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 <linux/falloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/fiemap.h>
43 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
48 * used by extent splitting.
50 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
52 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
53 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
55 static __le32 ext4_extent_block_csum(struct inode *inode,
56 struct ext4_extent_header *eh)
58 struct ext4_inode_info *ei = EXT4_I(inode);
59 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
63 EXT4_EXTENT_TAIL_OFFSET(eh));
64 return cpu_to_le32(csum);
67 static int ext4_extent_block_csum_verify(struct inode *inode,
68 struct ext4_extent_header *eh)
70 struct ext4_extent_tail *et;
72 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
73 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
76 et = find_ext4_extent_tail(eh);
77 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
82 static void ext4_extent_block_csum_set(struct inode *inode,
83 struct ext4_extent_header *eh)
85 struct ext4_extent_tail *et;
87 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
88 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
91 et = find_ext4_extent_tail(eh);
92 et->et_checksum = ext4_extent_block_csum(inode, eh);
95 static int ext4_split_extent(handle_t *handle,
97 struct ext4_ext_path *path,
98 struct ext4_map_blocks *map,
102 static int ext4_split_extent_at(handle_t *handle,
104 struct ext4_ext_path *path,
109 static int ext4_ext_truncate_extend_restart(handle_t *handle,
115 if (!ext4_handle_valid(handle))
117 if (handle->h_buffer_credits > needed)
119 err = ext4_journal_extend(handle, needed);
122 err = ext4_truncate_restart_trans(handle, inode, needed);
134 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
135 struct ext4_ext_path *path)
138 /* path points to block */
139 return ext4_journal_get_write_access(handle, path->p_bh);
141 /* path points to leaf/index in inode body */
142 /* we use in-core data, no need to protect them */
152 #define ext4_ext_dirty(handle, inode, path) \
153 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
154 static int __ext4_ext_dirty(const char *where, unsigned int line,
155 handle_t *handle, struct inode *inode,
156 struct ext4_ext_path *path)
160 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
161 /* path points to block */
162 err = __ext4_handle_dirty_metadata(where, line, handle,
165 /* path points to leaf/index in inode body */
166 err = ext4_mark_inode_dirty(handle, inode);
171 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
172 struct ext4_ext_path *path,
176 int depth = path->p_depth;
177 struct ext4_extent *ex;
180 * Try to predict block placement assuming that we are
181 * filling in a file which will eventually be
182 * non-sparse --- i.e., in the case of libbfd writing
183 * an ELF object sections out-of-order but in a way
184 * the eventually results in a contiguous object or
185 * executable file, or some database extending a table
186 * space file. However, this is actually somewhat
187 * non-ideal if we are writing a sparse file such as
188 * qemu or KVM writing a raw image file that is going
189 * to stay fairly sparse, since it will end up
190 * fragmenting the file system's free space. Maybe we
191 * should have some hueristics or some way to allow
192 * userspace to pass a hint to file system,
193 * especially if the latter case turns out to be
196 ex = path[depth].p_ext;
198 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
199 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
201 if (block > ext_block)
202 return ext_pblk + (block - ext_block);
204 return ext_pblk - (ext_block - block);
207 /* it looks like index is empty;
208 * try to find starting block from index itself */
209 if (path[depth].p_bh)
210 return path[depth].p_bh->b_blocknr;
213 /* OK. use inode's group */
214 return ext4_inode_to_goal_block(inode);
218 * Allocation for a meta data block
221 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
222 struct ext4_ext_path *path,
223 struct ext4_extent *ex, int *err, unsigned int flags)
225 ext4_fsblk_t goal, newblock;
227 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
228 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
233 static inline int ext4_ext_space_block(struct inode *inode, int check)
237 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
238 / sizeof(struct ext4_extent);
239 #ifdef AGGRESSIVE_TEST
240 if (!check && size > 6)
246 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
250 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
251 / sizeof(struct ext4_extent_idx);
252 #ifdef AGGRESSIVE_TEST
253 if (!check && size > 5)
259 static inline int ext4_ext_space_root(struct inode *inode, int check)
263 size = sizeof(EXT4_I(inode)->i_data);
264 size -= sizeof(struct ext4_extent_header);
265 size /= sizeof(struct ext4_extent);
266 #ifdef AGGRESSIVE_TEST
267 if (!check && size > 3)
273 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
277 size = sizeof(EXT4_I(inode)->i_data);
278 size -= sizeof(struct ext4_extent_header);
279 size /= sizeof(struct ext4_extent_idx);
280 #ifdef AGGRESSIVE_TEST
281 if (!check && size > 4)
288 * Calculate the number of metadata blocks needed
289 * to allocate @blocks
290 * Worse case is one block per extent
292 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
294 struct ext4_inode_info *ei = EXT4_I(inode);
297 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
298 / sizeof(struct ext4_extent_idx));
301 * If the new delayed allocation block is contiguous with the
302 * previous da block, it can share index blocks with the
303 * previous block, so we only need to allocate a new index
304 * block every idxs leaf blocks. At ldxs**2 blocks, we need
305 * an additional index block, and at ldxs**3 blocks, yet
306 * another index blocks.
308 if (ei->i_da_metadata_calc_len &&
309 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
312 if ((ei->i_da_metadata_calc_len % idxs) == 0)
314 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
316 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
318 ei->i_da_metadata_calc_len = 0;
320 ei->i_da_metadata_calc_len++;
321 ei->i_da_metadata_calc_last_lblock++;
326 * In the worst case we need a new set of index blocks at
327 * every level of the inode's extent tree.
