1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
6 * Architecture independence:
7 * Copyright (c) 2005, Bull S.A.
8 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
12 * Extents support for EXT4
15 * - ext4*_error() should be used in some situations
16 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17 * - smart tree reduction
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/backing-dev.h>
31 #include "ext4_jbd2.h"
32 #include "ext4_extents.h"
35 #include <trace/events/ext4.h>
38 * used by extent splitting.
40 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
42 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
43 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
45 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
46 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
48 static __le32 ext4_extent_block_csum(struct inode *inode,
49 struct ext4_extent_header *eh)
51 struct ext4_inode_info *ei = EXT4_I(inode);
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
55 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
56 EXT4_EXTENT_TAIL_OFFSET(eh));
57 return cpu_to_le32(csum);
60 static int ext4_extent_block_csum_verify(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_extent_tail *et;
65 if (!ext4_has_metadata_csum(inode->i_sb))
68 et = find_ext4_extent_tail(eh);
69 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
74 static void ext4_extent_block_csum_set(struct inode *inode,
75 struct ext4_extent_header *eh)
77 struct ext4_extent_tail *et;
79 if (!ext4_has_metadata_csum(inode->i_sb))
82 et = find_ext4_extent_tail(eh);
83 et->et_checksum = ext4_extent_block_csum(inode, eh);
86 static int ext4_split_extent(handle_t *handle,
88 struct ext4_ext_path **ppath,
89 struct ext4_map_blocks *map,
93 static int ext4_split_extent_at(handle_t *handle,
95 struct ext4_ext_path **ppath,
100 static int ext4_find_delayed_extent(struct inode *inode,
101 struct extent_status *newes);
103 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
106 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
107 * moment, get_block can be called only for blocks inside i_size since
108 * page cache has been already dropped and writes are blocked by
109 * i_mutex. So we can safely drop the i_data_sem here.
111 BUG_ON(EXT4_JOURNAL(inode) == NULL);
112 ext4_discard_preallocations(inode);
113 up_write(&EXT4_I(inode)->i_data_sem);
119 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
120 * transaction with 'restart_cred' credits. The function drops i_data_sem
121 * when restarting transaction and gets it after transaction is restarted.
123 * The function returns 0 on success, 1 if transaction had to be restarted,
124 * and < 0 in case of fatal error.
126 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
127 int check_cred, int restart_cred,
133 ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
134 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
136 down_write(&EXT4_I(inode)->i_data_sem);
145 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
146 struct ext4_ext_path *path)
149 /* path points to block */
150 BUFFER_TRACE(path->p_bh, "get_write_access");
151 return ext4_journal_get_write_access(handle, path->p_bh);
153 /* path points to leaf/index in inode body */
154 /* we use in-core data, no need to protect them */
164 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
165 struct inode *inode, struct ext4_ext_path *path)
169 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
171 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
172 /* path points to block */
173 err = __ext4_handle_dirty_metadata(where, line, handle,
176 /* path points to leaf/index in inode body */
177 err = ext4_mark_inode_dirty(handle, inode);
182 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
183 struct ext4_ext_path *path,
187 int depth = path->p_depth;
188 struct ext4_extent *ex;
191 * Try to predict block placement assuming that we are
192 * filling in a file which will eventually be
193 * non-sparse --- i.e., in the case of libbfd writing
194 * an ELF object sections out-of-order but in a way
195 * the eventually results in a contiguous object or
196 * executable file, or some database extending a table
197 * space file. However, this is actually somewhat
198 * non-ideal if we are writing a sparse file such as
199 * qemu or KVM writing a raw image file that is going
200 * to stay fairly sparse, since it will end up
201 * fragmenting the file system's free space. Maybe we
202 * should have some hueristics or some way to allow
203 * userspace to pass a hint to file system,
204 * especially if the latter case turns out to be
207 ex = path[depth].p_ext;
209 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
210 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
212 if (block > ext_block)
213 return ext_pblk + (block - ext_block);
215 return ext_pblk - (ext_block - block);
218 /* it looks like index is empty;
219 * try to find starting block from index itself */
220 if (path[depth].p_bh)
221 return path[depth].p_bh->b_blocknr;
224 /* OK. use inode's group */
225 return ext4_inode_to_goal_block(inode);
229 * Allocation for a meta data block
232 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
233 struct ext4_ext_path *path,
234 struct ext4_extent *ex, int *err, unsigned int flags)
236 ext4_fsblk_t goal, newblock;
238 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
239 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
244 static inline int ext4_ext_space_block(struct inode *inode, int check)
248 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
249 / sizeof(struct ext4_extent);
250 #ifdef AGGRESSIVE_TEST
251 if (!check && size > 6)
257 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
261 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
262 / sizeof(struct ext4_extent_idx);
263 #ifdef AGGRESSIVE_TEST
264 if (!check && size > 5)
270 static inline int ext4_ext_space_root(struct inode *inode, int check)
274 size = sizeof(EXT4_I(inode)->i_data);
275 size -= sizeof(struct ext4_extent_header);
276 size /= sizeof(struct ext4_extent);
277 #ifdef AGGRESSIVE_TEST
278 if (!check && size > 3)
284 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
288 size = sizeof(EXT4_I(inode)->i_data);
289 size -= sizeof(struct ext4_extent_header);
290 size /= sizeof(struct ext4_extent_idx);
291 #ifdef AGGRESSIVE_TEST
292 if (!check && size > 4)
299 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
300 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
303 struct ext4_ext_path *path = *ppath;
304 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
306 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
307 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
308 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
309 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
313 * Calculate the number of metadata blocks needed
314 * to allocate @blocks
315 * Worse case is one block per extent
317 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
319 struct ext4_inode_info *ei = EXT4_I(inode);
322 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
323 / sizeof(struct ext4_extent_idx));
326 * If the new delayed allocation block is contiguous with the
327 * previous da block, it can share index blocks with the
328 * previous block, so we only need to allocate a new index
329 * block every idxs leaf blocks. At ldxs**2 blocks, we need
330 * an additional index block, and at ldxs**3 blocks, yet
331 * another index blocks.
333 if (ei->i_da_metadata_calc_len &&
334 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
337 if ((ei->i_da_metadata_calc_len % idxs) == 0)
339 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
341 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
343 ei->i_da_metadata_calc_len = 0;
345 ei->i_da_metadata_calc_len++;
346 ei->i_da_metadata_calc_last_lblock++;
351 * In the worst case we need a new set of index blocks at
352 * every level of the inode's extent tree.
354 ei->i_da_metadata_calc_len = 1;
355 ei->i_da_metadata_calc_last_lblock = lblock;
356 return ext_depth(inode) + 1;
360 ext4_ext_max_entries(struct inode *inode, int depth)
364 if (depth == ext_depth(inode)) {
366 max = ext4_ext_space_root(inode, 1);
368 max = ext4_ext_space_root_idx(inode, 1);
371 max = ext4_ext_space_block(inode, 1);
373 max = ext4_ext_space_block_idx(inode, 1);
379 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
381 ext4_fsblk_t block = ext4_ext_pblock(ext);
382 int len = ext4_ext_get_actual_len(ext);
383 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
388 * - overflow/wrap-around
390 if (lblock + len <= lblock)
392 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
395 static int ext4_valid_extent_idx(struct inode *inode,
396 struct ext4_extent_idx *ext_idx)
398 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
400 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
403 static int ext4_valid_extent_entries(struct inode *inode,
404 struct ext4_extent_header *eh,
407 unsigned short entries;
408 if (eh->eh_entries == 0)
411 entries = le16_to_cpu(eh->eh_entries);
415 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
416 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
417 ext4_fsblk_t pblock = 0;
418 ext4_lblk_t lblock = 0;
419 ext4_lblk_t prev = 0;
422 if (!ext4_valid_extent(inode, ext))
425 /* Check for overlapping extents */
426 lblock = le32_to_cpu(ext->ee_block);
427 len = ext4_ext_get_actual_len(ext);
428 if ((lblock <= prev) && prev) {
429 pblock = ext4_ext_pblock(ext);
430 es->s_last_error_block = cpu_to_le64(pblock);
435 prev = lblock + len - 1;
438 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
440 if (!ext4_valid_extent_idx(inode, ext_idx))
449 static int __ext4_ext_check(const char *function, unsigned int line,
450 struct inode *inode, struct ext4_extent_header *eh,
451 int depth, ext4_fsblk_t pblk)
453 const char *error_msg;
454 int max = 0, err = -EFSCORRUPTED;
456 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
457 error_msg = "invalid magic";
460 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
461 error_msg = "unexpected eh_depth";
464 if (unlikely(eh->eh_max == 0)) {
465 error_msg = "invalid eh_max";
468 max = ext4_ext_max_entries(inode, depth);
469 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
470 error_msg = "too large eh_max";
473 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
474 error_msg = "invalid eh_entries";
477 if (!ext4_valid_extent_entries(inode, eh, depth)) {
478 error_msg = "invalid extent entries";
481 if (unlikely(depth > 32)) {
482 error_msg = "too large eh_depth";
485 /* Verify checksum on non-root extent tree nodes */
486 if (ext_depth(inode) != depth &&
487 !ext4_extent_block_csum_verify(inode, eh)) {
488 error_msg = "extent tree corrupted";
495 ext4_error_inode(inode, function, line, 0,
496 "pblk %llu bad header/extent: %s - magic %x, "
497 "entries %u, max %u(%u), depth %u(%u)",
498 (unsigned long long) pblk, error_msg,
499 le16_to_cpu(eh->eh_magic),
500 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
501 max, le16_to_cpu(eh->eh_depth), depth);
505 #define ext4_ext_check(inode, eh, depth, pblk) \
506 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
508 int ext4_ext_check_inode(struct inode *inode)
510 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
513 static struct buffer_head *
514 __read_extent_tree_block(const char *function, unsigned int line,
515 struct inode *inode, ext4_fsblk_t pblk, int depth,
518 struct buffer_head *bh;
521 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
523 return ERR_PTR(-ENOMEM);
525 if (!bh_uptodate_or_lock(bh)) {
526 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
527 err = bh_submit_read(bh);
531 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
533 if (!ext4_has_feature_journal(inode->i_sb) ||
535 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
536 err = __ext4_ext_check(function, line, inode,
537 ext_block_hdr(bh), depth, pblk);
541 set_buffer_verified(bh);
543 * If this is a leaf block, cache all of its entries
545 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
546 struct ext4_extent_header *eh = ext_block_hdr(bh);
547 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
548 ext4_lblk_t prev = 0;
551 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
552 unsigned int status = EXTENT_STATUS_WRITTEN;
553 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
554 int len = ext4_ext_get_actual_len(ex);
556 if (prev && (prev != lblk))
557 ext4_es_cache_extent(inode, prev,
561 if (ext4_ext_is_unwritten(ex))
562 status = EXTENT_STATUS_UNWRITTEN;
563 ext4_es_cache_extent(inode, lblk, len,
564 ext4_ext_pblock(ex), status);
575 #define read_extent_tree_block(inode, pblk, depth, flags) \
576 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
580 * This function is called to cache a file's extent information in the
583 int ext4_ext_precache(struct inode *inode)
585 struct ext4_inode_info *ei = EXT4_I(inode);
586 struct ext4_ext_path *path = NULL;
587 struct buffer_head *bh;
588 int i = 0, depth, ret = 0;
590 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
591 return 0; /* not an extent-mapped inode */
593 down_read(&ei->i_data_sem);
594 depth = ext_depth(inode);
596 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
599 up_read(&ei->i_data_sem);
603 /* Don't cache anything if there are no external extent blocks */
606 path[0].p_hdr = ext_inode_hdr(inode);
607 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
610 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
613 * If this is a leaf block or we've reached the end of
614 * the index block, go up
617 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
618 brelse(path[i].p_bh);
623 bh = read_extent_tree_block(inode,
624 ext4_idx_pblock(path[i].p_idx++),
626 EXT4_EX_FORCE_CACHE);
633 path[i].p_hdr = ext_block_hdr(bh);
634 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
636 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
638 up_read(&ei->i_data_sem);
639 ext4_ext_drop_refs(path);
645 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
647 int k, l = path->p_depth;
650 for (k = 0; k <= l; k++, path++) {
652 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
653 ext4_idx_pblock(path->p_idx));
654 } else if (path->p_ext) {
655 ext_debug(" %d:[%d]%d:%llu ",
656 le32_to_cpu(path->p_ext->ee_block),
657 ext4_ext_is_unwritten(path->p_ext),
658 ext4_ext_get_actual_len(path->p_ext),
659 ext4_ext_pblock(path->p_ext));
666 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
668 int depth = ext_depth(inode);
669 struct ext4_extent_header *eh;
670 struct ext4_extent *ex;
676 eh = path[depth].p_hdr;
677 ex = EXT_FIRST_EXTENT(eh);
679 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
681 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
682 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
683 ext4_ext_is_unwritten(ex),
684 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
689 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
690 ext4_fsblk_t newblock, int level)
692 int depth = ext_depth(inode);
693 struct ext4_extent *ex;
695 if (depth != level) {
696 struct ext4_extent_idx *idx;
697 idx = path[level].p_idx;
698 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
699 ext_debug("%d: move %d:%llu in new index %llu\n", level,
700 le32_to_cpu(idx->ei_block),
701 ext4_idx_pblock(idx),
709 ex = path[depth].p_ext;
710 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
711 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
712 le32_to_cpu(ex->ee_block),
714 ext4_ext_is_unwritten(ex),
715 ext4_ext_get_actual_len(ex),
722 #define ext4_ext_show_path(inode, path)
723 #define ext4_ext_show_leaf(inode, path)
724 #define ext4_ext_show_move(inode, path, newblock, level)
727 void ext4_ext_drop_refs(struct ext4_ext_path *path)
733 depth = path->p_depth;
734 for (i = 0; i <= depth; i++, path++)
742 * ext4_ext_binsearch_idx:
743 * binary search for the closest index of the given block
744 * the header must be checked before calling this
747 ext4_ext_binsearch_idx(struct inode *inode,
748 struct ext4_ext_path *path, ext4_lblk_t block)
750 struct ext4_extent_header *eh = path->p_hdr;
751 struct ext4_extent_idx *r, *l, *m;
754 ext_debug("binsearch for %u(idx): ", block);
756 l = EXT_FIRST_INDEX(eh) + 1;
757 r = EXT_LAST_INDEX(eh);
760 if (block < le32_to_cpu(m->ei_block))
764 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
765 m, le32_to_cpu(m->ei_block),
766 r, le32_to_cpu(r->ei_block));
770 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
771 ext4_idx_pblock(path->p_idx));
773 #ifdef CHECK_BINSEARCH
775 struct ext4_extent_idx *chix, *ix;
778 chix = ix = EXT_FIRST_INDEX(eh);
779 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
781 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
782 printk(KERN_DEBUG "k=%d, ix=0x%p, "
784 ix, EXT_FIRST_INDEX(eh));
785 printk(KERN_DEBUG "%u <= %u\n",
786 le32_to_cpu(ix->ei_block),
787 le32_to_cpu(ix[-1].ei_block));
789 BUG_ON(k && le32_to_cpu(ix->ei_block)
790 <= le32_to_cpu(ix[-1].ei_block));
791 if (block < le32_to_cpu(ix->ei_block))
795 BUG_ON(chix != path->p_idx);
802 * ext4_ext_binsearch:
803 * binary search for closest extent of the given block
804 * the header must be checked before calling this
807 ext4_ext_binsearch(struct inode *inode,
808 struct ext4_ext_path *path, ext4_lblk_t block)
810 struct ext4_extent_header *eh = path->p_hdr;
811 struct ext4_extent *r, *l, *m;
813 if (eh->eh_entries == 0) {
815 * this leaf is empty:
816 * we get such a leaf in split/add case
821 ext_debug("binsearch for %u: ", block);
823 l = EXT_FIRST_EXTENT(eh) + 1;
824 r = EXT_LAST_EXTENT(eh);
828 if (block < le32_to_cpu(m->ee_block))
832 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
833 m, le32_to_cpu(m->ee_block),
834 r, le32_to_cpu(r->ee_block));
838 ext_debug(" -> %d:%llu:[%d]%d ",
839 le32_to_cpu(path->p_ext->ee_block),
840 ext4_ext_pblock(path->p_ext),
841 ext4_ext_is_unwritten(path->p_ext),
842 ext4_ext_get_actual_len(path->p_ext));
844 #ifdef CHECK_BINSEARCH
846 struct ext4_extent *chex, *ex;
849 chex = ex = EXT_FIRST_EXTENT(eh);
850 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
851 BUG_ON(k && le32_to_cpu(ex->ee_block)
852 <= le32_to_cpu(ex[-1].ee_block));
853 if (block < le32_to_cpu(ex->ee_block))
857 BUG_ON(chex != path->p_ext);
863 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
865 struct ext4_extent_header *eh;
867 eh = ext_inode_hdr(inode);
870 eh->eh_magic = EXT4_EXT_MAGIC;
871 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
872 ext4_mark_inode_dirty(handle, inode);
876 struct ext4_ext_path *
877 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
878 struct ext4_ext_path **orig_path, int flags)
880 struct ext4_extent_header *eh;
881 struct buffer_head *bh;
882 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
883 short int depth, i, ppos = 0;
886 eh = ext_inode_hdr(inode);
887 depth = ext_depth(inode);
888 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
889 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
896 ext4_ext_drop_refs(path);
897 if (depth > path[0].p_maxdepth) {
899 *orig_path = path = NULL;
903 /* account possible depth increase */
904 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
907 return ERR_PTR(-ENOMEM);
908 path[0].p_maxdepth = depth + 1;
914 /* walk through the tree */
916 ext_debug("depth %d: num %d, max %d\n",
917 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
919 ext4_ext_binsearch_idx(inode, path + ppos, block);
920 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
921 path[ppos].p_depth = i;
922 path[ppos].p_ext = NULL;
924 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
931 eh = ext_block_hdr(bh);
933 path[ppos].p_bh = bh;
934 path[ppos].p_hdr = eh;
937 path[ppos].p_depth = i;
938 path[ppos].p_ext = NULL;
939 path[ppos].p_idx = NULL;
942 ext4_ext_binsearch(inode, path + ppos, block);
943 /* if not an empty leaf */
944 if (path[ppos].p_ext)
945 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
947 ext4_ext_show_path(inode, path);
952 ext4_ext_drop_refs(path);
960 * ext4_ext_insert_index:
961 * insert new index [@logical;@ptr] into the block at @curp;
962 * check where to insert: before @curp or after @curp
964 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
965 struct ext4_ext_path *curp,
966 int logical, ext4_fsblk_t ptr)
968 struct ext4_extent_idx *ix;
971 err = ext4_ext_get_access(handle, inode, curp);
975 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
976 EXT4_ERROR_INODE(inode,
977 "logical %d == ei_block %d!",
978 logical, le32_to_cpu(curp->p_idx->ei_block));
979 return -EFSCORRUPTED;
982 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
983 >= le16_to_cpu(curp->p_hdr->eh_max))) {
984 EXT4_ERROR_INODE(inode,
985 "eh_entries %d >= eh_max %d!",
986 le16_to_cpu(curp->p_hdr->eh_entries),
987 le16_to_cpu(curp->p_hdr->eh_max));
988 return -EFSCORRUPTED;
991 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
993 ext_debug("insert new index %d after: %llu\n", logical, ptr);
994 ix = curp->p_idx + 1;
997 ext_debug("insert new index %d before: %llu\n", logical, ptr);
1001 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1004 ext_debug("insert new index %d: "
1005 "move %d indices from 0x%p to 0x%p\n",
1006 logical, len, ix, ix + 1);
1007 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1010 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1011 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1012 return -EFSCORRUPTED;
1015 ix->ei_block = cpu_to_le32(logical);
1016 ext4_idx_store_pblock(ix, ptr);
1017 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1019 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1020 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1021 return -EFSCORRUPTED;
1024 err = ext4_ext_dirty(handle, inode, curp);
1025 ext4_std_error(inode->i_sb, err);
1032 * inserts new subtree into the path, using free index entry
1034 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1035 * - makes decision where to split
1036 * - moves remaining extents and index entries (right to the split point)
1037 * into the newly allocated blocks
1038 * - initializes subtree
1040 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1042 struct ext4_ext_path *path,
1043 struct ext4_extent *newext, int at)
1045 struct buffer_head *bh = NULL;
1046 int depth = ext_depth(inode);
1047 struct ext4_extent_header *neh;
1048 struct ext4_extent_idx *fidx;
1049 int i = at, k, m, a;
1050 ext4_fsblk_t newblock, oldblock;
1052 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1054 size_t ext_size = 0;
1056 /* make decision: where to split? */
1057 /* FIXME: now decision is simplest: at current extent */
1059 /* if current leaf will be split, then we should use
1060 * border from split point */
1061 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1062 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1063 return -EFSCORRUPTED;
1065 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1066 border = path[depth].p_ext[1].ee_block;
1067 ext_debug("leaf will be split."
