2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
71 static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
74 struct ext4_extent_tail *et;
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
86 static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
89 struct ext4_extent_tail *et;
91 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
95 et = find_ext4_extent_tail(eh);
96 et->et_checksum = ext4_extent_block_csum(inode, eh);
99 static int ext4_split_extent(handle_t *handle,
101 struct ext4_ext_path **ppath,
102 struct ext4_map_blocks *map,
106 static int ext4_split_extent_at(handle_t *handle,
108 struct ext4_ext_path **ppath,
113 static int ext4_find_delayed_extent(struct inode *inode,
114 struct extent_status *newes);
116 static int ext4_ext_truncate_extend_restart(handle_t *handle,
122 if (!ext4_handle_valid(handle))
124 if (handle->h_buffer_credits > needed)
126 err = ext4_journal_extend(handle, needed);
129 err = ext4_truncate_restart_trans(handle, inode, needed);
141 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
142 struct ext4_ext_path *path)
145 /* path points to block */
146 BUFFER_TRACE(path->p_bh, "get_write_access");
147 return ext4_journal_get_write_access(handle, path->p_bh);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
165 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
167 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
168 /* path points to block */
169 err = __ext4_handle_dirty_metadata(where, line, handle,
172 /* path points to leaf/index in inode body */
173 err = ext4_mark_inode_dirty(handle, inode);
178 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
179 struct ext4_ext_path *path,
183 int depth = path->p_depth;
184 struct ext4_extent *ex;
187 * Try to predict block placement assuming that we are
188 * filling in a file which will eventually be
189 * non-sparse --- i.e., in the case of libbfd writing
190 * an ELF object sections out-of-order but in a way
191 * the eventually results in a contiguous object or
192 * executable file, or some database extending a table
193 * space file. However, this is actually somewhat
194 * non-ideal if we are writing a sparse file such as
195 * qemu or KVM writing a raw image file that is going
196 * to stay fairly sparse, since it will end up
197 * fragmenting the file system's free space. Maybe we
198 * should have some hueristics or some way to allow
199 * userspace to pass a hint to file system,
200 * especially if the latter case turns out to be
203 ex = path[depth].p_ext;
205 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
206 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
208 if (block > ext_block)
209 return ext_pblk + (block - ext_block);
211 return ext_pblk - (ext_block - block);
214 /* it looks like index is empty;
215 * try to find starting block from index itself */
216 if (path[depth].p_bh)
217 return path[depth].p_bh->b_blocknr;
220 /* OK. use inode's group */
221 return ext4_inode_to_goal_block(inode);
225 * Allocation for a meta data block
228 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
229 struct ext4_ext_path *path,
230 struct ext4_extent *ex, int *err, unsigned int flags)
232 ext4_fsblk_t goal, newblock;
234 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
235 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
240 static inline int ext4_ext_space_block(struct inode *inode, int check)
244 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
245 / sizeof(struct ext4_extent);
246 #ifdef AGGRESSIVE_TEST
247 if (!check && size > 6)
253 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
257 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
258 / sizeof(struct ext4_extent_idx);
259 #ifdef AGGRESSIVE_TEST
260 if (!check && size > 5)
266 static inline int ext4_ext_space_root(struct inode *inode, int check)
270 size = sizeof(EXT4_I(inode)->i_data);
271 size -= sizeof(struct ext4_extent_header);
272 size /= sizeof(struct ext4_extent);
273 #ifdef AGGRESSIVE_TEST
274 if (!check && size > 3)
280 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
284 size = sizeof(EXT4_I(inode)->i_data);
285 size -= sizeof(struct ext4_extent_header);
286 size /= sizeof(struct ext4_extent_idx);
287 #ifdef AGGRESSIVE_TEST
288 if (!check && size > 4)
295 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
296 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
299 struct ext4_ext_path *path = *ppath;
300 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
302 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
303 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
304 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
305 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
309 * Calculate the number of metadata blocks needed
310 * to allocate @blocks
311 * Worse case is one block per extent
313 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
315 struct ext4_inode_info *ei = EXT4_I(inode);
318 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
319 / sizeof(struct ext4_extent_idx));
322 * If the new delayed allocation block is contiguous with the
323 * previous da block, it can share index blocks with the
324 * previous block, so we only need to allocate a new index
325 * block every idxs leaf blocks. At ldxs**2 blocks, we need
326 * an additional index block, and at ldxs**3 blocks, yet
327 * another index blocks.
329 if (ei->i_da_metadata_calc_len &&
330 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
333 if ((ei->i_da_metadata_calc_len % idxs) == 0)
335 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
337 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
339 ei->i_da_metadata_calc_len = 0;
341 ei->i_da_metadata_calc_len++;
342 ei->i_da_metadata_calc_last_lblock++;
347 * In the worst case we need a new set of index blocks at
348 * every level of the inode's extent tree.
350 ei->i_da_metadata_calc_len = 1;
351 ei->i_da_metadata_calc_last_lblock = lblock;
352 return ext_depth(inode) + 1;
356 ext4_ext_max_entries(struct inode *inode, int depth)
360 if (depth == ext_depth(inode)) {
362 max = ext4_ext_space_root(inode, 1);
364 max = ext4_ext_space_root_idx(inode, 1);
367 max = ext4_ext_space_block(inode, 1);
369 max = ext4_ext_space_block_idx(inode, 1);
375 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
377 ext4_fsblk_t block = ext4_ext_pblock(ext);
378 int len = ext4_ext_get_actual_len(ext);
379 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
380 ext4_lblk_t last = lblock + len - 1;
384 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
387 static int ext4_valid_extent_idx(struct inode *inode,
388 struct ext4_extent_idx *ext_idx)
390 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
392 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
395 static int ext4_valid_extent_entries(struct inode *inode,
396 struct ext4_extent_header *eh,
399 unsigned short entries;
400 if (eh->eh_entries == 0)
403 entries = le16_to_cpu(eh->eh_entries);
407 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
408 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
409 ext4_fsblk_t pblock = 0;
410 ext4_lblk_t lblock = 0;
411 ext4_lblk_t prev = 0;
414 if (!ext4_valid_extent(inode, ext))
417 /* Check for overlapping extents */
418 lblock = le32_to_cpu(ext->ee_block);
419 len = ext4_ext_get_actual_len(ext);
420 if ((lblock <= prev) && prev) {
421 pblock = ext4_ext_pblock(ext);
422 es->s_last_error_block = cpu_to_le64(pblock);
427 prev = lblock + len - 1;
430 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
432 if (!ext4_valid_extent_idx(inode, ext_idx))
441 static int __ext4_ext_check(const char *function, unsigned int line,
442 struct inode *inode, struct ext4_extent_header *eh,
443 int depth, ext4_fsblk_t pblk)
445 const char *error_msg;
448 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
449 error_msg = "invalid magic";
452 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
453 error_msg = "unexpected eh_depth";
456 if (unlikely(eh->eh_max == 0)) {
457 error_msg = "invalid eh_max";
460 max = ext4_ext_max_entries(inode, depth);
461 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
462 error_msg = "too large eh_max";
465 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
466 error_msg = "invalid eh_entries";
469 if (!ext4_valid_extent_entries(inode, eh, depth)) {
470 error_msg = "invalid extent entries";
473 /* Verify checksum on non-root extent tree nodes */
474 if (ext_depth(inode) != depth &&
475 !ext4_extent_block_csum_verify(inode, eh)) {
476 error_msg = "extent tree corrupted";
482 ext4_error_inode(inode, function, line, 0,
483 "pblk %llu bad header/extent: %s - magic %x, "
484 "entries %u, max %u(%u), depth %u(%u)",
485 (unsigned long long) pblk, error_msg,
486 le16_to_cpu(eh->eh_magic),
487 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
488 max, le16_to_cpu(eh->eh_depth), depth);
492 #define ext4_ext_check(inode, eh, depth, pblk) \
493 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
495 int ext4_ext_check_inode(struct inode *inode)
497 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
500 static struct buffer_head *
501 __read_extent_tree_block(const char *function, unsigned int line,
502 struct inode *inode, ext4_fsblk_t pblk, int depth,
505 struct buffer_head *bh;
508 bh = sb_getblk(inode->i_sb, pblk);
510 return ERR_PTR(-ENOMEM);
512 if (!bh_uptodate_or_lock(bh)) {
513 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
514 err = bh_submit_read(bh);
518 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
520 err = __ext4_ext_check(function, line, inode,
521 ext_block_hdr(bh), depth, pblk);
524 set_buffer_verified(bh);
526 * If this is a leaf block, cache all of its entries
528 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
529 struct ext4_extent_header *eh = ext_block_hdr(bh);
530 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
531 ext4_lblk_t prev = 0;
534 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
535 unsigned int status = EXTENT_STATUS_WRITTEN;
536 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
537 int len = ext4_ext_get_actual_len(ex);
539 if (prev && (prev != lblk))
540 ext4_es_cache_extent(inode, prev,
544 if (ext4_ext_is_unwritten(ex))
545 status = EXTENT_STATUS_UNWRITTEN;
546 ext4_es_cache_extent(inode, lblk, len,
547 ext4_ext_pblock(ex), status);
558 #define read_extent_tree_block(inode, pblk, depth, flags) \
559 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
563 * This function is called to cache a file's extent information in the
566 int ext4_ext_precache(struct inode *inode)
568 struct ext4_inode_info *ei = EXT4_I(inode);
569 struct ext4_ext_path *path = NULL;
570 struct buffer_head *bh;
571 int i = 0, depth, ret = 0;
573 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
574 return 0; /* not an extent-mapped inode */
576 down_read(&ei->i_data_sem);
577 depth = ext_depth(inode);
579 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
582 up_read(&ei->i_data_sem);
586 /* Don't cache anything if there are no external extent blocks */
589 path[0].p_hdr = ext_inode_hdr(inode);
590 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
593 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
596 * If this is a leaf block or we've reached the end of
597 * the index block, go up
600 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
601 brelse(path[i].p_bh);
606 bh = read_extent_tree_block(inode,
607 ext4_idx_pblock(path[i].p_idx++),
609 EXT4_EX_FORCE_CACHE);
616 path[i].p_hdr = ext_block_hdr(bh);
617 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
619 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
621 up_read(&ei->i_data_sem);
622 ext4_ext_drop_refs(path);
628 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
630 int k, l = path->p_depth;
633 for (k = 0; k <= l; k++, path++) {
635 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
636 ext4_idx_pblock(path->p_idx));
637 } else if (path->p_ext) {
638 ext_debug(" %d:[%d]%d:%llu ",
639 le32_to_cpu(path->p_ext->ee_block),
640 ext4_ext_is_unwritten(path->p_ext),
641 ext4_ext_get_actual_len(path->p_ext),
642 ext4_ext_pblock(path->p_ext));
649 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
651 int depth = ext_depth(inode);
652 struct ext4_extent_header *eh;
653 struct ext4_extent *ex;
659 eh = path[depth].p_hdr;
660 ex = EXT_FIRST_EXTENT(eh);
662 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
664 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
665 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
666 ext4_ext_is_unwritten(ex),
667 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
672 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
673 ext4_fsblk_t newblock, int level)
675 int depth = ext_depth(inode);
676 struct ext4_extent *ex;
678 if (depth != level) {
679 struct ext4_extent_idx *idx;
680 idx = path[level].p_idx;
681 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
682 ext_debug("%d: move %d:%llu in new index %llu\n", level,
683 le32_to_cpu(idx->ei_block),
684 ext4_idx_pblock(idx),
692 ex = path[depth].p_ext;
693 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
694 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
695 le32_to_cpu(ex->ee_block),
697 ext4_ext_is_unwritten(ex),
698 ext4_ext_get_actual_len(ex),
705 #define ext4_ext_show_path(inode, path)
706 #define ext4_ext_show_leaf(inode, path)
707 #define ext4_ext_show_move(inode, path, newblock, level)
710 void ext4_ext_drop_refs(struct ext4_ext_path *path)
712 int depth = path->p_depth;
715 for (i = 0; i <= depth; i++, path++)
723 * ext4_ext_binsearch_idx:
724 * binary search for the closest index of the given block
725 * the header must be checked before calling this
728 ext4_ext_binsearch_idx(struct inode *inode,
729 struct ext4_ext_path *path, ext4_lblk_t block)
731 struct ext4_extent_header *eh = path->p_hdr;
732 struct ext4_extent_idx *r, *l, *m;
735 ext_debug("binsearch for %u(idx): ", block);
737 l = EXT_FIRST_INDEX(eh) + 1;
738 r = EXT_LAST_INDEX(eh);
741 if (block < le32_to_cpu(m->ei_block))
745 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
746 m, le32_to_cpu(m->ei_block),
747 r, le32_to_cpu(r->ei_block));
751 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
752 ext4_idx_pblock(path->p_idx));
754 #ifdef CHECK_BINSEARCH
756 struct ext4_extent_idx *chix, *ix;
759 chix = ix = EXT_FIRST_INDEX(eh);
760 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
762 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
763 printk(KERN_DEBUG "k=%d, ix=0x%p, "
765 ix, EXT_FIRST_INDEX(eh));
766 printk(KERN_DEBUG "%u <= %u\n",
767 le32_to_cpu(ix->ei_block),
768 le32_to_cpu(ix[-1].ei_block));
770 BUG_ON(k && le32_to_cpu(ix->ei_block)
771 <= le32_to_cpu(ix[-1].ei_block));
772 if (block < le32_to_cpu(ix->ei_block))
776 BUG_ON(chix != path->p_idx);
783 * ext4_ext_binsearch:
784 * binary search for closest extent of the given block
785 * the header must be checked before calling this
788 ext4_ext_binsearch(struct inode *inode,
789 struct ext4_ext_path *path, ext4_lblk_t block)
791 struct ext4_extent_header *eh = path->p_hdr;
792 struct ext4_extent *r, *l, *m;
794 if (eh->eh_entries == 0) {
796 * this leaf is empty:
797 * we get such a leaf in split/add case
802 ext_debug("binsearch for %u: ", block);
804 l = EXT_FIRST_EXTENT(eh) + 1;
805 r = EXT_LAST_EXTENT(eh);
809 if (block < le32_to_cpu(m->ee_block))
813 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
814 m, le32_to_cpu(m->ee_block),
815 r, le32_to_cpu(r->ee_block));
819 ext_debug(" -> %d:%llu:[%d]%d ",
820 le32_to_cpu(path->p_ext->ee_block),
821 ext4_ext_pblock(path->p_ext),
822 ext4_ext_is_unwritten(path->p_ext),
823 ext4_ext_get_actual_len(path->p_ext));
825 #ifdef CHECK_BINSEARCH
827 struct ext4_extent *chex, *ex;
830 chex = ex = EXT_FIRST_EXTENT(eh);
831 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
832 BUG_ON(k && le32_to_cpu(ex->ee_block)
833 <= le32_to_cpu(ex[-1].ee_block));
834 if (block < le32_to_cpu(ex->ee_block))
838 BUG_ON(chex != path->p_ext);
844 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
846 struct ext4_extent_header *eh;
848 eh = ext_inode_hdr(inode);
851 eh->eh_magic = EXT4_EXT_MAGIC;
852 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
853 ext4_mark_inode_dirty(handle, inode);
857 struct ext4_ext_path *
858 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
859 struct ext4_ext_path **orig_path, int flags)
861 struct ext4_extent_header *eh;
862 struct buffer_head *bh;
863 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
864 short int depth, i, ppos = 0;
867 eh = ext_inode_hdr(inode);
868 depth = ext_depth(inode);
870 /* account possible depth increase */
872 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
875 return ERR_PTR(-ENOMEM);
881 /* walk through the tree */
883 ext_debug("depth %d: num %d, max %d\n",
884 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
886 ext4_ext_binsearch_idx(inode, path + ppos, block);
887 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
888 path[ppos].p_depth = i;
889 path[ppos].p_ext = NULL;
891 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
893 if (unlikely(IS_ERR(bh))) {
898 eh = ext_block_hdr(bh);
900 if (unlikely(ppos > depth)) {
902 EXT4_ERROR_INODE(inode,
903 "ppos %d > depth %d", ppos, depth);
907 path[ppos].p_bh = bh;
908 path[ppos].p_hdr = eh;
911 path[ppos].p_depth = i;
912 path[ppos].p_ext = NULL;
913 path[ppos].p_idx = NULL;
916 ext4_ext_binsearch(inode, path + ppos, block);
917 /* if not an empty leaf */
918 if (path[ppos].p_ext)
919 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
921 ext4_ext_show_path(inode, path);
926 ext4_ext_drop_refs(path);
934 * ext4_ext_insert_index:
935 * insert new index [@logical;@ptr] into the block at @curp;
936 * check where to insert: before @curp or after @curp
938 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
939 struct ext4_ext_path *curp,
940 int logical, ext4_fsblk_t ptr)
942 struct ext4_extent_idx *ix;
945 err = ext4_ext_get_access(handle, inode, curp);
949 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
950 EXT4_ERROR_INODE(inode,
951 "logical %d == ei_block %d!",
952 logical, le32_to_cpu(curp->p_idx->ei_block));
956 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
957 >= le16_to_cpu(curp->p_hdr->eh_max))) {
958 EXT4_ERROR_INODE(inode,
959 "eh_entries %d >= eh_max %d!",
960 le16_to_cpu(curp->p_hdr->eh_entries),
961 le16_to_cpu(curp->p_hdr->eh_max));
965 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
967 ext_debug("insert new index %d after: %llu\n", logical, ptr);
968 ix = curp->p_idx + 1;
971 ext_debug("insert new index %d before: %llu\n", logical, ptr);
975 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
978 ext_debug("insert new index %d: "
979 "move %d indices from 0x%p to 0x%p\n",
980 logical, len, ix, ix + 1);
981 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
984 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
985 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
989 ix->ei_block = cpu_to_le32(logical);
990 ext4_idx_store_pblock(ix, ptr);
991 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
993 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
994 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
998 err = ext4_ext_dirty(handle, inode, curp);
999 ext4_std_error(inode->i_sb, err);
1006 * inserts new subtree into the path, using free index entry
1008 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1009 * - makes decision where to split
1010 * - moves remaining extents and index entries (right to the split point)
1011 * into the newly allocated blocks
1012 * - initializes subtree
1014 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1016 struct ext4_ext_path *path,
1017 struct ext4_extent *newext, int at)
1019 struct buffer_head *bh = NULL;
1020 int depth = ext_depth(inode);
1021 struct ext4_extent_header *neh;
1022 struct ext4_extent_idx *fidx;
1023 int i = at, k, m, a;
1024 ext4_fsblk_t newblock, oldblock;
1026 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1029 /* make decision: where to split? */
1030 /* FIXME: now decision is simplest: at current extent */
1032 /* if current leaf will be split, then we should use
1033 * border from split point */
1034 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1035 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1038 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1039 border = path[depth].p_ext[1].ee_block;
1040 ext_debug("leaf will be split."
