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_truncate_extend_restart(handle_t *handle,
109 if (!ext4_handle_valid(handle))
111 if (handle->h_buffer_credits >= needed)
114 * If we need to extend the journal get a few extra blocks
115 * while we're at it for efficiency's sake.
118 err = ext4_journal_extend(handle, needed - handle->h_buffer_credits);
121 err = ext4_truncate_restart_trans(handle, inode, needed);
133 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
134 struct ext4_ext_path *path)
137 /* path points to block */
138 BUFFER_TRACE(path->p_bh, "get_write_access");
139 return ext4_journal_get_write_access(handle, path->p_bh);
141 /* path points to leaf/index in inode body */
142 /* we use in-core data, no need to protect them */
152 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
153 struct inode *inode, struct ext4_ext_path *path)
157 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
159 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
160 /* path points to block */
161 err = __ext4_handle_dirty_metadata(where, line, handle,
164 /* path points to leaf/index in inode body */
165 err = ext4_mark_inode_dirty(handle, inode);
170 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
171 struct ext4_ext_path *path,
175 int depth = path->p_depth;
176 struct ext4_extent *ex;
179 * Try to predict block placement assuming that we are
180 * filling in a file which will eventually be
181 * non-sparse --- i.e., in the case of libbfd writing
182 * an ELF object sections out-of-order but in a way
183 * the eventually results in a contiguous object or
184 * executable file, or some database extending a table
185 * space file. However, this is actually somewhat
186 * non-ideal if we are writing a sparse file such as
187 * qemu or KVM writing a raw image file that is going
188 * to stay fairly sparse, since it will end up
189 * fragmenting the file system's free space. Maybe we
190 * should have some hueristics or some way to allow
191 * userspace to pass a hint to file system,
192 * especially if the latter case turns out to be
195 ex = path[depth].p_ext;
197 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
198 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
200 if (block > ext_block)
201 return ext_pblk + (block - ext_block);
203 return ext_pblk - (ext_block - block);
206 /* it looks like index is empty;
207 * try to find starting block from index itself */
208 if (path[depth].p_bh)
209 return path[depth].p_bh->b_blocknr;
212 /* OK. use inode's group */
213 return ext4_inode_to_goal_block(inode);
217 * Allocation for a meta data block
220 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
221 struct ext4_ext_path *path,
222 struct ext4_extent *ex, int *err, unsigned int flags)
224 ext4_fsblk_t goal, newblock;
226 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
227 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
232 static inline int ext4_ext_space_block(struct inode *inode, int check)
236 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
237 / sizeof(struct ext4_extent);
238 #ifdef AGGRESSIVE_TEST
239 if (!check && size > 6)
245 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
249 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
250 / sizeof(struct ext4_extent_idx);
251 #ifdef AGGRESSIVE_TEST
252 if (!check && size > 5)
258 static inline int ext4_ext_space_root(struct inode *inode, int check)
262 size = sizeof(EXT4_I(inode)->i_data);
263 size -= sizeof(struct ext4_extent_header);
264 size /= sizeof(struct ext4_extent);
265 #ifdef AGGRESSIVE_TEST
266 if (!check && size > 3)
272 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
276 size = sizeof(EXT4_I(inode)->i_data);
277 size -= sizeof(struct ext4_extent_header);
278 size /= sizeof(struct ext4_extent_idx);
279 #ifdef AGGRESSIVE_TEST
280 if (!check && size > 4)
287 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
288 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
291 struct ext4_ext_path *path = *ppath;
292 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
294 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
295 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
296 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
297 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
301 * Calculate the number of metadata blocks needed
302 * to allocate @blocks
303 * Worse case is one block per extent
305 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
307 struct ext4_inode_info *ei = EXT4_I(inode);
310 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
311 / sizeof(struct ext4_extent_idx));
314 * If the new delayed allocation block is contiguous with the
315 * previous da block, it can share index blocks with the
316 * previous block, so we only need to allocate a new index
317 * block every idxs leaf blocks. At ldxs**2 blocks, we need
318 * an additional index block, and at ldxs**3 blocks, yet
319 * another index blocks.
321 if (ei->i_da_metadata_calc_len &&
322 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
325 if ((ei->i_da_metadata_calc_len % idxs) == 0)
327 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
329 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
331 ei->i_da_metadata_calc_len = 0;
333 ei->i_da_metadata_calc_len++;
334 ei->i_da_metadata_calc_last_lblock++;
339 * In the worst case we need a new set of index blocks at
340 * every level of the inode's extent tree.
342 ei->i_da_metadata_calc_len = 1;
343 ei->i_da_metadata_calc_last_lblock = lblock;
344 return ext_depth(inode) + 1;
348 ext4_ext_max_entries(struct inode *inode, int depth)
352 if (depth == ext_depth(inode)) {
354 max = ext4_ext_space_root(inode, 1);
356 max = ext4_ext_space_root_idx(inode, 1);
359 max = ext4_ext_space_block(inode, 1);
361 max = ext4_ext_space_block_idx(inode, 1);
367 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
369 ext4_fsblk_t block = ext4_ext_pblock(ext);
370 int len = ext4_ext_get_actual_len(ext);
371 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
376 * - overflow/wrap-around
378 if (lblock + len <= lblock)
380 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
383 static int ext4_valid_extent_idx(struct inode *inode,
384 struct ext4_extent_idx *ext_idx)
386 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
388 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
391 static int ext4_valid_extent_entries(struct inode *inode,
392 struct ext4_extent_header *eh,
395 unsigned short entries;
396 if (eh->eh_entries == 0)
399 entries = le16_to_cpu(eh->eh_entries);
403 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
404 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
405 ext4_fsblk_t pblock = 0;
406 ext4_lblk_t lblock = 0;
407 ext4_lblk_t prev = 0;
410 if (!ext4_valid_extent(inode, ext))
413 /* Check for overlapping extents */
414 lblock = le32_to_cpu(ext->ee_block);
415 len = ext4_ext_get_actual_len(ext);
416 if ((lblock <= prev) && prev) {
417 pblock = ext4_ext_pblock(ext);
418 es->s_last_error_block = cpu_to_le64(pblock);
423 prev = lblock + len - 1;
426 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
428 if (!ext4_valid_extent_idx(inode, ext_idx))
437 static int __ext4_ext_check(const char *function, unsigned int line,
438 struct inode *inode, struct ext4_extent_header *eh,
439 int depth, ext4_fsblk_t pblk)
441 const char *error_msg;
442 int max = 0, err = -EFSCORRUPTED;
444 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
445 error_msg = "invalid magic";
448 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
449 error_msg = "unexpected eh_depth";
452 if (unlikely(eh->eh_max == 0)) {
453 error_msg = "invalid eh_max";
456 max = ext4_ext_max_entries(inode, depth);
457 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
458 error_msg = "too large eh_max";
461 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
462 error_msg = "invalid eh_entries";
465 if (!ext4_valid_extent_entries(inode, eh, depth)) {
466 error_msg = "invalid extent entries";
469 if (unlikely(depth > 32)) {
470 error_msg = "too large eh_depth";
473 /* Verify checksum on non-root extent tree nodes */
474 if (ext_depth(inode) != depth &&
475 !ext4_extent_block_csum_verify(inode, eh)) {
476 error_msg = "extent tree corrupted";
483 ext4_error_inode(inode, function, line, 0,
484 "pblk %llu bad header/extent: %s - magic %x, "
485 "entries %u, max %u(%u), depth %u(%u)",
486 (unsigned long long) pblk, error_msg,
487 le16_to_cpu(eh->eh_magic),
488 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
489 max, le16_to_cpu(eh->eh_depth), depth);
493 #define ext4_ext_check(inode, eh, depth, pblk) \
494 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
496 int ext4_ext_check_inode(struct inode *inode)
498 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
501 static struct buffer_head *
502 __read_extent_tree_block(const char *function, unsigned int line,
503 struct inode *inode, ext4_fsblk_t pblk, int depth,
506 struct buffer_head *bh;
509 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
511 return ERR_PTR(-ENOMEM);
513 if (!bh_uptodate_or_lock(bh)) {
514 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
515 err = bh_submit_read(bh);
519 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
521 if (!ext4_has_feature_journal(inode->i_sb) ||
523 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
524 err = __ext4_ext_check(function, line, inode,
525 ext_block_hdr(bh), depth, pblk);
529 set_buffer_verified(bh);
531 * If this is a leaf block, cache all of its entries
533 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
534 struct ext4_extent_header *eh = ext_block_hdr(bh);
535 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
536 ext4_lblk_t prev = 0;
539 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
540 unsigned int status = EXTENT_STATUS_WRITTEN;
541 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
542 int len = ext4_ext_get_actual_len(ex);
544 if (prev && (prev != lblk))
545 ext4_es_cache_extent(inode, prev,
549 if (ext4_ext_is_unwritten(ex))
550 status = EXTENT_STATUS_UNWRITTEN;
551 ext4_es_cache_extent(inode, lblk, len,
552 ext4_ext_pblock(ex), status);
563 #define read_extent_tree_block(inode, pblk, depth, flags) \
564 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
568 * This function is called to cache a file's extent information in the
571 int ext4_ext_precache(struct inode *inode)
573 struct ext4_inode_info *ei = EXT4_I(inode);
574 struct ext4_ext_path *path = NULL;
575 struct buffer_head *bh;
576 int i = 0, depth, ret = 0;
578 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
579 return 0; /* not an extent-mapped inode */
581 down_read(&ei->i_data_sem);
582 depth = ext_depth(inode);
584 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
587 up_read(&ei->i_data_sem);
591 /* Don't cache anything if there are no external extent blocks */
594 path[0].p_hdr = ext_inode_hdr(inode);
595 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
598 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
601 * If this is a leaf block or we've reached the end of
602 * the index block, go up
605 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
606 brelse(path[i].p_bh);
611 bh = read_extent_tree_block(inode,
612 ext4_idx_pblock(path[i].p_idx++),
614 EXT4_EX_FORCE_CACHE);
621 path[i].p_hdr = ext_block_hdr(bh);
622 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
624 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
626 up_read(&ei->i_data_sem);
627 ext4_ext_drop_refs(path);
633 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
635 int k, l = path->p_depth;
638 for (k = 0; k <= l; k++, path++) {
640 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
641 ext4_idx_pblock(path->p_idx));
642 } else if (path->p_ext) {
643 ext_debug(" %d:[%d]%d:%llu ",
644 le32_to_cpu(path->p_ext->ee_block),
645 ext4_ext_is_unwritten(path->p_ext),
646 ext4_ext_get_actual_len(path->p_ext),
647 ext4_ext_pblock(path->p_ext));
654 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
656 int depth = ext_depth(inode);
657 struct ext4_extent_header *eh;
658 struct ext4_extent *ex;
664 eh = path[depth].p_hdr;
665 ex = EXT_FIRST_EXTENT(eh);
667 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
669 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
670 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
671 ext4_ext_is_unwritten(ex),
672 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
677 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
678 ext4_fsblk_t newblock, int level)
680 int depth = ext_depth(inode);
681 struct ext4_extent *ex;
683 if (depth != level) {
684 struct ext4_extent_idx *idx;
685 idx = path[level].p_idx;
686 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
687 ext_debug("%d: move %d:%llu in new index %llu\n", level,
688 le32_to_cpu(idx->ei_block),
689 ext4_idx_pblock(idx),
697 ex = path[depth].p_ext;
698 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
699 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
700 le32_to_cpu(ex->ee_block),
702 ext4_ext_is_unwritten(ex),
703 ext4_ext_get_actual_len(ex),
710 #define ext4_ext_show_path(inode, path)
711 #define ext4_ext_show_leaf(inode, path)
712 #define ext4_ext_show_move(inode, path, newblock, level)
715 void ext4_ext_drop_refs(struct ext4_ext_path *path)
721 depth = path->p_depth;
722 for (i = 0; i <= depth; i++, path++)
730 * ext4_ext_binsearch_idx:
731 * binary search for the closest index of the given block
732 * the header must be checked before calling this
735 ext4_ext_binsearch_idx(struct inode *inode,
736 struct ext4_ext_path *path, ext4_lblk_t block)
738 struct ext4_extent_header *eh = path->p_hdr;
739 struct ext4_extent_idx *r, *l, *m;
742 ext_debug("binsearch for %u(idx): ", block);
744 l = EXT_FIRST_INDEX(eh) + 1;
745 r = EXT_LAST_INDEX(eh);
748 if (block < le32_to_cpu(m->ei_block))
752 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
753 m, le32_to_cpu(m->ei_block),
754 r, le32_to_cpu(r->ei_block));
758 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
759 ext4_idx_pblock(path->p_idx));
761 #ifdef CHECK_BINSEARCH
763 struct ext4_extent_idx *chix, *ix;
766 chix = ix = EXT_FIRST_INDEX(eh);
767 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
769 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
770 printk(KERN_DEBUG "k=%d, ix=0x%p, "
772 ix, EXT_FIRST_INDEX(eh));
773 printk(KERN_DEBUG "%u <= %u\n",
774 le32_to_cpu(ix->ei_block),
775 le32_to_cpu(ix[-1].ei_block));
777 BUG_ON(k && le32_to_cpu(ix->ei_block)
778 <= le32_to_cpu(ix[-1].ei_block));
779 if (block < le32_to_cpu(ix->ei_block))
783 BUG_ON(chix != path->p_idx);
790 * ext4_ext_binsearch:
791 * binary search for closest extent of the given block
792 * the header must be checked before calling this
795 ext4_ext_binsearch(struct inode *inode,
796 struct ext4_ext_path *path, ext4_lblk_t block)
798 struct ext4_extent_header *eh = path->p_hdr;
799 struct ext4_extent *r, *l, *m;
801 if (eh->eh_entries == 0) {
803 * this leaf is empty:
804 * we get such a leaf in split/add case
809 ext_debug("binsearch for %u: ", block);
811 l = EXT_FIRST_EXTENT(eh) + 1;
812 r = EXT_LAST_EXTENT(eh);
816 if (block < le32_to_cpu(m->ee_block))
820 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
821 m, le32_to_cpu(m->ee_block),
822 r, le32_to_cpu(r->ee_block));
826 ext_debug(" -> %d:%llu:[%d]%d ",
827 le32_to_cpu(path->p_ext->ee_block),
828 ext4_ext_pblock(path->p_ext),
829 ext4_ext_is_unwritten(path->p_ext),
830 ext4_ext_get_actual_len(path->p_ext));
832 #ifdef CHECK_BINSEARCH
834 struct ext4_extent *chex, *ex;
837 chex = ex = EXT_FIRST_EXTENT(eh);
838 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
839 BUG_ON(k && le32_to_cpu(ex->ee_block)
840 <= le32_to_cpu(ex[-1].ee_block));
841 if (block < le32_to_cpu(ex->ee_block))
845 BUG_ON(chex != path->p_ext);
851 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
853 struct ext4_extent_header *eh;
855 eh = ext_inode_hdr(inode);
858 eh->eh_magic = EXT4_EXT_MAGIC;
859 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
860 ext4_mark_inode_dirty(handle, inode);
864 struct ext4_ext_path *
865 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
866 struct ext4_ext_path **orig_path, int flags)
868 struct ext4_extent_header *eh;
869 struct buffer_head *bh;
870 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
871 short int depth, i, ppos = 0;
874 eh = ext_inode_hdr(inode);
875 depth = ext_depth(inode);
876 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
877 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
884 ext4_ext_drop_refs(path);
885 if (depth > path[0].p_maxdepth) {
887 *orig_path = path = NULL;
891 /* account possible depth increase */
892 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
895 return ERR_PTR(-ENOMEM);
896 path[0].p_maxdepth = depth + 1;
902 /* walk through the tree */
904 ext_debug("depth %d: num %d, max %d\n",
905 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
907 ext4_ext_binsearch_idx(inode, path + ppos, block);
908 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
909 path[ppos].p_depth = i;
910 path[ppos].p_ext = NULL;
912 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
919 eh = ext_block_hdr(bh);
921 path[ppos].p_bh = bh;
922 path[ppos].p_hdr = eh;
925 path[ppos].p_depth = i;
926 path[ppos].p_ext = NULL;
927 path[ppos].p_idx = NULL;
930 ext4_ext_binsearch(inode, path + ppos, block);
931 /* if not an empty leaf */
932 if (path[ppos].p_ext)
933 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
935 ext4_ext_show_path(inode, path);
940 ext4_ext_drop_refs(path);
948 * ext4_ext_insert_index:
949 * insert new index [@logical;@ptr] into the block at @curp;
950 * check where to insert: before @curp or after @curp
952 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
953 struct ext4_ext_path *curp,
954 int logical, ext4_fsblk_t ptr)
956 struct ext4_extent_idx *ix;
959 err = ext4_ext_get_access(handle, inode, curp);
963 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
964 EXT4_ERROR_INODE(inode,
965 "logical %d == ei_block %d!",
966 logical, le32_to_cpu(curp->p_idx->ei_block));
967 return -EFSCORRUPTED;
970 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
971 >= le16_to_cpu(curp->p_hdr->eh_max))) {
972 EXT4_ERROR_INODE(inode,
973 "eh_entries %d >= eh_max %d!",
974 le16_to_cpu(curp->p_hdr->eh_entries),
975 le16_to_cpu(curp->p_hdr->eh_max));
976 return -EFSCORRUPTED;
979 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
981 ext_debug("insert new index %d after: %llu\n", logical, ptr);
982 ix = curp->p_idx + 1;
985 ext_debug("insert new index %d before: %llu\n", logical, ptr);
989 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
992 ext_debug("insert new index %d: "
993 "move %d indices from 0x%p to 0x%p\n",
994 logical, len, ix, ix + 1);
995 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
998 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
999 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1000 return -EFSCORRUPTED;
1003 ix->ei_block = cpu_to_le32(logical);
1004 ext4_idx_store_pblock(ix, ptr);
1005 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1007 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1008 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1009 return -EFSCORRUPTED;
1012 err = ext4_ext_dirty(handle, inode, curp);
1013 ext4_std_error(inode->i_sb, err);
1020 * inserts new subtree into the path, using free index entry
1022 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1023 * - makes decision where to split
1024 * - moves remaining extents and index entries (right to the split point)
1025 * into the newly allocated blocks
1026 * - initializes subtree
1028 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1030 struct ext4_ext_path *path,
1031 struct ext4_extent *newext, int at)
1033 struct buffer_head *bh = NULL;
1034 int depth = ext_depth(inode);
1035 struct ext4_extent_header *neh;
1036 struct ext4_extent_idx *fidx;
1037 int i = at, k, m, a;
1038 ext4_fsblk_t newblock, oldblock;
1040 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1042 size_t ext_size = 0;
1044 /* make decision: where to split? */
1045 /* FIXME: now decision is simplest: at current extent */
1047 /* if current leaf will be split, then we should use
1048 * border from split point */
1049 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1050 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1051 return -EFSCORRUPTED;
1053 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1054 border = path[depth].p_ext[1].ee_block;
1055 ext_debug("leaf will be split."