1068 " next leaf starts at %d\n",
1069 le32_to_cpu(border));
1071 border = newext->ee_block;
1072 ext_debug("leaf will be added."
1073 " next leaf starts at %d\n",
1074 le32_to_cpu(border));
1078 * If error occurs, then we break processing
1079 * and mark filesystem read-only. index won't
1080 * be inserted and tree will be in consistent
1081 * state. Next mount will repair buffers too.
1085 * Get array to track all allocated blocks.
1086 * We need this to handle errors and free blocks
1089 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1093 /* allocate all needed blocks */
1094 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1095 for (a = 0; a < depth - at; a++) {
1096 newblock = ext4_ext_new_meta_block(handle, inode, path,
1097 newext, &err, flags);
1100 ablocks[a] = newblock;
1103 /* initialize new leaf */
1104 newblock = ablocks[--a];
1105 if (unlikely(newblock == 0)) {
1106 EXT4_ERROR_INODE(inode, "newblock == 0!");
1107 err = -EFSCORRUPTED;
1110 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1111 if (unlikely(!bh)) {
1117 err = ext4_journal_get_create_access(handle, bh);
1121 neh = ext_block_hdr(bh);
1122 neh->eh_entries = 0;
1123 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1124 neh->eh_magic = EXT4_EXT_MAGIC;
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 fidx = EXT_FIRST_INDEX(neh);
1204 fidx->ei_block = border;
1205 ext4_idx_store_pblock(fidx, oldblock);
1207 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1208 i, newblock, le32_to_cpu(border), oldblock);
1210 /* move remainder of path[i] to the new index block */
1211 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1212 EXT_LAST_INDEX(path[i].p_hdr))) {
1213 EXT4_ERROR_INODE(inode,
1214 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1215 le32_to_cpu(path[i].p_ext->ee_block));
1216 err = -EFSCORRUPTED;
1219 /* start copy indexes */
1220 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1221 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1222 EXT_MAX_INDEX(path[i].p_hdr));
1223 ext4_ext_show_move(inode, path, newblock, i);
1225 memmove(++fidx, path[i].p_idx,
1226 sizeof(struct ext4_extent_idx) * m);
1227 le16_add_cpu(&neh->eh_entries, m);
1229 /* zero out unused area in the extent block */
1230 ext_size = sizeof(struct ext4_extent_header) +
1231 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1232 memset(bh->b_data + ext_size, 0,
1233 inode->i_sb->s_blocksize - ext_size);
1234 ext4_extent_block_csum_set(inode, neh);
1235 set_buffer_uptodate(bh);
1238 err = ext4_handle_dirty_metadata(handle, inode, bh);
1244 /* correct old index */
1246 err = ext4_ext_get_access(handle, inode, path + i);
1249 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1250 err = ext4_ext_dirty(handle, inode, path + i);
1258 /* insert new index */
1259 err = ext4_ext_insert_index(handle, inode, path + at,
1260 le32_to_cpu(border), newblock);
1264 if (buffer_locked(bh))
1270 /* free all allocated blocks in error case */
1271 for (i = 0; i < depth; i++) {
1274 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1275 EXT4_FREE_BLOCKS_METADATA);
1284 * ext4_ext_grow_indepth:
1285 * implements tree growing procedure:
1286 * - allocates new block
1287 * - moves top-level data (index block or leaf) into the new block
1288 * - initializes new top-level, creating index that points to the
1289 * just created block
1291 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1294 struct ext4_extent_header *neh;
1295 struct buffer_head *bh;
1296 ext4_fsblk_t newblock, goal = 0;
1297 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1299 size_t ext_size = 0;
1301 /* Try to prepend new index to old one */
1302 if (ext_depth(inode))
1303 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1304 if (goal > le32_to_cpu(es->s_first_data_block)) {
1305 flags |= EXT4_MB_HINT_TRY_GOAL;
1308 goal = ext4_inode_to_goal_block(inode);
1309 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1314 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1319 err = ext4_journal_get_create_access(handle, bh);
1325 ext_size = sizeof(EXT4_I(inode)->i_data);
1326 /* move top-level index/leaf into new block */
1327 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1328 /* zero out unused area in the extent block */
1329 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1331 /* set size of new block */
1332 neh = ext_block_hdr(bh);
1333 /* old root could have indexes or leaves
1334 * so calculate e_max right way */
1335 if (ext_depth(inode))
1336 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1338 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1339 neh->eh_magic = EXT4_EXT_MAGIC;
1340 ext4_extent_block_csum_set(inode, neh);
1341 set_buffer_uptodate(bh);
1344 err = ext4_handle_dirty_metadata(handle, inode, bh);
1348 /* Update top-level index: num,max,pointer */
1349 neh = ext_inode_hdr(inode);
1350 neh->eh_entries = cpu_to_le16(1);
1351 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1352 if (neh->eh_depth == 0) {
1353 /* Root extent block becomes index block */
1354 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1355 EXT_FIRST_INDEX(neh)->ei_block =
1356 EXT_FIRST_EXTENT(neh)->ee_block;
1358 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1359 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1360 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1361 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1363 le16_add_cpu(&neh->eh_depth, 1);
1364 ext4_mark_inode_dirty(handle, inode);
1372 * ext4_ext_create_new_leaf:
1373 * finds empty index and adds new leaf.
1374 * if no free index is found, then it requests in-depth growing.
1376 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1377 unsigned int mb_flags,
1378 unsigned int gb_flags,
1379 struct ext4_ext_path **ppath,
1380 struct ext4_extent *newext)
1382 struct ext4_ext_path *path = *ppath;
1383 struct ext4_ext_path *curp;
1384 int depth, i, err = 0;
1387 i = depth = ext_depth(inode);
1389 /* walk up to the tree and look for free index entry */
1390 curp = path + depth;
1391 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1396 /* we use already allocated block for index block,
1397 * so subsequent data blocks should be contiguous */
1398 if (EXT_HAS_FREE_INDEX(curp)) {
1399 /* if we found index with free entry, then use that
1400 * entry: create all needed subtree and add new leaf */
1401 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1406 path = ext4_find_extent(inode,
1407 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1410 err = PTR_ERR(path);
1412 /* tree is full, time to grow in depth */
1413 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1418 path = ext4_find_extent(inode,
1419 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1422 err = PTR_ERR(path);
1427 * only first (depth 0 -> 1) produces free space;
1428 * in all other cases we have to split the grown tree
1430 depth = ext_depth(inode);
1431 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1432 /* now we need to split */
1442 * search the closest allocated block to the left for *logical
1443 * and returns it at @logical + it's physical address at @phys
1444 * if *logical is the smallest allocated block, the function
1445 * returns 0 at @phys
1446 * return value contains 0 (success) or error code
1448 static int ext4_ext_search_left(struct inode *inode,
1449 struct ext4_ext_path *path,
1450 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1452 struct ext4_extent_idx *ix;
1453 struct ext4_extent *ex;
1456 if (unlikely(path == NULL)) {
1457 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1458 return -EFSCORRUPTED;
1460 depth = path->p_depth;
1463 if (depth == 0 && path->p_ext == NULL)
1466 /* usually extent in the path covers blocks smaller
1467 * then *logical, but it can be that extent is the
1468 * first one in the file */
1470 ex = path[depth].p_ext;
1471 ee_len = ext4_ext_get_actual_len(ex);
1472 if (*logical < le32_to_cpu(ex->ee_block)) {
1473 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1474 EXT4_ERROR_INODE(inode,
1475 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1476 *logical, le32_to_cpu(ex->ee_block));
1477 return -EFSCORRUPTED;
1479 while (--depth >= 0) {
1480 ix = path[depth].p_idx;
1481 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1482 EXT4_ERROR_INODE(inode,
1483 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1484 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1485 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1486 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1488 return -EFSCORRUPTED;
1494 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1495 EXT4_ERROR_INODE(inode,
1496 "logical %d < ee_block %d + ee_len %d!",
1497 *logical, le32_to_cpu(ex->ee_block), ee_len);
1498 return -EFSCORRUPTED;
1501 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1502 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1507 * search the closest allocated block to the right for *logical
1508 * and returns it at @logical + it's physical address at @phys
1509 * if *logical is the largest allocated block, the function
1510 * returns 0 at @phys
1511 * return value contains 0 (success) or error code
1513 static int ext4_ext_search_right(struct inode *inode,
1514 struct ext4_ext_path *path,
1515 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1516 struct ext4_extent **ret_ex)
1518 struct buffer_head *bh = NULL;
1519 struct ext4_extent_header *eh;
1520 struct ext4_extent_idx *ix;
1521 struct ext4_extent *ex;
1523 int depth; /* Note, NOT eh_depth; depth from top of tree */
1526 if (unlikely(path == NULL)) {
1527 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1528 return -EFSCORRUPTED;
1530 depth = path->p_depth;
1533 if (depth == 0 && path->p_ext == NULL)
1536 /* usually extent in the path covers blocks smaller
1537 * then *logical, but it can be that extent is the
1538 * first one in the file */
1540 ex = path[depth].p_ext;
1541 ee_len = ext4_ext_get_actual_len(ex);
1542 if (*logical < le32_to_cpu(ex->ee_block)) {
1543 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1544 EXT4_ERROR_INODE(inode,
1545 "first_extent(path[%d].p_hdr) != ex",
1547 return -EFSCORRUPTED;
1549 while (--depth >= 0) {
1550 ix = path[depth].p_idx;
1551 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1552 EXT4_ERROR_INODE(inode,
1553 "ix != EXT_FIRST_INDEX *logical %d!",
1555 return -EFSCORRUPTED;
1561 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1562 EXT4_ERROR_INODE(inode,
1563 "logical %d < ee_block %d + ee_len %d!",
1564 *logical, le32_to_cpu(ex->ee_block), ee_len);
1565 return -EFSCORRUPTED;
1568 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1569 /* next allocated block in this leaf */
1574 /* go up and search for index to the right */
1575 while (--depth >= 0) {
1576 ix = path[depth].p_idx;
1577 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1581 /* we've gone up to the root and found no index to the right */
1585 /* we've found index to the right, let's
1586 * follow it and find the closest allocated
1587 * block to the right */
1589 block = ext4_idx_pblock(ix);
1590 while (++depth < path->p_depth) {
1591 /* subtract from p_depth to get proper eh_depth */
1592 bh = read_extent_tree_block(inode, block,
1593 path->p_depth - depth, 0);
1596 eh = ext_block_hdr(bh);
1597 ix = EXT_FIRST_INDEX(eh);
1598 block = ext4_idx_pblock(ix);
1602 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1605 eh = ext_block_hdr(bh);
1606 ex = EXT_FIRST_EXTENT(eh);
1608 *logical = le32_to_cpu(ex->ee_block);
1609 *phys = ext4_ext_pblock(ex);
1617 * ext4_ext_next_allocated_block:
1618 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1619 * NOTE: it considers block number from index entry as
1620 * allocated block. Thus, index entries have to be consistent
1624 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1628 BUG_ON(path == NULL);
1629 depth = path->p_depth;
1631 if (depth == 0 && path->p_ext == NULL)
1632 return EXT_MAX_BLOCKS;
1634 while (depth >= 0) {
1635 if (depth == path->p_depth) {
1637 if (path[depth].p_ext &&
1638 path[depth].p_ext !=
1639 EXT_LAST_EXTENT(path[depth].p_hdr))
1640 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1643 if (path[depth].p_idx !=
1644 EXT_LAST_INDEX(path[depth].p_hdr))
1645 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1650 return EXT_MAX_BLOCKS;
1654 * ext4_ext_next_leaf_block:
1655 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1657 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1661 BUG_ON(path == NULL);
1662 depth = path->p_depth;
1664 /* zero-tree has no leaf blocks at all */
1666 return EXT_MAX_BLOCKS;
1668 /* go to index block */
1671 while (depth >= 0) {
1672 if (path[depth].p_idx !=
1673 EXT_LAST_INDEX(path[depth].p_hdr))
1674 return (ext4_lblk_t)
1675 le32_to_cpu(path[depth].p_idx[1].ei_block);
1679 return EXT_MAX_BLOCKS;
1683 * ext4_ext_correct_indexes:
1684 * if leaf gets modified and modified extent is first in the leaf,
1685 * then we have to correct all indexes above.
