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)
716 depth = path->p_depth;
717 for (i = 0; i <= depth; i++, path++)
725 * ext4_ext_binsearch_idx:
726 * binary search for the closest index of the given block
727 * the header must be checked before calling this
730 ext4_ext_binsearch_idx(struct inode *inode,
731 struct ext4_ext_path *path, ext4_lblk_t block)
733 struct ext4_extent_header *eh = path->p_hdr;
734 struct ext4_extent_idx *r, *l, *m;
737 ext_debug("binsearch for %u(idx): ", block);
739 l = EXT_FIRST_INDEX(eh) + 1;
740 r = EXT_LAST_INDEX(eh);
743 if (block < le32_to_cpu(m->ei_block))
747 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
748 m, le32_to_cpu(m->ei_block),
749 r, le32_to_cpu(r->ei_block));
753 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
754 ext4_idx_pblock(path->p_idx));
756 #ifdef CHECK_BINSEARCH
758 struct ext4_extent_idx *chix, *ix;
761 chix = ix = EXT_FIRST_INDEX(eh);
762 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
764 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
765 printk(KERN_DEBUG "k=%d, ix=0x%p, "
767 ix, EXT_FIRST_INDEX(eh));
768 printk(KERN_DEBUG "%u <= %u\n",
769 le32_to_cpu(ix->ei_block),
770 le32_to_cpu(ix[-1].ei_block));
772 BUG_ON(k && le32_to_cpu(ix->ei_block)
773 <= le32_to_cpu(ix[-1].ei_block));
774 if (block < le32_to_cpu(ix->ei_block))
778 BUG_ON(chix != path->p_idx);
785 * ext4_ext_binsearch:
786 * binary search for closest extent of the given block
787 * the header must be checked before calling this
790 ext4_ext_binsearch(struct inode *inode,
791 struct ext4_ext_path *path, ext4_lblk_t block)
793 struct ext4_extent_header *eh = path->p_hdr;
794 struct ext4_extent *r, *l, *m;
796 if (eh->eh_entries == 0) {
798 * this leaf is empty:
799 * we get such a leaf in split/add case
804 ext_debug("binsearch for %u: ", block);
806 l = EXT_FIRST_EXTENT(eh) + 1;
807 r = EXT_LAST_EXTENT(eh);
811 if (block < le32_to_cpu(m->ee_block))
815 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
816 m, le32_to_cpu(m->ee_block),
817 r, le32_to_cpu(r->ee_block));
821 ext_debug(" -> %d:%llu:[%d]%d ",
822 le32_to_cpu(path->p_ext->ee_block),
823 ext4_ext_pblock(path->p_ext),
824 ext4_ext_is_unwritten(path->p_ext),
825 ext4_ext_get_actual_len(path->p_ext));
827 #ifdef CHECK_BINSEARCH
829 struct ext4_extent *chex, *ex;
832 chex = ex = EXT_FIRST_EXTENT(eh);
833 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
834 BUG_ON(k && le32_to_cpu(ex->ee_block)
835 <= le32_to_cpu(ex[-1].ee_block));
836 if (block < le32_to_cpu(ex->ee_block))
840 BUG_ON(chex != path->p_ext);
846 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
848 struct ext4_extent_header *eh;
850 eh = ext_inode_hdr(inode);
853 eh->eh_magic = EXT4_EXT_MAGIC;
854 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
855 ext4_mark_inode_dirty(handle, inode);
859 struct ext4_ext_path *
860 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
861 struct ext4_ext_path **orig_path, int flags)
863 struct ext4_extent_header *eh;
864 struct buffer_head *bh;
865 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
866 short int depth, i, ppos = 0;
869 eh = ext_inode_hdr(inode);
870 depth = ext_depth(inode);
873 ext4_ext_drop_refs(path);
874 if (depth > path[0].p_maxdepth) {
876 *orig_path = path = NULL;
880 /* account possible depth increase */
881 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
884 return ERR_PTR(-ENOMEM);
885 path[0].p_maxdepth = depth + 1;
891 /* walk through the tree */
893 ext_debug("depth %d: num %d, max %d\n",
894 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
896 ext4_ext_binsearch_idx(inode, path + ppos, block);
897 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
898 path[ppos].p_depth = i;
899 path[ppos].p_ext = NULL;
901 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
903 if (unlikely(IS_ERR(bh))) {
908 eh = ext_block_hdr(bh);
910 if (unlikely(ppos > depth)) {
912 EXT4_ERROR_INODE(inode,
913 "ppos %d > depth %d", ppos, depth);
917 path[ppos].p_bh = bh;
918 path[ppos].p_hdr = eh;
921 path[ppos].p_depth = i;
922 path[ppos].p_ext = NULL;
923 path[ppos].p_idx = NULL;
926 ext4_ext_binsearch(inode, path + ppos, block);
927 /* if not an empty leaf */
928 if (path[ppos].p_ext)
929 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
931 ext4_ext_show_path(inode, path);
936 ext4_ext_drop_refs(path);
944 * ext4_ext_insert_index:
945 * insert new index [@logical;@ptr] into the block at @curp;
946 * check where to insert: before @curp or after @curp
948 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
949 struct ext4_ext_path *curp,
950 int logical, ext4_fsblk_t ptr)
952 struct ext4_extent_idx *ix;
955 err = ext4_ext_get_access(handle, inode, curp);
959 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
960 EXT4_ERROR_INODE(inode,
961 "logical %d == ei_block %d!",
962 logical, le32_to_cpu(curp->p_idx->ei_block));
966 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
967 >= le16_to_cpu(curp->p_hdr->eh_max))) {
968 EXT4_ERROR_INODE(inode,
969 "eh_entries %d >= eh_max %d!",
970 le16_to_cpu(curp->p_hdr->eh_entries),
971 le16_to_cpu(curp->p_hdr->eh_max));
975 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
977 ext_debug("insert new index %d after: %llu\n", logical, ptr);
978 ix = curp->p_idx + 1;
981 ext_debug("insert new index %d before: %llu\n", logical, ptr);
985 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
988 ext_debug("insert new index %d: "
989 "move %d indices from 0x%p to 0x%p\n",
990 logical, len, ix, ix + 1);
991 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
994 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
995 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
999 ix->ei_block = cpu_to_le32(logical);
1000 ext4_idx_store_pblock(ix, ptr);
1001 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1003 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1004 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1008 err = ext4_ext_dirty(handle, inode, curp);
1009 ext4_std_error(inode->i_sb, err);
1016 * inserts new subtree into the path, using free index entry
1018 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1019 * - makes decision where to split
1020 * - moves remaining extents and index entries (right to the split point)
1021 * into the newly allocated blocks
1022 * - initializes subtree
1024 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1026 struct ext4_ext_path *path,
1027 struct ext4_extent *newext, int at)
1029 struct buffer_head *bh = NULL;
1030 int depth = ext_depth(inode);
1031 struct ext4_extent_header *neh;
1032 struct ext4_extent_idx *fidx;
1033 int i = at, k, m, a;
1034 ext4_fsblk_t newblock, oldblock;
1036 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1039 /* make decision: where to split? */
1040 /* FIXME: now decision is simplest: at current extent */
1042 /* if current leaf will be split, then we should use
1043 * border from split point */
1044 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1045 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1048 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1049 border = path[depth].p_ext[1].ee_block;
1050 ext_debug("leaf will be split."
1051 " next leaf starts at %d\n",
1052 le32_to_cpu(border));
1054 border = newext->ee_block;
1055 ext_debug("leaf will be added."
1056 " next leaf starts at %d\n",
1057 le32_to_cpu(border));
1061 * If error occurs, then we break processing
1062 * and mark filesystem read-only. index won't
1063 * be inserted and tree will be in consistent
1064 * state. Next mount will repair buffers too.
1068 * Get array to track all allocated blocks.
1069 * We need this to handle errors and free blocks
1072 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1076 /* allocate all needed blocks */
1077 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1078 for (a = 0; a < depth - at; a++) {
1079 newblock = ext4_ext_new_meta_block(handle, inode, path,
1080 newext, &err, flags);
1083 ablocks[a] = newblock;
1086 /* initialize new leaf */
1087 newblock = ablocks[--a];
1088 if (unlikely(newblock == 0)) {
1089 EXT4_ERROR_INODE(inode, "newblock == 0!");
1093 bh = sb_getblk(inode->i_sb, newblock);
1094 if (unlikely(!bh)) {
1100 err = ext4_journal_get_create_access(handle, bh);
1104 neh = ext_block_hdr(bh);
1105 neh->eh_entries = 0;
1106 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1107 neh->eh_magic = EXT4_EXT_MAGIC;
1110 /* move remainder of path[depth] to the new leaf */
1111 if (unlikely(path[depth].p_hdr->eh_entries !=
1112 path[depth].p_hdr->eh_max)) {
1113 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1114 path[depth].p_hdr->eh_entries,
1115 path[depth].p_hdr->eh_max);
1119 /* start copy from next extent */
1120 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1121 ext4_ext_show_move(inode, path, newblock, depth);
1123 struct ext4_extent *ex;
1124 ex = EXT_FIRST_EXTENT(neh);
1125 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1126 le16_add_cpu(&neh->eh_entries, m);
1129 ext4_extent_block_csum_set(inode, neh);
1130 set_buffer_uptodate(bh);
1133 err = ext4_handle_dirty_metadata(handle, inode, bh);
1139 /* correct old leaf */
1141 err = ext4_ext_get_access(handle, inode, path + depth);
1144 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1145 err = ext4_ext_dirty(handle, inode, path + depth);
1151 /* create intermediate indexes */
1153 if (unlikely(k < 0)) {
1154 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1159 ext_debug("create %d intermediate indices\n", k);
1160 /* insert new index into current index block */
1161 /* current depth stored in i var */
1164 oldblock = newblock;
1165 newblock = ablocks[--a];
1166 bh = sb_getblk(inode->i_sb, newblock);
1167 if (unlikely(!bh)) {
1173 err = ext4_journal_get_create_access(handle, bh);
1177 neh = ext_block_hdr(bh);
1178 neh->eh_entries = cpu_to_le16(1);
1179 neh->eh_magic = EXT4_EXT_MAGIC;
1180 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1181 neh->eh_depth = cpu_to_le16(depth - i);
1182 fidx = EXT_FIRST_INDEX(neh);
1183 fidx->ei_block = border;
1184 ext4_idx_store_pblock(fidx, oldblock);
1186 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1187 i, newblock, le32_to_cpu(border), oldblock);
1189 /* move remainder of path[i] to the new index block */
1190 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1191 EXT_LAST_INDEX(path[i].p_hdr))) {
1192 EXT4_ERROR_INODE(inode,
1193 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1194 le32_to_cpu(path[i].p_ext->ee_block));
1198 /* start copy indexes */
1199 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1200 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1201 EXT_MAX_INDEX(path[i].p_hdr));
1202 ext4_ext_show_move(inode, path, newblock, i);
1204 memmove(++fidx, path[i].p_idx,
1205 sizeof(struct ext4_extent_idx) * m);
1206 le16_add_cpu(&neh->eh_entries, m);
1208 ext4_extent_block_csum_set(inode, neh);
1209 set_buffer_uptodate(bh);
1212 err = ext4_handle_dirty_metadata(handle, inode, bh);
1218 /* correct old index */
1220 err = ext4_ext_get_access(handle, inode, path + i);
1223 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1224 err = ext4_ext_dirty(handle, inode, path + i);
1232 /* insert new index */
1233 err = ext4_ext_insert_index(handle, inode, path + at,
1234 le32_to_cpu(border), newblock);
1238 if (buffer_locked(bh))
1244 /* free all allocated blocks in error case */
1245 for (i = 0; i < depth; i++) {
1248 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1249 EXT4_FREE_BLOCKS_METADATA);
1258 * ext4_ext_grow_indepth:
1259 * implements tree growing procedure:
1260 * - allocates new block
1261 * - moves top-level data (index block or leaf) into the new block
1262 * - initializes new top-level, creating index that points to the
1263 * just created block
1265 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1268 struct ext4_extent_header *neh;
1269 struct buffer_head *bh;
1270 ext4_fsblk_t newblock, goal = 0;
1271 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1274 /* Try to prepend new index to old one */
1275 if (ext_depth(inode))
1276 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1277 if (goal > le32_to_cpu(es->s_first_data_block)) {
1278 flags |= EXT4_MB_HINT_TRY_GOAL;
1281 goal = ext4_inode_to_goal_block(inode);
1282 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1287 bh = sb_getblk(inode->i_sb, newblock);
1292 err = ext4_journal_get_create_access(handle, bh);
1298 /* move top-level index/leaf into new block */
1299 memmove(bh->b_data, EXT4_I(inode)->i_data,
1300 sizeof(EXT4_I(inode)->i_data));
1302 /* set size of new block */
1303 neh = ext_block_hdr(bh);
1304 /* old root could have indexes or leaves
1305 * so calculate e_max right way */
1306 if (ext_depth(inode))
1307 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1309 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1310 neh->eh_magic = EXT4_EXT_MAGIC;
1311 ext4_extent_block_csum_set(inode, neh);
1312 set_buffer_uptodate(bh);
1315 err = ext4_handle_dirty_metadata(handle, inode, bh);
1319 /* Update top-level index: num,max,pointer */
1320 neh = ext_inode_hdr(inode);
1321 neh->eh_entries = cpu_to_le16(1);
1322 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1323 if (neh->eh_depth == 0) {
1324 /* Root extent block becomes index block */
1325 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1326 EXT_FIRST_INDEX(neh)->ei_block =
1327 EXT_FIRST_EXTENT(neh)->ee_block;
1329 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1330 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1331 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1332 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1334 le16_add_cpu(&neh->eh_depth, 1);
1335 ext4_mark_inode_dirty(handle, inode);
1343 * ext4_ext_create_new_leaf:
1344 * finds empty index and adds new leaf.
1345 * if no free index is found, then it requests in-depth growing.