329 ei->i_da_metadata_calc_len = 1;
330 ei->i_da_metadata_calc_last_lblock = lblock;
331 return ext_depth(inode) + 1;
335 ext4_ext_max_entries(struct inode *inode, int depth)
339 if (depth == ext_depth(inode)) {
341 max = ext4_ext_space_root(inode, 1);
343 max = ext4_ext_space_root_idx(inode, 1);
346 max = ext4_ext_space_block(inode, 1);
348 max = ext4_ext_space_block_idx(inode, 1);
354 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
356 ext4_fsblk_t block = ext4_ext_pblock(ext);
357 int len = ext4_ext_get_actual_len(ext);
361 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
364 static int ext4_valid_extent_idx(struct inode *inode,
365 struct ext4_extent_idx *ext_idx)
367 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
369 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
372 static int ext4_valid_extent_entries(struct inode *inode,
373 struct ext4_extent_header *eh,
376 unsigned short entries;
377 if (eh->eh_entries == 0)
380 entries = le16_to_cpu(eh->eh_entries);
384 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
386 if (!ext4_valid_extent(inode, ext))
392 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
394 if (!ext4_valid_extent_idx(inode, ext_idx))
403 static int __ext4_ext_check(const char *function, unsigned int line,
404 struct inode *inode, struct ext4_extent_header *eh,
407 const char *error_msg;
410 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
411 error_msg = "invalid magic";
414 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
415 error_msg = "unexpected eh_depth";
418 if (unlikely(eh->eh_max == 0)) {
419 error_msg = "invalid eh_max";
422 max = ext4_ext_max_entries(inode, depth);
423 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
424 error_msg = "too large eh_max";
427 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
428 error_msg = "invalid eh_entries";
431 if (!ext4_valid_extent_entries(inode, eh, depth)) {
432 error_msg = "invalid extent entries";
435 /* Verify checksum on non-root extent tree nodes */
436 if (ext_depth(inode) != depth &&
437 !ext4_extent_block_csum_verify(inode, eh)) {
438 error_msg = "extent tree corrupted";
444 ext4_error_inode(inode, function, line, 0,
445 "bad header/extent: %s - magic %x, "
446 "entries %u, max %u(%u), depth %u(%u)",
447 error_msg, le16_to_cpu(eh->eh_magic),
448 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
449 max, le16_to_cpu(eh->eh_depth), depth);
454 #define ext4_ext_check(inode, eh, depth) \
455 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
457 int ext4_ext_check_inode(struct inode *inode)
459 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
462 static int __ext4_ext_check_block(const char *function, unsigned int line,
464 struct ext4_extent_header *eh,
466 struct buffer_head *bh)
470 if (buffer_verified(bh))
472 ret = ext4_ext_check(inode, eh, depth);
475 set_buffer_verified(bh);
479 #define ext4_ext_check_block(inode, eh, depth, bh) \
480 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
483 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
485 int k, l = path->p_depth;
488 for (k = 0; k <= l; k++, path++) {
490 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
491 ext4_idx_pblock(path->p_idx));
492 } else if (path->p_ext) {
493 ext_debug(" %d:[%d]%d:%llu ",
494 le32_to_cpu(path->p_ext->ee_block),
495 ext4_ext_is_uninitialized(path->p_ext),
496 ext4_ext_get_actual_len(path->p_ext),
497 ext4_ext_pblock(path->p_ext));
504 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
506 int depth = ext_depth(inode);
507 struct ext4_extent_header *eh;
508 struct ext4_extent *ex;
514 eh = path[depth].p_hdr;
515 ex = EXT_FIRST_EXTENT(eh);
517 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
519 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
520 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
521 ext4_ext_is_uninitialized(ex),
522 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
527 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
528 ext4_fsblk_t newblock, int level)
530 int depth = ext_depth(inode);
531 struct ext4_extent *ex;
533 if (depth != level) {
534 struct ext4_extent_idx *idx;
535 idx = path[level].p_idx;
536 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
537 ext_debug("%d: move %d:%llu in new index %llu\n", level,
538 le32_to_cpu(idx->ei_block),
539 ext4_idx_pblock(idx),
547 ex = path[depth].p_ext;
548 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
549 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
550 le32_to_cpu(ex->ee_block),
552 ext4_ext_is_uninitialized(ex),
553 ext4_ext_get_actual_len(ex),
560 #define ext4_ext_show_path(inode, path)
561 #define ext4_ext_show_leaf(inode, path)
562 #define ext4_ext_show_move(inode, path, newblock, level)
565 void ext4_ext_drop_refs(struct ext4_ext_path *path)
567 int depth = path->p_depth;
570 for (i = 0; i <= depth; i++, path++)
578 * ext4_ext_binsearch_idx:
579 * binary search for the closest index of the given block
580 * the header must be checked before calling this
583 ext4_ext_binsearch_idx(struct inode *inode,
584 struct ext4_ext_path *path, ext4_lblk_t block)
586 struct ext4_extent_header *eh = path->p_hdr;
587 struct ext4_extent_idx *r, *l, *m;
590 ext_debug("binsearch for %u(idx): ", block);
592 l = EXT_FIRST_INDEX(eh) + 1;
593 r = EXT_LAST_INDEX(eh);
596 if (block < le32_to_cpu(m->ei_block))
600 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
601 m, le32_to_cpu(m->ei_block),
602 r, le32_to_cpu(r->ei_block));
606 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
607 ext4_idx_pblock(path->p_idx));
609 #ifdef CHECK_BINSEARCH
611 struct ext4_extent_idx *chix, *ix;
614 chix = ix = EXT_FIRST_INDEX(eh);
615 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
617 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
618 printk(KERN_DEBUG "k=%d, ix=0x%p, "
620 ix, EXT_FIRST_INDEX(eh));
621 printk(KERN_DEBUG "%u <= %u\n",
622 le32_to_cpu(ix->ei_block),
623 le32_to_cpu(ix[-1].ei_block));