1686 * TODO: do we need to correct tree in all cases?
1688 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1689 struct ext4_ext_path *path)
1691 struct ext4_extent_header *eh;
1692 int depth = ext_depth(inode);
1693 struct ext4_extent *ex;
1697 eh = path[depth].p_hdr;
1698 ex = path[depth].p_ext;
1700 if (unlikely(ex == NULL || eh == NULL)) {
1701 EXT4_ERROR_INODE(inode,
1702 "ex %p == NULL or eh %p == NULL", ex, eh);
1703 return -EFSCORRUPTED;
1707 /* there is no tree at all */
1711 if (ex != EXT_FIRST_EXTENT(eh)) {
1712 /* we correct tree if first leaf got modified only */
1717 * TODO: we need correction if border is smaller than current one
1720 border = path[depth].p_ext->ee_block;
1721 err = ext4_ext_get_access(handle, inode, path + k);
1724 path[k].p_idx->ei_block = border;
1725 err = ext4_ext_dirty(handle, inode, path + k);
1730 /* change all left-side indexes */
1731 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1733 err = ext4_ext_get_access(handle, inode, path + k);
1736 path[k].p_idx->ei_block = border;
1737 err = ext4_ext_dirty(handle, inode, path + k);
1746 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1747 struct ext4_extent *ex2)
1749 unsigned short ext1_ee_len, ext2_ee_len;
1751 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1754 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1755 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1757 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1758 le32_to_cpu(ex2->ee_block))
1762 * To allow future support for preallocated extents to be added
1763 * as an RO_COMPAT feature, refuse to merge to extents if
1764 * this can result in the top bit of ee_len being set.
1766 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1769 * The check for IO to unwritten extent is somewhat racy as we
1770 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1771 * dropping i_data_sem. But reserved blocks should save us in that
1774 if (ext4_ext_is_unwritten(ex1) &&
1775 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1776 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1777 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1779 #ifdef AGGRESSIVE_TEST
1780 if (ext1_ee_len >= 4)
1784 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1790 * This function tries to merge the "ex" extent to the next extent in the tree.
1791 * It always tries to merge towards right. If you want to merge towards
1792 * left, pass "ex - 1" as argument instead of "ex".
1793 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1794 * 1 if they got merged.
1796 static int ext4_ext_try_to_merge_right(struct inode *inode,
1797 struct ext4_ext_path *path,
1798 struct ext4_extent *ex)
1800 struct ext4_extent_header *eh;
1801 unsigned int depth, len;
1802 int merge_done = 0, unwritten;
1804 depth = ext_depth(inode);
1805 BUG_ON(path[depth].p_hdr == NULL);
1806 eh = path[depth].p_hdr;
1808 while (ex < EXT_LAST_EXTENT(eh)) {
1809 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1811 /* merge with next extent! */
1812 unwritten = ext4_ext_is_unwritten(ex);
1813 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1814 + ext4_ext_get_actual_len(ex + 1));
1816 ext4_ext_mark_unwritten(ex);
1818 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1819 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1820 * sizeof(struct ext4_extent);
1821 memmove(ex + 1, ex + 2, len);
1823 le16_add_cpu(&eh->eh_entries, -1);
1825 WARN_ON(eh->eh_entries == 0);
1826 if (!eh->eh_entries)
1827 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1834 * This function does a very simple check to see if we can collapse
1835 * an extent tree with a single extent tree leaf block into the inode.
1837 static void ext4_ext_try_to_merge_up(handle_t *handle,
1838 struct inode *inode,
1839 struct ext4_ext_path *path)
1842 unsigned max_root = ext4_ext_space_root(inode, 0);
1845 if ((path[0].p_depth != 1) ||
1846 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1847 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1851 * We need to modify the block allocation bitmap and the block
1852 * group descriptor to release the extent tree block. If we
1853 * can't get the journal credits, give up.
1855 if (ext4_journal_extend(handle, 2,
1856 ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
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);
2331 static int ext4_fill_es_cache_info(struct inode *inode,
2332 ext4_lblk_t block, ext4_lblk_t num,
2333 struct fiemap_extent_info *fieinfo)
2335 ext4_lblk_t next, end = block + num - 1;
2336 struct extent_status es;
2337 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2341 while (block <= end) {
2344 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2346 if (ext4_es_is_unwritten(&es))
2347 flags |= FIEMAP_EXTENT_UNWRITTEN;
2348 if (ext4_es_is_delayed(&es))
2349 flags |= (FIEMAP_EXTENT_DELALLOC |
2350 FIEMAP_EXTENT_UNKNOWN);
2351 if (ext4_es_is_hole(&es))
2352 flags |= EXT4_FIEMAP_EXTENT_HOLE;
2354 flags |= FIEMAP_EXTENT_LAST;
2355 if (flags & (FIEMAP_EXTENT_DELALLOC|
2356 EXT4_FIEMAP_EXTENT_HOLE))
2359 es.es_pblk = ext4_es_pblock(&es);
2360 err = fiemap_fill_next_extent(fieinfo,
2361 (__u64)es.es_lblk << blksize_bits,
2362 (__u64)es.es_pblk << blksize_bits,
2363 (__u64)es.es_len << blksize_bits,
2378 * ext4_ext_determine_hole - determine hole around given block
2379 * @inode: inode we lookup in
2380 * @path: path in extent tree to @lblk
2381 * @lblk: pointer to logical block around which we want to determine hole
2383 * Determine hole length (and start if easily possible) around given logical
2384 * block. We don't try too hard to find the beginning of the hole but @path
2385 * actually points to extent before @lblk, we provide it.
2387 * The function returns the length of a hole starting at @lblk. We update @lblk
2388 * to the beginning of the hole if we managed to find it.
2390 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2391 struct ext4_ext_path *path,
2394 int depth = ext_depth(inode);
2395 struct ext4_extent *ex;
2398 ex = path[depth].p_ext;
2400 /* there is no extent yet, so gap is [0;-] */
2402 len = EXT_MAX_BLOCKS;
2403 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2404 len = le32_to_cpu(ex->ee_block) - *lblk;
2405 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2406 + ext4_ext_get_actual_len(ex)) {
2409 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2410 next = ext4_ext_next_allocated_block(path);
2411 BUG_ON(next == *lblk);
2420 * ext4_ext_put_gap_in_cache:
2421 * calculate boundaries of the gap that the requested block fits into
2422 * and cache this gap
2425 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2426 ext4_lblk_t hole_len)
2428 struct extent_status es;
2430 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2431 hole_start + hole_len - 1, &es);
2433 /* There's delayed extent containing lblock? */
2434 if (es.es_lblk <= hole_start)
2436 hole_len = min(es.es_lblk - hole_start, hole_len);
2438 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2439 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2440 EXTENT_STATUS_HOLE);
2445 * removes index from the index block.
2447 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2448 struct ext4_ext_path *path, int depth)
2453 /* free index block */
2455 path = path + depth;
2456 leaf = ext4_idx_pblock(path->p_idx);
2457 if (unlikely(path->p_hdr->eh_entries == 0)) {
2458 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2459 return -EFSCORRUPTED;
2461 err = ext4_ext_get_access(handle, inode, path);
2465 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2466 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2467 len *= sizeof(struct ext4_extent_idx);
2468 memmove(path->p_idx, path->p_idx + 1, len);
2471 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2472 err = ext4_ext_dirty(handle, inode, path);
2475 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2476 trace_ext4_ext_rm_idx(inode, leaf);
2478 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2479 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2481 while (--depth >= 0) {
2482 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2485 err = ext4_ext_get_access(handle, inode, path);
2488 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2489 err = ext4_ext_dirty(handle, inode, path);
2497 * ext4_ext_calc_credits_for_single_extent:
2498 * This routine returns max. credits that needed to insert an extent
2499 * to the extent tree.
2500 * When pass the actual path, the caller should calculate credits
2503 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2504 struct ext4_ext_path *path)
2507 int depth = ext_depth(inode);
2510 /* probably there is space in leaf? */
2511 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2512 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2515 * There are some space in the leaf tree, no
2516 * need to account for leaf block credit
2518 * bitmaps and block group descriptor blocks
2519 * and other metadata blocks still need to be
2522 /* 1 bitmap, 1 block group descriptor */
2523 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2528 return ext4_chunk_trans_blocks(inode, nrblocks);
2532 * How many index/leaf blocks need to change/allocate to add @extents extents?
2534 * If we add a single extent, then in the worse case, each tree level
2535 * index/leaf need to be changed in case of the tree split.
2537 * If more extents are inserted, they could cause the whole tree split more
2538 * than once, but this is really rare.
2540 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2545 /* If we are converting the inline data, only one is needed here. */
2546 if (ext4_has_inline_data(inode))
2549 depth = ext_depth(inode);
2559 static inline int get_default_free_blocks_flags(struct inode *inode)
2561 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2562 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2563 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2564 else if (ext4_should_journal_data(inode))
2565 return EXT4_FREE_BLOCKS_FORGET;
2570 * ext4_rereserve_cluster - increment the reserved cluster count when
2571 * freeing a cluster with a pending reservation
2573 * @inode - file containing the cluster
2574 * @lblk - logical block in cluster to be reserved
2576 * Increments the reserved cluster count and adjusts quota in a bigalloc
2577 * file system when freeing a partial cluster containing at least one
2578 * delayed and unwritten block. A partial cluster meeting that
2579 * requirement will have a pending reservation. If so, the
2580 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2581 * defer reserved and allocated space accounting to a subsequent call
2584 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2586 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2587 struct ext4_inode_info *ei = EXT4_I(inode);
2589 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2591 spin_lock(&ei->i_block_reservation_lock);
2592 ei->i_reserved_data_blocks++;
2593 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2594 spin_unlock(&ei->i_block_reservation_lock);
2596 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2597 ext4_remove_pending(inode, lblk);
2600 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2601 struct ext4_extent *ex,
2602 struct partial_cluster *partial,
2603 ext4_lblk_t from, ext4_lblk_t to)
2605 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2606 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2607 ext4_fsblk_t last_pblk, pblk;
2611 /* only extent tail removal is allowed */
2612 if (from < le32_to_cpu(ex->ee_block) ||
2613 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2614 ext4_error(sbi->s_sb,
2615 "strange request: removal(2) %u-%u from %u:%u",
2616 from, to, le32_to_cpu(ex->ee_block), ee_len);
2620 #ifdef EXTENTS_STATS
2621 spin_lock(&sbi->s_ext_stats_lock);
2622 sbi->s_ext_blocks += ee_len;
2623 sbi->s_ext_extents++;
2624 if (ee_len < sbi->s_ext_min)
2625 sbi->s_ext_min = ee_len;
2626 if (ee_len > sbi->s_ext_max)
2627 sbi->s_ext_max = ee_len;
2628 if (ext_depth(inode) > sbi->s_depth_max)
2629 sbi->s_depth_max = ext_depth(inode);
2630 spin_unlock(&sbi->s_ext_stats_lock);
2633 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2636 * if we have a partial cluster, and it's different from the
2637 * cluster of the last block in the extent, we free it
2639 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2641 if (partial->state != initial &&
2642 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2643 if (partial->state == tofree) {
2644 flags = get_default_free_blocks_flags(inode);
2645 if (ext4_is_pending(inode, partial->lblk))
2646 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2647 ext4_free_blocks(handle, inode, NULL,
2648 EXT4_C2B(sbi, partial->pclu),
2649 sbi->s_cluster_ratio, flags);
2650 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2651 ext4_rereserve_cluster(inode, partial->lblk);
2653 partial->state = initial;
2656 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2657 pblk = ext4_ext_pblock(ex) + ee_len - num;
2660 * We free the partial cluster at the end of the extent (if any),
2661 * unless the cluster is used by another extent (partial_cluster
2662 * state is nofree). If a partial cluster exists here, it must be
2663 * shared with the last block in the extent.
2665 flags = get_default_free_blocks_flags(inode);
2667 /* partial, left end cluster aligned, right end unaligned */
2668 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2669 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2670 (partial->state != nofree)) {
2671 if (ext4_is_pending(inode, to))
2672 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2673 ext4_free_blocks(handle, inode, NULL,
2674 EXT4_PBLK_CMASK(sbi, last_pblk),
2675 sbi->s_cluster_ratio, flags);
2676 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2677 ext4_rereserve_cluster(inode, to);
2678 partial->state = initial;
2679 flags = get_default_free_blocks_flags(inode);
2682 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2685 * For bigalloc file systems, we never free a partial cluster
2686 * at the beginning of the extent. Instead, we check to see if we
2687 * need to free it on a subsequent call to ext4_remove_blocks,
2688 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2690 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2691 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2693 /* reset the partial cluster if we've freed past it */
2694 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2695 partial->state = initial;
2698 * If we've freed the entire extent but the beginning is not left
2699 * cluster aligned and is not marked as ineligible for freeing we
2700 * record the partial cluster at the beginning of the extent. It
2701 * wasn't freed by the preceding ext4_free_blocks() call, and we
2702 * need to look farther to the left to determine if it's to be freed
2703 * (not shared with another extent). Else, reset the partial
2704 * cluster - we're either done freeing or the beginning of the
2705 * extent is left cluster aligned.
2707 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2708 if (partial->state == initial) {
2709 partial->pclu = EXT4_B2C(sbi, pblk);
2710 partial->lblk = from;
2711 partial->state = tofree;
2714 partial->state = initial;
2721 * ext4_ext_rm_leaf() Removes the extents associated with the
2722 * blocks appearing between "start" and "end". Both "start"
2723 * and "end" must appear in the same extent or EIO is returned.
2725 * @handle: The journal handle
2726 * @inode: The files inode
2727 * @path: The path to the leaf
2728 * @partial_cluster: The cluster which we'll have to free if all extents
2729 * has been released from it. However, if this value is
2730 * negative, it's a cluster just to the right of the
2731 * punched region and it must not be freed.
2732 * @start: The first block to remove
2733 * @end: The last block to remove
2736 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2737 struct ext4_ext_path *path,
2738 struct partial_cluster *partial,
2739 ext4_lblk_t start, ext4_lblk_t end)
2741 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2742 int err = 0, correct_index = 0;
2743 int depth = ext_depth(inode), credits, revoke_credits;
2744 struct ext4_extent_header *eh;
2747 ext4_lblk_t ex_ee_block;
2748 unsigned short ex_ee_len;
2749 unsigned unwritten = 0;
2750 struct ext4_extent *ex;
2753 /* the header must be checked already in ext4_ext_remove_space() */
2754 ext_debug("truncate since %u in leaf to %u\n", start, end);
2755 if (!path[depth].p_hdr)
2756 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2757 eh = path[depth].p_hdr;
2758 if (unlikely(path[depth].p_hdr == NULL)) {
2759 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2760 return -EFSCORRUPTED;
2762 /* find where to start removing */
2763 ex = path[depth].p_ext;
2765 ex = EXT_LAST_EXTENT(eh);
2767 ex_ee_block = le32_to_cpu(ex->ee_block);
2768 ex_ee_len = ext4_ext_get_actual_len(ex);
2770 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2772 while (ex >= EXT_FIRST_EXTENT(eh) &&
2773 ex_ee_block + ex_ee_len > start) {
2775 if (ext4_ext_is_unwritten(ex))
2780 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2781 unwritten, ex_ee_len);
2782 path[depth].p_ext = ex;
2784 a = ex_ee_block > start ? ex_ee_block : start;
2785 b = ex_ee_block+ex_ee_len - 1 < end ?