1056 " next leaf starts at %d\n",
1057 le32_to_cpu(border));
1059 border = newext->ee_block;
1060 ext_debug("leaf will be added."
1061 " next leaf starts at %d\n",
1062 le32_to_cpu(border));
1066 * If error occurs, then we break processing
1067 * and mark filesystem read-only. index won't
1068 * be inserted and tree will be in consistent
1069 * state. Next mount will repair buffers too.
1073 * Get array to track all allocated blocks.
1074 * We need this to handle errors and free blocks
1077 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1081 /* allocate all needed blocks */
1082 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1083 for (a = 0; a < depth - at; a++) {
1084 newblock = ext4_ext_new_meta_block(handle, inode, path,
1085 newext, &err, flags);
1088 ablocks[a] = newblock;
1091 /* initialize new leaf */
1092 newblock = ablocks[--a];
1093 if (unlikely(newblock == 0)) {
1094 EXT4_ERROR_INODE(inode, "newblock == 0!");
1095 err = -EFSCORRUPTED;
1098 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1099 if (unlikely(!bh)) {
1105 err = ext4_journal_get_create_access(handle, bh);
1109 neh = ext_block_hdr(bh);
1110 neh->eh_entries = 0;
1111 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1112 neh->eh_magic = EXT4_EXT_MAGIC;
1115 /* move remainder of path[depth] to the new leaf */
1116 if (unlikely(path[depth].p_hdr->eh_entries !=
1117 path[depth].p_hdr->eh_max)) {
1118 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1119 path[depth].p_hdr->eh_entries,
1120 path[depth].p_hdr->eh_max);
1121 err = -EFSCORRUPTED;
1124 /* start copy from next extent */
1125 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1126 ext4_ext_show_move(inode, path, newblock, depth);
1128 struct ext4_extent *ex;
1129 ex = EXT_FIRST_EXTENT(neh);
1130 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1131 le16_add_cpu(&neh->eh_entries, m);
1134 /* zero out unused area in the extent block */
1135 ext_size = sizeof(struct ext4_extent_header) +
1136 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1137 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1138 ext4_extent_block_csum_set(inode, neh);
1139 set_buffer_uptodate(bh);
1142 err = ext4_handle_dirty_metadata(handle, inode, bh);
1148 /* correct old leaf */
1150 err = ext4_ext_get_access(handle, inode, path + depth);
1153 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1154 err = ext4_ext_dirty(handle, inode, path + depth);
1160 /* create intermediate indexes */
1162 if (unlikely(k < 0)) {
1163 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1164 err = -EFSCORRUPTED;
1168 ext_debug("create %d intermediate indices\n", k);
1169 /* insert new index into current index block */
1170 /* current depth stored in i var */
1173 oldblock = newblock;
1174 newblock = ablocks[--a];
1175 bh = sb_getblk(inode->i_sb, newblock);
1176 if (unlikely(!bh)) {
1182 err = ext4_journal_get_create_access(handle, bh);
1186 neh = ext_block_hdr(bh);
1187 neh->eh_entries = cpu_to_le16(1);
1188 neh->eh_magic = EXT4_EXT_MAGIC;
1189 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1190 neh->eh_depth = cpu_to_le16(depth - i);
1191 fidx = EXT_FIRST_INDEX(neh);
1192 fidx->ei_block = border;
1193 ext4_idx_store_pblock(fidx, oldblock);
1195 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1196 i, newblock, le32_to_cpu(border), oldblock);
1198 /* move remainder of path[i] to the new index block */
1199 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1200 EXT_LAST_INDEX(path[i].p_hdr))) {
1201 EXT4_ERROR_INODE(inode,
1202 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1203 le32_to_cpu(path[i].p_ext->ee_block));
1204 err = -EFSCORRUPTED;
1207 /* start copy indexes */
1208 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1209 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1210 EXT_MAX_INDEX(path[i].p_hdr));
1211 ext4_ext_show_move(inode, path, newblock, i);
1213 memmove(++fidx, path[i].p_idx,
1214 sizeof(struct ext4_extent_idx) * m);
1215 le16_add_cpu(&neh->eh_entries, m);
1217 /* zero out unused area in the extent block */
1218 ext_size = sizeof(struct ext4_extent_header) +
1219 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1220 memset(bh->b_data + ext_size, 0,
1221 inode->i_sb->s_blocksize - ext_size);
1222 ext4_extent_block_csum_set(inode, neh);
1223 set_buffer_uptodate(bh);
1226 err = ext4_handle_dirty_metadata(handle, inode, bh);
1232 /* correct old index */
1234 err = ext4_ext_get_access(handle, inode, path + i);
1237 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1238 err = ext4_ext_dirty(handle, inode, path + i);
1246 /* insert new index */
1247 err = ext4_ext_insert_index(handle, inode, path + at,
1248 le32_to_cpu(border), newblock);
1252 if (buffer_locked(bh))
1258 /* free all allocated blocks in error case */
1259 for (i = 0; i < depth; i++) {
1262 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1263 EXT4_FREE_BLOCKS_METADATA);
1272 * ext4_ext_grow_indepth:
1273 * implements tree growing procedure:
1274 * - allocates new block
1275 * - moves top-level data (index block or leaf) into the new block
1276 * - initializes new top-level, creating index that points to the
1277 * just created block
1279 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1282 struct ext4_extent_header *neh;
1283 struct buffer_head *bh;
1284 ext4_fsblk_t newblock, goal = 0;
1285 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1287 size_t ext_size = 0;
1289 /* Try to prepend new index to old one */
1290 if (ext_depth(inode))
1291 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1292 if (goal > le32_to_cpu(es->s_first_data_block)) {
1293 flags |= EXT4_MB_HINT_TRY_GOAL;
1296 goal = ext4_inode_to_goal_block(inode);
1297 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1302 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1307 err = ext4_journal_get_create_access(handle, bh);
1313 ext_size = sizeof(EXT4_I(inode)->i_data);
1314 /* move top-level index/leaf into new block */
1315 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1316 /* zero out unused area in the extent block */
1317 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1319 /* set size of new block */
1320 neh = ext_block_hdr(bh);
1321 /* old root could have indexes or leaves
1322 * so calculate e_max right way */
1323 if (ext_depth(inode))
1324 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1326 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1327 neh->eh_magic = EXT4_EXT_MAGIC;
1328 ext4_extent_block_csum_set(inode, neh);
1329 set_buffer_uptodate(bh);
1332 err = ext4_handle_dirty_metadata(handle, inode, bh);
1336 /* Update top-level index: num,max,pointer */
1337 neh = ext_inode_hdr(inode);
1338 neh->eh_entries = cpu_to_le16(1);
1339 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1340 if (neh->eh_depth == 0) {
1341 /* Root extent block becomes index block */
1342 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1343 EXT_FIRST_INDEX(neh)->ei_block =
1344 EXT_FIRST_EXTENT(neh)->ee_block;
1346 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1347 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1348 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1349 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1351 le16_add_cpu(&neh->eh_depth, 1);
1352 ext4_mark_inode_dirty(handle, inode);
1360 * ext4_ext_create_new_leaf:
1361 * finds empty index and adds new leaf.
1362 * if no free index is found, then it requests in-depth growing.
1364 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1365 unsigned int mb_flags,
1366 unsigned int gb_flags,
1367 struct ext4_ext_path **ppath,
1368 struct ext4_extent *newext)
1370 struct ext4_ext_path *path = *ppath;
1371 struct ext4_ext_path *curp;
1372 int depth, i, err = 0;
1375 i = depth = ext_depth(inode);
1377 /* walk up to the tree and look for free index entry */
1378 curp = path + depth;
1379 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1384 /* we use already allocated block for index block,
1385 * so subsequent data blocks should be contiguous */
1386 if (EXT_HAS_FREE_INDEX(curp)) {
1387 /* if we found index with free entry, then use that
1388 * entry: create all needed subtree and add new leaf */
1389 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1394 path = ext4_find_extent(inode,
1395 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1398 err = PTR_ERR(path);
1400 /* tree is full, time to grow in depth */
1401 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1406 path = ext4_find_extent(inode,
1407 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1410 err = PTR_ERR(path);
1415 * only first (depth 0 -> 1) produces free space;
1416 * in all other cases we have to split the grown tree
1418 depth = ext_depth(inode);
1419 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1420 /* now we need to split */
1430 * search the closest allocated block to the left for *logical
1431 * and returns it at @logical + it's physical address at @phys
1432 * if *logical is the smallest allocated block, the function
1433 * returns 0 at @phys
1434 * return value contains 0 (success) or error code
1436 static int ext4_ext_search_left(struct inode *inode,
1437 struct ext4_ext_path *path,
1438 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1440 struct ext4_extent_idx *ix;
1441 struct ext4_extent *ex;
1444 if (unlikely(path == NULL)) {
1445 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1446 return -EFSCORRUPTED;
1448 depth = path->p_depth;
1451 if (depth == 0 && path->p_ext == NULL)
1454 /* usually extent in the path covers blocks smaller
1455 * then *logical, but it can be that extent is the
1456 * first one in the file */
1458 ex = path[depth].p_ext;
1459 ee_len = ext4_ext_get_actual_len(ex);
1460 if (*logical < le32_to_cpu(ex->ee_block)) {
1461 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1462 EXT4_ERROR_INODE(inode,
1463 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1464 *logical, le32_to_cpu(ex->ee_block));
1465 return -EFSCORRUPTED;
1467 while (--depth >= 0) {
1468 ix = path[depth].p_idx;
1469 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1470 EXT4_ERROR_INODE(inode,
1471 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1472 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1473 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1474 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1476 return -EFSCORRUPTED;
1482 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1483 EXT4_ERROR_INODE(inode,
1484 "logical %d < ee_block %d + ee_len %d!",
1485 *logical, le32_to_cpu(ex->ee_block), ee_len);
1486 return -EFSCORRUPTED;
1489 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1490 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1495 * search the closest allocated block to the right for *logical
1496 * and returns it at @logical + it's physical address at @phys
1497 * if *logical is the largest allocated block, the function
1498 * returns 0 at @phys
1499 * return value contains 0 (success) or error code
1501 static int ext4_ext_search_right(struct inode *inode,
1502 struct ext4_ext_path *path,
1503 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1504 struct ext4_extent **ret_ex)
1506 struct buffer_head *bh = NULL;
1507 struct ext4_extent_header *eh;
1508 struct ext4_extent_idx *ix;
1509 struct ext4_extent *ex;
1511 int depth; /* Note, NOT eh_depth; depth from top of tree */
1514 if (unlikely(path == NULL)) {
1515 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1516 return -EFSCORRUPTED;
1518 depth = path->p_depth;
1521 if (depth == 0 && path->p_ext == NULL)
1524 /* usually extent in the path covers blocks smaller
1525 * then *logical, but it can be that extent is the
1526 * first one in the file */
1528 ex = path[depth].p_ext;
1529 ee_len = ext4_ext_get_actual_len(ex);
1530 if (*logical < le32_to_cpu(ex->ee_block)) {
1531 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1532 EXT4_ERROR_INODE(inode,
1533 "first_extent(path[%d].p_hdr) != ex",
1535 return -EFSCORRUPTED;
1537 while (--depth >= 0) {
1538 ix = path[depth].p_idx;
1539 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1540 EXT4_ERROR_INODE(inode,
1541 "ix != EXT_FIRST_INDEX *logical %d!",
1543 return -EFSCORRUPTED;
1549 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1550 EXT4_ERROR_INODE(inode,
1551 "logical %d < ee_block %d + ee_len %d!",
1552 *logical, le32_to_cpu(ex->ee_block), ee_len);
1553 return -EFSCORRUPTED;
1556 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1557 /* next allocated block in this leaf */
1562 /* go up and search for index to the right */
1563 while (--depth >= 0) {
1564 ix = path[depth].p_idx;
1565 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1569 /* we've gone up to the root and found no index to the right */
1573 /* we've found index to the right, let's
1574 * follow it and find the closest allocated
1575 * block to the right */
1577 block = ext4_idx_pblock(ix);
1578 while (++depth < path->p_depth) {
1579 /* subtract from p_depth to get proper eh_depth */
1580 bh = read_extent_tree_block(inode, block,
1581 path->p_depth - depth, 0);
1584 eh = ext_block_hdr(bh);
1585 ix = EXT_FIRST_INDEX(eh);
1586 block = ext4_idx_pblock(ix);
1590 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1593 eh = ext_block_hdr(bh);
1594 ex = EXT_FIRST_EXTENT(eh);
1596 *logical = le32_to_cpu(ex->ee_block);
1597 *phys = ext4_ext_pblock(ex);
1605 * ext4_ext_next_allocated_block:
1606 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1607 * NOTE: it considers block number from index entry as
1608 * allocated block. Thus, index entries have to be consistent
1612 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1616 BUG_ON(path == NULL);
1617 depth = path->p_depth;
1619 if (depth == 0 && path->p_ext == NULL)
1620 return EXT_MAX_BLOCKS;
1622 while (depth >= 0) {
1623 if (depth == path->p_depth) {
1625 if (path[depth].p_ext &&
1626 path[depth].p_ext !=
1627 EXT_LAST_EXTENT(path[depth].p_hdr))
1628 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1631 if (path[depth].p_idx !=
1632 EXT_LAST_INDEX(path[depth].p_hdr))
1633 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1638 return EXT_MAX_BLOCKS;
1642 * ext4_ext_next_leaf_block:
1643 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1645 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1649 BUG_ON(path == NULL);
1650 depth = path->p_depth;
1652 /* zero-tree has no leaf blocks at all */
1654 return EXT_MAX_BLOCKS;
1656 /* go to index block */
1659 while (depth >= 0) {
1660 if (path[depth].p_idx !=
1661 EXT_LAST_INDEX(path[depth].p_hdr))
1662 return (ext4_lblk_t)
1663 le32_to_cpu(path[depth].p_idx[1].ei_block);
1667 return EXT_MAX_BLOCKS;
1671 * ext4_ext_correct_indexes:
1672 * if leaf gets modified and modified extent is first in the leaf,
1673 * then we have to correct all indexes above.
1674 * TODO: do we need to correct tree in all cases?
1676 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1677 struct ext4_ext_path *path)
1679 struct ext4_extent_header *eh;
1680 int depth = ext_depth(inode);
1681 struct ext4_extent *ex;
1685 eh = path[depth].p_hdr;
1686 ex = path[depth].p_ext;
1688 if (unlikely(ex == NULL || eh == NULL)) {
1689 EXT4_ERROR_INODE(inode,
1690 "ex %p == NULL or eh %p == NULL", ex, eh);
1691 return -EFSCORRUPTED;
1695 /* there is no tree at all */
1699 if (ex != EXT_FIRST_EXTENT(eh)) {
1700 /* we correct tree if first leaf got modified only */
1705 * TODO: we need correction if border is smaller than current one
1708 border = path[depth].p_ext->ee_block;
1709 err = ext4_ext_get_access(handle, inode, path + k);
1712 path[k].p_idx->ei_block = border;
1713 err = ext4_ext_dirty(handle, inode, path + k);
1718 /* change all left-side indexes */
1719 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1721 err = ext4_ext_get_access(handle, inode, path + k);
1724 path[k].p_idx->ei_block = border;
1725 err = ext4_ext_dirty(handle, inode, path + k);
1734 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1735 struct ext4_extent *ex2)
1737 unsigned short ext1_ee_len, ext2_ee_len;
1739 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1742 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1743 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1745 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1746 le32_to_cpu(ex2->ee_block))
1750 * To allow future support for preallocated extents to be added
1751 * as an RO_COMPAT feature, refuse to merge to extents if
1752 * this can result in the top bit of ee_len being set.
1754 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1757 * The check for IO to unwritten extent is somewhat racy as we
1758 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1759 * dropping i_data_sem. But reserved blocks should save us in that
1762 if (ext4_ext_is_unwritten(ex1) &&
1763 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1764 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1765 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1767 #ifdef AGGRESSIVE_TEST
1768 if (ext1_ee_len >= 4)
1772 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1778 * This function tries to merge the "ex" extent to the next extent in the tree.
1779 * It always tries to merge towards right. If you want to merge towards
1780 * left, pass "ex - 1" as argument instead of "ex".
1781 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1782 * 1 if they got merged.