1347 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1348 unsigned int mb_flags,
1349 unsigned int gb_flags,
1350 struct ext4_ext_path **ppath,
1351 struct ext4_extent *newext)
1353 struct ext4_ext_path *path = *ppath;
1354 struct ext4_ext_path *curp;
1355 int depth, i, err = 0;
1358 i = depth = ext_depth(inode);
1360 /* walk up to the tree and look for free index entry */
1361 curp = path + depth;
1362 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1367 /* we use already allocated block for index block,
1368 * so subsequent data blocks should be contiguous */
1369 if (EXT_HAS_FREE_INDEX(curp)) {
1370 /* if we found index with free entry, then use that
1371 * entry: create all needed subtree and add new leaf */
1372 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1377 path = ext4_find_extent(inode,
1378 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1381 err = PTR_ERR(path);
1383 /* tree is full, time to grow in depth */
1384 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1389 path = ext4_find_extent(inode,
1390 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1393 err = PTR_ERR(path);
1398 * only first (depth 0 -> 1) produces free space;
1399 * in all other cases we have to split the grown tree
1401 depth = ext_depth(inode);
1402 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1403 /* now we need to split */
1413 * search the closest allocated block to the left for *logical
1414 * and returns it at @logical + it's physical address at @phys
1415 * if *logical is the smallest allocated block, the function
1416 * returns 0 at @phys
1417 * return value contains 0 (success) or error code
1419 static int ext4_ext_search_left(struct inode *inode,
1420 struct ext4_ext_path *path,
1421 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1423 struct ext4_extent_idx *ix;
1424 struct ext4_extent *ex;
1427 if (unlikely(path == NULL)) {
1428 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1431 depth = path->p_depth;
1434 if (depth == 0 && path->p_ext == NULL)
1437 /* usually extent in the path covers blocks smaller
1438 * then *logical, but it can be that extent is the
1439 * first one in the file */
1441 ex = path[depth].p_ext;
1442 ee_len = ext4_ext_get_actual_len(ex);
1443 if (*logical < le32_to_cpu(ex->ee_block)) {
1444 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1445 EXT4_ERROR_INODE(inode,
1446 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1447 *logical, le32_to_cpu(ex->ee_block));
1450 while (--depth >= 0) {
1451 ix = path[depth].p_idx;
1452 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1453 EXT4_ERROR_INODE(inode,
1454 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1455 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1456 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1457 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1465 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1466 EXT4_ERROR_INODE(inode,
1467 "logical %d < ee_block %d + ee_len %d!",
1468 *logical, le32_to_cpu(ex->ee_block), ee_len);
1472 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1473 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1478 * search the closest allocated block to the right for *logical
1479 * and returns it at @logical + it's physical address at @phys
1480 * if *logical is the largest allocated block, the function
1481 * returns 0 at @phys
1482 * return value contains 0 (success) or error code
1484 static int ext4_ext_search_right(struct inode *inode,
1485 struct ext4_ext_path *path,
1486 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1487 struct ext4_extent **ret_ex)
1489 struct buffer_head *bh = NULL;
1490 struct ext4_extent_header *eh;
1491 struct ext4_extent_idx *ix;
1492 struct ext4_extent *ex;
1494 int depth; /* Note, NOT eh_depth; depth from top of tree */
1497 if (unlikely(path == NULL)) {
1498 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1501 depth = path->p_depth;
1504 if (depth == 0 && path->p_ext == NULL)
1507 /* usually extent in the path covers blocks smaller
1508 * then *logical, but it can be that extent is the
1509 * first one in the file */
1511 ex = path[depth].p_ext;
1512 ee_len = ext4_ext_get_actual_len(ex);
1513 if (*logical < le32_to_cpu(ex->ee_block)) {
1514 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1515 EXT4_ERROR_INODE(inode,
1516 "first_extent(path[%d].p_hdr) != ex",
1520 while (--depth >= 0) {
1521 ix = path[depth].p_idx;
1522 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1523 EXT4_ERROR_INODE(inode,
1524 "ix != EXT_FIRST_INDEX *logical %d!",
1532 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1533 EXT4_ERROR_INODE(inode,
1534 "logical %d < ee_block %d + ee_len %d!",
1535 *logical, le32_to_cpu(ex->ee_block), ee_len);
1539 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1540 /* next allocated block in this leaf */
1545 /* go up and search for index to the right */
1546 while (--depth >= 0) {
1547 ix = path[depth].p_idx;
1548 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1552 /* we've gone up to the root and found no index to the right */
1556 /* we've found index to the right, let's
1557 * follow it and find the closest allocated
1558 * block to the right */
1560 block = ext4_idx_pblock(ix);
1561 while (++depth < path->p_depth) {
1562 /* subtract from p_depth to get proper eh_depth */
1563 bh = read_extent_tree_block(inode, block,
1564 path->p_depth - depth, 0);
1567 eh = ext_block_hdr(bh);
1568 ix = EXT_FIRST_INDEX(eh);
1569 block = ext4_idx_pblock(ix);
1573 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1576 eh = ext_block_hdr(bh);
1577 ex = EXT_FIRST_EXTENT(eh);
1579 *logical = le32_to_cpu(ex->ee_block);
1580 *phys = ext4_ext_pblock(ex);
1588 * ext4_ext_next_allocated_block:
1589 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1590 * NOTE: it considers block number from index entry as
1591 * allocated block. Thus, index entries have to be consistent
1595 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1599 BUG_ON(path == NULL);
1600 depth = path->p_depth;
1602 if (depth == 0 && path->p_ext == NULL)
1603 return EXT_MAX_BLOCKS;
1605 while (depth >= 0) {
1606 if (depth == path->p_depth) {
1608 if (path[depth].p_ext &&
1609 path[depth].p_ext !=
1610 EXT_LAST_EXTENT(path[depth].p_hdr))
1611 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1614 if (path[depth].p_idx !=
1615 EXT_LAST_INDEX(path[depth].p_hdr))
1616 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1621 return EXT_MAX_BLOCKS;
1625 * ext4_ext_next_leaf_block:
1626 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1628 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1632 BUG_ON(path == NULL);
1633 depth = path->p_depth;
1635 /* zero-tree has no leaf blocks at all */
1637 return EXT_MAX_BLOCKS;
1639 /* go to index block */
1642 while (depth >= 0) {
1643 if (path[depth].p_idx !=
1644 EXT_LAST_INDEX(path[depth].p_hdr))
1645 return (ext4_lblk_t)
1646 le32_to_cpu(path[depth].p_idx[1].ei_block);
1650 return EXT_MAX_BLOCKS;
1654 * ext4_ext_correct_indexes:
1655 * if leaf gets modified and modified extent is first in the leaf,
1656 * then we have to correct all indexes above.
1657 * TODO: do we need to correct tree in all cases?
1659 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1660 struct ext4_ext_path *path)
1662 struct ext4_extent_header *eh;
1663 int depth = ext_depth(inode);
1664 struct ext4_extent *ex;
1668 eh = path[depth].p_hdr;
1669 ex = path[depth].p_ext;
1671 if (unlikely(ex == NULL || eh == NULL)) {
1672 EXT4_ERROR_INODE(inode,
1673 "ex %p == NULL or eh %p == NULL", ex, eh);
1678 /* there is no tree at all */
1682 if (ex != EXT_FIRST_EXTENT(eh)) {
1683 /* we correct tree if first leaf got modified only */
1688 * TODO: we need correction if border is smaller than current one
1691 border = path[depth].p_ext->ee_block;
1692 err = ext4_ext_get_access(handle, inode, path + k);
1695 path[k].p_idx->ei_block = border;
1696 err = ext4_ext_dirty(handle, inode, path + k);
1701 /* change all left-side indexes */
1702 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1704 err = ext4_ext_get_access(handle, inode, path + k);
1707 path[k].p_idx->ei_block = border;
1708 err = ext4_ext_dirty(handle, inode, path + k);
1717 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1718 struct ext4_extent *ex2)
1720 unsigned short ext1_ee_len, ext2_ee_len;
1723 * Make sure that both extents are initialized. We don't merge
1724 * unwritten extents so that we can be sure that end_io code has
1725 * the extent that was written properly split out and conversion to
1726 * initialized is trivial.
1728 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1731 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1732 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1734 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1735 le32_to_cpu(ex2->ee_block))
1739 * To allow future support for preallocated extents to be added
1740 * as an RO_COMPAT feature, refuse to merge to extents if
1741 * this can result in the top bit of ee_len being set.
1743 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1745 if (ext4_ext_is_unwritten(ex1) &&
1746 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1747 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1748 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1750 #ifdef AGGRESSIVE_TEST
1751 if (ext1_ee_len >= 4)
1755 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1761 * This function tries to merge the "ex" extent to the next extent in the tree.
1762 * It always tries to merge towards right. If you want to merge towards
1763 * left, pass "ex - 1" as argument instead of "ex".
1764 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1765 * 1 if they got merged.
1767 static int ext4_ext_try_to_merge_right(struct inode *inode,
1768 struct ext4_ext_path *path,
1769 struct ext4_extent *ex)
1771 struct ext4_extent_header *eh;
1772 unsigned int depth, len;
1773 int merge_done = 0, unwritten;
1775 depth = ext_depth(inode);
1776 BUG_ON(path[depth].p_hdr == NULL);
1777 eh = path[depth].p_hdr;
1779 while (ex < EXT_LAST_EXTENT(eh)) {
1780 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1782 /* merge with next extent! */
1783 unwritten = ext4_ext_is_unwritten(ex);
1784 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1785 + ext4_ext_get_actual_len(ex + 1));
1787 ext4_ext_mark_unwritten(ex);
1789 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1790 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1791 * sizeof(struct ext4_extent);
1792 memmove(ex + 1, ex + 2, len);
1794 le16_add_cpu(&eh->eh_entries, -1);
1796 WARN_ON(eh->eh_entries == 0);
1797 if (!eh->eh_entries)
1798 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1805 * This function does a very simple check to see if we can collapse
1806 * an extent tree with a single extent tree leaf block into the inode.
1808 static void ext4_ext_try_to_merge_up(handle_t *handle,
1809 struct inode *inode,
1810 struct ext4_ext_path *path)
1813 unsigned max_root = ext4_ext_space_root(inode, 0);
1816 if ((path[0].p_depth != 1) ||
1817 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1818 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1822 * We need to modify the block allocation bitmap and the block
1823 * group descriptor to release the extent tree block. If we
1824 * can't get the journal credits, give up.
1826 if (ext4_journal_extend(handle, 2))
1830 * Copy the extent data up to the inode
1832 blk = ext4_idx_pblock(path[0].p_idx);
1833 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1834 sizeof(struct ext4_extent_idx);
1835 s += sizeof(struct ext4_extent_header);
1837 path[1].p_maxdepth = path[0].p_maxdepth;
1838 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1839 path[0].p_depth = 0;
1840 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1841 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1842 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1844 brelse(path[1].p_bh);
1845 ext4_free_blocks(handle, inode, NULL, blk, 1,
1846 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1850 * This function tries to merge the @ex extent to neighbours in the tree.
1851 * return 1 if merge left else 0.
1853 static void ext4_ext_try_to_merge(handle_t *handle,
1854 struct inode *inode,
1855 struct ext4_ext_path *path,
1856 struct ext4_extent *ex) {
1857 struct ext4_extent_header *eh;
1861 depth = ext_depth(inode);
1862 BUG_ON(path[depth].p_hdr == NULL);
1863 eh = path[depth].p_hdr;
1865 if (ex > EXT_FIRST_EXTENT(eh))
1866 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1869 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1871 ext4_ext_try_to_merge_up(handle, inode, path);
1875 * check if a portion of the "newext" extent overlaps with an
1878 * If there is an overlap discovered, it updates the length of the newext
1879 * such that there will be no overlap, and then returns 1.
1880 * If there is no overlap found, it returns 0.
1882 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1883 struct inode *inode,
1884 struct ext4_extent *newext,
1885 struct ext4_ext_path *path)
1888 unsigned int depth, len1;
1889 unsigned int ret = 0;
1891 b1 = le32_to_cpu(newext->ee_block);
1892 len1 = ext4_ext_get_actual_len(newext);
1893 depth = ext_depth(inode);
1894 if (!path[depth].p_ext)
1896 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1899 * get the next allocated block if the extent in the path
1900 * is before the requested block(s)
1903 b2 = ext4_ext_next_allocated_block(path);
1904 if (b2 == EXT_MAX_BLOCKS)
1906 b2 = EXT4_LBLK_CMASK(sbi, b2);
1909 /* check for wrap through zero on extent logical start block*/
1910 if (b1 + len1 < b1) {
1911 len1 = EXT_MAX_BLOCKS - b1;
1912 newext->ee_len = cpu_to_le16(len1);
1916 /* check for overlap */
1917 if (b1 + len1 > b2) {
1918 newext->ee_len = cpu_to_le16(b2 - b1);
1926 * ext4_ext_insert_extent:
1927 * tries to merge requsted extent into the existing extent or
1928 * inserts requested extent as new one into the tree,
1929 * creating new leaf in the no-space case.
1931 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1932 struct ext4_ext_path **ppath,
1933 struct ext4_extent *newext, int gb_flags)
1935 struct ext4_ext_path *path = *ppath;
1936 struct ext4_extent_header *eh;
1937 struct ext4_extent *ex, *fex;
1938 struct ext4_extent *nearex; /* nearest extent */
1939 struct ext4_ext_path *npath = NULL;
1940 int depth, len, err;
1942 int mb_flags = 0, unwritten;
1944 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1945 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1946 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1947 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1950 depth = ext_depth(inode);
1951 ex = path[depth].p_ext;
1952 eh = path[depth].p_hdr;
1953 if (unlikely(path[depth].p_hdr == NULL)) {
1954 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1958 /* try to insert block into found extent and return */
1959 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1962 * Try to see whether we should rather test the extent on
1963 * right from ex, or from the left of ex. This is because
1964 * ext4_find_extent() can return either extent on the
1965 * left, or on the right from the searched position. This
1966 * will make merging more effective.
1968 if (ex < EXT_LAST_EXTENT(eh) &&
1969 (le32_to_cpu(ex->ee_block) +
1970 ext4_ext_get_actual_len(ex) <
1971 le32_to_cpu(newext->ee_block))) {
1974 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1975 (le32_to_cpu(newext->ee_block) +
1976 ext4_ext_get_actual_len(newext) <
1977 le32_to_cpu(ex->ee_block)))
1980 /* Try to append newex to the ex */
1981 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1982 ext_debug("append [%d]%d block to %u:[%d]%d"
1984 ext4_ext_is_unwritten(newext),
1985 ext4_ext_get_actual_len(newext),
1986 le32_to_cpu(ex->ee_block),
1987 ext4_ext_is_unwritten(ex),
1988 ext4_ext_get_actual_len(ex),
1989 ext4_ext_pblock(ex));
1990 err = ext4_ext_get_access(handle, inode,
1994 unwritten = ext4_ext_is_unwritten(ex);
1995 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1996 + ext4_ext_get_actual_len(newext));
1998 ext4_ext_mark_unwritten(ex);
1999 eh = path[depth].p_hdr;
2005 /* Try to prepend newex to the ex */
2006 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2007 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2009 le32_to_cpu(newext->ee_block),
2010 ext4_ext_is_unwritten(newext),
2011 ext4_ext_get_actual_len(newext),
2012 le32_to_cpu(ex->ee_block),
2013 ext4_ext_is_unwritten(ex),
2014 ext4_ext_get_actual_len(ex),
2015 ext4_ext_pblock(ex));
2016 err = ext4_ext_get_access(handle, inode,
2021 unwritten = ext4_ext_is_unwritten(ex);
2022 ex->ee_block = newext->ee_block;
2023 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2024 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2025 + ext4_ext_get_actual_len(newext));
2027 ext4_ext_mark_unwritten(ex);
2028 eh = path[depth].p_hdr;
2034 depth = ext_depth(inode);
2035 eh = path[depth].p_hdr;
2036 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2039 /* probably next leaf has space for us? */
2040 fex = EXT_LAST_EXTENT(eh);
2041 next = EXT_MAX_BLOCKS;
2042 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2043 next = ext4_ext_next_leaf_block(path);
2044 if (next != EXT_MAX_BLOCKS) {
2045 ext_debug("next leaf block - %u\n", next);
2046 BUG_ON(npath != NULL);
2047 npath = ext4_find_extent(inode, next, NULL, 0);
2049 return PTR_ERR(npath);
2050 BUG_ON(npath->p_depth != path->p_depth);
2051 eh = npath[depth].p_hdr;
2052 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2053 ext_debug("next leaf isn't full(%d)\n",
2054 le16_to_cpu(eh->eh_entries));
2058 ext_debug("next leaf has no free space(%d,%d)\n",
2059 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2063 * There is no free space in the found leaf.