625 BUG_ON(k && le32_to_cpu(ix->ei_block)
626 <= le32_to_cpu(ix[-1].ei_block));
627 if (block < le32_to_cpu(ix->ei_block))
631 BUG_ON(chix != path->p_idx);
638 * ext4_ext_binsearch:
639 * binary search for closest extent of the given block
640 * the header must be checked before calling this
643 ext4_ext_binsearch(struct inode *inode,
644 struct ext4_ext_path *path, ext4_lblk_t block)
646 struct ext4_extent_header *eh = path->p_hdr;
647 struct ext4_extent *r, *l, *m;
649 if (eh->eh_entries == 0) {
651 * this leaf is empty:
652 * we get such a leaf in split/add case
657 ext_debug("binsearch for %u: ", block);
659 l = EXT_FIRST_EXTENT(eh) + 1;
660 r = EXT_LAST_EXTENT(eh);
664 if (block < le32_to_cpu(m->ee_block))
668 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
669 m, le32_to_cpu(m->ee_block),
670 r, le32_to_cpu(r->ee_block));
674 ext_debug(" -> %d:%llu:[%d]%d ",
675 le32_to_cpu(path->p_ext->ee_block),
676 ext4_ext_pblock(path->p_ext),
677 ext4_ext_is_uninitialized(path->p_ext),
678 ext4_ext_get_actual_len(path->p_ext));
680 #ifdef CHECK_BINSEARCH
682 struct ext4_extent *chex, *ex;
685 chex = ex = EXT_FIRST_EXTENT(eh);
686 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
687 BUG_ON(k && le32_to_cpu(ex->ee_block)
688 <= le32_to_cpu(ex[-1].ee_block));
689 if (block < le32_to_cpu(ex->ee_block))
693 BUG_ON(chex != path->p_ext);
699 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
701 struct ext4_extent_header *eh;
703 eh = ext_inode_hdr(inode);
706 eh->eh_magic = EXT4_EXT_MAGIC;
707 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
708 ext4_mark_inode_dirty(handle, inode);
709 ext4_ext_invalidate_cache(inode);
713 struct ext4_ext_path *
714 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
715 struct ext4_ext_path *path)
717 struct ext4_extent_header *eh;
718 struct buffer_head *bh;
719 short int depth, i, ppos = 0, alloc = 0;
721 eh = ext_inode_hdr(inode);
722 depth = ext_depth(inode);
724 /* account possible depth increase */
726 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
729 return ERR_PTR(-ENOMEM);
736 /* walk through the tree */
738 ext_debug("depth %d: num %d, max %d\n",
739 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
741 ext4_ext_binsearch_idx(inode, path + ppos, block);
742 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
743 path[ppos].p_depth = i;
744 path[ppos].p_ext = NULL;
746 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
749 if (!bh_uptodate_or_lock(bh)) {
750 trace_ext4_ext_load_extent(inode, block,
752 if (bh_submit_read(bh) < 0) {
757 eh = ext_block_hdr(bh);
759 if (unlikely(ppos > depth)) {
761 EXT4_ERROR_INODE(inode,
762 "ppos %d > depth %d", ppos, depth);
765 path[ppos].p_bh = bh;
766 path[ppos].p_hdr = eh;
769 if (ext4_ext_check_block(inode, eh, i, bh))
773 path[ppos].p_depth = i;
774 path[ppos].p_ext = NULL;
775 path[ppos].p_idx = NULL;
778 ext4_ext_binsearch(inode, path + ppos, block);
779 /* if not an empty leaf */
780 if (path[ppos].p_ext)
781 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
783 ext4_ext_show_path(inode, path);
788 ext4_ext_drop_refs(path);
791 return ERR_PTR(-EIO);
795 * ext4_ext_insert_index:
796 * insert new index [@logical;@ptr] into the block at @curp;
797 * check where to insert: before @curp or after @curp
799 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
800 struct ext4_ext_path *curp,
801 int logical, ext4_fsblk_t ptr)
803 struct ext4_extent_idx *ix;
806 err = ext4_ext_get_access(handle, inode, curp);
810 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
811 EXT4_ERROR_INODE(inode,
812 "logical %d == ei_block %d!",
813 logical, le32_to_cpu(curp->p_idx->ei_block));
817 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
818 >= le16_to_cpu(curp->p_hdr->eh_max))) {
819 EXT4_ERROR_INODE(inode,
820 "eh_entries %d >= eh_max %d!",
821 le16_to_cpu(curp->p_hdr->eh_entries),
822 le16_to_cpu(curp->p_hdr->eh_max));
826 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
828 ext_debug("insert new index %d after: %llu\n", logical, ptr);
829 ix = curp->p_idx + 1;
832 ext_debug("insert new index %d before: %llu\n", logical, ptr);
836 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
839 ext_debug("insert new index %d: "
840 "move %d indices from 0x%p to 0x%p\n",
841 logical, len, ix, ix + 1);
842 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
845 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
846 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
850 ix->ei_block = cpu_to_le32(logical);
851 ext4_idx_store_pblock(ix, ptr);
852 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
854 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
855 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
859 err = ext4_ext_dirty(handle, inode, curp);
860 ext4_std_error(inode->i_sb, err);
867 * inserts new subtree into the path, using free index entry
869 * - allocates all needed blocks (new leaf and all intermediate index blocks)
870 * - makes decision where to split
871 * - moves remaining extents and index entries (right to the split point)
872 * into the newly allocated blocks
873 * - initializes subtree
875 static int ext4_ext_split(handle_t *handle, struct inode *inode,
877 struct ext4_ext_path *path,
878 struct ext4_extent *newext, int at)
880 struct buffer_head *bh = NULL;
881 int depth = ext_depth(inode);
882 struct ext4_extent_header *neh;
883 struct ext4_extent_idx *fidx;
885 ext4_fsblk_t newblock, oldblock;
887 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
890 /* make decision: where to split? */
891 /* FIXME: now decision is simplest: at current extent */
893 /* if current leaf will be split, then we should use
894 * border from split point */
895 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
896 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
899 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
900 border = path[depth].p_ext[1].ee_block;
901 ext_debug("leaf will be split."