2786 ex_ee_block+ex_ee_len - 1 : end;
2788 ext_debug(" border %u:%u\n", a, b);
2790 /* If this extent is beyond the end of the hole, skip it */
2791 if (end < ex_ee_block) {
2793 * We're going to skip this extent and move to another,
2794 * so note that its first cluster is in use to avoid
2795 * freeing it when removing blocks. Eventually, the
2796 * right edge of the truncated/punched region will
2797 * be just to the left.
2799 if (sbi->s_cluster_ratio > 1) {
2800 pblk = ext4_ext_pblock(ex);
2801 partial->pclu = EXT4_B2C(sbi, pblk);
2802 partial->state = nofree;
2805 ex_ee_block = le32_to_cpu(ex->ee_block);
2806 ex_ee_len = ext4_ext_get_actual_len(ex);
2808 } else if (b != ex_ee_block + ex_ee_len - 1) {
2809 EXT4_ERROR_INODE(inode,
2810 "can not handle truncate %u:%u "
2812 start, end, ex_ee_block,
2813 ex_ee_block + ex_ee_len - 1);
2814 err = -EFSCORRUPTED;
2816 } else if (a != ex_ee_block) {
2817 /* remove tail of the extent */
2818 num = a - ex_ee_block;
2820 /* remove whole extent: excellent! */
2824 * 3 for leaf, sb, and inode plus 2 (bmap and group
2825 * descriptor) for each block group; assume two block
2826 * groups plus ex_ee_len/blocks_per_block_group for
2829 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2830 if (ex == EXT_FIRST_EXTENT(eh)) {
2832 credits += (ext_depth(inode)) + 1;
2834 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2836 * We may end up freeing some index blocks and data from the
2837 * punched range. Note that partial clusters are accounted for
2838 * by ext4_free_data_revoke_credits().
2841 ext4_free_metadata_revoke_credits(inode->i_sb,
2843 ext4_free_data_revoke_credits(inode, b - a + 1);
2845 err = ext4_datasem_ensure_credits(handle, inode, credits,
2846 credits, revoke_credits);
2853 err = ext4_ext_get_access(handle, inode, path + depth);
2857 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2862 /* this extent is removed; mark slot entirely unused */
2863 ext4_ext_store_pblock(ex, 0);
2865 ex->ee_len = cpu_to_le16(num);
2867 * Do not mark unwritten if all the blocks in the
2868 * extent have been removed.
2870 if (unwritten && num)
2871 ext4_ext_mark_unwritten(ex);
2873 * If the extent was completely released,
2874 * we need to remove it from the leaf
2877 if (end != EXT_MAX_BLOCKS - 1) {
2879 * For hole punching, we need to scoot all the
2880 * extents up when an extent is removed so that
2881 * we dont have blank extents in the middle
2883 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2884 sizeof(struct ext4_extent));
2886 /* Now get rid of the one at the end */
2887 memset(EXT_LAST_EXTENT(eh), 0,
2888 sizeof(struct ext4_extent));
2890 le16_add_cpu(&eh->eh_entries, -1);
2893 err = ext4_ext_dirty(handle, inode, path + depth);
2897 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2898 ext4_ext_pblock(ex));
2900 ex_ee_block = le32_to_cpu(ex->ee_block);
2901 ex_ee_len = ext4_ext_get_actual_len(ex);
2904 if (correct_index && eh->eh_entries)
2905 err = ext4_ext_correct_indexes(handle, inode, path);
2908 * If there's a partial cluster and at least one extent remains in
2909 * the leaf, free the partial cluster if it isn't shared with the
2910 * current extent. If it is shared with the current extent
2911 * we reset the partial cluster because we've reached the start of the
2912 * truncated/punched region and we're done removing blocks.
2914 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2915 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2916 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2917 int flags = get_default_free_blocks_flags(inode);
2919 if (ext4_is_pending(inode, partial->lblk))
2920 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2921 ext4_free_blocks(handle, inode, NULL,
2922 EXT4_C2B(sbi, partial->pclu),
2923 sbi->s_cluster_ratio, flags);
2924 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2925 ext4_rereserve_cluster(inode, partial->lblk);
2927 partial->state = initial;
2930 /* if this leaf is free, then we should
2931 * remove it from index block above */
2932 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2933 err = ext4_ext_rm_idx(handle, inode, path, depth);
2940 * ext4_ext_more_to_rm:
2941 * returns 1 if current index has to be freed (even partial)
2944 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2946 BUG_ON(path->p_idx == NULL);
2948 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2952 * if truncate on deeper level happened, it wasn't partial,
2953 * so we have to consider current index for truncation
2955 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2960 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2963 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2964 int depth = ext_depth(inode);
2965 struct ext4_ext_path *path = NULL;
2966 struct partial_cluster partial;
2972 partial.state = initial;
2974 ext_debug("truncate since %u to %u\n", start, end);
2976 /* probably first extent we're gonna free will be last in block */
2977 handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2979 ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2981 return PTR_ERR(handle);
2984 trace_ext4_ext_remove_space(inode, start, end, depth);
2987 * Check if we are removing extents inside the extent tree. If that
2988 * is the case, we are going to punch a hole inside the extent tree
2989 * so we have to check whether we need to split the extent covering
2990 * the last block to remove so we can easily remove the part of it
2991 * in ext4_ext_rm_leaf().
2993 if (end < EXT_MAX_BLOCKS - 1) {
2994 struct ext4_extent *ex;
2995 ext4_lblk_t ee_block, ex_end, lblk;
2998 /* find extent for or closest extent to this block */
2999 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
3001 ext4_journal_stop(handle);
3002 return PTR_ERR(path);
3004 depth = ext_depth(inode);
3005 /* Leaf not may not exist only if inode has no blocks at all */
3006 ex = path[depth].p_ext;
3009 EXT4_ERROR_INODE(inode,
3010 "path[%d].p_hdr == NULL",
3012 err = -EFSCORRUPTED;
3017 ee_block = le32_to_cpu(ex->ee_block);
3018 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
3021 * See if the last block is inside the extent, if so split
3022 * the extent at 'end' block so we can easily remove the
3023 * tail of the first part of the split extent in
3024 * ext4_ext_rm_leaf().
3026 if (end >= ee_block && end < ex_end) {
3029 * If we're going to split the extent, note that
3030 * the cluster containing the block after 'end' is
3031 * in use to avoid freeing it when removing blocks.
3033 if (sbi->s_cluster_ratio > 1) {
3034 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
3035 partial.pclu = EXT4_B2C(sbi, pblk);
3036 partial.state = nofree;
3040 * Split the extent in two so that 'end' is the last
3041 * block in the first new extent. Also we should not
3042 * fail removing space due to ENOSPC so try to use
3043 * reserved block if that happens.
3045 err = ext4_force_split_extent_at(handle, inode, &path,
3050 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
3051 partial.state == initial) {
3053 * If we're punching, there's an extent to the right.
3054 * If the partial cluster hasn't been set, set it to
3055 * that extent's first cluster and its state to nofree
3056 * so it won't be freed should it contain blocks to be
3057 * removed. If it's already set (tofree/nofree), we're
3058 * retrying and keep the original partial cluster info
3059 * so a cluster marked tofree as a result of earlier
3060 * extent removal is not lost.
3063 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3068 partial.pclu = EXT4_B2C(sbi, pblk);
3069 partial.state = nofree;
3074 * We start scanning from right side, freeing all the blocks
3075 * after i_size and walking into the tree depth-wise.
3077 depth = ext_depth(inode);
3082 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3084 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3087 ext4_journal_stop(handle);
3090 path[0].p_maxdepth = path[0].p_depth = depth;
3091 path[0].p_hdr = ext_inode_hdr(inode);
3094 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3095 err = -EFSCORRUPTED;
3101 while (i >= 0 && err == 0) {
3103 /* this is leaf block */
3104 err = ext4_ext_rm_leaf(handle, inode, path,
3105 &partial, start, end);
3106 /* root level has p_bh == NULL, brelse() eats this */
3107 brelse(path[i].p_bh);
3108 path[i].p_bh = NULL;
3113 /* this is index block */
3114 if (!path[i].p_hdr) {
3115 ext_debug("initialize header\n");
3116 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3119 if (!path[i].p_idx) {
3120 /* this level hasn't been touched yet */
3121 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3122 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3123 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3125 le16_to_cpu(path[i].p_hdr->eh_entries));
3127 /* we were already here, see at next index */
3131 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3132 i, EXT_FIRST_INDEX(path[i].p_hdr),
3134 if (ext4_ext_more_to_rm(path + i)) {
3135 struct buffer_head *bh;
3136 /* go to the next level */
3137 ext_debug("move to level %d (block %llu)\n",
3138 i + 1, ext4_idx_pblock(path[i].p_idx));
3139 memset(path + i + 1, 0, sizeof(*path));
3140 bh = read_extent_tree_block(inode,
3141 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3144 /* should we reset i_size? */
3148 /* Yield here to deal with large extent trees.
3149 * Should be a no-op if we did IO above. */
3151 if (WARN_ON(i + 1 > depth)) {
3152 err = -EFSCORRUPTED;
3155 path[i + 1].p_bh = bh;
3157 /* save actual number of indexes since this
3158 * number is changed at the next iteration */
3159 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3162 /* we finished processing this index, go up */
3163 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3164 /* index is empty, remove it;
3165 * handle must be already prepared by the
3166 * truncatei_leaf() */
3167 err = ext4_ext_rm_idx(handle, inode, path, i);
3169 /* root level has p_bh == NULL, brelse() eats this */
3170 brelse(path[i].p_bh);
3171 path[i].p_bh = NULL;
3173 ext_debug("return to level %d\n", i);
3177 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3178 path->p_hdr->eh_entries);
3181 * if there's a partial cluster and we have removed the first extent
3182 * in the file, then we also free the partial cluster, if any
3184 if (partial.state == tofree && err == 0) {
3185 int flags = get_default_free_blocks_flags(inode);
3187 if (ext4_is_pending(inode, partial.lblk))
3188 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3189 ext4_free_blocks(handle, inode, NULL,
3190 EXT4_C2B(sbi, partial.pclu),
3191 sbi->s_cluster_ratio, flags);
3192 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3193 ext4_rereserve_cluster(inode, partial.lblk);
3194 partial.state = initial;
3197 /* TODO: flexible tree reduction should be here */
3198 if (path->p_hdr->eh_entries == 0) {
3200 * truncate to zero freed all the tree,
3201 * so we need to correct eh_depth
3203 err = ext4_ext_get_access(handle, inode, path);
3205 ext_inode_hdr(inode)->eh_depth = 0;
3206 ext_inode_hdr(inode)->eh_max =
3207 cpu_to_le16(ext4_ext_space_root(inode, 0));
3208 err = ext4_ext_dirty(handle, inode, path);
3212 ext4_ext_drop_refs(path);
3217 ext4_journal_stop(handle);
3223 * called at mount time
3225 void ext4_ext_init(struct super_block *sb)
3228 * possible initialization would be here
3231 if (ext4_has_feature_extents(sb)) {
3232 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3233 printk(KERN_INFO "EXT4-fs: file extents enabled"
3234 #ifdef AGGRESSIVE_TEST
3235 ", aggressive tests"
3237 #ifdef CHECK_BINSEARCH
3240 #ifdef EXTENTS_STATS
3245 #ifdef EXTENTS_STATS
3246 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3247 EXT4_SB(sb)->s_ext_min = 1 << 30;
3248 EXT4_SB(sb)->s_ext_max = 0;
3254 * called at umount time
3256 void ext4_ext_release(struct super_block *sb)
3258 if (!ext4_has_feature_extents(sb))
3261 #ifdef EXTENTS_STATS
3262 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3263 struct ext4_sb_info *sbi = EXT4_SB(sb);
3264 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3265 sbi->s_ext_blocks, sbi->s_ext_extents,
3266 sbi->s_ext_blocks / sbi->s_ext_extents);
3267 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3268 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3273 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3275 ext4_lblk_t ee_block;
3276 ext4_fsblk_t ee_pblock;
3277 unsigned int ee_len;
3279 ee_block = le32_to_cpu(ex->ee_block);
3280 ee_len = ext4_ext_get_actual_len(ex);
3281 ee_pblock = ext4_ext_pblock(ex);
3286 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3287 EXTENT_STATUS_WRITTEN);
3290 /* FIXME!! we need to try to merge to left or right after zero-out */
3291 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3293 ext4_fsblk_t ee_pblock;
3294 unsigned int ee_len;
3296 ee_len = ext4_ext_get_actual_len(ex);
3297 ee_pblock = ext4_ext_pblock(ex);
3298 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3303 * ext4_split_extent_at() splits an extent at given block.
3305 * @handle: the journal handle
3306 * @inode: the file inode
3307 * @path: the path to the extent
3308 * @split: the logical block where the extent is splitted.
3309 * @split_flags: indicates if the extent could be zeroout if split fails, and
3310 * the states(init or unwritten) of new extents.
3311 * @flags: flags used to insert new extent to extent tree.
3314 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3315 * of which are deterimined by split_flag.
3317 * There are two cases:
3318 * a> the extent are splitted into two extent.
3319 * b> split is not needed, and just mark the extent.
3321 * return 0 on success.
3323 static int ext4_split_extent_at(handle_t *handle,
3324 struct inode *inode,
3325 struct ext4_ext_path **ppath,
3330 struct ext4_ext_path *path = *ppath;
3331 ext4_fsblk_t newblock;
3332 ext4_lblk_t ee_block;
3333 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3334 struct ext4_extent *ex2 = NULL;
3335 unsigned int ee_len, depth;
3338 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3339 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3341 ext_debug("ext4_split_extents_at: inode %lu, logical"
3342 "block %llu\n", inode->i_ino, (unsigned long long)split);
3344 ext4_ext_show_leaf(inode, path);
3346 depth = ext_depth(inode);
3347 ex = path[depth].p_ext;
3348 ee_block = le32_to_cpu(ex->ee_block);
3349 ee_len = ext4_ext_get_actual_len(ex);
3350 newblock = split - ee_block + ext4_ext_pblock(ex);
3352 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3353 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3354 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3355 EXT4_EXT_MARK_UNWRIT1 |
3356 EXT4_EXT_MARK_UNWRIT2));
3358 err = ext4_ext_get_access(handle, inode, path + depth);
3362 if (split == ee_block) {
3364 * case b: block @split is the block that the extent begins with
3365 * then we just change the state of the extent, and splitting
3368 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3369 ext4_ext_mark_unwritten(ex);
3371 ext4_ext_mark_initialized(ex);
3373 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3374 ext4_ext_try_to_merge(handle, inode, path, ex);
3376 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3381 memcpy(&orig_ex, ex, sizeof(orig_ex));
3382 ex->ee_len = cpu_to_le16(split - ee_block);
3383 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3384 ext4_ext_mark_unwritten(ex);
3387 * path may lead to new leaf, not to original leaf any more
3388 * after ext4_ext_insert_extent() returns,
3390 err = ext4_ext_dirty(handle, inode, path + depth);
3392 goto fix_extent_len;
3395 ex2->ee_block = cpu_to_le32(split);
3396 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3397 ext4_ext_store_pblock(ex2, newblock);
3398 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3399 ext4_ext_mark_unwritten(ex2);
3401 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3402 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3403 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3404 if (split_flag & EXT4_EXT_DATA_VALID1) {
3405 err = ext4_ext_zeroout(inode, ex2);
3406 zero_ex.ee_block = ex2->ee_block;
3407 zero_ex.ee_len = cpu_to_le16(
3408 ext4_ext_get_actual_len(ex2));
3409 ext4_ext_store_pblock(&zero_ex,
3410 ext4_ext_pblock(ex2));
3412 err = ext4_ext_zeroout(inode, ex);
3413 zero_ex.ee_block = ex->ee_block;
3414 zero_ex.ee_len = cpu_to_le16(
3415 ext4_ext_get_actual_len(ex));
3416 ext4_ext_store_pblock(&zero_ex,
3417 ext4_ext_pblock(ex));
3420 err = ext4_ext_zeroout(inode, &orig_ex);
3421 zero_ex.ee_block = orig_ex.ee_block;
3422 zero_ex.ee_len = cpu_to_le16(
3423 ext4_ext_get_actual_len(&orig_ex));
3424 ext4_ext_store_pblock(&zero_ex,
3425 ext4_ext_pblock(&orig_ex));
3429 goto fix_extent_len;
3430 /* update the extent length and mark as initialized */
3431 ex->ee_len = cpu_to_le16(ee_len);
3432 ext4_ext_try_to_merge(handle, inode, path, ex);
3433 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3435 goto fix_extent_len;
3437 /* update extent status tree */
3438 err = ext4_zeroout_es(inode, &zero_ex);
3442 goto fix_extent_len;
3445 ext4_ext_show_leaf(inode, path);
3449 ex->ee_len = orig_ex.ee_len;
3450 ext4_ext_dirty(handle, inode, path + path->p_depth);
3455 * ext4_split_extents() splits an extent and mark extent which is covered
3456 * by @map as split_flags indicates
3458 * It may result in splitting the extent into multiple extents (up to three)
3459 * There are three possibilities:
3460 * a> There is no split required
3461 * b> Splits in two extents: Split is happening at either end of the extent
3462 * c> Splits in three extents: Somone is splitting in middle of the extent
3465 static int ext4_split_extent(handle_t *handle,
3466 struct inode *inode,
3467 struct ext4_ext_path **ppath,
3468 struct ext4_map_blocks *map,
3472 struct ext4_ext_path *path = *ppath;
3473 ext4_lblk_t ee_block;
3474 struct ext4_extent *ex;
3475 unsigned int ee_len, depth;
3478 int split_flag1, flags1;
3479 int allocated = map->m_len;
3481 depth = ext_depth(inode);
3482 ex = path[depth].p_ext;
3483 ee_block = le32_to_cpu(ex->ee_block);
3484 ee_len = ext4_ext_get_actual_len(ex);
3485 unwritten = ext4_ext_is_unwritten(ex);
3487 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3488 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3489 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3491 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3492 EXT4_EXT_MARK_UNWRIT2;
3493 if (split_flag & EXT4_EXT_DATA_VALID2)
3494 split_flag1 |= EXT4_EXT_DATA_VALID1;
3495 err = ext4_split_extent_at(handle, inode, ppath,
3496 map->m_lblk + map->m_len, split_flag1, flags1);
3500 allocated = ee_len - (map->m_lblk - ee_block);
3503 * Update path is required because previous ext4_split_extent_at() may
3504 * result in split of original leaf or extent zeroout.