1784 static int ext4_ext_try_to_merge_right(struct inode *inode,
1785 struct ext4_ext_path *path,
1786 struct ext4_extent *ex)
1788 struct ext4_extent_header *eh;
1789 unsigned int depth, len;
1790 int merge_done = 0, unwritten;
1792 depth = ext_depth(inode);
1793 BUG_ON(path[depth].p_hdr == NULL);
1794 eh = path[depth].p_hdr;
1796 while (ex < EXT_LAST_EXTENT(eh)) {
1797 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1799 /* merge with next extent! */
1800 unwritten = ext4_ext_is_unwritten(ex);
1801 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1802 + ext4_ext_get_actual_len(ex + 1));
1804 ext4_ext_mark_unwritten(ex);
1806 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1807 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1808 * sizeof(struct ext4_extent);
1809 memmove(ex + 1, ex + 2, len);
1811 le16_add_cpu(&eh->eh_entries, -1);
1813 WARN_ON(eh->eh_entries == 0);
1814 if (!eh->eh_entries)
1815 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1822 * This function does a very simple check to see if we can collapse
1823 * an extent tree with a single extent tree leaf block into the inode.
1825 static void ext4_ext_try_to_merge_up(handle_t *handle,
1826 struct inode *inode,
1827 struct ext4_ext_path *path)
1830 unsigned max_root = ext4_ext_space_root(inode, 0);
1833 if ((path[0].p_depth != 1) ||
1834 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1835 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1839 * We need to modify the block allocation bitmap and the block
1840 * group descriptor to release the extent tree block. If we
1841 * can't get the journal credits, give up.
1843 if (ext4_journal_extend(handle, 2))
1847 * Copy the extent data up to the inode
1849 blk = ext4_idx_pblock(path[0].p_idx);
1850 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1851 sizeof(struct ext4_extent_idx);
1852 s += sizeof(struct ext4_extent_header);
1854 path[1].p_maxdepth = path[0].p_maxdepth;
1855 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1856 path[0].p_depth = 0;
1857 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1858 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1859 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1861 brelse(path[1].p_bh);
1862 ext4_free_blocks(handle, inode, NULL, blk, 1,
1863 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1867 * This function tries to merge the @ex extent to neighbours in the tree.
1868 * return 1 if merge left else 0.
1870 static void ext4_ext_try_to_merge(handle_t *handle,
1871 struct inode *inode,
1872 struct ext4_ext_path *path,
1873 struct ext4_extent *ex) {
1874 struct ext4_extent_header *eh;
1878 depth = ext_depth(inode);
1879 BUG_ON(path[depth].p_hdr == NULL);
1880 eh = path[depth].p_hdr;
1882 if (ex > EXT_FIRST_EXTENT(eh))
1883 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1886 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1888 ext4_ext_try_to_merge_up(handle, inode, path);
1892 * check if a portion of the "newext" extent overlaps with an
1895 * If there is an overlap discovered, it updates the length of the newext
1896 * such that there will be no overlap, and then returns 1.
1897 * If there is no overlap found, it returns 0.
1899 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1900 struct inode *inode,
1901 struct ext4_extent *newext,
1902 struct ext4_ext_path *path)
1905 unsigned int depth, len1;
1906 unsigned int ret = 0;
1908 b1 = le32_to_cpu(newext->ee_block);
1909 len1 = ext4_ext_get_actual_len(newext);
1910 depth = ext_depth(inode);
1911 if (!path[depth].p_ext)
1913 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1916 * get the next allocated block if the extent in the path
1917 * is before the requested block(s)
1920 b2 = ext4_ext_next_allocated_block(path);
1921 if (b2 == EXT_MAX_BLOCKS)
1923 b2 = EXT4_LBLK_CMASK(sbi, b2);
1926 /* check for wrap through zero on extent logical start block*/
1927 if (b1 + len1 < b1) {
1928 len1 = EXT_MAX_BLOCKS - b1;
1929 newext->ee_len = cpu_to_le16(len1);
1933 /* check for overlap */
1934 if (b1 + len1 > b2) {
1935 newext->ee_len = cpu_to_le16(b2 - b1);
1943 * ext4_ext_insert_extent:
1944 * tries to merge requsted extent into the existing extent or
1945 * inserts requested extent as new one into the tree,
1946 * creating new leaf in the no-space case.
1948 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1949 struct ext4_ext_path **ppath,
1950 struct ext4_extent *newext, int gb_flags)
1952 struct ext4_ext_path *path = *ppath;
1953 struct ext4_extent_header *eh;
1954 struct ext4_extent *ex, *fex;
1955 struct ext4_extent *nearex; /* nearest extent */
1956 struct ext4_ext_path *npath = NULL;
1957 int depth, len, err;
1959 int mb_flags = 0, unwritten;
1961 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1962 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1963 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1964 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1965 return -EFSCORRUPTED;
1967 depth = ext_depth(inode);
1968 ex = path[depth].p_ext;
1969 eh = path[depth].p_hdr;
1970 if (unlikely(path[depth].p_hdr == NULL)) {
1971 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1972 return -EFSCORRUPTED;
1975 /* try to insert block into found extent and return */
1976 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1979 * Try to see whether we should rather test the extent on
1980 * right from ex, or from the left of ex. This is because
1981 * ext4_find_extent() can return either extent on the
1982 * left, or on the right from the searched position. This
1983 * will make merging more effective.
1985 if (ex < EXT_LAST_EXTENT(eh) &&
1986 (le32_to_cpu(ex->ee_block) +
1987 ext4_ext_get_actual_len(ex) <
1988 le32_to_cpu(newext->ee_block))) {
1991 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1992 (le32_to_cpu(newext->ee_block) +
1993 ext4_ext_get_actual_len(newext) <
1994 le32_to_cpu(ex->ee_block)))
1997 /* Try to append newex to the ex */
1998 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1999 ext_debug("append [%d]%d block to %u:[%d]%d"
2001 ext4_ext_is_unwritten(newext),
2002 ext4_ext_get_actual_len(newext),
2003 le32_to_cpu(ex->ee_block),
2004 ext4_ext_is_unwritten(ex),
2005 ext4_ext_get_actual_len(ex),
2006 ext4_ext_pblock(ex));
2007 err = ext4_ext_get_access(handle, inode,
2011 unwritten = ext4_ext_is_unwritten(ex);
2012 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2013 + ext4_ext_get_actual_len(newext));
2015 ext4_ext_mark_unwritten(ex);
2016 eh = path[depth].p_hdr;
2022 /* Try to prepend newex to the ex */
2023 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2024 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2026 le32_to_cpu(newext->ee_block),
2027 ext4_ext_is_unwritten(newext),
2028 ext4_ext_get_actual_len(newext),
2029 le32_to_cpu(ex->ee_block),
2030 ext4_ext_is_unwritten(ex),
2031 ext4_ext_get_actual_len(ex),
2032 ext4_ext_pblock(ex));
2033 err = ext4_ext_get_access(handle, inode,
2038 unwritten = ext4_ext_is_unwritten(ex);
2039 ex->ee_block = newext->ee_block;
2040 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2041 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2042 + ext4_ext_get_actual_len(newext));
2044 ext4_ext_mark_unwritten(ex);
2045 eh = path[depth].p_hdr;
2051 depth = ext_depth(inode);
2052 eh = path[depth].p_hdr;
2053 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2056 /* probably next leaf has space for us? */
2057 fex = EXT_LAST_EXTENT(eh);
2058 next = EXT_MAX_BLOCKS;
2059 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2060 next = ext4_ext_next_leaf_block(path);
2061 if (next != EXT_MAX_BLOCKS) {
2062 ext_debug("next leaf block - %u\n", next);
2063 BUG_ON(npath != NULL);
2064 npath = ext4_find_extent(inode, next, NULL, 0);
2066 return PTR_ERR(npath);
2067 BUG_ON(npath->p_depth != path->p_depth);
2068 eh = npath[depth].p_hdr;
2069 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2070 ext_debug("next leaf isn't full(%d)\n",
2071 le16_to_cpu(eh->eh_entries));
2075 ext_debug("next leaf has no free space(%d,%d)\n",
2076 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2080 * There is no free space in the found leaf.
2081 * We're gonna add a new leaf in the tree.
2083 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2084 mb_flags |= EXT4_MB_USE_RESERVED;
2085 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2089 depth = ext_depth(inode);
2090 eh = path[depth].p_hdr;
2093 nearex = path[depth].p_ext;
2095 err = ext4_ext_get_access(handle, inode, path + depth);
2100 /* there is no extent in this leaf, create first one */
2101 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2102 le32_to_cpu(newext->ee_block),
2103 ext4_ext_pblock(newext),
2104 ext4_ext_is_unwritten(newext),
2105 ext4_ext_get_actual_len(newext));
2106 nearex = EXT_FIRST_EXTENT(eh);
2108 if (le32_to_cpu(newext->ee_block)
2109 > le32_to_cpu(nearex->ee_block)) {
2111 ext_debug("insert %u:%llu:[%d]%d before: "
2113 le32_to_cpu(newext->ee_block),
2114 ext4_ext_pblock(newext),
2115 ext4_ext_is_unwritten(newext),
2116 ext4_ext_get_actual_len(newext),
2121 BUG_ON(newext->ee_block == nearex->ee_block);
2122 ext_debug("insert %u:%llu:[%d]%d after: "
2124 le32_to_cpu(newext->ee_block),
2125 ext4_ext_pblock(newext),
2126 ext4_ext_is_unwritten(newext),
2127 ext4_ext_get_actual_len(newext),
2130 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2132 ext_debug("insert %u:%llu:[%d]%d: "
2133 "move %d extents from 0x%p to 0x%p\n",
2134 le32_to_cpu(newext->ee_block),
2135 ext4_ext_pblock(newext),
2136 ext4_ext_is_unwritten(newext),
2137 ext4_ext_get_actual_len(newext),
2138 len, nearex, nearex + 1);
2139 memmove(nearex + 1, nearex,
2140 len * sizeof(struct ext4_extent));
2144 le16_add_cpu(&eh->eh_entries, 1);
2145 path[depth].p_ext = nearex;
2146 nearex->ee_block = newext->ee_block;
2147 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2148 nearex->ee_len = newext->ee_len;
2151 /* try to merge extents */
2152 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2153 ext4_ext_try_to_merge(handle, inode, path, nearex);
2156 /* time to correct all indexes above */
2157 err = ext4_ext_correct_indexes(handle, inode, path);
2161 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2164 ext4_ext_drop_refs(npath);
2169 static int ext4_fill_fiemap_extents(struct inode *inode,
2170 ext4_lblk_t block, ext4_lblk_t num,
2171 struct fiemap_extent_info *fieinfo)
2173 struct ext4_ext_path *path = NULL;
2174 struct ext4_extent *ex;
2175 struct extent_status es;
2176 ext4_lblk_t next, next_del, start = 0, end = 0;
2177 ext4_lblk_t last = block + num;
2178 int exists, depth = 0, err = 0;
2179 unsigned int flags = 0;
2180 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2182 while (block < last && block != EXT_MAX_BLOCKS) {
2184 /* find extent for this block */
2185 down_read(&EXT4_I(inode)->i_data_sem);
2187 path = ext4_find_extent(inode, block, &path, 0);
2189 up_read(&EXT4_I(inode)->i_data_sem);
2190 err = PTR_ERR(path);
2195 depth = ext_depth(inode);
2196 if (unlikely(path[depth].p_hdr == NULL)) {
2197 up_read(&EXT4_I(inode)->i_data_sem);
2198 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2199 err = -EFSCORRUPTED;
2202 ex = path[depth].p_ext;
2203 next = ext4_ext_next_allocated_block(path);
2208 /* there is no extent yet, so try to allocate
2209 * all requested space */
2212 } else if (le32_to_cpu(ex->ee_block) > block) {
2213 /* need to allocate space before found extent */
2215 end = le32_to_cpu(ex->ee_block);
2216 if (block + num < end)
2218 } else if (block >= le32_to_cpu(ex->ee_block)
2219 + ext4_ext_get_actual_len(ex)) {
2220 /* need to allocate space after found extent */
2225 } else if (block >= le32_to_cpu(ex->ee_block)) {
2227 * some part of requested space is covered
2231 end = le32_to_cpu(ex->ee_block)
2232 + ext4_ext_get_actual_len(ex);
2233 if (block + num < end)
2239 BUG_ON(end <= start);
2243 es.es_len = end - start;
2246 es.es_lblk = le32_to_cpu(ex->ee_block);
2247 es.es_len = ext4_ext_get_actual_len(ex);
2248 es.es_pblk = ext4_ext_pblock(ex);
2249 if (ext4_ext_is_unwritten(ex))
2250 flags |= FIEMAP_EXTENT_UNWRITTEN;
2254 * Find delayed extent and update es accordingly. We call
2255 * it even in !exists case to find out whether es is the
2256 * last existing extent or not.
2258 next_del = ext4_find_delayed_extent(inode, &es);
2259 if (!exists && next_del) {
2261 flags |= (FIEMAP_EXTENT_DELALLOC |
2262 FIEMAP_EXTENT_UNKNOWN);
2264 up_read(&EXT4_I(inode)->i_data_sem);
2266 if (unlikely(es.es_len == 0)) {
2267 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2268 err = -EFSCORRUPTED;
2273 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2274 * we need to check next == EXT_MAX_BLOCKS because it is
2275 * possible that an extent is with unwritten and delayed
2276 * status due to when an extent is delayed allocated and
2277 * is allocated by fallocate status tree will track both of
2280 * So we could return a unwritten and delayed extent, and
2281 * its block is equal to 'next'.
2283 if (next == next_del && next == EXT_MAX_BLOCKS) {
2284 flags |= FIEMAP_EXTENT_LAST;
2285 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2286 next != EXT_MAX_BLOCKS)) {
2287 EXT4_ERROR_INODE(inode,
2288 "next extent == %u, next "
2289 "delalloc extent = %u",
2291 err = -EFSCORRUPTED;
2297 err = fiemap_fill_next_extent(fieinfo,
2298 (__u64)es.es_lblk << blksize_bits,
2299 (__u64)es.es_pblk << blksize_bits,
2300 (__u64)es.es_len << blksize_bits,
2310 block = es.es_lblk + es.es_len;
2313 ext4_ext_drop_refs(path);
2318 static int ext4_fill_es_cache_info(struct inode *inode,
2319 ext4_lblk_t block, ext4_lblk_t num,
2320 struct fiemap_extent_info *fieinfo)
2322 ext4_lblk_t next, end = block + num - 1;
2323 struct extent_status es;
2324 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2328 while (block <= end) {
2331 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2333 if (ext4_es_is_unwritten(&es))
2334 flags |= FIEMAP_EXTENT_UNWRITTEN;
2335 if (ext4_es_is_delayed(&es))
2336 flags |= (FIEMAP_EXTENT_DELALLOC |
2337 FIEMAP_EXTENT_UNKNOWN);
2338 if (ext4_es_is_hole(&es))
2339 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2341 flags |= FIEMAP_EXTENT_LAST;
2342 if (flags & (FIEMAP_EXTENT_DELALLOC|
2343 EXT4_FIEMAP_EXTENT_HOLE))
2346 es.es_pblk = ext4_es_pblock(&es);
2347 err = fiemap_fill_next_extent(fieinfo,
2348 (__u64)es.es_lblk << blksize_bits,
2349 (__u64)es.es_pblk << blksize_bits,
2350 (__u64)es.es_len << blksize_bits,
2365 * ext4_ext_determine_hole - determine hole around given block
2366 * @inode: inode we lookup in
2367 * @path: path in extent tree to @lblk
2368 * @lblk: pointer to logical block around which we want to determine hole
2370 * Determine hole length (and start if easily possible) around given logical
2371 * block. We don't try too hard to find the beginning of the hole but @path
2372 * actually points to extent before @lblk, we provide it.
2374 * The function returns the length of a hole starting at @lblk. We update @lblk
2375 * to the beginning of the hole if we managed to find it.
2377 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2378 struct ext4_ext_path *path,
2381 int depth = ext_depth(inode);
2382 struct ext4_extent *ex;
2385 ex = path[depth].p_ext;
2387 /* there is no extent yet, so gap is [0;-] */
2389 len = EXT_MAX_BLOCKS;
2390 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2391 len = le32_to_cpu(ex->ee_block) - *lblk;
2392 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2393 + ext4_ext_get_actual_len(ex)) {
2396 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2397 next = ext4_ext_next_allocated_block(path);
2398 BUG_ON(next == *lblk);
2407 * ext4_ext_put_gap_in_cache:
2408 * calculate boundaries of the gap that the requested block fits into
2409 * and cache this gap
2412 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2413 ext4_lblk_t hole_len)
2415 struct extent_status es;
2417 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2418 hole_start + hole_len - 1, &es);
2420 /* There's delayed extent containing lblock? */
2421 if (es.es_lblk <= hole_start)
2423 hole_len = min(es.es_lblk - hole_start, hole_len);
2425 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2426 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2427 EXTENT_STATUS_HOLE);
2432 * removes index from the index block.
2434 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2435 struct ext4_ext_path *path, int depth)
2440 /* free index block */
2442 path = path + depth;
2443 leaf = ext4_idx_pblock(path->p_idx);
2444 if (unlikely(path->p_hdr->eh_entries == 0)) {
2445 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2446 return -EFSCORRUPTED;
2448 err = ext4_ext_get_access(handle, inode, path);
2452 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2453 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2454 len *= sizeof(struct ext4_extent_idx);
2455 memmove(path->p_idx, path->p_idx + 1, len);
2458 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2459 err = ext4_ext_dirty(handle, inode, path);
2462 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2463 trace_ext4_ext_rm_idx(inode, leaf);
2465 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2466 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2468 while (--depth >= 0) {
2469 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2472 err = ext4_ext_get_access(handle, inode, path);
2475 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2476 err = ext4_ext_dirty(handle, inode, path);
2484 * ext4_ext_calc_credits_for_single_extent:
2485 * This routine returns max. credits that needed to insert an extent
2486 * to the extent tree.