2064 * We're gonna add a new leaf in the tree.
2066 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2067 mb_flags |= EXT4_MB_USE_RESERVED;
2068 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2072 depth = ext_depth(inode);
2073 eh = path[depth].p_hdr;
2076 nearex = path[depth].p_ext;
2078 err = ext4_ext_get_access(handle, inode, path + depth);
2083 /* there is no extent in this leaf, create first one */
2084 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2085 le32_to_cpu(newext->ee_block),
2086 ext4_ext_pblock(newext),
2087 ext4_ext_is_unwritten(newext),
2088 ext4_ext_get_actual_len(newext));
2089 nearex = EXT_FIRST_EXTENT(eh);
2091 if (le32_to_cpu(newext->ee_block)
2092 > le32_to_cpu(nearex->ee_block)) {
2094 ext_debug("insert %u:%llu:[%d]%d before: "
2096 le32_to_cpu(newext->ee_block),
2097 ext4_ext_pblock(newext),
2098 ext4_ext_is_unwritten(newext),
2099 ext4_ext_get_actual_len(newext),
2104 BUG_ON(newext->ee_block == nearex->ee_block);
2105 ext_debug("insert %u:%llu:[%d]%d after: "
2107 le32_to_cpu(newext->ee_block),
2108 ext4_ext_pblock(newext),
2109 ext4_ext_is_unwritten(newext),
2110 ext4_ext_get_actual_len(newext),
2113 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2115 ext_debug("insert %u:%llu:[%d]%d: "
2116 "move %d extents from 0x%p to 0x%p\n",
2117 le32_to_cpu(newext->ee_block),
2118 ext4_ext_pblock(newext),
2119 ext4_ext_is_unwritten(newext),
2120 ext4_ext_get_actual_len(newext),
2121 len, nearex, nearex + 1);
2122 memmove(nearex + 1, nearex,
2123 len * sizeof(struct ext4_extent));
2127 le16_add_cpu(&eh->eh_entries, 1);
2128 path[depth].p_ext = nearex;
2129 nearex->ee_block = newext->ee_block;
2130 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2131 nearex->ee_len = newext->ee_len;
2134 /* try to merge extents */
2135 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2136 ext4_ext_try_to_merge(handle, inode, path, nearex);
2139 /* time to correct all indexes above */
2140 err = ext4_ext_correct_indexes(handle, inode, path);
2144 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2147 ext4_ext_drop_refs(npath);
2152 static int ext4_fill_fiemap_extents(struct inode *inode,
2153 ext4_lblk_t block, ext4_lblk_t num,
2154 struct fiemap_extent_info *fieinfo)
2156 struct ext4_ext_path *path = NULL;
2157 struct ext4_extent *ex;
2158 struct extent_status es;
2159 ext4_lblk_t next, next_del, start = 0, end = 0;
2160 ext4_lblk_t last = block + num;
2161 int exists, depth = 0, err = 0;
2162 unsigned int flags = 0;
2163 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2165 while (block < last && block != EXT_MAX_BLOCKS) {
2167 /* find extent for this block */
2168 down_read(&EXT4_I(inode)->i_data_sem);
2170 path = ext4_find_extent(inode, block, &path, 0);
2172 up_read(&EXT4_I(inode)->i_data_sem);
2173 err = PTR_ERR(path);
2178 depth = ext_depth(inode);
2179 if (unlikely(path[depth].p_hdr == NULL)) {
2180 up_read(&EXT4_I(inode)->i_data_sem);
2181 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2185 ex = path[depth].p_ext;
2186 next = ext4_ext_next_allocated_block(path);
2191 /* there is no extent yet, so try to allocate
2192 * all requested space */
2195 } else if (le32_to_cpu(ex->ee_block) > block) {
2196 /* need to allocate space before found extent */
2198 end = le32_to_cpu(ex->ee_block);
2199 if (block + num < end)
2201 } else if (block >= le32_to_cpu(ex->ee_block)
2202 + ext4_ext_get_actual_len(ex)) {
2203 /* need to allocate space after found extent */
2208 } else if (block >= le32_to_cpu(ex->ee_block)) {
2210 * some part of requested space is covered
2214 end = le32_to_cpu(ex->ee_block)
2215 + ext4_ext_get_actual_len(ex);
2216 if (block + num < end)
2222 BUG_ON(end <= start);
2226 es.es_len = end - start;
2229 es.es_lblk = le32_to_cpu(ex->ee_block);
2230 es.es_len = ext4_ext_get_actual_len(ex);
2231 es.es_pblk = ext4_ext_pblock(ex);
2232 if (ext4_ext_is_unwritten(ex))
2233 flags |= FIEMAP_EXTENT_UNWRITTEN;
2237 * Find delayed extent and update es accordingly. We call
2238 * it even in !exists case to find out whether es is the
2239 * last existing extent or not.
2241 next_del = ext4_find_delayed_extent(inode, &es);
2242 if (!exists && next_del) {
2244 flags |= (FIEMAP_EXTENT_DELALLOC |
2245 FIEMAP_EXTENT_UNKNOWN);
2247 up_read(&EXT4_I(inode)->i_data_sem);
2249 if (unlikely(es.es_len == 0)) {
2250 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2256 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2257 * we need to check next == EXT_MAX_BLOCKS because it is
2258 * possible that an extent is with unwritten and delayed
2259 * status due to when an extent is delayed allocated and
2260 * is allocated by fallocate status tree will track both of
2263 * So we could return a unwritten and delayed extent, and
2264 * its block is equal to 'next'.
2266 if (next == next_del && next == EXT_MAX_BLOCKS) {
2267 flags |= FIEMAP_EXTENT_LAST;
2268 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2269 next != EXT_MAX_BLOCKS)) {
2270 EXT4_ERROR_INODE(inode,
2271 "next extent == %u, next "
2272 "delalloc extent = %u",
2280 err = fiemap_fill_next_extent(fieinfo,
2281 (__u64)es.es_lblk << blksize_bits,
2282 (__u64)es.es_pblk << blksize_bits,
2283 (__u64)es.es_len << blksize_bits,
2293 block = es.es_lblk + es.es_len;
2296 ext4_ext_drop_refs(path);
2302 * ext4_ext_put_gap_in_cache:
2303 * calculate boundaries of the gap that the requested block fits into
2304 * and cache this gap
2307 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2310 int depth = ext_depth(inode);
2311 unsigned long len = 0;
2312 ext4_lblk_t lblock = 0;
2313 struct ext4_extent *ex;
2315 ex = path[depth].p_ext;
2318 * there is no extent yet, so gap is [0;-] and we
2321 ext_debug("cache gap(whole file):");
2322 } else if (block < le32_to_cpu(ex->ee_block)) {
2324 len = le32_to_cpu(ex->ee_block) - block;
2325 ext_debug("cache gap(before): %u [%u:%u]",
2327 le32_to_cpu(ex->ee_block),
2328 ext4_ext_get_actual_len(ex));
2329 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2330 ext4_es_insert_extent(inode, lblock, len, ~0,
2331 EXTENT_STATUS_HOLE);
2332 } else if (block >= le32_to_cpu(ex->ee_block)
2333 + ext4_ext_get_actual_len(ex)) {
2335 lblock = le32_to_cpu(ex->ee_block)
2336 + ext4_ext_get_actual_len(ex);
2338 next = ext4_ext_next_allocated_block(path);
2339 ext_debug("cache gap(after): [%u:%u] %u",
2340 le32_to_cpu(ex->ee_block),
2341 ext4_ext_get_actual_len(ex),
2343 BUG_ON(next == lblock);
2344 len = next - lblock;
2345 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2346 ext4_es_insert_extent(inode, lblock, len, ~0,
2347 EXTENT_STATUS_HOLE);
2352 ext_debug(" -> %u:%lu\n", lblock, len);
2357 * removes index from the index block.
2359 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2360 struct ext4_ext_path *path, int depth)
2365 /* free index block */
2367 path = path + depth;
2368 leaf = ext4_idx_pblock(path->p_idx);
2369 if (unlikely(path->p_hdr->eh_entries == 0)) {
2370 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2373 err = ext4_ext_get_access(handle, inode, path);
2377 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2378 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2379 len *= sizeof(struct ext4_extent_idx);
2380 memmove(path->p_idx, path->p_idx + 1, len);
2383 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2384 err = ext4_ext_dirty(handle, inode, path);
2387 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2388 trace_ext4_ext_rm_idx(inode, leaf);
2390 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2391 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2393 while (--depth >= 0) {
2394 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2397 err = ext4_ext_get_access(handle, inode, path);
2400 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2401 err = ext4_ext_dirty(handle, inode, path);
2409 * ext4_ext_calc_credits_for_single_extent:
2410 * This routine returns max. credits that needed to insert an extent
2411 * to the extent tree.
2412 * When pass the actual path, the caller should calculate credits
2415 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2416 struct ext4_ext_path *path)
2419 int depth = ext_depth(inode);
2422 /* probably there is space in leaf? */
2423 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2424 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2427 * There are some space in the leaf tree, no
2428 * need to account for leaf block credit
2430 * bitmaps and block group descriptor blocks
2431 * and other metadata blocks still need to be
2434 /* 1 bitmap, 1 block group descriptor */
2435 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2440 return ext4_chunk_trans_blocks(inode, nrblocks);
2444 * How many index/leaf blocks need to change/allocate to add @extents extents?
2446 * If we add a single extent, then in the worse case, each tree level
2447 * index/leaf need to be changed in case of the tree split.
2449 * If more extents are inserted, they could cause the whole tree split more
2450 * than once, but this is really rare.
2452 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2457 /* If we are converting the inline data, only one is needed here. */
2458 if (ext4_has_inline_data(inode))
2461 depth = ext_depth(inode);
2471 static inline int get_default_free_blocks_flags(struct inode *inode)
2473 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2474 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2475 else if (ext4_should_journal_data(inode))
2476 return EXT4_FREE_BLOCKS_FORGET;
2480 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2481 struct ext4_extent *ex,
2482 long long *partial_cluster,
2483 ext4_lblk_t from, ext4_lblk_t to)
2485 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2486 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2488 int flags = get_default_free_blocks_flags(inode);
2491 * For bigalloc file systems, we never free a partial cluster
2492 * at the beginning of the extent. Instead, we make a note
2493 * that we tried freeing the cluster, and check to see if we
2494 * need to free it on a subsequent call to ext4_remove_blocks,
2495 * or at the end of the ext4_truncate() operation.
2497 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2499 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2501 * If we have a partial cluster, and it's different from the
2502 * cluster of the last block, we need to explicitly free the
2503 * partial cluster here.
2505 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2506 if ((*partial_cluster > 0) &&
2507 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2508 ext4_free_blocks(handle, inode, NULL,
2509 EXT4_C2B(sbi, *partial_cluster),
2510 sbi->s_cluster_ratio, flags);
2511 *partial_cluster = 0;
2514 #ifdef EXTENTS_STATS
2516 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2517 spin_lock(&sbi->s_ext_stats_lock);
2518 sbi->s_ext_blocks += ee_len;
2519 sbi->s_ext_extents++;
2520 if (ee_len < sbi->s_ext_min)
2521 sbi->s_ext_min = ee_len;
2522 if (ee_len > sbi->s_ext_max)
2523 sbi->s_ext_max = ee_len;
2524 if (ext_depth(inode) > sbi->s_depth_max)
2525 sbi->s_depth_max = ext_depth(inode);
2526 spin_unlock(&sbi->s_ext_stats_lock);
2529 if (from >= le32_to_cpu(ex->ee_block)
2530 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2533 unsigned int unaligned;
2535 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2536 pblk = ext4_ext_pblock(ex) + ee_len - num;
2538 * Usually we want to free partial cluster at the end of the
2539 * extent, except for the situation when the cluster is still
2540 * used by any other extent (partial_cluster is negative).
2542 if (*partial_cluster < 0 &&
2543 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2544 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2546 ext_debug("free last %u blocks starting %llu partial %lld\n",
2547 num, pblk, *partial_cluster);
2548 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2550 * If the block range to be freed didn't start at the
2551 * beginning of a cluster, and we removed the entire
2552 * extent and the cluster is not used by any other extent,
2553 * save the partial cluster here, since we might need to
2554 * delete if we determine that the truncate operation has
2555 * removed all of the blocks in the cluster.
2557 * On the other hand, if we did not manage to free the whole
2558 * extent, we have to mark the cluster as used (store negative
2559 * cluster number in partial_cluster).
2561 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2562 if (unaligned && (ee_len == num) &&
2563 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2564 *partial_cluster = EXT4_B2C(sbi, pblk);
2566 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2567 else if (*partial_cluster > 0)
2568 *partial_cluster = 0;
2570 ext4_error(sbi->s_sb, "strange request: removal(2) "
2571 "%u-%u from %u:%u\n",
2572 from, to, le32_to_cpu(ex->ee_block), ee_len);
2578 * ext4_ext_rm_leaf() Removes the extents associated with the
2579 * blocks appearing between "start" and "end", and splits the extents
2580 * if "start" and "end" appear in the same extent
2582 * @handle: The journal handle
2583 * @inode: The files inode
2584 * @path: The path to the leaf
2585 * @partial_cluster: The cluster which we'll have to free if all extents
2586 * has been released from it. It gets negative in case
2587 * that the cluster is still used.
2588 * @start: The first block to remove
2589 * @end: The last block to remove
2592 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2593 struct ext4_ext_path *path,
2594 long long *partial_cluster,
2595 ext4_lblk_t start, ext4_lblk_t end)
2597 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2598 int err = 0, correct_index = 0;
2599 int depth = ext_depth(inode), credits;
2600 struct ext4_extent_header *eh;
2603 ext4_lblk_t ex_ee_block;
2604 unsigned short ex_ee_len;
2605 unsigned unwritten = 0;
2606 struct ext4_extent *ex;
2609 /* the header must be checked already in ext4_ext_remove_space() */
2610 ext_debug("truncate since %u in leaf to %u\n", start, end);
2611 if (!path[depth].p_hdr)
2612 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2613 eh = path[depth].p_hdr;
2614 if (unlikely(path[depth].p_hdr == NULL)) {
2615 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2618 /* find where to start removing */
2619 ex = path[depth].p_ext;
2621 ex = EXT_LAST_EXTENT(eh);
2623 ex_ee_block = le32_to_cpu(ex->ee_block);
2624 ex_ee_len = ext4_ext_get_actual_len(ex);
2627 * If we're starting with an extent other than the last one in the
2628 * node, we need to see if it shares a cluster with the extent to
2629 * the right (towards the end of the file). If its leftmost cluster
2630 * is this extent's rightmost cluster and it is not cluster aligned,
2631 * we'll mark it as a partial that is not to be deallocated.
