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_inode_block_valid(inode, 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_inode_block_valid(inode, block, 1);
391 static int ext4_valid_extent_entries(struct inode *inode,
392 struct ext4_extent_header *eh,
393 ext4_fsblk_t *pblk, int depth)
395 unsigned short entries;
396 ext4_lblk_t lblock = 0;
397 ext4_lblk_t prev = 0;
399 if (eh->eh_entries == 0)
402 entries = le16_to_cpu(eh->eh_entries);
406 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
407 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
408 ext4_fsblk_t pblock = 0;
410 if (!ext4_valid_extent(inode, ext))
413 /* Check for overlapping extents */
414 lblock = le32_to_cpu(ext->ee_block);
415 if ((lblock <= prev) && prev) {
416 pblock = ext4_ext_pblock(ext);
417 es->s_last_error_block = cpu_to_le64(pblock);
420 prev = lblock + ext4_ext_get_actual_len(ext) - 1;
425 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
427 if (!ext4_valid_extent_idx(inode, ext_idx))
430 /* Check for overlapping index extents */
431 lblock = le32_to_cpu(ext_idx->ei_block);
432 if ((lblock <= prev) && prev) {
433 *pblk = ext4_idx_pblock(ext_idx);
444 static int __ext4_ext_check(const char *function, unsigned int line,
445 struct inode *inode, struct ext4_extent_header *eh,
446 int depth, ext4_fsblk_t pblk)
448 const char *error_msg;
449 int max = 0, err = -EFSCORRUPTED;
451 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
452 error_msg = "invalid magic";
455 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
456 error_msg = "unexpected eh_depth";
459 if (unlikely(eh->eh_max == 0)) {
460 error_msg = "invalid eh_max";
463 max = ext4_ext_max_entries(inode, depth);
464 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
465 error_msg = "too large eh_max";
468 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
469 error_msg = "invalid eh_entries";
472 if (!ext4_valid_extent_entries(inode, eh, &pblk, depth)) {
473 error_msg = "invalid extent entries";
476 if (unlikely(depth > 32)) {
477 error_msg = "too large eh_depth";
480 /* Verify checksum on non-root extent tree nodes */
481 if (ext_depth(inode) != depth &&
482 !ext4_extent_block_csum_verify(inode, eh)) {
483 error_msg = "extent tree corrupted";
490 ext4_error_inode(inode, function, line, 0,
491 "pblk %llu bad header/extent: %s - magic %x, "
492 "entries %u, max %u(%u), depth %u(%u)",
493 (unsigned long long) pblk, error_msg,
494 le16_to_cpu(eh->eh_magic),
495 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
496 max, le16_to_cpu(eh->eh_depth), depth);
500 #define ext4_ext_check(inode, eh, depth, pblk) \
501 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
503 int ext4_ext_check_inode(struct inode *inode)
505 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
508 static void ext4_cache_extents(struct inode *inode,
509 struct ext4_extent_header *eh)
511 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
512 ext4_lblk_t prev = 0;
515 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
516 unsigned int status = EXTENT_STATUS_WRITTEN;
517 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
518 int len = ext4_ext_get_actual_len(ex);
520 if (prev && (prev != lblk))
521 ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
524 if (ext4_ext_is_unwritten(ex))
525 status = EXTENT_STATUS_UNWRITTEN;
526 ext4_es_cache_extent(inode, lblk, len,
527 ext4_ext_pblock(ex), status);
532 static struct buffer_head *
533 __read_extent_tree_block(const char *function, unsigned int line,
534 struct inode *inode, ext4_fsblk_t pblk, int depth,
537 struct buffer_head *bh;
540 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
542 return ERR_PTR(-ENOMEM);
544 if (!bh_uptodate_or_lock(bh)) {
545 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
546 err = bh_submit_read(bh);
550 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
552 err = __ext4_ext_check(function, line, inode,
553 ext_block_hdr(bh), depth, pblk);
556 set_buffer_verified(bh);
558 * If this is a leaf block, cache all of its entries
560 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
561 struct ext4_extent_header *eh = ext_block_hdr(bh);
562 ext4_cache_extents(inode, eh);
571 #define read_extent_tree_block(inode, pblk, depth, flags) \
572 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
576 * This function is called to cache a file's extent information in the
579 int ext4_ext_precache(struct inode *inode)
581 struct ext4_inode_info *ei = EXT4_I(inode);
582 struct ext4_ext_path *path = NULL;
583 struct buffer_head *bh;
584 int i = 0, depth, ret = 0;
586 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
587 return 0; /* not an extent-mapped inode */
589 down_read(&ei->i_data_sem);
590 depth = ext_depth(inode);
592 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
595 up_read(&ei->i_data_sem);
599 /* Don't cache anything if there are no external extent blocks */
602 path[0].p_hdr = ext_inode_hdr(inode);
603 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
606 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
609 * If this is a leaf block or we've reached the end of
610 * the index block, go up
613 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
614 brelse(path[i].p_bh);
619 bh = read_extent_tree_block(inode,
620 ext4_idx_pblock(path[i].p_idx++),
622 EXT4_EX_FORCE_CACHE);
629 path[i].p_hdr = ext_block_hdr(bh);
630 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
632 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
634 up_read(&ei->i_data_sem);
635 ext4_ext_drop_refs(path);
641 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
643 int k, l = path->p_depth;
646 for (k = 0; k <= l; k++, path++) {
648 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
649 ext4_idx_pblock(path->p_idx));
650 } else if (path->p_ext) {
651 ext_debug(" %d:[%d]%d:%llu ",
652 le32_to_cpu(path->p_ext->ee_block),
653 ext4_ext_is_unwritten(path->p_ext),
654 ext4_ext_get_actual_len(path->p_ext),
655 ext4_ext_pblock(path->p_ext));
662 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
664 int depth = ext_depth(inode);
665 struct ext4_extent_header *eh;
666 struct ext4_extent *ex;
672 eh = path[depth].p_hdr;
673 ex = EXT_FIRST_EXTENT(eh);
675 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
677 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
678 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
679 ext4_ext_is_unwritten(ex),
680 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
685 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
686 ext4_fsblk_t newblock, int level)
688 int depth = ext_depth(inode);
689 struct ext4_extent *ex;
691 if (depth != level) {
692 struct ext4_extent_idx *idx;
693 idx = path[level].p_idx;
694 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
695 ext_debug("%d: move %d:%llu in new index %llu\n", level,
696 le32_to_cpu(idx->ei_block),
697 ext4_idx_pblock(idx),
705 ex = path[depth].p_ext;
706 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
707 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
708 le32_to_cpu(ex->ee_block),
710 ext4_ext_is_unwritten(ex),
711 ext4_ext_get_actual_len(ex),
718 #define ext4_ext_show_path(inode, path)
719 #define ext4_ext_show_leaf(inode, path)
720 #define ext4_ext_show_move(inode, path, newblock, level)
723 void ext4_ext_drop_refs(struct ext4_ext_path *path)
729 depth = path->p_depth;
730 for (i = 0; i <= depth; i++, path++)
738 * ext4_ext_binsearch_idx:
739 * binary search for the closest index of the given block
740 * the header must be checked before calling this
743 ext4_ext_binsearch_idx(struct inode *inode,
744 struct ext4_ext_path *path, ext4_lblk_t block)
746 struct ext4_extent_header *eh = path->p_hdr;
747 struct ext4_extent_idx *r, *l, *m;
750 ext_debug("binsearch for %u(idx): ", block);
752 l = EXT_FIRST_INDEX(eh) + 1;
753 r = EXT_LAST_INDEX(eh);
756 if (block < le32_to_cpu(m->ei_block))
760 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
761 m, le32_to_cpu(m->ei_block),
762 r, le32_to_cpu(r->ei_block));
766 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
767 ext4_idx_pblock(path->p_idx));
769 #ifdef CHECK_BINSEARCH
771 struct ext4_extent_idx *chix, *ix;
774 chix = ix = EXT_FIRST_INDEX(eh);
775 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
777 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
778 printk(KERN_DEBUG "k=%d, ix=0x%p, "
780 ix, EXT_FIRST_INDEX(eh));
781 printk(KERN_DEBUG "%u <= %u\n",
782 le32_to_cpu(ix->ei_block),
783 le32_to_cpu(ix[-1].ei_block));
785 BUG_ON(k && le32_to_cpu(ix->ei_block)
786 <= le32_to_cpu(ix[-1].ei_block));
787 if (block < le32_to_cpu(ix->ei_block))
791 BUG_ON(chix != path->p_idx);
798 * ext4_ext_binsearch:
799 * binary search for closest extent of the given block
800 * the header must be checked before calling this
803 ext4_ext_binsearch(struct inode *inode,
804 struct ext4_ext_path *path, ext4_lblk_t block)
806 struct ext4_extent_header *eh = path->p_hdr;
807 struct ext4_extent *r, *l, *m;
809 if (eh->eh_entries == 0) {
811 * this leaf is empty:
812 * we get such a leaf in split/add case
817 ext_debug("binsearch for %u: ", block);
819 l = EXT_FIRST_EXTENT(eh) + 1;
820 r = EXT_LAST_EXTENT(eh);
824 if (block < le32_to_cpu(m->ee_block))
828 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
829 m, le32_to_cpu(m->ee_block),
830 r, le32_to_cpu(r->ee_block));
834 ext_debug(" -> %d:%llu:[%d]%d ",
835 le32_to_cpu(path->p_ext->ee_block),
836 ext4_ext_pblock(path->p_ext),
837 ext4_ext_is_unwritten(path->p_ext),
838 ext4_ext_get_actual_len(path->p_ext));
840 #ifdef CHECK_BINSEARCH
842 struct ext4_extent *chex, *ex;
845 chex = ex = EXT_FIRST_EXTENT(eh);
846 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
847 BUG_ON(k && le32_to_cpu(ex->ee_block)
848 <= le32_to_cpu(ex[-1].ee_block));
849 if (block < le32_to_cpu(ex->ee_block))
853 BUG_ON(chex != path->p_ext);
859 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
861 struct ext4_extent_header *eh;
863 eh = ext_inode_hdr(inode);
866 eh->eh_magic = EXT4_EXT_MAGIC;
867 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
868 eh->eh_generation = 0;
869 ext4_mark_inode_dirty(handle, inode);
873 struct ext4_ext_path *
874 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
875 struct ext4_ext_path **orig_path, int flags)
877 struct ext4_extent_header *eh;
878 struct buffer_head *bh;
879 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
880 short int depth, i, ppos = 0;
883 eh = ext_inode_hdr(inode);
884 depth = ext_depth(inode);
885 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
886 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
893 ext4_ext_drop_refs(path);
894 if (depth > path[0].p_maxdepth) {
896 *orig_path = path = NULL;
900 /* account possible depth increase */
901 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
904 return ERR_PTR(-ENOMEM);
905 path[0].p_maxdepth = depth + 1;
911 if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
912 ext4_cache_extents(inode, eh);
913 /* walk through the tree */
915 ext_debug("depth %d: num %d, max %d\n",
916 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
918 ext4_ext_binsearch_idx(inode, path + ppos, block);
919 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
920 path[ppos].p_depth = i;
921 path[ppos].p_ext = NULL;
923 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
930 eh = ext_block_hdr(bh);
932 path[ppos].p_bh = bh;
933 path[ppos].p_hdr = eh;
936 path[ppos].p_depth = i;
937 path[ppos].p_ext = NULL;
938 path[ppos].p_idx = NULL;
941 ext4_ext_binsearch(inode, path + ppos, block);
942 /* if not an empty leaf */
943 if (path[ppos].p_ext)
944 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
946 ext4_ext_show_path(inode, path);
951 ext4_ext_drop_refs(path);
959 * ext4_ext_insert_index:
960 * insert new index [@logical;@ptr] into the block at @curp;
961 * check where to insert: before @curp or after @curp
963 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
964 struct ext4_ext_path *curp,
965 int logical, ext4_fsblk_t ptr)
967 struct ext4_extent_idx *ix;
970 err = ext4_ext_get_access(handle, inode, curp);
974 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
975 EXT4_ERROR_INODE(inode,
976 "logical %d == ei_block %d!",
977 logical, le32_to_cpu(curp->p_idx->ei_block));
978 return -EFSCORRUPTED;
981 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
982 >= le16_to_cpu(curp->p_hdr->eh_max))) {
983 EXT4_ERROR_INODE(inode,
984 "eh_entries %d >= eh_max %d!",
985 le16_to_cpu(curp->p_hdr->eh_entries),
986 le16_to_cpu(curp->p_hdr->eh_max));
987 return -EFSCORRUPTED;
990 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
992 ext_debug("insert new index %d after: %llu\n", logical, ptr);
993 ix = curp->p_idx + 1;
996 ext_debug("insert new index %d before: %llu\n", logical, ptr);
1000 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1003 ext_debug("insert new index %d: "
1004 "move %d indices from 0x%p to 0x%p\n",
1005 logical, len, ix, ix + 1);
1006 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1009 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1010 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1011 return -EFSCORRUPTED;
1014 ix->ei_block = cpu_to_le32(logical);
1015 ext4_idx_store_pblock(ix, ptr);
1016 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1018 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1019 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1020 return -EFSCORRUPTED;
1023 err = ext4_ext_dirty(handle, inode, curp);
1024 ext4_std_error(inode->i_sb, err);
1031 * inserts new subtree into the path, using free index entry
1033 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1034 * - makes decision where to split
1035 * - moves remaining extents and index entries (right to the split point)
1036 * into the newly allocated blocks
1037 * - initializes subtree
1039 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1041 struct ext4_ext_path *path,
1042 struct ext4_extent *newext, int at)
1044 struct buffer_head *bh = NULL;
1045 int depth = ext_depth(inode);
1046 struct ext4_extent_header *neh;
1047 struct ext4_extent_idx *fidx;
1048 int i = at, k, m, a;
1049 ext4_fsblk_t newblock, oldblock;
1051 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1053 size_t ext_size = 0;
1055 /* make decision: where to split? */
1056 /* FIXME: now decision is simplest: at current extent */
1058 /* if current leaf will be split, then we should use
1059 * border from split point */
1060 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1061 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1062 return -EFSCORRUPTED;
1064 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1065 border = path[depth].p_ext[1].ee_block;
1066 ext_debug("leaf will be split."
1067 " next leaf starts at %d\n",
1068 le32_to_cpu(border));
1070 border = newext->ee_block;
1071 ext_debug("leaf will be added."
1072 " next leaf starts at %d\n",
1073 le32_to_cpu(border));
1077 * If error occurs, then we break processing
1078 * and mark filesystem read-only. index won't
1079 * be inserted and tree will be in consistent
1080 * state. Next mount will repair buffers too.
1084 * Get array to track all allocated blocks.