2487 * When pass the actual path, the caller should calculate credits
2490 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2491 struct ext4_ext_path *path)
2494 int depth = ext_depth(inode);
2497 /* probably there is space in leaf? */
2498 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2499 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2502 * There are some space in the leaf tree, no
2503 * need to account for leaf block credit
2505 * bitmaps and block group descriptor blocks
2506 * and other metadata blocks still need to be
2509 /* 1 bitmap, 1 block group descriptor */
2510 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2515 return ext4_chunk_trans_blocks(inode, nrblocks);
2519 * How many index/leaf blocks need to change/allocate to add @extents extents?
2521 * If we add a single extent, then in the worse case, each tree level
2522 * index/leaf need to be changed in case of the tree split.
2524 * If more extents are inserted, they could cause the whole tree split more
2525 * than once, but this is really rare.
2527 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2532 /* If we are converting the inline data, only one is needed here. */
2533 if (ext4_has_inline_data(inode))
2536 depth = ext_depth(inode);
2546 static inline int get_default_free_blocks_flags(struct inode *inode)
2548 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2549 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2550 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2551 else if (ext4_should_journal_data(inode))
2552 return EXT4_FREE_BLOCKS_FORGET;
2557 * ext4_rereserve_cluster - increment the reserved cluster count when
2558 * freeing a cluster with a pending reservation
2560 * @inode - file containing the cluster
2561 * @lblk - logical block in cluster to be reserved
2563 * Increments the reserved cluster count and adjusts quota in a bigalloc
2564 * file system when freeing a partial cluster containing at least one
2565 * delayed and unwritten block. A partial cluster meeting that
2566 * requirement will have a pending reservation. If so, the
2567 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2568 * defer reserved and allocated space accounting to a subsequent call
2571 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2573 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2574 struct ext4_inode_info *ei = EXT4_I(inode);
2576 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2578 spin_lock(&ei->i_block_reservation_lock);
2579 ei->i_reserved_data_blocks++;
2580 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2581 spin_unlock(&ei->i_block_reservation_lock);
2583 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2584 ext4_remove_pending(inode, lblk);
2587 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2588 struct ext4_extent *ex,
2589 struct partial_cluster *partial,
2590 ext4_lblk_t from, ext4_lblk_t to)
2592 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2593 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2594 ext4_fsblk_t last_pblk, pblk;
2598 /* only extent tail removal is allowed */
2599 if (from < le32_to_cpu(ex->ee_block) ||
2600 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2601 ext4_error(sbi->s_sb,
2602 "strange request: removal(2) %u-%u from %u:%u",
2603 from, to, le32_to_cpu(ex->ee_block), ee_len);
2607 #ifdef EXTENTS_STATS
2608 spin_lock(&sbi->s_ext_stats_lock);
2609 sbi->s_ext_blocks += ee_len;
2610 sbi->s_ext_extents++;
2611 if (ee_len < sbi->s_ext_min)
2612 sbi->s_ext_min = ee_len;
2613 if (ee_len > sbi->s_ext_max)
2614 sbi->s_ext_max = ee_len;
2615 if (ext_depth(inode) > sbi->s_depth_max)
2616 sbi->s_depth_max = ext_depth(inode);
2617 spin_unlock(&sbi->s_ext_stats_lock);
2620 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2623 * if we have a partial cluster, and it's different from the
2624 * cluster of the last block in the extent, we free it
2626 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2628 if (partial->state != initial &&
2629 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2630 if (partial->state == tofree) {
2631 flags = get_default_free_blocks_flags(inode);
2632 if (ext4_is_pending(inode, partial->lblk))
2633 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2634 ext4_free_blocks(handle, inode, NULL,
2635 EXT4_C2B(sbi, partial->pclu),
2636 sbi->s_cluster_ratio, flags);
2637 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2638 ext4_rereserve_cluster(inode, partial->lblk);
2640 partial->state = initial;
2643 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2644 pblk = ext4_ext_pblock(ex) + ee_len - num;
2647 * We free the partial cluster at the end of the extent (if any),
2648 * unless the cluster is used by another extent (partial_cluster
2649 * state is nofree). If a partial cluster exists here, it must be
2650 * shared with the last block in the extent.
2652 flags = get_default_free_blocks_flags(inode);
2654 /* partial, left end cluster aligned, right end unaligned */
2655 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2656 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2657 (partial->state != nofree)) {
2658 if (ext4_is_pending(inode, to))
2659 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2660 ext4_free_blocks(handle, inode, NULL,
2661 EXT4_PBLK_CMASK(sbi, last_pblk),
2662 sbi->s_cluster_ratio, flags);
2663 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2664 ext4_rereserve_cluster(inode, to);
2665 partial->state = initial;
2666 flags = get_default_free_blocks_flags(inode);
2669 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2672 * For bigalloc file systems, we never free a partial cluster
2673 * at the beginning of the extent. Instead, we check to see if we
2674 * need to free it on a subsequent call to ext4_remove_blocks,
2675 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2677 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2678 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2680 /* reset the partial cluster if we've freed past it */
2681 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2682 partial->state = initial;
2685 * If we've freed the entire extent but the beginning is not left
2686 * cluster aligned and is not marked as ineligible for freeing we
2687 * record the partial cluster at the beginning of the extent. It
2688 * wasn't freed by the preceding ext4_free_blocks() call, and we
2689 * need to look farther to the left to determine if it's to be freed
2690 * (not shared with another extent). Else, reset the partial
2691 * cluster - we're either done freeing or the beginning of the
2692 * extent is left cluster aligned.
2694 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2695 if (partial->state == initial) {
2696 partial->pclu = EXT4_B2C(sbi, pblk);
2697 partial->lblk = from;
2698 partial->state = tofree;
2701 partial->state = initial;
2708 * ext4_ext_rm_leaf() Removes the extents associated with the
2709 * blocks appearing between "start" and "end". Both "start"
2710 * and "end" must appear in the same extent or EIO is returned.
2712 * @handle: The journal handle
2713 * @inode: The files inode
2714 * @path: The path to the leaf
2715 * @partial_cluster: The cluster which we'll have to free if all extents
2716 * has been released from it. However, if this value is
2717 * negative, it's a cluster just to the right of the
2718 * punched region and it must not be freed.
2719 * @start: The first block to remove
2720 * @end: The last block to remove
2723 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2724 struct ext4_ext_path *path,
2725 struct partial_cluster *partial,
2726 ext4_lblk_t start, ext4_lblk_t end)
2728 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2729 int err = 0, correct_index = 0;
2730 int depth = ext_depth(inode), credits;
2731 struct ext4_extent_header *eh;
2734 ext4_lblk_t ex_ee_block;
2735 unsigned short ex_ee_len;
2736 unsigned unwritten = 0;
2737 struct ext4_extent *ex;
2740 /* the header must be checked already in ext4_ext_remove_space() */
2741 ext_debug("truncate since %u in leaf to %u\n", start, end);
2742 if (!path[depth].p_hdr)
2743 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2744 eh = path[depth].p_hdr;
2745 if (unlikely(path[depth].p_hdr == NULL)) {
2746 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2747 return -EFSCORRUPTED;
2749 /* find where to start removing */
2750 ex = path[depth].p_ext;
2752 ex = EXT_LAST_EXTENT(eh);
2754 ex_ee_block = le32_to_cpu(ex->ee_block);
2755 ex_ee_len = ext4_ext_get_actual_len(ex);
2757 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2759 while (ex >= EXT_FIRST_EXTENT(eh) &&
2760 ex_ee_block + ex_ee_len > start) {
2762 if (ext4_ext_is_unwritten(ex))
2767 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2768 unwritten, ex_ee_len);
2769 path[depth].p_ext = ex;
2771 a = ex_ee_block > start ? ex_ee_block : start;
2772 b = ex_ee_block+ex_ee_len - 1 < end ?
2773 ex_ee_block+ex_ee_len - 1 : end;
2775 ext_debug(" border %u:%u\n", a, b);
2777 /* If this extent is beyond the end of the hole, skip it */
2778 if (end < ex_ee_block) {
2780 * We're going to skip this extent and move to another,
2781 * so note that its first cluster is in use to avoid
2782 * freeing it when removing blocks. Eventually, the
2783 * right edge of the truncated/punched region will
2784 * be just to the left.
2786 if (sbi->s_cluster_ratio > 1) {
2787 pblk = ext4_ext_pblock(ex);
2788 partial->pclu = EXT4_B2C(sbi, pblk);
2789 partial->state = nofree;
2792 ex_ee_block = le32_to_cpu(ex->ee_block);
2793 ex_ee_len = ext4_ext_get_actual_len(ex);
2795 } else if (b != ex_ee_block + ex_ee_len - 1) {
2796 EXT4_ERROR_INODE(inode,
2797 "can not handle truncate %u:%u "
2799 start, end, ex_ee_block,
2800 ex_ee_block + ex_ee_len - 1);
2801 err = -EFSCORRUPTED;
2803 } else if (a != ex_ee_block) {
2804 /* remove tail of the extent */
2805 num = a - ex_ee_block;
2807 /* remove whole extent: excellent! */
2811 * 3 for leaf, sb, and inode plus 2 (bmap and group
2812 * descriptor) for each block group; assume two block
2813 * groups plus ex_ee_len/blocks_per_block_group for
2816 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2817 if (ex == EXT_FIRST_EXTENT(eh)) {
2819 credits += (ext_depth(inode)) + 1;
2821 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2823 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2827 err = ext4_ext_get_access(handle, inode, path + depth);
2831 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2836 /* this extent is removed; mark slot entirely unused */
2837 ext4_ext_store_pblock(ex, 0);
2839 ex->ee_len = cpu_to_le16(num);
2841 * Do not mark unwritten if all the blocks in the
2842 * extent have been removed.
2844 if (unwritten && num)
2845 ext4_ext_mark_unwritten(ex);
2847 * If the extent was completely released,
2848 * we need to remove it from the leaf
2851 if (end != EXT_MAX_BLOCKS - 1) {
2853 * For hole punching, we need to scoot all the
2854 * extents up when an extent is removed so that
2855 * we dont have blank extents in the middle
2857 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2858 sizeof(struct ext4_extent));
2860 /* Now get rid of the one at the end */
2861 memset(EXT_LAST_EXTENT(eh), 0,
2862 sizeof(struct ext4_extent));
2864 le16_add_cpu(&eh->eh_entries, -1);
2867 err = ext4_ext_dirty(handle, inode, path + depth);
2871 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2872 ext4_ext_pblock(ex));
2874 ex_ee_block = le32_to_cpu(ex->ee_block);
2875 ex_ee_len = ext4_ext_get_actual_len(ex);
2878 if (correct_index && eh->eh_entries)
2879 err = ext4_ext_correct_indexes(handle, inode, path);
2882 * If there's a partial cluster and at least one extent remains in
2883 * the leaf, free the partial cluster if it isn't shared with the
2884 * current extent. If it is shared with the current extent
2885 * we reset the partial cluster because we've reached the start of the
2886 * truncated/punched region and we're done removing blocks.
2888 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2889 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2890 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2891 int flags = get_default_free_blocks_flags(inode);
2893 if (ext4_is_pending(inode, partial->lblk))
2894 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2895 ext4_free_blocks(handle, inode, NULL,
2896 EXT4_C2B(sbi, partial->pclu),
2897 sbi->s_cluster_ratio, flags);
2898 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2899 ext4_rereserve_cluster(inode, partial->lblk);
2901 partial->state = initial;
2904 /* if this leaf is free, then we should
2905 * remove it from index block above */
2906 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2907 err = ext4_ext_rm_idx(handle, inode, path, depth);
2914 * ext4_ext_more_to_rm:
2915 * returns 1 if current index has to be freed (even partial)
2918 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2920 BUG_ON(path->p_idx == NULL);
2922 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2926 * if truncate on deeper level happened, it wasn't partial,
2927 * so we have to consider current index for truncation
2929 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2934 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2937 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2938 int depth = ext_depth(inode);
2939 struct ext4_ext_path *path = NULL;
2940 struct partial_cluster partial;
2946 partial.state = initial;
2948 ext_debug("truncate since %u to %u\n", start, end);
2950 /* probably first extent we're gonna free will be last in block */
2951 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2953 return PTR_ERR(handle);
2956 trace_ext4_ext_remove_space(inode, start, end, depth);
2959 * Check if we are removing extents inside the extent tree. If that
2960 * is the case, we are going to punch a hole inside the extent tree
2961 * so we have to check whether we need to split the extent covering
2962 * the last block to remove so we can easily remove the part of it
2963 * in ext4_ext_rm_leaf().
2965 if (end < EXT_MAX_BLOCKS - 1) {
2966 struct ext4_extent *ex;
2967 ext4_lblk_t ee_block, ex_end, lblk;
2970 /* find extent for or closest extent to this block */
2971 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2973 ext4_journal_stop(handle);
2974 return PTR_ERR(path);
2976 depth = ext_depth(inode);
2977 /* Leaf not may not exist only if inode has no blocks at all */
2978 ex = path[depth].p_ext;
2981 EXT4_ERROR_INODE(inode,
2982 "path[%d].p_hdr == NULL",
2984 err = -EFSCORRUPTED;
2989 ee_block = le32_to_cpu(ex->ee_block);
2990 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2993 * See if the last block is inside the extent, if so split
2994 * the extent at 'end' block so we can easily remove the
2995 * tail of the first part of the split extent in
2996 * ext4_ext_rm_leaf().
2998 if (end >= ee_block && end < ex_end) {
3001 * If we're going to split the extent, note that
3002 * the cluster containing the block after 'end' is
3003 * in use to avoid freeing it when removing blocks.
3005 if (sbi->s_cluster_ratio > 1) {
3006 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
3007 partial.pclu = EXT4_B2C(sbi, pblk);
3008 partial.state = nofree;
3012 * Split the extent in two so that 'end' is the last
3013 * block in the first new extent. Also we should not
3014 * fail removing space due to ENOSPC so try to use
3015 * reserved block if that happens.
3017 err = ext4_force_split_extent_at(handle, inode, &path,
3022 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
3023 partial.state == initial) {
3025 * If we're punching, there's an extent to the right.
3026 * If the partial cluster hasn't been set, set it to
3027 * that extent's first cluster and its state to nofree
3028 * so it won't be freed should it contain blocks to be
3029 * removed. If it's already set (tofree/nofree), we're
3030 * retrying and keep the original partial cluster info
3031 * so a cluster marked tofree as a result of earlier
3032 * extent removal is not lost.
3035 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3040 partial.pclu = EXT4_B2C(sbi, pblk);
3041 partial.state = nofree;
3046 * We start scanning from right side, freeing all the blocks
3047 * after i_size and walking into the tree depth-wise.
3049 depth = ext_depth(inode);
3054 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3056 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3059 ext4_journal_stop(handle);
3062 path[0].p_maxdepth = path[0].p_depth = depth;
3063 path[0].p_hdr = ext_inode_hdr(inode);
3066 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3067 err = -EFSCORRUPTED;
3073 while (i >= 0 && err == 0) {
3075 /* this is leaf block */
3076 err = ext4_ext_rm_leaf(handle, inode, path,
3077 &partial, start, end);
3078 /* root level has p_bh == NULL, brelse() eats this */
3079 brelse(path[i].p_bh);
3080 path[i].p_bh = NULL;
3085 /* this is index block */
3086 if (!path[i].p_hdr) {
3087 ext_debug("initialize header\n");
3088 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3091 if (!path[i].p_idx) {
3092 /* this level hasn't been touched yet */
3093 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3094 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3095 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3097 le16_to_cpu(path[i].p_hdr->eh_entries));
3099 /* we were already here, see at next index */
3103 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3104 i, EXT_FIRST_INDEX(path[i].p_hdr),
3106 if (ext4_ext_more_to_rm(path + i)) {
3107 struct buffer_head *bh;
3108 /* go to the next level */
3109 ext_debug("move to level %d (block %llu)\n",
3110 i + 1, ext4_idx_pblock(path[i].p_idx));
3111 memset(path + i + 1, 0, sizeof(*path));
3112 bh = read_extent_tree_block(inode,
3113 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3116 /* should we reset i_size? */
3120 /* Yield here to deal with large extent trees.