902 " next leaf starts at %d\n",
903 le32_to_cpu(border));
905 border = newext->ee_block;
906 ext_debug("leaf will be added."
907 " next leaf starts at %d\n",
908 le32_to_cpu(border));
912 * If error occurs, then we break processing
913 * and mark filesystem read-only. index won't
914 * be inserted and tree will be in consistent
915 * state. Next mount will repair buffers too.
919 * Get array to track all allocated blocks.
920 * We need this to handle errors and free blocks
923 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
927 /* allocate all needed blocks */
928 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
929 for (a = 0; a < depth - at; a++) {
930 newblock = ext4_ext_new_meta_block(handle, inode, path,
931 newext, &err, flags);
934 ablocks[a] = newblock;
937 /* initialize new leaf */
938 newblock = ablocks[--a];
939 if (unlikely(newblock == 0)) {
940 EXT4_ERROR_INODE(inode, "newblock == 0!");
944 bh = sb_getblk(inode->i_sb, newblock);
951 err = ext4_journal_get_create_access(handle, bh);
955 neh = ext_block_hdr(bh);
957 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
958 neh->eh_magic = EXT4_EXT_MAGIC;
961 /* move remainder of path[depth] to the new leaf */
962 if (unlikely(path[depth].p_hdr->eh_entries !=
963 path[depth].p_hdr->eh_max)) {
964 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
965 path[depth].p_hdr->eh_entries,
966 path[depth].p_hdr->eh_max);
970 /* start copy from next extent */
971 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
972 ext4_ext_show_move(inode, path, newblock, depth);
974 struct ext4_extent *ex;
975 ex = EXT_FIRST_EXTENT(neh);
976 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
977 le16_add_cpu(&neh->eh_entries, m);
980 ext4_extent_block_csum_set(inode, neh);
981 set_buffer_uptodate(bh);
984 err = ext4_handle_dirty_metadata(handle, inode, bh);
990 /* correct old leaf */
992 err = ext4_ext_get_access(handle, inode, path + depth);
995 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
996 err = ext4_ext_dirty(handle, inode, path + depth);
1002 /* create intermediate indexes */
1004 if (unlikely(k < 0)) {
1005 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1010 ext_debug("create %d intermediate indices\n", k);
1011 /* insert new index into current index block */
1012 /* current depth stored in i var */
1015 oldblock = newblock;
1016 newblock = ablocks[--a];
1017 bh = sb_getblk(inode->i_sb, newblock);
1024 err = ext4_journal_get_create_access(handle, bh);
1028 neh = ext_block_hdr(bh);
1029 neh->eh_entries = cpu_to_le16(1);
1030 neh->eh_magic = EXT4_EXT_MAGIC;
1031 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1032 neh->eh_depth = cpu_to_le16(depth - i);
1033 fidx = EXT_FIRST_INDEX(neh);
1034 fidx->ei_block = border;
1035 ext4_idx_store_pblock(fidx, oldblock);
1037 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1038 i, newblock, le32_to_cpu(border), oldblock);
1040 /* move remainder of path[i] to the new index block */
1041 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1042 EXT_LAST_INDEX(path[i].p_hdr))) {
1043 EXT4_ERROR_INODE(inode,
1044 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1045 le32_to_cpu(path[i].p_ext->ee_block));
1049 /* start copy indexes */
1050 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1051 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1052 EXT_MAX_INDEX(path[i].p_hdr));
1053 ext4_ext_show_move(inode, path, newblock, i);
1055 memmove(++fidx, path[i].p_idx,
1056 sizeof(struct ext4_extent_idx) * m);
1057 le16_add_cpu(&neh->eh_entries, m);
1059 ext4_extent_block_csum_set(inode, neh);
1060 set_buffer_uptodate(bh);
1063 err = ext4_handle_dirty_metadata(handle, inode, bh);
1069 /* correct old index */
1071 err = ext4_ext_get_access(handle, inode, path + i);
1074 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1075 err = ext4_ext_dirty(handle, inode, path + i);
1083 /* insert new index */
1084 err = ext4_ext_insert_index(handle, inode, path + at,
1085 le32_to_cpu(border), newblock);
1089 if (buffer_locked(bh))
1095 /* free all allocated blocks in error case */
1096 for (i = 0; i < depth; i++) {
1099 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1100 EXT4_FREE_BLOCKS_METADATA);
1109 * ext4_ext_grow_indepth:
1110 * implements tree growing procedure:
1111 * - allocates new block
1112 * - moves top-level data (index block or leaf) into the new block
1113 * - initializes new top-level, creating index that points to the
1114 * just created block
1116 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1118 struct ext4_extent *newext)
1120 struct ext4_extent_header *neh;
1121 struct buffer_head *bh;
1122 ext4_fsblk_t newblock;
1125 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1126 newext, &err, flags);
1130 bh = sb_getblk(inode->i_sb, newblock);
1133 ext4_std_error(inode->i_sb, err);
1138 err = ext4_journal_get_create_access(handle, bh);
1144 /* move top-level index/leaf into new block */
1145 memmove(bh->b_data, EXT4_I(inode)->i_data,
1146 sizeof(EXT4_I(inode)->i_data));
1148 /* set size of new block */
1149 neh = ext_block_hdr(bh);
1150 /* old root could have indexes or leaves
1151 * so calculate e_max right way */
1152 if (ext_depth(inode))
1153 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1155 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1156 neh->eh_magic = EXT4_EXT_MAGIC;
1157 ext4_extent_block_csum_set(inode, neh);
1158 set_buffer_uptodate(bh);
1161 err = ext4_handle_dirty_metadata(handle, inode, bh);
1165 /* Update top-level index: num,max,pointer */
1166 neh = ext_inode_hdr(inode);
1167 neh->eh_entries = cpu_to_le16(1);
1168 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1169 if (neh->eh_depth == 0) {
1170 /* Root extent block becomes index block */
1171 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1172 EXT_FIRST_INDEX(neh)->ei_block =
1173 EXT_FIRST_EXTENT(neh)->ee_block;
1175 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1176 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1177 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1178 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1180 neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1181 ext4_mark_inode_dirty(handle, inode);
1189 * ext4_ext_create_new_leaf:
1190 * finds empty index and adds new leaf.