1041 " next leaf starts at %d\n",
1042 le32_to_cpu(border));
1044 border = newext->ee_block;
1045 ext_debug("leaf will be added."
1046 " next leaf starts at %d\n",
1047 le32_to_cpu(border));
1051 * If error occurs, then we break processing
1052 * and mark filesystem read-only. index won't
1053 * be inserted and tree will be in consistent
1054 * state. Next mount will repair buffers too.
1058 * Get array to track all allocated blocks.
1059 * We need this to handle errors and free blocks
1062 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1066 /* allocate all needed blocks */
1067 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1068 for (a = 0; a < depth - at; a++) {
1069 newblock = ext4_ext_new_meta_block(handle, inode, path,
1070 newext, &err, flags);
1073 ablocks[a] = newblock;
1076 /* initialize new leaf */
1077 newblock = ablocks[--a];
1078 if (unlikely(newblock == 0)) {
1079 EXT4_ERROR_INODE(inode, "newblock == 0!");
1083 bh = sb_getblk(inode->i_sb, newblock);
1084 if (unlikely(!bh)) {
1090 err = ext4_journal_get_create_access(handle, bh);
1094 neh = ext_block_hdr(bh);
1095 neh->eh_entries = 0;
1096 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1097 neh->eh_magic = EXT4_EXT_MAGIC;
1100 /* move remainder of path[depth] to the new leaf */
1101 if (unlikely(path[depth].p_hdr->eh_entries !=
1102 path[depth].p_hdr->eh_max)) {
1103 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1104 path[depth].p_hdr->eh_entries,
1105 path[depth].p_hdr->eh_max);
1109 /* start copy from next extent */
1110 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1111 ext4_ext_show_move(inode, path, newblock, depth);
1113 struct ext4_extent *ex;
1114 ex = EXT_FIRST_EXTENT(neh);
1115 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1116 le16_add_cpu(&neh->eh_entries, m);
1119 ext4_extent_block_csum_set(inode, neh);
1120 set_buffer_uptodate(bh);
1123 err = ext4_handle_dirty_metadata(handle, inode, bh);
1129 /* correct old leaf */
1131 err = ext4_ext_get_access(handle, inode, path + depth);
1134 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1135 err = ext4_ext_dirty(handle, inode, path + depth);
1141 /* create intermediate indexes */
1143 if (unlikely(k < 0)) {
1144 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1149 ext_debug("create %d intermediate indices\n", k);
1150 /* insert new index into current index block */
1151 /* current depth stored in i var */
1154 oldblock = newblock;
1155 newblock = ablocks[--a];
1156 bh = sb_getblk(inode->i_sb, newblock);
1157 if (unlikely(!bh)) {
1163 err = ext4_journal_get_create_access(handle, bh);
1167 neh = ext_block_hdr(bh);
1168 neh->eh_entries = cpu_to_le16(1);
1169 neh->eh_magic = EXT4_EXT_MAGIC;
1170 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1171 neh->eh_depth = cpu_to_le16(depth - i);
1172 fidx = EXT_FIRST_INDEX(neh);
1173 fidx->ei_block = border;
1174 ext4_idx_store_pblock(fidx, oldblock);
1176 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1177 i, newblock, le32_to_cpu(border), oldblock);
1179 /* move remainder of path[i] to the new index block */
1180 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1181 EXT_LAST_INDEX(path[i].p_hdr))) {
1182 EXT4_ERROR_INODE(inode,
1183 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1184 le32_to_cpu(path[i].p_ext->ee_block));
1188 /* start copy indexes */
1189 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1190 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1191 EXT_MAX_INDEX(path[i].p_hdr));
1192 ext4_ext_show_move(inode, path, newblock, i);
1194 memmove(++fidx, path[i].p_idx,
1195 sizeof(struct ext4_extent_idx) * m);
1196 le16_add_cpu(&neh->eh_entries, m);
1198 ext4_extent_block_csum_set(inode, neh);
1199 set_buffer_uptodate(bh);
1202 err = ext4_handle_dirty_metadata(handle, inode, bh);
1208 /* correct old index */
1210 err = ext4_ext_get_access(handle, inode, path + i);
1213 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1214 err = ext4_ext_dirty(handle, inode, path + i);
1222 /* insert new index */
1223 err = ext4_ext_insert_index(handle, inode, path + at,
1224 le32_to_cpu(border), newblock);
1228 if (buffer_locked(bh))
1234 /* free all allocated blocks in error case */
1235 for (i = 0; i < depth; i++) {
1238 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1239 EXT4_FREE_BLOCKS_METADATA);
1248 * ext4_ext_grow_indepth:
1249 * implements tree growing procedure:
1250 * - allocates new block
1251 * - moves top-level data (index block or leaf) into the new block
1252 * - initializes new top-level, creating index that points to the
1253 * just created block
1255 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1257 struct ext4_extent *newext)
1259 struct ext4_extent_header *neh;
1260 struct buffer_head *bh;
1261 ext4_fsblk_t newblock;
1264 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1265 newext, &err, flags);
1269 bh = sb_getblk(inode->i_sb, newblock);
1274 err = ext4_journal_get_create_access(handle, bh);
1280 /* move top-level index/leaf into new block */
1281 memmove(bh->b_data, EXT4_I(inode)->i_data,
1282 sizeof(EXT4_I(inode)->i_data));
1284 /* set size of new block */
1285 neh = ext_block_hdr(bh);
1286 /* old root could have indexes or leaves
1287 * so calculate e_max right way */
1288 if (ext_depth(inode))
1289 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1291 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1292 neh->eh_magic = EXT4_EXT_MAGIC;
1293 ext4_extent_block_csum_set(inode, neh);
1294 set_buffer_uptodate(bh);
1297 err = ext4_handle_dirty_metadata(handle, inode, bh);
1301 /* Update top-level index: num,max,pointer */
1302 neh = ext_inode_hdr(inode);
1303 neh->eh_entries = cpu_to_le16(1);
1304 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1305 if (neh->eh_depth == 0) {
1306 /* Root extent block becomes index block */
1307 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1308 EXT_FIRST_INDEX(neh)->ei_block =
1309 EXT_FIRST_EXTENT(neh)->ee_block;
1311 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1312 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1313 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1314 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1316 le16_add_cpu(&neh->eh_depth, 1);
1317 ext4_mark_inode_dirty(handle, inode);
1325 * ext4_ext_create_new_leaf:
1326 * finds empty index and adds new leaf.
1327 * if no free index is found, then it requests in-depth growing.
1329 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1330 unsigned int mb_flags,
1331 unsigned int gb_flags,
1332 struct ext4_ext_path **ppath,
1333 struct ext4_extent *newext)
1335 struct ext4_ext_path *path = *ppath;
1336 struct ext4_ext_path *curp;
1337 int depth, i, err = 0;
1340 i = depth = ext_depth(inode);
1342 /* walk up to the tree and look for free index entry */
1343 curp = path + depth;
1344 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1349 /* we use already allocated block for index block,
1350 * so subsequent data blocks should be contiguous */
1351 if (EXT_HAS_FREE_INDEX(curp)) {
1352 /* if we found index with free entry, then use that
1353 * entry: create all needed subtree and add new leaf */
1354 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1359 ext4_ext_drop_refs(path);
1360 path = ext4_ext_find_extent(inode,
1361 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1364 err = PTR_ERR(path);
1366 /* tree is full, time to grow in depth */
1367 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1372 ext4_ext_drop_refs(path);
1373 path = ext4_ext_find_extent(inode,
1374 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1377 err = PTR_ERR(path);
1382 * only first (depth 0 -> 1) produces free space;
1383 * in all other cases we have to split the grown tree
1385 depth = ext_depth(inode);
1386 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1387 /* now we need to split */
1397 * search the closest allocated block to the left for *logical
1398 * and returns it at @logical + it's physical address at @phys
1399 * if *logical is the smallest allocated block, the function
1400 * returns 0 at @phys
1401 * return value contains 0 (success) or error code
1403 static int ext4_ext_search_left(struct inode *inode,
1404 struct ext4_ext_path *path,
1405 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1407 struct ext4_extent_idx *ix;
1408 struct ext4_extent *ex;
1411 if (unlikely(path == NULL)) {
1412 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1415 depth = path->p_depth;
1418 if (depth == 0 && path->p_ext == NULL)
1421 /* usually extent in the path covers blocks smaller
1422 * then *logical, but it can be that extent is the
1423 * first one in the file */
1425 ex = path[depth].p_ext;
1426 ee_len = ext4_ext_get_actual_len(ex);
1427 if (*logical < le32_to_cpu(ex->ee_block)) {
1428 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1429 EXT4_ERROR_INODE(inode,
1430 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1431 *logical, le32_to_cpu(ex->ee_block));
1434 while (--depth >= 0) {
1435 ix = path[depth].p_idx;
1436 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1437 EXT4_ERROR_INODE(inode,
1438 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1439 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1440 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1441 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1449 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1450 EXT4_ERROR_INODE(inode,
1451 "logical %d < ee_block %d + ee_len %d!",
1452 *logical, le32_to_cpu(ex->ee_block), ee_len);
1456 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1457 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1462 * search the closest allocated block to the right for *logical
1463 * and returns it at @logical + it's physical address at @phys
1464 * if *logical is the largest allocated block, the function
1465 * returns 0 at @phys
1466 * return value contains 0 (success) or error code
1468 static int ext4_ext_search_right(struct inode *inode,
1469 struct ext4_ext_path *path,
1470 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1471 struct ext4_extent **ret_ex)
1473 struct buffer_head *bh = NULL;
1474 struct ext4_extent_header *eh;
1475 struct ext4_extent_idx *ix;
1476 struct ext4_extent *ex;
1478 int depth; /* Note, NOT eh_depth; depth from top of tree */
1481 if (unlikely(path == NULL)) {
1482 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1485 depth = path->p_depth;
1488 if (depth == 0 && path->p_ext == NULL)
1491 /* usually extent in the path covers blocks smaller
1492 * then *logical, but it can be that extent is the
1493 * first one in the file */
1495 ex = path[depth].p_ext;
1496 ee_len = ext4_ext_get_actual_len(ex);
1497 if (*logical < le32_to_cpu(ex->ee_block)) {
1498 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1499 EXT4_ERROR_INODE(inode,
1500 "first_extent(path[%d].p_hdr) != ex",
1504 while (--depth >= 0) {
1505 ix = path[depth].p_idx;
1506 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1507 EXT4_ERROR_INODE(inode,
1508 "ix != EXT_FIRST_INDEX *logical %d!",
1516 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1517 EXT4_ERROR_INODE(inode,
1518 "logical %d < ee_block %d + ee_len %d!",
1519 *logical, le32_to_cpu(ex->ee_block), ee_len);
1523 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1524 /* next allocated block in this leaf */
1529 /* go up and search for index to the right */
1530 while (--depth >= 0) {
1531 ix = path[depth].p_idx;
1532 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1536 /* we've gone up to the root and found no index to the right */
1540 /* we've found index to the right, let's
1541 * follow it and find the closest allocated
1542 * block to the right */
1544 block = ext4_idx_pblock(ix);
1545 while (++depth < path->p_depth) {
1546 /* subtract from p_depth to get proper eh_depth */
1547 bh = read_extent_tree_block(inode, block,
1548 path->p_depth - depth, 0);
1551 eh = ext_block_hdr(bh);
1552 ix = EXT_FIRST_INDEX(eh);
1553 block = ext4_idx_pblock(ix);
1557 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1560 eh = ext_block_hdr(bh);
1561 ex = EXT_FIRST_EXTENT(eh);
1563 *logical = le32_to_cpu(ex->ee_block);
1564 *phys = ext4_ext_pblock(ex);
1572 * ext4_ext_next_allocated_block:
1573 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1574 * NOTE: it considers block number from index entry as
1575 * allocated block. Thus, index entries have to be consistent
1579 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1583 BUG_ON(path == NULL);
1584 depth = path->p_depth;
1586 if (depth == 0 && path->p_ext == NULL)
1587 return EXT_MAX_BLOCKS;
1589 while (depth >= 0) {
1590 if (depth == path->p_depth) {
1592 if (path[depth].p_ext &&
1593 path[depth].p_ext !=
1594 EXT_LAST_EXTENT(path[depth].p_hdr))
1595 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1598 if (path[depth].p_idx !=
1599 EXT_LAST_INDEX(path[depth].p_hdr))
1600 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1605 return EXT_MAX_BLOCKS;
1609 * ext4_ext_next_leaf_block:
1610 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1612 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1616 BUG_ON(path == NULL);
1617 depth = path->p_depth;
1619 /* zero-tree has no leaf blocks at all */
1621 return EXT_MAX_BLOCKS;
1623 /* go to index block */
1626 while (depth >= 0) {
1627 if (path[depth].p_idx !=
1628 EXT_LAST_INDEX(path[depth].p_hdr))
1629 return (ext4_lblk_t)
1630 le32_to_cpu(path[depth].p_idx[1].ei_block);
1634 return EXT_MAX_BLOCKS;
1638 * ext4_ext_correct_indexes:
1639 * if leaf gets modified and modified extent is first in the leaf,
1640 * then we have to correct all indexes above.
1641 * TODO: do we need to correct tree in all cases?
1643 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1644 struct ext4_ext_path *path)
1646 struct ext4_extent_header *eh;
1647 int depth = ext_depth(inode);
1648 struct ext4_extent *ex;
1652 eh = path[depth].p_hdr;
1653 ex = path[depth].p_ext;
1655 if (unlikely(ex == NULL || eh == NULL)) {
1656 EXT4_ERROR_INODE(inode,
1657 "ex %p == NULL or eh %p == NULL", ex, eh);
1662 /* there is no tree at all */
1666 if (ex != EXT_FIRST_EXTENT(eh)) {
1667 /* we correct tree if first leaf got modified only */
1672 * TODO: we need correction if border is smaller than current one
1675 border = path[depth].p_ext->ee_block;
1676 err = ext4_ext_get_access(handle, inode, path + k);
1679 path[k].p_idx->ei_block = border;
1680 err = ext4_ext_dirty(handle, inode, path + k);
1685 /* change all left-side indexes */
1686 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1688 err = ext4_ext_get_access(handle, inode, path + k);
1691 path[k].p_idx->ei_block = border;
1692 err = ext4_ext_dirty(handle, inode, path + k);
1701 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1702 struct ext4_extent *ex2)
1704 unsigned short ext1_ee_len, ext2_ee_len;
1707 * Make sure that both extents are initialized. We don't merge
1708 * unwritten extents so that we can be sure that end_io code has
1709 * the extent that was written properly split out and conversion to
1710 * initialized is trivial.