3121 * Should be a no-op if we did IO above. */
3123 if (WARN_ON(i + 1 > depth)) {
3124 err = -EFSCORRUPTED;
3127 path[i + 1].p_bh = bh;
3129 /* save actual number of indexes since this
3130 * number is changed at the next iteration */
3131 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3134 /* we finished processing this index, go up */
3135 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3136 /* index is empty, remove it;
3137 * handle must be already prepared by the
3138 * truncatei_leaf() */
3139 err = ext4_ext_rm_idx(handle, inode, path, i);
3141 /* root level has p_bh == NULL, brelse() eats this */
3142 brelse(path[i].p_bh);
3143 path[i].p_bh = NULL;
3145 ext_debug("return to level %d\n", i);
3149 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3150 path->p_hdr->eh_entries);
3153 * if there's a partial cluster and we have removed the first extent
3154 * in the file, then we also free the partial cluster, if any
3156 if (partial.state == tofree && err == 0) {
3157 int flags = get_default_free_blocks_flags(inode);
3159 if (ext4_is_pending(inode, partial.lblk))
3160 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3161 ext4_free_blocks(handle, inode, NULL,
3162 EXT4_C2B(sbi, partial.pclu),
3163 sbi->s_cluster_ratio, flags);
3164 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3165 ext4_rereserve_cluster(inode, partial.lblk);
3166 partial.state = initial;
3169 /* TODO: flexible tree reduction should be here */
3170 if (path->p_hdr->eh_entries == 0) {
3172 * truncate to zero freed all the tree,
3173 * so we need to correct eh_depth
3175 err = ext4_ext_get_access(handle, inode, path);
3177 ext_inode_hdr(inode)->eh_depth = 0;
3178 ext_inode_hdr(inode)->eh_max =
3179 cpu_to_le16(ext4_ext_space_root(inode, 0));
3180 err = ext4_ext_dirty(handle, inode, path);
3184 ext4_ext_drop_refs(path);
3189 ext4_journal_stop(handle);
3195 * called at mount time
3197 void ext4_ext_init(struct super_block *sb)
3200 * possible initialization would be here
3203 if (ext4_has_feature_extents(sb)) {
3204 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3205 printk(KERN_INFO "EXT4-fs: file extents enabled"
3206 #ifdef AGGRESSIVE_TEST
3207 ", aggressive tests"
3209 #ifdef CHECK_BINSEARCH
3212 #ifdef EXTENTS_STATS
3217 #ifdef EXTENTS_STATS
3218 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3219 EXT4_SB(sb)->s_ext_min = 1 << 30;
3220 EXT4_SB(sb)->s_ext_max = 0;
3226 * called at umount time
3228 void ext4_ext_release(struct super_block *sb)
3230 if (!ext4_has_feature_extents(sb))
3233 #ifdef EXTENTS_STATS
3234 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3235 struct ext4_sb_info *sbi = EXT4_SB(sb);
3236 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3237 sbi->s_ext_blocks, sbi->s_ext_extents,
3238 sbi->s_ext_blocks / sbi->s_ext_extents);
3239 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3240 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3245 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3247 ext4_lblk_t ee_block;
3248 ext4_fsblk_t ee_pblock;
3249 unsigned int ee_len;
3251 ee_block = le32_to_cpu(ex->ee_block);
3252 ee_len = ext4_ext_get_actual_len(ex);
3253 ee_pblock = ext4_ext_pblock(ex);
3258 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3259 EXTENT_STATUS_WRITTEN);
3262 /* FIXME!! we need to try to merge to left or right after zero-out */
3263 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3265 ext4_fsblk_t ee_pblock;
3266 unsigned int ee_len;
3268 ee_len = ext4_ext_get_actual_len(ex);
3269 ee_pblock = ext4_ext_pblock(ex);
3270 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3275 * ext4_split_extent_at() splits an extent at given block.
3277 * @handle: the journal handle
3278 * @inode: the file inode
3279 * @path: the path to the extent
3280 * @split: the logical block where the extent is splitted.
3281 * @split_flags: indicates if the extent could be zeroout if split fails, and
3282 * the states(init or unwritten) of new extents.
3283 * @flags: flags used to insert new extent to extent tree.
3286 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3287 * of which are deterimined by split_flag.
3289 * There are two cases:
3290 * a> the extent are splitted into two extent.
3291 * b> split is not needed, and just mark the extent.
3293 * return 0 on success.
3295 static int ext4_split_extent_at(handle_t *handle,
3296 struct inode *inode,
3297 struct ext4_ext_path **ppath,
3302 struct ext4_ext_path *path = *ppath;
3303 ext4_fsblk_t newblock;
3304 ext4_lblk_t ee_block;
3305 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3306 struct ext4_extent *ex2 = NULL;
3307 unsigned int ee_len, depth;
3310 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3311 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3313 ext_debug("ext4_split_extents_at: inode %lu, logical"
3314 "block %llu\n", inode->i_ino, (unsigned long long)split);
3316 ext4_ext_show_leaf(inode, path);
3318 depth = ext_depth(inode);
3319 ex = path[depth].p_ext;
3320 ee_block = le32_to_cpu(ex->ee_block);
3321 ee_len = ext4_ext_get_actual_len(ex);
3322 newblock = split - ee_block + ext4_ext_pblock(ex);
3324 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3325 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3326 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3327 EXT4_EXT_MARK_UNWRIT1 |
3328 EXT4_EXT_MARK_UNWRIT2));
3330 err = ext4_ext_get_access(handle, inode, path + depth);
3334 if (split == ee_block) {
3336 * case b: block @split is the block that the extent begins with
3337 * then we just change the state of the extent, and splitting
3340 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3341 ext4_ext_mark_unwritten(ex);
3343 ext4_ext_mark_initialized(ex);
3345 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3346 ext4_ext_try_to_merge(handle, inode, path, ex);
3348 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3353 memcpy(&orig_ex, ex, sizeof(orig_ex));
3354 ex->ee_len = cpu_to_le16(split - ee_block);
3355 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3356 ext4_ext_mark_unwritten(ex);
3359 * path may lead to new leaf, not to original leaf any more
3360 * after ext4_ext_insert_extent() returns,
3362 err = ext4_ext_dirty(handle, inode, path + depth);
3364 goto fix_extent_len;
3367 ex2->ee_block = cpu_to_le32(split);
3368 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3369 ext4_ext_store_pblock(ex2, newblock);
3370 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3371 ext4_ext_mark_unwritten(ex2);
3373 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3374 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3375 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3376 if (split_flag & EXT4_EXT_DATA_VALID1) {
3377 err = ext4_ext_zeroout(inode, ex2);
3378 zero_ex.ee_block = ex2->ee_block;
3379 zero_ex.ee_len = cpu_to_le16(
3380 ext4_ext_get_actual_len(ex2));
3381 ext4_ext_store_pblock(&zero_ex,
3382 ext4_ext_pblock(ex2));
3384 err = ext4_ext_zeroout(inode, ex);
3385 zero_ex.ee_block = ex->ee_block;
3386 zero_ex.ee_len = cpu_to_le16(
3387 ext4_ext_get_actual_len(ex));
3388 ext4_ext_store_pblock(&zero_ex,
3389 ext4_ext_pblock(ex));
3392 err = ext4_ext_zeroout(inode, &orig_ex);
3393 zero_ex.ee_block = orig_ex.ee_block;
3394 zero_ex.ee_len = cpu_to_le16(
3395 ext4_ext_get_actual_len(&orig_ex));
3396 ext4_ext_store_pblock(&zero_ex,
3397 ext4_ext_pblock(&orig_ex));
3401 goto fix_extent_len;
3402 /* update the extent length and mark as initialized */
3403 ex->ee_len = cpu_to_le16(ee_len);
3404 ext4_ext_try_to_merge(handle, inode, path, ex);
3405 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3407 goto fix_extent_len;
3409 /* update extent status tree */
3410 err = ext4_zeroout_es(inode, &zero_ex);
3414 goto fix_extent_len;
3417 ext4_ext_show_leaf(inode, path);
3421 ex->ee_len = orig_ex.ee_len;
3422 ext4_ext_dirty(handle, inode, path + path->p_depth);
3427 * ext4_split_extents() splits an extent and mark extent which is covered
3428 * by @map as split_flags indicates
3430 * It may result in splitting the extent into multiple extents (up to three)
3431 * There are three possibilities:
3432 * a> There is no split required
3433 * b> Splits in two extents: Split is happening at either end of the extent
3434 * c> Splits in three extents: Somone is splitting in middle of the extent
3437 static int ext4_split_extent(handle_t *handle,
3438 struct inode *inode,
3439 struct ext4_ext_path **ppath,
3440 struct ext4_map_blocks *map,
3444 struct ext4_ext_path *path = *ppath;
3445 ext4_lblk_t ee_block;
3446 struct ext4_extent *ex;
3447 unsigned int ee_len, depth;
3450 int split_flag1, flags1;
3451 int allocated = map->m_len;
3453 depth = ext_depth(inode);
3454 ex = path[depth].p_ext;
3455 ee_block = le32_to_cpu(ex->ee_block);
3456 ee_len = ext4_ext_get_actual_len(ex);
3457 unwritten = ext4_ext_is_unwritten(ex);
3459 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3460 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3461 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3463 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3464 EXT4_EXT_MARK_UNWRIT2;
3465 if (split_flag & EXT4_EXT_DATA_VALID2)
3466 split_flag1 |= EXT4_EXT_DATA_VALID1;
3467 err = ext4_split_extent_at(handle, inode, ppath,
3468 map->m_lblk + map->m_len, split_flag1, flags1);
3472 allocated = ee_len - (map->m_lblk - ee_block);
3475 * Update path is required because previous ext4_split_extent_at() may
3476 * result in split of original leaf or extent zeroout.
3478 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3480 return PTR_ERR(path);
3481 depth = ext_depth(inode);
3482 ex = path[depth].p_ext;
3484 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3485 (unsigned long) map->m_lblk);
3486 return -EFSCORRUPTED;
3488 unwritten = ext4_ext_is_unwritten(ex);
3491 if (map->m_lblk >= ee_block) {
3492 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3494 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3495 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3496 EXT4_EXT_MARK_UNWRIT2);
3498 err = ext4_split_extent_at(handle, inode, ppath,
3499 map->m_lblk, split_flag1, flags);
3504 ext4_ext_show_leaf(inode, path);
3506 return err ? err : allocated;
3510 * This function is called by ext4_ext_map_blocks() if someone tries to write
3511 * to an unwritten extent. It may result in splitting the unwritten
3512 * extent into multiple extents (up to three - one initialized and two
3514 * There are three possibilities:
3515 * a> There is no split required: Entire extent should be initialized
3516 * b> Splits in two extents: Write is happening at either end of the extent
3517 * c> Splits in three extents: Somone is writing in middle of the extent
3520 * - The extent pointed to by 'path' is unwritten.
3521 * - The extent pointed to by 'path' contains a superset
3522 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3524 * Post-conditions on success:
3525 * - the returned value is the number of blocks beyond map->l_lblk
3526 * that are allocated and initialized.
3527 * It is guaranteed to be >= map->m_len.
3529 static int ext4_ext_convert_to_initialized(handle_t *handle,
3530 struct inode *inode,
3531 struct ext4_map_blocks *map,
3532 struct ext4_ext_path **ppath,
3535 struct ext4_ext_path *path = *ppath;
3536 struct ext4_sb_info *sbi;
3537 struct ext4_extent_header *eh;
3538 struct ext4_map_blocks split_map;
3539 struct ext4_extent zero_ex1, zero_ex2;
3540 struct ext4_extent *ex, *abut_ex;
3541 ext4_lblk_t ee_block, eof_block;
3542 unsigned int ee_len, depth, map_len = map->m_len;
3543 int allocated = 0, max_zeroout = 0;
3545 int split_flag = EXT4_EXT_DATA_VALID2;
3547 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3548 "block %llu, max_blocks %u\n", inode->i_ino,
3549 (unsigned long long)map->m_lblk, map_len);
3551 sbi = EXT4_SB(inode->i_sb);
3552 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3553 inode->i_sb->s_blocksize_bits;
3554 if (eof_block < map->m_lblk + map_len)
3555 eof_block = map->m_lblk + map_len;
3557 depth = ext_depth(inode);
3558 eh = path[depth].p_hdr;
3559 ex = path[depth].p_ext;
3560 ee_block = le32_to_cpu(ex->ee_block);
3561 ee_len = ext4_ext_get_actual_len(ex);
3562 zero_ex1.ee_len = 0;
3563 zero_ex2.ee_len = 0;
3565 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3567 /* Pre-conditions */
3568 BUG_ON(!ext4_ext_is_unwritten(ex));
3569 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3572 * Attempt to transfer newly initialized blocks from the currently
3573 * unwritten extent to its neighbor. This is much cheaper
3574 * than an insertion followed by a merge as those involve costly
3575 * memmove() calls. Transferring to the left is the common case in
3576 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3577 * followed by append writes.
3579 * Limitations of the current logic:
3580 * - L1: we do not deal with writes covering the whole extent.
3581 * This would require removing the extent if the transfer
3583 * - L2: we only attempt to merge with an extent stored in the
3584 * same extent tree node.
3586 if ((map->m_lblk == ee_block) &&
3587 /* See if we can merge left */
3588 (map_len < ee_len) && /*L1*/
3589 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3590 ext4_lblk_t prev_lblk;
3591 ext4_fsblk_t prev_pblk, ee_pblk;
3592 unsigned int prev_len;
3595 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3596 prev_len = ext4_ext_get_actual_len(abut_ex);
3597 prev_pblk = ext4_ext_pblock(abut_ex);
3598 ee_pblk = ext4_ext_pblock(ex);
3601 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3602 * upon those conditions:
3603 * - C1: abut_ex is initialized,
3604 * - C2: abut_ex is logically abutting ex,
3605 * - C3: abut_ex is physically abutting ex,
3606 * - C4: abut_ex can receive the additional blocks without
3607 * overflowing the (initialized) length limit.
3609 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3610 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3611 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3612 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3613 err = ext4_ext_get_access(handle, inode, path + depth);
3617 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3620 /* Shift the start of ex by 'map_len' blocks */
3621 ex->ee_block = cpu_to_le32(ee_block + map_len);
3622 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3623 ex->ee_len = cpu_to_le16(ee_len - map_len);
3624 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3626 /* Extend abut_ex by 'map_len' blocks */
3627 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3629 /* Result: number of initialized blocks past m_lblk */
3630 allocated = map_len;
3632 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3633 (map_len < ee_len) && /*L1*/
3634 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3635 /* See if we can merge right */
3636 ext4_lblk_t next_lblk;
3637 ext4_fsblk_t next_pblk, ee_pblk;
3638 unsigned int next_len;
3641 next_lblk = le32_to_cpu(abut_ex->ee_block);
3642 next_len = ext4_ext_get_actual_len(abut_ex);
3643 next_pblk = ext4_ext_pblock(abut_ex);
3644 ee_pblk = ext4_ext_pblock(ex);
3647 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3648 * upon those conditions:
3649 * - C1: abut_ex is initialized,
3650 * - C2: abut_ex is logically abutting ex,
3651 * - C3: abut_ex is physically abutting ex,
3652 * - C4: abut_ex can receive the additional blocks without
3653 * overflowing the (initialized) length limit.
3655 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3656 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3657 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3658 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3659 err = ext4_ext_get_access(handle, inode, path + depth);
3663 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3666 /* Shift the start of abut_ex by 'map_len' blocks */
3667 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3668 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3669 ex->ee_len = cpu_to_le16(ee_len - map_len);
3670 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3672 /* Extend abut_ex by 'map_len' blocks */
3673 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3675 /* Result: number of initialized blocks past m_lblk */
3676 allocated = map_len;
3680 /* Mark the block containing both extents as dirty */
3681 ext4_ext_dirty(handle, inode, path + depth);
3683 /* Update path to point to the right extent */
3684 path[depth].p_ext = abut_ex;
3687 allocated = ee_len - (map->m_lblk - ee_block);
3689 WARN_ON(map->m_lblk < ee_block);
3691 * It is safe to convert extent to initialized via explicit
3692 * zeroout only if extent is fully inside i_size or new_size.
3694 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3696 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3697 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3698 (inode->i_sb->s_blocksize_bits - 10);
3700 if (IS_ENCRYPTED(inode))
3705 * 1. split the extent into three extents.
3706 * 2. split the extent into two extents, zeroout the head of the first
3708 * 3. split the extent into two extents, zeroout the tail of the second
3710 * 4. split the extent into two extents with out zeroout.
3711 * 5. no splitting needed, just possibly zeroout the head and / or the
3712 * tail of the extent.
3714 split_map.m_lblk = map->m_lblk;
3715 split_map.m_len = map->m_len;
3717 if (max_zeroout && (allocated > split_map.m_len)) {
3718 if (allocated <= max_zeroout) {
3721 cpu_to_le32(split_map.m_lblk +
3724 cpu_to_le16(allocated - split_map.m_len);
3725 ext4_ext_store_pblock(&zero_ex1,
3726 ext4_ext_pblock(ex) + split_map.m_lblk +
3727 split_map.m_len - ee_block);
3728 err = ext4_ext_zeroout(inode, &zero_ex1);
3731 split_map.m_len = allocated;
3733 if (split_map.m_lblk - ee_block + split_map.m_len <
3736 if (split_map.m_lblk != ee_block) {
3737 zero_ex2.ee_block = ex->ee_block;
3738 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3740 ext4_ext_store_pblock(&zero_ex2,
3741 ext4_ext_pblock(ex));
3742 err = ext4_ext_zeroout(inode, &zero_ex2);
3747 split_map.m_len += split_map.m_lblk - ee_block;
3748 split_map.m_lblk = ee_block;
3749 allocated = map->m_len;
3753 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3758 /* If we have gotten a failure, don't zero out status tree */
3760 err = ext4_zeroout_es(inode, &zero_ex1);
3762 err = ext4_zeroout_es(inode, &zero_ex2);
3764 return err ? err : allocated;
3768 * This function is called by ext4_ext_map_blocks() from
3769 * ext4_get_blocks_dio_write() when DIO to write
3770 * to an unwritten extent.
3772 * Writing to an unwritten extent may result in splitting the unwritten
3773 * extent into multiple initialized/unwritten extents (up to three)
3774 * There are three possibilities:
3775 * a> There is no split required: Entire extent should be unwritten
3776 * b> Splits in two extents: Write is happening at either end of the extent
3777 * c> Splits in three extents: Somone is writing in middle of the extent
3779 * This works the same way in the case of initialized -> unwritten conversion.
3781 * One of more index blocks maybe needed if the extent tree grow after
3782 * the unwritten extent split. To prevent ENOSPC occur at the IO
3783 * complete, we need to split the unwritten extent before DIO submit
3784 * the IO. The unwritten extent called at this time will be split
3785 * into three unwritten extent(at most). After IO complete, the part
3786 * being filled will be convert to initialized by the end_io callback function
3787 * via ext4_convert_unwritten_extents().
3789 * Returns the size of unwritten extent to be written on success.