3506 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3508 return PTR_ERR(path);
3509 depth = ext_depth(inode);
3510 ex = path[depth].p_ext;
3512 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3513 (unsigned long) map->m_lblk);
3514 return -EFSCORRUPTED;
3516 unwritten = ext4_ext_is_unwritten(ex);
3519 if (map->m_lblk >= ee_block) {
3520 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3522 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3523 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3524 EXT4_EXT_MARK_UNWRIT2);
3526 err = ext4_split_extent_at(handle, inode, ppath,
3527 map->m_lblk, split_flag1, flags);
3532 ext4_ext_show_leaf(inode, path);
3534 return err ? err : allocated;
3538 * This function is called by ext4_ext_map_blocks() if someone tries to write
3539 * to an unwritten extent. It may result in splitting the unwritten
3540 * extent into multiple extents (up to three - one initialized and two
3542 * There are three possibilities:
3543 * a> There is no split required: Entire extent should be initialized
3544 * b> Splits in two extents: Write is happening at either end of the extent
3545 * c> Splits in three extents: Somone is writing in middle of the extent
3548 * - The extent pointed to by 'path' is unwritten.
3549 * - The extent pointed to by 'path' contains a superset
3550 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3552 * Post-conditions on success:
3553 * - the returned value is the number of blocks beyond map->l_lblk
3554 * that are allocated and initialized.
3555 * It is guaranteed to be >= map->m_len.
3557 static int ext4_ext_convert_to_initialized(handle_t *handle,
3558 struct inode *inode,
3559 struct ext4_map_blocks *map,
3560 struct ext4_ext_path **ppath,
3563 struct ext4_ext_path *path = *ppath;
3564 struct ext4_sb_info *sbi;
3565 struct ext4_extent_header *eh;
3566 struct ext4_map_blocks split_map;
3567 struct ext4_extent zero_ex1, zero_ex2;
3568 struct ext4_extent *ex, *abut_ex;
3569 ext4_lblk_t ee_block, eof_block;
3570 unsigned int ee_len, depth, map_len = map->m_len;
3571 int allocated = 0, max_zeroout = 0;
3573 int split_flag = EXT4_EXT_DATA_VALID2;
3575 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3576 "block %llu, max_blocks %u\n", inode->i_ino,
3577 (unsigned long long)map->m_lblk, map_len);
3579 sbi = EXT4_SB(inode->i_sb);
3580 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3581 inode->i_sb->s_blocksize_bits;
3582 if (eof_block < map->m_lblk + map_len)
3583 eof_block = map->m_lblk + map_len;
3585 depth = ext_depth(inode);
3586 eh = path[depth].p_hdr;
3587 ex = path[depth].p_ext;
3588 ee_block = le32_to_cpu(ex->ee_block);
3589 ee_len = ext4_ext_get_actual_len(ex);
3590 zero_ex1.ee_len = 0;
3591 zero_ex2.ee_len = 0;
3593 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3595 /* Pre-conditions */
3596 BUG_ON(!ext4_ext_is_unwritten(ex));
3597 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3600 * Attempt to transfer newly initialized blocks from the currently
3601 * unwritten extent to its neighbor. This is much cheaper
3602 * than an insertion followed by a merge as those involve costly
3603 * memmove() calls. Transferring to the left is the common case in
3604 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3605 * followed by append writes.
3607 * Limitations of the current logic:
3608 * - L1: we do not deal with writes covering the whole extent.
3609 * This would require removing the extent if the transfer
3611 * - L2: we only attempt to merge with an extent stored in the
3612 * same extent tree node.
3614 if ((map->m_lblk == ee_block) &&
3615 /* See if we can merge left */
3616 (map_len < ee_len) && /*L1*/
3617 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3618 ext4_lblk_t prev_lblk;
3619 ext4_fsblk_t prev_pblk, ee_pblk;
3620 unsigned int prev_len;
3623 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3624 prev_len = ext4_ext_get_actual_len(abut_ex);
3625 prev_pblk = ext4_ext_pblock(abut_ex);
3626 ee_pblk = ext4_ext_pblock(ex);
3629 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3630 * upon those conditions:
3631 * - C1: abut_ex is initialized,
3632 * - C2: abut_ex is logically abutting ex,
3633 * - C3: abut_ex is physically abutting ex,
3634 * - C4: abut_ex can receive the additional blocks without
3635 * overflowing the (initialized) length limit.
3637 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3638 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3639 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3640 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3641 err = ext4_ext_get_access(handle, inode, path + depth);
3645 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3648 /* Shift the start of ex by 'map_len' blocks */
3649 ex->ee_block = cpu_to_le32(ee_block + map_len);
3650 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3651 ex->ee_len = cpu_to_le16(ee_len - map_len);
3652 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3654 /* Extend abut_ex by 'map_len' blocks */
3655 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3657 /* Result: number of initialized blocks past m_lblk */
3658 allocated = map_len;
3660 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3661 (map_len < ee_len) && /*L1*/
3662 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3663 /* See if we can merge right */
3664 ext4_lblk_t next_lblk;
3665 ext4_fsblk_t next_pblk, ee_pblk;
3666 unsigned int next_len;
3669 next_lblk = le32_to_cpu(abut_ex->ee_block);
3670 next_len = ext4_ext_get_actual_len(abut_ex);
3671 next_pblk = ext4_ext_pblock(abut_ex);
3672 ee_pblk = ext4_ext_pblock(ex);
3675 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3676 * upon those conditions:
3677 * - C1: abut_ex is initialized,
3678 * - C2: abut_ex is logically abutting ex,
3679 * - C3: abut_ex is physically abutting ex,
3680 * - C4: abut_ex can receive the additional blocks without
3681 * overflowing the (initialized) length limit.
3683 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3684 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3685 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3686 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3687 err = ext4_ext_get_access(handle, inode, path + depth);
3691 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3694 /* Shift the start of abut_ex by 'map_len' blocks */
3695 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3696 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3697 ex->ee_len = cpu_to_le16(ee_len - map_len);
3698 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3700 /* Extend abut_ex by 'map_len' blocks */
3701 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3703 /* Result: number of initialized blocks past m_lblk */
3704 allocated = map_len;
3708 /* Mark the block containing both extents as dirty */
3709 ext4_ext_dirty(handle, inode, path + depth);
3711 /* Update path to point to the right extent */
3712 path[depth].p_ext = abut_ex;
3715 allocated = ee_len - (map->m_lblk - ee_block);
3717 WARN_ON(map->m_lblk < ee_block);
3719 * It is safe to convert extent to initialized via explicit
3720 * zeroout only if extent is fully inside i_size or new_size.
3722 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3724 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3725 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3726 (inode->i_sb->s_blocksize_bits - 10);
3728 if (IS_ENCRYPTED(inode))
3733 * 1. split the extent into three extents.
3734 * 2. split the extent into two extents, zeroout the head of the first
3736 * 3. split the extent into two extents, zeroout the tail of the second
3738 * 4. split the extent into two extents with out zeroout.
3739 * 5. no splitting needed, just possibly zeroout the head and / or the
3740 * tail of the extent.
3742 split_map.m_lblk = map->m_lblk;
3743 split_map.m_len = map->m_len;
3745 if (max_zeroout && (allocated > split_map.m_len)) {
3746 if (allocated <= max_zeroout) {
3749 cpu_to_le32(split_map.m_lblk +
3752 cpu_to_le16(allocated - split_map.m_len);
3753 ext4_ext_store_pblock(&zero_ex1,
3754 ext4_ext_pblock(ex) + split_map.m_lblk +
3755 split_map.m_len - ee_block);
3756 err = ext4_ext_zeroout(inode, &zero_ex1);
3759 split_map.m_len = allocated;
3761 if (split_map.m_lblk - ee_block + split_map.m_len <
3764 if (split_map.m_lblk != ee_block) {
3765 zero_ex2.ee_block = ex->ee_block;
3766 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3768 ext4_ext_store_pblock(&zero_ex2,
3769 ext4_ext_pblock(ex));
3770 err = ext4_ext_zeroout(inode, &zero_ex2);
3775 split_map.m_len += split_map.m_lblk - ee_block;
3776 split_map.m_lblk = ee_block;
3777 allocated = map->m_len;
3781 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3786 /* If we have gotten a failure, don't zero out status tree */
3788 err = ext4_zeroout_es(inode, &zero_ex1);
3790 err = ext4_zeroout_es(inode, &zero_ex2);
3792 return err ? err : allocated;
3796 * This function is called by ext4_ext_map_blocks() from
3797 * ext4_get_blocks_dio_write() when DIO to write
3798 * to an unwritten extent.
3800 * Writing to an unwritten extent may result in splitting the unwritten
3801 * extent into multiple initialized/unwritten extents (up to three)
3802 * There are three possibilities:
3803 * a> There is no split required: Entire extent should be unwritten
3804 * b> Splits in two extents: Write is happening at either end of the extent
3805 * c> Splits in three extents: Somone is writing in middle of the extent
3807 * This works the same way in the case of initialized -> unwritten conversion.
3809 * One of more index blocks maybe needed if the extent tree grow after
3810 * the unwritten extent split. To prevent ENOSPC occur at the IO
3811 * complete, we need to split the unwritten extent before DIO submit
3812 * the IO. The unwritten extent called at this time will be split
3813 * into three unwritten extent(at most). After IO complete, the part
3814 * being filled will be convert to initialized by the end_io callback function
3815 * via ext4_convert_unwritten_extents().
3817 * Returns the size of unwritten extent to be written on success.
3819 static int ext4_split_convert_extents(handle_t *handle,
3820 struct inode *inode,
3821 struct ext4_map_blocks *map,
3822 struct ext4_ext_path **ppath,
3825 struct ext4_ext_path *path = *ppath;
3826 ext4_lblk_t eof_block;
3827 ext4_lblk_t ee_block;
3828 struct ext4_extent *ex;
3829 unsigned int ee_len;
3830 int split_flag = 0, depth;
3832 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3833 __func__, inode->i_ino,
3834 (unsigned long long)map->m_lblk, map->m_len);
3836 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3837 inode->i_sb->s_blocksize_bits;
3838 if (eof_block < map->m_lblk + map->m_len)
3839 eof_block = map->m_lblk + map->m_len;
3841 * It is safe to convert extent to initialized via explicit
3842 * zeroout only if extent is fully insde i_size or new_size.
3844 depth = ext_depth(inode);
3845 ex = path[depth].p_ext;
3846 ee_block = le32_to_cpu(ex->ee_block);
3847 ee_len = ext4_ext_get_actual_len(ex);
3849 /* Convert to unwritten */
3850 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3851 split_flag |= EXT4_EXT_DATA_VALID1;
3852 /* Convert to initialized */
3853 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3854 split_flag |= ee_block + ee_len <= eof_block ?
3855 EXT4_EXT_MAY_ZEROOUT : 0;
3856 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3858 flags |= EXT4_GET_BLOCKS_PRE_IO;
3859 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3862 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3863 struct inode *inode,
3864 struct ext4_map_blocks *map,
3865 struct ext4_ext_path **ppath)
3867 struct ext4_ext_path *path = *ppath;
3868 struct ext4_extent *ex;
3869 ext4_lblk_t ee_block;
3870 unsigned int ee_len;
3874 depth = ext_depth(inode);
3875 ex = path[depth].p_ext;
3876 ee_block = le32_to_cpu(ex->ee_block);
3877 ee_len = ext4_ext_get_actual_len(ex);
3879 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3880 "block %llu, max_blocks %u\n", inode->i_ino,
3881 (unsigned long long)ee_block, ee_len);
3883 /* If extent is larger than requested it is a clear sign that we still
3884 * have some extent state machine issues left. So extent_split is still
3886 * TODO: Once all related issues will be fixed this situation should be
3889 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3890 #ifdef CONFIG_EXT4_DEBUG
3891 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3892 " len %u; IO logical block %llu, len %u",
3893 inode->i_ino, (unsigned long long)ee_block, ee_len,
3894 (unsigned long long)map->m_lblk, map->m_len);
3896 err = ext4_split_convert_extents(handle, inode, map, ppath,
3897 EXT4_GET_BLOCKS_CONVERT);
3900 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3902 return PTR_ERR(path);
3903 depth = ext_depth(inode);
3904 ex = path[depth].p_ext;
3907 err = ext4_ext_get_access(handle, inode, path + depth);
3910 /* first mark the extent as initialized */
3911 ext4_ext_mark_initialized(ex);
3913 /* note: ext4_ext_correct_indexes() isn't needed here because
3914 * borders are not changed
3916 ext4_ext_try_to_merge(handle, inode, path, ex);
3918 /* Mark modified extent as dirty */
3919 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3921 ext4_ext_show_leaf(inode, path);
3926 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3928 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3930 struct ext4_ext_path *path,
3934 struct ext4_extent_header *eh;
3935 struct ext4_extent *last_ex;
3937 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3940 depth = ext_depth(inode);
3941 eh = path[depth].p_hdr;
3944 * We're going to remove EOFBLOCKS_FL entirely in future so we
3945 * do not care for this case anymore. Simply remove the flag
3946 * if there are no extents.
3948 if (unlikely(!eh->eh_entries))
3950 last_ex = EXT_LAST_EXTENT(eh);
3952 * We should clear the EOFBLOCKS_FL flag if we are writing the
3953 * last block in the last extent in the file. We test this by
3954 * first checking to see if the caller to
3955 * ext4_ext_get_blocks() was interested in the last block (or
3956 * a block beyond the last block) in the current extent. If
3957 * this turns out to be false, we can bail out from this
3958 * function immediately.
3960 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3961 ext4_ext_get_actual_len(last_ex))
3964 * If the caller does appear to be planning to write at or
3965 * beyond the end of the current extent, we then test to see
3966 * if the current extent is the last extent in the file, by
3967 * checking to make sure it was reached via the rightmost node
3968 * at each level of the tree.