1712 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1715 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1716 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1718 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1719 le32_to_cpu(ex2->ee_block))
1723 * To allow future support for preallocated extents to be added
1724 * as an RO_COMPAT feature, refuse to merge to extents if
1725 * this can result in the top bit of ee_len being set.
1727 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1729 if (ext4_ext_is_unwritten(ex1) &&
1730 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1731 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1732 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1734 #ifdef AGGRESSIVE_TEST
1735 if (ext1_ee_len >= 4)
1739 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1745 * This function tries to merge the "ex" extent to the next extent in the tree.
1746 * It always tries to merge towards right. If you want to merge towards
1747 * left, pass "ex - 1" as argument instead of "ex".
1748 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1749 * 1 if they got merged.
1751 static int ext4_ext_try_to_merge_right(struct inode *inode,
1752 struct ext4_ext_path *path,
1753 struct ext4_extent *ex)
1755 struct ext4_extent_header *eh;
1756 unsigned int depth, len;
1757 int merge_done = 0, unwritten;
1759 depth = ext_depth(inode);
1760 BUG_ON(path[depth].p_hdr == NULL);
1761 eh = path[depth].p_hdr;
1763 while (ex < EXT_LAST_EXTENT(eh)) {
1764 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1766 /* merge with next extent! */
1767 unwritten = ext4_ext_is_unwritten(ex);
1768 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1769 + ext4_ext_get_actual_len(ex + 1));
1771 ext4_ext_mark_unwritten(ex);
1773 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1774 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1775 * sizeof(struct ext4_extent);
1776 memmove(ex + 1, ex + 2, len);
1778 le16_add_cpu(&eh->eh_entries, -1);
1780 WARN_ON(eh->eh_entries == 0);
1781 if (!eh->eh_entries)
1782 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1789 * This function does a very simple check to see if we can collapse
1790 * an extent tree with a single extent tree leaf block into the inode.
1792 static void ext4_ext_try_to_merge_up(handle_t *handle,
1793 struct inode *inode,
1794 struct ext4_ext_path *path)
1797 unsigned max_root = ext4_ext_space_root(inode, 0);
1800 if ((path[0].p_depth != 1) ||
1801 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1802 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1806 * We need to modify the block allocation bitmap and the block
1807 * group descriptor to release the extent tree block. If we
1808 * can't get the journal credits, give up.
1810 if (ext4_journal_extend(handle, 2))
1814 * Copy the extent data up to the inode
1816 blk = ext4_idx_pblock(path[0].p_idx);
1817 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1818 sizeof(struct ext4_extent_idx);
1819 s += sizeof(struct ext4_extent_header);
1821 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1822 path[0].p_depth = 0;
1823 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1824 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1825 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1827 brelse(path[1].p_bh);
1828 ext4_free_blocks(handle, inode, NULL, blk, 1,
1829 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1833 * This function tries to merge the @ex extent to neighbours in the tree.
1834 * return 1 if merge left else 0.
1836 static void ext4_ext_try_to_merge(handle_t *handle,
1837 struct inode *inode,
1838 struct ext4_ext_path *path,
1839 struct ext4_extent *ex) {
1840 struct ext4_extent_header *eh;
1844 depth = ext_depth(inode);
1845 BUG_ON(path[depth].p_hdr == NULL);
1846 eh = path[depth].p_hdr;
1848 if (ex > EXT_FIRST_EXTENT(eh))
1849 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1852 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1854 ext4_ext_try_to_merge_up(handle, inode, path);
1858 * check if a portion of the "newext" extent overlaps with an
1861 * If there is an overlap discovered, it updates the length of the newext
1862 * such that there will be no overlap, and then returns 1.
1863 * If there is no overlap found, it returns 0.
1865 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1866 struct inode *inode,
1867 struct ext4_extent *newext,
1868 struct ext4_ext_path *path)
1871 unsigned int depth, len1;
1872 unsigned int ret = 0;
1874 b1 = le32_to_cpu(newext->ee_block);
1875 len1 = ext4_ext_get_actual_len(newext);
1876 depth = ext_depth(inode);
1877 if (!path[depth].p_ext)
1879 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1882 * get the next allocated block if the extent in the path
1883 * is before the requested block(s)
1886 b2 = ext4_ext_next_allocated_block(path);
1887 if (b2 == EXT_MAX_BLOCKS)
1889 b2 = EXT4_LBLK_CMASK(sbi, b2);
1892 /* check for wrap through zero on extent logical start block*/
1893 if (b1 + len1 < b1) {
1894 len1 = EXT_MAX_BLOCKS - b1;
1895 newext->ee_len = cpu_to_le16(len1);
1899 /* check for overlap */
1900 if (b1 + len1 > b2) {
1901 newext->ee_len = cpu_to_le16(b2 - b1);
1909 * ext4_ext_insert_extent:
1910 * tries to merge requsted extent into the existing extent or
1911 * inserts requested extent as new one into the tree,
1912 * creating new leaf in the no-space case.
1914 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1915 struct ext4_ext_path **ppath,
1916 struct ext4_extent *newext, int gb_flags)
1918 struct ext4_ext_path *path = *ppath;
1919 struct ext4_extent_header *eh;
1920 struct ext4_extent *ex, *fex;
1921 struct ext4_extent *nearex; /* nearest extent */
1922 struct ext4_ext_path *npath = NULL;
1923 int depth, len, err;
1925 int mb_flags = 0, unwritten;
1927 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1928 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1931 depth = ext_depth(inode);
1932 ex = path[depth].p_ext;
1933 eh = path[depth].p_hdr;
1934 if (unlikely(path[depth].p_hdr == NULL)) {
1935 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1939 /* try to insert block into found extent and return */
1940 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1943 * Try to see whether we should rather test the extent on
1944 * right from ex, or from the left of ex. This is because
1945 * ext4_ext_find_extent() can return either extent on the
1946 * left, or on the right from the searched position. This
1947 * will make merging more effective.
1949 if (ex < EXT_LAST_EXTENT(eh) &&
1950 (le32_to_cpu(ex->ee_block) +
1951 ext4_ext_get_actual_len(ex) <
1952 le32_to_cpu(newext->ee_block))) {
1955 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1956 (le32_to_cpu(newext->ee_block) +
1957 ext4_ext_get_actual_len(newext) <
1958 le32_to_cpu(ex->ee_block)))
1961 /* Try to append newex to the ex */
1962 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1963 ext_debug("append [%d]%d block to %u:[%d]%d"
1965 ext4_ext_is_unwritten(newext),
1966 ext4_ext_get_actual_len(newext),
1967 le32_to_cpu(ex->ee_block),
1968 ext4_ext_is_unwritten(ex),
1969 ext4_ext_get_actual_len(ex),
1970 ext4_ext_pblock(ex));
1971 err = ext4_ext_get_access(handle, inode,
1975 unwritten = ext4_ext_is_unwritten(ex);
1976 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1977 + ext4_ext_get_actual_len(newext));
1979 ext4_ext_mark_unwritten(ex);
1980 eh = path[depth].p_hdr;
1986 /* Try to prepend newex to the ex */
1987 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1988 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1990 le32_to_cpu(newext->ee_block),
1991 ext4_ext_is_unwritten(newext),
1992 ext4_ext_get_actual_len(newext),
1993 le32_to_cpu(ex->ee_block),
1994 ext4_ext_is_unwritten(ex),
1995 ext4_ext_get_actual_len(ex),
1996 ext4_ext_pblock(ex));
1997 err = ext4_ext_get_access(handle, inode,
2002 unwritten = ext4_ext_is_unwritten(ex);
2003 ex->ee_block = newext->ee_block;
2004 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2005 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2006 + ext4_ext_get_actual_len(newext));
2008 ext4_ext_mark_unwritten(ex);
2009 eh = path[depth].p_hdr;
2015 depth = ext_depth(inode);
2016 eh = path[depth].p_hdr;
2017 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2020 /* probably next leaf has space for us? */
2021 fex = EXT_LAST_EXTENT(eh);
2022 next = EXT_MAX_BLOCKS;
2023 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2024 next = ext4_ext_next_leaf_block(path);
2025 if (next != EXT_MAX_BLOCKS) {
2026 ext_debug("next leaf block - %u\n", next);
2027 BUG_ON(npath != NULL);
2028 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2030 return PTR_ERR(npath);
2031 BUG_ON(npath->p_depth != path->p_depth);
2032 eh = npath[depth].p_hdr;
2033 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2034 ext_debug("next leaf isn't full(%d)\n",
2035 le16_to_cpu(eh->eh_entries));
2039 ext_debug("next leaf has no free space(%d,%d)\n",
2040 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2044 * There is no free space in the found leaf.
2045 * We're gonna add a new leaf in the tree.
2047 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2048 mb_flags = EXT4_MB_USE_RESERVED;
2049 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2053 depth = ext_depth(inode);
2054 eh = path[depth].p_hdr;
2057 nearex = path[depth].p_ext;
2059 err = ext4_ext_get_access(handle, inode, path + depth);
2064 /* there is no extent in this leaf, create first one */
2065 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2066 le32_to_cpu(newext->ee_block),
2067 ext4_ext_pblock(newext),
2068 ext4_ext_is_unwritten(newext),
2069 ext4_ext_get_actual_len(newext));
2070 nearex = EXT_FIRST_EXTENT(eh);
2072 if (le32_to_cpu(newext->ee_block)
2073 > le32_to_cpu(nearex->ee_block)) {
2075 ext_debug("insert %u:%llu:[%d]%d before: "
2077 le32_to_cpu(newext->ee_block),
2078 ext4_ext_pblock(newext),
2079 ext4_ext_is_unwritten(newext),
2080 ext4_ext_get_actual_len(newext),
2085 BUG_ON(newext->ee_block == nearex->ee_block);
2086 ext_debug("insert %u:%llu:[%d]%d after: "
2088 le32_to_cpu(newext->ee_block),
2089 ext4_ext_pblock(newext),
2090 ext4_ext_is_unwritten(newext),
2091 ext4_ext_get_actual_len(newext),
2094 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2096 ext_debug("insert %u:%llu:[%d]%d: "
2097 "move %d extents from 0x%p to 0x%p\n",
2098 le32_to_cpu(newext->ee_block),
2099 ext4_ext_pblock(newext),
2100 ext4_ext_is_unwritten(newext),
2101 ext4_ext_get_actual_len(newext),
2102 len, nearex, nearex + 1);
2103 memmove(nearex + 1, nearex,
2104 len * sizeof(struct ext4_extent));
2108 le16_add_cpu(&eh->eh_entries, 1);
2109 path[depth].p_ext = nearex;
2110 nearex->ee_block = newext->ee_block;
2111 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2112 nearex->ee_len = newext->ee_len;
2115 /* try to merge extents */
2116 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2117 ext4_ext_try_to_merge(handle, inode, path, nearex);
2120 /* time to correct all indexes above */
2121 err = ext4_ext_correct_indexes(handle, inode, path);
2125 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2129 ext4_ext_drop_refs(npath);
2135 static int ext4_fill_fiemap_extents(struct inode *inode,
2136 ext4_lblk_t block, ext4_lblk_t num,
2137 struct fiemap_extent_info *fieinfo)
2139 struct ext4_ext_path *path = NULL;
2140 struct ext4_extent *ex;
2141 struct extent_status es;
2142 ext4_lblk_t next, next_del, start = 0, end = 0;
2143 ext4_lblk_t last = block + num;
2144 int exists, depth = 0, err = 0;
2145 unsigned int flags = 0;
2146 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2148 while (block < last && block != EXT_MAX_BLOCKS) {
2150 /* find extent for this block */
2151 down_read(&EXT4_I(inode)->i_data_sem);
2153 if (path && ext_depth(inode) != depth) {
2154 /* depth was changed. we have to realloc path */
2159 path = ext4_ext_find_extent(inode, block, &path, 0);
2161 up_read(&EXT4_I(inode)->i_data_sem);
2162 err = PTR_ERR(path);
2167 depth = ext_depth(inode);
2168 if (unlikely(path[depth].p_hdr == NULL)) {
2169 up_read(&EXT4_I(inode)->i_data_sem);
2170 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2174 ex = path[depth].p_ext;
2175 next = ext4_ext_next_allocated_block(path);
2176 ext4_ext_drop_refs(path);
2181 /* there is no extent yet, so try to allocate
2182 * all requested space */
2185 } else if (le32_to_cpu(ex->ee_block) > block) {
2186 /* need to allocate space before found extent */
2188 end = le32_to_cpu(ex->ee_block);
2189 if (block + num < end)
2191 } else if (block >= le32_to_cpu(ex->ee_block)
2192 + ext4_ext_get_actual_len(ex)) {
2193 /* need to allocate space after found extent */
2198 } else if (block >= le32_to_cpu(ex->ee_block)) {
2200 * some part of requested space is covered
2204 end = le32_to_cpu(ex->ee_block)
2205 + ext4_ext_get_actual_len(ex);
2206 if (block + num < end)
2212 BUG_ON(end <= start);
2216 es.es_len = end - start;
2219 es.es_lblk = le32_to_cpu(ex->ee_block);
2220 es.es_len = ext4_ext_get_actual_len(ex);
2221 es.es_pblk = ext4_ext_pblock(ex);
2222 if (ext4_ext_is_unwritten(ex))
2223 flags |= FIEMAP_EXTENT_UNWRITTEN;
2227 * Find delayed extent and update es accordingly. We call
2228 * it even in !exists case to find out whether es is the
2229 * last existing extent or not.
2231 next_del = ext4_find_delayed_extent(inode, &es);
2232 if (!exists && next_del) {
2234 flags |= (FIEMAP_EXTENT_DELALLOC |
2235 FIEMAP_EXTENT_UNKNOWN);
2237 up_read(&EXT4_I(inode)->i_data_sem);
2239 if (unlikely(es.es_len == 0)) {
2240 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2246 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2247 * we need to check next == EXT_MAX_BLOCKS because it is
2248 * possible that an extent is with unwritten and delayed
2249 * status due to when an extent is delayed allocated and
2250 * is allocated by fallocate status tree will track both of
2253 * So we could return a unwritten and delayed extent, and
2254 * its block is equal to 'next'.
2256 if (next == next_del && next == EXT_MAX_BLOCKS) {
2257 flags |= FIEMAP_EXTENT_LAST;
2258 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2259 next != EXT_MAX_BLOCKS)) {
2260 EXT4_ERROR_INODE(inode,
2261 "next extent == %u, next "
2262 "delalloc extent = %u",
2270 err = fiemap_fill_next_extent(fieinfo,
2271 (__u64)es.es_lblk << blksize_bits,
2272 (__u64)es.es_pblk << blksize_bits,
2273 (__u64)es.es_len << blksize_bits,
2283 block = es.es_lblk + es.es_len;
2287 ext4_ext_drop_refs(path);
2295 * ext4_ext_put_gap_in_cache:
2296 * calculate boundaries of the gap that the requested block fits into
2297 * and cache this gap
2300 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2303 int depth = ext_depth(inode);
2304 unsigned long len = 0;
2305 ext4_lblk_t lblock = 0;
2306 struct ext4_extent *ex;
2308 ex = path[depth].p_ext;
2311 * there is no extent yet, so gap is [0;-] and we
2314 ext_debug("cache gap(whole file):");
2315 } else if (block < le32_to_cpu(ex->ee_block)) {
2317 len = le32_to_cpu(ex->ee_block) - block;
2318 ext_debug("cache gap(before): %u [%u:%u]",
2320 le32_to_cpu(ex->ee_block),
2321 ext4_ext_get_actual_len(ex));
2322 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2323 ext4_es_insert_extent(inode, lblock, len, ~0,
2324 EXTENT_STATUS_HOLE);
2325 } else if (block >= le32_to_cpu(ex->ee_block)
2326 + ext4_ext_get_actual_len(ex)) {
2328 lblock = le32_to_cpu(ex->ee_block)
2329 + ext4_ext_get_actual_len(ex);
2331 next = ext4_ext_next_allocated_block(path);
2332 ext_debug("cache gap(after): [%u:%u] %u",
2333 le32_to_cpu(ex->ee_block),
2334 ext4_ext_get_actual_len(ex),
2336 BUG_ON(next == lblock);
2337 len = next - lblock;
2338 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2339 ext4_es_insert_extent(inode, lblock, len, ~0,
2340 EXTENT_STATUS_HOLE);
2345 ext_debug(" -> %u:%lu\n", lblock, len);
2350 * removes index from the index block.
