding
5502 * page cache below. We are also protected from pages becoming dirty
5505 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5509 truncate_pagecache(inode, ioffset);
5511 credits = ext4_writepage_trans_blocks(inode);
5512 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5513 if (IS_ERR(handle)) {
5514 ret = PTR_ERR(handle);
5518 down_write(&EXT4_I(inode)->i_data_sem);
5519 ext4_discard_preallocations(inode);
5521 ret = ext4_es_remove_extent(inode, punch_start,
5522 EXT_MAX_BLOCKS - punch_start);
5524 up_write(&EXT4_I(inode)->i_data_sem);
5528 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5530 up_write(&EXT4_I(inode)->i_data_sem);
5533 ext4_discard_preallocations(inode);
5535 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5536 punch_stop - punch_start, SHIFT_LEFT);
5538 up_write(&EXT4_I(inode)->i_data_sem);
5542 new_size = inode->i_size - len;
5543 i_size_write(inode, new_size);
5544 EXT4_I(inode)->i_disksize = new_size;
5546 up_write(&EXT4_I(inode)->i_data_sem);
5548 ext4_handle_sync(handle);
5549 inode->i_mtime = inode->i_ctime = current_time(inode);
5550 ext4_mark_inode_dirty(handle, inode);
5551 ext4_update_inode_fsync_trans(handle, inode, 1);
5554 ext4_journal_stop(handle);
5556 up_write(&EXT4_I(inode)->i_mmap_sem);
5558 inode_unlock(inode);
5563 * ext4_insert_range:
5564 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5565 * The data blocks starting from @offset to the EOF are shifted by @len
5566 * towards right to create a hole in the @inode. Inode size is increased
5568 * Returns 0 on success, error otherwise.
5570 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5572 struct super_block *sb = inode->i_sb;
5574 struct ext4_ext_path *path;
5575 struct ext4_extent *extent;
5576 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5577 unsigned int credits, ee_len;
5578 int ret = 0, depth, split_flag = 0;
5582 * We need to test this early because xfstests assumes that an
5583 * insert range of (0, 1) will return EOPNOTSUPP if the file
5584 * system does not support insert range.
5586 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5589 /* Insert range works only on fs cluster size aligned regions. */
5590 if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5593 trace_ext4_insert_range(inode, offset, len);
5595 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5596 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5598 /* Call ext4_force_commit to flush all data in case of data=journal */
5599 if (ext4_should_journal_data(inode)) {
5600 ret = ext4_force_commit(inode->i_sb);
5606 /* Currently just for extent based files */
5607 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5612 /* Check whether the maximum file size would be exceeded */
5613 if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5618 /* Offset must be less than i_size */
5619 if (offset >= inode->i_size) {
5624 /* Wait for existing dio to complete */
5625 inode_dio_wait(inode);
5628 * Prevent page faults from reinstantiating pages we have released from
5631 down_write(&EXT4_I(inode)->i_mmap_sem);
5633 ret = ext4_break_layouts(inode);
5638 * Need to round down to align start offset to page size boundary
5639 * for page size > block size.