1191 * if no free index is found, then it requests in-depth growing.
1193 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1195 struct ext4_ext_path *path,
1196 struct ext4_extent *newext)
1198 struct ext4_ext_path *curp;
1199 int depth, i, err = 0;
1202 i = depth = ext_depth(inode);
1204 /* walk up to the tree and look for free index entry */
1205 curp = path + depth;
1206 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1211 /* we use already allocated block for index block,
1212 * so subsequent data blocks should be contiguous */
1213 if (EXT_HAS_FREE_INDEX(curp)) {
1214 /* if we found index with free entry, then use that
1215 * entry: create all needed subtree and add new leaf */
1216 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1221 ext4_ext_drop_refs(path);
1222 path = ext4_ext_find_extent(inode,
1223 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1226 err = PTR_ERR(path);
1228 /* tree is full, time to grow in depth */
1229 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1234 ext4_ext_drop_refs(path);
1235 path = ext4_ext_find_extent(inode,
1236 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1239 err = PTR_ERR(path);
1244 * only first (depth 0 -> 1) produces free space;
1245 * in all other cases we have to split the grown tree
1247 depth = ext_depth(inode);
1248 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1249 /* now we need to split */
1259 * search the closest allocated block to the left for *logical
1260 * and returns it at @logical + it's physical address at @phys
1261 * if *logical is the smallest allocated block, the function
1262 * returns 0 at @phys
1263 * return value contains 0 (success) or error code
1265 static int ext4_ext_search_left(struct inode *inode,
1266 struct ext4_ext_path *path,
1267 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1269 struct ext4_extent_idx *ix;
1270 struct ext4_extent *ex;
1273 if (unlikely(path == NULL)) {
1274 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1277 depth = path->p_depth;
1280 if (depth == 0 && path->p_ext == NULL)
1283 /* usually extent in the path covers blocks smaller
1284 * then *logical, but it can be that extent is the
1285 * first one in the file */
1287 ex = path[depth].p_ext;
1288 ee_len = ext4_ext_get_actual_len(ex);
1289 if (*logical < le32_to_cpu(ex->ee_block)) {
1290 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1291 EXT4_ERROR_INODE(inode,
1292 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1293 *logical, le32_to_cpu(ex->ee_block));
1296 while (--depth >= 0) {
1297 ix = path[depth].p_idx;
1298 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1299 EXT4_ERROR_INODE(inode,
1300 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1301 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1302 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1303 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1311 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1312 EXT4_ERROR_INODE(inode,
1313 "logical %d < ee_block %d + ee_len %d!",
1314 *logical, le32_to_cpu(ex->ee_block), ee_len);
1318 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1319 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1324 * search the closest allocated block to the right for *logical
1325 * and returns it at @logical + it's physical address at @phys
1326 * if *logical is the largest allocated block, the function
1327 * returns 0 at @phys
1328 * return value contains 0 (success) or error code
1330 static int ext4_ext_search_right(struct inode *inode,
1331 struct ext4_ext_path *path,
1332 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1333 struct ext4_extent **ret_ex)
1335 struct buffer_head *bh = NULL;
1336 struct ext4_extent_header *eh;
1337 struct ext4_extent_idx *ix;
1338 struct ext4_extent *ex;
1340 int depth; /* Note, NOT eh_depth; depth from top of tree */
1343 if (unlikely(path == NULL)) {
1344 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1347 depth = path->p_depth;
1350 if (depth == 0 && path->p_ext == NULL)
1353 /* usually extent in the path covers blocks smaller
1354 * then *logical, but it can be that extent is the
1355 * first one in the file */
1357 ex = path[depth].p_ext;
1358 ee_len = ext4_ext_get_actual_len(ex);
1359 if (*logical < le32_to_cpu(ex->ee_block)) {
1360 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1361 EXT4_ERROR_INODE(inode,
1362 "first_extent(path[%d].p_hdr) != ex",
1366 while (--depth >= 0) {
1367 ix = path[depth].p_idx;
1368 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1369 EXT4_ERROR_INODE(inode,
1370 "ix != EXT_FIRST_INDEX *logical %d!",
1378 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1379 EXT4_ERROR_INODE(inode,
1380 "logical %d < ee_block %d + ee_len %d!",
1381 *logical, le32_to_cpu(ex->ee_block), ee_len);
1385 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1386 /* next allocated block in this leaf */
1391 /* go up and search for index to the right */
1392 while (--depth >= 0) {
1393 ix = path[depth].p_idx;
1394 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1398 /* we've gone up to the root and found no index to the right */
1402 /* we've found index to the right, let's
1403 * follow it and find the closest allocated
1404 * block to the right */
1406 block = ext4_idx_pblock(ix);
1407 while (++depth < path->p_depth) {