2352 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2353 struct ext4_ext_path *path, int depth)
2358 /* free index block */
2360 path = path + depth;
2361 leaf = ext4_idx_pblock(path->p_idx);
2362 if (unlikely(path->p_hdr->eh_entries == 0)) {
2363 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2366 err = ext4_ext_get_access(handle, inode, path);
2370 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2371 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2372 len *= sizeof(struct ext4_extent_idx);
2373 memmove(path->p_idx, path->p_idx + 1, len);
2376 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2377 err = ext4_ext_dirty(handle, inode, path);
2380 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2381 trace_ext4_ext_rm_idx(inode, leaf);
2383 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2384 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2386 while (--depth >= 0) {
2387 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2390 err = ext4_ext_get_access(handle, inode, path);
2393 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2394 err = ext4_ext_dirty(handle, inode, path);
2402 * ext4_ext_calc_credits_for_single_extent:
2403 * This routine returns max. credits that needed to insert an extent
2404 * to the extent tree.
2405 * When pass the actual path, the caller should calculate credits
2408 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2409 struct ext4_ext_path *path)
2412 int depth = ext_depth(inode);
2415 /* probably there is space in leaf? */
2416 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2417 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2420 * There are some space in the leaf tree, no
2421 * need to account for leaf block credit
2423 * bitmaps and block group descriptor blocks
2424 * and other metadata blocks still need to be
2427 /* 1 bitmap, 1 block group descriptor */
2428 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2433 return ext4_chunk_trans_blocks(inode, nrblocks);
2437 * How many index/leaf blocks need to change/allocate to add @extents extents?
2439 * If we add a single extent, then in the worse case, each tree level
2440 * index/leaf need to be changed in case of the tree split.
2442 * If more extents are inserted, they could cause the whole tree split more
2443 * than once, but this is really rare.
2445 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2450 /* If we are converting the inline data, only one is needed here. */
2451 if (ext4_has_inline_data(inode))
2454 depth = ext_depth(inode);
2464 static inline int get_default_free_blocks_flags(struct inode *inode)
2466 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2467 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2468 else if (ext4_should_journal_data(inode))
2469 return EXT4_FREE_BLOCKS_FORGET;
2473 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2474 struct ext4_extent *ex,
2475 long long *partial_cluster,
2476 ext4_lblk_t from, ext4_lblk_t to)
2478 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2479 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2481 int flags = get_default_free_blocks_flags(inode);
2484 * For bigalloc file systems, we never free a partial cluster
2485 * at the beginning of the extent. Instead, we make a note
2486 * that we tried freeing the cluster, and check to see if we
2487 * need to free it on a subsequent call to ext4_remove_blocks,
2488 * or at the end of the ext4_truncate() operation.
2490 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2492 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2494 * If we have a partial cluster, and it's different from the
2495 * cluster of the last block, we need to explicitly free the
2496 * partial cluster here.
2498 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2499 if ((*partial_cluster > 0) &&
2500 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2501 ext4_free_blocks(handle, inode, NULL,
2502 EXT4_C2B(sbi, *partial_cluster),
2503 sbi->s_cluster_ratio, flags);
2504 *partial_cluster = 0;
2507 #ifdef EXTENTS_STATS
2509 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2510 spin_lock(&sbi->s_ext_stats_lock);
2511 sbi->s_ext_blocks += ee_len;
2512 sbi->s_ext_extents++;
2513 if (ee_len < sbi->s_ext_min)
2514 sbi->s_ext_min = ee_len;
2515 if (ee_len > sbi->s_ext_max)
2516 sbi->s_ext_max = ee_len;
2517 if (ext_depth(inode) > sbi->s_depth_max)
2518 sbi->s_depth_max = ext_depth(inode);
2519 spin_unlock(&sbi->s_ext_stats_lock);
2522 if (from >= le32_to_cpu(ex->ee_block)
2523 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2526 unsigned int unaligned;
2528 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2529 pblk = ext4_ext_pblock(ex) + ee_len - num;
2531 * Usually we want to free partial cluster at the end of the
2532 * extent, except for the situation when the cluster is still
2533 * used by any other extent (partial_cluster is negative).
2535 if (*partial_cluster < 0 &&
2536 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2537 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2539 ext_debug("free last %u blocks starting %llu partial %lld\n",
2540 num, pblk, *partial_cluster);
2541 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2543 * If the block range to be freed didn't start at the
2544 * beginning of a cluster, and we removed the entire
2545 * extent and the cluster is not used by any other extent,
2546 * save the partial cluster here, since we might need to
2547 * delete if we determine that the truncate operation has
2548 * removed all of the blocks in the cluster.
2550 * On the other hand, if we did not manage to free the whole
2551 * extent, we have to mark the cluster as used (store negative
2552 * cluster number in partial_cluster).
2554 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2555 if (unaligned && (ee_len == num) &&
2556 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2557 *partial_cluster = EXT4_B2C(sbi, pblk);
2559 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2560 else if (*partial_cluster > 0)
2561 *partial_cluster = 0;
2563 ext4_error(sbi->s_sb, "strange request: removal(2) "
2564 "%u-%u from %u:%u\n",
2565 from, to, le32_to_cpu(ex->ee_block), ee_len);
2571 * ext4_ext_rm_leaf() Removes the extents associated with the
2572 * blocks appearing between "start" and "end", and splits the extents
2573 * if "start" and "end" appear in the same extent
2575 * @handle: The journal handle
2576 * @inode: The files inode
2577 * @path: The path to the leaf
2578 * @partial_cluster: The cluster which we'll have to free if all extents
2579 * has been released from it. It gets negative in case
2580 * that the cluster is still used.
2581 * @start: The first block to remove
2582 * @end: The last block to remove
2585 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2586 struct ext4_ext_path *path,
2587 long long *partial_cluster,
2588 ext4_lblk_t start, ext4_lblk_t end)
2590 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2591 int err = 0, correct_index = 0;
2592 int depth = ext_depth(inode), credits;
2593 struct ext4_extent_header *eh;
2596 ext4_lblk_t ex_ee_block;
2597 unsigned short ex_ee_len;
2598 unsigned unwritten = 0;
2599 struct ext4_extent *ex;
2602 /* the header must be checked already in ext4_ext_remove_space() */
2603 ext_debug("truncate since %u in leaf to %u\n", start, end);
2604 if (!path[depth].p_hdr)
2605 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2606 eh = path[depth].p_hdr;
2607 if (unlikely(path[depth].p_hdr == NULL)) {
2608 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2611 /* find where to start removing */
2612 ex = path[depth].p_ext;
2614 ex = EXT_LAST_EXTENT(eh);
2616 ex_ee_block = le32_to_cpu(ex->ee_block);
2617 ex_ee_len = ext4_ext_get_actual_len(ex);
2620 * If we're starting with an extent other than the last one in the
2621 * node, we need to see if it shares a cluster with the extent to
2622 * the right (towards the end of the file). If its leftmost cluster
2623 * is this extent's rightmost cluster and it is not cluster aligned,
2624 * we'll mark it as a partial that is not to be deallocated.
2627 if (ex != EXT_LAST_EXTENT(eh)) {
2628 ext4_fsblk_t current_pblk, right_pblk;
2629 long long current_cluster, right_cluster;
2631 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2632 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2633 right_pblk = ext4_ext_pblock(ex + 1);
2634 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2635 if (current_cluster == right_cluster &&
2636 EXT4_PBLK_COFF(sbi, right_pblk))
2637 *partial_cluster = -right_cluster;
2640 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2642 while (ex >= EXT_FIRST_EXTENT(eh) &&
2643 ex_ee_block + ex_ee_len > start) {
2645 if (ext4_ext_is_unwritten(ex))
2650 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2651 unwritten, ex_ee_len);
2652 path[depth].p_ext = ex;
2654 a = ex_ee_block > start ? ex_ee_block : start;
2655 b = ex_ee_block+ex_ee_len - 1 < end ?
2656 ex_ee_block+ex_ee_len - 1 : end;
2658 ext_debug(" border %u:%u\n", a, b);
2660 /* If this extent is beyond the end of the hole, skip it */
2661 if (end < ex_ee_block) {
2663 * We're going to skip this extent and move to another,
2664 * so if this extent is not cluster aligned we have
2665 * to mark the current cluster as used to avoid
2666 * accidentally freeing it later on
2668 pblk = ext4_ext_pblock(ex);
2669 if (EXT4_PBLK_COFF(sbi, pblk))
2671 -((long long)EXT4_B2C(sbi, pblk));
2673 ex_ee_block = le32_to_cpu(ex->ee_block);
2674 ex_ee_len = ext4_ext_get_actual_len(ex);
2676 } else if (b != ex_ee_block + ex_ee_len - 1) {
2677 EXT4_ERROR_INODE(inode,
2678 "can not handle truncate %u:%u "
2680 start, end, ex_ee_block,
2681 ex_ee_block + ex_ee_len - 1);
2684 } else if (a != ex_ee_block) {
2685 /* remove tail of the extent */
2686 num = a - ex_ee_block;
2688 /* remove whole extent: excellent! */
2692 * 3 for leaf, sb, and inode plus 2 (bmap and group
2693 * descriptor) for each block group; assume two block
2694 * groups plus ex_ee_len/blocks_per_block_group for
2697 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2698 if (ex == EXT_FIRST_EXTENT(eh)) {
2700 credits += (ext_depth(inode)) + 1;
2702 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2704 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2708 err = ext4_ext_get_access(handle, inode, path + depth);
2712 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2718 /* this extent is removed; mark slot entirely unused */
2719 ext4_ext_store_pblock(ex, 0);
2721 ex->ee_len = cpu_to_le16(num);
2723 * Do not mark unwritten if all the blocks in the
2724 * extent have been removed.
2726 if (unwritten && num)
2727 ext4_ext_mark_unwritten(ex);
2729 * If the extent was completely released,
2730 * we need to remove it from the leaf
2733 if (end != EXT_MAX_BLOCKS - 1) {
2735 * For hole punching, we need to scoot all the
2736 * extents up when an extent is removed so that
2737 * we dont have blank extents in the middle
2739 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2740 sizeof(struct ext4_extent));
2742 /* Now get rid of the one at the end */
2743 memset(EXT_LAST_EXTENT(eh), 0,
2744 sizeof(struct ext4_extent));
2746 le16_add_cpu(&eh->eh_entries, -1);
2747 } else if (*partial_cluster > 0)
2748 *partial_cluster = 0;
2750 err = ext4_ext_dirty(handle, inode, path + depth);
2754 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2755 ext4_ext_pblock(ex));
2757 ex_ee_block = le32_to_cpu(ex->ee_block);
2758 ex_ee_len = ext4_ext_get_actual_len(ex);
2761 if (correct_index && eh->eh_entries)
2762 err = ext4_ext_correct_indexes(handle, inode, path);
2765 * If there's a partial cluster and at least one extent remains in
2766 * the leaf, free the partial cluster if it isn't shared with the
2767 * current extent. If there's a partial cluster and no extents
2768 * remain in the leaf, it can't be freed here. It can only be
2769 * freed when it's possible to determine if it's not shared with
2770 * any other extent - when the next leaf is processed or when space
2771 * removal is complete.
2773 if (*partial_cluster > 0 && eh->eh_entries &&
2774 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2775 *partial_cluster)) {
2776 int flags = get_default_free_blocks_flags(inode);
2778 ext4_free_blocks(handle, inode, NULL,
2779 EXT4_C2B(sbi, *partial_cluster),
2780 sbi->s_cluster_ratio, flags);
2781 *partial_cluster = 0;
2784 /* if this leaf is free, then we should
2785 * remove it from index block above */
2786 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2787 err = ext4_ext_rm_idx(handle, inode, path, depth);
2794 * ext4_ext_more_to_rm:
2795 * returns 1 if current index has to be freed (even partial)
2798 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2800 BUG_ON(path->p_idx == NULL);
2802 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2806 * if truncate on deeper level happened, it wasn't partial,
2807 * so we have to consider current index for truncation
2809 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2814 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2817 struct super_block *sb = inode->i_sb;
2818 int depth = ext_depth(inode);
2819 struct ext4_ext_path *path = NULL;
2820 long long partial_cluster = 0;
2824 ext_debug("truncate since %u to %u\n", start, end);
2826 /* probably first extent we're gonna free will be last in block */
2827 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2829 return PTR_ERR(handle);
2832 trace_ext4_ext_remove_space(inode, start, end, depth);
2835 * Check if we are removing extents inside the extent tree. If that
2836 * is the case, we are going to punch a hole inside the extent tree
2837 * so we have to check whether we need to split the extent covering
2838 * the last block to remove so we can easily remove the part of it
2839 * in ext4_ext_rm_leaf().
2841 if (end < EXT_MAX_BLOCKS - 1) {
2842 struct ext4_extent *ex;
2843 ext4_lblk_t ee_block;
2845 /* find extent for this block */
2846 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2848 ext4_journal_stop(handle);
2849 return PTR_ERR(path);
2851 depth = ext_depth(inode);
2852 /* Leaf not may not exist only if inode has no blocks at all */
2853 ex = path[depth].p_ext;
2856 EXT4_ERROR_INODE(inode,
2857 "path[%d].p_hdr == NULL",
2864 ee_block = le32_to_cpu(ex->ee_block);
2867 * See if the last block is inside the extent, if so split
2868 * the extent at 'end' block so we can easily remove the
2869 * tail of the first part of the split extent in
2870 * ext4_ext_rm_leaf().
2872 if (end >= ee_block &&
2873 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2875 * Split the extent in two so that 'end' is the last
2876 * block in the first new extent. Also we should not
2877 * fail removing space due to ENOSPC so try to use
2878 * reserved block if that happens.
2880 err = ext4_force_split_extent_at(handle, inode, &path,
2887 * We start scanning from right side, freeing all the blocks
2888 * after i_size and walking into the tree depth-wise.
2890 depth = ext_depth(inode);
2895 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2897 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2900 ext4_journal_stop(handle);
2903 path[0].p_depth = depth;
2904 path[0].p_hdr = ext_inode_hdr(inode);
2907 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2914 while (i >= 0 && err == 0) {
2916 /* this is leaf block */
2917 err = ext4_ext_rm_leaf(handle, inode, path,
2918 &partial_cluster, start,
2920 /* root level has p_bh == NULL, brelse() eats this */
2921 brelse(path[i].p_bh);
2922 path[i].p_bh = NULL;
2927 /* this is index block */
2928 if (!path[i].p_hdr) {
2929 ext_debug("initialize header\n");
2930 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2933 if (!path[i].p_idx) {
2934 /* this level hasn't been touched yet */
2935 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2936 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2937 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2939 le16_to_cpu(path[i].p_hdr->eh_entries));
2941 /* we were already here, see at next index */
2945 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2946 i, EXT_FIRST_INDEX(path[i].p_hdr),
2948 if (ext4_ext_more_to_rm(path + i)) {
2949 struct buffer_head *bh;
2950 /* go to the next level */
2951 ext_debug("move to level %d (block %llu)\n",
2952 i + 1, ext4_idx_pblock(path[i].p_idx));
2953 memset(path + i + 1, 0, sizeof(*path));
2954 bh = read_extent_tree_block(inode,
2955 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2958 /* should we reset i_size? */
2962 /* Yield here to deal with large extent trees.