1085 * We need this to handle errors and free blocks
1088 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1092 /* allocate all needed blocks */
1093 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1094 for (a = 0; a < depth - at; a++) {
1095 newblock = ext4_ext_new_meta_block(handle, inode, path,
1096 newext, &err, flags);
1099 ablocks[a] = newblock;
1102 /* initialize new leaf */
1103 newblock = ablocks[--a];
1104 if (unlikely(newblock == 0)) {
1105 EXT4_ERROR_INODE(inode, "newblock == 0!");
1106 err = -EFSCORRUPTED;
1109 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1110 if (unlikely(!bh)) {
1116 err = ext4_journal_get_create_access(handle, bh);
1120 neh = ext_block_hdr(bh);
1121 neh->eh_entries = 0;
1122 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1123 neh->eh_magic = EXT4_EXT_MAGIC;
1125 neh->eh_generation = 0;
1127 /* move remainder of path[depth] to the new leaf */
1128 if (unlikely(path[depth].p_hdr->eh_entries !=
1129 path[depth].p_hdr->eh_max)) {
1130 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1131 path[depth].p_hdr->eh_entries,
1132 path[depth].p_hdr->eh_max);
1133 err = -EFSCORRUPTED;
1136 /* start copy from next extent */
1137 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1138 ext4_ext_show_move(inode, path, newblock, depth);
1140 struct ext4_extent *ex;
1141 ex = EXT_FIRST_EXTENT(neh);
1142 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1143 le16_add_cpu(&neh->eh_entries, m);
1146 /* zero out unused area in the extent block */
1147 ext_size = sizeof(struct ext4_extent_header) +
1148 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1149 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1150 ext4_extent_block_csum_set(inode, neh);
1151 set_buffer_uptodate(bh);
1154 err = ext4_handle_dirty_metadata(handle, inode, bh);
1160 /* correct old leaf */
1162 err = ext4_ext_get_access(handle, inode, path + depth);
1165 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1166 err = ext4_ext_dirty(handle, inode, path + depth);
1172 /* create intermediate indexes */
1174 if (unlikely(k < 0)) {
1175 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1176 err = -EFSCORRUPTED;
1180 ext_debug("create %d intermediate indices\n", k);
1181 /* insert new index into current index block */
1182 /* current depth stored in i var */
1185 oldblock = newblock;
1186 newblock = ablocks[--a];
1187 bh = sb_getblk(inode->i_sb, newblock);
1188 if (unlikely(!bh)) {
1194 err = ext4_journal_get_create_access(handle, bh);
1198 neh = ext_block_hdr(bh);
1199 neh->eh_entries = cpu_to_le16(1);
1200 neh->eh_magic = EXT4_EXT_MAGIC;
1201 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1202 neh->eh_depth = cpu_to_le16(depth - i);
1203 neh->eh_generation = 0;
1204 fidx = EXT_FIRST_INDEX(neh);
1205 fidx->ei_block = border;
1206 ext4_idx_store_pblock(fidx, oldblock);
1208 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1209 i, newblock, le32_to_cpu(border), oldblock);
1211 /* move remainder of path[i] to the new index block */
1212 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1213 EXT_LAST_INDEX(path[i].p_hdr))) {
1214 EXT4_ERROR_INODE(inode,
1215 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1216 le32_to_cpu(path[i].p_ext->ee_block));
1217 err = -EFSCORRUPTED;
1220 /* start copy indexes */
1221 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1222 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1223 EXT_MAX_INDEX(path[i].p_hdr));
1224 ext4_ext_show_move(inode, path, newblock, i);
1226 memmove(++fidx, path[i].p_idx,
1227 sizeof(struct ext4_extent_idx) * m);
1228 le16_add_cpu(&neh->eh_entries, m);
1230 /* zero out unused area in the extent block */
1231 ext_size = sizeof(struct ext4_extent_header) +
1232 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1233 memset(bh->b_data + ext_size, 0,
1234 inode->i_sb->s_blocksize - ext_size);
1235 ext4_extent_block_csum_set(inode, neh);
1236 set_buffer_uptodate(bh);
1239 err = ext4_handle_dirty_metadata(handle, inode, bh);
1245 /* correct old index */
1247 err = ext4_ext_get_access(handle, inode, path + i);
1250 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1251 err = ext4_ext_dirty(handle, inode, path + i);
1259 /* insert new index */
1260 err = ext4_ext_insert_index(handle, inode, path + at,
1261 le32_to_cpu(border), newblock);
1265 if (buffer_locked(bh))
1271 /* free all allocated blocks in error case */
1272 for (i = 0; i < depth; i++) {
1275 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1276 EXT4_FREE_BLOCKS_METADATA);
1285 * ext4_ext_grow_indepth:
1286 * implements tree growing procedure:
1287 * - allocates new block
1288 * - moves top-level data (index block or leaf) into the new block
1289 * - initializes new top-level, creating index that points to the
1290 * just created block
1292 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1295 struct ext4_extent_header *neh;
1296 struct buffer_head *bh;
1297 ext4_fsblk_t newblock, goal = 0;
1298 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1300 size_t ext_size = 0;
1302 /* Try to prepend new index to old one */
1303 if (ext_depth(inode))
1304 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1305 if (goal > le32_to_cpu(es->s_first_data_block)) {
1306 flags |= EXT4_MB_HINT_TRY_GOAL;
1309 goal = ext4_inode_to_goal_block(inode);
1310 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1315 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1320 err = ext4_journal_get_create_access(handle, bh);
1326 ext_size = sizeof(EXT4_I(inode)->i_data);
1327 /* move top-level index/leaf into new block */
1328 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1329 /* zero out unused area in the extent block */
1330 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1332 /* set size of new block */
1333 neh = ext_block_hdr(bh);
1334 /* old root could have indexes or leaves
1335 * so calculate e_max right way */
1336 if (ext_depth(inode))
1337 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1339 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1340 neh->eh_magic = EXT4_EXT_MAGIC;
1341 ext4_extent_block_csum_set(inode, neh);
1342 set_buffer_uptodate(bh);
1345 err = ext4_handle_dirty_metadata(handle, inode, bh);
1349 /* Update top-level index: num,max,pointer */
1350 neh = ext_inode_hdr(inode);
1351 neh->eh_entries = cpu_to_le16(1);
1352 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1353 if (neh->eh_depth == 0) {
1354 /* Root extent block becomes index block */
1355 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1356 EXT_FIRST_INDEX(neh)->ei_block =
1357 EXT_FIRST_EXTENT(neh)->ee_block;
1359 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1360 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1361 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1362 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1364 le16_add_cpu(&neh->eh_depth, 1);
1365 ext4_mark_inode_dirty(handle, inode);
1373 * ext4_ext_create_new_leaf:
1374 * finds empty index and adds new leaf.
1375 * if no free index is found, then it requests in-depth growing.
1377 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1378 unsigned int mb_flags,
1379 unsigned int gb_flags,
1380 struct ext4_ext_path **ppath,
1381 struct ext4_extent *newext)
1383 struct ext4_ext_path *path = *ppath;
1384 struct ext4_ext_path *curp;
1385 int depth, i, err = 0;
1388 i = depth = ext_depth(inode);
1390 /* walk up to the tree and look for free index entry */
1391 curp = path + depth;
1392 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1397 /* we use already allocated block for index block,
1398 * so subsequent data blocks should be contiguous */
1399 if (EXT_HAS_FREE_INDEX(curp)) {
1400 /* if we found index with free entry, then use that
1401 * entry: create all needed subtree and add new leaf */
1402 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1407 path = ext4_find_extent(inode,
1408 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1411 err = PTR_ERR(path);
1413 /* tree is full, time to grow in depth */
1414 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1419 path = ext4_find_extent(inode,
1420 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1423 err = PTR_ERR(path);
1428 * only first (depth 0 -> 1) produces free space;
1429 * in all other cases we have to split the grown tree
1431 depth = ext_depth(inode);
1432 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1433 /* now we need to split */
1443 * search the closest allocated block to the left for *logical
1444 * and returns it at @logical + it's physical address at @phys
1445 * if *logical is the smallest allocated block, the function
1446 * returns 0 at @phys
1447 * return value contains 0 (success) or error code
1449 static int ext4_ext_search_left(struct inode *inode,
1450 struct ext4_ext_path *path,
1451 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1453 struct ext4_extent_idx *ix;
1454 struct ext4_extent *ex;
1457 if (unlikely(path == NULL)) {
1458 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1459 return -EFSCORRUPTED;
1461 depth = path->p_depth;
1464 if (depth == 0 && path->p_ext == NULL)
1467 /* usually extent in the path covers blocks smaller
1468 * then *logical, but it can be that extent is the
1469 * first one in the file */
1471 ex = path[depth].p_ext;
1472 ee_len = ext4_ext_get_actual_len(ex);
1473 if (*logical < le32_to_cpu(ex->ee_block)) {
1474 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1475 EXT4_ERROR_INODE(inode,
1476 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1477 *logical, le32_to_cpu(ex->ee_block));
1478 return -EFSCORRUPTED;
1480 while (--depth >= 0) {
1481 ix = path[depth].p_idx;
1482 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1483 EXT4_ERROR_INODE(inode,
1484 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1485 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1486 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1487 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1489 return -EFSCORRUPTED;
1495 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1496 EXT4_ERROR_INODE(inode,
1497 "logical %d < ee_block %d + ee_len %d!",
1498 *logical, le32_to_cpu(ex->ee_block), ee_len);
1499 return -EFSCORRUPTED;
1502 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1503 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1508 * search the closest allocated block to the right for *logical
1509 * and returns it at @logical + it's physical address at @phys
1510 * if *logical is the largest allocated block, the function
1511 * returns 0 at @phys
1512 * return value contains 0 (success) or error code
1514 static int ext4_ext_search_right(struct inode *inode,
1515 struct ext4_ext_path *path,
1516 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1517 struct ext4_extent **ret_ex)
1519 struct buffer_head *bh = NULL;
1520 struct ext4_extent_header *eh;
1521 struct ext4_extent_idx *ix;
1522 struct ext4_extent *ex;
1524 int depth; /* Note, NOT eh_depth; depth from top of tree */
1527 if (unlikely(path == NULL)) {
1528 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1529 return -EFSCORRUPTED;
1531 depth = path->p_depth;
1534 if (depth == 0 && path->p_ext == NULL)
1537 /* usually extent in the path covers blocks smaller
1538 * then *logical, but it can be that extent is the
1539 * first one in the file */
1541 ex = path[depth].p_ext;
1542 ee_len = ext4_ext_get_actual_len(ex);
1543 if (*logical < le32_to_cpu(ex->ee_block)) {
1544 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1545 EXT4_ERROR_INODE(inode,
1546 "first_extent(path[%d].p_hdr) != ex",
1548 return -EFSCORRUPTED;
1550 while (--depth >= 0) {
1551 ix = path[depth].p_idx;
1552 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1553 EXT4_ERROR_INODE(inode,
1554 "ix != EXT_FIRST_INDEX *logical %d!",
1556 return -EFSCORRUPTED;
1562 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1563 EXT4_ERROR_INODE(inode,
1564 "logical %d < ee_block %d + ee_len %d!",
1565 *logical, le32_to_cpu(ex->ee_block), ee_len);
1566 return -EFSCORRUPTED;
1569 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1570 /* next allocated block in this leaf */
1575 /* go up and search for index to the right */
1576 while (--depth >= 0) {
1577 ix = path[depth].p_idx;
1578 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1582 /* we've gone up to the root and found no index to the right */
1586 /* we've found index to the right, let's
1587 * follow it and find the closest allocated
1588 * block to the right */
1590 block = ext4_idx_pblock(ix);
1591 while (++depth < path->p_depth) {
1592 /* subtract from p_depth to get proper eh_depth */
1593 bh = read_extent_tree_block(inode, block,
1594 path->p_depth - depth, 0);
1597 eh = ext_block_hdr(bh);
1598 ix = EXT_FIRST_INDEX(eh);
1599 block = ext4_idx_pblock(ix);
1603 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1606 eh = ext_block_hdr(bh);
1607 ex = EXT_FIRST_EXTENT(eh);
1609 *logical = le32_to_cpu(ex->ee_block);
1610 *phys = ext4_ext_pblock(ex);
1618 * ext4_ext_next_allocated_block:
1619 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1620 * NOTE: it considers block number from index entry as
1621 * allocated block. Thus, index entries have to be consistent
1625 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1629 BUG_ON(path == NULL);
1630 depth = path->p_depth;
1632 if (depth == 0 && path->p_ext == NULL)
1633 return EXT_MAX_BLOCKS;
1635 while (depth >= 0) {
1636 if (depth == path->p_depth) {
1638 if (path[depth].p_ext &&
1639 path[depth].p_ext !=
1640 EXT_LAST_EXTENT(path[depth].p_hdr))
1641 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1644 if (path[depth].p_idx !=
1645 EXT_LAST_INDEX(path[depth].p_hdr))
1646 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1651 return EXT_MAX_BLOCKS;
1655 * ext4_ext_next_leaf_block:
1656 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1658 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1662 BUG_ON(path == NULL);
1663 depth = path->p_depth;
1665 /* zero-tree has no leaf blocks at all */
1667 return EXT_MAX_BLOCKS;
1669 /* go to index block */
1672 while (depth >= 0) {
1673 if (path[depth].p_idx !=
1674 EXT_LAST_INDEX(path[depth].p_hdr))
1675 return (ext4_lblk_t)
1676 le32_to_cpu(path[depth].p_idx[1].ei_block);
1680 return EXT_MAX_BLOCKS;
1684 * ext4_ext_correct_indexes:
1685 * if leaf gets modified and modified extent is first in the leaf,
1686 * then we have to correct all indexes above.
1687 * TODO: do we need to correct tree in all cases?
1689 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1690 struct ext4_ext_path *path)
1692 struct ext4_extent_header *eh;
1693 int depth = ext_depth(inode);
1694 struct ext4_extent *ex;
1698 eh = path[depth].p_hdr;
1699 ex = path[depth].p_ext;
1701 if (unlikely(ex == NULL || eh == NULL)) {
1702 EXT4_ERROR_INODE(inode,
1703 "ex %p == NULL or eh %p == NULL", ex, eh);
1704 return -EFSCORRUPTED;
1708 /* there is no tree at all */
1712 if (ex != EXT_FIRST_EXTENT(eh)) {
1713 /* we correct tree if first leaf got modified only */
1718 * TODO: we need correction if border is smaller than current one
1721 border = path[depth].p_ext->ee_block;
1722 err = ext4_ext_get_access(handle, inode, path + k);
1725 path[k].p_idx->ei_block = border;
1726 err = ext4_ext_dirty(handle, inode, path + k);
1731 /* change all left-side indexes */
1732 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1734 err = ext4_ext_get_access(handle, inode, path + k);
1737 path[k].p_idx->ei_block = border;
1738 err = ext4_ext_dirty(handle, inode, path + k);
1747 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1748 struct ext4_extent *ex2)
1750 unsigned short ext1_ee_len, ext2_ee_len;
1752 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1755 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1756 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1758 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1759 le32_to_cpu(ex2->ee_block))
1763 * To allow future support for preallocated extents to be added
1764 * as an RO_COMPAT feature, refuse to merge to extents if
1765 * this can result in the top bit of ee_len being set.
1767 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1770 * The check for IO to unwritten extent is somewhat racy as we
1771 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1772 * dropping i_data_sem. But reserved blocks should save us in that
1775 if (ext4_ext_is_unwritten(ex1) &&
1776 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1777 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1778 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1780 #ifdef AGGRESSIVE_TEST
1781 if (ext1_ee_len >= 4)
1785 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1791 * This function tries to merge the "ex" extent to the next extent in the tree.
1792 * It always tries to merge towards right. If you want to merge towards
1793 * left, pass "ex - 1" as argument instead of "ex".
1794 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1795 * 1 if they got merged.
1797 static int ext4_ext_try_to_merge_right(struct inode *inode,
1798 struct ext4_ext_path *path,
1799 struct ext4_extent *ex)
1801 struct ext4_extent_header *eh;
1802 unsigned int depth, len;
1803 int merge_done = 0, unwritten;
1805 depth = ext_depth(inode);
1806 BUG_ON(path[depth].p_hdr == NULL);
1807 eh = path[depth].p_hdr;
1809 while (ex < EXT_LAST_EXTENT(eh)) {
1810 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1812 /* merge with next extent! */
1813 unwritten = ext4_ext_is_unwritten(ex);
1814 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1815 + ext4_ext_get_actual_len(ex + 1));
1817 ext4_ext_mark_unwritten(ex);
1819 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1820 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1821 * sizeof(struct ext4_extent);
1822 memmove(ex + 1, ex + 2, len);
1824 le16_add_cpu(&eh->eh_entries, -1);
1826 WARN_ON(eh->eh_entries == 0);
1827 if (!eh->eh_entries)
1828 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1835 * This function does a very simple check to see if we can collapse
1836 * an extent tree with a single extent tree leaf block into the inode.
1838 static void ext4_ext_try_to_merge_up(handle_t *handle,
1839 struct inode *inode,
1840 struct ext4_ext_path *path)
1843 unsigned max_root = ext4_ext_space_root(inode, 0);
1846 if ((path[0].p_depth != 1) ||
1847 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1848 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1852 * We need to modify the block allocation bitmap and the block
1853 * group descriptor to release the extent tree block. If we
1854 * can't get the journal credits, give up.
1856 if (ext4_journal_extend(handle, 2))
1860 * Copy the extent data up to the inode
1862 blk = ext4_idx_pblock(path[0].p_idx);
1863 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1864 sizeof(struct ext4_extent_idx);
1865 s += sizeof(struct ext4_extent_header);
1867 path[1].p_maxdepth = path[0].p_maxdepth;
1868 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1869 path[0].p_depth = 0;
1870 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1871 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1872 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1874 brelse(path[1].p_bh);
1875 ext4_free_blocks(handle, inode, NULL, blk, 1,
1876 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1880 * This function tries to merge the @ex extent to neighbours in the tree.
1881 * return 1 if merge left else 0.
1883 static void ext4_ext_try_to_merge(handle_t *handle,
1884 struct inode *inode,
1885 struct ext4_ext_path *path,
1886 struct ext4_extent *ex) {
1887 struct ext4_extent_header *eh;
1891 depth = ext_depth(inode);
1892 BUG_ON(path[depth].p_hdr == NULL);
1893 eh = path[depth].p_hdr;
1895 if (ex > EXT_FIRST_EXTENT(eh))
1896 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1899 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1901 ext4_ext_try_to_merge_up(handle, inode, path);
1905 * check if a portion of the "newext" extent overlaps with an
1908 * If there is an overlap discovered, it updates the length of the newext
1909 * such that there will be no overlap, and then returns 1.
1910 * If there is no overlap found, it returns 0.