1408 bh = sb_bread(inode->i_sb, block);
1411 eh = ext_block_hdr(bh);
1412 /* subtract from p_depth to get proper eh_depth */
1413 if (ext4_ext_check_block(inode, eh,
1414 path->p_depth - depth, bh)) {
1418 ix = EXT_FIRST_INDEX(eh);
1419 block = ext4_idx_pblock(ix);
1423 bh = sb_bread(inode->i_sb, block);
1426 eh = ext_block_hdr(bh);
1427 if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
1431 ex = EXT_FIRST_EXTENT(eh);
1433 *logical = le32_to_cpu(ex->ee_block);
1434 *phys = ext4_ext_pblock(ex);
1442 * ext4_ext_next_allocated_block:
1443 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1444 * NOTE: it considers block number from index entry as
1445 * allocated block. Thus, index entries have to be consistent
1449 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1453 BUG_ON(path == NULL);
1454 depth = path->p_depth;
1456 if (depth == 0 && path->p_ext == NULL)
1457 return EXT_MAX_BLOCKS;
1459 while (depth >= 0) {
1460 if (depth == path->p_depth) {
1462 if (path[depth].p_ext &&
1463 path[depth].p_ext !=
1464 EXT_LAST_EXTENT(path[depth].p_hdr))
1465 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1468 if (path[depth].p_idx !=
1469 EXT_LAST_INDEX(path[depth].p_hdr))
1470 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1475 return EXT_MAX_BLOCKS;
1479 * ext4_ext_next_leaf_block:
1480 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1482 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1486 BUG_ON(path == NULL);
1487 depth = path->p_depth;
1489 /* zero-tree has no leaf blocks at all */
1491 return EXT_MAX_BLOCKS;
1493 /* go to index block */
1496 while (depth >= 0) {
1497 if (path[depth].p_idx !=
1498 EXT_LAST_INDEX(path[depth].p_hdr))
1499 return (ext4_lblk_t)
1500 le32_to_cpu(path[depth].p_idx[1].ei_block);
1504 return EXT_MAX_BLOCKS;
1508 * ext4_ext_correct_indexes:
1509 * if leaf gets modified and modified extent is first in the leaf,
1510 * then we have to correct all indexes above.
1511 * TODO: do we need to correct tree in all cases?
1513 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1514 struct ext4_ext_path *path)
1516 struct ext4_extent_header *eh;
1517 int depth = ext_depth(inode);
1518 struct ext4_extent *ex;
1522 eh = path[depth].p_hdr;
1523 ex = path[depth].p_ext;
1525 if (unlikely(ex == NULL || eh == NULL)) {
1526 EXT4_ERROR_INODE(inode,
1527 "ex %p == NULL or eh %p == NULL", ex, eh);
1532 /* there is no tree at all */
1536 if (ex != EXT_FIRST_EXTENT(eh)) {
1537 /* we correct tree if first leaf got modified only */
1542 * TODO: we need correction if border is smaller than current one
1545 border = path[depth].p_ext->ee_block;
1546 err = ext4_ext_get_access(handle, inode, path + k);
1549 path[k].p_idx->ei_block = border;
1550 err = ext4_ext_dirty(handle, inode, path + k);
1555 /* change all left-side indexes */
1556 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1558 err = ext4_ext_get_access(handle, inode, path + k);
1561 path[k].p_idx->ei_block = border;
1562 err = ext4_ext_dirty(handle, inode, path + k);
1571 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1572 struct ext4_extent *ex2)
1574 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1577 * Make sure that either both extents are uninitialized, or
1580 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1583 if (ext4_ext_is_uninitialized(ex1))
1584 max_len = EXT_UNINIT_MAX_LEN;
1586 max_len = EXT_INIT_MAX_LEN;
1588 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1589 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1591 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1592 le32_to_cpu(ex2->ee_block))
1596 * To allow future support for preallocated extents to be added
1597 * as an RO_COMPAT feature, refuse to merge to extents if
1598 * this can result in the top bit of ee_len being set.
1600 if (ext1_ee_len + ext2_ee_len > max_len)
1602 #ifdef AGGRESSIVE_TEST
1603 if (ext1_ee_len >= 4)
1607 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1613 * This function tries to merge the "ex" extent to the next extent in the tree.
1614 * It always tries to merge towards right. If you want to merge towards
1615 * left, pass "ex - 1" as argument instead of "ex".
1616 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1617 * 1 if they got merged.
1619 static int ext4_ext_try_to_merge_right(struct inode *inode,
1620 struct ext4_ext_path *path,
1621 struct ext4_extent *ex)
1623 struct ext4_extent_header *eh;
1624 unsigned int depth, len;
1626 int uninitialized = 0;
1628 depth = ext_depth(inode);
1629 BUG_ON(path[depth].p_hdr == NULL);
1630 eh = path[depth].p_hdr;
1632 while (ex < EXT_LAST_EXTENT(eh)) {
1633 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1635 /* merge with next extent! */
1636 if (ext4_ext_is_uninitialized(ex))
1638 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1639 + ext4_ext_get_actual_len(ex + 1));
1641 ext4_ext_mark_uninitialized(ex);
1643 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1644 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1645 * sizeof(struct ext4_extent);
1646 memmove(ex + 1, ex + 2, len);
1648 le16_add_cpu(&eh->eh_entries, -1);
1650 WARN_ON(eh->eh_entries == 0);
1651 if (!eh->eh_entries)
1652 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1659 * This function tries to merge the @ex extent to neighbours in the tree.
1660 * return 1 if merge left else 0.