2963 * Should be a no-op if we did IO above. */
2965 if (WARN_ON(i + 1 > depth)) {
2969 path[i + 1].p_bh = bh;
2971 /* save actual number of indexes since this
2972 * number is changed at the next iteration */
2973 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2976 /* we finished processing this index, go up */
2977 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2978 /* index is empty, remove it;
2979 * handle must be already prepared by the
2980 * truncatei_leaf() */
2981 err = ext4_ext_rm_idx(handle, inode, path, i);
2983 /* root level has p_bh == NULL, brelse() eats this */
2984 brelse(path[i].p_bh);
2985 path[i].p_bh = NULL;
2987 ext_debug("return to level %d\n", i);
2991 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2992 partial_cluster, path->p_hdr->eh_entries);
2994 /* If we still have something in the partial cluster and we have removed
2995 * even the first extent, then we should free the blocks in the partial
2996 * cluster as well. */
2997 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2998 int flags = get_default_free_blocks_flags(inode);
3000 ext4_free_blocks(handle, inode, NULL,
3001 EXT4_C2B(EXT4_SB(sb), partial_cluster),
3002 EXT4_SB(sb)->s_cluster_ratio, flags);
3003 partial_cluster = 0;
3006 /* TODO: flexible tree reduction should be here */
3007 if (path->p_hdr->eh_entries == 0) {
3009 * truncate to zero freed all the tree,
3010 * so we need to correct eh_depth
3012 err = ext4_ext_get_access(handle, inode, path);
3014 ext_inode_hdr(inode)->eh_depth = 0;
3015 ext_inode_hdr(inode)->eh_max =
3016 cpu_to_le16(ext4_ext_space_root(inode, 0));
3017 err = ext4_ext_dirty(handle, inode, path);
3022 ext4_ext_drop_refs(path);
3028 ext4_journal_stop(handle);
3034 * called at mount time
3036 void ext4_ext_init(struct super_block *sb)
3039 * possible initialization would be here
3042 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3043 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3044 printk(KERN_INFO "EXT4-fs: file extents enabled"
3045 #ifdef AGGRESSIVE_TEST
3046 ", aggressive tests"
3048 #ifdef CHECK_BINSEARCH
3051 #ifdef EXTENTS_STATS
3056 #ifdef EXTENTS_STATS
3057 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3058 EXT4_SB(sb)->s_ext_min = 1 << 30;
3059 EXT4_SB(sb)->s_ext_max = 0;
3065 * called at umount time
3067 void ext4_ext_release(struct super_block *sb)
3069 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3072 #ifdef EXTENTS_STATS
3073 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3074 struct ext4_sb_info *sbi = EXT4_SB(sb);
3075 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3076 sbi->s_ext_blocks, sbi->s_ext_extents,
3077 sbi->s_ext_blocks / sbi->s_ext_extents);
3078 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3079 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3084 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3086 ext4_lblk_t ee_block;
3087 ext4_fsblk_t ee_pblock;
3088 unsigned int ee_len;
3090 ee_block = le32_to_cpu(ex->ee_block);
3091 ee_len = ext4_ext_get_actual_len(ex);
3092 ee_pblock = ext4_ext_pblock(ex);
3097 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3098 EXTENT_STATUS_WRITTEN);
3101 /* FIXME!! we need to try to merge to left or right after zero-out */
3102 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3104 ext4_fsblk_t ee_pblock;
3105 unsigned int ee_len;
3108 ee_len = ext4_ext_get_actual_len(ex);
3109 ee_pblock = ext4_ext_pblock(ex);
3111 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3119 * ext4_split_extent_at() splits an extent at given block.
3121 * @handle: the journal handle
3122 * @inode: the file inode
3123 * @path: the path to the extent
3124 * @split: the logical block where the extent is splitted.
3125 * @split_flags: indicates if the extent could be zeroout if split fails, and
3126 * the states(init or unwritten) of new extents.
3127 * @flags: flags used to insert new extent to extent tree.
3130 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3131 * of which are deterimined by split_flag.
3133 * There are two cases:
3134 * a> the extent are splitted into two extent.
3135 * b> split is not needed, and just mark the extent.
3137 * return 0 on success.
3139 static int ext4_split_extent_at(handle_t *handle,
3140 struct inode *inode,
3141 struct ext4_ext_path **ppath,
3146 struct ext4_ext_path *path = *ppath;
3147 ext4_fsblk_t newblock;
3148 ext4_lblk_t ee_block;
3149 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3150 struct ext4_extent *ex2 = NULL;
3151 unsigned int ee_len, depth;
3154 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3155 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3157 ext_debug("ext4_split_extents_at: inode %lu, logical"
3158 "block %llu\n", inode->i_ino, (unsigned long long)split);
3160 ext4_ext_show_leaf(inode, path);
3162 depth = ext_depth(inode);
3163 ex = path[depth].p_ext;
3164 ee_block = le32_to_cpu(ex->ee_block);
3165 ee_len = ext4_ext_get_actual_len(ex);
3166 newblock = split - ee_block + ext4_ext_pblock(ex);
3168 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3169 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3170 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3171 EXT4_EXT_MARK_UNWRIT1 |
3172 EXT4_EXT_MARK_UNWRIT2));
3174 err = ext4_ext_get_access(handle, inode, path + depth);
3178 if (split == ee_block) {
3180 * case b: block @split is the block that the extent begins with
3181 * then we just change the state of the extent, and splitting
3184 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3185 ext4_ext_mark_unwritten(ex);
3187 ext4_ext_mark_initialized(ex);
3189 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3190 ext4_ext_try_to_merge(handle, inode, path, ex);
3192 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3197 memcpy(&orig_ex, ex, sizeof(orig_ex));
3198 ex->ee_len = cpu_to_le16(split - ee_block);
3199 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3200 ext4_ext_mark_unwritten(ex);
3203 * path may lead to new leaf, not to original leaf any more
3204 * after ext4_ext_insert_extent() returns,
3206 err = ext4_ext_dirty(handle, inode, path + depth);
3208 goto fix_extent_len;
3211 ex2->ee_block = cpu_to_le32(split);
3212 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3213 ext4_ext_store_pblock(ex2, newblock);
3214 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3215 ext4_ext_mark_unwritten(ex2);
3217 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3218 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3219 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3220 if (split_flag & EXT4_EXT_DATA_VALID1) {
3221 err = ext4_ext_zeroout(inode, ex2);
3222 zero_ex.ee_block = ex2->ee_block;
3223 zero_ex.ee_len = cpu_to_le16(
3224 ext4_ext_get_actual_len(ex2));
3225 ext4_ext_store_pblock(&zero_ex,
3226 ext4_ext_pblock(ex2));
3228 err = ext4_ext_zeroout(inode, ex);
3229 zero_ex.ee_block = ex->ee_block;
3230 zero_ex.ee_len = cpu_to_le16(
3231 ext4_ext_get_actual_len(ex));
3232 ext4_ext_store_pblock(&zero_ex,
3233 ext4_ext_pblock(ex));
3236 err = ext4_ext_zeroout(inode, &orig_ex);
3237 zero_ex.ee_block = orig_ex.ee_block;
3238 zero_ex.ee_len = cpu_to_le16(
3239 ext4_ext_get_actual_len(&orig_ex));
3240 ext4_ext_store_pblock(&zero_ex,
3241 ext4_ext_pblock(&orig_ex));
3245 goto fix_extent_len;
3246 /* update the extent length and mark as initialized */
3247 ex->ee_len = cpu_to_le16(ee_len);
3248 ext4_ext_try_to_merge(handle, inode, path, ex);
3249 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3251 goto fix_extent_len;
3253 /* update extent status tree */
3254 err = ext4_zeroout_es(inode, &zero_ex);
3258 goto fix_extent_len;
3261 ext4_ext_show_leaf(inode, path);
3265 ex->ee_len = orig_ex.ee_len;
3266 ext4_ext_dirty(handle, inode, path + path->p_depth);
3271 * ext4_split_extents() splits an extent and mark extent which is covered
3272 * by @map as split_flags indicates
3274 * It may result in splitting the extent into multiple extents (up to three)
3275 * There are three possibilities:
3276 * a> There is no split required
3277 * b> Splits in two extents: Split is happening at either end of the extent
3278 * c> Splits in three extents: Somone is splitting in middle of the extent
3281 static int ext4_split_extent(handle_t *handle,
3282 struct inode *inode,
3283 struct ext4_ext_path **ppath,
3284 struct ext4_map_blocks *map,
3288 struct ext4_ext_path *path = *ppath;
3289 ext4_lblk_t ee_block;
3290 struct ext4_extent *ex;
3291 unsigned int ee_len, depth;
3294 int split_flag1, flags1;
3295 int allocated = map->m_len;
3297 depth = ext_depth(inode);
3298 ex = path[depth].p_ext;
3299 ee_block = le32_to_cpu(ex->ee_block);
3300 ee_len = ext4_ext_get_actual_len(ex);
3301 unwritten = ext4_ext_is_unwritten(ex);
3303 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3304 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3305 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3307 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3308 EXT4_EXT_MARK_UNWRIT2;
3309 if (split_flag & EXT4_EXT_DATA_VALID2)
3310 split_flag1 |= EXT4_EXT_DATA_VALID1;
3311 err = ext4_split_extent_at(handle, inode, ppath,
3312 map->m_lblk + map->m_len, split_flag1, flags1);
3316 allocated = ee_len - (map->m_lblk - ee_block);
3319 * Update path is required because previous ext4_split_extent_at() may
3320 * result in split of original leaf or extent zeroout.
3322 ext4_ext_drop_refs(path);
3323 path = ext4_ext_find_extent(inode, map->m_lblk, ppath, 0);
3325 return PTR_ERR(path);
3326 depth = ext_depth(inode);
3327 ex = path[depth].p_ext;
3329 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3330 (unsigned long) map->m_lblk);
3333 unwritten = ext4_ext_is_unwritten(ex);
3336 if (map->m_lblk >= ee_block) {
3337 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3339 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3340 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3341 EXT4_EXT_MARK_UNWRIT2);
3343 err = ext4_split_extent_at(handle, inode, ppath,
3344 map->m_lblk, split_flag1, flags);
3349 ext4_ext_show_leaf(inode, path);
3351 return err ? err : allocated;
3355 * This function is called by ext4_ext_map_blocks() if someone tries to write
3356 * to an unwritten extent. It may result in splitting the unwritten
3357 * extent into multiple extents (up to three - one initialized and two
3359 * There are three possibilities:
3360 * a> There is no split required: Entire extent should be initialized
3361 * b> Splits in two extents: Write is happening at either end of the extent
3362 * c> Splits in three extents: Somone is writing in middle of the extent
3365 * - The extent pointed to by 'path' is unwritten.
3366 * - The extent pointed to by 'path' contains a superset
3367 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3369 * Post-conditions on success:
3370 * - the returned value is the number of blocks beyond map->l_lblk
3371 * that are allocated and initialized.
3372 * It is guaranteed to be >= map->m_len.
3374 static int ext4_ext_convert_to_initialized(handle_t *handle,
3375 struct inode *inode,
3376 struct ext4_map_blocks *map,
3377 struct ext4_ext_path **ppath,
3380 struct ext4_ext_path *path = *ppath;
3381 struct ext4_sb_info *sbi;
3382 struct ext4_extent_header *eh;
3383 struct ext4_map_blocks split_map;
3384 struct ext4_extent zero_ex;
3385 struct ext4_extent *ex, *abut_ex;
3386 ext4_lblk_t ee_block, eof_block;
3387 unsigned int ee_len, depth, map_len = map->m_len;
3388 int allocated = 0, max_zeroout = 0;
3392 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3393 "block %llu, max_blocks %u\n", inode->i_ino,
3394 (unsigned long long)map->m_lblk, map_len);
3396 sbi = EXT4_SB(inode->i_sb);
3397 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3398 inode->i_sb->s_blocksize_bits;
3399 if (eof_block < map->m_lblk + map_len)
3400 eof_block = map->m_lblk + map_len;
3402 depth = ext_depth(inode);
3403 eh = path[depth].p_hdr;
3404 ex = path[depth].p_ext;
3405 ee_block = le32_to_cpu(ex->ee_block);
3406 ee_len = ext4_ext_get_actual_len(ex);
3409 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3411 /* Pre-conditions */
3412 BUG_ON(!ext4_ext_is_unwritten(ex));
3413 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3416 * Attempt to transfer newly initialized blocks from the currently
3417 * unwritten extent to its neighbor. This is much cheaper
3418 * than an insertion followed by a merge as those involve costly
3419 * memmove() calls. Transferring to the left is the common case in
3420 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3421 * followed by append writes.
3423 * Limitations of the current logic:
3424 * - L1: we do not deal with writes covering the whole extent.
3425 * This would require removing the extent if the transfer
3427 * - L2: we only attempt to merge with an extent stored in the
3428 * same extent tree node.
3430 if ((map->m_lblk == ee_block) &&
3431 /* See if we can merge left */
3432 (map_len < ee_len) && /*L1*/
3433 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3434 ext4_lblk_t prev_lblk;
3435 ext4_fsblk_t prev_pblk, ee_pblk;
3436 unsigned int prev_len;
3439 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3440 prev_len = ext4_ext_get_actual_len(abut_ex);
3441 prev_pblk = ext4_ext_pblock(abut_ex);
3442 ee_pblk = ext4_ext_pblock(ex);
3445 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3446 * upon those conditions:
3447 * - C1: abut_ex is initialized,
3448 * - C2: abut_ex is logically abutting ex,
3449 * - C3: abut_ex is physically abutting ex,
3450 * - C4: abut_ex can receive the additional blocks without
3451 * overflowing the (initialized) length limit.
3453 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3454 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3455 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3456 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3457 err = ext4_ext_get_access(handle, inode, path + depth);
3461 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3464 /* Shift the start of ex by 'map_len' blocks */
3465 ex->ee_block = cpu_to_le32(ee_block + map_len);
3466 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3467 ex->ee_len = cpu_to_le16(ee_len - map_len);
3468 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3470 /* Extend abut_ex by 'map_len' blocks */
3471 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3473 /* Result: number of initialized blocks past m_lblk */
3474 allocated = map_len;
3476 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3477 (map_len < ee_len) && /*L1*/
3478 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3479 /* See if we can merge right */
3480 ext4_lblk_t next_lblk;
3481 ext4_fsblk_t next_pblk, ee_pblk;
3482 unsigned int next_len;
3485 next_lblk = le32_to_cpu(abut_ex->ee_block);
3486 next_len = ext4_ext_get_actual_len(abut_ex);
3487 next_pblk = ext4_ext_pblock(abut_ex);
3488 ee_pblk = ext4_ext_pblock(ex);
3491 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3492 * upon those conditions:
3493 * - C1: abut_ex is initialized,
3494 * - C2: abut_ex is logically abutting ex,
3495 * - C3: abut_ex is physically abutting ex,
3496 * - C4: abut_ex can receive the additional blocks without
3497 * overflowing the (initialized) length limit.
3499 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3500 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3501 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3502 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3503 err = ext4_ext_get_access(handle, inode, path + depth);
3507 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3510 /* Shift the start of abut_ex by 'map_len' blocks */
3511 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3512 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3513 ex->ee_len = cpu_to_le16(ee_len - map_len);
3514 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3516 /* Extend abut_ex by 'map_len' blocks */
3517 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3519 /* Result: number of initialized blocks past m_lblk */
3520 allocated = map_len;
3524 /* Mark the block containing both extents as dirty */
3525 ext4_ext_dirty(handle, inode, path + depth);
3527 /* Update path to point to the right extent */
3528 path[depth].p_ext = abut_ex;
3531 allocated = ee_len - (map->m_lblk - ee_block);
3533 WARN_ON(map->m_lblk < ee_block);
3535 * It is safe to convert extent to initialized via explicit
3536 * zeroout only if extent is fully inside i_size or new_size.
3538 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3540 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3541 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3542 (inode->i_sb->s_blocksize_bits - 10);
3544 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3545 if (max_zeroout && (ee_len <= max_zeroout)) {
3546 err = ext4_ext_zeroout(inode, ex);
3549 zero_ex.ee_block = ex->ee_block;
3550 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3551 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3553 err = ext4_ext_get_access(handle, inode, path + depth);
3556 ext4_ext_mark_initialized(ex);
3557 ext4_ext_try_to_merge(handle, inode, path, ex);
3558 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3564 * 1. split the extent into three extents.
3565 * 2. split the extent into two extents, zeroout the first half.
3566 * 3. split the extent into two extents, zeroout the second half.
3567 * 4. split the extent into two extents with out zeroout.
3569 split_map.m_lblk = map->m_lblk;
3570 split_map.m_len = map->m_len;
3572 if (max_zeroout && (allocated > map->m_len)) {
3573 if (allocated <= max_zeroout) {
3576 cpu_to_le32(map->m_lblk);
3577 zero_ex.ee_len = cpu_to_le16(allocated);
3578 ext4_ext_store_pblock(&zero_ex,
3579 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3580 err = ext4_ext_zeroout(inode, &zero_ex);
3583 split_map.m_lblk = map->m_lblk;
3584 split_map.m_len = allocated;
3585 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3587 if (map->m_lblk != ee_block) {
3588 zero_ex.ee_block = ex->ee_block;
3589 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3591 ext4_ext_store_pblock(&zero_ex,
3592 ext4_ext_pblock(ex));
3593 err = ext4_ext_zeroout(inode, &zero_ex);
3598 split_map.m_lblk = ee_block;
3599 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3600 allocated = map->m_len;
3604 allocated = ext4_split_extent(handle, inode, ppath,
3605 &split_map, split_flag, flags);
3610 /* If we have gotten a failure, don't zero out status tree */
3612 err = ext4_zeroout_es(inode, &zero_ex);
3613 return err ? err : allocated;
3617 * This function is called by ext4_ext_map_blocks() from
3618 * ext4_get_blocks_dio_write() when DIO to write
3619 * to an unwritten extent.
3621 * Writing to an unwritten extent may result in splitting the unwritten
3622 * extent into multiple initialized/unwritten extents (up to three)
3623 * There are three possibilities:
3624 * a> There is no split required: Entire extent should be unwritten
3625 * b> Splits in two extents: Write is happening at either end of the extent
3626 * c> Splits in three extents: Somone is writing in middle of the extent
3628 * This works the same way in the case of initialized -> unwritten conversion.