3970 for (i = depth-1; i >= 0; i--)
3971 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3974 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3975 return ext4_mark_inode_dirty(handle, inode);
3979 convert_initialized_extent(handle_t *handle, struct inode *inode,
3980 struct ext4_map_blocks *map,
3981 struct ext4_ext_path **ppath,
3982 unsigned int allocated)
3984 struct ext4_ext_path *path = *ppath;
3985 struct ext4_extent *ex;
3986 ext4_lblk_t ee_block;
3987 unsigned int ee_len;
3992 * Make sure that the extent is no bigger than we support with
3995 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3996 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3998 depth = ext_depth(inode);
3999 ex = path[depth].p_ext;
4000 ee_block = le32_to_cpu(ex->ee_block);
4001 ee_len = ext4_ext_get_actual_len(ex);
4003 ext_debug("%s: inode %lu, logical"
4004 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
4005 (unsigned long long)ee_block, ee_len);
4007 if (ee_block != map->m_lblk || ee_len > map->m_len) {
4008 err = ext4_split_convert_extents(handle, inode, map, ppath,
4009 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
4012 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
4014 return PTR_ERR(path);
4015 depth = ext_depth(inode);
4016 ex = path[depth].p_ext;
4018 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
4019 (unsigned long) map->m_lblk);
4020 return -EFSCORRUPTED;
4024 err = ext4_ext_get_access(handle, inode, path + depth);
4027 /* first mark the extent as unwritten */
4028 ext4_ext_mark_unwritten(ex);
4030 /* note: ext4_ext_correct_indexes() isn't needed here because
4031 * borders are not changed
4033 ext4_ext_try_to_merge(handle, inode, path, ex);
4035 /* Mark modified extent as dirty */
4036 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4039 ext4_ext_show_leaf(inode, path);
4041 ext4_update_inode_fsync_trans(handle, inode, 1);
4042 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4045 map->m_flags |= EXT4_MAP_UNWRITTEN;
4046 if (allocated > map->m_len)
4047 allocated = map->m_len;
4048 map->m_len = allocated;
4053 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4054 struct ext4_map_blocks *map,
4055 struct ext4_ext_path **ppath, int flags,
4056 unsigned int allocated, ext4_fsblk_t newblock)
4058 struct ext4_ext_path *path = *ppath;
4062 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4063 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4064 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4066 ext4_ext_show_leaf(inode, path);
4069 * When writing into unwritten space, we should not fail to
4070 * allocate metadata blocks for the new extent block if needed.
4072 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4074 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4075 allocated, newblock);
4077 /* get_block() before submit the IO, split the extent */
4078 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4079 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4080 flags | EXT4_GET_BLOCKS_CONVERT);
4083 map->m_flags |= EXT4_MAP_UNWRITTEN;
4086 /* IO end_io complete, convert the filled extent to written */
4087 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4088 if (flags & EXT4_GET_BLOCKS_ZERO) {
4089 if (allocated > map->m_len)
4090 allocated = map->m_len;
4091 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4096 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4099 ext4_update_inode_fsync_trans(handle, inode, 1);
4100 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4104 map->m_flags |= EXT4_MAP_MAPPED;
4105 map->m_pblk = newblock;
4106 if (allocated > map->m_len)
4107 allocated = map->m_len;
4108 map->m_len = allocated;
4111 /* buffered IO case */
4113 * repeat fallocate creation request
4114 * we already have an unwritten extent
4116 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4117 map->m_flags |= EXT4_MAP_UNWRITTEN;
4121 /* buffered READ or buffered write_begin() lookup */
4122 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4124 * We have blocks reserved already. We
4125 * return allocated blocks so that delalloc
4126 * won't do block reservation for us. But
4127 * the buffer head will be unmapped so that
4128 * a read from the block returns 0s.
4130 map->m_flags |= EXT4_MAP_UNWRITTEN;
4134 /* buffered write, writepage time, convert*/
4135 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4137 ext4_update_inode_fsync_trans(handle, inode, 1);
4144 map->m_flags |= EXT4_MAP_NEW;
4145 if (allocated > map->m_len)
4146 allocated = map->m_len;
4147 map->m_len = allocated;
4150 map->m_flags |= EXT4_MAP_MAPPED;
4151 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4152 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4158 if (allocated > map->m_len)
4159 allocated = map->m_len;
4160 ext4_ext_show_leaf(inode, path);
4161 map->m_pblk = newblock;
4162 map->m_len = allocated;
4164 return err ? err : allocated;
4168 * get_implied_cluster_alloc - check to see if the requested
4169 * allocation (in the map structure) overlaps with a cluster already
4170 * allocated in an extent.
4171 * @sb The filesystem superblock structure
4172 * @map The requested lblk->pblk mapping
4173 * @ex The extent structure which might contain an implied
4174 * cluster allocation
4176 * This function is called by ext4_ext_map_blocks() after we failed to
4177 * find blocks that were already in the inode's extent tree. Hence,
4178 * we know that the beginning of the requested region cannot overlap
4179 * the extent from the inode's extent tree. There are three cases we
4180 * want to catch. The first is this case:
4182 * |--- cluster # N--|
4183 * |--- extent ---| |---- requested region ---|
4186 * The second case that we need to test for is this one:
4188 * |--------- cluster # N ----------------|
4189 * |--- requested region --| |------- extent ----|
4190 * |=======================|
4192 * The third case is when the requested region lies between two extents
4193 * within the same cluster:
4194 * |------------- cluster # N-------------|
4195 * |----- ex -----| |---- ex_right ----|
4196 * |------ requested region ------|
4197 * |================|
4199 * In each of the above cases, we need to set the map->m_pblk and
4200 * map->m_len so it corresponds to the return the extent labelled as
4201 * "|====|" from cluster #N, since it is already in use for data in
4202 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4203 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4204 * as a new "allocated" block region. Otherwise, we will return 0 and
4205 * ext4_ext_map_blocks() will then allocate one or more new clusters
4206 * by calling ext4_mb_new_blocks().
4208 static int get_implied_cluster_alloc(struct super_block *sb,
4209 struct ext4_map_blocks *map,
4210 struct ext4_extent *ex,
4211 struct ext4_ext_path *path)
4213 struct ext4_sb_info *sbi = EXT4_SB(sb);
4214 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4215 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4216 ext4_lblk_t rr_cluster_start;
4217 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4218 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4219 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4221 /* The extent passed in that we are trying to match */
4222 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4223 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4225 /* The requested region passed into ext4_map_blocks() */
4226 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4228 if ((rr_cluster_start == ex_cluster_end) ||
4229 (rr_cluster_start == ex_cluster_start)) {
4230 if (rr_cluster_start == ex_cluster_end)
4231 ee_start += ee_len - 1;
4232 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4233 map->m_len = min(map->m_len,
4234 (unsigned) sbi->s_cluster_ratio - c_offset);
4236 * Check for and handle this case:
4238 * |--------- cluster # N-------------|
4239 * |------- extent ----|
4240 * |--- requested region ---|
4244 if (map->m_lblk < ee_block)
4245 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4248 * Check for the case where there is already another allocated
4249 * block to the right of 'ex' but before the end of the cluster.
4251 * |------------- cluster # N-------------|
4252 * |----- ex -----| |---- ex_right ----|
4253 * |------ requested region ------|
4254 * |================|
4256 if (map->m_lblk > ee_block) {
4257 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4258 map->m_len = min(map->m_len, next - map->m_lblk);
4261 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4265 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4271 * Block allocation/map/preallocation routine for extents based files
4274 * Need to be called with
4275 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4276 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4278 * return > 0, number of of blocks already mapped/allocated
4279 * if create == 0 and these are pre-allocated blocks
4280 * buffer head is unmapped
4281 * otherwise blocks are mapped
4283 * return = 0, if plain look up failed (blocks have not been allocated)
4284 * buffer head is unmapped
4286 * return < 0, error case.
4288 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4289 struct ext4_map_blocks *map, int flags)
4291 struct ext4_ext_path *path = NULL;
4292 struct ext4_extent newex, *ex, *ex2;
4293 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4294 ext4_fsblk_t newblock = 0;
4295 int free_on_err = 0, err = 0, depth, ret;
4296 unsigned int allocated = 0, offset = 0;
4297 unsigned int allocated_clusters = 0;
4298 struct ext4_allocation_request ar;
4299 ext4_lblk_t cluster_offset;
4300 bool map_from_cluster = false;
4302 ext_debug("blocks %u/%u requested for inode %lu\n",
4303 map->m_lblk, map->m_len, inode->i_ino);
4304 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4306 /* find extent for this block */
4307 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4309 err = PTR_ERR(path);
4314 depth = ext_depth(inode);
4317 * consistent leaf must not be empty;
4318 * this situation is possible, though, _during_ tree modification;
4319 * this is why assert can't be put in ext4_find_extent()
4321 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4322 EXT4_ERROR_INODE(inode, "bad extent address "
4323 "lblock: %lu, depth: %d pblock %lld",
4324 (unsigned long) map->m_lblk, depth,
4325 path[depth].p_block);
4326 err = -EFSCORRUPTED;
4330 ex = path[depth].p_ext;
4332 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4333 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4334 unsigned short ee_len;
4338 * unwritten extents are treated as holes, except that
4339 * we split out initialized portions during a write.
4341 ee_len = ext4_ext_get_actual_len(ex);
4343 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4345 /* if found extent covers block, simply return it */
4346 if (in_range(map->m_lblk, ee_block, ee_len)) {
4347 newblock = map->m_lblk - ee_block + ee_start;
4348 /* number of remaining blocks in the extent */
4349 allocated = ee_len - (map->m_lblk - ee_block);
4350 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4351 ee_block, ee_len, newblock);
4354 * If the extent is initialized check whether the
4355 * caller wants to convert it to unwritten.
4357 if ((!ext4_ext_is_unwritten(ex)) &&
4358 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4359 allocated = convert_initialized_extent(
4360 handle, inode, map, &path,
4363 } else if (!ext4_ext_is_unwritten(ex))
4366 ret = ext4_ext_handle_unwritten_extents(
4367 handle, inode, map, &path, flags,
4368 allocated, newblock);
4378 * requested block isn't allocated yet;
4379 * we couldn't try to create block if create flag is zero
4381 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4382 ext4_lblk_t hole_start, hole_len;
4384 hole_start = map->m_lblk;
4385 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4387 * put just found gap into cache to speed up
4388 * subsequent requests
4390 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4392 /* Update hole_len to reflect hole size after map->m_lblk */
4393 if (hole_start != map->m_lblk)
4394 hole_len -= map->m_lblk - hole_start;
4396 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4402 * Okay, we need to do block allocation.
4404 newex.ee_block = cpu_to_le32(map->m_lblk);
4405 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4408 * If we are doing bigalloc, check to see if the extent returned
4409 * by ext4_find_extent() implies a cluster we can use.
4411 if (cluster_offset && ex &&
4412 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4413 ar.len = allocated = map->m_len;
4414 newblock = map->m_pblk;
4415 map_from_cluster = true;
4416 goto got_allocated_blocks;
4419 /* find neighbour allocated blocks */
4420 ar.lleft = map->m_lblk;
4421 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4424 ar.lright = map->m_lblk;
4426 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4430 /* Check if the extent after searching to the right implies a
4431 * cluster we can use. */
4432 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4433 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4434 ar.len = allocated = map->m_len;
4435 newblock = map->m_pblk;
4436 map_from_cluster = true;
4437 goto got_allocated_blocks;
4441 * See if request is beyond maximum number of blocks we can have in
4442 * a single extent. For an initialized extent this limit is
4443 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4444 * EXT_UNWRITTEN_MAX_LEN.
4446 if (map->m_len > EXT_INIT_MAX_LEN &&
4447 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4448 map->m_len = EXT_INIT_MAX_LEN;
4449 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4450 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4451 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4453 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4454 newex.ee_len = cpu_to_le16(map->m_len);
4455 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4457 allocated = ext4_ext_get_actual_len(&newex);
4459 allocated = map->m_len;
4461 /* allocate new block */
4463 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4464 ar.logical = map->m_lblk;
4466 * We calculate the offset from the beginning of the cluster
4467 * for the logical block number, since when we allocate a
4468 * physical cluster, the physical block should start at the
4469 * same offset from the beginning of the cluster. This is
4470 * needed so that future calls to get_implied_cluster_alloc()
4473 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4474 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4476 ar.logical -= offset;
4477 if (S_ISREG(inode->i_mode))
4478 ar.flags = EXT4_MB_HINT_DATA;
4480 /* disable in-core preallocation for non-regular files */
4482 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4483 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4484 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4485 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4486 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4487 ar.flags |= EXT4_MB_USE_RESERVED;
4488 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4491 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4492 ar.goal, newblock, allocated);
4494 allocated_clusters = ar.len;
4495 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4496 if (ar.len > allocated)
4499 got_allocated_blocks:
4500 /* try to insert new extent into found leaf and return */
4501 ext4_ext_store_pblock(&newex, newblock + offset);
4502 newex.ee_len = cpu_to_le16(ar.len);
4503 /* Mark unwritten */
4504 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4505 ext4_ext_mark_unwritten(&newex);
4506 map->m_flags |= EXT4_MAP_UNWRITTEN;
4510 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4511 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4514 err = ext4_ext_insert_extent(handle, inode, &path,
4517 if (err && free_on_err) {
4518 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4519 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4520 /* free data blocks we just allocated */
4521 /* not a good idea to call discard here directly,
4522 * but otherwise we'd need to call it every free() */
4523 ext4_discard_preallocations(inode);
4524 ext4_free_blocks(handle, inode, NULL, newblock,
4525 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4529 /* previous routine could use block we allocated */
4530 newblock = ext4_ext_pblock(&newex);
4531 allocated = ext4_ext_get_actual_len(&newex);
4532 if (allocated > map->m_len)
4533 allocated = map->m_len;
4534 map->m_flags |= EXT4_MAP_NEW;
4537 * Reduce the reserved cluster count to reflect successful deferred
4538 * allocation of delayed allocated clusters or direct allocation of
4539 * clusters discovered to be delayed allocated. Once allocated, a
4540 * cluster is not included in the reserved count.
4542 if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4543 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4545 * When allocating delayed allocated clusters, simply
4546 * reduce the reserved cluster count and claim quota
4548 ext4_da_update_reserve_space(inode, allocated_clusters,
4551 ext4_lblk_t lblk, len;
4555 * When allocating non-delayed allocated clusters
4556 * (from fallocate, filemap, DIO, or clusters
4557 * allocated when delalloc has been disabled by
4558 * ext4_nonda_switch), reduce the reserved cluster
4559 * count by the number of allocated clusters that
4560 * have previously been delayed allocated. Quota
4561 * has been claimed by ext4_mb_new_blocks() above,
4562 * so release the quota reservations made for any
4563 * previously delayed allocated clusters.
4565 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4566 len = allocated_clusters << sbi->s_cluster_bits;
4567 n = ext4_es_delayed_clu(inode, lblk, len);
4569 ext4_da_update_reserve_space(inode, (int) n, 0);
4574 * Cache the extent and update transaction to commit on fdatasync only
4575 * when it is _not_ an unwritten extent.
4577 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4578 ext4_update_inode_fsync_trans(handle, inode, 1);
4580 ext4_update_inode_fsync_trans(handle, inode, 0);
4582 if (allocated > map->m_len)
4583 allocated = map->m_len;
4584 ext4_ext_show_leaf(inode, path);
4585 map->m_flags |= EXT4_MAP_MAPPED;
4586 map->m_pblk = newblock;
4587 map->m_len = allocated;
4589 ext4_ext_drop_refs(path);
4592 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4593 err ? err : allocated);
4594 return err ? err : allocated;
4597 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4599 struct super_block *sb = inode->i_sb;
4600 ext4_lblk_t last_block;
4604 * TODO: optimization is possible here.
4605 * Probably we need not scan at all,
4606 * because page truncation is enough.
4609 /* we have to know where to truncate from in crash case */
4610 EXT4_I(inode)->i_disksize = inode->i_size;
4611 err = ext4_mark_inode_dirty(handle, inode);
4615 last_block = (inode->i_size + sb->s_blocksize - 1)
4616 >> EXT4_BLOCK_SIZE_BITS(sb);
4618 err = ext4_es_remove_extent(inode, last_block,
4619 EXT_MAX_BLOCKS - last_block);
4620 if (err == -ENOMEM) {
4622 congestion_wait(BLK_RW_ASYNC, HZ/50);
4627 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4630 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4631 ext4_lblk_t len, loff_t new_size,
4634 struct inode *inode = file_inode(file);
4640 struct ext4_map_blocks map;
4641 unsigned int credits;
4644 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4645 map.m_lblk = offset;
4648 * Don't normalize the request if it can fit in one extent so
4649 * that it doesn't get unnecessarily split into multiple
4652 if (len <= EXT_UNWRITTEN_MAX_LEN)
4653 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4656 * credits to insert 1 extent into extent tree
4658 credits = ext4_chunk_trans_blocks(inode, len);
4659 depth = ext_depth(inode);
4662 while (ret >= 0 && len) {
4664 * Recalculate credits when extent tree depth changes.