1662 static int ext4_ext_try_to_merge(struct inode *inode,
1663 struct ext4_ext_path *path,
1664 struct ext4_extent *ex) {
1665 struct ext4_extent_header *eh;
1670 depth = ext_depth(inode);
1671 BUG_ON(path[depth].p_hdr == NULL);
1672 eh = path[depth].p_hdr;
1674 if (ex > EXT_FIRST_EXTENT(eh))
1675 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1678 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1684 * check if a portion of the "newext" extent overlaps with an
1687 * If there is an overlap discovered, it updates the length of the newext
1688 * such that there will be no overlap, and then returns 1.
1689 * If there is no overlap found, it returns 0.
1691 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1692 struct inode *inode,
1693 struct ext4_extent *newext,
1694 struct ext4_ext_path *path)
1697 unsigned int depth, len1;
1698 unsigned int ret = 0;
1700 b1 = le32_to_cpu(newext->ee_block);
1701 len1 = ext4_ext_get_actual_len(newext);
1702 depth = ext_depth(inode);
1703 if (!path[depth].p_ext)
1705 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1706 b2 &= ~(sbi->s_cluster_ratio - 1);
1709 * get the next allocated block if the extent in the path
1710 * is before the requested block(s)
1713 b2 = ext4_ext_next_allocated_block(path);
1714 if (b2 == EXT_MAX_BLOCKS)
1716 b2 &= ~(sbi->s_cluster_ratio - 1);
1719 /* check for wrap through zero on extent logical start block*/
1720 if (b1 + len1 < b1) {
1721 len1 = EXT_MAX_BLOCKS - b1;
1722 newext->ee_len = cpu_to_le16(len1);
1726 /* check for overlap */
1727 if (b1 + len1 > b2) {
1728 newext->ee_len = cpu_to_le16(b2 - b1);
1736 * ext4_ext_insert_extent:
1737 * tries to merge requsted extent into the existing extent or
1738 * inserts requested extent as new one into the tree,
1739 * creating new leaf in the no-space case.
1741 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1742 struct ext4_ext_path *path,
1743 struct ext4_extent *newext, int flag)
1745 struct ext4_extent_header *eh;
1746 struct ext4_extent *ex, *fex;
1747 struct ext4_extent *nearex; /* nearest extent */
1748 struct ext4_ext_path *npath = NULL;
1749 int depth, len, err;
1751 unsigned uninitialized = 0;
1754 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1755 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1758 depth = ext_depth(inode);
1759 ex = path[depth].p_ext;
1760 if (unlikely(path[depth].p_hdr == NULL)) {
1761 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1765 /* try to insert block into found extent and return */
1766 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1767 && ext4_can_extents_be_merged(inode, ex, newext)) {
1768 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1769 ext4_ext_is_uninitialized(newext),
1770 ext4_ext_get_actual_len(newext),
1771 le32_to_cpu(ex->ee_block),
1772 ext4_ext_is_uninitialized(ex),
1773 ext4_ext_get_actual_len(ex),
1774 ext4_ext_pblock(ex));
1775 err = ext4_ext_get_access(handle, inode, path + depth);
1780 * ext4_can_extents_be_merged should have checked that either
1781 * both extents are uninitialized, or both aren't. Thus we
1782 * need to check only one of them here.
1784 if (ext4_ext_is_uninitialized(ex))
1786 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1787 + ext4_ext_get_actual_len(newext));
1789 ext4_ext_mark_uninitialized(ex);
1790 eh = path[depth].p_hdr;
1795 depth = ext_depth(inode);
1796 eh = path[depth].p_hdr;
1797 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1800 /* probably next leaf has space for us? */
1801 fex = EXT_LAST_EXTENT(eh);
1802 next = EXT_MAX_BLOCKS;
1803 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1804 next = ext4_ext_next_leaf_block(path);
1805 if (next != EXT_MAX_BLOCKS) {
1806 ext_debug("next leaf block - %u\n", next);
1807 BUG_ON(npath != NULL);
1808 npath = ext4_ext_find_extent(inode, next, NULL);
1810 return PTR_ERR(npath);
1811 BUG_ON(npath->p_depth != path->p_depth);
1812 eh = npath[depth].p_hdr;
1813 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1814 ext_debug("next leaf isn't full(%d)\n",
1815 le16_to_cpu(eh->eh_entries));
1819 ext_debug("next leaf has no free space(%d,%d)\n",
1820 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1824 * There is no free space in the found leaf.
1825 * We're gonna add a new leaf in the tree.