2634 if (ex != EXT_LAST_EXTENT(eh)) {
2635 ext4_fsblk_t current_pblk, right_pblk;
2636 long long current_cluster, right_cluster;
2638 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2639 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2640 right_pblk = ext4_ext_pblock(ex + 1);
2641 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2642 if (current_cluster == right_cluster &&
2643 EXT4_PBLK_COFF(sbi, right_pblk))
2644 *partial_cluster = -right_cluster;
2647 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2649 while (ex >= EXT_FIRST_EXTENT(eh) &&
2650 ex_ee_block + ex_ee_len > start) {
2652 if (ext4_ext_is_unwritten(ex))
2657 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2658 unwritten, ex_ee_len);
2659 path[depth].p_ext = ex;
2661 a = ex_ee_block > start ? ex_ee_block : start;
2662 b = ex_ee_block+ex_ee_len - 1 < end ?
2663 ex_ee_block+ex_ee_len - 1 : end;
2665 ext_debug(" border %u:%u\n", a, b);
2667 /* If this extent is beyond the end of the hole, skip it */
2668 if (end < ex_ee_block) {
2670 * We're going to skip this extent and move to another,
2671 * so if this extent is not cluster aligned we have
2672 * to mark the current cluster as used to avoid
2673 * accidentally freeing it later on
2675 pblk = ext4_ext_pblock(ex);
2676 if (EXT4_PBLK_COFF(sbi, pblk))
2678 -((long long)EXT4_B2C(sbi, pblk));
2680 ex_ee_block = le32_to_cpu(ex->ee_block);
2681 ex_ee_len = ext4_ext_get_actual_len(ex);
2683 } else if (b != ex_ee_block + ex_ee_len - 1) {
2684 EXT4_ERROR_INODE(inode,
2685 "can not handle truncate %u:%u "
2687 start, end, ex_ee_block,
2688 ex_ee_block + ex_ee_len - 1);
2691 } else if (a != ex_ee_block) {
2692 /* remove tail of the extent */
2693 num = a - ex_ee_block;
2695 /* remove whole extent: excellent! */
2699 * 3 for leaf, sb, and inode plus 2 (bmap and group
2700 * descriptor) for each block group; assume two block
2701 * groups plus ex_ee_len/blocks_per_block_group for
2704 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2705 if (ex == EXT_FIRST_EXTENT(eh)) {
2707 credits += (ext_depth(inode)) + 1;
2709 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2711 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2715 err = ext4_ext_get_access(handle, inode, path + depth);
2719 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2725 /* this extent is removed; mark slot entirely unused */
2726 ext4_ext_store_pblock(ex, 0);
2728 ex->ee_len = cpu_to_le16(num);
2730 * Do not mark unwritten if all the blocks in the
2731 * extent have been removed.
2733 if (unwritten && num)
2734 ext4_ext_mark_unwritten(ex);
2736 * If the extent was completely released,
2737 * we need to remove it from the leaf
2740 if (end != EXT_MAX_BLOCKS - 1) {
2742 * For hole punching, we need to scoot all the
2743 * extents up when an extent is removed so that
2744 * we dont have blank extents in the middle
2746 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2747 sizeof(struct ext4_extent));
2749 /* Now get rid of the one at the end */
2750 memset(EXT_LAST_EXTENT(eh), 0,
2751 sizeof(struct ext4_extent));
2753 le16_add_cpu(&eh->eh_entries, -1);
2754 } else if (*partial_cluster > 0)
2755 *partial_cluster = 0;
2757 err = ext4_ext_dirty(handle, inode, path + depth);
2761 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2762 ext4_ext_pblock(ex));
2764 ex_ee_block = le32_to_cpu(ex->ee_block);
2765 ex_ee_len = ext4_ext_get_actual_len(ex);
2768 if (correct_index && eh->eh_entries)
2769 err = ext4_ext_correct_indexes(handle, inode, path);
2772 * If there's a partial cluster and at least one extent remains in
2773 * the leaf, free the partial cluster if it isn't shared with the
2774 * current extent. If there's a partial cluster and no extents
2775 * remain in the leaf, it can't be freed here. It can only be
2776 * freed when it's possible to determine if it's not shared with
2777 * any other extent - when the next leaf is processed or when space
2778 * removal is complete.
2780 if (*partial_cluster > 0 && eh->eh_entries &&
2781 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2782 *partial_cluster)) {
2783 int flags = get_default_free_blocks_flags(inode);
2785 ext4_free_blocks(handle, inode, NULL,
2786 EXT4_C2B(sbi, *partial_cluster),
2787 sbi->s_cluster_ratio, flags);
2788 *partial_cluster = 0;
2791 /* if this leaf is free, then we should
2792 * remove it from index block above */
2793 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2794 err = ext4_ext_rm_idx(handle, inode, path, depth);
2801 * ext4_ext_more_to_rm:
2802 * returns 1 if current index has to be freed (even partial)
2805 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2807 BUG_ON(path->p_idx == NULL);
2809 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2813 * if truncate on deeper level happened, it wasn't partial,
2814 * so we have to consider current index for truncation
2816 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2821 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2824 struct super_block *sb = inode->i_sb;
2825 int depth = ext_depth(inode);
2826 struct ext4_ext_path *path = NULL;
2827 long long partial_cluster = 0;
2831 ext_debug("truncate since %u to %u\n", start, end);
2833 /* probably first extent we're gonna free will be last in block */
2834 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2836 return PTR_ERR(handle);
2839 trace_ext4_ext_remove_space(inode, start, end, depth);
2842 * Check if we are removing extents inside the extent tree. If that
2843 * is the case, we are going to punch a hole inside the extent tree
2844 * so we have to check whether we need to split the extent covering
2845 * the last block to remove so we can easily remove the part of it
2846 * in ext4_ext_rm_leaf().
2848 if (end < EXT_MAX_BLOCKS - 1) {
2849 struct ext4_extent *ex;
2850 ext4_lblk_t ee_block;
2852 /* find extent for this block */
2853 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2855 ext4_journal_stop(handle);
2856 return PTR_ERR(path);
2858 depth = ext_depth(inode);
2859 /* Leaf not may not exist only if inode has no blocks at all */
2860 ex = path[depth].p_ext;
2863 EXT4_ERROR_INODE(inode,
2864 "path[%d].p_hdr == NULL",
2871 ee_block = le32_to_cpu(ex->ee_block);
2874 * See if the last block is inside the extent, if so split
2875 * the extent at 'end' block so we can easily remove the
2876 * tail of the first part of the split extent in
2877 * ext4_ext_rm_leaf().
2879 if (end >= ee_block &&
2880 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2882 * Split the extent in two so that 'end' is the last
2883 * block in the first new extent. Also we should not
2884 * fail removing space due to ENOSPC so try to use
2885 * reserved block if that happens.
2887 err = ext4_force_split_extent_at(handle, inode, &path,
2894 * We start scanning from right side, freeing all the blocks
2895 * after i_size and walking into the tree depth-wise.
2897 depth = ext_depth(inode);
2902 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2904 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2907 ext4_journal_stop(handle);
2910 path[0].p_maxdepth = path[0].p_depth = depth;
2911 path[0].p_hdr = ext_inode_hdr(inode);
2914 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2921 while (i >= 0 && err == 0) {
2923 /* this is leaf block */
2924 err = ext4_ext_rm_leaf(handle, inode, path,
2925 &partial_cluster, start,
2927 /* root level has p_bh == NULL, brelse() eats this */
2928 brelse(path[i].p_bh);
2929 path[i].p_bh = NULL;
2934 /* this is index block */
2935 if (!path[i].p_hdr) {
2936 ext_debug("initialize header\n");
2937 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2940 if (!path[i].p_idx) {
2941 /* this level hasn't been touched yet */
2942 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2943 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2944 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2946 le16_to_cpu(path[i].p_hdr->eh_entries));
2948 /* we were already here, see at next index */
2952 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2953 i, EXT_FIRST_INDEX(path[i].p_hdr),
2955 if (ext4_ext_more_to_rm(path + i)) {
2956 struct buffer_head *bh;
2957 /* go to the next level */
2958 ext_debug("move to level %d (block %llu)\n",
2959 i + 1, ext4_idx_pblock(path[i].p_idx));
2960 memset(path + i + 1, 0, sizeof(*path));
2961 bh = read_extent_tree_block(inode,
2962 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2965 /* should we reset i_size? */
2969 /* Yield here to deal with large extent trees.
2970 * Should be a no-op if we did IO above. */
2972 if (WARN_ON(i + 1 > depth)) {
2976 path[i + 1].p_bh = bh;
2978 /* save actual number of indexes since this
2979 * number is changed at the next iteration */
2980 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2983 /* we finished processing this index, go up */
2984 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2985 /* index is empty, remove it;
2986 * handle must be already prepared by the
2987 * truncatei_leaf() */
2988 err = ext4_ext_rm_idx(handle, inode, path, i);
2990 /* root level has p_bh == NULL, brelse() eats this */
2991 brelse(path[i].p_bh);
2992 path[i].p_bh = NULL;
2994 ext_debug("return to level %d\n", i);
2998 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2999 partial_cluster, path->p_hdr->eh_entries);
3001 /* If we still have something in the partial cluster and we have removed
3002 * even the first extent, then we should free the blocks in the partial
3003 * cluster as well. */
3004 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
3005 int flags = get_default_free_blocks_flags(inode);
3007 ext4_free_blocks(handle, inode, NULL,
3008 EXT4_C2B(EXT4_SB(sb), partial_cluster),
3009 EXT4_SB(sb)->s_cluster_ratio, flags);
3010 partial_cluster = 0;
3013 /* TODO: flexible tree reduction should be here */
3014 if (path->p_hdr->eh_entries == 0) {
3016 * truncate to zero freed all the tree,
3017 * so we need to correct eh_depth
3019 err = ext4_ext_get_access(handle, inode, path);
3021 ext_inode_hdr(inode)->eh_depth = 0;
3022 ext_inode_hdr(inode)->eh_max =
3023 cpu_to_le16(ext4_ext_space_root(inode, 0));
3024 err = ext4_ext_dirty(handle, inode, path);
3028 ext4_ext_drop_refs(path);
3033 ext4_journal_stop(handle);
3039 * called at mount time
3041 void ext4_ext_init(struct super_block *sb)
3044 * possible initialization would be here
3047 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3048 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3049 printk(KERN_INFO "EXT4-fs: file extents enabled"
3050 #ifdef AGGRESSIVE_TEST
3051 ", aggressive tests"
3053 #ifdef CHECK_BINSEARCH
3056 #ifdef EXTENTS_STATS
3061 #ifdef EXTENTS_STATS
3062 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3063 EXT4_SB(sb)->s_ext_min = 1 << 30;
3064 EXT4_SB(sb)->s_ext_max = 0;
3070 * called at umount time
3072 void ext4_ext_release(struct super_block *sb)
3074 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3077 #ifdef EXTENTS_STATS
3078 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3079 struct ext4_sb_info *sbi = EXT4_SB(sb);
3080 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3081 sbi->s_ext_blocks, sbi->s_ext_extents,
3082 sbi->s_ext_blocks / sbi->s_ext_extents);
3083 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3084 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3089 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3091 ext4_lblk_t ee_block;
3092 ext4_fsblk_t ee_pblock;
3093 unsigned int ee_len;
3095 ee_block = le32_to_cpu(ex->ee_block);
3096 ee_len = ext4_ext_get_actual_len(ex);
3097 ee_pblock = ext4_ext_pblock(ex);
3102 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3103 EXTENT_STATUS_WRITTEN);
3106 /* FIXME!! we need to try to merge to left or right after zero-out */
3107 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3109 ext4_fsblk_t ee_pblock;
3110 unsigned int ee_len;
3113 ee_len = ext4_ext_get_actual_len(ex);
3114 ee_pblock = ext4_ext_pblock(ex);
3116 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3124 * ext4_split_extent_at() splits an extent at given block.
3126 * @handle: the journal handle
3127 * @inode: the file inode
3128 * @path: the path to the extent
3129 * @split: the logical block where the extent is splitted.
3130 * @split_flags: indicates if the extent could be zeroout if split fails, and
3131 * the states(init or unwritten) of new extents.
3132 * @flags: flags used to insert new extent to extent tree.
3135 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3136 * of which are deterimined by split_flag.
3138 * There are two cases:
3139 * a> the extent are splitted into two extent.
3140 * b> split is not needed, and just mark the extent.
3142 * return 0 on success.
3144 static int ext4_split_extent_at(handle_t *handle,
3145 struct inode *inode,
3146 struct ext4_ext_path **ppath,
3151 struct ext4_ext_path *path = *ppath;
3152 ext4_fsblk_t newblock;
3153 ext4_lblk_t ee_block;
3154 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3155 struct ext4_extent *ex2 = NULL;
3156 unsigned int ee_len, depth;
3159 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3160 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3162 ext_debug("ext4_split_extents_at: inode %lu, logical"
3163 "block %llu\n", inode->i_ino, (unsigned long long)split);
3165 ext4_ext_show_leaf(inode, path);
3167 depth = ext_depth(inode);
3168 ex = path[depth].p_ext;
3169 ee_block = le32_to_cpu(ex->ee_block);
3170 ee_len = ext4_ext_get_actual_len(ex);
3171 newblock = split - ee_block + ext4_ext_pblock(ex);
3173 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3174 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3175 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3176 EXT4_EXT_MARK_UNWRIT1 |
3177 EXT4_EXT_MARK_UNWRIT2));
3179 err = ext4_ext_get_access(handle, inode, path + depth);
3183 if (split == ee_block) {
3185 * case b: block @split is the block that the extent begins with
3186 * then we just change the state of the extent, and splitting
3189 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3190 ext4_ext_mark_unwritten(ex);
3192 ext4_ext_mark_initialized(ex);
3194 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3195 ext4_ext_try_to_merge(handle, inode, path, ex);
3197 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3202 memcpy(&orig_ex, ex, sizeof(orig_ex));
3203 ex->ee_len = cpu_to_le16(split - ee_block);
3204 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3205 ext4_ext_mark_unwritten(ex);
3208 * path may lead to new leaf, not to original leaf any more
3209 * after ext4_ext_insert_extent() returns,
3211 err = ext4_ext_dirty(handle, inode, path + depth);
3213 goto fix_extent_len;
3216 ex2->ee_block = cpu_to_le32(split);
3217 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3218 ext4_ext_store_pblock(ex2, newblock);
3219 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3220 ext4_ext_mark_unwritten(ex2);
3222 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3223 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3224 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3225 if (split_flag & EXT4_EXT_DATA_VALID1) {
3226 err = ext4_ext_zeroout(inode, ex2);
3227 zero_ex.ee_block = ex2->ee_block;
3228 zero_ex.ee_len = cpu_to_le16(
3229 ext4_ext_get_actual_len(ex2));
3230 ext4_ext_store_pblock(&zero_ex,
3231 ext4_ext_pblock(ex2));
3233 err = ext4_ext_zeroout(inode, ex);
3234 zero_ex.ee_block = ex->ee_block;
3235 zero_ex.ee_len = cpu_to_le16(
3236 ext4_ext_get_actual_len(ex));
3237 ext4_ext_store_pblock(&zero_ex,
3238 ext4_ext_pblock(ex));
3241 err = ext4_ext_zeroout(inode, &orig_ex);
3242 zero_ex.ee_block = orig_ex.ee_block;
3243 zero_ex.ee_len = cpu_to_le16(
3244 ext4_ext_get_actual_len(&orig_ex));
3245 ext4_ext_store_pblock(&zero_ex,
3246 ext4_ext_pblock(&orig_ex));
3250 goto fix_extent_len;
3251 /* update the extent length and mark as initialized */
3252 ex->ee_len = cpu_to_le16(ee_len);
3253 ext4_ext_try_to_merge(handle, inode, path, ex);
3254 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3256 goto fix_extent_len;
3258 /* update extent status tree */
3259 err = ext4_zeroout_es(inode, &zero_ex);
3263 goto fix_extent_len;
3266 ext4_ext_show_leaf(inode, path);
3270 ex->ee_len = orig_ex.ee_len;
3271 ext4_ext_dirty(handle, inode, path + path->p_depth);
3276 * ext4_split_extents() splits an extent and mark extent which is covered
3277 * by @map as split_flags indicates
3279 * It may result in splitting the extent into multiple extents (up to three)
3280 * There are three possibilities:
3281 * a> There is no split required
3282 * b> Splits in two extents: Split is happening at either end of the extent
3283 * c> Splits in three extents: Somone is splitting in middle of the extent
3286 static int ext4_split_extent(handle_t *handle,
3287 struct inode *inode,
3288 struct ext4_ext_path **ppath,
3289 struct ext4_map_blocks *map,
3293 struct ext4_ext_path *path = *ppath;
3294 ext4_lblk_t ee_block;
3295 struct ext4_extent *ex;
3296 unsigned int ee_len, depth;
3299 int split_flag1, flags1;
3300 int allocated = map->m_len;
3302 depth = ext_depth(inode);
3303 ex = path[depth].p_ext;
3304 ee_block = le32_to_cpu(ex->ee_block);
3305 ee_len = ext4_ext_get_actual_len(ex);
3306 unwritten = ext4_ext_is_unwritten(ex);
3308 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3309 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3310 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3312 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3313 EXT4_EXT_MARK_UNWRIT2;
3314 if (split_flag & EXT4_EXT_DATA_VALID2)
3315 split_flag1 |= EXT4_EXT_DATA_VALID1;
3316 err = ext4_split_extent_at(handle, inode, ppath,
3317 map->m_lblk + map->m_len, split_flag1, flags1);
3321 allocated = ee_len - (map->m_lblk - ee_block);
3324 * Update path is required because previous ext4_split_extent_at() may
3325 * result in split of original leaf or extent zeroout.