3791 static int ext4_split_convert_extents(handle_t *handle,
3792 struct inode *inode,
3793 struct ext4_map_blocks *map,
3794 struct ext4_ext_path **ppath,
3797 struct ext4_ext_path *path = *ppath;
3798 ext4_lblk_t eof_block;
3799 ext4_lblk_t ee_block;
3800 struct ext4_extent *ex;
3801 unsigned int ee_len;
3802 int split_flag = 0, depth;
3804 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3805 __func__, inode->i_ino,
3806 (unsigned long long)map->m_lblk, map->m_len);
3808 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3809 inode->i_sb->s_blocksize_bits;
3810 if (eof_block < map->m_lblk + map->m_len)
3811 eof_block = map->m_lblk + map->m_len;
3813 * It is safe to convert extent to initialized via explicit
3814 * zeroout only if extent is fully insde i_size or new_size.
3816 depth = ext_depth(inode);
3817 ex = path[depth].p_ext;
3818 ee_block = le32_to_cpu(ex->ee_block);
3819 ee_len = ext4_ext_get_actual_len(ex);
3821 /* Convert to unwritten */
3822 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3823 split_flag |= EXT4_EXT_DATA_VALID1;
3824 /* Convert to initialized */
3825 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3826 split_flag |= ee_block + ee_len <= eof_block ?
3827 EXT4_EXT_MAY_ZEROOUT : 0;
3828 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3830 flags |= EXT4_GET_BLOCKS_PRE_IO;
3831 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3834 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3835 struct inode *inode,
3836 struct ext4_map_blocks *map,
3837 struct ext4_ext_path **ppath)
3839 struct ext4_ext_path *path = *ppath;
3840 struct ext4_extent *ex;
3841 ext4_lblk_t ee_block;
3842 unsigned int ee_len;
3846 depth = ext_depth(inode);
3847 ex = path[depth].p_ext;
3848 ee_block = le32_to_cpu(ex->ee_block);
3849 ee_len = ext4_ext_get_actual_len(ex);
3851 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3852 "block %llu, max_blocks %u\n", inode->i_ino,
3853 (unsigned long long)ee_block, ee_len);
3855 /* If extent is larger than requested it is a clear sign that we still
3856 * have some extent state machine issues left. So extent_split is still
3858 * TODO: Once all related issues will be fixed this situation should be
3861 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3862 #ifdef CONFIG_EXT4_DEBUG
3863 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3864 " len %u; IO logical block %llu, len %u",
3865 inode->i_ino, (unsigned long long)ee_block, ee_len,
3866 (unsigned long long)map->m_lblk, map->m_len);
3868 err = ext4_split_convert_extents(handle, inode, map, ppath,
3869 EXT4_GET_BLOCKS_CONVERT);
3872 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3874 return PTR_ERR(path);
3875 depth = ext_depth(inode);
3876 ex = path[depth].p_ext;
3879 err = ext4_ext_get_access(handle, inode, path + depth);
3882 /* first mark the extent as initialized */
3883 ext4_ext_mark_initialized(ex);
3885 /* note: ext4_ext_correct_indexes() isn't needed here because
3886 * borders are not changed
3888 ext4_ext_try_to_merge(handle, inode, path, ex);
3890 /* Mark modified extent as dirty */
3891 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3893 ext4_ext_show_leaf(inode, path);
3898 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3900 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3902 struct ext4_ext_path *path,
3906 struct ext4_extent_header *eh;
3907 struct ext4_extent *last_ex;
3909 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3912 depth = ext_depth(inode);
3913 eh = path[depth].p_hdr;
3916 * We're going to remove EOFBLOCKS_FL entirely in future so we
3917 * do not care for this case anymore. Simply remove the flag
3918 * if there are no extents.
3920 if (unlikely(!eh->eh_entries))
3922 last_ex = EXT_LAST_EXTENT(eh);
3924 * We should clear the EOFBLOCKS_FL flag if we are writing the
3925 * last block in the last extent in the file. We test this by
3926 * first checking to see if the caller to
3927 * ext4_ext_get_blocks() was interested in the last block (or
3928 * a block beyond the last block) in the current extent. If
3929 * this turns out to be false, we can bail out from this
3930 * function immediately.
3932 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3933 ext4_ext_get_actual_len(last_ex))
3936 * If the caller does appear to be planning to write at or
3937 * beyond the end of the current extent, we then test to see
3938 * if the current extent is the last extent in the file, by
3939 * checking to make sure it was reached via the rightmost node
3940 * at each level of the tree.
3942 for (i = depth-1; i >= 0; i--)
3943 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3946 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3947 return ext4_mark_inode_dirty(handle, inode);
3951 convert_initialized_extent(handle_t *handle, struct inode *inode,
3952 struct ext4_map_blocks *map,
3953 struct ext4_ext_path **ppath,
3954 unsigned int allocated)
3956 struct ext4_ext_path *path = *ppath;
3957 struct ext4_extent *ex;
3958 ext4_lblk_t ee_block;
3959 unsigned int ee_len;
3964 * Make sure that the extent is no bigger than we support with
3967 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3968 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3970 depth = ext_depth(inode);
3971 ex = path[depth].p_ext;
3972 ee_block = le32_to_cpu(ex->ee_block);
3973 ee_len = ext4_ext_get_actual_len(ex);
3975 ext_debug("%s: inode %lu, logical"
3976 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3977 (unsigned long long)ee_block, ee_len);
3979 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3980 err = ext4_split_convert_extents(handle, inode, map, ppath,
3981 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3984 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3986 return PTR_ERR(path);
3987 depth = ext_depth(inode);
3988 ex = path[depth].p_ext;
3990 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3991 (unsigned long) map->m_lblk);
3992 return -EFSCORRUPTED;
3996 err = ext4_ext_get_access(handle, inode, path + depth);
3999 /* first mark the extent as unwritten */
4000 ext4_ext_mark_unwritten(ex);
4002 /* note: ext4_ext_correct_indexes() isn't needed here because
4003 * borders are not changed
4005 ext4_ext_try_to_merge(handle, inode, path, ex);
4007 /* Mark modified extent as dirty */
4008 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4011 ext4_ext_show_leaf(inode, path);
4013 ext4_update_inode_fsync_trans(handle, inode, 1);
4014 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4017 map->m_flags |= EXT4_MAP_UNWRITTEN;
4018 if (allocated > map->m_len)
4019 allocated = map->m_len;
4020 map->m_len = allocated;
4025 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4026 struct ext4_map_blocks *map,
4027 struct ext4_ext_path **ppath, int flags,
4028 unsigned int allocated, ext4_fsblk_t newblock)
4030 struct ext4_ext_path *path = *ppath;
4034 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4035 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4036 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4038 ext4_ext_show_leaf(inode, path);
4041 * When writing into unwritten space, we should not fail to
4042 * allocate metadata blocks for the new extent block if needed.
4044 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4046 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4047 allocated, newblock);
4049 /* get_block() before submit the IO, split the extent */
4050 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4051 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4052 flags | EXT4_GET_BLOCKS_CONVERT);
4055 map->m_flags |= EXT4_MAP_UNWRITTEN;
4058 /* IO end_io complete, convert the filled extent to written */
4059 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4060 if (flags & EXT4_GET_BLOCKS_ZERO) {
4061 if (allocated > map->m_len)
4062 allocated = map->m_len;
4063 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4068 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4071 ext4_update_inode_fsync_trans(handle, inode, 1);
4072 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4076 map->m_flags |= EXT4_MAP_MAPPED;
4077 map->m_pblk = newblock;
4078 if (allocated > map->m_len)
4079 allocated = map->m_len;
4080 map->m_len = allocated;
4083 /* buffered IO case */
4085 * repeat fallocate creation request
4086 * we already have an unwritten extent
4088 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4089 map->m_flags |= EXT4_MAP_UNWRITTEN;
4093 /* buffered READ or buffered write_begin() lookup */
4094 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4096 * We have blocks reserved already. We
4097 * return allocated blocks so that delalloc
4098 * won't do block reservation for us. But
4099 * the buffer head will be unmapped so that
4100 * a read from the block returns 0s.
4102 map->m_flags |= EXT4_MAP_UNWRITTEN;
4106 /* buffered write, writepage time, convert*/
4107 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4109 ext4_update_inode_fsync_trans(handle, inode, 1);
4116 map->m_flags |= EXT4_MAP_NEW;
4117 if (allocated > map->m_len)
4118 allocated = map->m_len;
4119 map->m_len = allocated;
4122 map->m_flags |= EXT4_MAP_MAPPED;
4123 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4124 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4130 if (allocated > map->m_len)
4131 allocated = map->m_len;
4132 ext4_ext_show_leaf(inode, path);
4133 map->m_pblk = newblock;
4134 map->m_len = allocated;
4136 return err ? err : allocated;
4140 * get_implied_cluster_alloc - check to see if the requested
4141 * allocation (in the map structure) overlaps with a cluster already
4142 * allocated in an extent.
4143 * @sb The filesystem superblock structure
4144 * @map The requested lblk->pblk mapping
4145 * @ex The extent structure which might contain an implied
4146 * cluster allocation
4148 * This function is called by ext4_ext_map_blocks() after we failed to
4149 * find blocks that were already in the inode's extent tree. Hence,
4150 * we know that the beginning of the requested region cannot overlap
4151 * the extent from the inode's extent tree. There are three cases we
4152 * want to catch. The first is this case:
4154 * |--- cluster # N--|
4155 * |--- extent ---| |---- requested region ---|
4158 * The second case that we need to test for is this one:
4160 * |--------- cluster # N ----------------|
4161 * |--- requested region --| |------- extent ----|
4162 * |=======================|
4164 * The third case is when the requested region lies between two extents
4165 * within the same cluster:
4166 * |------------- cluster # N-------------|
4167 * |----- ex -----| |---- ex_right ----|
4168 * |------ requested region ------|
4169 * |================|
4171 * In each of the above cases, we need to set the map->m_pblk and
4172 * map->m_len so it corresponds to the return the extent labelled as
4173 * "|====|" from cluster #N, since it is already in use for data in
4174 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4175 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4176 * as a new "allocated" block region. Otherwise, we will return 0 and
4177 * ext4_ext_map_blocks() will then allocate one or more new clusters
4178 * by calling ext4_mb_new_blocks().
4180 static int get_implied_cluster_alloc(struct super_block *sb,
4181 struct ext4_map_blocks *map,
4182 struct ext4_extent *ex,
4183 struct ext4_ext_path *path)
4185 struct ext4_sb_info *sbi = EXT4_SB(sb);
4186 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4187 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4188 ext4_lblk_t rr_cluster_start;
4189 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4190 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4191 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4193 /* The extent passed in that we are trying to match */
4194 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4195 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4197 /* The requested region passed into ext4_map_blocks() */
4198 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4200 if ((rr_cluster_start == ex_cluster_end) ||
4201 (rr_cluster_start == ex_cluster_start)) {
4202 if (rr_cluster_start == ex_cluster_end)
4203 ee_start += ee_len - 1;
4204 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4205 map->m_len = min(map->m_len,
4206 (unsigned) sbi->s_cluster_ratio - c_offset);
4208 * Check for and handle this case:
4210 * |--------- cluster # N-------------|
4211 * |------- extent ----|
4212 * |--- requested region ---|
4216 if (map->m_lblk < ee_block)
4217 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4220 * Check for the case where there is already another allocated
4221 * block to the right of 'ex' but before the end of the cluster.
4223 * |------------- cluster # N-------------|
4224 * |----- ex -----| |---- ex_right ----|
4225 * |------ requested region ------|
4226 * |================|
4228 if (map->m_lblk > ee_block) {
4229 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4230 map->m_len = min(map->m_len, next - map->m_lblk);
4233 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4237 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4243 * Block allocation/map/preallocation routine for extents based files
4246 * Need to be called with
4247 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4248 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4250 * return > 0, number of of blocks already mapped/allocated
4251 * if create == 0 and these are pre-allocated blocks
4252 * buffer head is unmapped
4253 * otherwise blocks are mapped
4255 * return = 0, if plain look up failed (blocks have not been allocated)
4256 * buffer head is unmapped
4258 * return < 0, error case.
4260 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4261 struct ext4_map_blocks *map, int flags)
4263 struct ext4_ext_path *path = NULL;
4264 struct ext4_extent newex, *ex, *ex2;
4265 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4266 ext4_fsblk_t newblock = 0;
4267 int free_on_err = 0, err = 0, depth, ret;
4268 unsigned int allocated = 0, offset = 0;
4269 unsigned int allocated_clusters = 0;
4270 struct ext4_allocation_request ar;
4271 ext4_lblk_t cluster_offset;
4272 bool map_from_cluster = false;
4274 ext_debug("blocks %u/%u requested for inode %lu\n",
4275 map->m_lblk, map->m_len, inode->i_ino);
4276 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4278 /* find extent for this block */
4279 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4281 err = PTR_ERR(path);
4286 depth = ext_depth(inode);
4289 * consistent leaf must not be empty;
4290 * this situation is possible, though, _during_ tree modification;
4291 * this is why assert can't be put in ext4_find_extent()
4293 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4294 EXT4_ERROR_INODE(inode, "bad extent address "
4295 "lblock: %lu, depth: %d pblock %lld",
4296 (unsigned long) map->m_lblk, depth,
4297 path[depth].p_block);
4298 err = -EFSCORRUPTED;
4302 ex = path[depth].p_ext;
4304 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4305 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4306 unsigned short ee_len;
4310 * unwritten extents are treated as holes, except that
4311 * we split out initialized portions during a write.
4313 ee_len = ext4_ext_get_actual_len(ex);
4315 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4317 /* if found extent covers block, simply return it */
4318 if (in_range(map->m_lblk, ee_block, ee_len)) {
4319 newblock = map->m_lblk - ee_block + ee_start;
4320 /* number of remaining blocks in the extent */
4321 allocated = ee_len - (map->m_lblk - ee_block);
4322 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4323 ee_block, ee_len, newblock);
4326 * If the extent is initialized check whether the
4327 * caller wants to convert it to unwritten.
4329 if ((!ext4_ext_is_unwritten(ex)) &&
4330 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4331 allocated = convert_initialized_extent(
4332 handle, inode, map, &path,
4335 } else if (!ext4_ext_is_unwritten(ex))
4338 ret = ext4_ext_handle_unwritten_extents(
4339 handle, inode, map, &path, flags,
4340 allocated, newblock);
4350 * requested block isn't allocated yet;
4351 * we couldn't try to create block if create flag is zero
4353 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4354 ext4_lblk_t hole_start, hole_len;
4356 hole_start = map->m_lblk;
4357 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4359 * put just found gap into cache to speed up
4360 * subsequent requests
4362 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4364 /* Update hole_len to reflect hole size after map->m_lblk */
4365 if (hole_start != map->m_lblk)
4366 hole_len -= map->m_lblk - hole_start;
4368 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4374 * Okay, we need to do block allocation.
4376 newex.ee_block = cpu_to_le32(map->m_lblk);
4377 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4380 * If we are doing bigalloc, check to see if the extent returned
4381 * by ext4_find_extent() implies a cluster we can use.
4383 if (cluster_offset && ex &&
4384 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4385 ar.len = allocated = map->m_len;
4386 newblock = map->m_pblk;
4387 map_from_cluster = true;
4388 goto got_allocated_blocks;
4391 /* find neighbour allocated blocks */
4392 ar.lleft = map->m_lblk;
4393 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4396 ar.lright = map->m_lblk;
4398 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4402 /* Check if the extent after searching to the right implies a
4403 * cluster we can use. */
4404 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4405 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4406 ar.len = allocated = map->m_len;
4407 newblock = map->m_pblk;
4408 map_from_cluster = true;
4409 goto got_allocated_blocks;
4413 * See if request is beyond maximum number of blocks we can have in
4414 * a single extent. For an initialized extent this limit is
4415 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4416 * EXT_UNWRITTEN_MAX_LEN.
4418 if (map->m_len > EXT_INIT_MAX_LEN &&
4419 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4420 map->m_len = EXT_INIT_MAX_LEN;
4421 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4422 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4423 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4425 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4426 newex.ee_len = cpu_to_le16(map->m_len);
4427 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4429 allocated = ext4_ext_get_actual_len(&newex);
4431 allocated = map->m_len;
4433 /* allocate new block */
4435 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4436 ar.logical = map->m_lblk;
4438 * We calculate the offset from the beginning of the cluster
4439 * for the logical block number, since when we allocate a
4440 * physical cluster, the physical block should start at the
4441 * same offset from the beginning of the cluster. This is
4442 * needed so that future calls to get_implied_cluster_alloc()
4445 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4446 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4448 ar.logical -= offset;
4449 if (S_ISREG(inode->i_mode))
4450 ar.flags = EXT4_MB_HINT_DATA;
4452 /* disable in-core preallocation for non-regular files */
4454 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4455 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4456 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4457 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4458 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4459 ar.flags |= EXT4_MB_USE_RESERVED;
4460 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4463 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4464 ar.goal, newblock, allocated);
4466 allocated_clusters = ar.len;
4467 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4468 if (ar.len > allocated)
4471 got_allocated_blocks:
4472 /* try to insert new extent into found leaf and return */
4473 ext4_ext_store_pblock(&newex, newblock + offset);
4474 newex.ee_len = cpu_to_le16(ar.len);
4475 /* Mark unwritten */
4476 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4477 ext4_ext_mark_unwritten(&newex);
4478 map->m_flags |= EXT4_MAP_UNWRITTEN;
4482 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4483 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4486 err = ext4_ext_insert_extent(handle, inode, &path,
4489 if (err && free_on_err) {
4490 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4491 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4492 /* free data blocks we just allocated */
4493 /* not a good idea to call discard here directly,
4494 * but otherwise we'd need to call it every free() */
4495 ext4_discard_preallocations(inode);
4496 ext4_free_blocks(handle, inode, NULL, newblock,
4497 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4501 /* previous routine could use block we allocated */
4502 newblock = ext4_ext_pblock(&newex);
4503 allocated = ext4_ext_get_actual_len(&newex);
4504 if (allocated > map->m_len)
4505 allocated = map->m_len;
4506 map->m_flags |= EXT4_MAP_NEW;
4509 * Reduce the reserved cluster count to reflect successful deferred
4510 * allocation of delayed allocated clusters or direct allocation of
4511 * clusters discovered to be delayed allocated. Once allocated, a
4512 * cluster is not included in the reserved count.