3630 * One of more index blocks maybe needed if the extent tree grow after
3631 * the unwritten extent split. To prevent ENOSPC occur at the IO
3632 * complete, we need to split the unwritten extent before DIO submit
3633 * the IO. The unwritten extent called at this time will be split
3634 * into three unwritten extent(at most). After IO complete, the part
3635 * being filled will be convert to initialized by the end_io callback function
3636 * via ext4_convert_unwritten_extents().
3638 * Returns the size of unwritten extent to be written on success.
3640 static int ext4_split_convert_extents(handle_t *handle,
3641 struct inode *inode,
3642 struct ext4_map_blocks *map,
3643 struct ext4_ext_path **ppath,
3646 struct ext4_ext_path *path = *ppath;
3647 ext4_lblk_t eof_block;
3648 ext4_lblk_t ee_block;
3649 struct ext4_extent *ex;
3650 unsigned int ee_len;
3651 int split_flag = 0, depth;
3653 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3654 __func__, inode->i_ino,
3655 (unsigned long long)map->m_lblk, map->m_len);
3657 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3658 inode->i_sb->s_blocksize_bits;
3659 if (eof_block < map->m_lblk + map->m_len)
3660 eof_block = map->m_lblk + map->m_len;
3662 * It is safe to convert extent to initialized via explicit
3663 * zeroout only if extent is fully insde i_size or new_size.
3665 depth = ext_depth(inode);
3666 ex = path[depth].p_ext;
3667 ee_block = le32_to_cpu(ex->ee_block);
3668 ee_len = ext4_ext_get_actual_len(ex);
3670 /* Convert to unwritten */
3671 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3672 split_flag |= EXT4_EXT_DATA_VALID1;
3673 /* Convert to initialized */
3674 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3675 split_flag |= ee_block + ee_len <= eof_block ?
3676 EXT4_EXT_MAY_ZEROOUT : 0;
3677 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3679 flags |= EXT4_GET_BLOCKS_PRE_IO;
3680 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3683 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3684 struct inode *inode,
3685 struct ext4_map_blocks *map,
3686 struct ext4_ext_path **ppath)
3688 struct ext4_ext_path *path = *ppath;
3689 struct ext4_extent *ex;
3690 ext4_lblk_t ee_block;
3691 unsigned int ee_len;
3695 depth = ext_depth(inode);
3696 ex = path[depth].p_ext;
3697 ee_block = le32_to_cpu(ex->ee_block);
3698 ee_len = ext4_ext_get_actual_len(ex);
3700 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3701 "block %llu, max_blocks %u\n", inode->i_ino,
3702 (unsigned long long)ee_block, ee_len);
3704 /* If extent is larger than requested it is a clear sign that we still
3705 * have some extent state machine issues left. So extent_split is still
3707 * TODO: Once all related issues will be fixed this situation should be
3710 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3712 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3713 " len %u; IO logical block %llu, len %u\n",
3714 inode->i_ino, (unsigned long long)ee_block, ee_len,
3715 (unsigned long long)map->m_lblk, map->m_len);
3717 err = ext4_split_convert_extents(handle, inode, map, ppath,
3718 EXT4_GET_BLOCKS_CONVERT);
3721 ext4_ext_drop_refs(path);
3722 path = ext4_ext_find_extent(inode, map->m_lblk, ppath, 0);
3724 return PTR_ERR(path);
3725 depth = ext_depth(inode);
3726 ex = path[depth].p_ext;
3729 err = ext4_ext_get_access(handle, inode, path + depth);
3732 /* first mark the extent as initialized */
3733 ext4_ext_mark_initialized(ex);
3735 /* note: ext4_ext_correct_indexes() isn't needed here because
3736 * borders are not changed
3738 ext4_ext_try_to_merge(handle, inode, path, ex);
3740 /* Mark modified extent as dirty */
3741 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3743 ext4_ext_show_leaf(inode, path);
3747 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3748 sector_t block, int count)
3751 for (i = 0; i < count; i++)
3752 unmap_underlying_metadata(bdev, block + i);
3756 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3758 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3760 struct ext4_ext_path *path,
3764 struct ext4_extent_header *eh;
3765 struct ext4_extent *last_ex;
3767 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3770 depth = ext_depth(inode);
3771 eh = path[depth].p_hdr;
3774 * We're going to remove EOFBLOCKS_FL entirely in future so we
3775 * do not care for this case anymore. Simply remove the flag
3776 * if there are no extents.
3778 if (unlikely(!eh->eh_entries))
3780 last_ex = EXT_LAST_EXTENT(eh);
3782 * We should clear the EOFBLOCKS_FL flag if we are writing the
3783 * last block in the last extent in the file. We test this by
3784 * first checking to see if the caller to
3785 * ext4_ext_get_blocks() was interested in the last block (or
3786 * a block beyond the last block) in the current extent. If
3787 * this turns out to be false, we can bail out from this
3788 * function immediately.
3790 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3791 ext4_ext_get_actual_len(last_ex))
3794 * If the caller does appear to be planning to write at or
3795 * beyond the end of the current extent, we then test to see
3796 * if the current extent is the last extent in the file, by
3797 * checking to make sure it was reached via the rightmost node
3798 * at each level of the tree.
3800 for (i = depth-1; i >= 0; i--)
3801 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3804 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3805 return ext4_mark_inode_dirty(handle, inode);
3809 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3811 * Return 1 if there is a delalloc block in the range, otherwise 0.
3813 int ext4_find_delalloc_range(struct inode *inode,
3814 ext4_lblk_t lblk_start,
3815 ext4_lblk_t lblk_end)
3817 struct extent_status es;
3819 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3821 return 0; /* there is no delay extent in this tree */
3822 else if (es.es_lblk <= lblk_start &&
3823 lblk_start < es.es_lblk + es.es_len)
3825 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3831 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3833 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3834 ext4_lblk_t lblk_start, lblk_end;
3835 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3836 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3838 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3842 * Determines how many complete clusters (out of those specified by the 'map')
3843 * are under delalloc and were reserved quota for.
3844 * This function is called when we are writing out the blocks that were
3845 * originally written with their allocation delayed, but then the space was
3846 * allocated using fallocate() before the delayed allocation could be resolved.
3847 * The cases to look for are:
3848 * ('=' indicated delayed allocated blocks
3849 * '-' indicates non-delayed allocated blocks)
3850 * (a) partial clusters towards beginning and/or end outside of allocated range
3851 * are not delalloc'ed.
3853 * |----c---=|====c====|====c====|===-c----|
3854 * |++++++ allocated ++++++|
3855 * ==> 4 complete clusters in above example
3857 * (b) partial cluster (outside of allocated range) towards either end is
3858 * marked for delayed allocation. In this case, we will exclude that
3861 * |----====c========|========c========|
3862 * |++++++ allocated ++++++|
3863 * ==> 1 complete clusters in above example
3866 * |================c================|
3867 * |++++++ allocated ++++++|
3868 * ==> 0 complete clusters in above example
3870 * The ext4_da_update_reserve_space will be called only if we
3871 * determine here that there were some "entire" clusters that span
3872 * this 'allocated' range.
3873 * In the non-bigalloc case, this function will just end up returning num_blks
3874 * without ever calling ext4_find_delalloc_range.
3877 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3878 unsigned int num_blks)
3880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3881 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3882 ext4_lblk_t lblk_from, lblk_to, c_offset;
3883 unsigned int allocated_clusters = 0;
3885 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3886 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3888 /* max possible clusters for this allocation */
3889 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3891 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3893 /* Check towards left side */
3894 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3896 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3897 lblk_to = lblk_from + c_offset - 1;
3899 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3900 allocated_clusters--;
3903 /* Now check towards right. */
3904 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3905 if (allocated_clusters && c_offset) {
3906 lblk_from = lblk_start + num_blks;
3907 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3909 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3910 allocated_clusters--;
3913 return allocated_clusters;
3917 convert_initialized_extent(handle_t *handle, struct inode *inode,
3918 struct ext4_map_blocks *map,
3919 struct ext4_ext_path **ppath, int flags,
3920 unsigned int allocated, ext4_fsblk_t newblock)
3922 struct ext4_ext_path *path = *ppath;
3923 struct ext4_extent *ex;
3924 ext4_lblk_t ee_block;
3925 unsigned int ee_len;
3930 * Make sure that the extent is no bigger than we support with
3933 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3934 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3936 depth = ext_depth(inode);
3937 ex = path[depth].p_ext;
3938 ee_block = le32_to_cpu(ex->ee_block);
3939 ee_len = ext4_ext_get_actual_len(ex);
3941 ext_debug("%s: inode %lu, logical"
3942 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3943 (unsigned long long)ee_block, ee_len);
3945 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3946 err = ext4_split_convert_extents(handle, inode, map, ppath,
3947 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3950 ext4_ext_drop_refs(path);
3951 path = ext4_ext_find_extent(inode, map->m_lblk, ppath, 0);
3953 return PTR_ERR(path);
3954 depth = ext_depth(inode);
3955 ex = path[depth].p_ext;
3957 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3958 (unsigned long) map->m_lblk);
3963 err = ext4_ext_get_access(handle, inode, path + depth);
3966 /* first mark the extent as unwritten */
3967 ext4_ext_mark_unwritten(ex);
3969 /* note: ext4_ext_correct_indexes() isn't needed here because
3970 * borders are not changed
3972 ext4_ext_try_to_merge(handle, inode, path, ex);
3974 /* Mark modified extent as dirty */
3975 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3978 ext4_ext_show_leaf(inode, path);
3980 ext4_update_inode_fsync_trans(handle, inode, 1);
3981 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3984 map->m_flags |= EXT4_MAP_UNWRITTEN;
3985 if (allocated > map->m_len)
3986 allocated = map->m_len;
3987 map->m_len = allocated;
3992 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3993 struct ext4_map_blocks *map,
3994 struct ext4_ext_path **ppath, int flags,
3995 unsigned int allocated, ext4_fsblk_t newblock)
3997 struct ext4_ext_path *path = *ppath;
4000 ext4_io_end_t *io = ext4_inode_aio(inode);
4002 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4003 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4004 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4006 ext4_ext_show_leaf(inode, path);
4009 * When writing into unwritten space, we should not fail to
4010 * allocate metadata blocks for the new extent block if needed.
4012 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4014 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4015 allocated, newblock);
4017 /* get_block() before submit the IO, split the extent */
4018 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4019 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4020 flags | EXT4_GET_BLOCKS_CONVERT);
4024 * Flag the inode(non aio case) or end_io struct (aio case)
4025 * that this IO needs to conversion to written when IO is
4029 ext4_set_io_unwritten_flag(inode, io);
4031 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4032 map->m_flags |= EXT4_MAP_UNWRITTEN;
4035 /* IO end_io complete, convert the filled extent to written */
4036 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4037 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4040 ext4_update_inode_fsync_trans(handle, inode, 1);
4041 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4045 map->m_flags |= EXT4_MAP_MAPPED;
4046 map->m_pblk = newblock;
4047 if (allocated > map->m_len)
4048 allocated = map->m_len;
4049 map->m_len = allocated;
4052 /* buffered IO case */
4054 * repeat fallocate creation request
4055 * we already have an unwritten extent
4057 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4058 map->m_flags |= EXT4_MAP_UNWRITTEN;
4062 /* buffered READ or buffered write_begin() lookup */
4063 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4065 * We have blocks reserved already. We
4066 * return allocated blocks so that delalloc
4067 * won't do block reservation for us. But
4068 * the buffer head will be unmapped so that
4069 * a read from the block returns 0s.
4071 map->m_flags |= EXT4_MAP_UNWRITTEN;
4075 /* buffered write, writepage time, convert*/
4076 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4078 ext4_update_inode_fsync_trans(handle, inode, 1);
4085 map->m_flags |= EXT4_MAP_NEW;
4087 * if we allocated more blocks than requested
4088 * we need to make sure we unmap the extra block
4089 * allocated. The actual needed block will get
4090 * unmapped later when we find the buffer_head marked
4093 if (allocated > map->m_len) {
4094 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4095 newblock + map->m_len,
4096 allocated - map->m_len);
4097 allocated = map->m_len;
4099 map->m_len = allocated;
4102 * If we have done fallocate with the offset that is already
4103 * delayed allocated, we would have block reservation
4104 * and quota reservation done in the delayed write path.
4105 * But fallocate would have already updated quota and block
4106 * count for this offset. So cancel these reservation
4108 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4109 unsigned int reserved_clusters;
4110 reserved_clusters = get_reserved_cluster_alloc(inode,
4111 map->m_lblk, map->m_len);
4112 if (reserved_clusters)
4113 ext4_da_update_reserve_space(inode,
4119 map->m_flags |= EXT4_MAP_MAPPED;
4120 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4121 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4127 if (allocated > map->m_len)
4128 allocated = map->m_len;
4129 ext4_ext_show_leaf(inode, path);
4130 map->m_pblk = newblock;
4131 map->m_len = allocated;
4133 return err ? err : allocated;
4137 * get_implied_cluster_alloc - check to see if the requested
4138 * allocation (in the map structure) overlaps with a cluster already
4139 * allocated in an extent.
4140 * @sb The filesystem superblock structure
4141 * @map The requested lblk->pblk mapping
4142 * @ex The extent structure which might contain an implied
4143 * cluster allocation
4145 * This function is called by ext4_ext_map_blocks() after we failed to
4146 * find blocks that were already in the inode's extent tree. Hence,
4147 * we know that the beginning of the requested region cannot overlap
4148 * the extent from the inode's extent tree. There are three cases we
4149 * want to catch. The first is this case:
4151 * |--- cluster # N--|
4152 * |--- extent ---| |---- requested region ---|
4155 * The second case that we need to test for is this one:
4157 * |--------- cluster # N ----------------|
4158 * |--- requested region --| |------- extent ----|
4159 * |=======================|
4161 * The third case is when the requested region lies between two extents
4162 * within the same cluster:
4163 * |------------- cluster # N-------------|
4164 * |----- ex -----| |---- ex_right ----|
4165 * |------ requested region ------|
4166 * |================|
4168 * In each of the above cases, we need to set the map->m_pblk and
4169 * map->m_len so it corresponds to the return the extent labelled as
4170 * "|====|" from cluster #N, since it is already in use for data in
4171 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4172 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4173 * as a new "allocated" block region. Otherwise, we will return 0 and
4174 * ext4_ext_map_blocks() will then allocate one or more new clusters
4175 * by calling ext4_mb_new_blocks().
4177 static int get_implied_cluster_alloc(struct super_block *sb,
4178 struct ext4_map_blocks *map,
4179 struct ext4_extent *ex,
4180 struct ext4_ext_path *path)
4182 struct ext4_sb_info *sbi = EXT4_SB(sb);
4183 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4184 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4185 ext4_lblk_t rr_cluster_start;
4186 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4187 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4188 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4190 /* The extent passed in that we are trying to match */
4191 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4192 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4194 /* The requested region passed into ext4_map_blocks() */
4195 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4197 if ((rr_cluster_start == ex_cluster_end) ||
4198 (rr_cluster_start == ex_cluster_start)) {
4199 if (rr_cluster_start == ex_cluster_end)
4200 ee_start += ee_len - 1;
4201 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4202 map->m_len = min(map->m_len,
4203 (unsigned) sbi->s_cluster_ratio - c_offset);
4205 * Check for and handle this case:
4207 * |--------- cluster # N-------------|
4208 * |------- extent ----|
4209 * |--- requested region ---|
4213 if (map->m_lblk < ee_block)
4214 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4217 * Check for the case where there is already another allocated
4218 * block to the right of 'ex' but before the end of the cluster.
4220 * |------------- cluster # N-------------|
4221 * |----- ex -----| |---- ex_right ----|
4222 * |------ requested region ------|
4223 * |================|
4225 if (map->m_lblk > ee_block) {
4226 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4227 map->m_len = min(map->m_len, next - map->m_lblk);
4230 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4234 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4240 * Block allocation/map/preallocation routine for extents based files
4243 * Need to be called with
4244 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4245 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4247 * return > 0, number of of blocks already mapped/allocated
4248 * if create == 0 and these are pre-allocated blocks
4249 * buffer head is unmapped
4250 * otherwise blocks are mapped
4252 * return = 0, if plain look up failed (blocks have not been allocated)
4253 * buffer head is unmapped
4255 * return < 0, error case.