3327 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3329 return PTR_ERR(path);
3330 depth = ext_depth(inode);
3331 ex = path[depth].p_ext;
3333 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3334 (unsigned long) map->m_lblk);
3337 unwritten = ext4_ext_is_unwritten(ex);
3340 if (map->m_lblk >= ee_block) {
3341 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3343 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3344 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3345 EXT4_EXT_MARK_UNWRIT2);
3347 err = ext4_split_extent_at(handle, inode, ppath,
3348 map->m_lblk, split_flag1, flags);
3353 ext4_ext_show_leaf(inode, path);
3355 return err ? err : allocated;
3359 * This function is called by ext4_ext_map_blocks() if someone tries to write
3360 * to an unwritten extent. It may result in splitting the unwritten
3361 * extent into multiple extents (up to three - one initialized and two
3363 * There are three possibilities:
3364 * a> There is no split required: Entire extent should be initialized
3365 * b> Splits in two extents: Write is happening at either end of the extent
3366 * c> Splits in three extents: Somone is writing in middle of the extent
3369 * - The extent pointed to by 'path' is unwritten.
3370 * - The extent pointed to by 'path' contains a superset
3371 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3373 * Post-conditions on success:
3374 * - the returned value is the number of blocks beyond map->l_lblk
3375 * that are allocated and initialized.
3376 * It is guaranteed to be >= map->m_len.
3378 static int ext4_ext_convert_to_initialized(handle_t *handle,
3379 struct inode *inode,
3380 struct ext4_map_blocks *map,
3381 struct ext4_ext_path **ppath,
3384 struct ext4_ext_path *path = *ppath;
3385 struct ext4_sb_info *sbi;
3386 struct ext4_extent_header *eh;
3387 struct ext4_map_blocks split_map;
3388 struct ext4_extent zero_ex;
3389 struct ext4_extent *ex, *abut_ex;
3390 ext4_lblk_t ee_block, eof_block;
3391 unsigned int ee_len, depth, map_len = map->m_len;
3392 int allocated = 0, max_zeroout = 0;
3396 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3397 "block %llu, max_blocks %u\n", inode->i_ino,
3398 (unsigned long long)map->m_lblk, map_len);
3400 sbi = EXT4_SB(inode->i_sb);
3401 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3402 inode->i_sb->s_blocksize_bits;
3403 if (eof_block < map->m_lblk + map_len)
3404 eof_block = map->m_lblk + map_len;
3406 depth = ext_depth(inode);
3407 eh = path[depth].p_hdr;
3408 ex = path[depth].p_ext;
3409 ee_block = le32_to_cpu(ex->ee_block);
3410 ee_len = ext4_ext_get_actual_len(ex);
3413 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3415 /* Pre-conditions */
3416 BUG_ON(!ext4_ext_is_unwritten(ex));
3417 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3420 * Attempt to transfer newly initialized blocks from the currently
3421 * unwritten extent to its neighbor. This is much cheaper
3422 * than an insertion followed by a merge as those involve costly
3423 * memmove() calls. Transferring to the left is the common case in
3424 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3425 * followed by append writes.
3427 * Limitations of the current logic:
3428 * - L1: we do not deal with writes covering the whole extent.
3429 * This would require removing the extent if the transfer
3431 * - L2: we only attempt to merge with an extent stored in the
3432 * same extent tree node.
3434 if ((map->m_lblk == ee_block) &&
3435 /* See if we can merge left */
3436 (map_len < ee_len) && /*L1*/
3437 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3438 ext4_lblk_t prev_lblk;
3439 ext4_fsblk_t prev_pblk, ee_pblk;
3440 unsigned int prev_len;
3443 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3444 prev_len = ext4_ext_get_actual_len(abut_ex);
3445 prev_pblk = ext4_ext_pblock(abut_ex);
3446 ee_pblk = ext4_ext_pblock(ex);
3449 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3450 * upon those conditions:
3451 * - C1: abut_ex is initialized,
3452 * - C2: abut_ex is logically abutting ex,
3453 * - C3: abut_ex is physically abutting ex,
3454 * - C4: abut_ex can receive the additional blocks without
3455 * overflowing the (initialized) length limit.
3457 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3458 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3459 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3460 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3461 err = ext4_ext_get_access(handle, inode, path + depth);
3465 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3468 /* Shift the start of ex by 'map_len' blocks */
3469 ex->ee_block = cpu_to_le32(ee_block + map_len);
3470 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3471 ex->ee_len = cpu_to_le16(ee_len - map_len);
3472 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3474 /* Extend abut_ex by 'map_len' blocks */
3475 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3477 /* Result: number of initialized blocks past m_lblk */
3478 allocated = map_len;
3480 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3481 (map_len < ee_len) && /*L1*/
3482 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3483 /* See if we can merge right */
3484 ext4_lblk_t next_lblk;
3485 ext4_fsblk_t next_pblk, ee_pblk;
3486 unsigned int next_len;
3489 next_lblk = le32_to_cpu(abut_ex->ee_block);
3490 next_len = ext4_ext_get_actual_len(abut_ex);
3491 next_pblk = ext4_ext_pblock(abut_ex);
3492 ee_pblk = ext4_ext_pblock(ex);
3495 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3496 * upon those conditions:
3497 * - C1: abut_ex is initialized,
3498 * - C2: abut_ex is logically abutting ex,
3499 * - C3: abut_ex is physically abutting ex,
3500 * - C4: abut_ex can receive the additional blocks without
3501 * overflowing the (initialized) length limit.
3503 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3504 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3505 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3506 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3507 err = ext4_ext_get_access(handle, inode, path + depth);
3511 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3514 /* Shift the start of abut_ex by 'map_len' blocks */
3515 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3516 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3517 ex->ee_len = cpu_to_le16(ee_len - map_len);
3518 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3520 /* Extend abut_ex by 'map_len' blocks */
3521 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3523 /* Result: number of initialized blocks past m_lblk */
3524 allocated = map_len;
3528 /* Mark the block containing both extents as dirty */
3529 ext4_ext_dirty(handle, inode, path + depth);
3531 /* Update path to point to the right extent */
3532 path[depth].p_ext = abut_ex;
3535 allocated = ee_len - (map->m_lblk - ee_block);
3537 WARN_ON(map->m_lblk < ee_block);
3539 * It is safe to convert extent to initialized via explicit
3540 * zeroout only if extent is fully inside i_size or new_size.
3542 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3544 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3545 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3546 (inode->i_sb->s_blocksize_bits - 10);
3548 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3549 if (max_zeroout && (ee_len <= max_zeroout)) {
3550 err = ext4_ext_zeroout(inode, ex);
3553 zero_ex.ee_block = ex->ee_block;
3554 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3555 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3557 err = ext4_ext_get_access(handle, inode, path + depth);
3560 ext4_ext_mark_initialized(ex);
3561 ext4_ext_try_to_merge(handle, inode, path, ex);
3562 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3568 * 1. split the extent into three extents.
3569 * 2. split the extent into two extents, zeroout the first half.
3570 * 3. split the extent into two extents, zeroout the second half.
3571 * 4. split the extent into two extents with out zeroout.
3573 split_map.m_lblk = map->m_lblk;
3574 split_map.m_len = map->m_len;
3576 if (max_zeroout && (allocated > map->m_len)) {
3577 if (allocated <= max_zeroout) {
3580 cpu_to_le32(map->m_lblk);
3581 zero_ex.ee_len = cpu_to_le16(allocated);
3582 ext4_ext_store_pblock(&zero_ex,
3583 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3584 err = ext4_ext_zeroout(inode, &zero_ex);
3587 split_map.m_lblk = map->m_lblk;
3588 split_map.m_len = allocated;
3589 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3591 if (map->m_lblk != ee_block) {
3592 zero_ex.ee_block = ex->ee_block;
3593 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3595 ext4_ext_store_pblock(&zero_ex,
3596 ext4_ext_pblock(ex));
3597 err = ext4_ext_zeroout(inode, &zero_ex);
3602 split_map.m_lblk = ee_block;
3603 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3604 allocated = map->m_len;
3608 allocated = ext4_split_extent(handle, inode, ppath,
3609 &split_map, split_flag, flags);
3614 /* If we have gotten a failure, don't zero out status tree */
3616 err = ext4_zeroout_es(inode, &zero_ex);
3617 return err ? err : allocated;
3621 * This function is called by ext4_ext_map_blocks() from
3622 * ext4_get_blocks_dio_write() when DIO to write
3623 * to an unwritten extent.
3625 * Writing to an unwritten extent may result in splitting the unwritten
3626 * extent into multiple initialized/unwritten extents (up to three)
3627 * There are three possibilities:
3628 * a> There is no split required: Entire extent should be unwritten
3629 * b> Splits in two extents: Write is happening at either end of the extent
3630 * c> Splits in three extents: Somone is writing in middle of the extent
3632 * This works the same way in the case of initialized -> unwritten conversion.
3634 * One of more index blocks maybe needed if the extent tree grow after
3635 * the unwritten extent split. To prevent ENOSPC occur at the IO
3636 * complete, we need to split the unwritten extent before DIO submit
3637 * the IO. The unwritten extent called at this time will be split
3638 * into three unwritten extent(at most). After IO complete, the part
3639 * being filled will be convert to initialized by the end_io callback function
3640 * via ext4_convert_unwritten_extents().
3642 * Returns the size of unwritten extent to be written on success.
3644 static int ext4_split_convert_extents(handle_t *handle,
3645 struct inode *inode,
3646 struct ext4_map_blocks *map,
3647 struct ext4_ext_path **ppath,
3650 struct ext4_ext_path *path = *ppath;
3651 ext4_lblk_t eof_block;
3652 ext4_lblk_t ee_block;
3653 struct ext4_extent *ex;
3654 unsigned int ee_len;
3655 int split_flag = 0, depth;
3657 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3658 __func__, inode->i_ino,
3659 (unsigned long long)map->m_lblk, map->m_len);
3661 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3662 inode->i_sb->s_blocksize_bits;
3663 if (eof_block < map->m_lblk + map->m_len)
3664 eof_block = map->m_lblk + map->m_len;
3666 * It is safe to convert extent to initialized via explicit
3667 * zeroout only if extent is fully insde i_size or new_size.
3669 depth = ext_depth(inode);
3670 ex = path[depth].p_ext;
3671 ee_block = le32_to_cpu(ex->ee_block);
3672 ee_len = ext4_ext_get_actual_len(ex);
3674 /* Convert to unwritten */
3675 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3676 split_flag |= EXT4_EXT_DATA_VALID1;
3677 /* Convert to initialized */
3678 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3679 split_flag |= ee_block + ee_len <= eof_block ?
3680 EXT4_EXT_MAY_ZEROOUT : 0;
3681 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3683 flags |= EXT4_GET_BLOCKS_PRE_IO;
3684 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3687 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3688 struct inode *inode,
3689 struct ext4_map_blocks *map,
3690 struct ext4_ext_path **ppath)
3692 struct ext4_ext_path *path = *ppath;
3693 struct ext4_extent *ex;
3694 ext4_lblk_t ee_block;
3695 unsigned int ee_len;
3699 depth = ext_depth(inode);
3700 ex = path[depth].p_ext;
3701 ee_block = le32_to_cpu(ex->ee_block);
3702 ee_len = ext4_ext_get_actual_len(ex);
3704 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3705 "block %llu, max_blocks %u\n", inode->i_ino,
3706 (unsigned long long)ee_block, ee_len);
3708 /* If extent is larger than requested it is a clear sign that we still
3709 * have some extent state machine issues left. So extent_split is still
3711 * TODO: Once all related issues will be fixed this situation should be
3714 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3716 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3717 " len %u; IO logical block %llu, len %u\n",
3718 inode->i_ino, (unsigned long long)ee_block, ee_len,
3719 (unsigned long long)map->m_lblk, map->m_len);
3721 err = ext4_split_convert_extents(handle, inode, map, ppath,
3722 EXT4_GET_BLOCKS_CONVERT);
3725 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3727 return PTR_ERR(path);
3728 depth = ext_depth(inode);
3729 ex = path[depth].p_ext;
3732 err = ext4_ext_get_access(handle, inode, path + depth);
3735 /* first mark the extent as initialized */
3736 ext4_ext_mark_initialized(ex);
3738 /* note: ext4_ext_correct_indexes() isn't needed here because
3739 * borders are not changed
3741 ext4_ext_try_to_merge(handle, inode, path, ex);
3743 /* Mark modified extent as dirty */
3744 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3746 ext4_ext_show_leaf(inode, path);
3750 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3751 sector_t block, int count)
3754 for (i = 0; i < count; i++)
3755 unmap_underlying_metadata(bdev, block + i);
3759 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3761 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3763 struct ext4_ext_path *path,
3767 struct ext4_extent_header *eh;
3768 struct ext4_extent *last_ex;
3770 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3773 depth = ext_depth(inode);
3774 eh = path[depth].p_hdr;
3777 * We're going to remove EOFBLOCKS_FL entirely in future so we
3778 * do not care for this case anymore. Simply remove the flag
3779 * if there are no extents.