1912 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1913 struct inode *inode,
1914 struct ext4_extent *newext,
1915 struct ext4_ext_path *path)
1918 unsigned int depth, len1;
1919 unsigned int ret = 0;
1921 b1 = le32_to_cpu(newext->ee_block);
1922 len1 = ext4_ext_get_actual_len(newext);
1923 depth = ext_depth(inode);
1924 if (!path[depth].p_ext)
1926 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1929 * get the next allocated block if the extent in the path
1930 * is before the requested block(s)
1933 b2 = ext4_ext_next_allocated_block(path);
1934 if (b2 == EXT_MAX_BLOCKS)
1936 b2 = EXT4_LBLK_CMASK(sbi, b2);
1939 /* check for wrap through zero on extent logical start block*/
1940 if (b1 + len1 < b1) {
1941 len1 = EXT_MAX_BLOCKS - b1;
1942 newext->ee_len = cpu_to_le16(len1);
1946 /* check for overlap */
1947 if (b1 + len1 > b2) {
1948 newext->ee_len = cpu_to_le16(b2 - b1);
1956 * ext4_ext_insert_extent:
1957 * tries to merge requsted extent into the existing extent or
1958 * inserts requested extent as new one into the tree,
1959 * creating new leaf in the no-space case.
1961 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1962 struct ext4_ext_path **ppath,
1963 struct ext4_extent *newext, int gb_flags)
1965 struct ext4_ext_path *path = *ppath;
1966 struct ext4_extent_header *eh;
1967 struct ext4_extent *ex, *fex;
1968 struct ext4_extent *nearex; /* nearest extent */
1969 struct ext4_ext_path *npath = NULL;
1970 int depth, len, err;
1972 int mb_flags = 0, unwritten;
1974 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1975 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1976 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1977 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1978 return -EFSCORRUPTED;
1980 depth = ext_depth(inode);
1981 ex = path[depth].p_ext;
1982 eh = path[depth].p_hdr;
1983 if (unlikely(path[depth].p_hdr == NULL)) {
1984 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1985 return -EFSCORRUPTED;
1988 /* try to insert block into found extent and return */
1989 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1992 * Try to see whether we should rather test the extent on
1993 * right from ex, or from the left of ex. This is because
1994 * ext4_find_extent() can return either extent on the
1995 * left, or on the right from the searched position. This
1996 * will make merging more effective.
1998 if (ex < EXT_LAST_EXTENT(eh) &&
1999 (le32_to_cpu(ex->ee_block) +
2000 ext4_ext_get_actual_len(ex) <
2001 le32_to_cpu(newext->ee_block))) {
2004 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2005 (le32_to_cpu(newext->ee_block) +
2006 ext4_ext_get_actual_len(newext) <
2007 le32_to_cpu(ex->ee_block)))
2010 /* Try to append newex to the ex */
2011 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2012 ext_debug("append [%d]%d block to %u:[%d]%d"
2014 ext4_ext_is_unwritten(newext),
2015 ext4_ext_get_actual_len(newext),
2016 le32_to_cpu(ex->ee_block),
2017 ext4_ext_is_unwritten(ex),
2018 ext4_ext_get_actual_len(ex),
2019 ext4_ext_pblock(ex));
2020 err = ext4_ext_get_access(handle, inode,
2024 unwritten = ext4_ext_is_unwritten(ex);
2025 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2026 + ext4_ext_get_actual_len(newext));
2028 ext4_ext_mark_unwritten(ex);
2029 eh = path[depth].p_hdr;
2035 /* Try to prepend newex to the ex */
2036 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2037 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2039 le32_to_cpu(newext->ee_block),
2040 ext4_ext_is_unwritten(newext),
2041 ext4_ext_get_actual_len(newext),
2042 le32_to_cpu(ex->ee_block),
2043 ext4_ext_is_unwritten(ex),
2044 ext4_ext_get_actual_len(ex),
2045 ext4_ext_pblock(ex));
2046 err = ext4_ext_get_access(handle, inode,
2051 unwritten = ext4_ext_is_unwritten(ex);
2052 ex->ee_block = newext->ee_block;
2053 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2054 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2055 + ext4_ext_get_actual_len(newext));
2057 ext4_ext_mark_unwritten(ex);
2058 eh = path[depth].p_hdr;
2064 depth = ext_depth(inode);
2065 eh = path[depth].p_hdr;
2066 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2069 /* probably next leaf has space for us? */
2070 fex = EXT_LAST_EXTENT(eh);
2071 next = EXT_MAX_BLOCKS;
2072 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2073 next = ext4_ext_next_leaf_block(path);
2074 if (next != EXT_MAX_BLOCKS) {
2075 ext_debug("next leaf block - %u\n", next);
2076 BUG_ON(npath != NULL);
2077 npath = ext4_find_extent(inode, next, NULL, 0);
2079 return PTR_ERR(npath);
2080 BUG_ON(npath->p_depth != path->p_depth);
2081 eh = npath[depth].p_hdr;
2082 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2083 ext_debug("next leaf isn't full(%d)\n",
2084 le16_to_cpu(eh->eh_entries));
2088 ext_debug("next leaf has no free space(%d,%d)\n",
2089 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2093 * There is no free space in the found leaf.
2094 * We're gonna add a new leaf in the tree.
2096 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2097 mb_flags |= EXT4_MB_USE_RESERVED;
2098 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2102 depth = ext_depth(inode);
2103 eh = path[depth].p_hdr;
2106 nearex = path[depth].p_ext;
2108 err = ext4_ext_get_access(handle, inode, path + depth);
2113 /* there is no extent in this leaf, create first one */
2114 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2115 le32_to_cpu(newext->ee_block),
2116 ext4_ext_pblock(newext),
2117 ext4_ext_is_unwritten(newext),
2118 ext4_ext_get_actual_len(newext));
2119 nearex = EXT_FIRST_EXTENT(eh);
2121 if (le32_to_cpu(newext->ee_block)
2122 > le32_to_cpu(nearex->ee_block)) {
2124 ext_debug("insert %u:%llu:[%d]%d before: "
2126 le32_to_cpu(newext->ee_block),
2127 ext4_ext_pblock(newext),
2128 ext4_ext_is_unwritten(newext),
2129 ext4_ext_get_actual_len(newext),
2134 BUG_ON(newext->ee_block == nearex->ee_block);
2135 ext_debug("insert %u:%llu:[%d]%d after: "
2137 le32_to_cpu(newext->ee_block),
2138 ext4_ext_pblock(newext),
2139 ext4_ext_is_unwritten(newext),
2140 ext4_ext_get_actual_len(newext),
2143 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2145 ext_debug("insert %u:%llu:[%d]%d: "
2146 "move %d extents from 0x%p to 0x%p\n",
2147 le32_to_cpu(newext->ee_block),
2148 ext4_ext_pblock(newext),
2149 ext4_ext_is_unwritten(newext),
2150 ext4_ext_get_actual_len(newext),
2151 len, nearex, nearex + 1);
2152 memmove(nearex + 1, nearex,
2153 len * sizeof(struct ext4_extent));
2157 le16_add_cpu(&eh->eh_entries, 1);
2158 path[depth].p_ext = nearex;
2159 nearex->ee_block = newext->ee_block;
2160 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2161 nearex->ee_len = newext->ee_len;
2164 /* try to merge extents */
2165 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2166 ext4_ext_try_to_merge(handle, inode, path, nearex);
2169 /* time to correct all indexes above */
2170 err = ext4_ext_correct_indexes(handle, inode, path);
2174 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2177 ext4_ext_drop_refs(npath);
2182 static int ext4_fill_fiemap_extents(struct inode *inode,
2183 ext4_lblk_t block, ext4_lblk_t num,
2184 struct fiemap_extent_info *fieinfo)
2186 struct ext4_ext_path *path = NULL;
2187 struct ext4_extent *ex;
2188 struct extent_status es;
2189 ext4_lblk_t next, next_del, start = 0, end = 0;
2190 ext4_lblk_t last = block + num;
2191 int exists, depth = 0, err = 0;
2192 unsigned int flags = 0;
2193 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2195 while (block < last && block != EXT_MAX_BLOCKS) {
2197 /* find extent for this block */
2198 down_read(&EXT4_I(inode)->i_data_sem);
2200 path = ext4_find_extent(inode, block, &path, 0);
2202 up_read(&EXT4_I(inode)->i_data_sem);
2203 err = PTR_ERR(path);
2208 depth = ext_depth(inode);
2209 if (unlikely(path[depth].p_hdr == NULL)) {
2210 up_read(&EXT4_I(inode)->i_data_sem);
2211 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2212 err = -EFSCORRUPTED;
2215 ex = path[depth].p_ext;
2216 next = ext4_ext_next_allocated_block(path);
2221 /* there is no extent yet, so try to allocate
2222 * all requested space */
2225 } else if (le32_to_cpu(ex->ee_block) > block) {
2226 /* need to allocate space before found extent */
2228 end = le32_to_cpu(ex->ee_block);
2229 if (block + num < end)
2231 } else if (block >= le32_to_cpu(ex->ee_block)
2232 + ext4_ext_get_actual_len(ex)) {
2233 /* need to allocate space after found extent */
2238 } else if (block >= le32_to_cpu(ex->ee_block)) {
2240 * some part of requested space is covered
2244 end = le32_to_cpu(ex->ee_block)
2245 + ext4_ext_get_actual_len(ex);
2246 if (block + num < end)
2252 BUG_ON(end <= start);
2256 es.es_len = end - start;
2259 es.es_lblk = le32_to_cpu(ex->ee_block);
2260 es.es_len = ext4_ext_get_actual_len(ex);
2261 es.es_pblk = ext4_ext_pblock(ex);
2262 if (ext4_ext_is_unwritten(ex))
2263 flags |= FIEMAP_EXTENT_UNWRITTEN;
2267 * Find delayed extent and update es accordingly. We call
2268 * it even in !exists case to find out whether es is the
2269 * last existing extent or not.
2271 next_del = ext4_find_delayed_extent(inode, &es);
2272 if (!exists && next_del) {
2274 flags |= (FIEMAP_EXTENT_DELALLOC |
2275 FIEMAP_EXTENT_UNKNOWN);
2277 up_read(&EXT4_I(inode)->i_data_sem);
2279 if (unlikely(es.es_len == 0)) {
2280 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2281 err = -EFSCORRUPTED;
2286 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2287 * we need to check next == EXT_MAX_BLOCKS because it is
2288 * possible that an extent is with unwritten and delayed
2289 * status due to when an extent is delayed allocated and
2290 * is allocated by fallocate status tree will track both of
2293 * So we could return a unwritten and delayed extent, and
2294 * its block is equal to 'next'.
2296 if (next == next_del && next == EXT_MAX_BLOCKS) {
2297 flags |= FIEMAP_EXTENT_LAST;
2298 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2299 next != EXT_MAX_BLOCKS)) {
2300 EXT4_ERROR_INODE(inode,
2301 "next extent == %u, next "
2302 "delalloc extent = %u",
2304 err = -EFSCORRUPTED;
2310 err = fiemap_fill_next_extent(fieinfo,
2311 (__u64)es.es_lblk << blksize_bits,
2312 (__u64)es.es_pblk << blksize_bits,
2313 (__u64)es.es_len << blksize_bits,
2323 block = es.es_lblk + es.es_len;
2326 ext4_ext_drop_refs(path);
2332 * ext4_ext_determine_hole - determine hole around given block
2333 * @inode: inode we lookup in
2334 * @path: path in extent tree to @lblk
2335 * @lblk: pointer to logical block around which we want to determine hole
2337 * Determine hole length (and start if easily possible) around given logical
2338 * block. We don't try too hard to find the beginning of the hole but @path
2339 * actually points to extent before @lblk, we provide it.
2341 * The function returns the length of a hole starting at @lblk. We update @lblk
2342 * to the beginning of the hole if we managed to find it.
2344 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2345 struct ext4_ext_path *path,
2348 int depth = ext_depth(inode);
2349 struct ext4_extent *ex;
2352 ex = path[depth].p_ext;
2354 /* there is no extent yet, so gap is [0;-] */
2356 len = EXT_MAX_BLOCKS;
2357 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2358 len = le32_to_cpu(ex->ee_block) - *lblk;
2359 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2360 + ext4_ext_get_actual_len(ex)) {
2363 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2364 next = ext4_ext_next_allocated_block(path);
2365 BUG_ON(next == *lblk);
2374 * ext4_ext_put_gap_in_cache:
2375 * calculate boundaries of the gap that the requested block fits into
2376 * and cache this gap
2379 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2380 ext4_lblk_t hole_len)
2382 struct extent_status es;
2384 ext4_es_find_delayed_extent_range(inode, hole_start,
2385 hole_start + hole_len - 1, &es);
2387 /* There's delayed extent containing lblock? */
2388 if (es.es_lblk <= hole_start)
2390 hole_len = min(es.es_lblk - hole_start, hole_len);
2392 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2393 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2394 EXTENT_STATUS_HOLE);
2399 * removes index from the index block.
2401 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2402 struct ext4_ext_path *path, int depth)
2407 /* free index block */
2409 path = path + depth;
2410 leaf = ext4_idx_pblock(path->p_idx);
2411 if (unlikely(path->p_hdr->eh_entries == 0)) {
2412 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2413 return -EFSCORRUPTED;
2415 err = ext4_ext_get_access(handle, inode, path);
2419 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2420 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2421 len *= sizeof(struct ext4_extent_idx);
2422 memmove(path->p_idx, path->p_idx + 1, len);
2425 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2426 err = ext4_ext_dirty(handle, inode, path);
2429 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2430 trace_ext4_ext_rm_idx(inode, leaf);
2432 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2433 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2435 while (--depth >= 0) {
2436 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2439 err = ext4_ext_get_access(handle, inode, path);
2442 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2443 err = ext4_ext_dirty(handle, inode, path);
2451 * ext4_ext_calc_credits_for_single_extent:
2452 * This routine returns max. credits that needed to insert an extent
2453 * to the extent tree.
2454 * When pass the actual path, the caller should calculate credits
2457 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2458 struct ext4_ext_path *path)
2461 int depth = ext_depth(inode);
2464 /* probably there is space in leaf? */
2465 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2466 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2469 * There are some space in the leaf tree, no
2470 * need to account for leaf block credit
2472 * bitmaps and block group descriptor blocks
2473 * and other metadata blocks still need to be
2476 /* 1 bitmap, 1 block group descriptor */
2477 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2482 return ext4_chunk_trans_blocks(inode, nrblocks);
2486 * How many index/leaf blocks need to change/allocate to add @extents extents?
2488 * If we add a single extent, then in the worse case, each tree level
2489 * index/leaf need to be changed in case of the tree split.
2491 * If more extents are inserted, they could cause the whole tree split more
2492 * than once, but this is really rare.
2494 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2499 /* If we are converting the inline data, only one is needed here. */
2500 if (ext4_has_inline_data(inode))
2503 depth = ext_depth(inode);
2513 static inline int get_default_free_blocks_flags(struct inode *inode)
2515 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2516 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2517 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2518 else if (ext4_should_journal_data(inode))
2519 return EXT4_FREE_BLOCKS_FORGET;
2523 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2524 struct ext4_extent *ex,
2525 long long *partial_cluster,
2526 ext4_lblk_t from, ext4_lblk_t to)
2528 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2529 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2531 int flags = get_default_free_blocks_flags(inode);
2534 * For bigalloc file systems, we never free a partial cluster
2535 * at the beginning of the extent. Instead, we make a note
2536 * that we tried freeing the cluster, and check to see if we
2537 * need to free it on a subsequent call to ext4_remove_blocks,
2538 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2540 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2542 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2544 * If we have a partial cluster, and it's different from the
2545 * cluster of the last block, we need to explicitly free the
2546 * partial cluster here.
2548 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2549 if (*partial_cluster > 0 &&
2550 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2551 ext4_free_blocks(handle, inode, NULL,
2552 EXT4_C2B(sbi, *partial_cluster),
2553 sbi->s_cluster_ratio, flags);
2554 *partial_cluster = 0;
2557 #ifdef EXTENTS_STATS
2559 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2560 spin_lock(&sbi->s_ext_stats_lock);
2561 sbi->s_ext_blocks += ee_len;
2562 sbi->s_ext_extents++;
2563 if (ee_len < sbi->s_ext_min)
2564 sbi->s_ext_min = ee_len;
2565 if (ee_len > sbi->s_ext_max)
2566 sbi->s_ext_max = ee_len;
2567 if (ext_depth(inode) > sbi->s_depth_max)
2568 sbi->s_depth_max = ext_depth(inode);
2569 spin_unlock(&sbi->s_ext_stats_lock);
2572 if (from >= le32_to_cpu(ex->ee_block)
2573 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2576 long long first_cluster;
2578 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2579 pblk = ext4_ext_pblock(ex) + ee_len - num;
2581 * Usually we want to free partial cluster at the end of the
2582 * extent, except for the situation when the cluster is still
2583 * used by any other extent (partial_cluster is negative).
2585 if (*partial_cluster < 0 &&
2586 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2587 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2589 ext_debug("free last %u blocks starting %llu partial %lld\n",
2590 num, pblk, *partial_cluster);
2591 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2593 * If the block range to be freed didn't start at the
2594 * beginning of a cluster, and we removed the entire
2595 * extent and the cluster is not used by any other extent,
2596 * save the partial cluster here, since we might need to
2597 * delete if we determine that the truncate or punch hole
2598 * operation has removed all of the blocks in the cluster.
2599 * If that cluster is used by another extent, preserve its
2600 * negative value so it isn't freed later on.