4257 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4258 struct ext4_map_blocks *map, int flags)
4260 struct ext4_ext_path *path = NULL;
4261 struct ext4_extent newex, *ex, *ex2;
4262 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4263 ext4_fsblk_t newblock = 0;
4264 int free_on_err = 0, err = 0, depth, ret;
4265 unsigned int allocated = 0, offset = 0;
4266 unsigned int allocated_clusters = 0;
4267 struct ext4_allocation_request ar;
4268 ext4_io_end_t *io = ext4_inode_aio(inode);
4269 ext4_lblk_t cluster_offset;
4270 int set_unwritten = 0;
4272 ext_debug("blocks %u/%u requested for inode %lu\n",
4273 map->m_lblk, map->m_len, inode->i_ino);
4274 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4276 /* find extent for this block */
4277 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4279 err = PTR_ERR(path);
4284 depth = ext_depth(inode);
4287 * consistent leaf must not be empty;
4288 * this situation is possible, though, _during_ tree modification;
4289 * this is why assert can't be put in ext4_ext_find_extent()
4291 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4292 EXT4_ERROR_INODE(inode, "bad extent address "
4293 "lblock: %lu, depth: %d pblock %lld",
4294 (unsigned long) map->m_lblk, depth,
4295 path[depth].p_block);
4300 ex = path[depth].p_ext;
4302 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4303 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4304 unsigned short ee_len;
4308 * unwritten extents are treated as holes, except that
4309 * we split out initialized portions during a write.
4311 ee_len = ext4_ext_get_actual_len(ex);
4313 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4315 /* if found extent covers block, simply return it */
4316 if (in_range(map->m_lblk, ee_block, ee_len)) {
4317 newblock = map->m_lblk - ee_block + ee_start;
4318 /* number of remaining blocks in the extent */
4319 allocated = ee_len - (map->m_lblk - ee_block);
4320 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4321 ee_block, ee_len, newblock);
4324 * If the extent is initialized check whether the
4325 * caller wants to convert it to unwritten.
4327 if ((!ext4_ext_is_unwritten(ex)) &&
4328 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4329 allocated = convert_initialized_extent(
4330 handle, inode, map, &path,
4331 flags, allocated, newblock);
4333 } else if (!ext4_ext_is_unwritten(ex))
4336 ret = ext4_ext_handle_unwritten_extents(
4337 handle, inode, map, &path, flags,
4338 allocated, newblock);
4347 if ((sbi->s_cluster_ratio > 1) &&
4348 ext4_find_delalloc_cluster(inode, map->m_lblk))
4349 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4352 * requested block isn't allocated yet;
4353 * we couldn't try to create block if create flag is zero
4355 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4357 * put just found gap into cache to speed up
4358 * subsequent requests
4360 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4361 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4366 * Okay, we need to do block allocation.
4368 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4369 newex.ee_block = cpu_to_le32(map->m_lblk);
4370 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4373 * If we are doing bigalloc, check to see if the extent returned
4374 * by ext4_ext_find_extent() implies a cluster we can use.
4376 if (cluster_offset && ex &&
4377 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4378 ar.len = allocated = map->m_len;
4379 newblock = map->m_pblk;
4380 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4381 goto got_allocated_blocks;
4384 /* find neighbour allocated blocks */
4385 ar.lleft = map->m_lblk;
4386 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4389 ar.lright = map->m_lblk;
4391 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4395 /* Check if the extent after searching to the right implies a
4396 * cluster we can use. */
4397 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4398 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4399 ar.len = allocated = map->m_len;
4400 newblock = map->m_pblk;
4401 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4402 goto got_allocated_blocks;
4406 * See if request is beyond maximum number of blocks we can have in
4407 * a single extent. For an initialized extent this limit is
4408 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4409 * EXT_UNWRITTEN_MAX_LEN.
4411 if (map->m_len > EXT_INIT_MAX_LEN &&
4412 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4413 map->m_len = EXT_INIT_MAX_LEN;
4414 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4415 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4416 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4418 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4419 newex.ee_len = cpu_to_le16(map->m_len);
4420 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4422 allocated = ext4_ext_get_actual_len(&newex);
4424 allocated = map->m_len;
4426 /* allocate new block */
4428 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4429 ar.logical = map->m_lblk;
4431 * We calculate the offset from the beginning of the cluster
4432 * for the logical block number, since when we allocate a
4433 * physical cluster, the physical block should start at the
4434 * same offset from the beginning of the cluster. This is
4435 * needed so that future calls to get_implied_cluster_alloc()
4438 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4439 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4441 ar.logical -= offset;
4442 if (S_ISREG(inode->i_mode))
4443 ar.flags = EXT4_MB_HINT_DATA;
4445 /* disable in-core preallocation for non-regular files */
4447 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4448 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4449 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4452 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4453 ar.goal, newblock, allocated);
4455 allocated_clusters = ar.len;
4456 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4457 if (ar.len > allocated)
4460 got_allocated_blocks:
4461 /* try to insert new extent into found leaf and return */
4462 ext4_ext_store_pblock(&newex, newblock + offset);
4463 newex.ee_len = cpu_to_le16(ar.len);
4464 /* Mark unwritten */
4465 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4466 ext4_ext_mark_unwritten(&newex);
4467 map->m_flags |= EXT4_MAP_UNWRITTEN;
4469 * io_end structure was created for every IO write to an
4470 * unwritten extent. To avoid unnecessary conversion,
4471 * here we flag the IO that really needs the conversion.
4472 * For non asycn direct IO case, flag the inode state
4473 * that we need to perform conversion when IO is done.
4475 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4480 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4481 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4484 err = ext4_ext_insert_extent(handle, inode, &path,
4487 if (!err && set_unwritten) {
4489 ext4_set_io_unwritten_flag(inode, io);
4491 ext4_set_inode_state(inode,
4492 EXT4_STATE_DIO_UNWRITTEN);
4495 if (err && free_on_err) {
4496 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4497 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4498 /* free data blocks we just allocated */
4499 /* not a good idea to call discard here directly,
4500 * but otherwise we'd need to call it every free() */
4501 ext4_discard_preallocations(inode);
4502 ext4_free_blocks(handle, inode, NULL, newblock,
4503 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4507 /* previous routine could use block we allocated */
4508 newblock = ext4_ext_pblock(&newex);
4509 allocated = ext4_ext_get_actual_len(&newex);
4510 if (allocated > map->m_len)
4511 allocated = map->m_len;
4512 map->m_flags |= EXT4_MAP_NEW;
4515 * Update reserved blocks/metadata blocks after successful
4516 * block allocation which had been deferred till now.
4518 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4519 unsigned int reserved_clusters;
4521 * Check how many clusters we had reserved this allocated range
4523 reserved_clusters = get_reserved_cluster_alloc(inode,
4524 map->m_lblk, allocated);
4525 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4526 if (reserved_clusters) {
4528 * We have clusters reserved for this range.
4529 * But since we are not doing actual allocation
4530 * and are simply using blocks from previously
4531 * allocated cluster, we should release the
4532 * reservation and not claim quota.
4534 ext4_da_update_reserve_space(inode,
4535 reserved_clusters, 0);
4538 BUG_ON(allocated_clusters < reserved_clusters);
4539 if (reserved_clusters < allocated_clusters) {
4540 struct ext4_inode_info *ei = EXT4_I(inode);
4541 int reservation = allocated_clusters -
4544 * It seems we claimed few clusters outside of
4545 * the range of this allocation. We should give
4546 * it back to the reservation pool. This can
4547 * happen in the following case:
4549 * * Suppose s_cluster_ratio is 4 (i.e., each
4550 * cluster has 4 blocks. Thus, the clusters
4551 * are [0-3],[4-7],[8-11]...
4552 * * First comes delayed allocation write for
4553 * logical blocks 10 & 11. Since there were no
4554 * previous delayed allocated blocks in the
4555 * range [8-11], we would reserve 1 cluster
4557 * * Next comes write for logical blocks 3 to 8.
4558 * In this case, we will reserve 2 clusters
4559 * (for [0-3] and [4-7]; and not for [8-11] as
4560 * that range has a delayed allocated blocks.
4561 * Thus total reserved clusters now becomes 3.
4562 * * Now, during the delayed allocation writeout
4563 * time, we will first write blocks [3-8] and
4564 * allocate 3 clusters for writing these
4565 * blocks. Also, we would claim all these
4566 * three clusters above.
4567 * * Now when we come here to writeout the
4568 * blocks [10-11], we would expect to claim
4569 * the reservation of 1 cluster we had made
4570 * (and we would claim it since there are no
4571 * more delayed allocated blocks in the range
4572 * [8-11]. But our reserved cluster count had
4573 * already gone to 0.
4575 * Thus, at the step 4 above when we determine
4576 * that there are still some unwritten delayed
4577 * allocated blocks outside of our current
4578 * block range, we should increment the
4579 * reserved clusters count so that when the
4580 * remaining blocks finally gets written, we
4583 dquot_reserve_block(inode,
4584 EXT4_C2B(sbi, reservation));
4585 spin_lock(&ei->i_block_reservation_lock);
4586 ei->i_reserved_data_blocks += reservation;
4587 spin_unlock(&ei->i_block_reservation_lock);
4590 * We will claim quota for all newly allocated blocks.
4591 * We're updating the reserved space *after* the
4592 * correction above so we do not accidentally free
4593 * all the metadata reservation because we might
4594 * actually need it later on.
4596 ext4_da_update_reserve_space(inode, allocated_clusters,
4602 * Cache the extent and update transaction to commit on fdatasync only
4603 * when it is _not_ an unwritten extent.
4605 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4606 ext4_update_inode_fsync_trans(handle, inode, 1);
4608 ext4_update_inode_fsync_trans(handle, inode, 0);
4610 if (allocated > map->m_len)
4611 allocated = map->m_len;
4612 ext4_ext_show_leaf(inode, path);
4613 map->m_flags |= EXT4_MAP_MAPPED;
4614 map->m_pblk = newblock;
4615 map->m_len = allocated;
4618 ext4_ext_drop_refs(path);
4622 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4623 err ? err : allocated);
4624 ext4_es_lru_add(inode);
4625 return err ? err : allocated;
4628 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4630 struct super_block *sb = inode->i_sb;
4631 ext4_lblk_t last_block;
4635 * TODO: optimization is possible here.
4636 * Probably we need not scan at all,
4637 * because page truncation is enough.
4640 /* we have to know where to truncate from in crash case */
4641 EXT4_I(inode)->i_disksize = inode->i_size;
4642 ext4_mark_inode_dirty(handle, inode);
4644 last_block = (inode->i_size + sb->s_blocksize - 1)
4645 >> EXT4_BLOCK_SIZE_BITS(sb);
4647 err = ext4_es_remove_extent(inode, last_block,
4648 EXT_MAX_BLOCKS - last_block);
4649 if (err == -ENOMEM) {
4651 congestion_wait(BLK_RW_ASYNC, HZ/50);
4655 ext4_std_error(inode->i_sb, err);
4658 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4659 ext4_std_error(inode->i_sb, err);
4662 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4663 ext4_lblk_t len, loff_t new_size,
4664 int flags, int mode)
4666 struct inode *inode = file_inode(file);
4671 struct ext4_map_blocks map;
4672 unsigned int credits;
4675 map.m_lblk = offset;
4678 * Don't normalize the request if it can fit in one extent so
4679 * that it doesn't get unnecessarily split into multiple
4682 if (len <= EXT_UNWRITTEN_MAX_LEN)
4683 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4686 * credits to insert 1 extent into extent tree
4688 credits = ext4_chunk_trans_blocks(inode, len);
4691 while (ret >= 0 && len) {
4692 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4694 if (IS_ERR(handle)) {
4695 ret = PTR_ERR(handle);
4698 ret = ext4_map_blocks(handle, inode, &map, flags);
4700 ext4_debug("inode #%lu: block %u: len %u: "
4701 "ext4_ext_map_blocks returned %d",
4702 inode->i_ino, map.m_lblk,
4704 ext4_mark_inode_dirty(handle, inode);
4705 ret2 = ext4_journal_stop(handle);
4709 map.m_len = len = len - ret;
4710 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4711 inode->i_ctime = ext4_current_time(inode);
4713 if (epos > new_size)
4715 if (ext4_update_inode_size(inode, epos) & 0x1)
4716 inode->i_mtime = inode->i_ctime;
4718 if (epos > inode->i_size)
4719 ext4_set_inode_flag(inode,
4720 EXT4_INODE_EOFBLOCKS);
4722 ext4_mark_inode_dirty(handle, inode);
4723 ret2 = ext4_journal_stop(handle);
4727 if (ret == -ENOSPC &&
4728 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4733 return ret > 0 ? ret2 : ret;
4736 static long ext4_zero_range(struct file *file, loff_t offset,
4737 loff_t len, int mode)
4739 struct inode *inode = file_inode(file);
4740 handle_t *handle = NULL;
4741 unsigned int max_blocks;
4742 loff_t new_size = 0;
4746 int partial_begin, partial_end;
4749 struct address_space *mapping = inode->i_mapping;
4750 unsigned int blkbits = inode->i_blkbits;
4752 trace_ext4_zero_range(inode, offset, len, mode);
4754 if (!S_ISREG(inode->i_mode))
4757 /* Call ext4_force_commit to flush all data in case of data=journal. */
4758 if (ext4_should_journal_data(inode)) {
4759 ret = ext4_force_commit(inode->i_sb);
4765 * Write out all dirty pages to avoid race conditions
4766 * Then release them.
4768 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4769 ret = filemap_write_and_wait_range(mapping, offset,
4776 * Round up offset. This is not fallocate, we neet to zero out
4777 * blocks, so convert interior block aligned part of the range to
4778 * unwritten and possibly manually zero out unaligned parts of the
4781 start = round_up(offset, 1 << blkbits);
4782 end = round_down((offset + len), 1 << blkbits);
4784 if (start < offset || end > offset + len)
4786 partial_begin = offset & ((1 << blkbits) - 1);
4787 partial_end = (offset + len) & ((1 << blkbits) - 1);
4789 lblk = start >> blkbits;
4790 max_blocks = (end >> blkbits);
4791 if (max_blocks < lblk)
4796 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4797 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4799 if (mode & FALLOC_FL_KEEP_SIZE)
4800 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4802 mutex_lock(&inode->i_mutex);
4805 * Indirect files do not support unwritten extnets
4807 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4812 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4813 offset + len > i_size_read(inode)) {
4814 new_size = offset + len;
4815 ret = inode_newsize_ok(inode, new_size);
4819 * If we have a partial block after EOF we have to allocate
4826 if (max_blocks > 0) {
4828 /* Now release the pages and zero block aligned part of pages*/
4829 truncate_pagecache_range(inode, start, end - 1);
4830 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4832 /* Wait all existing dio workers, newcomers will block on i_mutex */
4833 ext4_inode_block_unlocked_dio(inode);
4834 inode_dio_wait(inode);
4836 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4841 * Remove entire range from the extent status tree.
4843 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4844 * NOT sufficient. I'm not sure why this is the case,
4845 * but let's be conservative and remove the extent
4846 * status tree for the entire inode. There should be
4847 * no outstanding delalloc extents thanks to the
4848 * filemap_write_and_wait_range() call above.
4850 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4854 if (!partial_begin && !partial_end)
4858 * In worst case we have to writeout two nonadjacent unwritten
4859 * blocks and update the inode
4861 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4862 if (ext4_should_journal_data(inode))
4864 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4865 if (IS_ERR(handle)) {
4866 ret = PTR_ERR(handle);
4867 ext4_std_error(inode->i_sb, ret);
4871 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4873 ext4_update_inode_size(inode, new_size);
4876 * Mark that we allocate beyond EOF so the subsequent truncate
4877 * can proceed even if the new size is the same as i_size.
4879 if ((offset + len) > i_size_read(inode))
4880 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4882 ext4_mark_inode_dirty(handle, inode);
4884 /* Zero out partial block at the edges of the range */
4885 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4887 if (file->f_flags & O_SYNC)
4888 ext4_handle_sync(handle);
4890 ext4_journal_stop(handle);
4892 ext4_inode_resume_unlocked_dio(inode);
4894 mutex_unlock(&inode->i_mutex);
4899 * preallocate space for a file. This implements ext4's fallocate file
4900 * operation, which gets called from sys_fallocate system call.
4901 * For block-mapped files, posix_fallocate should fall back to the method
4902 * of writing zeroes to the required new blocks (the same behavior which is
4903 * expected for file systems which do not support fallocate() system call).