4666 if (depth != ext_depth(inode)) {
4667 credits = ext4_chunk_trans_blocks(inode, len);
4668 depth = ext_depth(inode);
4671 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4673 if (IS_ERR(handle)) {
4674 ret = PTR_ERR(handle);
4677 ret = ext4_map_blocks(handle, inode, &map, flags);
4679 ext4_debug("inode #%lu: block %u: len %u: "
4680 "ext4_ext_map_blocks returned %d",
4681 inode->i_ino, map.m_lblk,
4683 ext4_mark_inode_dirty(handle, inode);
4684 ret2 = ext4_journal_stop(handle);
4688 map.m_len = len = len - ret;
4689 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4690 inode->i_ctime = current_time(inode);
4692 if (epos > new_size)
4694 if (ext4_update_inode_size(inode, epos) & 0x1)
4695 inode->i_mtime = inode->i_ctime;
4697 if (epos > inode->i_size)
4698 ext4_set_inode_flag(inode,
4699 EXT4_INODE_EOFBLOCKS);
4701 ext4_mark_inode_dirty(handle, inode);
4702 ext4_update_inode_fsync_trans(handle, inode, 1);
4703 ret2 = ext4_journal_stop(handle);
4707 if (ret == -ENOSPC &&
4708 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4713 return ret > 0 ? ret2 : ret;
4716 static long ext4_zero_range(struct file *file, loff_t offset,
4717 loff_t len, int mode)
4719 struct inode *inode = file_inode(file);
4720 handle_t *handle = NULL;
4721 unsigned int max_blocks;
4722 loff_t new_size = 0;
4726 int partial_begin, partial_end;
4729 unsigned int blkbits = inode->i_blkbits;
4731 trace_ext4_zero_range(inode, offset, len, mode);
4733 if (!S_ISREG(inode->i_mode))
4736 /* Call ext4_force_commit to flush all data in case of data=journal. */
4737 if (ext4_should_journal_data(inode)) {
4738 ret = ext4_force_commit(inode->i_sb);
4744 * Round up offset. This is not fallocate, we neet to zero out
4745 * blocks, so convert interior block aligned part of the range to
4746 * unwritten and possibly manually zero out unaligned parts of the
4749 start = round_up(offset, 1 << blkbits);
4750 end = round_down((offset + len), 1 << blkbits);
4752 if (start < offset || end > offset + len)
4754 partial_begin = offset & ((1 << blkbits) - 1);
4755 partial_end = (offset + len) & ((1 << blkbits) - 1);
4757 lblk = start >> blkbits;
4758 max_blocks = (end >> blkbits);
4759 if (max_blocks < lblk)
4767 * Indirect files do not support unwritten extnets
4769 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4774 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4775 (offset + len > i_size_read(inode) ||
4776 offset + len > EXT4_I(inode)->i_disksize)) {
4777 new_size = offset + len;
4778 ret = inode_newsize_ok(inode, new_size);
4783 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4784 if (mode & FALLOC_FL_KEEP_SIZE)
4785 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4787 /* Wait all existing dio workers, newcomers will block on i_mutex */
4788 inode_dio_wait(inode);
4790 /* Preallocate the range including the unaligned edges */
4791 if (partial_begin || partial_end) {
4792 ret = ext4_alloc_file_blocks(file,
4793 round_down(offset, 1 << blkbits) >> blkbits,
4794 (round_up((offset + len), 1 << blkbits) -
4795 round_down(offset, 1 << blkbits)) >> blkbits,
4802 /* Zero range excluding the unaligned edges */
4803 if (max_blocks > 0) {
4804 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4808 * Prevent page faults from reinstantiating pages we have
4809 * released from page cache.
4811 down_write(&EXT4_I(inode)->i_mmap_sem);
4813 ret = ext4_break_layouts(inode);
4815 up_write(&EXT4_I(inode)->i_mmap_sem);
4819 ret = ext4_update_disksize_before_punch(inode, offset, len);
4821 up_write(&EXT4_I(inode)->i_mmap_sem);
4824 /* Now release the pages and zero block aligned part of pages */
4825 truncate_pagecache_range(inode, start, end - 1);
4826 inode->i_mtime = inode->i_ctime = current_time(inode);
4828 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4830 up_write(&EXT4_I(inode)->i_mmap_sem);
4834 if (!partial_begin && !partial_end)
4838 * In worst case we have to writeout two nonadjacent unwritten
4839 * blocks and update the inode
4841 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4842 if (ext4_should_journal_data(inode))
4844 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4845 if (IS_ERR(handle)) {
4846 ret = PTR_ERR(handle);
4847 ext4_std_error(inode->i_sb, ret);
4851 inode->i_mtime = inode->i_ctime = current_time(inode);
4853 ext4_update_inode_size(inode, new_size);
4856 * Mark that we allocate beyond EOF so the subsequent truncate
4857 * can proceed even if the new size is the same as i_size.
4859 if ((offset + len) > i_size_read(inode))
4860 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4862 ext4_mark_inode_dirty(handle, inode);
4864 /* Zero out partial block at the edges of the range */
4865 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4867 ext4_update_inode_fsync_trans(handle, inode, 1);
4869 if (file->f_flags & O_SYNC)
4870 ext4_handle_sync(handle);
4872 ext4_journal_stop(handle);
4874 inode_unlock(inode);
4879 * preallocate space for a file. This implements ext4's fallocate file
4880 * operation, which gets called from sys_fallocate system call.
4881 * For block-mapped files, posix_fallocate should fall back to the method
4882 * of writing zeroes to the required new blocks (the same behavior which is
4883 * expected for file systems which do not support fallocate() system call).
4885 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4887 struct inode *inode = file_inode(file);
4888 loff_t new_size = 0;
4889 unsigned int max_blocks;
4893 unsigned int blkbits = inode->i_blkbits;
4896 * Encrypted inodes can't handle collapse range or insert
4897 * range since we would need to re-encrypt blocks with a
4898 * different IV or XTS tweak (which are based on the logical
4901 * XXX It's not clear why zero range isn't working, but we'll
4902 * leave it disabled for encrypted inodes for now. This is a
4903 * bug we should fix....
4905 if (IS_ENCRYPTED(inode) &&
4906 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4907 FALLOC_FL_ZERO_RANGE)))
4910 /* Return error if mode is not supported */
4911 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4912 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4913 FALLOC_FL_INSERT_RANGE))
4916 if (mode & FALLOC_FL_PUNCH_HOLE)
4917 return ext4_punch_hole(inode, offset, len);
4919 ret = ext4_convert_inline_data(inode);
4923 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4924 return ext4_collapse_range(inode, offset, len);
4926 if (mode & FALLOC_FL_INSERT_RANGE)
4927 return ext4_insert_range(inode, offset, len);
4929 if (mode & FALLOC_FL_ZERO_RANGE)
4930 return ext4_zero_range(file, offset, len, mode);
4932 trace_ext4_fallocate_enter(inode, offset, len, mode);
4933 lblk = offset >> blkbits;
4935 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4936 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4937 if (mode & FALLOC_FL_KEEP_SIZE)
4938 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4943 * We only support preallocation for extent-based files only
4945 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4950 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4951 (offset + len > i_size_read(inode) ||
4952 offset + len > EXT4_I(inode)->i_disksize)) {
4953 new_size = offset + len;
4954 ret = inode_newsize_ok(inode, new_size);
4959 /* Wait all existing dio workers, newcomers will block on i_mutex */
4960 inode_dio_wait(inode);
4962 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4966 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4967 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4968 EXT4_I(inode)->i_sync_tid);
4971 inode_unlock(inode);
4972 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4977 * This function convert a range of blocks to written extents
4978 * The caller of this function will pass the start offset and the size.
4979 * all unwritten extents within this range will be converted to
4982 * This function is called from the direct IO end io call back
4983 * function, to convert the fallocated extents after IO is completed.
4984 * Returns 0 on success.
4986 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4987 loff_t offset, ssize_t len)
4989 unsigned int max_blocks;
4992 struct ext4_map_blocks map;
4993 unsigned int credits, blkbits = inode->i_blkbits;
4995 map.m_lblk = offset >> blkbits;
4996 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4999 * This is somewhat ugly but the idea is clear: When transaction is
5000 * reserved, everything goes into it. Otherwise we rather start several
5001 * smaller transactions for conversion of each extent separately.
5004 handle = ext4_journal_start_reserved(handle,
5005 EXT4_HT_EXT_CONVERT);
5007 return PTR_ERR(handle);
5011 * credits to insert 1 extent into extent tree
5013 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5015 while (ret >= 0 && ret < max_blocks) {
5017 map.m_len = (max_blocks -= ret);
5019 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5021 if (IS_ERR(handle)) {
5022 ret = PTR_ERR(handle);
5026 ret = ext4_map_blocks(handle, inode, &map,
5027 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5029 ext4_warning(inode->i_sb,
5030 "inode #%lu: block %u: len %u: "
5031 "ext4_ext_map_blocks returned %d",
5032 inode->i_ino, map.m_lblk,
5034 ext4_mark_inode_dirty(handle, inode);
5036 ret2 = ext4_journal_stop(handle);
5037 if (ret <= 0 || ret2)
5041 ret2 = ext4_journal_stop(handle);
5042 return ret > 0 ? ret2 : ret;
5046 * If newes is not existing extent (newes->ec_pblk equals zero) find
5047 * delayed extent at start of newes and update newes accordingly and
5048 * return start of the next delayed extent.
5050 * If newes is existing extent (newes->ec_pblk is not equal zero)
5051 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5052 * extent found. Leave newes unmodified.
5054 static int ext4_find_delayed_extent(struct inode *inode,
5055 struct extent_status *newes)
5057 struct extent_status es;
5058 ext4_lblk_t block, next_del;
5060 if (newes->es_pblk == 0) {
5061 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5063 newes->es_lblk + newes->es_len - 1,
5067 * No extent in extent-tree contains block @newes->es_pblk,
5068 * then the block may stay in 1)a hole or 2)delayed-extent.
5074 if (es.es_lblk > newes->es_lblk) {
5076 newes->es_len = min(es.es_lblk - newes->es_lblk,
5081 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5084 block = newes->es_lblk + newes->es_len;
5085 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5086 EXT_MAX_BLOCKS, &es);
5088 next_del = EXT_MAX_BLOCKS;
5090 next_del = es.es_lblk;
5095 static int ext4_xattr_fiemap(struct inode *inode,
5096 struct fiemap_extent_info *fieinfo)
5100 __u32 flags = FIEMAP_EXTENT_LAST;
5101 int blockbits = inode->i_sb->s_blocksize_bits;
5105 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5106 struct ext4_iloc iloc;
5107 int offset; /* offset of xattr in inode */
5109 error = ext4_get_inode_loc(inode, &iloc);
5112 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5113 offset = EXT4_GOOD_OLD_INODE_SIZE +
5114 EXT4_I(inode)->i_extra_isize;
5116 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5117 flags |= FIEMAP_EXTENT_DATA_INLINE;
5119 } else { /* external block */
5120 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5121 length = inode->i_sb->s_blocksize;
5125 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5127 return (error < 0 ? error : 0);
5130 static int _ext4_fiemap(struct inode *inode,
5131 struct fiemap_extent_info *fieinfo,
5132 __u64 start, __u64 len,
5133 int (*fill)(struct inode *, ext4_lblk_t,
5135 struct fiemap_extent_info *))
5137 ext4_lblk_t start_blk;
5138 u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5142 if (ext4_has_inline_data(inode)) {
5145 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5152 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5153 error = ext4_ext_precache(inode);
5156 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5159 /* fallback to generic here if not in extents fmt */
5160 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5161 fill == ext4_fill_fiemap_extents)
5162 return generic_block_fiemap(inode, fieinfo, start, len,
5165 if (fill == ext4_fill_es_cache_info)
5166 ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5167 if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5170 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5171 error = ext4_xattr_fiemap(inode, fieinfo);
5173 ext4_lblk_t len_blks;
5176 start_blk = start >> inode->i_sb->s_blocksize_bits;
5177 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5178 if (last_blk >= EXT_MAX_BLOCKS)
5179 last_blk = EXT_MAX_BLOCKS-1;
5180 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5183 * Walk the extent tree gathering extent information
5184 * and pushing extents back to the user.
5186 error = fill(inode, start_blk, len_blks, fieinfo);
5191 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5192 __u64 start, __u64 len)
5194 return _ext4_fiemap(inode, fieinfo, start, len,
5195 ext4_fill_fiemap_extents);
5198 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5199 __u64 start, __u64 len)
5201 if (ext4_has_inline_data(inode)) {
5204 down_read(&EXT4_I(inode)->xattr_sem);
5205 has_inline = ext4_has_inline_data(inode);
5206 up_read(&EXT4_I(inode)->xattr_sem);
5211 return _ext4_fiemap(inode, fieinfo, start, len,
5212 ext4_fill_es_cache_info);
5218 * Function to access the path buffer for marking it dirty.
5219 * It also checks if there are sufficient credits left in the journal handle
5223 ext4_access_path(handle_t *handle, struct inode *inode,
5224 struct ext4_ext_path *path)
5228 if (!ext4_handle_valid(handle))
5232 * Check if need to extend journal credits
5233 * 3 for leaf, sb, and inode plus 2 (bmap and group
5234 * descriptor) for each block group; assume two block
5237 credits = ext4_writepage_trans_blocks(inode);
5238 err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
5242 err = ext4_ext_get_access(handle, inode, path);
5247 * ext4_ext_shift_path_extents:
5248 * Shift the extents of a path structure lying between path[depth].p_ext
5249 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5250 * if it is right shift or left shift operation.
5253 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5254 struct inode *inode, handle_t *handle,
5255 enum SHIFT_DIRECTION SHIFT)
5258 struct ext4_extent *ex_start, *ex_last;
5260 depth = path->p_depth;
5262 while (depth >= 0) {
5263 if (depth == path->p_depth) {
5264 ex_start = path[depth].p_ext;
5266 return -EFSCORRUPTED;
5268 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5270 err = ext4_access_path(handle, inode, path + depth);
5274 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5277 while (ex_start <= ex_last) {
5278 if (SHIFT == SHIFT_LEFT) {
5279 le32_add_cpu(&ex_start->ee_block,
5281 /* Try to merge to the left. */
5283 EXT_FIRST_EXTENT(path[depth].p_hdr))
5285 ext4_ext_try_to_merge_right(inode,
5286 path, ex_start - 1))
5291 le32_add_cpu(&ex_last->ee_block, shift);
5292 ext4_ext_try_to_merge_right(inode, path,
5297 err = ext4_ext_dirty(handle, inode, path + depth);
5301 if (--depth < 0 || !update)
5305 /* Update index too */
5306 err = ext4_access_path(handle, inode, path + depth);
5310 if (SHIFT == SHIFT_LEFT)
5311 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5313 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5314 err = ext4_ext_dirty(handle, inode, path + depth);
5318 /* we are done if current index is not a starting index */
5319 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5330 * ext4_ext_shift_extents:
5331 * All the extents which lies in the range from @start to the last allocated
5332 * block for the @inode are shifted either towards left or right (depending
5333 * upon @SHIFT) by @shift blocks.
5334 * On success, 0 is returned, error otherwise.
5337 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5338 ext4_lblk_t start, ext4_lblk_t shift,
5339 enum SHIFT_DIRECTION SHIFT)
5341 struct ext4_ext_path *path;
5343 struct ext4_extent *extent;
5344 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5346 /* Let path point to the last extent */
5347 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5350 return PTR_ERR(path);
5352 depth = path->p_depth;
5353 extent = path[depth].p_ext;
5357 stop = le32_to_cpu(extent->ee_block);
5360 * For left shifts, make sure the hole on the left is big enough to
5361 * accommodate the shift. For right shifts, make sure the last extent
5362 * won't be shifted beyond EXT_MAX_BLOCKS.