5641 ioffset = round_down(offset, PAGE_SIZE);
5642 /* Write out all dirty pages */
5643 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5647 truncate_pagecache(inode, ioffset);
5649 credits = ext4_writepage_trans_blocks(inode);
5650 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5651 if (IS_ERR(handle)) {
5652 ret = PTR_ERR(handle);
5656 /* Expand file to avoid data loss if there is error while shifting */
5657 inode->i_size += len;
5658 EXT4_I(inode)->i_disksize += len;
5659 inode->i_mtime = inode->i_ctime = current_time(inode);
5660 ret = ext4_mark_inode_dirty(handle, inode);
5664 down_write(&EXT4_I(inode)->i_data_sem);
5665 ext4_discard_preallocations(inode);
5667 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5669 up_write(&EXT4_I(inode)->i_data_sem);
5673 depth = ext_depth(inode);
5674 extent = path[depth].p_ext;
5676 ee_start_lblk = le32_to_cpu(extent->ee_block);
5677 ee_len = ext4_ext_get_actual_len(extent);
5680 * If offset_lblk is not the starting block of extent, split
5681 * the extent @offset_lblk
5683 if ((offset_lblk > ee_start_lblk) &&
5684 (offset_lblk < (ee_start_lblk + ee_len))) {
5685 if (ext4_ext_is_unwritten(extent))
5686 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5687 EXT4_EXT_MARK_UNWRIT2;
5688 ret = ext4_split_extent_at(handle, inode, &path,
5689 offset_lblk, split_flag,
5691 EXT4_GET_BLOCKS_PRE_IO |
5692 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5695 ext4_ext_drop_refs(path);
5698 up_write(&EXT4_I(inode)->i_data_sem);
5702 ext4_ext_drop_refs(path);
5706 ret = ext4_es_remove_extent(inode, offset_lblk,
5707 EXT_MAX_BLOCKS - offset_lblk);
5709 up_write(&EXT4_I(inode)->i_data_sem);
5714 * if offset_lblk lies in a hole which is at start of file, use
5715 * ee_start_lblk to shift extents
5717 ret = ext4_ext_shift_extents(inode, handle,
5718 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5719 len_lblk, SHIFT_RIGHT);
5721 up_write(&EXT4_I(inode)->i_data_sem);
5723 ext4_handle_sync(handle);
5725 ext4_update_inode_fsync_trans(handle, inode, 1);
5728 ext4_journal_stop(handle);
5730 up_write(&EXT4_I(inode)->i_mmap_sem);
5732 inode_unlock(inode);
5737 * ext4_swap_extents() - Swap extents between two inodes
5738 * @handle: handle for this transaction
5739 * @inode1: First inode
5740 * @inode2: Second inode
5741 * @lblk1: Start block for first inode
5742 * @lblk2: Start block for second inode
5743 * @count: Number of blocks to swap
5744 * @unwritten: Mark second inode's extents as unwritten after swap
5745 * @erp: Pointer to save error value
5747 * This helper routine does exactly what is promise "swap extents". All other
5748 * stuff such as page-cache locking consistency, bh mapping consistency or
5749 * extent's data copying must be performed by caller.
5751 * i_mutex is held for both inodes
5752 * i_data_sem is locked for write for both inodes
5754 * All pages from requested range are locked for both inodes
5757 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5758 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5759 ext4_lblk_t count, int unwritten, int *erp)
5761 struct ext4_ext_path *path1 = NULL;
5762 struct ext4_ext_path *path2 = NULL;
5763 int replaced_count = 0;
5765 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5766 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5767 BUG_ON(!inode_is_locked(inode1));
5768 BUG_ON(!inode_is_locked(inode2));
5770 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5773 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5778 struct ext4_extent *ex1, *ex2, tmp_ex;
5779 ext4_lblk_t e1_blk, e2_blk;
5780 int e1_len, e2_len, len;
5783 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5784 if (IS_ERR(path1)) {
5785 *erp = PTR_ERR(path1);
5791 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5792 if (IS_ERR(path2)) {
5793 *erp = PTR_ERR(path2);
5797 ex1 = path1[path1->p_depth].p_ext;
5798 ex2 = path2[path2->p_depth].p_ext;
5799 /* Do we have somthing to swap ? */
5800 if (unlikely(!ex2 || !ex1))
5803 e1_blk = le32_to_cpu(ex1->ee_block);
5804 e2_blk = le32_to_cpu(ex2->ee_block);
5805 e1_len = ext4_ext_get_actual_len(ex1);
5806 e2_len = ext4_ext_get_actual_len(ex2);
5809 if (!in_range(lblk1, e1_blk, e1_len) ||
5810 !in_range(lblk2, e2_blk, e2_len)) {
5811 ext4_lblk_t next1, next2;
5813 /* if hole after extent, then go to next extent */
5814 next1 = ext4_ext_next_allocated_block(path1);