4905 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4907 struct inode *inode = file_inode(file);
4908 loff_t new_size = 0;
4909 unsigned int max_blocks;
4913 unsigned int blkbits = inode->i_blkbits;
4915 /* Return error if mode is not supported */
4916 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4917 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4920 if (mode & FALLOC_FL_PUNCH_HOLE)
4921 return ext4_punch_hole(inode, offset, len);
4923 ret = ext4_convert_inline_data(inode);
4928 * currently supporting (pre)allocate mode for extent-based
4931 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4934 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4935 return ext4_collapse_range(inode, offset, len);
4937 if (mode & FALLOC_FL_ZERO_RANGE)
4938 return ext4_zero_range(file, offset, len, mode);
4940 trace_ext4_fallocate_enter(inode, offset, len, mode);
4941 lblk = offset >> blkbits;
4943 * We can't just convert len to max_blocks because
4944 * If blocksize = 4096 offset = 3072 and len = 2048
4946 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4949 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4950 if (mode & FALLOC_FL_KEEP_SIZE)
4951 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4953 mutex_lock(&inode->i_mutex);
4955 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4956 offset + len > i_size_read(inode)) {
4957 new_size = offset + len;
4958 ret = inode_newsize_ok(inode, new_size);
4963 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4968 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4969 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4970 EXT4_I(inode)->i_sync_tid);
4973 mutex_unlock(&inode->i_mutex);
4974 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4979 * This function convert a range of blocks to written extents
4980 * The caller of this function will pass the start offset and the size.
4981 * all unwritten extents within this range will be converted to
4984 * This function is called from the direct IO end io call back
4985 * function, to convert the fallocated extents after IO is completed.
4986 * Returns 0 on success.
4988 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4989 loff_t offset, ssize_t len)
4991 unsigned int max_blocks;
4994 struct ext4_map_blocks map;
4995 unsigned int credits, blkbits = inode->i_blkbits;
4997 map.m_lblk = offset >> blkbits;
4999 * We can't just convert len to max_blocks because
5000 * If blocksize = 4096 offset = 3072 and len = 2048
5002 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5005 * This is somewhat ugly but the idea is clear: When transaction is
5006 * reserved, everything goes into it. Otherwise we rather start several
5007 * smaller transactions for conversion of each extent separately.
5010 handle = ext4_journal_start_reserved(handle,
5011 EXT4_HT_EXT_CONVERT);
5013 return PTR_ERR(handle);
5017 * credits to insert 1 extent into extent tree
5019 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5021 while (ret >= 0 && ret < max_blocks) {
5023 map.m_len = (max_blocks -= ret);
5025 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5027 if (IS_ERR(handle)) {
5028 ret = PTR_ERR(handle);
5032 ret = ext4_map_blocks(handle, inode, &map,
5033 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5035 ext4_warning(inode->i_sb,
5036 "inode #%lu: block %u: len %u: "
5037 "ext4_ext_map_blocks returned %d",
5038 inode->i_ino, map.m_lblk,
5040 ext4_mark_inode_dirty(handle, inode);
5042 ret2 = ext4_journal_stop(handle);
5043 if (ret <= 0 || ret2)
5047 ret2 = ext4_journal_stop(handle);
5048 return ret > 0 ? ret2 : ret;
5052 * If newes is not existing extent (newes->ec_pblk equals zero) find
5053 * delayed extent at start of newes and update newes accordingly and
5054 * return start of the next delayed extent.
5056 * If newes is existing extent (newes->ec_pblk is not equal zero)
5057 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5058 * extent found. Leave newes unmodified.
5060 static int ext4_find_delayed_extent(struct inode *inode,
5061 struct extent_status *newes)
5063 struct extent_status es;
5064 ext4_lblk_t block, next_del;
5066 if (newes->es_pblk == 0) {
5067 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5068 newes->es_lblk + newes->es_len - 1, &es);
5071 * No extent in extent-tree contains block @newes->es_pblk,
5072 * then the block may stay in 1)a hole or 2)delayed-extent.
5078 if (es.es_lblk > newes->es_lblk) {
5080 newes->es_len = min(es.es_lblk - newes->es_lblk,
5085 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5088 block = newes->es_lblk + newes->es_len;
5089 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5091 next_del = EXT_MAX_BLOCKS;
5093 next_del = es.es_lblk;
5097 /* fiemap flags we can handle specified here */
5098 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5100 static int ext4_xattr_fiemap(struct inode *inode,
5101 struct fiemap_extent_info *fieinfo)
5105 __u32 flags = FIEMAP_EXTENT_LAST;
5106 int blockbits = inode->i_sb->s_blocksize_bits;
5110 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5111 struct ext4_iloc iloc;
5112 int offset; /* offset of xattr in inode */
5114 error = ext4_get_inode_loc(inode, &iloc);
5117 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5118 offset = EXT4_GOOD_OLD_INODE_SIZE +
5119 EXT4_I(inode)->i_extra_isize;
5121 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5122 flags |= FIEMAP_EXTENT_DATA_INLINE;
5124 } else { /* external block */
5125 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5126 length = inode->i_sb->s_blocksize;
5130 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5132 return (error < 0 ? error : 0);
5135 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5136 __u64 start, __u64 len)
5138 ext4_lblk_t start_blk;
5141 if (ext4_has_inline_data(inode)) {
5144 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5150 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5151 error = ext4_ext_precache(inode);
5156 /* fallback to generic here if not in extents fmt */
5157 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5158 return generic_block_fiemap(inode, fieinfo, start, len,
5161 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5164 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5165 error = ext4_xattr_fiemap(inode, fieinfo);
5167 ext4_lblk_t len_blks;
5170 start_blk = start >> inode->i_sb->s_blocksize_bits;
5171 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5172 if (last_blk >= EXT_MAX_BLOCKS)
5173 last_blk = EXT_MAX_BLOCKS-1;
5174 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5177 * Walk the extent tree gathering extent information
5178 * and pushing extents back to the user.
5180 error = ext4_fill_fiemap_extents(inode, start_blk,
5183 ext4_es_lru_add(inode);
5189 * Function to access the path buffer for marking it dirty.
5190 * It also checks if there are sufficient credits left in the journal handle
5194 ext4_access_path(handle_t *handle, struct inode *inode,
5195 struct ext4_ext_path *path)
5199 if (!ext4_handle_valid(handle))
5203 * Check if need to extend journal credits
5204 * 3 for leaf, sb, and inode plus 2 (bmap and group
5205 * descriptor) for each block group; assume two block
5208 if (handle->h_buffer_credits < 7) {
5209 credits = ext4_writepage_trans_blocks(inode);
5210 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5211 /* EAGAIN is success */
5212 if (err && err != -EAGAIN)
5216 err = ext4_ext_get_access(handle, inode, path);
5221 * ext4_ext_shift_path_extents:
5222 * Shift the extents of a path structure lying between path[depth].p_ext
5223 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5224 * from starting block for each extent.
5227 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5228 struct inode *inode, handle_t *handle,
5232 struct ext4_extent *ex_start, *ex_last;
5234 depth = path->p_depth;
5236 while (depth >= 0) {
5237 if (depth == path->p_depth) {
5238 ex_start = path[depth].p_ext;
5242 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5246 err = ext4_access_path(handle, inode, path + depth);
5250 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5253 *start = le32_to_cpu(ex_last->ee_block) +
5254 ext4_ext_get_actual_len(ex_last);
5256 while (ex_start <= ex_last) {
5257 le32_add_cpu(&ex_start->ee_block, -shift);
5258 /* Try to merge to the left. */
5260 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5261 ext4_ext_try_to_merge_right(inode,
5262 path, ex_start - 1))
5267 err = ext4_ext_dirty(handle, inode, path + depth);
5271 if (--depth < 0 || !update)
5275 /* Update index too */
5276 err = ext4_access_path(handle, inode, path + depth);
5280 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5281 err = ext4_ext_dirty(handle, inode, path + depth);
5285 /* we are done if current index is not a starting index */
5286 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5297 * ext4_ext_shift_extents:
5298 * All the extents which lies in the range from start to the last allocated
5299 * block for the file are shifted downwards by shift blocks.
5300 * On success, 0 is returned, error otherwise.
5303 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5304 ext4_lblk_t start, ext4_lblk_t shift)
5306 struct ext4_ext_path *path;
5308 struct ext4_extent *extent;
5309 ext4_lblk_t stop_block;
5310 ext4_lblk_t ex_start, ex_end;
5312 /* Let path point to the last extent */
5313 path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5315 return PTR_ERR(path);
5317 depth = path->p_depth;
5318 extent = path[depth].p_ext;
5320 ext4_ext_drop_refs(path);
5325 stop_block = le32_to_cpu(extent->ee_block) +
5326 ext4_ext_get_actual_len(extent);
5327 ext4_ext_drop_refs(path);
5330 /* Nothing to shift, if hole is at the end of file */
5331 if (start >= stop_block)
5335 * Don't start shifting extents until we make sure the hole is big
5336 * enough to accomodate the shift.
5338 path = ext4_ext_find_extent(inode, start - 1, NULL, 0);
5340 return PTR_ERR(path);
5341 depth = path->p_depth;
5342 extent = path[depth].p_ext;
5344 ex_start = le32_to_cpu(extent->ee_block);
5345 ex_end = le32_to_cpu(extent->ee_block) +
5346 ext4_ext_get_actual_len(extent);
5351 ext4_ext_drop_refs(path);
5354 if ((start == ex_start && shift > ex_start) ||
5355 (shift > start - ex_end))
5358 /* Its safe to start updating extents */
5359 while (start < stop_block) {
5360 path = ext4_ext_find_extent(inode, start, NULL, 0);
5362 return PTR_ERR(path);
5363 depth = path->p_depth;
5364 extent = path[depth].p_ext;
5366 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5367 (unsigned long) start);
5370 if (start > le32_to_cpu(extent->ee_block)) {
5371 /* Hole, move to the next extent */
5372 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5373 path[depth].p_ext++;
5375 start = ext4_ext_next_allocated_block(path);
5376 ext4_ext_drop_refs(path);
5381 ret = ext4_ext_shift_path_extents(path, shift, inode,
5383 ext4_ext_drop_refs(path);
5393 * ext4_collapse_range:
5394 * This implements the fallocate's collapse range functionality for ext4
5395 * Returns: 0 and non-zero on error.
5397 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5399 struct super_block *sb = inode->i_sb;
5400 ext4_lblk_t punch_start, punch_stop;
5402 unsigned int credits;
5403 loff_t new_size, ioffset;
5406 /* Collapse range works only on fs block size aligned offsets. */
5407 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5408 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5411 if (!S_ISREG(inode->i_mode))
5414 trace_ext4_collapse_range(inode, offset, len);
5416 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5417 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5419 /* Call ext4_force_commit to flush all data in case of data=journal. */
5420 if (ext4_should_journal_data(inode)) {
5421 ret = ext4_force_commit(inode->i_sb);
5427 * Need to round down offset to be aligned with page size boundary
5428 * for page size > block size.
5430 ioffset = round_down(offset, PAGE_SIZE);
5432 /* Write out all dirty pages */
5433 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5438 /* Take mutex lock */
5439 mutex_lock(&inode->i_mutex);
5442 * There is no need to overlap collapse range with EOF, in which case
5443 * it is effectively a truncate operation
5445 if (offset + len >= i_size_read(inode)) {
5450 /* Currently just for extent based files */
5451 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5456 truncate_pagecache(inode, ioffset);
5458 /* Wait for existing dio to complete */
5459 ext4_inode_block_unlocked_dio(inode);
5460 inode_dio_wait(inode);
5462 credits = ext4_writepage_trans_blocks(inode);
5463 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5464 if (IS_ERR(handle)) {
5465 ret = PTR_ERR(handle);
5469 down_write(&EXT4_I(inode)->i_data_sem);
5470 ext4_discard_preallocations(inode);
5472 ret = ext4_es_remove_extent(inode, punch_start,
5473 EXT_MAX_BLOCKS - punch_start);
5475 up_write(&EXT4_I(inode)->i_data_sem);
5479 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5481 up_write(&EXT4_I(inode)->i_data_sem);
5484 ext4_discard_preallocations(inode);
5486 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5487 punch_stop - punch_start);
5489 up_write(&EXT4_I(inode)->i_data_sem);
5493 new_size = i_size_read(inode) - len;
5494 i_size_write(inode, new_size);
5495 EXT4_I(inode)->i_disksize = new_size;
5497 up_write(&EXT4_I(inode)->i_data_sem);
5499 ext4_handle_sync(handle);
5500 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5501 ext4_mark_inode_dirty(handle, inode);
5504 ext4_journal_stop(handle);
5506 ext4_inode_resume_unlocked_dio(inode);
5508 mutex_unlock(&inode->i_mutex);
5513 * ext4_swap_extents - Swap extents between two inodes
5515 * @inode1: First inode
5516 * @inode2: Second inode
5517 * @lblk1: Start block for first inode
5518 * @lblk2: Start block for second inode
5519 * @count: Number of blocks to swap
5520 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5521 * @erp: Pointer to save error value
5523 * This helper routine does exactly what is promise "swap extents". All other
5524 * stuff such as page-cache locking consistency, bh mapping consistency or
5525 * extent's data copying must be performed by caller.
5527 * i_mutex is held for both inodes
5528 * i_data_sem is locked for write for both inodes
5530 * All pages from requested range are locked for both inodes
5533 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5534 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5535 ext4_lblk_t count, int unwritten, int *erp)
5537 struct ext4_ext_path *path1 = NULL;
5538 struct ext4_ext_path *path2 = NULL;
5539 int replaced_count = 0;
5541 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5542 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5543 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5544 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5546 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5549 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5554 struct ext4_extent *ex1, *ex2, tmp_ex;
5555 ext4_lblk_t e1_blk, e2_blk;
5556 int e1_len, e2_len, len;
5559 path1 = ext4_ext_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5560 if (unlikely(IS_ERR(path1))) {
5561 *erp = PTR_ERR(path1);
5567 path2 = ext4_ext_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5568 if (unlikely(IS_ERR(path2))) {
5569 *erp = PTR_ERR(path2);
5573 ex1 = path1[path1->p_depth].p_ext;
5574 ex2 = path2[path2->p_depth].p_ext;
5575 /* Do we have somthing to swap ? */
5576 if (unlikely(!ex2 || !ex1))
5579 e1_blk = le32_to_cpu(ex1->ee_block);
5580 e2_blk = le32_to_cpu(ex2->ee_block);
5581 e1_len = ext4_ext_get_actual_len(ex1);
5582 e2_len = ext4_ext_get_actual_len(ex2);
5585 if (!in_range(lblk1, e1_blk, e1_len) ||
5586 !in_range(lblk2, e2_blk, e2_len)) {
5587 ext4_lblk_t next1, next2;
5589 /* if hole after extent, then go to next extent */
5590 next1 = ext4_ext_next_allocated_block(path1);
5591 next2 = ext4_ext_next_allocated_block(path2);
5592 /* If hole before extent, then shift to that extent */
5597 /* Do we have something to swap */
5598 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5600 /* Move to the rightest boundary */
5601 len = next1 - lblk1;
5602 if (len < next2 - lblk2)
5603 len = next2 - lblk2;
5612 /* Prepare left boundary */
5613 if (e1_blk < lblk1) {
5615 *erp = ext4_force_split_extent_at(handle, inode1,
5620 if (e2_blk < lblk2) {
5622 *erp = ext4_force_split_extent_at(handle, inode2,
5627 /* ext4_split_extent_at() may result in leaf extent split,
5628 * path must to be revalidated. */
5632 /* Prepare right boundary */
5634 if (len > e1_blk + e1_len - lblk1)
5635 len = e1_blk + e1_len - lblk1;
5636 if (len > e2_blk + e2_len - lblk2)
5637 len = e2_blk + e2_len - lblk2;
5639 if (len != e1_len) {
5641 *erp = ext4_force_split_extent_at(handle, inode1,
5642 &path1, lblk1 + len, 0);
5646 if (len != e2_len) {
5648 *erp = ext4_force_split_extent_at(handle, inode2,
5649 &path2, lblk2 + len, 0);
5653 /* ext4_split_extent_at() may result in leaf extent split,
5654 * path must to be revalidated. */
5658 BUG_ON(e2_len != e1_len);
5659 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5662 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5666 /* Both extents are fully inside boundaries. Swap it now */
5668 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5669 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5670 ex1->ee_len = cpu_to_le16(e2_len);
5671 ex2->ee_len = cpu_to_le16(e1_len);
5673 ext4_ext_mark_unwritten(ex2);
5674 if (ext4_ext_is_unwritten(&tmp_ex))
5675 ext4_ext_mark_unwritten(ex1);
5677 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5678 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5679 *erp = ext4_ext_dirty(handle, inode2, path2 +
5683 *erp = ext4_ext_dirty(handle, inode1, path1 +
5686 * Looks scarry ah..? second inode already points to new blocks,
5687 * and it was successfully dirtied. But luckily error may happen
5688 * only due to journal error, so full transaction will be
5695 replaced_count += len;
5700 ext4_ext_drop_refs(path1);
5705 ext4_ext_drop_refs(path2);
5710 return replaced_count;