3781 if (unlikely(!eh->eh_entries))
3783 last_ex = EXT_LAST_EXTENT(eh);
3785 * We should clear the EOFBLOCKS_FL flag if we are writing the
3786 * last block in the last extent in the file. We test this by
3787 * first checking to see if the caller to
3788 * ext4_ext_get_blocks() was interested in the last block (or
3789 * a block beyond the last block) in the current extent. If
3790 * this turns out to be false, we can bail out from this
3791 * function immediately.
3793 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3794 ext4_ext_get_actual_len(last_ex))
3797 * If the caller does appear to be planning to write at or
3798 * beyond the end of the current extent, we then test to see
3799 * if the current extent is the last extent in the file, by
3800 * checking to make sure it was reached via the rightmost node
3801 * at each level of the tree.
3803 for (i = depth-1; i >= 0; i--)
3804 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3807 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3808 return ext4_mark_inode_dirty(handle, inode);
3812 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3814 * Return 1 if there is a delalloc block in the range, otherwise 0.
3816 int ext4_find_delalloc_range(struct inode *inode,
3817 ext4_lblk_t lblk_start,
3818 ext4_lblk_t lblk_end)
3820 struct extent_status es;
3822 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3824 return 0; /* there is no delay extent in this tree */
3825 else if (es.es_lblk <= lblk_start &&
3826 lblk_start < es.es_lblk + es.es_len)
3828 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3834 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3836 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3837 ext4_lblk_t lblk_start, lblk_end;
3838 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3839 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3841 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3845 * Determines how many complete clusters (out of those specified by the 'map')
3846 * are under delalloc and were reserved quota for.
3847 * This function is called when we are writing out the blocks that were
3848 * originally written with their allocation delayed, but then the space was
3849 * allocated using fallocate() before the delayed allocation could be resolved.
3850 * The cases to look for are:
3851 * ('=' indicated delayed allocated blocks
3852 * '-' indicates non-delayed allocated blocks)
3853 * (a) partial clusters towards beginning and/or end outside of allocated range
3854 * are not delalloc'ed.
3856 * |----c---=|====c====|====c====|===-c----|
3857 * |++++++ allocated ++++++|
3858 * ==> 4 complete clusters in above example
3860 * (b) partial cluster (outside of allocated range) towards either end is
3861 * marked for delayed allocation. In this case, we will exclude that
3864 * |----====c========|========c========|
3865 * |++++++ allocated ++++++|
3866 * ==> 1 complete clusters in above example
3869 * |================c================|
3870 * |++++++ allocated ++++++|
3871 * ==> 0 complete clusters in above example
3873 * The ext4_da_update_reserve_space will be called only if we
3874 * determine here that there were some "entire" clusters that span
3875 * this 'allocated' range.
3876 * In the non-bigalloc case, this function will just end up returning num_blks
3877 * without ever calling ext4_find_delalloc_range.
3880 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3881 unsigned int num_blks)
3883 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3884 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3885 ext4_lblk_t lblk_from, lblk_to, c_offset;
3886 unsigned int allocated_clusters = 0;
3888 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3889 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3891 /* max possible clusters for this allocation */
3892 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3894 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3896 /* Check towards left side */
3897 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3899 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3900 lblk_to = lblk_from + c_offset - 1;
3902 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3903 allocated_clusters--;
3906 /* Now check towards right. */
3907 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3908 if (allocated_clusters && c_offset) {
3909 lblk_from = lblk_start + num_blks;
3910 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3912 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3913 allocated_clusters--;
3916 return allocated_clusters;
3920 convert_initialized_extent(handle_t *handle, struct inode *inode,
3921 struct ext4_map_blocks *map,
3922 struct ext4_ext_path **ppath, int flags,
3923 unsigned int allocated, ext4_fsblk_t newblock)
3925 struct ext4_ext_path *path = *ppath;
3926 struct ext4_extent *ex;
3927 ext4_lblk_t ee_block;
3928 unsigned int ee_len;
3933 * Make sure that the extent is no bigger than we support with
3936 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3937 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3939 depth = ext_depth(inode);
3940 ex = path[depth].p_ext;
3941 ee_block = le32_to_cpu(ex->ee_block);
3942 ee_len = ext4_ext_get_actual_len(ex);
3944 ext_debug("%s: inode %lu, logical"
3945 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3946 (unsigned long long)ee_block, ee_len);
3948 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3949 err = ext4_split_convert_extents(handle, inode, map, ppath,
3950 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3953 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3955 return PTR_ERR(path);
3956 depth = ext_depth(inode);
3957 ex = path[depth].p_ext;
3959 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3960 (unsigned long) map->m_lblk);
3965 err = ext4_ext_get_access(handle, inode, path + depth);
3968 /* first mark the extent as unwritten */
3969 ext4_ext_mark_unwritten(ex);
3971 /* note: ext4_ext_correct_indexes() isn't needed here because
3972 * borders are not changed
3974 ext4_ext_try_to_merge(handle, inode, path, ex);
3976 /* Mark modified extent as dirty */
3977 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3980 ext4_ext_show_leaf(inode, path);
3982 ext4_update_inode_fsync_trans(handle, inode, 1);
3983 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3986 map->m_flags |= EXT4_MAP_UNWRITTEN;
3987 if (allocated > map->m_len)
3988 allocated = map->m_len;
3989 map->m_len = allocated;
3994 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3995 struct ext4_map_blocks *map,
3996 struct ext4_ext_path **ppath, int flags,
3997 unsigned int allocated, ext4_fsblk_t newblock)
3999 struct ext4_ext_path *path = *ppath;
4002 ext4_io_end_t *io = ext4_inode_aio(inode);
4004 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4005 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4006 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4008 ext4_ext_show_leaf(inode, path);
4011 * When writing into unwritten space, we should not fail to
4012 * allocate metadata blocks for the new extent block if needed.
4014 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4016 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4017 allocated, newblock);
4019 /* get_block() before submit the IO, split the extent */
4020 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4021 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4022 flags | EXT4_GET_BLOCKS_CONVERT);
4026 * Flag the inode(non aio case) or end_io struct (aio case)
4027 * that this IO needs to conversion to written when IO is
4031 ext4_set_io_unwritten_flag(inode, io);
4033 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4034 map->m_flags |= EXT4_MAP_UNWRITTEN;
4037 /* IO end_io complete, convert the filled extent to written */
4038 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4039 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4042 ext4_update_inode_fsync_trans(handle, inode, 1);
4043 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4047 map->m_flags |= EXT4_MAP_MAPPED;
4048 map->m_pblk = newblock;
4049 if (allocated > map->m_len)
4050 allocated = map->m_len;
4051 map->m_len = allocated;
4054 /* buffered IO case */
4056 * repeat fallocate creation request
4057 * we already have an unwritten extent
4059 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4060 map->m_flags |= EXT4_MAP_UNWRITTEN;
4064 /* buffered READ or buffered write_begin() lookup */
4065 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4067 * We have blocks reserved already. We
4068 * return allocated blocks so that delalloc
4069 * won't do block reservation for us. But
4070 * the buffer head will be unmapped so that
4071 * a read from the block returns 0s.
4073 map->m_flags |= EXT4_MAP_UNWRITTEN;
4077 /* buffered write, writepage time, convert*/
4078 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4080 ext4_update_inode_fsync_trans(handle, inode, 1);
4087 map->m_flags |= EXT4_MAP_NEW;
4089 * if we allocated more blocks than requested
4090 * we need to make sure we unmap the extra block
4091 * allocated. The actual needed block will get
4092 * unmapped later when we find the buffer_head marked
4095 if (allocated > map->m_len) {
4096 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4097 newblock + map->m_len,
4098 allocated - map->m_len);
4099 allocated = map->m_len;
4101 map->m_len = allocated;
4104 * If we have done fallocate with the offset that is already
4105 * delayed allocated, we would have block reservation
4106 * and quota reservation done in the delayed write path.
4107 * But fallocate would have already updated quota and block
4108 * count for this offset. So cancel these reservation
4110 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4111 unsigned int reserved_clusters;
4112 reserved_clusters = get_reserved_cluster_alloc(inode,
4113 map->m_lblk, map->m_len);
4114 if (reserved_clusters)
4115 ext4_da_update_reserve_space(inode,
4121 map->m_flags |= EXT4_MAP_MAPPED;
4122 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4123 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4129 if (allocated > map->m_len)
4130 allocated = map->m_len;
4131 ext4_ext_show_leaf(inode, path);
4132 map->m_pblk = newblock;
4133 map->m_len = allocated;
4135 return err ? err : allocated;
4139 * get_implied_cluster_alloc - check to see if the requested
4140 * allocation (in the map structure) overlaps with a cluster already
4141 * allocated in an extent.
4142 * @sb The filesystem superblock structure
4143 * @map The requested lblk->pblk mapping
4144 * @ex The extent structure which might contain an implied
4145 * cluster allocation
4147 * This function is called by ext4_ext_map_blocks() after we failed to
4148 * find blocks that were already in the inode's extent tree. Hence,
4149 * we know that the beginning of the requested region cannot overlap
4150 * the extent from the inode's extent tree. There are three cases we
4151 * want to catch. The first is this case:
4153 * |--- cluster # N--|
4154 * |--- extent ---| |---- requested region ---|
4157 * The second case that we need to test for is this one:
4159 * |--------- cluster # N ----------------|
4160 * |--- requested region --| |------- extent ----|
4161 * |=======================|
4163 * The third case is when the requested region lies between two extents
4164 * within the same cluster:
4165 * |------------- cluster # N-------------|
4166 * |----- ex -----| |---- ex_right ----|
4167 * |------ requested region ------|
4168 * |================|
4170 * In each of the above cases, we need to set the map->m_pblk and
4171 * map->m_len so it corresponds to the return the extent labelled as
4172 * "|====|" from cluster #N, since it is already in use for data in
4173 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4174 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4175 * as a new "allocated" block region. Otherwise, we will return 0 and
4176 * ext4_ext_map_blocks() will then allocate one or more new clusters
4177 * by calling ext4_mb_new_blocks().
4179 static int get_implied_cluster_alloc(struct super_block *sb,
4180 struct ext4_map_blocks *map,
4181 struct ext4_extent *ex,
4182 struct ext4_ext_path *path)
4184 struct ext4_sb_info *sbi = EXT4_SB(sb);
4185 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4186 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4187 ext4_lblk_t rr_cluster_start;
4188 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4189 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4190 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4192 /* The extent passed in that we are trying to match */
4193 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4194 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4196 /* The requested region passed into ext4_map_blocks() */
4197 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4199 if ((rr_cluster_start == ex_cluster_end) ||
4200 (rr_cluster_start == ex_cluster_start)) {
4201 if (rr_cluster_start == ex_cluster_end)
4202 ee_start += ee_len - 1;
4203 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4204 map->m_len = min(map->m_len,
4205 (unsigned) sbi->s_cluster_ratio - c_offset);
4207 * Check for and handle this case:
4209 * |--------- cluster # N-------------|
4210 * |------- extent ----|
4211 * |--- requested region ---|
4215 if (map->m_lblk < ee_block)
4216 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4219 * Check for the case where there is already another allocated
4220 * block to the right of 'ex' but before the end of the cluster.
4222 * |------------- cluster # N-------------|
4223 * |----- ex -----| |---- ex_right ----|
4224 * |------ requested region ------|
4225 * |================|
4227 if (map->m_lblk > ee_block) {
4228 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4229 map->m_len = min(map->m_len, next - map->m_lblk);
4232 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4236 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4242 * Block allocation/map/preallocation routine for extents based files
4245 * Need to be called with
4246 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4247 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4249 * return > 0, number of of blocks already mapped/allocated
4250 * if create == 0 and these are pre-allocated blocks
4251 * buffer head is unmapped
4252 * otherwise blocks are mapped
4254 * return = 0, if plain look up failed (blocks have not been allocated)
4255 * buffer head is unmapped
4257 * return < 0, error case.
4259 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4260 struct ext4_map_blocks *map, int flags)
4262 struct ext4_ext_path *path = NULL;
4263 struct ext4_extent newex, *ex, *ex2;
4264 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4265 ext4_fsblk_t newblock = 0;
4266 int free_on_err = 0, err = 0, depth, ret;
4267 unsigned int allocated = 0, offset = 0;
4268 unsigned int allocated_clusters = 0;
4269 struct ext4_allocation_request ar;
4270 ext4_io_end_t *io = ext4_inode_aio(inode);
4271 ext4_lblk_t cluster_offset;
4272 int set_unwritten = 0;
4274 ext_debug("blocks %u/%u requested for inode %lu\n",
4275 map->m_lblk, map->m_len, inode->i_ino);
4276 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4278 /* find extent for this block */
4279 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4281 err = PTR_ERR(path);
4286 depth = ext_depth(inode);
4289 * consistent leaf must not be empty;
4290 * this situation is possible, though, _during_ tree modification;
4291 * this is why assert can't be put in ext4_find_extent()
4293 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4294 EXT4_ERROR_INODE(inode, "bad extent address "
4295 "lblock: %lu, depth: %d pblock %lld",
4296 (unsigned long) map->m_lblk, depth,
4297 path[depth].p_block);
4302 ex = path[depth].p_ext;
4304 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4305 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4306 unsigned short ee_len;
4310 * unwritten extents are treated as holes, except that
4311 * we split out initialized portions during a write.
4313 ee_len = ext4_ext_get_actual_len(ex);
4315 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4317 /* if found extent covers block, simply return it */
4318 if (in_range(map->m_lblk, ee_block, ee_len)) {
4319 newblock = map->m_lblk - ee_block + ee_start;
4320 /* number of remaining blocks in the extent */
4321 allocated = ee_len - (map->m_lblk - ee_block);
4322 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4323 ee_block, ee_len, newblock);
4326 * If the extent is initialized check whether the
4327 * caller wants to convert it to unwritten.
4329 if ((!ext4_ext_is_unwritten(ex)) &&
4330 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4331 allocated = convert_initialized_extent(
4332 handle, inode, map, &path,
4333 flags, allocated, newblock);
4335 } else if (!ext4_ext_is_unwritten(ex))
4338 ret = ext4_ext_handle_unwritten_extents(
4339 handle, inode, map, &path, flags,
4340 allocated, newblock);
4349 if ((sbi->s_cluster_ratio > 1) &&
4350 ext4_find_delalloc_cluster(inode, map->m_lblk))
4351 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4354 * requested block isn't allocated yet;
4355 * we couldn't try to create block if create flag is zero
4357 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4359 * put just found gap into cache to speed up
4360 * subsequent requests
4362 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4363 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4368 * Okay, we need to do block allocation.
4370 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4371 newex.ee_block = cpu_to_le32(map->m_lblk);
4372 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4375 * If we are doing bigalloc, check to see if the extent returned
4376 * by ext4_find_extent() implies a cluster we can use.