5364 if (SHIFT == SHIFT_LEFT) {
5365 path = ext4_find_extent(inode, start - 1, &path,
5368 return PTR_ERR(path);
5369 depth = path->p_depth;
5370 extent = path[depth].p_ext;
5372 ex_start = le32_to_cpu(extent->ee_block);
5373 ex_end = le32_to_cpu(extent->ee_block) +
5374 ext4_ext_get_actual_len(extent);
5380 if ((start == ex_start && shift > ex_start) ||
5381 (shift > start - ex_end)) {
5386 if (shift > EXT_MAX_BLOCKS -
5387 (stop + ext4_ext_get_actual_len(extent))) {
5394 * In case of left shift, iterator points to start and it is increased
5395 * till we reach stop. In case of right shift, iterator points to stop
5396 * and it is decreased till we reach start.
5398 if (SHIFT == SHIFT_LEFT)
5404 * Its safe to start updating extents. Start and stop are unsigned, so
5405 * in case of right shift if extent with 0 block is reached, iterator
5406 * becomes NULL to indicate the end of the loop.
5408 while (iterator && start <= stop) {
5409 path = ext4_find_extent(inode, *iterator, &path,
5412 return PTR_ERR(path);
5413 depth = path->p_depth;
5414 extent = path[depth].p_ext;
5416 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5417 (unsigned long) *iterator);
5418 return -EFSCORRUPTED;
5420 if (SHIFT == SHIFT_LEFT && *iterator >
5421 le32_to_cpu(extent->ee_block)) {
5422 /* Hole, move to the next extent */
5423 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5424 path[depth].p_ext++;
5426 *iterator = ext4_ext_next_allocated_block(path);
5431 if (SHIFT == SHIFT_LEFT) {
5432 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5433 *iterator = le32_to_cpu(extent->ee_block) +
5434 ext4_ext_get_actual_len(extent);
5436 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5437 if (le32_to_cpu(extent->ee_block) > 0)
5438 *iterator = le32_to_cpu(extent->ee_block) - 1;
5440 /* Beginning is reached, end of the loop */
5442 /* Update path extent in case we need to stop */
5443 while (le32_to_cpu(extent->ee_block) < start)
5445 path[depth].p_ext = extent;
5447 ret = ext4_ext_shift_path_extents(path, shift, inode,
5453 ext4_ext_drop_refs(path);
5459 * ext4_collapse_range:
5460 * This implements the fallocate's collapse range functionality for ext4
5461 * Returns: 0 and non-zero on error.
5463 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5465 struct super_block *sb = inode->i_sb;
5466 ext4_lblk_t punch_start, punch_stop;
5468 unsigned int credits;
5469 loff_t new_size, ioffset;
5473 * We need to test this early because xfstests assumes that a
5474 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5475 * system does not support collapse range.
5477 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5480 /* Collapse range works only on fs block size aligned offsets. */
5481 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5482 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5485 if (!S_ISREG(inode->i_mode))
5488 trace_ext4_collapse_range(inode, offset, len);
5490 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5491 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5493 /* Call ext4_force_commit to flush all data in case of data=journal. */
5494 if (ext4_should_journal_data(inode)) {
5495 ret = ext4_force_commit(inode->i_sb);
5502 * There is no need to overlap collapse range with EOF, in which case
5503 * it is effectively a truncate operation
5505 if (offset + len >= i_size_read(inode)) {
5510 /* Currently just for extent based files */
5511 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5516 /* Wait for existing dio to complete */
5517 inode_dio_wait(inode);
5520 * Prevent page faults from reinstantiating pages we have released from
5523 down_write(&EXT4_I(inode)->i_mmap_sem);
5525 ret = ext4_break_layouts(inode);
5530 * Need to round down offset to be aligned with page size boundary
5531 * for page size > block size.
5533 ioffset = round_down(offset, PAGE_SIZE);
5535 * Write tail of the last page before removed range since it will get
5536 * removed from the page cache below.
5538 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5542 * Write data that will be shifted to preserve them when discarding
5543 * page cache below. We are also protected from pages becoming dirty
5546 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5550 truncate_pagecache(inode, ioffset);
5552 credits = ext4_writepage_trans_blocks(inode);
5553 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5554 if (IS_ERR(handle)) {
5555 ret = PTR_ERR(handle);
5559 down_write(&EXT4_I(inode)->i_data_sem);
5560 ext4_discard_preallocations(inode);
5562 ret = ext4_es_remove_extent(inode, punch_start,
5563 EXT_MAX_BLOCKS - punch_start);
5565 up_write(&EXT4_I(inode)->i_data_sem);
5569 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5571 up_write(&EXT4_I(inode)->i_data_sem);
5574 ext4_discard_preallocations(inode);
5576 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5577 punch_stop - punch_start, SHIFT_LEFT);
5579 up_write(&EXT4_I(inode)->i_data_sem);
5583 new_size = i_size_read(inode) - len;
5584 i_size_write(inode, new_size);
5585 EXT4_I(inode)->i_disksize = new_size;
5587 up_write(&EXT4_I(inode)->i_data_sem);
5589 ext4_handle_sync(handle);
5590 inode->i_mtime = inode->i_ctime = current_time(inode);
5591 ext4_mark_inode_dirty(handle, inode);
5592 ext4_update_inode_fsync_trans(handle, inode, 1);
5595 ext4_journal_stop(handle);
5597 up_write(&EXT4_I(inode)->i_mmap_sem);
5599 inode_unlock(inode);
5604 * ext4_insert_range:
5605 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5606 * The data blocks starting from @offset to the EOF are shifted by @len
5607 * towards right to create a hole in the @inode. Inode size is increased
5609 * Returns 0 on success, error otherwise.
5611 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5613 struct super_block *sb = inode->i_sb;
5615 struct ext4_ext_path *path;
5616 struct ext4_extent *extent;
5617 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5618 unsigned int credits, ee_len;
5619 int ret = 0, depth, split_flag = 0;
5623 * We need to test this early because xfstests assumes that an
5624 * insert range of (0, 1) will return EOPNOTSUPP if the file
5625 * system does not support insert range.
5627 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5630 /* Insert range works only on fs block size aligned offsets. */
5631 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5632 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5635 if (!S_ISREG(inode->i_mode))
5638 trace_ext4_insert_range(inode, offset, len);
5640 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5641 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5643 /* Call ext4_force_commit to flush all data in case of data=journal */
5644 if (ext4_should_journal_data(inode)) {
5645 ret = ext4_force_commit(inode->i_sb);
5651 /* Currently just for extent based files */
5652 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5657 /* Check for wrap through zero */
5658 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5663 /* Offset should be less than i_size */
5664 if (offset >= i_size_read(inode)) {
5669 /* Wait for existing dio to complete */
5670 inode_dio_wait(inode);
5673 * Prevent page faults from reinstantiating pages we have released from
5676 down_write(&EXT4_I(inode)->i_mmap_sem);
5678 ret = ext4_break_layouts(inode);
5683 * Need to round down to align start offset to page size boundary
5684 * for page size > block size.
5686 ioffset = round_down(offset, PAGE_SIZE);
5687 /* Write out all dirty pages */
5688 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5692 truncate_pagecache(inode, ioffset);
5694 credits = ext4_writepage_trans_blocks(inode);
5695 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5696 if (IS_ERR(handle)) {
5697 ret = PTR_ERR(handle);
5701 /* Expand file to avoid data loss if there is error while shifting */
5702 inode->i_size += len;
5703 EXT4_I(inode)->i_disksize += len;
5704 inode->i_mtime = inode->i_ctime = current_time(inode);
5705 ret = ext4_mark_inode_dirty(handle, inode);
5709 down_write(&EXT4_I(inode)->i_data_sem);
5710 ext4_discard_preallocations(inode);
5712 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5714 up_write(&EXT4_I(inode)->i_data_sem);
5718 depth = ext_depth(inode);
5719 extent = path[depth].p_ext;
5721 ee_start_lblk = le32_to_cpu(extent->ee_block);
5722 ee_len = ext4_ext_get_actual_len(extent);
5725 * If offset_lblk is not the starting block of extent, split
5726 * the extent @offset_lblk
5728 if ((offset_lblk > ee_start_lblk) &&
5729 (offset_lblk < (ee_start_lblk + ee_len))) {
5730 if (ext4_ext_is_unwritten(extent))
5731 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5732 EXT4_EXT_MARK_UNWRIT2;
5733 ret = ext4_split_extent_at(handle, inode, &path,
5734 offset_lblk, split_flag,
5736 EXT4_GET_BLOCKS_PRE_IO |
5737 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5740 ext4_ext_drop_refs(path);
5743 up_write(&EXT4_I(inode)->i_data_sem);
5747 ext4_ext_drop_refs(path);
5751 ret = ext4_es_remove_extent(inode, offset_lblk,
5752 EXT_MAX_BLOCKS - offset_lblk);
5754 up_write(&EXT4_I(inode)->i_data_sem);
5759 * if offset_lblk lies in a hole which is at start of file, use
5760 * ee_start_lblk to shift extents
5762 ret = ext4_ext_shift_extents(inode, handle,
5763 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5764 len_lblk, SHIFT_RIGHT);
5766 up_write(&EXT4_I(inode)->i_data_sem);
5768 ext4_handle_sync(handle);
5770 ext4_update_inode_fsync_trans(handle, inode, 1);
5773 ext4_journal_stop(handle);
5775 up_write(&EXT4_I(inode)->i_mmap_sem);
5777 inode_unlock(inode);
5782 * ext4_swap_extents() - Swap extents between two inodes
5783 * @handle: handle for this transaction
5784 * @inode1: First inode
5785 * @inode2: Second inode
5786 * @lblk1: Start block for first inode
5787 * @lblk2: Start block for second inode
5788 * @count: Number of blocks to swap
5789 * @unwritten: Mark second inode's extents as unwritten after swap
5790 * @erp: Pointer to save error value
5792 * This helper routine does exactly what is promise "swap extents". All other
5793 * stuff such as page-cache locking consistency, bh mapping consistency or
5794 * extent's data copying must be performed by caller.
5796 * i_mutex is held for both inodes
5797 * i_data_sem is locked for write for both inodes
5799 * All pages from requested range are locked for both inodes
5802 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5803 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5804 ext4_lblk_t count, int unwritten, int *erp)
5806 struct ext4_ext_path *path1 = NULL;
5807 struct ext4_ext_path *path2 = NULL;
5808 int replaced_count = 0;
5810 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5811 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5812 BUG_ON(!inode_is_locked(inode1));
5813 BUG_ON(!inode_is_locked(inode2));
5815 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5818 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5823 struct ext4_extent *ex1, *ex2, tmp_ex;
5824 ext4_lblk_t e1_blk, e2_blk;
5825 int e1_len, e2_len, len;
5828 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5829 if (IS_ERR(path1)) {
5830 *erp = PTR_ERR(path1);
5836 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5837 if (IS_ERR(path2)) {
5838 *erp = PTR_ERR(path2);
5842 ex1 = path1[path1->p_depth].p_ext;
5843 ex2 = path2[path2->p_depth].p_ext;
5844 /* Do we have somthing to swap ? */
5845 if (unlikely(!ex2 || !ex1))
5848 e1_blk = le32_to_cpu(ex1->ee_block);
5849 e2_blk = le32_to_cpu(ex2->ee_block);
5850 e1_len = ext4_ext_get_actual_len(ex1);
5851 e2_len = ext4_ext_get_actual_len(ex2);
5854 if (!in_range(lblk1, e1_blk, e1_len) ||
5855 !in_range(lblk2, e2_blk, e2_len)) {
5856 ext4_lblk_t next1, next2;
5858 /* if hole after extent, then go to next extent */
5859 next1 = ext4_ext_next_allocated_block(path1);
5860 next2 = ext4_ext_next_allocated_block(path2);
5861 /* If hole before extent, then shift to that extent */
5866 /* Do we have something to swap */
5867 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5869 /* Move to the rightest boundary */
5870 len = next1 - lblk1;
5871 if (len < next2 - lblk2)
5872 len = next2 - lblk2;
5881 /* Prepare left boundary */
5882 if (e1_blk < lblk1) {
5884 *erp = ext4_force_split_extent_at(handle, inode1,
5889 if (e2_blk < lblk2) {
5891 *erp = ext4_force_split_extent_at(handle, inode2,
5896 /* ext4_split_extent_at() may result in leaf extent split,
5897 * path must to be revalidated. */
5901 /* Prepare right boundary */
5903 if (len > e1_blk + e1_len - lblk1)
5904 len = e1_blk + e1_len - lblk1;
5905 if (len > e2_blk + e2_len - lblk2)
5906 len = e2_blk + e2_len - lblk2;
5908 if (len != e1_len) {
5910 *erp = ext4_force_split_extent_at(handle, inode1,
5911 &path1, lblk1 + len, 0);
5915 if (len != e2_len) {
5917 *erp = ext4_force_split_extent_at(handle, inode2,
5918 &path2, lblk2 + len, 0);
5922 /* ext4_split_extent_at() may result in leaf extent split,
5923 * path must to be revalidated. */
5927 BUG_ON(e2_len != e1_len);
5928 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5931 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5935 /* Both extents are fully inside boundaries. Swap it now */
5937 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5938 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5939 ex1->ee_len = cpu_to_le16(e2_len);
5940 ex2->ee_len = cpu_to_le16(e1_len);
5942 ext4_ext_mark_unwritten(ex2);
5943 if (ext4_ext_is_unwritten(&tmp_ex))
5944 ext4_ext_mark_unwritten(ex1);
5946 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5947 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5948 *erp = ext4_ext_dirty(handle, inode2, path2 +
5952 *erp = ext4_ext_dirty(handle, inode1, path1 +
5955 * Looks scarry ah..? second inode already points to new blocks,
5956 * and it was successfully dirtied. But luckily error may happen
5957 * only due to journal error, so full transaction will be
5964 replaced_count += len;
5968 ext4_ext_drop_refs(path1);
5970 ext4_ext_drop_refs(path2);
5972 path1 = path2 = NULL;
5974 return replaced_count;
5978 * ext4_clu_mapped - determine whether any block in a logical cluster has
5979 * been mapped to a physical cluster
5981 * @inode - file containing the logical cluster
5982 * @lclu - logical cluster of interest
5984 * Returns 1 if any block in the logical cluster is mapped, signifying
5985 * that a physical cluster has been allocated for it. Otherwise,
5986 * returns 0. Can also return negative error codes. Derived from
5987 * ext4_ext_map_blocks().
5989 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5991 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5992 struct ext4_ext_path *path;
5993 int depth, mapped = 0, err = 0;
5994 struct ext4_extent *extent;
5995 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5997 /* search for the extent closest to the first block in the cluster */
5998 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
6000 err = PTR_ERR(path);
6005 depth = ext_depth(inode);
6008 * A consistent leaf must not be empty. This situation is possible,
6009 * though, _during_ tree modification, and it's why an assert can't
6010 * be put in ext4_find_extent().
6012 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
6013 EXT4_ERROR_INODE(inode,
6014 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
6015 (unsigned long) EXT4_C2B(sbi, lclu),
6016 depth, path[depth].p_block);
6017 err = -EFSCORRUPTED;
6021 extent = path[depth].p_ext;
6023 /* can't be mapped if the extent tree is empty */
6027 first_lblk = le32_to_cpu(extent->ee_block);
6028 first_lclu = EXT4_B2C(sbi, first_lblk);
6031 * Three possible outcomes at this point - found extent spanning
6032 * the target cluster, to the left of the target cluster, or to the
6033 * right of the target cluster. The first two cases are handled here.
6034 * The last case indicates the target cluster is not mapped.
6036 if (lclu >= first_lclu) {
6037 last_lclu = EXT4_B2C(sbi, first_lblk +
6038 ext4_ext_get_actual_len(extent) - 1);
6039 if (lclu <= last_lclu) {
6042 first_lblk = ext4_ext_next_allocated_block(path);
6043 first_lclu = EXT4_B2C(sbi, first_lblk);
6044 if (lclu == first_lclu)
6050 ext4_ext_drop_refs(path);
6053 return err ? err : mapped;