2602 * If the whole extent wasn't freed, we've reached the
2603 * start of the truncated/punched region and have finished
2604 * removing blocks. If there's a partial cluster here it's
2605 * shared with the remainder of the extent and is no longer
2606 * a candidate for removal.
2608 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2609 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2610 if (first_cluster != -*partial_cluster)
2611 *partial_cluster = first_cluster;
2613 *partial_cluster = 0;
2616 ext4_error(sbi->s_sb, "strange request: removal(2) "
2618 from, to, le32_to_cpu(ex->ee_block), ee_len);
2624 * ext4_ext_rm_leaf() Removes the extents associated with the
2625 * blocks appearing between "start" and "end". Both "start"
2626 * and "end" must appear in the same extent or EIO is returned.
2628 * @handle: The journal handle
2629 * @inode: The files inode
2630 * @path: The path to the leaf
2631 * @partial_cluster: The cluster which we'll have to free if all extents
2632 * has been released from it. However, if this value is
2633 * negative, it's a cluster just to the right of the
2634 * punched region and it must not be freed.
2635 * @start: The first block to remove
2636 * @end: The last block to remove
2639 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2640 struct ext4_ext_path *path,
2641 long long *partial_cluster,
2642 ext4_lblk_t start, ext4_lblk_t end)
2644 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2645 int err = 0, correct_index = 0;
2646 int depth = ext_depth(inode), credits;
2647 struct ext4_extent_header *eh;
2650 ext4_lblk_t ex_ee_block;
2651 unsigned short ex_ee_len;
2652 unsigned unwritten = 0;
2653 struct ext4_extent *ex;
2656 /* the header must be checked already in ext4_ext_remove_space() */
2657 ext_debug("truncate since %u in leaf to %u\n", start, end);
2658 if (!path[depth].p_hdr)
2659 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2660 eh = path[depth].p_hdr;
2661 if (unlikely(path[depth].p_hdr == NULL)) {
2662 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2663 return -EFSCORRUPTED;
2665 /* find where to start removing */
2666 ex = path[depth].p_ext;
2668 ex = EXT_LAST_EXTENT(eh);
2670 ex_ee_block = le32_to_cpu(ex->ee_block);
2671 ex_ee_len = ext4_ext_get_actual_len(ex);
2673 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2675 while (ex >= EXT_FIRST_EXTENT(eh) &&
2676 ex_ee_block + ex_ee_len > start) {
2678 if (ext4_ext_is_unwritten(ex))
2683 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2684 unwritten, ex_ee_len);
2685 path[depth].p_ext = ex;
2687 a = ex_ee_block > start ? ex_ee_block : start;
2688 b = ex_ee_block+ex_ee_len - 1 < end ?
2689 ex_ee_block+ex_ee_len - 1 : end;
2691 ext_debug(" border %u:%u\n", a, b);
2693 /* If this extent is beyond the end of the hole, skip it */
2694 if (end < ex_ee_block) {
2696 * We're going to skip this extent and move to another,
2697 * so note that its first cluster is in use to avoid
2698 * freeing it when removing blocks. Eventually, the
2699 * right edge of the truncated/punched region will
2700 * be just to the left.
2702 if (sbi->s_cluster_ratio > 1) {
2703 pblk = ext4_ext_pblock(ex);
2705 -(long long) EXT4_B2C(sbi, pblk);
2708 ex_ee_block = le32_to_cpu(ex->ee_block);
2709 ex_ee_len = ext4_ext_get_actual_len(ex);
2711 } else if (b != ex_ee_block + ex_ee_len - 1) {
2712 EXT4_ERROR_INODE(inode,
2713 "can not handle truncate %u:%u "
2715 start, end, ex_ee_block,
2716 ex_ee_block + ex_ee_len - 1);
2717 err = -EFSCORRUPTED;
2719 } else if (a != ex_ee_block) {
2720 /* remove tail of the extent */
2721 num = a - ex_ee_block;
2723 /* remove whole extent: excellent! */
2727 * 3 for leaf, sb, and inode plus 2 (bmap and group
2728 * descriptor) for each block group; assume two block
2729 * groups plus ex_ee_len/blocks_per_block_group for
2732 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2733 if (ex == EXT_FIRST_EXTENT(eh)) {
2735 credits += (ext_depth(inode)) + 1;
2737 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2739 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2743 err = ext4_ext_get_access(handle, inode, path + depth);
2747 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2753 /* this extent is removed; mark slot entirely unused */
2754 ext4_ext_store_pblock(ex, 0);
2756 ex->ee_len = cpu_to_le16(num);
2758 * Do not mark unwritten if all the blocks in the
2759 * extent have been removed.
2761 if (unwritten && num)
2762 ext4_ext_mark_unwritten(ex);
2764 * If the extent was completely released,
2765 * we need to remove it from the leaf
2768 if (end != EXT_MAX_BLOCKS - 1) {
2770 * For hole punching, we need to scoot all the
2771 * extents up when an extent is removed so that
2772 * we dont have blank extents in the middle
2774 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2775 sizeof(struct ext4_extent));
2777 /* Now get rid of the one at the end */
2778 memset(EXT_LAST_EXTENT(eh), 0,
2779 sizeof(struct ext4_extent));
2781 le16_add_cpu(&eh->eh_entries, -1);
2784 err = ext4_ext_dirty(handle, inode, path + depth);
2788 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2789 ext4_ext_pblock(ex));
2791 ex_ee_block = le32_to_cpu(ex->ee_block);
2792 ex_ee_len = ext4_ext_get_actual_len(ex);
2795 if (correct_index && eh->eh_entries)
2796 err = ext4_ext_correct_indexes(handle, inode, path);
2799 * If there's a partial cluster and at least one extent remains in
2800 * the leaf, free the partial cluster if it isn't shared with the
2801 * current extent. If it is shared with the current extent
2802 * we zero partial_cluster because we've reached the start of the
2803 * truncated/punched region and we're done removing blocks.
2805 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2806 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2807 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2808 ext4_free_blocks(handle, inode, NULL,
2809 EXT4_C2B(sbi, *partial_cluster),
2810 sbi->s_cluster_ratio,
2811 get_default_free_blocks_flags(inode));
2813 *partial_cluster = 0;
2816 /* if this leaf is free, then we should
2817 * remove it from index block above */
2818 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2819 err = ext4_ext_rm_idx(handle, inode, path, depth);
2826 * ext4_ext_more_to_rm:
2827 * returns 1 if current index has to be freed (even partial)
2830 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2832 BUG_ON(path->p_idx == NULL);
2834 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2838 * if truncate on deeper level happened, it wasn't partial,
2839 * so we have to consider current index for truncation
2841 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2846 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2849 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2850 int depth = ext_depth(inode);
2851 struct ext4_ext_path *path = NULL;
2852 long long partial_cluster = 0;
2856 ext_debug("truncate since %u to %u\n", start, end);
2858 /* probably first extent we're gonna free will be last in block */
2859 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2861 return PTR_ERR(handle);
2864 trace_ext4_ext_remove_space(inode, start, end, depth);
2867 * Check if we are removing extents inside the extent tree. If that
2868 * is the case, we are going to punch a hole inside the extent tree
2869 * so we have to check whether we need to split the extent covering
2870 * the last block to remove so we can easily remove the part of it
2871 * in ext4_ext_rm_leaf().
2873 if (end < EXT_MAX_BLOCKS - 1) {
2874 struct ext4_extent *ex;
2875 ext4_lblk_t ee_block, ex_end, lblk;
2878 /* find extent for or closest extent to this block */
2879 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2881 ext4_journal_stop(handle);
2882 return PTR_ERR(path);
2884 depth = ext_depth(inode);
2885 /* Leaf not may not exist only if inode has no blocks at all */
2886 ex = path[depth].p_ext;
2889 EXT4_ERROR_INODE(inode,
2890 "path[%d].p_hdr == NULL",
2892 err = -EFSCORRUPTED;
2897 ee_block = le32_to_cpu(ex->ee_block);
2898 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2901 * See if the last block is inside the extent, if so split
2902 * the extent at 'end' block so we can easily remove the
2903 * tail of the first part of the split extent in
2904 * ext4_ext_rm_leaf().
2906 if (end >= ee_block && end < ex_end) {
2909 * If we're going to split the extent, note that
2910 * the cluster containing the block after 'end' is
2911 * in use to avoid freeing it when removing blocks.
2913 if (sbi->s_cluster_ratio > 1) {
2914 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2916 -(long long) EXT4_B2C(sbi, pblk);
2920 * Split the extent in two so that 'end' is the last
2921 * block in the first new extent. Also we should not
2922 * fail removing space due to ENOSPC so try to use
2923 * reserved block if that happens.
2925 err = ext4_force_split_extent_at(handle, inode, &path,
2930 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2932 * If there's an extent to the right its first cluster
2933 * contains the immediate right boundary of the
2934 * truncated/punched region. Set partial_cluster to
2935 * its negative value so it won't be freed if shared
2936 * with the current extent. The end < ee_block case
2937 * is handled in ext4_ext_rm_leaf().
2940 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2946 -(long long) EXT4_B2C(sbi, pblk);
2950 * We start scanning from right side, freeing all the blocks
2951 * after i_size and walking into the tree depth-wise.
2953 depth = ext_depth(inode);
2958 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2960 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2963 ext4_journal_stop(handle);
2966 path[0].p_maxdepth = path[0].p_depth = depth;
2967 path[0].p_hdr = ext_inode_hdr(inode);
2970 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2971 err = -EFSCORRUPTED;
2977 while (i >= 0 && err == 0) {
2979 /* this is leaf block */
2980 err = ext4_ext_rm_leaf(handle, inode, path,
2981 &partial_cluster, start,
2983 /* root level has p_bh == NULL, brelse() eats this */
2984 brelse(path[i].p_bh);
2985 path[i].p_bh = NULL;
2990 /* this is index block */
2991 if (!path[i].p_hdr) {
2992 ext_debug("initialize header\n");
2993 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2996 if (!path[i].p_idx) {
2997 /* this level hasn't been touched yet */
2998 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2999 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3000 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3002 le16_to_cpu(path[i].p_hdr->eh_entries));
3004 /* we were already here, see at next index */
3008 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3009 i, EXT_FIRST_INDEX(path[i].p_hdr),
3011 if (ext4_ext_more_to_rm(path + i)) {
3012 struct buffer_head *bh;
3013 /* go to the next level */
3014 ext_debug("move to level %d (block %llu)\n",
3015 i + 1, ext4_idx_pblock(path[i].p_idx));
3016 memset(path + i + 1, 0, sizeof(*path));
3017 bh = read_extent_tree_block(inode,
3018 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3021 /* should we reset i_size? */
3025 /* Yield here to deal with large extent trees.
3026 * Should be a no-op if we did IO above. */
3028 if (WARN_ON(i + 1 > depth)) {
3029 err = -EFSCORRUPTED;
3032 path[i + 1].p_bh = bh;
3034 /* save actual number of indexes since this
3035 * number is changed at the next iteration */
3036 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3039 /* we finished processing this index, go up */
3040 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3041 /* index is empty, remove it;
3042 * handle must be already prepared by the
3043 * truncatei_leaf() */
3044 err = ext4_ext_rm_idx(handle, inode, path, i);
3046 /* root level has p_bh == NULL, brelse() eats this */
3047 brelse(path[i].p_bh);
3048 path[i].p_bh = NULL;
3050 ext_debug("return to level %d\n", i);
3054 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3055 partial_cluster, path->p_hdr->eh_entries);
3058 * If we still have something in the partial cluster and we have removed
3059 * even the first extent, then we should free the blocks in the partial
3060 * cluster as well. (This code will only run when there are no leaves
3061 * to the immediate left of the truncated/punched region.)
3063 if (partial_cluster > 0 && err == 0) {
3064 /* don't zero partial_cluster since it's not used afterwards */
3065 ext4_free_blocks(handle, inode, NULL,
3066 EXT4_C2B(sbi, partial_cluster),
3067 sbi->s_cluster_ratio,
3068 get_default_free_blocks_flags(inode));
3071 /* TODO: flexible tree reduction should be here */
3072 if (path->p_hdr->eh_entries == 0) {
3074 * truncate to zero freed all the tree,
3075 * so we need to correct eh_depth
3077 err = ext4_ext_get_access(handle, inode, path);
3079 ext_inode_hdr(inode)->eh_depth = 0;
3080 ext_inode_hdr(inode)->eh_max =
3081 cpu_to_le16(ext4_ext_space_root(inode, 0));
3082 err = ext4_ext_dirty(handle, inode, path);
3086 ext4_ext_drop_refs(path);
3091 ext4_journal_stop(handle);
3097 * called at mount time
3099 void ext4_ext_init(struct super_block *sb)
3102 * possible initialization would be here
3105 if (ext4_has_feature_extents(sb)) {
3106 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3107 printk(KERN_INFO "EXT4-fs: file extents enabled"
3108 #ifdef AGGRESSIVE_TEST
3109 ", aggressive tests"
3111 #ifdef CHECK_BINSEARCH
3114 #ifdef EXTENTS_STATS
3119 #ifdef EXTENTS_STATS
3120 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3121 EXT4_SB(sb)->s_ext_min = 1 << 30;
3122 EXT4_SB(sb)->s_ext_max = 0;
3128 * called at umount time
3130 void ext4_ext_release(struct super_block *sb)
3132 if (!ext4_has_feature_extents(sb))
3135 #ifdef EXTENTS_STATS
3136 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3137 struct ext4_sb_info *sbi = EXT4_SB(sb);
3138 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3139 sbi->s_ext_blocks, sbi->s_ext_extents,
3140 sbi->s_ext_blocks / sbi->s_ext_extents);
3141 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3142 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3147 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3149 ext4_lblk_t ee_block;
3150 ext4_fsblk_t ee_pblock;
3151 unsigned int ee_len;
3153 ee_block = le32_to_cpu(ex->ee_block);
3154 ee_len = ext4_ext_get_actual_len(ex);
3155 ee_pblock = ext4_ext_pblock(ex);
3160 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3161 EXTENT_STATUS_WRITTEN);
3164 /* FIXME!! we need to try to merge to left or right after zero-out */
3165 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3167 ext4_fsblk_t ee_pblock;
3168 unsigned int ee_len;
3170 ee_len = ext4_ext_get_actual_len(ex);
3171 ee_pblock = ext4_ext_pblock(ex);
3172 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3177 * ext4_split_extent_at() splits an extent at given block.
3179 * @handle: the journal handle
3180 * @inode: the file inode
3181 * @path: the path to the extent
3182 * @split: the logical block where the extent is splitted.
3183 * @split_flags: indicates if the extent could be zeroout if split fails, and
3184 * the states(init or unwritten) of new extents.
3185 * @flags: flags used to insert new extent to extent tree.
3188 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3189 * of which are deterimined by split_flag.
3191 * There are two cases:
3192 * a> the extent are splitted into two extent.
3193 * b> split is not needed, and just mark the extent.
3195 * return 0 on success.