4378 if (cluster_offset && ex &&
4379 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4380 ar.len = allocated = map->m_len;
4381 newblock = map->m_pblk;
4382 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4383 goto got_allocated_blocks;
4386 /* find neighbour allocated blocks */
4387 ar.lleft = map->m_lblk;
4388 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4391 ar.lright = map->m_lblk;
4393 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4397 /* Check if the extent after searching to the right implies a
4398 * cluster we can use. */
4399 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4400 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4401 ar.len = allocated = map->m_len;
4402 newblock = map->m_pblk;
4403 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4404 goto got_allocated_blocks;
4408 * See if request is beyond maximum number of blocks we can have in
4409 * a single extent. For an initialized extent this limit is
4410 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4411 * EXT_UNWRITTEN_MAX_LEN.
4413 if (map->m_len > EXT_INIT_MAX_LEN &&
4414 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4415 map->m_len = EXT_INIT_MAX_LEN;
4416 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4417 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4418 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4420 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4421 newex.ee_len = cpu_to_le16(map->m_len);
4422 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4424 allocated = ext4_ext_get_actual_len(&newex);
4426 allocated = map->m_len;
4428 /* allocate new block */
4430 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4431 ar.logical = map->m_lblk;
4433 * We calculate the offset from the beginning of the cluster
4434 * for the logical block number, since when we allocate a
4435 * physical cluster, the physical block should start at the
4436 * same offset from the beginning of the cluster. This is
4437 * needed so that future calls to get_implied_cluster_alloc()
4440 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4441 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4443 ar.logical -= offset;
4444 if (S_ISREG(inode->i_mode))
4445 ar.flags = EXT4_MB_HINT_DATA;
4447 /* disable in-core preallocation for non-regular files */
4449 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4450 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4451 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4452 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4453 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4456 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4457 ar.goal, newblock, allocated);
4459 allocated_clusters = ar.len;
4460 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4461 if (ar.len > allocated)
4464 got_allocated_blocks:
4465 /* try to insert new extent into found leaf and return */
4466 ext4_ext_store_pblock(&newex, newblock + offset);
4467 newex.ee_len = cpu_to_le16(ar.len);
4468 /* Mark unwritten */
4469 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4470 ext4_ext_mark_unwritten(&newex);
4471 map->m_flags |= EXT4_MAP_UNWRITTEN;
4473 * io_end structure was created for every IO write to an
4474 * unwritten extent. To avoid unnecessary conversion,
4475 * here we flag the IO that really needs the conversion.
4476 * For non asycn direct IO case, flag the inode state
4477 * that we need to perform conversion when IO is done.
4479 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4484 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4485 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4488 err = ext4_ext_insert_extent(handle, inode, &path,
4491 if (!err && set_unwritten) {
4493 ext4_set_io_unwritten_flag(inode, io);
4495 ext4_set_inode_state(inode,
4496 EXT4_STATE_DIO_UNWRITTEN);
4499 if (err && free_on_err) {
4500 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4501 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4502 /* free data blocks we just allocated */
4503 /* not a good idea to call discard here directly,
4504 * but otherwise we'd need to call it every free() */
4505 ext4_discard_preallocations(inode);
4506 ext4_free_blocks(handle, inode, NULL, newblock,
4507 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4511 /* previous routine could use block we allocated */
4512 newblock = ext4_ext_pblock(&newex);
4513 allocated = ext4_ext_get_actual_len(&newex);
4514 if (allocated > map->m_len)
4515 allocated = map->m_len;
4516 map->m_flags |= EXT4_MAP_NEW;
4519 * Update reserved blocks/metadata blocks after successful
4520 * block allocation which had been deferred till now.
4522 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4523 unsigned int reserved_clusters;
4525 * Check how many clusters we had reserved this allocated range
4527 reserved_clusters = get_reserved_cluster_alloc(inode,
4528 map->m_lblk, allocated);
4529 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4530 if (reserved_clusters) {
4532 * We have clusters reserved for this range.
4533 * But since we are not doing actual allocation
4534 * and are simply using blocks from previously
4535 * allocated cluster, we should release the
4536 * reservation and not claim quota.
4538 ext4_da_update_reserve_space(inode,
4539 reserved_clusters, 0);
4542 BUG_ON(allocated_clusters < reserved_clusters);
4543 if (reserved_clusters < allocated_clusters) {
4544 struct ext4_inode_info *ei = EXT4_I(inode);
4545 int reservation = allocated_clusters -
4548 * It seems we claimed few clusters outside of
4549 * the range of this allocation. We should give
4550 * it back to the reservation pool. This can
4551 * happen in the following case:
4553 * * Suppose s_cluster_ratio is 4 (i.e., each
4554 * cluster has 4 blocks. Thus, the clusters
4555 * are [0-3],[4-7],[8-11]...
4556 * * First comes delayed allocation write for
4557 * logical blocks 10 & 11. Since there were no
4558 * previous delayed allocated blocks in the
4559 * range [8-11], we would reserve 1 cluster
4561 * * Next comes write for logical blocks 3 to 8.
4562 * In this case, we will reserve 2 clusters
4563 * (for [0-3] and [4-7]; and not for [8-11] as
4564 * that range has a delayed allocated blocks.
4565 * Thus total reserved clusters now becomes 3.
4566 * * Now, during the delayed allocation writeout
4567 * time, we will first write blocks [3-8] and
4568 * allocate 3 clusters for writing these
4569 * blocks. Also, we would claim all these
4570 * three clusters above.
4571 * * Now when we come here to writeout the
4572 * blocks [10-11], we would expect to claim
4573 * the reservation of 1 cluster we had made
4574 * (and we would claim it since there are no
4575 * more delayed allocated blocks in the range
4576 * [8-11]. But our reserved cluster count had
4577 * already gone to 0.
4579 * Thus, at the step 4 above when we determine
4580 * that there are still some unwritten delayed
4581 * allocated blocks outside of our current
4582 * block range, we should increment the
4583 * reserved clusters count so that when the
4584 * remaining blocks finally gets written, we
4587 dquot_reserve_block(inode,
4588 EXT4_C2B(sbi, reservation));
4589 spin_lock(&ei->i_block_reservation_lock);
4590 ei->i_reserved_data_blocks += reservation;
4591 spin_unlock(&ei->i_block_reservation_lock);
4594 * We will claim quota for all newly allocated blocks.
4595 * We're updating the reserved space *after* the
4596 * correction above so we do not accidentally free
4597 * all the metadata reservation because we might
4598 * actually need it later on.
4600 ext4_da_update_reserve_space(inode, allocated_clusters,
4606 * Cache the extent and update transaction to commit on fdatasync only
4607 * when it is _not_ an unwritten extent.
4609 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4610 ext4_update_inode_fsync_trans(handle, inode, 1);
4612 ext4_update_inode_fsync_trans(handle, inode, 0);
4614 if (allocated > map->m_len)
4615 allocated = map->m_len;
4616 ext4_ext_show_leaf(inode, path);
4617 map->m_flags |= EXT4_MAP_MAPPED;
4618 map->m_pblk = newblock;
4619 map->m_len = allocated;
4621 ext4_ext_drop_refs(path);
4624 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4625 err ? err : allocated);
4626 ext4_es_lru_add(inode);
4627 return err ? err : allocated;
4630 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4632 struct super_block *sb = inode->i_sb;
4633 ext4_lblk_t last_block;
4637 * TODO: optimization is possible here.
4638 * Probably we need not scan at all,
4639 * because page truncation is enough.
4642 /* we have to know where to truncate from in crash case */
4643 EXT4_I(inode)->i_disksize = inode->i_size;
4644 ext4_mark_inode_dirty(handle, inode);
4646 last_block = (inode->i_size + sb->s_blocksize - 1)
4647 >> EXT4_BLOCK_SIZE_BITS(sb);
4649 err = ext4_es_remove_extent(inode, last_block,
4650 EXT_MAX_BLOCKS - last_block);
4651 if (err == -ENOMEM) {
4653 congestion_wait(BLK_RW_ASYNC, HZ/50);
4657 ext4_std_error(inode->i_sb, err);
4660 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4661 ext4_std_error(inode->i_sb, err);
4664 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4665 ext4_lblk_t len, loff_t new_size,
4666 int flags, int mode)
4668 struct inode *inode = file_inode(file);
4673 struct ext4_map_blocks map;
4674 unsigned int credits;
4677 map.m_lblk = offset;
4680 * Don't normalize the request if it can fit in one extent so
4681 * that it doesn't get unnecessarily split into multiple
4684 if (len <= EXT_UNWRITTEN_MAX_LEN)
4685 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4688 * credits to insert 1 extent into extent tree
4690 credits = ext4_chunk_trans_blocks(inode, len);
4693 while (ret >= 0 && len) {
4694 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4696 if (IS_ERR(handle)) {
4697 ret = PTR_ERR(handle);
4700 ret = ext4_map_blocks(handle, inode, &map, flags);
4702 ext4_debug("inode #%lu: block %u: len %u: "
4703 "ext4_ext_map_blocks returned %d",
4704 inode->i_ino, map.m_lblk,
4706 ext4_mark_inode_dirty(handle, inode);
4707 ret2 = ext4_journal_stop(handle);
4711 map.m_len = len = len - ret;
4712 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4713 inode->i_ctime = ext4_current_time(inode);
4715 if (epos > new_size)
4717 if (ext4_update_inode_size(inode, epos) & 0x1)
4718 inode->i_mtime = inode->i_ctime;
4720 if (epos > inode->i_size)
4721 ext4_set_inode_flag(inode,
4722 EXT4_INODE_EOFBLOCKS);
4724 ext4_mark_inode_dirty(handle, inode);
4725 ret2 = ext4_journal_stop(handle);
4729 if (ret == -ENOSPC &&
4730 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4735 return ret > 0 ? ret2 : ret;
4738 static long ext4_zero_range(struct file *file, loff_t offset,
4739 loff_t len, int mode)
4741 struct inode *inode = file_inode(file);
4742 handle_t *handle = NULL;
4743 unsigned int max_blocks;
4744 loff_t new_size = 0;
4748 int partial_begin, partial_end;
4751 struct address_space *mapping = inode->i_mapping;
4752 unsigned int blkbits = inode->i_blkbits;
4754 trace_ext4_zero_range(inode, offset, len, mode);
4756 if (!S_ISREG(inode->i_mode))
4759 /* Call ext4_force_commit to flush all data in case of data=journal. */
4760 if (ext4_should_journal_data(inode)) {
4761 ret = ext4_force_commit(inode->i_sb);
4767 * Write out all dirty pages to avoid race conditions
4768 * Then release them.
4770 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4771 ret = filemap_write_and_wait_range(mapping, offset,
4778 * Round up offset. This is not fallocate, we neet to zero out
4779 * blocks, so convert interior block aligned part of the range to
4780 * unwritten and possibly manually zero out unaligned parts of the
4783 start = round_up(offset, 1 << blkbits);
4784 end = round_down((offset + len), 1 << blkbits);
4786 if (start < offset || end > offset + len)
4788 partial_begin = offset & ((1 << blkbits) - 1);
4789 partial_end = (offset + len) & ((1 << blkbits) - 1);
4791 lblk = start >> blkbits;
4792 max_blocks = (end >> blkbits);
4793 if (max_blocks < lblk)
4798 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4799 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4801 if (mode & FALLOC_FL_KEEP_SIZE)
4802 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4804 mutex_lock(&inode->i_mutex);
4807 * Indirect files do not support unwritten extnets
4809 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4814 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4815 offset + len > i_size_read(inode)) {
4816 new_size = offset + len;
4817 ret = inode_newsize_ok(inode, new_size);
4821 * If we have a partial block after EOF we have to allocate
4828 if (max_blocks > 0) {
4830 /* Now release the pages and zero block aligned part of pages*/
4831 truncate_pagecache_range(inode, start, end - 1);
4832 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4834 /* Wait all existing dio workers, newcomers will block on i_mutex */
4835 ext4_inode_block_unlocked_dio(inode);
4836 inode_dio_wait(inode);
4838 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4843 * Remove entire range from the extent status tree.
4845 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4846 * NOT sufficient. I'm not sure why this is the case,
4847 * but let's be conservative and remove the extent
4848 * status tree for the entire inode. There should be
4849 * no outstanding delalloc extents thanks to the
4850 * filemap_write_and_wait_range() call above.
4852 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4856 if (!partial_begin && !partial_end)
4860 * In worst case we have to writeout two nonadjacent unwritten
4861 * blocks and update the inode
4863 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4864 if (ext4_should_journal_data(inode))
4866 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4867 if (IS_ERR(handle)) {
4868 ret = PTR_ERR(handle);
4869 ext4_std_error(inode->i_sb, ret);
4873 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4875 ext4_update_inode_size(inode, new_size);
4878 * Mark that we allocate beyond EOF so the subsequent truncate
4879 * can proceed even if the new size is the same as i_size.
4881 if ((offset + len) > i_size_read(inode))
4882 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4884 ext4_mark_inode_dirty(handle, inode);
4886 /* Zero out partial block at the edges of the range */
4887 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4889 if (file->f_flags & O_SYNC)
4890 ext4_handle_sync(handle);
4892 ext4_journal_stop(handle);
4894 ext4_inode_resume_unlocked_dio(inode);
4896 mutex_unlock(&inode->i_mutex);
4901 * preallocate space for a file. This implements ext4's fallocate file
4902 * operation, which gets called from sys_fallocate system call.
4903 * For block-mapped files, posix_fallocate should fall back to the method
4904 * of writing zeroes to the required new blocks (the same behavior which is
4905 * expected for file systems which do not support fallocate() system call).
4907 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4909 struct inode *inode = file_inode(file);
4910 loff_t new_size = 0;
4911 unsigned int max_blocks;
4915 unsigned int blkbits = inode->i_blkbits;
4917 /* Return error if mode is not supported */
4918 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4919 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4922 if (mode & FALLOC_FL_PUNCH_HOLE)
4923 return ext4_punch_hole(inode, offset, len);
4925 ret = ext4_convert_inline_data(inode);
4930 * currently supporting (pre)allocate mode for extent-based
4933 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4936 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4937 return ext4_collapse_range(inode, offset, len);
4939 if (mode & FALLOC_FL_ZERO_RANGE)
4940 return ext4_zero_range(file, offset, len, mode);
4942 trace_ext4_fallocate_enter(inode, offset, len, mode);
4943 lblk = offset >> blkbits;
4945 * We can't just convert len to max_blocks because
4946 * If blocksize = 4096 offset = 3072 and len = 2048
4948 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4951 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4952 if (mode & FALLOC_FL_KEEP_SIZE)
4953 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4955 mutex_lock(&inode->i_mutex);
4957 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4958 offset + len > i_size_read(inode)) {
4959 new_size = offset + len;
4960 ret = inode_newsize_ok(inode, new_size);
4965 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4970 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4971 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4972 EXT4_I(inode)->i_sync_tid);
4975 mutex_unlock(&inode->i_mutex);
4976 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4981 * This function convert a range of blocks to written extents
4982 * The caller of this function will pass the start offset and the size.
4983 * all unwritten extents within this range will be converted to
4986 * This function is called from the direct IO end io call back
4987 * function, to convert the fallocated extents after IO is completed.
4988 * Returns 0 on success.