5815 next2 = ext4_ext_next_allocated_block(path2);
5816 /* If hole before extent, then shift to that extent */
5821 /* Do we have something to swap */
5822 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5824 /* Move to the rightest boundary */
5825 len = next1 - lblk1;
5826 if (len < next2 - lblk2)
5827 len = next2 - lblk2;
5836 /* Prepare left boundary */
5837 if (e1_blk < lblk1) {
5839 *erp = ext4_force_split_extent_at(handle, inode1,
5844 if (e2_blk < lblk2) {
5846 *erp = ext4_force_split_extent_at(handle, inode2,
5851 /* ext4_split_extent_at() may result in leaf extent split,
5852 * path must to be revalidated. */
5856 /* Prepare right boundary */
5858 if (len > e1_blk + e1_len - lblk1)
5859 len = e1_blk + e1_len - lblk1;
5860 if (len > e2_blk + e2_len - lblk2)
5861 len = e2_blk + e2_len - lblk2;
5863 if (len != e1_len) {
5865 *erp = ext4_force_split_extent_at(handle, inode1,
5866 &path1, lblk1 + len, 0);
5870 if (len != e2_len) {
5872 *erp = ext4_force_split_extent_at(handle, inode2,
5873 &path2, lblk2 + len, 0);
5877 /* ext4_split_extent_at() may result in leaf extent split,
5878 * path must to be revalidated. */
5882 BUG_ON(e2_len != e1_len);
5883 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5886 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5890 /* Both extents are fully inside boundaries. Swap it now */
5892 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5893 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5894 ex1->ee_len = cpu_to_le16(e2_len);
5895 ex2->ee_len = cpu_to_le16(e1_len);
5897 ext4_ext_mark_unwritten(ex2);
5898 if (ext4_ext_is_unwritten(&tmp_ex))
5899 ext4_ext_mark_unwritten(ex1);
5901 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5902 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5903 *erp = ext4_ext_dirty(handle, inode2, path2 +
5907 *erp = ext4_ext_dirty(handle, inode1, path1 +
5910 * Looks scarry ah..? second inode already points to new blocks,
5911 * and it was successfully dirtied. But luckily error may happen
5912 * only due to journal error, so full transaction will be
5919 replaced_count += len;
5923 ext4_ext_drop_refs(path1);
5925 ext4_ext_drop_refs(path2);
5927 path1 = path2 = NULL;
5929 return replaced_count;
5933 * ext4_clu_mapped - determine whether any block in a logical cluster has
5934 * been mapped to a physical cluster
5936 * @inode - file containing the logical cluster
5937 * @lclu - logical cluster of interest
5939 * Returns 1 if any block in the logical cluster is mapped, signifying
5940 * that a physical cluster has been allocated for it. Otherwise,
5941 * returns 0. Can also return negative error codes. Derived from
5942 * ext4_ext_map_blocks().
5944 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5946 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5947 struct ext4_ext_path *path;
5948 int depth, mapped = 0, err = 0;
5949 struct ext4_extent *extent;
5950 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5952 /* search for the extent closest to the first block in the cluster */
5953 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5955 err = PTR_ERR(path);
5960 depth = ext_depth(inode);
5963 * A consistent leaf must not be empty. This situation is possible,
5964 * though, _during_ tree modification, and it's why an assert can't
5965 * be put in ext4_find_extent().
5967 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5968 EXT4_ERROR_INODE(inode,
5969 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5970 (unsigned long) EXT4_C2B(sbi, lclu),
5971 depth, path[depth].p_block);
5972 err = -EFSCORRUPTED;
5976 extent = path[depth].p_ext;
5978 /* can't be mapped if the extent tree is empty */
5982 first_lblk = le32_to_cpu(extent->ee_block);
5983 first_lclu = EXT4_B2C(sbi, first_lblk);
5986 * Three possible outcomes at this point - found extent spanning
5987 * the target cluster, to the left of the target cluster, or to the
5988 * right of the target cluster. The first two cases are handled here.
5989 * The last case indicates the target cluster is not mapped.
5991 if (lclu >= first_lclu) {
5992 last_lclu = EXT4_B2C(sbi, first_lblk +
5993 ext4_ext_get_actual_len(extent) - 1);
5994 if (lclu <= last_lclu) {
5997 first_lblk = ext4_ext_next_allocated_block(path);
5998 first_lclu = EXT4_B2C(sbi, first_lblk);
5999 if (lclu == first_lclu)
6005 ext4_ext_drop_refs(path);
6008 return err ? err : mapped;