3197 static int ext4_split_extent_at(handle_t *handle,
3198 struct inode *inode,
3199 struct ext4_ext_path **ppath,
3204 struct ext4_ext_path *path = *ppath;
3205 ext4_fsblk_t newblock;
3206 ext4_lblk_t ee_block;
3207 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3208 struct ext4_extent *ex2 = NULL;
3209 unsigned int ee_len, depth;
3212 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3213 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3215 ext_debug("ext4_split_extents_at: inode %lu, logical"
3216 "block %llu\n", inode->i_ino, (unsigned long long)split);
3218 ext4_ext_show_leaf(inode, path);
3220 depth = ext_depth(inode);
3221 ex = path[depth].p_ext;
3222 ee_block = le32_to_cpu(ex->ee_block);
3223 ee_len = ext4_ext_get_actual_len(ex);
3224 newblock = split - ee_block + ext4_ext_pblock(ex);
3226 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3227 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3228 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3229 EXT4_EXT_MARK_UNWRIT1 |
3230 EXT4_EXT_MARK_UNWRIT2));
3232 err = ext4_ext_get_access(handle, inode, path + depth);
3236 if (split == ee_block) {
3238 * case b: block @split is the block that the extent begins with
3239 * then we just change the state of the extent, and splitting
3242 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3243 ext4_ext_mark_unwritten(ex);
3245 ext4_ext_mark_initialized(ex);
3247 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3248 ext4_ext_try_to_merge(handle, inode, path, ex);
3250 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3255 memcpy(&orig_ex, ex, sizeof(orig_ex));
3256 ex->ee_len = cpu_to_le16(split - ee_block);
3257 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3258 ext4_ext_mark_unwritten(ex);
3261 * path may lead to new leaf, not to original leaf any more
3262 * after ext4_ext_insert_extent() returns,
3264 err = ext4_ext_dirty(handle, inode, path + depth);
3266 goto fix_extent_len;
3269 ex2->ee_block = cpu_to_le32(split);
3270 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3271 ext4_ext_store_pblock(ex2, newblock);
3272 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3273 ext4_ext_mark_unwritten(ex2);
3275 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3276 if (err != -ENOSPC && err != -EDQUOT)
3279 if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3280 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3281 if (split_flag & EXT4_EXT_DATA_VALID1) {
3282 err = ext4_ext_zeroout(inode, ex2);
3283 zero_ex.ee_block = ex2->ee_block;
3284 zero_ex.ee_len = cpu_to_le16(
3285 ext4_ext_get_actual_len(ex2));
3286 ext4_ext_store_pblock(&zero_ex,
3287 ext4_ext_pblock(ex2));
3289 err = ext4_ext_zeroout(inode, ex);
3290 zero_ex.ee_block = ex->ee_block;
3291 zero_ex.ee_len = cpu_to_le16(
3292 ext4_ext_get_actual_len(ex));
3293 ext4_ext_store_pblock(&zero_ex,
3294 ext4_ext_pblock(ex));
3297 err = ext4_ext_zeroout(inode, &orig_ex);
3298 zero_ex.ee_block = orig_ex.ee_block;
3299 zero_ex.ee_len = cpu_to_le16(
3300 ext4_ext_get_actual_len(&orig_ex));
3301 ext4_ext_store_pblock(&zero_ex,
3302 ext4_ext_pblock(&orig_ex));
3306 /* update the extent length and mark as initialized */
3307 ex->ee_len = cpu_to_le16(ee_len);
3308 ext4_ext_try_to_merge(handle, inode, path, ex);
3309 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3311 /* update extent status tree */
3312 err = ext4_zeroout_es(inode, &zero_ex);
3313 /* If we failed at this point, we don't know in which
3314 * state the extent tree exactly is so don't try to fix
3315 * length of the original extent as it may do even more
3323 ex->ee_len = orig_ex.ee_len;
3324 ext4_ext_dirty(handle, inode, path + path->p_depth);
3327 ext4_ext_show_leaf(inode, path);
3332 * ext4_split_extents() splits an extent and mark extent which is covered
3333 * by @map as split_flags indicates
3335 * It may result in splitting the extent into multiple extents (up to three)
3336 * There are three possibilities:
3337 * a> There is no split required
3338 * b> Splits in two extents: Split is happening at either end of the extent
3339 * c> Splits in three extents: Somone is splitting in middle of the extent
3342 static int ext4_split_extent(handle_t *handle,
3343 struct inode *inode,
3344 struct ext4_ext_path **ppath,
3345 struct ext4_map_blocks *map,
3349 struct ext4_ext_path *path = *ppath;
3350 ext4_lblk_t ee_block;
3351 struct ext4_extent *ex;
3352 unsigned int ee_len, depth;
3355 int split_flag1, flags1;
3356 int allocated = map->m_len;
3358 depth = ext_depth(inode);
3359 ex = path[depth].p_ext;
3360 ee_block = le32_to_cpu(ex->ee_block);
3361 ee_len = ext4_ext_get_actual_len(ex);
3362 unwritten = ext4_ext_is_unwritten(ex);
3364 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3365 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3366 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3368 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3369 EXT4_EXT_MARK_UNWRIT2;
3370 if (split_flag & EXT4_EXT_DATA_VALID2)
3371 split_flag1 |= EXT4_EXT_DATA_VALID1;
3372 err = ext4_split_extent_at(handle, inode, ppath,
3373 map->m_lblk + map->m_len, split_flag1, flags1);
3377 allocated = ee_len - (map->m_lblk - ee_block);
3380 * Update path is required because previous ext4_split_extent_at() may
3381 * result in split of original leaf or extent zeroout.
3383 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3385 return PTR_ERR(path);
3386 depth = ext_depth(inode);
3387 ex = path[depth].p_ext;
3389 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3390 (unsigned long) map->m_lblk);
3391 return -EFSCORRUPTED;
3393 unwritten = ext4_ext_is_unwritten(ex);
3396 if (map->m_lblk >= ee_block) {
3397 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3399 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3400 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3401 EXT4_EXT_MARK_UNWRIT2);
3403 err = ext4_split_extent_at(handle, inode, ppath,
3404 map->m_lblk, split_flag1, flags);
3409 ext4_ext_show_leaf(inode, path);
3411 return err ? err : allocated;
3415 * This function is called by ext4_ext_map_blocks() if someone tries to write
3416 * to an unwritten extent. It may result in splitting the unwritten
3417 * extent into multiple extents (up to three - one initialized and two
3419 * There are three possibilities:
3420 * a> There is no split required: Entire extent should be initialized
3421 * b> Splits in two extents: Write is happening at either end of the extent
3422 * c> Splits in three extents: Somone is writing in middle of the extent
3425 * - The extent pointed to by 'path' is unwritten.
3426 * - The extent pointed to by 'path' contains a superset
3427 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3429 * Post-conditions on success:
3430 * - the returned value is the number of blocks beyond map->l_lblk
3431 * that are allocated and initialized.
3432 * It is guaranteed to be >= map->m_len.
3434 static int ext4_ext_convert_to_initialized(handle_t *handle,
3435 struct inode *inode,
3436 struct ext4_map_blocks *map,
3437 struct ext4_ext_path **ppath,
3440 struct ext4_ext_path *path = *ppath;
3441 struct ext4_sb_info *sbi;
3442 struct ext4_extent_header *eh;
3443 struct ext4_map_blocks split_map;
3444 struct ext4_extent zero_ex1, zero_ex2;
3445 struct ext4_extent *ex, *abut_ex;
3446 ext4_lblk_t ee_block, eof_block;
3447 unsigned int ee_len, depth, map_len = map->m_len;
3448 int allocated = 0, max_zeroout = 0;
3450 int split_flag = EXT4_EXT_DATA_VALID2;
3452 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3453 "block %llu, max_blocks %u\n", inode->i_ino,
3454 (unsigned long long)map->m_lblk, map_len);
3456 sbi = EXT4_SB(inode->i_sb);
3457 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3458 >> inode->i_sb->s_blocksize_bits;
3459 if (eof_block < map->m_lblk + map_len)
3460 eof_block = map->m_lblk + map_len;
3462 depth = ext_depth(inode);
3463 eh = path[depth].p_hdr;
3464 ex = path[depth].p_ext;
3465 ee_block = le32_to_cpu(ex->ee_block);
3466 ee_len = ext4_ext_get_actual_len(ex);
3467 zero_ex1.ee_len = 0;
3468 zero_ex2.ee_len = 0;
3470 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3472 /* Pre-conditions */
3473 BUG_ON(!ext4_ext_is_unwritten(ex));
3474 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3477 * Attempt to transfer newly initialized blocks from the currently
3478 * unwritten extent to its neighbor. This is much cheaper
3479 * than an insertion followed by a merge as those involve costly
3480 * memmove() calls. Transferring to the left is the common case in
3481 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3482 * followed by append writes.
3484 * Limitations of the current logic:
3485 * - L1: we do not deal with writes covering the whole extent.
3486 * This would require removing the extent if the transfer
3488 * - L2: we only attempt to merge with an extent stored in the
3489 * same extent tree node.
3491 if ((map->m_lblk == ee_block) &&
3492 /* See if we can merge left */
3493 (map_len < ee_len) && /*L1*/
3494 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3495 ext4_lblk_t prev_lblk;
3496 ext4_fsblk_t prev_pblk, ee_pblk;
3497 unsigned int prev_len;
3500 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3501 prev_len = ext4_ext_get_actual_len(abut_ex);
3502 prev_pblk = ext4_ext_pblock(abut_ex);
3503 ee_pblk = ext4_ext_pblock(ex);
3506 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3507 * upon those conditions:
3508 * - C1: abut_ex is initialized,
3509 * - C2: abut_ex is logically abutting ex,
3510 * - C3: abut_ex is physically abutting ex,
3511 * - C4: abut_ex can receive the additional blocks without
3512 * overflowing the (initialized) length limit.
3514 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3515 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3516 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3517 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3518 err = ext4_ext_get_access(handle, inode, path + depth);
3522 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3525 /* Shift the start of ex by 'map_len' blocks */
3526 ex->ee_block = cpu_to_le32(ee_block + map_len);
3527 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3528 ex->ee_len = cpu_to_le16(ee_len - map_len);
3529 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3531 /* Extend abut_ex by 'map_len' blocks */
3532 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3534 /* Result: number of initialized blocks past m_lblk */
3535 allocated = map_len;
3537 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3538 (map_len < ee_len) && /*L1*/
3539 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3540 /* See if we can merge right */
3541 ext4_lblk_t next_lblk;
3542 ext4_fsblk_t next_pblk, ee_pblk;
3543 unsigned int next_len;
3546 next_lblk = le32_to_cpu(abut_ex->ee_block);
3547 next_len = ext4_ext_get_actual_len(abut_ex);
3548 next_pblk = ext4_ext_pblock(abut_ex);
3549 ee_pblk = ext4_ext_pblock(ex);
3552 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3553 * upon those conditions:
3554 * - C1: abut_ex is initialized,
3555 * - C2: abut_ex is logically abutting ex,
3556 * - C3: abut_ex is physically abutting ex,
3557 * - C4: abut_ex can receive the additional blocks without
3558 * overflowing the (initialized) length limit.
3560 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3561 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3562 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3563 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3564 err = ext4_ext_get_access(handle, inode, path + depth);
3568 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3571 /* Shift the start of abut_ex by 'map_len' blocks */
3572 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3573 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3574 ex->ee_len = cpu_to_le16(ee_len - map_len);
3575 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3577 /* Extend abut_ex by 'map_len' blocks */
3578 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3580 /* Result: number of initialized blocks past m_lblk */
3581 allocated = map_len;
3585 /* Mark the block containing both extents as dirty */
3586 ext4_ext_dirty(handle, inode, path + depth);
3588 /* Update path to point to the right extent */
3589 path[depth].p_ext = abut_ex;
3592 allocated = ee_len - (map->m_lblk - ee_block);
3594 WARN_ON(map->m_lblk < ee_block);
3596 * It is safe to convert extent to initialized via explicit
3597 * zeroout only if extent is fully inside i_size or new_size.
3599 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3601 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3602 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3603 (inode->i_sb->s_blocksize_bits - 10);
3605 if (ext4_encrypted_inode(inode))
3610 * 1. split the extent into three extents.
3611 * 2. split the extent into two extents, zeroout the head of the first
3613 * 3. split the extent into two extents, zeroout the tail of the second
3615 * 4. split the extent into two extents with out zeroout.
3616 * 5. no splitting needed, just possibly zeroout the head and / or the
3617 * tail of the extent.
3619 split_map.m_lblk = map->m_lblk;
3620 split_map.m_len = map->m_len;
3622 if (max_zeroout && (allocated > split_map.m_len)) {
3623 if (allocated <= max_zeroout) {
3626 cpu_to_le32(split_map.m_lblk +
3629 cpu_to_le16(allocated - split_map.m_len);
3630 ext4_ext_store_pblock(&zero_ex1,
3631 ext4_ext_pblock(ex) + split_map.m_lblk +
3632 split_map.m_len - ee_block);
3633 err = ext4_ext_zeroout(inode, &zero_ex1);
3636 split_map.m_len = allocated;
3638 if (split_map.m_lblk - ee_block + split_map.m_len <
3641 if (split_map.m_lblk != ee_block) {
3642 zero_ex2.ee_block = ex->ee_block;
3643 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3645 ext4_ext_store_pblock(&zero_ex2,
3646 ext4_ext_pblock(ex));
3647 err = ext4_ext_zeroout(inode, &zero_ex2);
3652 split_map.m_len += split_map.m_lblk - ee_block;
3653 split_map.m_lblk = ee_block;
3654 allocated = map->m_len;
3658 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3663 /* If we have gotten a failure, don't zero out status tree */
3665 err = ext4_zeroout_es(inode, &zero_ex1);
3667 err = ext4_zeroout_es(inode, &zero_ex2);
3669 return err ? err : allocated;
3673 * This function is called by ext4_ext_map_blocks() from
3674 * ext4_get_blocks_dio_write() when DIO to write
3675 * to an unwritten extent.
3677 * Writing to an unwritten extent may result in splitting the unwritten
3678 * extent into multiple initialized/unwritten extents (up to three)
3679 * There are three possibilities:
3680 * a> There is no split required: Entire extent should be unwritten
3681 * b> Splits in two extents: Write is happening at either end of the extent
3682 * c> Splits in three extents: Somone is writing in middle of the extent
3684 * This works the same way in the case of initialized -> unwritten conversion.
3686 * One of more index blocks maybe needed if the extent tree grow after
3687 * the unwritten extent split. To prevent ENOSPC occur at the IO
3688 * complete, we need to split the unwritten extent before DIO submit
3689 * the IO. The unwritten extent called at this time will be split
3690 * into three unwritten extent(at most). After IO complete, the part
3691 * being filled will be convert to initialized by the end_io callback function
3692 * via ext4_convert_unwritten_extents().
3694 * Returns the size of unwritten extent to be written on success.
3696 static int ext4_split_convert_extents(handle_t *handle,
3697 struct inode *inode,
3698 struct ext4_map_blocks *map,
3699 struct ext4_ext_path **ppath,
3702 struct ext4_ext_path *path = *ppath;
3703 ext4_lblk_t eof_block;
3704 ext4_lblk_t ee_block;
3705 struct ext4_extent *ex;
3706 unsigned int ee_len;
3707 int split_flag = 0, depth;
3709 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3710 __func__, inode->i_ino,
3711 (unsigned long long)map->m_lblk, map->m_len);
3713 eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3714 >> inode->i_sb->s_blocksize_bits;
3715 if (eof_block < map->m_lblk + map->m_len)
3716 eof_block = map->m_lblk + map->m_len;
3718 * It is safe to convert extent to initialized via explicit
3719 * zeroout only if extent is fully insde i_size or new_size.
3721 depth = ext_depth(inode);
3722 ex = path[depth].p_ext;
3723 ee_block = le32_to_cpu(ex->ee_block);
3724 ee_len = ext4_ext_get_actual_len(ex);
3726 /* Convert to unwritten */
3727 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3728 split_flag |= EXT4_EXT_DATA_VALID1;
3729 /* Convert to initialized */
3730 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3731 split_flag |= ee_block + ee_len <= eof_block ?
3732 EXT4_EXT_MAY_ZEROOUT : 0;
3733 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3735 flags |= EXT4_GET_BLOCKS_PRE_IO;
3736 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3739 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3740 struct inode *inode,
3741 struct ext4_map_blocks *map,
3742 struct ext4_ext_path **ppath)
3744 struct ext4_ext_path *path = *ppath;
3745 struct ext4_extent *ex;
3746 ext4_lblk_t ee_block;
3747 unsigned int ee_len;
3751 depth = ext_depth(inode);
3752 ex = path[depth].p_ext;
3753 ee_block = le32_to_cpu(ex->ee_block);
3754 ee_len = ext4_ext_get_actual_len(ex);
3756 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3757 "block %llu, max_blocks %u\n", inode->i_ino,
3758 (unsigned long long)ee_block, ee_len);
3760 /* If extent is larger than requested it is a clear sign that we still
3761 * have some extent state machine issues left. So extent_split is still
3763 * TODO: Once all related issues will be fixed this situation should be
3766 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3767 #ifdef CONFIG_EXT4_DEBUG
3768 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3769 " len %u; IO logical block %llu, len %u",
3770 inode->i_ino, (unsigned long long)ee_block, ee_len,
3771 (unsigned long long)map->m_lblk, map->m_len);
3773 err = ext4_split_convert_extents(handle, inode, map, ppath,
3774 EXT4_GET_BLOCKS_CONVERT);
3777 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3779 return PTR_ERR(path);
3780 depth = ext_depth(inode);
3781 ex = path[depth].p_ext;
3784 err = ext4_ext_get_access(handle, inode, path + depth);
3787 /* first mark the extent as initialized */
3788 ext4_ext_mark_initialized(ex);
3790 /* note: ext4_ext_correct_indexes() isn't needed here because
3791 * borders are not changed
3793 ext4_ext_try_to_merge(handle, inode, path, ex);
3795 /* Mark modified extent as dirty */
3796 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3798 ext4_ext_show_leaf(inode, path);
3803 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3805 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3807 struct ext4_ext_path *path,
3811 struct ext4_extent_header *eh;
3812 struct ext4_extent *last_ex;
3814 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3817 depth = ext_depth(inode);
3818 eh = path[depth].p_hdr;
3821 * We're going to remove EOFBLOCKS_FL entirely in future so we
3822 * do not care for this case anymore. Simply remove the flag
3823 * if there are no extents.
3825 if (unlikely(!eh->eh_entries))
3827 last_ex = EXT_LAST_EXTENT(eh);
3829 * We should clear the EOFBLOCKS_FL flag if we are writing the
3830 * last block in the last extent in the file. We test this by
3831 * first checking to see if the caller to
3832 * ext4_ext_get_blocks() was interested in the last block (or
3833 * a block beyond the last block) in the current extent. If
3834 * this turns out to be false, we can bail out from this
3835 * function immediately.