4514 if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4515 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4517 * When allocating delayed allocated clusters, simply
4518 * reduce the reserved cluster count and claim quota
4520 ext4_da_update_reserve_space(inode, allocated_clusters,
4523 ext4_lblk_t lblk, len;
4527 * When allocating non-delayed allocated clusters
4528 * (from fallocate, filemap, DIO, or clusters
4529 * allocated when delalloc has been disabled by
4530 * ext4_nonda_switch), reduce the reserved cluster
4531 * count by the number of allocated clusters that
4532 * have previously been delayed allocated. Quota
4533 * has been claimed by ext4_mb_new_blocks() above,
4534 * so release the quota reservations made for any
4535 * previously delayed allocated clusters.
4537 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4538 len = allocated_clusters << sbi->s_cluster_bits;
4539 n = ext4_es_delayed_clu(inode, lblk, len);
4541 ext4_da_update_reserve_space(inode, (int) n, 0);
4546 * Cache the extent and update transaction to commit on fdatasync only
4547 * when it is _not_ an unwritten extent.
4549 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4550 ext4_update_inode_fsync_trans(handle, inode, 1);
4552 ext4_update_inode_fsync_trans(handle, inode, 0);
4554 if (allocated > map->m_len)
4555 allocated = map->m_len;
4556 ext4_ext_show_leaf(inode, path);
4557 map->m_flags |= EXT4_MAP_MAPPED;
4558 map->m_pblk = newblock;
4559 map->m_len = allocated;
4561 ext4_ext_drop_refs(path);
4564 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4565 err ? err : allocated);
4566 return err ? err : allocated;
4569 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4571 struct super_block *sb = inode->i_sb;
4572 ext4_lblk_t last_block;
4576 * TODO: optimization is possible here.
4577 * Probably we need not scan at all,
4578 * because page truncation is enough.
4581 /* we have to know where to truncate from in crash case */
4582 EXT4_I(inode)->i_disksize = inode->i_size;
4583 err = ext4_mark_inode_dirty(handle, inode);
4587 last_block = (inode->i_size + sb->s_blocksize - 1)
4588 >> EXT4_BLOCK_SIZE_BITS(sb);
4590 err = ext4_es_remove_extent(inode, last_block,
4591 EXT_MAX_BLOCKS - last_block);
4592 if (err == -ENOMEM) {
4594 congestion_wait(BLK_RW_ASYNC, HZ/50);
4599 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4602 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4603 ext4_lblk_t len, loff_t new_size,
4606 struct inode *inode = file_inode(file);
4612 struct ext4_map_blocks map;
4613 unsigned int credits;
4616 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4617 map.m_lblk = offset;
4620 * Don't normalize the request if it can fit in one extent so
4621 * that it doesn't get unnecessarily split into multiple
4624 if (len <= EXT_UNWRITTEN_MAX_LEN)
4625 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4628 * credits to insert 1 extent into extent tree
4630 credits = ext4_chunk_trans_blocks(inode, len);
4631 depth = ext_depth(inode);
4634 while (ret >= 0 && len) {
4636 * Recalculate credits when extent tree depth changes.
4638 if (depth != ext_depth(inode)) {
4639 credits = ext4_chunk_trans_blocks(inode, len);
4640 depth = ext_depth(inode);
4643 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4645 if (IS_ERR(handle)) {
4646 ret = PTR_ERR(handle);
4649 ret = ext4_map_blocks(handle, inode, &map, flags);
4651 ext4_debug("inode #%lu: block %u: len %u: "
4652 "ext4_ext_map_blocks returned %d",
4653 inode->i_ino, map.m_lblk,
4655 ext4_mark_inode_dirty(handle, inode);
4656 ret2 = ext4_journal_stop(handle);
4660 map.m_len = len = len - ret;
4661 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4662 inode->i_ctime = current_time(inode);
4664 if (epos > new_size)
4666 if (ext4_update_inode_size(inode, epos) & 0x1)
4667 inode->i_mtime = inode->i_ctime;
4669 if (epos > inode->i_size)
4670 ext4_set_inode_flag(inode,
4671 EXT4_INODE_EOFBLOCKS);
4673 ext4_mark_inode_dirty(handle, inode);
4674 ext4_update_inode_fsync_trans(handle, inode, 1);
4675 ret2 = ext4_journal_stop(handle);
4679 if (ret == -ENOSPC &&
4680 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4685 return ret > 0 ? ret2 : ret;
4688 static long ext4_zero_range(struct file *file, loff_t offset,
4689 loff_t len, int mode)
4691 struct inode *inode = file_inode(file);
4692 handle_t *handle = NULL;
4693 unsigned int max_blocks;
4694 loff_t new_size = 0;
4698 int partial_begin, partial_end;
4701 unsigned int blkbits = inode->i_blkbits;
4703 trace_ext4_zero_range(inode, offset, len, mode);
4705 if (!S_ISREG(inode->i_mode))
4708 /* Call ext4_force_commit to flush all data in case of data=journal. */
4709 if (ext4_should_journal_data(inode)) {
4710 ret = ext4_force_commit(inode->i_sb);
4716 * Round up offset. This is not fallocate, we neet to zero out
4717 * blocks, so convert interior block aligned part of the range to
4718 * unwritten and possibly manually zero out unaligned parts of the
4721 start = round_up(offset, 1 << blkbits);
4722 end = round_down((offset + len), 1 << blkbits);
4724 if (start < offset || end > offset + len)
4726 partial_begin = offset & ((1 << blkbits) - 1);
4727 partial_end = (offset + len) & ((1 << blkbits) - 1);
4729 lblk = start >> blkbits;
4730 max_blocks = (end >> blkbits);
4731 if (max_blocks < lblk)
4739 * Indirect files do not support unwritten extnets
4741 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4746 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4747 (offset + len > i_size_read(inode) ||
4748 offset + len > EXT4_I(inode)->i_disksize)) {
4749 new_size = offset + len;
4750 ret = inode_newsize_ok(inode, new_size);
4755 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4756 if (mode & FALLOC_FL_KEEP_SIZE)
4757 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4759 /* Wait all existing dio workers, newcomers will block on i_mutex */
4760 inode_dio_wait(inode);
4762 /* Preallocate the range including the unaligned edges */
4763 if (partial_begin || partial_end) {
4764 ret = ext4_alloc_file_blocks(file,
4765 round_down(offset, 1 << blkbits) >> blkbits,
4766 (round_up((offset + len), 1 << blkbits) -
4767 round_down(offset, 1 << blkbits)) >> blkbits,
4774 /* Zero range excluding the unaligned edges */
4775 if (max_blocks > 0) {
4776 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4780 * Prevent page faults from reinstantiating pages we have
4781 * released from page cache.
4783 down_write(&EXT4_I(inode)->i_mmap_sem);
4785 ret = ext4_break_layouts(inode);
4787 up_write(&EXT4_I(inode)->i_mmap_sem);
4791 ret = ext4_update_disksize_before_punch(inode, offset, len);
4793 up_write(&EXT4_I(inode)->i_mmap_sem);
4796 /* Now release the pages and zero block aligned part of pages */
4797 truncate_pagecache_range(inode, start, end - 1);
4798 inode->i_mtime = inode->i_ctime = current_time(inode);
4800 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4802 up_write(&EXT4_I(inode)->i_mmap_sem);
4806 if (!partial_begin && !partial_end)
4810 * In worst case we have to writeout two nonadjacent unwritten
4811 * blocks and update the inode
4813 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4814 if (ext4_should_journal_data(inode))
4816 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4817 if (IS_ERR(handle)) {
4818 ret = PTR_ERR(handle);
4819 ext4_std_error(inode->i_sb, ret);
4823 inode->i_mtime = inode->i_ctime = current_time(inode);
4825 ext4_update_inode_size(inode, new_size);
4828 * Mark that we allocate beyond EOF so the subsequent truncate
4829 * can proceed even if the new size is the same as i_size.
4831 if ((offset + len) > i_size_read(inode))
4832 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4834 ext4_mark_inode_dirty(handle, inode);
4836 /* Zero out partial block at the edges of the range */
4837 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4839 ext4_update_inode_fsync_trans(handle, inode, 1);
4841 if (file->f_flags & O_SYNC)
4842 ext4_handle_sync(handle);
4844 ext4_journal_stop(handle);
4846 inode_unlock(inode);
4851 * preallocate space for a file. This implements ext4's fallocate file
4852 * operation, which gets called from sys_fallocate system call.
4853 * For block-mapped files, posix_fallocate should fall back to the method
4854 * of writing zeroes to the required new blocks (the same behavior which is
4855 * expected for file systems which do not support fallocate() system call).
4857 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4859 struct inode *inode = file_inode(file);
4860 loff_t new_size = 0;
4861 unsigned int max_blocks;
4865 unsigned int blkbits = inode->i_blkbits;
4868 * Encrypted inodes can't handle collapse range or insert
4869 * range since we would need to re-encrypt blocks with a
4870 * different IV or XTS tweak (which are based on the logical
4873 * XXX It's not clear why zero range isn't working, but we'll
4874 * leave it disabled for encrypted inodes for now. This is a
4875 * bug we should fix....
4877 if (IS_ENCRYPTED(inode) &&
4878 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4879 FALLOC_FL_ZERO_RANGE)))
4882 /* Return error if mode is not supported */
4883 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4884 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4885 FALLOC_FL_INSERT_RANGE))
4888 if (mode & FALLOC_FL_PUNCH_HOLE)
4889 return ext4_punch_hole(inode, offset, len);
4891 ret = ext4_convert_inline_data(inode);
4895 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4896 return ext4_collapse_range(inode, offset, len);
4898 if (mode & FALLOC_FL_INSERT_RANGE)
4899 return ext4_insert_range(inode, offset, len);
4901 if (mode & FALLOC_FL_ZERO_RANGE)
4902 return ext4_zero_range(file, offset, len, mode);
4904 trace_ext4_fallocate_enter(inode, offset, len, mode);
4905 lblk = offset >> blkbits;
4907 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4908 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4909 if (mode & FALLOC_FL_KEEP_SIZE)
4910 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4915 * We only support preallocation for extent-based files only
4917 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4922 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4923 (offset + len > i_size_read(inode) ||
4924 offset + len > EXT4_I(inode)->i_disksize)) {
4925 new_size = offset + len;
4926 ret = inode_newsize_ok(inode, new_size);
4931 /* Wait all existing dio workers, newcomers will block on i_mutex */
4932 inode_dio_wait(inode);
4934 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4938 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4939 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4940 EXT4_I(inode)->i_sync_tid);
4943 inode_unlock(inode);
4944 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4949 * This function convert a range of blocks to written extents
4950 * The caller of this function will pass the start offset and the size.
4951 * all unwritten extents within this range will be converted to
4954 * This function is called from the direct IO end io call back
4955 * function, to convert the fallocated extents after IO is completed.
4956 * Returns 0 on success.
4958 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4959 loff_t offset, ssize_t len)
4961 unsigned int max_blocks;
4964 struct ext4_map_blocks map;
4965 unsigned int credits, blkbits = inode->i_blkbits;
4967 map.m_lblk = offset >> blkbits;
4968 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4971 * This is somewhat ugly but the idea is clear: When transaction is
4972 * reserved, everything goes into it. Otherwise we rather start several
4973 * smaller transactions for conversion of each extent separately.
4976 handle = ext4_journal_start_reserved(handle,
4977 EXT4_HT_EXT_CONVERT);
4979 return PTR_ERR(handle);
4983 * credits to insert 1 extent into extent tree
4985 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4987 while (ret >= 0 && ret < max_blocks) {
4989 map.m_len = (max_blocks -= ret);
4991 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4993 if (IS_ERR(handle)) {
4994 ret = PTR_ERR(handle);
4998 ret = ext4_map_blocks(handle, inode, &map,
4999 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5001 ext4_warning(inode->i_sb,
5002 "inode #%lu: block %u: len %u: "
5003 "ext4_ext_map_blocks returned %d",
5004 inode->i_ino, map.m_lblk,
5006 ext4_mark_inode_dirty(handle, inode);
5008 ret2 = ext4_journal_stop(handle);
5009 if (ret <= 0 || ret2)
5013 ret2 = ext4_journal_stop(handle);
5014 return ret > 0 ? ret2 : ret;
5018 * If newes is not existing extent (newes->ec_pblk equals zero) find
5019 * delayed extent at start of newes and update newes accordingly and
5020 * return start of the next delayed extent.
5022 * If newes is existing extent (newes->ec_pblk is not equal zero)
5023 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5024 * extent found. Leave newes unmodified.
5026 static int ext4_find_delayed_extent(struct inode *inode,
5027 struct extent_status *newes)
5029 struct extent_status es;
5030 ext4_lblk_t block, next_del;
5032 if (newes->es_pblk == 0) {
5033 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5035 newes->es_lblk + newes->es_len - 1,
5039 * No extent in extent-tree contains block @newes->es_pblk,
5040 * then the block may stay in 1)a hole or 2)delayed-extent.
5046 if (es.es_lblk > newes->es_lblk) {
5048 newes->es_len = min(es.es_lblk - newes->es_lblk,
5053 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5056 block = newes->es_lblk + newes->es_len;
5057 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5058 EXT_MAX_BLOCKS, &es);
5060 next_del = EXT_MAX_BLOCKS;
5062 next_del = es.es_lblk;
5067 static int ext4_xattr_fiemap(struct inode *inode,
5068 struct fiemap_extent_info *fieinfo)
5072 __u32 flags = FIEMAP_EXTENT_LAST;
5073 int blockbits = inode->i_sb->s_blocksize_bits;
5077 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5078 struct ext4_iloc iloc;
5079 int offset; /* offset of xattr in inode */
5081 error = ext4_get_inode_loc(inode, &iloc);
5084 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5085 offset = EXT4_GOOD_OLD_INODE_SIZE +
5086 EXT4_I(inode)->i_extra_isize;
5088 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5089 flags |= FIEMAP_EXTENT_DATA_INLINE;
5091 } else { /* external block */
5092 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5093 length = inode->i_sb->s_blocksize;
5097 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5099 return (error < 0 ? error : 0);
5102 static int _ext4_fiemap(struct inode *inode,
5103 struct fiemap_extent_info *fieinfo,
5104 __u64 start, __u64 len,
5105 int (*fill)(struct inode *, ext4_lblk_t,
5107 struct fiemap_extent_info *))
5109 ext4_lblk_t start_blk;
5110 u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5114 if (ext4_has_inline_data(inode)) {
5117 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5124 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5125 error = ext4_ext_precache(inode);
5128 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5131 /* fallback to generic here if not in extents fmt */
5132 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5133 fill == ext4_fill_fiemap_extents)
5134 return generic_block_fiemap(inode, fieinfo, start, len,
5137 if (fill == ext4_fill_es_cache_info)
5138 ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5139 if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5142 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5143 error = ext4_xattr_fiemap(inode, fieinfo);
5145 ext4_lblk_t len_blks;
5148 start_blk = start >> inode->i_sb->s_blocksize_bits;
5149 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5150 if (last_blk >= EXT_MAX_BLOCKS)
5151 last_blk = EXT_MAX_BLOCKS-1;
5152 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5155 * Walk the extent tree gathering extent information
5156 * and pushing extents back to the user.
5158 error = fill(inode, start_blk, len_blks, fieinfo);
5163 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5164 __u64 start, __u64 len)
5166 return _ext4_fiemap(inode, fieinfo, start, len,
5167 ext4_fill_fiemap_extents);
5170 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5171 __u64 start, __u64 len)
5173 if (ext4_has_inline_data(inode)) {
5176 down_read(&EXT4_I(inode)->xattr_sem);
5177 has_inline = ext4_has_inline_data(inode);
5178 up_read(&EXT4_I(inode)->xattr_sem);
5183 return _ext4_fiemap(inode, fieinfo, start, len,
5184 ext4_fill_es_cache_info);
5190 * Function to access the path buffer for marking it dirty.
5191 * It also checks if there are sufficient credits left in the journal handle
5195 ext4_access_path(handle_t *handle, struct inode *inode,
5196 struct ext4_ext_path *path)
5200 if (!ext4_handle_valid(handle))
5204 * Check if need to extend journal credits
5205 * 3 for leaf, sb, and inode plus 2 (bmap and group
5206 * descriptor) for each block group; assume two block
5209 if (handle->h_buffer_credits < 7) {
5210 credits = ext4_writepage_trans_blocks(inode);
5211 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5212 /* EAGAIN is success */
5213 if (err && err != -EAGAIN)
5217 err = ext4_ext_get_access(handle, inode, path);
5222 * ext4_ext_shift_path_extents:
5223 * Shift the extents of a path structure lying between path[depth].p_ext
5224 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5225 * if it is right shift or left shift operation.