4990 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4991 loff_t offset, ssize_t len)
4993 unsigned int max_blocks;
4996 struct ext4_map_blocks map;
4997 unsigned int credits, blkbits = inode->i_blkbits;
4999 map.m_lblk = offset >> blkbits;
5001 * We can't just convert len to max_blocks because
5002 * If blocksize = 4096 offset = 3072 and len = 2048
5004 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5007 * This is somewhat ugly but the idea is clear: When transaction is
5008 * reserved, everything goes into it. Otherwise we rather start several
5009 * smaller transactions for conversion of each extent separately.
5012 handle = ext4_journal_start_reserved(handle,
5013 EXT4_HT_EXT_CONVERT);
5015 return PTR_ERR(handle);
5019 * credits to insert 1 extent into extent tree
5021 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5023 while (ret >= 0 && ret < max_blocks) {
5025 map.m_len = (max_blocks -= ret);
5027 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5029 if (IS_ERR(handle)) {
5030 ret = PTR_ERR(handle);
5034 ret = ext4_map_blocks(handle, inode, &map,
5035 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5037 ext4_warning(inode->i_sb,
5038 "inode #%lu: block %u: len %u: "
5039 "ext4_ext_map_blocks returned %d",
5040 inode->i_ino, map.m_lblk,
5042 ext4_mark_inode_dirty(handle, inode);
5044 ret2 = ext4_journal_stop(handle);
5045 if (ret <= 0 || ret2)
5049 ret2 = ext4_journal_stop(handle);
5050 return ret > 0 ? ret2 : ret;
5054 * If newes is not existing extent (newes->ec_pblk equals zero) find
5055 * delayed extent at start of newes and update newes accordingly and
5056 * return start of the next delayed extent.
5058 * If newes is existing extent (newes->ec_pblk is not equal zero)
5059 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5060 * extent found. Leave newes unmodified.
5062 static int ext4_find_delayed_extent(struct inode *inode,
5063 struct extent_status *newes)
5065 struct extent_status es;
5066 ext4_lblk_t block, next_del;
5068 if (newes->es_pblk == 0) {
5069 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5070 newes->es_lblk + newes->es_len - 1, &es);
5073 * No extent in extent-tree contains block @newes->es_pblk,
5074 * then the block may stay in 1)a hole or 2)delayed-extent.
5080 if (es.es_lblk > newes->es_lblk) {
5082 newes->es_len = min(es.es_lblk - newes->es_lblk,
5087 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5090 block = newes->es_lblk + newes->es_len;
5091 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5093 next_del = EXT_MAX_BLOCKS;
5095 next_del = es.es_lblk;
5099 /* fiemap flags we can handle specified here */
5100 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5102 static int ext4_xattr_fiemap(struct inode *inode,
5103 struct fiemap_extent_info *fieinfo)
5107 __u32 flags = FIEMAP_EXTENT_LAST;
5108 int blockbits = inode->i_sb->s_blocksize_bits;
5112 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5113 struct ext4_iloc iloc;
5114 int offset; /* offset of xattr in inode */
5116 error = ext4_get_inode_loc(inode, &iloc);
5119 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5120 offset = EXT4_GOOD_OLD_INODE_SIZE +
5121 EXT4_I(inode)->i_extra_isize;
5123 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5124 flags |= FIEMAP_EXTENT_DATA_INLINE;
5126 } else { /* external block */
5127 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5128 length = inode->i_sb->s_blocksize;
5132 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5134 return (error < 0 ? error : 0);
5137 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5138 __u64 start, __u64 len)
5140 ext4_lblk_t start_blk;
5143 if (ext4_has_inline_data(inode)) {
5146 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5152 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5153 error = ext4_ext_precache(inode);
5158 /* fallback to generic here if not in extents fmt */
5159 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5160 return generic_block_fiemap(inode, fieinfo, start, len,
5163 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5166 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5167 error = ext4_xattr_fiemap(inode, fieinfo);
5169 ext4_lblk_t len_blks;
5172 start_blk = start >> inode->i_sb->s_blocksize_bits;
5173 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5174 if (last_blk >= EXT_MAX_BLOCKS)
5175 last_blk = EXT_MAX_BLOCKS-1;
5176 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5179 * Walk the extent tree gathering extent information
5180 * and pushing extents back to the user.
5182 error = ext4_fill_fiemap_extents(inode, start_blk,
5185 ext4_es_lru_add(inode);
5191 * Function to access the path buffer for marking it dirty.
5192 * It also checks if there are sufficient credits left in the journal handle
5196 ext4_access_path(handle_t *handle, struct inode *inode,
5197 struct ext4_ext_path *path)
5201 if (!ext4_handle_valid(handle))
5205 * Check if need to extend journal credits
5206 * 3 for leaf, sb, and inode plus 2 (bmap and group
5207 * descriptor) for each block group; assume two block
5210 if (handle->h_buffer_credits < 7) {
5211 credits = ext4_writepage_trans_blocks(inode);
5212 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5213 /* EAGAIN is success */
5214 if (err && err != -EAGAIN)
5218 err = ext4_ext_get_access(handle, inode, path);
5223 * ext4_ext_shift_path_extents:
5224 * Shift the extents of a path structure lying between path[depth].p_ext
5225 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5226 * from starting block for each extent.
5229 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5230 struct inode *inode, handle_t *handle,
5234 struct ext4_extent *ex_start, *ex_last;
5236 depth = path->p_depth;
5238 while (depth >= 0) {
5239 if (depth == path->p_depth) {
5240 ex_start = path[depth].p_ext;
5244 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5248 err = ext4_access_path(handle, inode, path + depth);
5252 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5255 *start = le32_to_cpu(ex_last->ee_block) +
5256 ext4_ext_get_actual_len(ex_last);
5258 while (ex_start <= ex_last) {
5259 le32_add_cpu(&ex_start->ee_block, -shift);
5260 /* Try to merge to the left. */
5262 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5263 ext4_ext_try_to_merge_right(inode,
5264 path, ex_start - 1))
5269 err = ext4_ext_dirty(handle, inode, path + depth);
5273 if (--depth < 0 || !update)
5277 /* Update index too */
5278 err = ext4_access_path(handle, inode, path + depth);
5282 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5283 err = ext4_ext_dirty(handle, inode, path + depth);
5287 /* we are done if current index is not a starting index */
5288 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5299 * ext4_ext_shift_extents:
5300 * All the extents which lies in the range from start to the last allocated
5301 * block for the file are shifted downwards by shift blocks.
5302 * On success, 0 is returned, error otherwise.
5305 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5306 ext4_lblk_t start, ext4_lblk_t shift)
5308 struct ext4_ext_path *path;
5310 struct ext4_extent *extent;
5311 ext4_lblk_t stop_block;
5312 ext4_lblk_t ex_start, ex_end;
5314 /* Let path point to the last extent */
5315 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5317 return PTR_ERR(path);
5319 depth = path->p_depth;
5320 extent = path[depth].p_ext;
5324 stop_block = le32_to_cpu(extent->ee_block) +
5325 ext4_ext_get_actual_len(extent);
5327 /* Nothing to shift, if hole is at the end of file */
5328 if (start >= stop_block)
5332 * Don't start shifting extents until we make sure the hole is big
5333 * enough to accomodate the shift.
5335 path = ext4_find_extent(inode, start - 1, &path, 0);
5337 return PTR_ERR(path);
5338 depth = path->p_depth;
5339 extent = path[depth].p_ext;
5341 ex_start = le32_to_cpu(extent->ee_block);
5342 ex_end = le32_to_cpu(extent->ee_block) +
5343 ext4_ext_get_actual_len(extent);
5349 if ((start == ex_start && shift > ex_start) ||
5350 (shift > start - ex_end))
5353 /* Its safe to start updating extents */
5354 while (start < stop_block) {
5355 path = ext4_find_extent(inode, start, &path, 0);
5357 return PTR_ERR(path);
5358 depth = path->p_depth;
5359 extent = path[depth].p_ext;
5361 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5362 (unsigned long) start);
5365 if (start > le32_to_cpu(extent->ee_block)) {
5366 /* Hole, move to the next extent */
5367 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5368 path[depth].p_ext++;
5370 start = ext4_ext_next_allocated_block(path);
5374 ret = ext4_ext_shift_path_extents(path, shift, inode,
5380 ext4_ext_drop_refs(path);
5386 * ext4_collapse_range:
5387 * This implements the fallocate's collapse range functionality for ext4
5388 * Returns: 0 and non-zero on error.
5390 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5392 struct super_block *sb = inode->i_sb;
5393 ext4_lblk_t punch_start, punch_stop;
5395 unsigned int credits;
5396 loff_t new_size, ioffset;
5399 /* Collapse range works only on fs block size aligned offsets. */
5400 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5401 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5404 if (!S_ISREG(inode->i_mode))
5407 trace_ext4_collapse_range(inode, offset, len);
5409 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5410 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5412 /* Call ext4_force_commit to flush all data in case of data=journal. */
5413 if (ext4_should_journal_data(inode)) {
5414 ret = ext4_force_commit(inode->i_sb);
5420 * Need to round down offset to be aligned with page size boundary
5421 * for page size > block size.
5423 ioffset = round_down(offset, PAGE_SIZE);
5425 /* Write out all dirty pages */
5426 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5431 /* Take mutex lock */
5432 mutex_lock(&inode->i_mutex);
5435 * There is no need to overlap collapse range with EOF, in which case
5436 * it is effectively a truncate operation
5438 if (offset + len >= i_size_read(inode)) {
5443 /* Currently just for extent based files */
5444 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5449 truncate_pagecache(inode, ioffset);
5451 /* Wait for existing dio to complete */
5452 ext4_inode_block_unlocked_dio(inode);
5453 inode_dio_wait(inode);
5455 credits = ext4_writepage_trans_blocks(inode);
5456 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5457 if (IS_ERR(handle)) {
5458 ret = PTR_ERR(handle);
5462 down_write(&EXT4_I(inode)->i_data_sem);
5463 ext4_discard_preallocations(inode);
5465 ret = ext4_es_remove_extent(inode, punch_start,
5466 EXT_MAX_BLOCKS - punch_start);
5468 up_write(&EXT4_I(inode)->i_data_sem);
5472 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5474 up_write(&EXT4_I(inode)->i_data_sem);
5477 ext4_discard_preallocations(inode);
5479 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5480 punch_stop - punch_start);
5482 up_write(&EXT4_I(inode)->i_data_sem);
5486 new_size = i_size_read(inode) - len;
5487 i_size_write(inode, new_size);
5488 EXT4_I(inode)->i_disksize = new_size;
5490 up_write(&EXT4_I(inode)->i_data_sem);
5492 ext4_handle_sync(handle);
5493 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5494 ext4_mark_inode_dirty(handle, inode);
5497 ext4_journal_stop(handle);
5499 ext4_inode_resume_unlocked_dio(inode);
5501 mutex_unlock(&inode->i_mutex);
5506 * ext4_swap_extents - Swap extents between two inodes
5508 * @inode1: First inode
5509 * @inode2: Second inode
5510 * @lblk1: Start block for first inode
5511 * @lblk2: Start block for second inode
5512 * @count: Number of blocks to swap
5513 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5514 * @erp: Pointer to save error value
5516 * This helper routine does exactly what is promise "swap extents". All other
5517 * stuff such as page-cache locking consistency, bh mapping consistency or
5518 * extent's data copying must be performed by caller.
5520 * i_mutex is held for both inodes
5521 * i_data_sem is locked for write for both inodes
5523 * All pages from requested range are locked for both inodes
5526 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5527 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5528 ext4_lblk_t count, int unwritten, int *erp)
5530 struct ext4_ext_path *path1 = NULL;
5531 struct ext4_ext_path *path2 = NULL;
5532 int replaced_count = 0;
5534 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5535 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5536 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5537 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5539 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5542 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5547 struct ext4_extent *ex1, *ex2, tmp_ex;
5548 ext4_lblk_t e1_blk, e2_blk;
5549 int e1_len, e2_len, len;
5552 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5553 if (unlikely(IS_ERR(path1))) {
5554 *erp = PTR_ERR(path1);
5560 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5561 if (unlikely(IS_ERR(path2))) {
5562 *erp = PTR_ERR(path2);
5566 ex1 = path1[path1->p_depth].p_ext;
5567 ex2 = path2[path2->p_depth].p_ext;
5568 /* Do we have somthing to swap ? */
5569 if (unlikely(!ex2 || !ex1))
5572 e1_blk = le32_to_cpu(ex1->ee_block);
5573 e2_blk = le32_to_cpu(ex2->ee_block);
5574 e1_len = ext4_ext_get_actual_len(ex1);
5575 e2_len = ext4_ext_get_actual_len(ex2);
5578 if (!in_range(lblk1, e1_blk, e1_len) ||
5579 !in_range(lblk2, e2_blk, e2_len)) {
5580 ext4_lblk_t next1, next2;
5582 /* if hole after extent, then go to next extent */
5583 next1 = ext4_ext_next_allocated_block(path1);
5584 next2 = ext4_ext_next_allocated_block(path2);
5585 /* If hole before extent, then shift to that extent */
5590 /* Do we have something to swap */
5591 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5593 /* Move to the rightest boundary */
5594 len = next1 - lblk1;
5595 if (len < next2 - lblk2)
5596 len = next2 - lblk2;
5605 /* Prepare left boundary */
5606 if (e1_blk < lblk1) {
5608 *erp = ext4_force_split_extent_at(handle, inode1,
5613 if (e2_blk < lblk2) {
5615 *erp = ext4_force_split_extent_at(handle, inode2,
5620 /* ext4_split_extent_at() may result in leaf extent split,
5621 * path must to be revalidated. */
5625 /* Prepare right boundary */
5627 if (len > e1_blk + e1_len - lblk1)
5628 len = e1_blk + e1_len - lblk1;
5629 if (len > e2_blk + e2_len - lblk2)
5630 len = e2_blk + e2_len - lblk2;
5632 if (len != e1_len) {
5634 *erp = ext4_force_split_extent_at(handle, inode1,
5635 &path1, lblk1 + len, 0);
5639 if (len != e2_len) {
5641 *erp = ext4_force_split_extent_at(handle, inode2,
5642 &path2, lblk2 + len, 0);
5646 /* ext4_split_extent_at() may result in leaf extent split,
5647 * path must to be revalidated. */
5651 BUG_ON(e2_len != e1_len);
5652 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5655 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5659 /* Both extents are fully inside boundaries. Swap it now */
5661 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5662 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5663 ex1->ee_len = cpu_to_le16(e2_len);
5664 ex2->ee_len = cpu_to_le16(e1_len);
5666 ext4_ext_mark_unwritten(ex2);
5667 if (ext4_ext_is_unwritten(&tmp_ex))
5668 ext4_ext_mark_unwritten(ex1);
5670 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5671 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5672 *erp = ext4_ext_dirty(handle, inode2, path2 +
5676 *erp = ext4_ext_dirty(handle, inode1, path1 +
5679 * Looks scarry ah..? second inode already points to new blocks,
5680 * and it was successfully dirtied. But luckily error may happen
5681 * only due to journal error, so full transaction will be
5688 replaced_count += len;
5692 ext4_ext_drop_refs(path1);
5694 ext4_ext_drop_refs(path2);
5696 path1 = path2 = NULL;
5698 return replaced_count;