3837 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3838 ext4_ext_get_actual_len(last_ex))
3841 * If the caller does appear to be planning to write at or
3842 * beyond the end of the current extent, we then test to see
3843 * if the current extent is the last extent in the file, by
3844 * checking to make sure it was reached via the rightmost node
3845 * at each level of the tree.
3847 for (i = depth-1; i >= 0; i--)
3848 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3851 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3852 return ext4_mark_inode_dirty(handle, inode);
3856 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3858 * Return 1 if there is a delalloc block in the range, otherwise 0.
3860 int ext4_find_delalloc_range(struct inode *inode,
3861 ext4_lblk_t lblk_start,
3862 ext4_lblk_t lblk_end)
3864 struct extent_status es;
3866 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3868 return 0; /* there is no delay extent in this tree */
3869 else if (es.es_lblk <= lblk_start &&
3870 lblk_start < es.es_lblk + es.es_len)
3872 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3878 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3881 ext4_lblk_t lblk_start, lblk_end;
3882 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3883 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3885 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3889 * Determines how many complete clusters (out of those specified by the 'map')
3890 * are under delalloc and were reserved quota for.
3891 * This function is called when we are writing out the blocks that were
3892 * originally written with their allocation delayed, but then the space was
3893 * allocated using fallocate() before the delayed allocation could be resolved.
3894 * The cases to look for are:
3895 * ('=' indicated delayed allocated blocks
3896 * '-' indicates non-delayed allocated blocks)
3897 * (a) partial clusters towards beginning and/or end outside of allocated range
3898 * are not delalloc'ed.
3900 * |----c---=|====c====|====c====|===-c----|
3901 * |++++++ allocated ++++++|
3902 * ==> 4 complete clusters in above example
3904 * (b) partial cluster (outside of allocated range) towards either end is
3905 * marked for delayed allocation. In this case, we will exclude that
3908 * |----====c========|========c========|
3909 * |++++++ allocated ++++++|
3910 * ==> 1 complete clusters in above example
3913 * |================c================|
3914 * |++++++ allocated ++++++|
3915 * ==> 0 complete clusters in above example
3917 * The ext4_da_update_reserve_space will be called only if we
3918 * determine here that there were some "entire" clusters that span
3919 * this 'allocated' range.
3920 * In the non-bigalloc case, this function will just end up returning num_blks
3921 * without ever calling ext4_find_delalloc_range.
3924 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3925 unsigned int num_blks)
3927 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3928 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3929 ext4_lblk_t lblk_from, lblk_to, c_offset;
3930 unsigned int allocated_clusters = 0;
3932 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3933 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3935 /* max possible clusters for this allocation */
3936 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3938 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3940 /* Check towards left side */
3941 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3943 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3944 lblk_to = lblk_from + c_offset - 1;
3946 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3947 allocated_clusters--;
3950 /* Now check towards right. */
3951 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3952 if (allocated_clusters && c_offset) {
3953 lblk_from = lblk_start + num_blks;
3954 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3956 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3957 allocated_clusters--;
3960 return allocated_clusters;
3964 convert_initialized_extent(handle_t *handle, struct inode *inode,
3965 struct ext4_map_blocks *map,
3966 struct ext4_ext_path **ppath,
3967 unsigned int allocated)
3969 struct ext4_ext_path *path = *ppath;
3970 struct ext4_extent *ex;
3971 ext4_lblk_t ee_block;
3972 unsigned int ee_len;
3977 * Make sure that the extent is no bigger than we support with
3980 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3981 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3983 depth = ext_depth(inode);
3984 ex = path[depth].p_ext;
3985 ee_block = le32_to_cpu(ex->ee_block);
3986 ee_len = ext4_ext_get_actual_len(ex);
3988 ext_debug("%s: inode %lu, logical"
3989 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3990 (unsigned long long)ee_block, ee_len);
3992 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3993 err = ext4_split_convert_extents(handle, inode, map, ppath,
3994 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3997 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3999 return PTR_ERR(path);
4000 depth = ext_depth(inode);
4001 ex = path[depth].p_ext;
4003 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
4004 (unsigned long) map->m_lblk);
4005 return -EFSCORRUPTED;
4009 err = ext4_ext_get_access(handle, inode, path + depth);
4012 /* first mark the extent as unwritten */
4013 ext4_ext_mark_unwritten(ex);
4015 /* note: ext4_ext_correct_indexes() isn't needed here because
4016 * borders are not changed
4018 ext4_ext_try_to_merge(handle, inode, path, ex);
4020 /* Mark modified extent as dirty */
4021 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4024 ext4_ext_show_leaf(inode, path);
4026 ext4_update_inode_fsync_trans(handle, inode, 1);
4027 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4030 map->m_flags |= EXT4_MAP_UNWRITTEN;
4031 if (allocated > map->m_len)
4032 allocated = map->m_len;
4033 map->m_len = allocated;
4038 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4039 struct ext4_map_blocks *map,
4040 struct ext4_ext_path **ppath, int flags,
4041 unsigned int allocated, ext4_fsblk_t newblock)
4043 struct ext4_ext_path *path = *ppath;
4047 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4048 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4049 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4051 ext4_ext_show_leaf(inode, path);
4054 * When writing into unwritten space, we should not fail to
4055 * allocate metadata blocks for the new extent block if needed.
4057 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4059 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4060 allocated, newblock);
4062 /* get_block() before submit the IO, split the extent */
4063 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4064 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4065 flags | EXT4_GET_BLOCKS_CONVERT);
4068 map->m_flags |= EXT4_MAP_UNWRITTEN;
4071 /* IO end_io complete, convert the filled extent to written */
4072 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4073 if (flags & EXT4_GET_BLOCKS_ZERO) {
4074 if (allocated > map->m_len)
4075 allocated = map->m_len;
4076 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4081 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4084 ext4_update_inode_fsync_trans(handle, inode, 1);
4085 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4089 map->m_flags |= EXT4_MAP_MAPPED;
4090 map->m_pblk = newblock;
4091 if (allocated > map->m_len)
4092 allocated = map->m_len;
4093 map->m_len = allocated;
4096 /* buffered IO case */
4098 * repeat fallocate creation request
4099 * we already have an unwritten extent
4101 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4102 map->m_flags |= EXT4_MAP_UNWRITTEN;
4106 /* buffered READ or buffered write_begin() lookup */
4107 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4109 * We have blocks reserved already. We
4110 * return allocated blocks so that delalloc
4111 * won't do block reservation for us. But
4112 * the buffer head will be unmapped so that
4113 * a read from the block returns 0s.
4115 map->m_flags |= EXT4_MAP_UNWRITTEN;
4119 /* buffered write, writepage time, convert*/
4120 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4122 ext4_update_inode_fsync_trans(handle, inode, 1);
4129 map->m_flags |= EXT4_MAP_NEW;
4131 * if we allocated more blocks than requested
4132 * we need to make sure we unmap the extra block
4133 * allocated. The actual needed block will get
4134 * unmapped later when we find the buffer_head marked
4137 if (allocated > map->m_len) {
4138 clean_bdev_aliases(inode->i_sb->s_bdev, newblock + map->m_len,
4139 allocated - map->m_len);
4140 allocated = map->m_len;
4142 map->m_len = allocated;
4145 * If we have done fallocate with the offset that is already
4146 * delayed allocated, we would have block reservation
4147 * and quota reservation done in the delayed write path.
4148 * But fallocate would have already updated quota and block
4149 * count for this offset. So cancel these reservation
4151 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4152 unsigned int reserved_clusters;
4153 reserved_clusters = get_reserved_cluster_alloc(inode,
4154 map->m_lblk, map->m_len);
4155 if (reserved_clusters)
4156 ext4_da_update_reserve_space(inode,
4162 map->m_flags |= EXT4_MAP_MAPPED;
4163 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4164 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4170 if (allocated > map->m_len)
4171 allocated = map->m_len;
4172 ext4_ext_show_leaf(inode, path);
4173 map->m_pblk = newblock;
4174 map->m_len = allocated;
4176 return err ? err : allocated;
4180 * get_implied_cluster_alloc - check to see if the requested
4181 * allocation (in the map structure) overlaps with a cluster already
4182 * allocated in an extent.
4183 * @sb The filesystem superblock structure
4184 * @map The requested lblk->pblk mapping
4185 * @ex The extent structure which might contain an implied
4186 * cluster allocation
4188 * This function is called by ext4_ext_map_blocks() after we failed to
4189 * find blocks that were already in the inode's extent tree. Hence,
4190 * we know that the beginning of the requested region cannot overlap
4191 * the extent from the inode's extent tree. There are three cases we
4192 * want to catch. The first is this case:
4194 * |--- cluster # N--|
4195 * |--- extent ---| |---- requested region ---|
4198 * The second case that we need to test for is this one:
4200 * |--------- cluster # N ----------------|
4201 * |--- requested region --| |------- extent ----|
4202 * |=======================|
4204 * The third case is when the requested region lies between two extents
4205 * within the same cluster:
4206 * |------------- cluster # N-------------|
4207 * |----- ex -----| |---- ex_right ----|
4208 * |------ requested region ------|
4209 * |================|
4211 * In each of the above cases, we need to set the map->m_pblk and
4212 * map->m_len so it corresponds to the return the extent labelled as
4213 * "|====|" from cluster #N, since it is already in use for data in
4214 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4215 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4216 * as a new "allocated" block region. Otherwise, we will return 0 and
4217 * ext4_ext_map_blocks() will then allocate one or more new clusters
4218 * by calling ext4_mb_new_blocks().
4220 static int get_implied_cluster_alloc(struct super_block *sb,
4221 struct ext4_map_blocks *map,
4222 struct ext4_extent *ex,
4223 struct ext4_ext_path *path)
4225 struct ext4_sb_info *sbi = EXT4_SB(sb);
4226 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4227 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4228 ext4_lblk_t rr_cluster_start;
4229 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4230 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4231 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4233 /* The extent passed in that we are trying to match */
4234 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4235 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4237 /* The requested region passed into ext4_map_blocks() */
4238 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4240 if ((rr_cluster_start == ex_cluster_end) ||
4241 (rr_cluster_start == ex_cluster_start)) {
4242 if (rr_cluster_start == ex_cluster_end)
4243 ee_start += ee_len - 1;
4244 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4245 map->m_len = min(map->m_len,
4246 (unsigned) sbi->s_cluster_ratio - c_offset);
4248 * Check for and handle this case:
4250 * |--------- cluster # N-------------|
4251 * |------- extent ----|
4252 * |--- requested region ---|
4256 if (map->m_lblk < ee_block)
4257 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4260 * Check for the case where there is already another allocated
4261 * block to the right of 'ex' but before the end of the cluster.
4263 * |------------- cluster # N-------------|
4264 * |----- ex -----| |---- ex_right ----|
4265 * |------ requested region ------|
4266 * |================|
4268 if (map->m_lblk > ee_block) {
4269 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4270 map->m_len = min(map->m_len, next - map->m_lblk);
4273 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4277 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4283 * Block allocation/map/preallocation routine for extents based files
4286 * Need to be called with
4287 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4288 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4290 * return > 0, number of of blocks already mapped/allocated
4291 * if create == 0 and these are pre-allocated blocks
4292 * buffer head is unmapped
4293 * otherwise blocks are mapped
4295 * return = 0, if plain look up failed (blocks have not been allocated)
4296 * buffer head is unmapped
4298 * return < 0, error case.
4300 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4301 struct ext4_map_blocks *map, int flags)
4303 struct ext4_ext_path *path = NULL;
4304 struct ext4_extent newex, *ex, *ex2;
4305 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4306 ext4_fsblk_t newblock = 0;
4307 int free_on_err = 0, err = 0, depth, ret;
4308 unsigned int allocated = 0, offset = 0;
4309 unsigned int allocated_clusters = 0;
4310 struct ext4_allocation_request ar;
4311 ext4_lblk_t cluster_offset;
4312 bool map_from_cluster = false;
4314 ext_debug("blocks %u/%u requested for inode %lu\n",
4315 map->m_lblk, map->m_len, inode->i_ino);
4316 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4318 /* find extent for this block */
4319 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4321 err = PTR_ERR(path);
4326 depth = ext_depth(inode);
4329 * consistent leaf must not be empty;
4330 * this situation is possible, though, _during_ tree modification;
4331 * this is why assert can't be put in ext4_find_extent()
4333 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4334 EXT4_ERROR_INODE(inode, "bad extent address "
4335 "lblock: %lu, depth: %d pblock %lld",
4336 (unsigned long) map->m_lblk, depth,
4337 path[depth].p_block);
4338 err = -EFSCORRUPTED;
4342 ex = path[depth].p_ext;
4344 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4345 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4346 unsigned short ee_len;
4350 * unwritten extents are treated as holes, except that
4351 * we split out initialized portions during a write.
4353 ee_len = ext4_ext_get_actual_len(ex);
4355 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4357 /* if found extent covers block, simply return it */
4358 if (in_range(map->m_lblk, ee_block, ee_len)) {
4359 newblock = map->m_lblk - ee_block + ee_start;
4360 /* number of remaining blocks in the extent */
4361 allocated = ee_len - (map->m_lblk - ee_block);
4362 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4363 ee_block, ee_len, newblock);
4366 * If the extent is initialized check whether the
4367 * caller wants to convert it to unwritten.
4369 if ((!ext4_ext_is_unwritten(ex)) &&
4370 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4371 allocated = convert_initialized_extent(
4372 handle, inode, map, &path,
4375 } else if (!ext4_ext_is_unwritten(ex))
4378 ret = ext4_ext_handle_unwritten_extents(
4379 handle, inode, map, &path, flags,
4380 allocated, newblock);
4390 * requested block isn't allocated yet;
4391 * we couldn't try to create block if create flag is zero
4393 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4394 ext4_lblk_t hole_start, hole_len;
4396 hole_start = map->m_lblk;
4397 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4399 * put just found gap into cache to speed up
4400 * subsequent requests
4402 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4404 /* Update hole_len to reflect hole size after map->m_lblk */
4405 if (hole_start != map->m_lblk)
4406 hole_len -= map->m_lblk - hole_start;
4408 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4414 * Okay, we need to do block allocation.
4416 newex.ee_block = cpu_to_le32(map->m_lblk);
4417 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4420 * If we are doing bigalloc, check to see if the extent returned
4421 * by ext4_find_extent() implies a cluster we can use.
4423 if (cluster_offset && ex &&
4424 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4425 ar.len = allocated = map->m_len;
4426 newblock = map->m_pblk;
4427 map_from_cluster = true;
4428 goto got_allocated_blocks;
4431 /* find neighbour allocated blocks */
4432 ar.lleft = map->m_lblk;
4433 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4436 ar.lright = map->m_lblk;
4438 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4442 /* Check if the extent after searching to the right implies a
4443 * cluster we can use. */
4444 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4445 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4446 ar.len = allocated = map->m_len;
4447 newblock = map->m_pblk;
4448 map_from_cluster = true;
4449 goto got_allocated_blocks;
4453 * See if request is beyond maximum number of blocks we can have in
4454 * a single extent. For an initialized extent this limit is
4455 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4456 * EXT_UNWRITTEN_MAX_LEN.
4458 if (map->m_len > EXT_INIT_MAX_LEN &&
4459 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4460 map->m_len = EXT_INIT_MAX_LEN;
4461 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4462 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4463 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4465 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4466 newex.ee_len = cpu_to_le16(map->m_len);
4467 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4469 allocated = ext4_ext_get_actual_len(&newex);
4471 allocated = map->m_len;
4473 /* allocate new block */
4475 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4476 ar.logical = map->m_lblk;
4478 * We calculate the offset from the beginning of the cluster
4479 * for the logical block number, since when we allocate a
4480 * physical cluster, the physical block should start at the
4481 * same offset from the beginning of the cluster. This is
4482 * needed so that future calls to get_implied_cluster_alloc()
4485 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4486 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4488 ar.logical -= offset;
4489 if (S_ISREG(inode->i_mode))
4490 ar.flags = EXT4_MB_HINT_DATA;
4492 /* disable in-core preallocation for non-regular files */
4494 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4495 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4496 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4497 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4498 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4499 ar.flags |= EXT4_MB_USE_RESERVED;
4500 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4503 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4504 ar.goal, newblock, allocated);
4506 allocated_clusters = ar.len;
4507 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4508 if (ar.len > allocated)
4511 got_allocated_blocks:
4512 /* try to insert new extent into found leaf and return */
4513 ext4_ext_store_pblock(&newex, newblock + offset);
4514 newex.ee_len = cpu_to_le16(ar.len);
4515 /* Mark unwritten */
4516 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4517 ext4_ext_mark_unwritten(&newex);
4518 map->m_flags |= EXT4_MAP_UNWRITTEN;
4522 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4523 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4526 err = ext4_ext_insert_extent(handle, inode, &path,
4529 if (err && free_on_err) {
4530 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4531 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4532 /* free data blocks we just allocated */
4533 /* not a good idea to call discard here directly,
4534 * but otherwise we'd need to call it every free() */
4535 ext4_discard_preallocations(inode);
4536 ext4_free_blocks(handle, inode, NULL, newblock,
4537 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4541 /* previous routine could use block we allocated */
4542 newblock = ext4_ext_pblock(&newex);
4543 allocated = ext4_ext_get_actual_len(&newex);
4544 if (allocated > map->m_len)
4545 allocated = map->m_len;
4546 map->m_flags |= EXT4_MAP_NEW;
4549 * Update reserved blocks/metadata blocks after successful
4550 * block allocation which had been deferred till now.