5228 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5229 struct inode *inode, handle_t *handle,
5230 enum SHIFT_DIRECTION SHIFT)
5233 struct ext4_extent *ex_start, *ex_last;
5235 depth = path->p_depth;
5237 while (depth >= 0) {
5238 if (depth == path->p_depth) {
5239 ex_start = path[depth].p_ext;
5241 return -EFSCORRUPTED;
5243 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5245 err = ext4_access_path(handle, inode, path + depth);
5249 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5252 while (ex_start <= ex_last) {
5253 if (SHIFT == SHIFT_LEFT) {
5254 le32_add_cpu(&ex_start->ee_block,
5256 /* Try to merge to the left. */
5258 EXT_FIRST_EXTENT(path[depth].p_hdr))
5260 ext4_ext_try_to_merge_right(inode,
5261 path, ex_start - 1))
5266 le32_add_cpu(&ex_last->ee_block, shift);
5267 ext4_ext_try_to_merge_right(inode, path,
5272 err = ext4_ext_dirty(handle, inode, path + depth);
5276 if (--depth < 0 || !update)
5280 /* Update index too */
5281 err = ext4_access_path(handle, inode, path + depth);
5285 if (SHIFT == SHIFT_LEFT)
5286 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5288 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5289 err = ext4_ext_dirty(handle, inode, path + depth);
5293 /* we are done if current index is not a starting index */
5294 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5305 * ext4_ext_shift_extents:
5306 * All the extents which lies in the range from @start to the last allocated
5307 * block for the @inode are shifted either towards left or right (depending
5308 * upon @SHIFT) by @shift blocks.
5309 * On success, 0 is returned, error otherwise.
5312 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5313 ext4_lblk_t start, ext4_lblk_t shift,
5314 enum SHIFT_DIRECTION SHIFT)
5316 struct ext4_ext_path *path;
5318 struct ext4_extent *extent;
5319 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5321 /* Let path point to the last extent */
5322 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5325 return PTR_ERR(path);
5327 depth = path->p_depth;
5328 extent = path[depth].p_ext;
5332 stop = le32_to_cpu(extent->ee_block);
5335 * For left shifts, make sure the hole on the left is big enough to
5336 * accommodate the shift. For right shifts, make sure the last extent
5337 * won't be shifted beyond EXT_MAX_BLOCKS.
5339 if (SHIFT == SHIFT_LEFT) {
5340 path = ext4_find_extent(inode, start - 1, &path,
5343 return PTR_ERR(path);
5344 depth = path->p_depth;
5345 extent = path[depth].p_ext;
5347 ex_start = le32_to_cpu(extent->ee_block);
5348 ex_end = le32_to_cpu(extent->ee_block) +
5349 ext4_ext_get_actual_len(extent);
5355 if ((start == ex_start && shift > ex_start) ||
5356 (shift > start - ex_end)) {
5361 if (shift > EXT_MAX_BLOCKS -
5362 (stop + ext4_ext_get_actual_len(extent))) {
5369 * In case of left shift, iterator points to start and it is increased
5370 * till we reach stop. In case of right shift, iterator points to stop
5371 * and it is decreased till we reach start.
5373 if (SHIFT == SHIFT_LEFT)
5379 * Its safe to start updating extents. Start and stop are unsigned, so
5380 * in case of right shift if extent with 0 block is reached, iterator
5381 * becomes NULL to indicate the end of the loop.
5383 while (iterator && start <= stop) {
5384 path = ext4_find_extent(inode, *iterator, &path,
5387 return PTR_ERR(path);
5388 depth = path->p_depth;
5389 extent = path[depth].p_ext;
5391 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5392 (unsigned long) *iterator);
5393 return -EFSCORRUPTED;
5395 if (SHIFT == SHIFT_LEFT && *iterator >
5396 le32_to_cpu(extent->ee_block)) {
5397 /* Hole, move to the next extent */
5398 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5399 path[depth].p_ext++;
5401 *iterator = ext4_ext_next_allocated_block(path);
5406 if (SHIFT == SHIFT_LEFT) {
5407 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5408 *iterator = le32_to_cpu(extent->ee_block) +
5409 ext4_ext_get_actual_len(extent);
5411 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5412 if (le32_to_cpu(extent->ee_block) > 0)
5413 *iterator = le32_to_cpu(extent->ee_block) - 1;
5415 /* Beginning is reached, end of the loop */
5417 /* Update path extent in case we need to stop */
5418 while (le32_to_cpu(extent->ee_block) < start)
5420 path[depth].p_ext = extent;
5422 ret = ext4_ext_shift_path_extents(path, shift, inode,
5428 ext4_ext_drop_refs(path);
5434 * ext4_collapse_range:
5435 * This implements the fallocate's collapse range functionality for ext4
5436 * Returns: 0 and non-zero on error.
5438 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5440 struct super_block *sb = inode->i_sb;
5441 ext4_lblk_t punch_start, punch_stop;
5443 unsigned int credits;
5444 loff_t new_size, ioffset;
5448 * We need to test this early because xfstests assumes that a
5449 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5450 * system does not support collapse range.
5452 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5455 /* Collapse range works only on fs block size aligned offsets. */
5456 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5457 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5460 if (!S_ISREG(inode->i_mode))
5463 trace_ext4_collapse_range(inode, offset, len);
5465 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5466 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5468 /* Call ext4_force_commit to flush all data in case of data=journal. */
5469 if (ext4_should_journal_data(inode)) {
5470 ret = ext4_force_commit(inode->i_sb);
5477 * There is no need to overlap collapse range with EOF, in which case
5478 * it is effectively a truncate operation
5480 if (offset + len >= i_size_read(inode)) {
5485 /* Currently just for extent based files */
5486 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5491 /* Wait for existing dio to complete */
5492 inode_dio_wait(inode);
5495 * Prevent page faults from reinstantiating pages we have released from
5498 down_write(&EXT4_I(inode)->i_mmap_sem);
5500 ret = ext4_break_layouts(inode);
5505 * Need to round down offset to be aligned with page size boundary
5506 * for page size > block size.
5508 ioffset = round_down(offset, PAGE_SIZE);
5510 * Write tail of the last page before removed range since it will get
5511 * removed from the page cache below.
5513 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5517 * Write data that will be shifted to preserve them when discarding
5518 * page cache below. We are also protected from pages becoming dirty
5521 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5525 truncate_pagecache(inode, ioffset);
5527 credits = ext4_writepage_trans_blocks(inode);
5528 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5529 if (IS_ERR(handle)) {
5530 ret = PTR_ERR(handle);
5534 down_write(&EXT4_I(inode)->i_data_sem);
5535 ext4_discard_preallocations(inode);
5537 ret = ext4_es_remove_extent(inode, punch_start,
5538 EXT_MAX_BLOCKS - punch_start);
5540 up_write(&EXT4_I(inode)->i_data_sem);
5544 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5546 up_write(&EXT4_I(inode)->i_data_sem);
5549 ext4_discard_preallocations(inode);
5551 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5552 punch_stop - punch_start, SHIFT_LEFT);
5554 up_write(&EXT4_I(inode)->i_data_sem);
5558 new_size = i_size_read(inode) - len;
5559 i_size_write(inode, new_size);
5560 EXT4_I(inode)->i_disksize = new_size;
5562 up_write(&EXT4_I(inode)->i_data_sem);
5564 ext4_handle_sync(handle);
5565 inode->i_mtime = inode->i_ctime = current_time(inode);
5566 ext4_mark_inode_dirty(handle, inode);
5567 ext4_update_inode_fsync_trans(handle, inode, 1);
5570 ext4_journal_stop(handle);
5572 up_write(&EXT4_I(inode)->i_mmap_sem);
5574 inode_unlock(inode);
5579 * ext4_insert_range:
5580 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5581 * The data blocks starting from @offset to the EOF are shifted by @len
5582 * towards right to create a hole in the @inode. Inode size is increased
5584 * Returns 0 on success, error otherwise.
5586 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5588 struct super_block *sb = inode->i_sb;
5590 struct ext4_ext_path *path;
5591 struct ext4_extent *extent;
5592 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5593 unsigned int credits, ee_len;
5594 int ret = 0, depth, split_flag = 0;
5598 * We need to test this early because xfstests assumes that an
5599 * insert range of (0, 1) will return EOPNOTSUPP if the file
5600 * system does not support insert range.
5602 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5605 /* Insert range works only on fs block size aligned offsets. */
5606 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5607 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5610 if (!S_ISREG(inode->i_mode))
5613 trace_ext4_insert_range(inode, offset, len);
5615 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5616 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5618 /* Call ext4_force_commit to flush all data in case of data=journal */
5619 if (ext4_should_journal_data(inode)) {
5620 ret = ext4_force_commit(inode->i_sb);
5626 /* Currently just for extent based files */
5627 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5632 /* Check for wrap through zero */
5633 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5638 /* Offset should be less than i_size */
5639 if (offset >= i_size_read(inode)) {
5644 /* Wait for existing dio to complete */
5645 inode_dio_wait(inode);
5648 * Prevent page faults from reinstantiating pages we have released from
5651 down_write(&EXT4_I(inode)->i_mmap_sem);
5653 ret = ext4_break_layouts(inode);
5658 * Need to round down to align start offset to page size boundary
5659 * for page size > block size.
5661 ioffset = round_down(offset, PAGE_SIZE);
5662 /* Write out all dirty pages */
5663 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5667 truncate_pagecache(inode, ioffset);
5669 credits = ext4_writepage_trans_blocks(inode);
5670 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5671 if (IS_ERR(handle)) {
5672 ret = PTR_ERR(handle);
5676 /* Expand file to avoid data loss if there is error while shifting */
5677 inode->i_size += len;
5678 EXT4_I(inode)->i_disksize += len;
5679 inode->i_mtime = inode->i_ctime = current_time(inode);
5680 ret = ext4_mark_inode_dirty(handle, inode);
5684 down_write(&EXT4_I(inode)->i_data_sem);
5685 ext4_discard_preallocations(inode);
5687 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5689 up_write(&EXT4_I(inode)->i_data_sem);
5693 depth = ext_depth(inode);
5694 extent = path[depth].p_ext;
5696 ee_start_lblk = le32_to_cpu(extent->ee_block);
5697 ee_len = ext4_ext_get_actual_len(extent);
5700 * If offset_lblk is not the starting block of extent, split
5701 * the extent @offset_lblk
5703 if ((offset_lblk > ee_start_lblk) &&
5704 (offset_lblk < (ee_start_lblk + ee_len))) {
5705 if (ext4_ext_is_unwritten(extent))
5706 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5707 EXT4_EXT_MARK_UNWRIT2;
5708 ret = ext4_split_extent_at(handle, inode, &path,
5709 offset_lblk, split_flag,
5711 EXT4_GET_BLOCKS_PRE_IO |
5712 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5715 ext4_ext_drop_refs(path);
5718 up_write(&EXT4_I(inode)->i_data_sem);
5722 ext4_ext_drop_refs(path);
5726 ret = ext4_es_remove_extent(inode, offset_lblk,
5727 EXT_MAX_BLOCKS - offset_lblk);
5729 up_write(&EXT4_I(inode)->i_data_sem);
5734 * if offset_lblk lies in a hole which is at start of file, use
5735 * ee_start_lblk to shift extents
5737 ret = ext4_ext_shift_extents(inode, handle,
5738 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5739 len_lblk, SHIFT_RIGHT);
5741 up_write(&EXT4_I(inode)->i_data_sem);
5743 ext4_handle_sync(handle);
5745 ext4_update_inode_fsync_trans(handle, inode, 1);
5748 ext4_journal_stop(handle);
5750 up_write(&EXT4_I(inode)->i_mmap_sem);
5752 inode_unlock(inode);
5757 * ext4_swap_extents() - Swap extents between two inodes
5758 * @handle: handle for this transaction
5759 * @inode1: First inode
5760 * @inode2: Second inode
5761 * @lblk1: Start block for first inode
5762 * @lblk2: Start block for second inode
5763 * @count: Number of blocks to swap
5764 * @unwritten: Mark second inode's extents as unwritten after swap
5765 * @erp: Pointer to save error value
5767 * This helper routine does exactly what is promise "swap extents". All other
5768 * stuff such as page-cache locking consistency, bh mapping consistency or
5769 * extent's data copying must be performed by caller.
5771 * i_mutex is held for both inodes
5772 * i_data_sem is locked for write for both inodes
5774 * All pages from requested range are locked for both inodes
5777 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5778 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5779 ext4_lblk_t count, int unwritten, int *erp)
5781 struct ext4_ext_path *path1 = NULL;
5782 struct ext4_ext_path *path2 = NULL;
5783 int replaced_count = 0;
5785 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5786 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5787 BUG_ON(!inode_is_locked(inode1));
5788 BUG_ON(!inode_is_locked(inode2));
5790 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5793 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5798 struct ext4_extent *ex1, *ex2, tmp_ex;
5799 ext4_lblk_t e1_blk, e2_blk;
5800 int e1_len, e2_len, len;
5803 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5804 if (IS_ERR(path1)) {
5805 *erp = PTR_ERR(path1);
5811 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5812 if (IS_ERR(path2)) {
5813 *erp = PTR_ERR(path2);
5817 ex1 = path1[path1->p_depth].p_ext;
5818 ex2 = path2[path2->p_depth].p_ext;
5819 /* Do we have somthing to swap ? */
5820 if (unlikely(!ex2 || !ex1))
5823 e1_blk = le32_to_cpu(ex1->ee_block);
5824 e2_blk = le32_to_cpu(ex2->ee_block);
5825 e1_len = ext4_ext_get_actual_len(ex1);
5826 e2_len = ext4_ext_get_actual_len(ex2);
5829 if (!in_range(lblk1, e1_blk, e1_len) ||
5830 !in_range(lblk2, e2_blk, e2_len)) {
5831 ext4_lblk_t next1, next2;
5833 /* if hole after extent, then go to next extent */
5834 next1 = ext4_ext_next_allocated_block(path1);
5835 next2 = ext4_ext_next_allocated_block(path2);
5836 /* If hole before extent, then shift to that extent */
5841 /* Do we have something to swap */
5842 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5844 /* Move to the rightest boundary */
5845 len = next1 - lblk1;
5846 if (len < next2 - lblk2)
5847 len = next2 - lblk2;
5856 /* Prepare left boundary */
5857 if (e1_blk < lblk1) {
5859 *erp = ext4_force_split_extent_at(handle, inode1,
5864 if (e2_blk < lblk2) {
5866 *erp = ext4_force_split_extent_at(handle, inode2,
5871 /* ext4_split_extent_at() may result in leaf extent split,
5872 * path must to be revalidated. */
5876 /* Prepare right boundary */
5878 if (len > e1_blk + e1_len - lblk1)
5879 len = e1_blk + e1_len - lblk1;
5880 if (len > e2_blk + e2_len - lblk2)
5881 len = e2_blk + e2_len - lblk2;
5883 if (len != e1_len) {
5885 *erp = ext4_force_split_extent_at(handle, inode1,
5886 &path1, lblk1 + len, 0);
5890 if (len != e2_len) {
5892 *erp = ext4_force_split_extent_at(handle, inode2,
5893 &path2, lblk2 + len, 0);
5897 /* ext4_split_extent_at() may result in leaf extent split,
5898 * path must to be revalidated. */
5902 BUG_ON(e2_len != e1_len);
5903 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5906 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5910 /* Both extents are fully inside boundaries. Swap it now */
5912 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5913 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5914 ex1->ee_len = cpu_to_le16(e2_len);
5915 ex2->ee_len = cpu_to_le16(e1_len);
5917 ext4_ext_mark_unwritten(ex2);
5918 if (ext4_ext_is_unwritten(&tmp_ex))
5919 ext4_ext_mark_unwritten(ex1);
5921 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5922 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5923 *erp = ext4_ext_dirty(handle, inode2, path2 +
5927 *erp = ext4_ext_dirty(handle, inode1, path1 +
5930 * Looks scarry ah..? second inode already points to new blocks,
5931 * and it was successfully dirtied. But luckily error may happen
5932 * only due to journal error, so full transaction will be
5939 replaced_count += len;
5943 ext4_ext_drop_refs(path1);
5945 ext4_ext_drop_refs(path2);
5947 path1 = path2 = NULL;
5949 return replaced_count;
5953 * ext4_clu_mapped - determine whether any block in a logical cluster has
5954 * been mapped to a physical cluster
5956 * @inode - file containing the logical cluster
5957 * @lclu - logical cluster of interest
5959 * Returns 1 if any block in the logical cluster is mapped, signifying
5960 * that a physical cluster has been allocated for it. Otherwise,
5961 * returns 0. Can also return negative error codes. Derived from
5962 * ext4_ext_map_blocks().
5964 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5966 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5967 struct ext4_ext_path *path;
5968 int depth, mapped = 0, err = 0;
5969 struct ext4_extent *extent;
5970 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5972 /* search for the extent closest to the first block in the cluster */
5973 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5975 err = PTR_ERR(path);
5980 depth = ext_depth(inode);
5983 * A consistent leaf must not be empty. This situation is possible,
5984 * though, _during_ tree modification, and it's why an assert can't
5985 * be put in ext4_find_extent().
5987 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5988 EXT4_ERROR_INODE(inode,
5989 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5990 (unsigned long) EXT4_C2B(sbi, lclu),
5991 depth, path[depth].p_block);
5992 err = -EFSCORRUPTED;
5996 extent = path[depth].p_ext;
5998 /* can't be mapped if the extent tree is empty */
6002 first_lblk = le32_to_cpu(extent->ee_block);
6003 first_lclu = EXT4_B2C(sbi, first_lblk);
6006 * Three possible outcomes at this point - found extent spanning
6007 * the target cluster, to the left of the target cluster, or to the
6008 * right of the target cluster. The first two cases are handled here.
6009 * The last case indicates the target cluster is not mapped.
6011 if (lclu >= first_lclu) {
6012 last_lclu = EXT4_B2C(sbi, first_lblk +
6013 ext4_ext_get_actual_len(extent) - 1);
6014 if (lclu <= last_lclu) {
6017 first_lblk = ext4_ext_next_allocated_block(path);
6018 first_lclu = EXT4_B2C(sbi, first_lblk);
6019 if (lclu == first_lclu)
6025 ext4_ext_drop_refs(path);
6028 return err ? err : mapped;