4552 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4553 unsigned int reserved_clusters;
4555 * Check how many clusters we had reserved this allocated range
4557 reserved_clusters = get_reserved_cluster_alloc(inode,
4558 map->m_lblk, allocated);
4559 if (!map_from_cluster) {
4560 BUG_ON(allocated_clusters < reserved_clusters);
4561 if (reserved_clusters < allocated_clusters) {
4562 struct ext4_inode_info *ei = EXT4_I(inode);
4563 int reservation = allocated_clusters -
4566 * It seems we claimed few clusters outside of
4567 * the range of this allocation. We should give
4568 * it back to the reservation pool. This can
4569 * happen in the following case:
4571 * * Suppose s_cluster_ratio is 4 (i.e., each
4572 * cluster has 4 blocks. Thus, the clusters
4573 * are [0-3],[4-7],[8-11]...
4574 * * First comes delayed allocation write for
4575 * logical blocks 10 & 11. Since there were no
4576 * previous delayed allocated blocks in the
4577 * range [8-11], we would reserve 1 cluster
4579 * * Next comes write for logical blocks 3 to 8.
4580 * In this case, we will reserve 2 clusters
4581 * (for [0-3] and [4-7]; and not for [8-11] as
4582 * that range has a delayed allocated blocks.
4583 * Thus total reserved clusters now becomes 3.
4584 * * Now, during the delayed allocation writeout
4585 * time, we will first write blocks [3-8] and
4586 * allocate 3 clusters for writing these
4587 * blocks. Also, we would claim all these
4588 * three clusters above.
4589 * * Now when we come here to writeout the
4590 * blocks [10-11], we would expect to claim
4591 * the reservation of 1 cluster we had made
4592 * (and we would claim it since there are no
4593 * more delayed allocated blocks in the range
4594 * [8-11]. But our reserved cluster count had
4595 * already gone to 0.
4597 * Thus, at the step 4 above when we determine
4598 * that there are still some unwritten delayed
4599 * allocated blocks outside of our current
4600 * block range, we should increment the
4601 * reserved clusters count so that when the
4602 * remaining blocks finally gets written, we
4605 dquot_reserve_block(inode,
4606 EXT4_C2B(sbi, reservation));
4607 spin_lock(&ei->i_block_reservation_lock);
4608 ei->i_reserved_data_blocks += reservation;
4609 spin_unlock(&ei->i_block_reservation_lock);
4612 * We will claim quota for all newly allocated blocks.
4613 * We're updating the reserved space *after* the
4614 * correction above so we do not accidentally free
4615 * all the metadata reservation because we might
4616 * actually need it later on.
4618 ext4_da_update_reserve_space(inode, allocated_clusters,
4624 * Cache the extent and update transaction to commit on fdatasync only
4625 * when it is _not_ an unwritten extent.
4627 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4628 ext4_update_inode_fsync_trans(handle, inode, 1);
4630 ext4_update_inode_fsync_trans(handle, inode, 0);
4632 if (allocated > map->m_len)
4633 allocated = map->m_len;
4634 ext4_ext_show_leaf(inode, path);
4635 map->m_flags |= EXT4_MAP_MAPPED;
4636 map->m_pblk = newblock;
4637 map->m_len = allocated;
4639 ext4_ext_drop_refs(path);
4642 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4643 err ? err : allocated);
4644 return err ? err : allocated;
4647 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4649 struct super_block *sb = inode->i_sb;
4650 ext4_lblk_t last_block;
4654 * TODO: optimization is possible here.
4655 * Probably we need not scan at all,
4656 * because page truncation is enough.
4659 /* we have to know where to truncate from in crash case */
4660 EXT4_I(inode)->i_disksize = inode->i_size;
4661 err = ext4_mark_inode_dirty(handle, inode);
4665 last_block = (inode->i_size + sb->s_blocksize - 1)
4666 >> EXT4_BLOCK_SIZE_BITS(sb);
4668 err = ext4_es_remove_extent(inode, last_block,
4669 EXT_MAX_BLOCKS - last_block);
4670 if (err == -ENOMEM) {
4672 congestion_wait(BLK_RW_ASYNC, HZ/50);
4677 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4680 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4681 ext4_lblk_t len, loff_t new_size,
4684 struct inode *inode = file_inode(file);
4690 struct ext4_map_blocks map;
4691 unsigned int credits;
4694 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4695 map.m_lblk = offset;
4698 * Don't normalize the request if it can fit in one extent so
4699 * that it doesn't get unnecessarily split into multiple
4702 if (len <= EXT_UNWRITTEN_MAX_LEN)
4703 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4706 * credits to insert 1 extent into extent tree
4708 credits = ext4_chunk_trans_blocks(inode, len);
4709 depth = ext_depth(inode);
4712 while (ret >= 0 && len) {
4714 * Recalculate credits when extent tree depth changes.
4716 if (depth != ext_depth(inode)) {
4717 credits = ext4_chunk_trans_blocks(inode, len);
4718 depth = ext_depth(inode);
4721 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4723 if (IS_ERR(handle)) {
4724 ret = PTR_ERR(handle);
4727 ret = ext4_map_blocks(handle, inode, &map, flags);
4729 ext4_debug("inode #%lu: block %u: len %u: "
4730 "ext4_ext_map_blocks returned %d",
4731 inode->i_ino, map.m_lblk,
4733 ext4_mark_inode_dirty(handle, inode);
4734 ret2 = ext4_journal_stop(handle);
4738 map.m_len = len = len - ret;
4739 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4740 inode->i_ctime = current_time(inode);
4742 if (epos > new_size)
4744 if (ext4_update_inode_size(inode, epos) & 0x1)
4745 inode->i_mtime = inode->i_ctime;
4747 if (epos > inode->i_size)
4748 ext4_set_inode_flag(inode,
4749 EXT4_INODE_EOFBLOCKS);
4751 ext4_mark_inode_dirty(handle, inode);
4752 ext4_update_inode_fsync_trans(handle, inode, 1);
4753 ret2 = ext4_journal_stop(handle);
4757 if (ret == -ENOSPC &&
4758 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4763 return ret > 0 ? ret2 : ret;
4766 static long ext4_zero_range(struct file *file, loff_t offset,
4767 loff_t len, int mode)
4769 struct inode *inode = file_inode(file);
4770 handle_t *handle = NULL;
4771 unsigned int max_blocks;
4772 loff_t new_size = 0;
4776 int partial_begin, partial_end;
4779 unsigned int blkbits = inode->i_blkbits;
4781 trace_ext4_zero_range(inode, offset, len, mode);
4783 if (!S_ISREG(inode->i_mode))
4786 /* Call ext4_force_commit to flush all data in case of data=journal. */
4787 if (ext4_should_journal_data(inode)) {
4788 ret = ext4_force_commit(inode->i_sb);
4794 * Round up offset. This is not fallocate, we neet to zero out
4795 * blocks, so convert interior block aligned part of the range to
4796 * unwritten and possibly manually zero out unaligned parts of the
4799 start = round_up(offset, 1 << blkbits);
4800 end = round_down((offset + len), 1 << blkbits);
4802 if (start < offset || end > offset + len)
4804 partial_begin = offset & ((1 << blkbits) - 1);
4805 partial_end = (offset + len) & ((1 << blkbits) - 1);
4807 lblk = start >> blkbits;
4808 max_blocks = (end >> blkbits);
4809 if (max_blocks < lblk)
4817 * Indirect files do not support unwritten extnets
4819 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4824 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4825 (offset + len > i_size_read(inode) ||
4826 offset + len > EXT4_I(inode)->i_disksize)) {
4827 new_size = offset + len;
4828 ret = inode_newsize_ok(inode, new_size);
4833 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4834 if (mode & FALLOC_FL_KEEP_SIZE)
4835 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4837 /* Wait all existing dio workers, newcomers will block on i_mutex */
4838 inode_dio_wait(inode);
4840 /* Preallocate the range including the unaligned edges */
4841 if (partial_begin || partial_end) {
4842 ret = ext4_alloc_file_blocks(file,
4843 round_down(offset, 1 << blkbits) >> blkbits,
4844 (round_up((offset + len), 1 << blkbits) -
4845 round_down(offset, 1 << blkbits)) >> blkbits,
4852 /* Zero range excluding the unaligned edges */
4853 if (max_blocks > 0) {
4854 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4858 * Prevent page faults from reinstantiating pages we have
4859 * released from page cache.
4861 down_write(&EXT4_I(inode)->i_mmap_sem);
4863 ret = ext4_break_layouts(inode);
4865 up_write(&EXT4_I(inode)->i_mmap_sem);
4869 ret = ext4_update_disksize_before_punch(inode, offset, len);
4871 up_write(&EXT4_I(inode)->i_mmap_sem);
4874 /* Now release the pages and zero block aligned part of pages */
4875 truncate_pagecache_range(inode, start, end - 1);
4876 inode->i_mtime = inode->i_ctime = current_time(inode);
4878 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4880 up_write(&EXT4_I(inode)->i_mmap_sem);
4884 if (!partial_begin && !partial_end)
4888 * In worst case we have to writeout two nonadjacent unwritten
4889 * blocks and update the inode
4891 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4892 if (ext4_should_journal_data(inode))
4894 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4895 if (IS_ERR(handle)) {
4896 ret = PTR_ERR(handle);
4897 ext4_std_error(inode->i_sb, ret);
4901 inode->i_mtime = inode->i_ctime = current_time(inode);
4903 ext4_update_inode_size(inode, new_size);
4906 * Mark that we allocate beyond EOF so the subsequent truncate
4907 * can proceed even if the new size is the same as i_size.
4909 if ((offset + len) > i_size_read(inode))
4910 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4912 ext4_mark_inode_dirty(handle, inode);
4914 /* Zero out partial block at the edges of the range */
4915 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4917 ext4_update_inode_fsync_trans(handle, inode, 1);
4919 if (file->f_flags & O_SYNC)
4920 ext4_handle_sync(handle);
4922 ext4_journal_stop(handle);
4924 inode_unlock(inode);
4929 * preallocate space for a file. This implements ext4's fallocate file
4930 * operation, which gets called from sys_fallocate system call.
4931 * For block-mapped files, posix_fallocate should fall back to the method
4932 * of writing zeroes to the required new blocks (the same behavior which is
4933 * expected for file systems which do not support fallocate() system call).
4935 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4937 struct inode *inode = file_inode(file);
4938 loff_t new_size = 0;
4939 unsigned int max_blocks;
4943 unsigned int blkbits = inode->i_blkbits;
4946 * Encrypted inodes can't handle collapse range or insert
4947 * range since we would need to re-encrypt blocks with a
4948 * different IV or XTS tweak (which are based on the logical
4951 * XXX It's not clear why zero range isn't working, but we'll
4952 * leave it disabled for encrypted inodes for now. This is a
4953 * bug we should fix....
4955 if (ext4_encrypted_inode(inode) &&
4956 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4957 FALLOC_FL_ZERO_RANGE)))
4960 /* Return error if mode is not supported */
4961 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4962 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4963 FALLOC_FL_INSERT_RANGE))
4966 if (mode & FALLOC_FL_PUNCH_HOLE)
4967 return ext4_punch_hole(inode, offset, len);
4969 ret = ext4_convert_inline_data(inode);
4973 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4974 return ext4_collapse_range(inode, offset, len);
4976 if (mode & FALLOC_FL_INSERT_RANGE)
4977 return ext4_insert_range(inode, offset, len);
4979 if (mode & FALLOC_FL_ZERO_RANGE)
4980 return ext4_zero_range(file, offset, len, mode);
4982 trace_ext4_fallocate_enter(inode, offset, len, mode);
4983 lblk = offset >> blkbits;
4985 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4986 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4987 if (mode & FALLOC_FL_KEEP_SIZE)
4988 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4993 * We only support preallocation for extent-based files only
4995 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5000 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5001 (offset + len > i_size_read(inode) ||
5002 offset + len > EXT4_I(inode)->i_disksize)) {
5003 new_size = offset + len;
5004 ret = inode_newsize_ok(inode, new_size);
5009 /* Wait all existing dio workers, newcomers will block on i_mutex */
5010 inode_dio_wait(inode);
5012 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
5016 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
5017 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5018 EXT4_I(inode)->i_sync_tid);
5021 inode_unlock(inode);
5022 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
5027 * This function convert a range of blocks to written extents
5028 * The caller of this function will pass the start offset and the size.
5029 * all unwritten extents within this range will be converted to
5032 * This function is called from the direct IO end io call back
5033 * function, to convert the fallocated extents after IO is completed.
5034 * Returns 0 on success.
5036 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5037 loff_t offset, ssize_t len)
5039 unsigned int max_blocks;
5042 struct ext4_map_blocks map;
5043 unsigned int credits, blkbits = inode->i_blkbits;
5045 map.m_lblk = offset >> blkbits;
5046 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
5049 * This is somewhat ugly but the idea is clear: When transaction is
5050 * reserved, everything goes into it. Otherwise we rather start several
5051 * smaller transactions for conversion of each extent separately.
5054 handle = ext4_journal_start_reserved(handle,
5055 EXT4_HT_EXT_CONVERT);
5057 return PTR_ERR(handle);
5061 * credits to insert 1 extent into extent tree
5063 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5065 while (ret >= 0 && ret < max_blocks) {
5067 map.m_len = (max_blocks -= ret);
5069 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5071 if (IS_ERR(handle)) {
5072 ret = PTR_ERR(handle);
5076 ret = ext4_map_blocks(handle, inode, &map,
5077 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5079 ext4_warning(inode->i_sb,
5080 "inode #%lu: block %u: len %u: "
5081 "ext4_ext_map_blocks returned %d",
5082 inode->i_ino, map.m_lblk,
5084 ext4_mark_inode_dirty(handle, inode);
5086 ret2 = ext4_journal_stop(handle);
5087 if (ret <= 0 || ret2)
5091 ret2 = ext4_journal_stop(handle);
5092 return ret > 0 ? ret2 : ret;
5096 * If newes is not existing extent (newes->ec_pblk equals zero) find
5097 * delayed extent at start of newes and update newes accordingly and
5098 * return start of the next delayed extent.
5100 * If newes is existing extent (newes->ec_pblk is not equal zero)
5101 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5102 * extent found. Leave newes unmodified.
5104 static int ext4_find_delayed_extent(struct inode *inode,
5105 struct extent_status *newes)
5107 struct extent_status es;
5108 ext4_lblk_t block, next_del;
5110 if (newes->es_pblk == 0) {
5111 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5112 newes->es_lblk + newes->es_len - 1, &es);
5115 * No extent in extent-tree contains block @newes->es_pblk,
5116 * then the block may stay in 1)a hole or 2)delayed-extent.
5122 if (es.es_lblk > newes->es_lblk) {
5124 newes->es_len = min(es.es_lblk - newes->es_lblk,
5129 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5132 block = newes->es_lblk + newes->es_len;
5133 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5135 next_del = EXT_MAX_BLOCKS;
5137 next_del = es.es_lblk;
5141 /* fiemap flags we can handle specified here */
5142 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5144 static int ext4_xattr_fiemap(struct inode *inode,
5145 struct fiemap_extent_info *fieinfo)
5149 __u32 flags = FIEMAP_EXTENT_LAST;
5150 int blockbits = inode->i_sb->s_blocksize_bits;
5154 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5155 struct ext4_iloc iloc;
5156 int offset; /* offset of xattr in inode */
5158 error = ext4_get_inode_loc(inode, &iloc);
5161 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5162 offset = EXT4_GOOD_OLD_INODE_SIZE +
5163 EXT4_I(inode)->i_extra_isize;
5165 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5166 flags |= FIEMAP_EXTENT_DATA_INLINE;
5168 } else { /* external block */
5169 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5170 length = inode->i_sb->s_blocksize;
5174 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5176 return (error < 0 ? error : 0);
5179 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5180 __u64 start, __u64 len)
5182 ext4_lblk_t start_blk;
5185 if (ext4_has_inline_data(inode)) {
5188 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5195 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5196 error = ext4_ext_precache(inode);
5201 /* fallback to generic here if not in extents fmt */
5202 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5203 return generic_block_fiemap(inode, fieinfo, start, len,
5206 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5209 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5210 error = ext4_xattr_fiemap(inode, fieinfo);
5212 ext4_lblk_t len_blks;
5215 start_blk = start >> inode->i_sb->s_blocksize_bits;
5216 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5217 if (last_blk >= EXT_MAX_BLOCKS)
5218 last_blk = EXT_MAX_BLOCKS-1;
5219 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5222 * Walk the extent tree gathering extent information
5223 * and pushing extents back to the user.
5225 error = ext4_fill_fiemap_extents(inode, start_blk,
5233 * Function to access the path buffer for marking it dirty.
5234 * It also checks if there are sufficient credits left in the journal handle
5238 ext4_access_path(handle_t *handle, struct inode *inode,
5239 struct ext4_ext_path *path)
5243 if (!ext4_handle_valid(handle))
5247 * Check if need to extend journal credits
5248 * 3 for leaf, sb, and inode plus 2 (bmap and group
5249 * descriptor) for each block group; assume two block
5252 if (handle->h_buffer_credits < 7) {
5253 credits = ext4_writepage_trans_blocks(inode);
5254 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5255 /* EAGAIN is success */
5256 if (err && err != -EAGAIN)
5260 err = ext4_ext_get_access(handle, inode, path);
5265 * ext4_ext_shift_path_extents:
5266 * Shift the extents of a path structure lying between path[depth].p_ext
5267 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5268 * if it is right shift or left shift operation.
5271 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5272 struct inode *inode, handle_t *handle,
5273 enum SHIFT_DIRECTION SHIFT)
5276 struct ext4_extent *ex_start, *ex_last;
5278 depth = path->p_depth;
5280 while (depth >= 0) {
5281 if (depth == path->p_depth) {
5282 ex_start = path[depth].p_ext;
5284 return -EFSCORRUPTED;
5286 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5288 err = ext4_access_path(handle, inode, path + depth);
5292 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5295 while (ex_start <= ex_last) {
5296 if (SHIFT == SHIFT_LEFT) {
5297 le32_add_cpu(&ex_start->ee_block,
5299 /* Try to merge to the left. */
5301 EXT_FIRST_EXTENT(path[depth].p_hdr))
5303 ext4_ext_try_to_merge_right(inode,
5304 path, ex_start - 1))
5309 le32_add_cpu(&ex_last->ee_block, shift);
5310 ext4_ext_try_to_merge_right(inode, path,
5315 err = ext4_ext_dirty(handle, inode, path + depth);
5319 if (--depth < 0 || !update)
5323 /* Update index too */
5324 err = ext4_access_path(handle, inode, path + depth);
5328 if (SHIFT == SHIFT_LEFT)
5329 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5331 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5332 err = ext4_ext_dirty(handle, inode, path + depth);
5336 /* we are done if current index is not a starting index */
5337 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5348 * ext4_ext_shift_extents:
5349 * All the extents which lies in the range from @start to the last allocated
5350 * block for the @inode are shifted either towards left or right (depending
5351 * upon @SHIFT) by @shift blocks.
5352 * On success, 0 is returned, error otherwise.
5355 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5356 ext4_lblk_t start, ext4_lblk_t shift,
5357 enum SHIFT_DIRECTION SHIFT)
5359 struct ext4_ext_path *path;
5361 struct ext4_extent *extent;
5362 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5364 /* Let path point to the last extent */
5365 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5368 return PTR_ERR(path);
5370 depth = path->p_depth;
5371 extent = path[depth].p_ext;
5375 stop = le32_to_cpu(extent->ee_block);
5378 * For left shifts, make sure the hole on the left is big enough to
5379 * accommodate the shift. For right shifts, make sure the last extent
5380 * won't be shifted beyond EXT_MAX_BLOCKS.
5382 if (SHIFT == SHIFT_LEFT) {
5383 path = ext4_find_extent(inode, start - 1, &path,
5386 return PTR_ERR(path);
5387 depth = path->p_depth;
5388 extent = path[depth].p_ext;
5390 ex_start = le32_to_cpu(extent->ee_block);
5391 ex_end = le32_to_cpu(extent->ee_block) +
5392 ext4_ext_get_actual_len(extent);
5398 if ((start == ex_start && shift > ex_start) ||
5399 (shift > start - ex_end)) {
5404 if (shift > EXT_MAX_BLOCKS -
5405 (stop + ext4_ext_get_actual_len(extent))) {
5412 * In case of left shift, iterator points to start and it is increased
5413 * till we reach stop. In case of right shift, iterator points to stop
5414 * and it is decreased till we reach start.
5416 if (SHIFT == SHIFT_LEFT)
5422 * Its safe to start updating extents. Start and stop are unsigned, so
5423 * in case of right shift if extent with 0 block is reached, iterator
5424 * becomes NULL to indicate the end of the loop.
5426 while (iterator && start <= stop) {
5427 path = ext4_find_extent(inode, *iterator, &path,
5430 return PTR_ERR(path);
5431 depth = path->p_depth;
5432 extent = path[depth].p_ext;
5434 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5435 (unsigned long) *iterator);
5436 return -EFSCORRUPTED;
5438 if (SHIFT == SHIFT_LEFT && *iterator >
5439 le32_to_cpu(extent->ee_block)) {
5440 /* Hole, move to the next extent */
5441 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5442 path[depth].p_ext++;
5444 *iterator = ext4_ext_next_allocated_block(path);
5449 if (SHIFT == SHIFT_LEFT) {
5450 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5451 *iterator = le32_to_cpu(extent->ee_block) +
5452 ext4_ext_get_actual_len(extent);
5454 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5455 if (le32_to_cpu(extent->ee_block) > 0)
5456 *iterator = le32_to_cpu(extent->ee_block) - 1;
5458 /* Beginning is reached, end of the loop */
5460 /* Update path extent in case we need to stop */
5461 while (le32_to_cpu(extent->ee_block) < start)
5463 path[depth].p_ext = extent;
5465 ret = ext4_ext_shift_path_extents(path, shift, inode,
5471 ext4_ext_drop_refs(path);
5477 * ext4_collapse_range:
5478 * This implements the fallocate's collapse range functionality for ext4
5479 * Returns: 0 and non-zero on error.
5481 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5483 struct super_block *sb = inode->i_sb;
5484 ext4_lblk_t punch_start, punch_stop;
5486 unsigned int credits;
5487 loff_t new_size, ioffset;
5491 * We need to test this early because xfstests assumes that a
5492 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5493 * system does not support collapse range.
5495 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5498 /* Collapse range works only on fs block size aligned offsets. */
5499 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5500 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5503 if (!S_ISREG(inode->i_mode))
5506 trace_ext4_collapse_range(inode, offset, len);
5508 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5509 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5511 /* Call ext4_force_commit to flush all data in case of data=journal. */
5512 if (ext4_should_journal_data(inode)) {
5513 ret = ext4_force_commit(inode->i_sb);
5520 * There is no need to overlap collapse range with EOF, in which case
5521 * it is effectively a truncate operation
5523 if (offset + len >= i_size_read(inode)) {
5528 /* Currently just for extent based files */
5529 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5534 /* Wait for existing dio to complete */
5535 inode_dio_wait(inode);
5538 * Prevent page faults from reinstantiating pages we have released from
5541 down_write(&EXT4_I(inode)->i_mmap_sem);
5543 ret = ext4_break_layouts(inode);
5548 * Need to round down offset to be aligned with page size boundary
5549 * for page size > block size.
5551 ioffset = round_down(offset, PAGE_SIZE);
5553 * Write tail of the last page before removed range since it will get
5554 * removed from the page cache below.
5556 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5560 * Write data that will be shifted to preserve them when discarding
5561 * page cache below. We are also protected from pages becoming dirty
5564 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5568 truncate_pagecache(inode, ioffset);
5570 credits = ext4_writepage_trans_blocks(inode);
5571 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5572 if (IS_ERR(handle)) {
5573 ret = PTR_ERR(handle);
5577 down_write(&EXT4_I(inode)->i_data_sem);
5578 ext4_discard_preallocations(inode);
5580 ret = ext4_es_remove_extent(inode, punch_start,
5581 EXT_MAX_BLOCKS - punch_start);
5583 up_write(&EXT4_I(inode)->i_data_sem);
5587 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5589 up_write(&EXT4_I(inode)->i_data_sem);
5592 ext4_discard_preallocations(inode);
5594 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5595 punch_stop - punch_start, SHIFT_LEFT);
5597 up_write(&EXT4_I(inode)->i_data_sem);
5601 new_size = i_size_read(inode) - len;
5602 i_size_write(inode, new_size);
5603 EXT4_I(inode)->i_disksize = new_size;
5605 up_write(&EXT4_I(inode)->i_data_sem);
5607 ext4_handle_sync(handle);
5608 inode->i_mtime = inode->i_ctime = current_time(inode);
5609 ext4_mark_inode_dirty(handle, inode);
5610 ext4_update_inode_fsync_trans(handle, inode, 1);
5613 ext4_journal_stop(handle);
5615 up_write(&EXT4_I(inode)->i_mmap_sem);
5617 inode_unlock(inode);
5622 * ext4_insert_range:
5623 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5624 * The data blocks starting from @offset to the EOF are shifted by @len
5625 * towards right to create a hole in the @inode. Inode size is increased
5627 * Returns 0 on success, error otherwise.
5629 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5631 struct super_block *sb = inode->i_sb;
5633 struct ext4_ext_path *path;
5634 struct ext4_extent *extent;
5635 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5636 unsigned int credits, ee_len;
5637 int ret = 0, depth, split_flag = 0;
5641 * We need to test this early because xfstests assumes that an
5642 * insert range of (0, 1) will return EOPNOTSUPP if the file
5643 * system does not support insert range.
5645 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5648 /* Insert range works only on fs block size aligned offsets. */
5649 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5650 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5653 if (!S_ISREG(inode->i_mode))
5656 trace_ext4_insert_range(inode, offset, len);
5658 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5659 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5661 /* Call ext4_force_commit to flush all data in case of data=journal */
5662 if (ext4_should_journal_data(inode)) {
5663 ret = ext4_force_commit(inode->i_sb);
5669 /* Currently just for extent based files */
5670 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5675 /* Check for wrap through zero */
5676 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5681 /* Offset should be less than i_size */
5682 if (offset >= i_size_read(inode)) {
5687 /* Wait for existing dio to complete */
5688 inode_dio_wait(inode);
5691 * Prevent page faults from reinstantiating pages we have released from
5694 down_write(&EXT4_I(inode)->i_mmap_sem);
5696 ret = ext4_break_layouts(inode);
5701 * Need to round down to align start offset to page size boundary
5702 * for page size > block size.
5704 ioffset = round_down(offset, PAGE_SIZE);
5705 /* Write out all dirty pages */
5706 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5710 truncate_pagecache(inode, ioffset);
5712 credits = ext4_writepage_trans_blocks(inode);
5713 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5714 if (IS_ERR(handle)) {
5715 ret = PTR_ERR(handle);
5719 /* Expand file to avoid data loss if there is error while shifting */
5720 inode->i_size += len;
5721 EXT4_I(inode)->i_disksize += len;
5722 inode->i_mtime = inode->i_ctime = current_time(inode);
5723 ret = ext4_mark_inode_dirty(handle, inode);
5727 down_write(&EXT4_I(inode)->i_data_sem);
5728 ext4_discard_preallocations(inode);
5730 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5732 up_write(&EXT4_I(inode)->i_data_sem);
5736 depth = ext_depth(inode);
5737 extent = path[depth].p_ext;
5739 ee_start_lblk = le32_to_cpu(extent->ee_block);
5740 ee_len = ext4_ext_get_actual_len(extent);
5743 * If offset_lblk is not the starting block of extent, split
5744 * the extent @offset_lblk
5746 if ((offset_lblk > ee_start_lblk) &&
5747 (offset_lblk < (ee_start_lblk + ee_len))) {
5748 if (ext4_ext_is_unwritten(extent))
5749 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5750 EXT4_EXT_MARK_UNWRIT2;
5751 ret = ext4_split_extent_at(handle, inode, &path,
5752 offset_lblk, split_flag,
5754 EXT4_GET_BLOCKS_PRE_IO |
5755 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5758 ext4_ext_drop_refs(path);
5761 up_write(&EXT4_I(inode)->i_data_sem);
5765 ext4_ext_drop_refs(path);
5769 ret = ext4_es_remove_extent(inode, offset_lblk,
5770 EXT_MAX_BLOCKS - offset_lblk);
5772 up_write(&EXT4_I(inode)->i_data_sem);
5777 * if offset_lblk lies in a hole which is at start of file, use
5778 * ee_start_lblk to shift extents
5780 ret = ext4_ext_shift_extents(inode, handle,
5781 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5782 len_lblk, SHIFT_RIGHT);
5784 up_write(&EXT4_I(inode)->i_data_sem);
5786 ext4_handle_sync(handle);
5788 ext4_update_inode_fsync_trans(handle, inode, 1);
5791 ext4_journal_stop(handle);
5793 up_write(&EXT4_I(inode)->i_mmap_sem);
5795 inode_unlock(inode);
5800 * ext4_swap_extents - Swap extents between two inodes
5802 * @inode1: First inode
5803 * @inode2: Second inode
5804 * @lblk1: Start block for first inode
5805 * @lblk2: Start block for second inode
5806 * @count: Number of blocks to swap
5807 * @unwritten: Mark second inode's extents as unwritten after swap
5808 * @erp: Pointer to save error value
5810 * This helper routine does exactly what is promise "swap extents". All other
5811 * stuff such as page-cache locking consistency, bh mapping consistency or
5812 * extent's data copying must be performed by caller.
5814 * i_mutex is held for both inodes
5815 * i_data_sem is locked for write for both inodes
5817 * All pages from requested range are locked for both inodes
5820 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5821 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5822 ext4_lblk_t count, int unwritten, int *erp)
5824 struct ext4_ext_path *path1 = NULL;
5825 struct ext4_ext_path *path2 = NULL;
5826 int replaced_count = 0;
5828 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5829 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5830 BUG_ON(!inode_is_locked(inode1));
5831 BUG_ON(!inode_is_locked(inode2));
5833 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5836 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5841 struct ext4_extent *ex1, *ex2, tmp_ex;
5842 ext4_lblk_t e1_blk, e2_blk;
5843 int e1_len, e2_len, len;
5846 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5847 if (IS_ERR(path1)) {
5848 *erp = PTR_ERR(path1);
5854 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5855 if (IS_ERR(path2)) {
5856 *erp = PTR_ERR(path2);
5860 ex1 = path1[path1->p_depth].p_ext;
5861 ex2 = path2[path2->p_depth].p_ext;
5862 /* Do we have somthing to swap ? */
5863 if (unlikely(!ex2 || !ex1))
5866 e1_blk = le32_to_cpu(ex1->ee_block);
5867 e2_blk = le32_to_cpu(ex2->ee_block);
5868 e1_len = ext4_ext_get_actual_len(ex1);
5869 e2_len = ext4_ext_get_actual_len(ex2);
5872 if (!in_range(lblk1, e1_blk, e1_len) ||
5873 !in_range(lblk2, e2_blk, e2_len)) {
5874 ext4_lblk_t next1, next2;
5876 /* if hole after extent, then go to next extent */
5877 next1 = ext4_ext_next_allocated_block(path1);
5878 next2 = ext4_ext_next_allocated_block(path2);
5879 /* If hole before extent, then shift to that extent */
5884 /* Do we have something to swap */
5885 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5887 /* Move to the rightest boundary */
5888 len = next1 - lblk1;
5889 if (len < next2 - lblk2)
5890 len = next2 - lblk2;
5899 /* Prepare left boundary */
5900 if (e1_blk < lblk1) {
5902 *erp = ext4_force_split_extent_at(handle, inode1,
5907 if (e2_blk < lblk2) {
5909 *erp = ext4_force_split_extent_at(handle, inode2,
5914 /* ext4_split_extent_at() may result in leaf extent split,
5915 * path must to be revalidated. */
5919 /* Prepare right boundary */
5921 if (len > e1_blk + e1_len - lblk1)
5922 len = e1_blk + e1_len - lblk1;
5923 if (len > e2_blk + e2_len - lblk2)
5924 len = e2_blk + e2_len - lblk2;
5926 if (len != e1_len) {
5928 *erp = ext4_force_split_extent_at(handle, inode1,
5929 &path1, lblk1 + len, 0);
5933 if (len != e2_len) {
5935 *erp = ext4_force_split_extent_at(handle, inode2,
5936 &path2, lblk2 + len, 0);
5940 /* ext4_split_extent_at() may result in leaf extent split,
5941 * path must to be revalidated. */
5945 BUG_ON(e2_len != e1_len);
5946 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5949 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5953 /* Both extents are fully inside boundaries. Swap it now */
5955 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5956 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5957 ex1->ee_len = cpu_to_le16(e2_len);
5958 ex2->ee_len = cpu_to_le16(e1_len);
5960 ext4_ext_mark_unwritten(ex2);
5961 if (ext4_ext_is_unwritten(&tmp_ex))
5962 ext4_ext_mark_unwritten(ex1);
5964 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5965 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5966 *erp = ext4_ext_dirty(handle, inode2, path2 +
5970 *erp = ext4_ext_dirty(handle, inode1, path1 +
5973 * Looks scarry ah..? second inode already points to new blocks,
5974 * and it was successfully dirtied. But luckily error may happen
5975 * only due to journal error, so full transaction will be
5982 replaced_count += len;
5986 ext4_ext_drop_refs(path1);
5988 ext4_ext_drop_refs(path2);
5990 path1 = path2 = NULL;
5992 return replaced_count;