1827 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1828 flags = EXT4_MB_USE_ROOT_BLOCKS;
1829 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1832 depth = ext_depth(inode);
1833 eh = path[depth].p_hdr;
1836 nearex = path[depth].p_ext;
1838 err = ext4_ext_get_access(handle, inode, path + depth);
1843 /* there is no extent in this leaf, create first one */
1844 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1845 le32_to_cpu(newext->ee_block),
1846 ext4_ext_pblock(newext),
1847 ext4_ext_is_uninitialized(newext),
1848 ext4_ext_get_actual_len(newext));
1849 nearex = EXT_FIRST_EXTENT(eh);
1851 if (le32_to_cpu(newext->ee_block)
1852 > le32_to_cpu(nearex->ee_block)) {
1854 ext_debug("insert %u:%llu:[%d]%d before: "
1856 le32_to_cpu(newext->ee_block),
1857 ext4_ext_pblock(newext),
1858 ext4_ext_is_uninitialized(newext),
1859 ext4_ext_get_actual_len(newext),
1864 BUG_ON(newext->ee_block == nearex->ee_block);
1865 ext_debug("insert %u:%llu:[%d]%d after: "
1867 le32_to_cpu(newext->ee_block),
1868 ext4_ext_pblock(newext),
1869 ext4_ext_is_uninitialized(newext),
1870 ext4_ext_get_actual_len(newext),
1873 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1875 ext_debug("insert %u:%llu:[%d]%d: "
1876 "move %d extents from 0x%p to 0x%p\n",
1877 le32_to_cpu(newext->ee_block),
1878 ext4_ext_pblock(newext),
1879 ext4_ext_is_uninitialized(newext),
1880 ext4_ext_get_actual_len(newext),
1881 len, nearex, nearex + 1);
1882 memmove(nearex + 1, nearex,
1883 len * sizeof(struct ext4_extent));
1887 le16_add_cpu(&eh->eh_entries, 1);
1888 path[depth].p_ext = nearex;
1889 nearex->ee_block = newext->ee_block;
1890 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1891 nearex->ee_len = newext->ee_len;
1894 /* try to merge extents */
1895 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1896 ext4_ext_try_to_merge(inode, path, nearex);
1899 /* time to correct all indexes above */
1900 err = ext4_ext_correct_indexes(handle, inode, path);
1904 err = ext4_ext_dirty(handle, inode, path + depth);
1908 ext4_ext_drop_refs(npath);
1911 ext4_ext_invalidate_cache(inode);
1915 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1916 ext4_lblk_t num, ext_prepare_callback func,
1919 struct ext4_ext_path *path = NULL;
1920 struct ext4_ext_cache cbex;
1921 struct ext4_extent *ex;
1922 ext4_lblk_t next, start = 0, end = 0;
1923 ext4_lblk_t last = block + num;
1924 int depth, exists, err = 0;
1926 BUG_ON(func == NULL);
1927 BUG_ON(inode == NULL);
1929 while (block < last && block != EXT_MAX_BLOCKS) {
1931 /* find extent for this block */
1932 down_read(&EXT4_I(inode)->i_data_sem);
1933 path = ext4_ext_find_extent(inode, block, path);
1934 up_read(&EXT4_I(inode)->i_data_sem);
1936 err = PTR_ERR(path);
1941 depth = ext_depth(inode);
1942 if (unlikely(path[depth].p_hdr == NULL)) {
1943 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1947 ex = path[depth].p_ext;
1948 next = ext4_ext_next_allocated_block(path);
1952 /* there is no extent yet, so try to allocate
1953 * all requested space */
1956 } else if (le32_to_cpu(ex->ee_block) > block) {
1957 /* need to allocate space before found extent */
1959 end = le32_to_cpu(ex->ee_block);
1960 if (block + num < end)
1962 } else if (block >= le32_to_cpu(ex->ee_block)
1963 + ext4_ext_get_actual_len(ex)) {
1964 /* need to allocate space after found extent */
1969 } else if (block >= le32_to_cpu(ex->ee_block)) {
1971 * some part of requested space is covered
1975 end = le32_to_cpu(ex->ee_block)
1976 + ext4_ext_get_actual_len(ex);
1977 if (block + num < end)
1983 BUG_ON(end <= start);
1986 cbex.ec_block = start;
1987 cbex.ec_len = end - start;
1990 cbex.ec_block = le32_to_cpu(ex->ee_block);
1991 cbex.ec_len = ext4_ext_get_actual_len(ex);
1992 cbex.ec_start = ext4_ext_pblock(ex);
1995 if (unlikely(cbex.ec_len == 0)) {
1996 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
2000 err = func(inode, next, &cbex, ex, cbdata);
2001 ext4_ext_drop_refs(path);
2006 if (err == EXT_REPEAT)
2008 else if (err == EXT_BREAK) {
2013 if (ext_depth(inode) != depth) {
2014 /* depth was changed. we have to realloc path */
2019 block = cbex.ec_block + cbex.ec_len;
2023 ext4_ext_drop_refs(path);
2031 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
2032 __u32 len, ext4_fsblk_t start)
2034 struct ext4_ext_cache *cex;
2036 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2037 trace_ext4_ext_put_in_cache(inode, block, len, start);
2038 cex = &EXT4_I(inode)->i_cached_extent;
2039 cex->ec_block = block;
2041 cex->ec_start = start;
2042 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2046 * ext4_ext_put_gap_in_cache:
2047 * calculate boundaries of the gap that the requested block fits into
2048 * and cache this gap
2051 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2054 int depth = ext_depth(inode);
2057 struct ext4_extent *ex;
2059 ex = path[depth].p_ext;
2061 /* there is no extent yet, so gap is [0;-] */
2063 len = EXT_MAX_BLOCKS;
2064 ext_debug("cache gap(whole file):");
2065 } else if (block < le32_to_cpu(ex->ee_block)) {
2067 len = le32_to_cpu(ex->ee_block) - block;
2068 ext_debug("cache gap(before): %u [%u:%u]",
2070 le32_to_cpu(ex->ee_block),
2071 ext4_ext_get_actual_len(ex));
2072 } else if (block >= le32_to_cpu(ex->ee_block)
2073 + ext4_ext_get_actual_len(ex)) {
2075 lblock = le32_to_cpu(ex->ee_block)
2076 + ext4_ext_get_actual_len(ex);
2078 next = ext4_ext_next_allocated_block(path);
2079 ext_debug("cache gap(after): [%u:%u] %u",
2080 le32_to_cpu(ex->ee_block),
2081 ext4_ext_get_actual_len(ex),
2083 BUG_ON(next == lblock);
2084 len = next - lblock;
2090 ext_debug(" -> %u:%lu\n", lblock, len);
2091 ext4_ext_put_in_cache(inode, lblock, len, 0);
2095 * ext4_ext_check_cache()
2096 * Checks to see if the given block is in the cache.
2097 * If it is, the cached extent is stored in the given
2098 * cache extent pointer. If the cached extent is a hole,
2099 * this routine should be used instead of
2100 * ext4_ext_in_cache if the calling function needs to
2101 * know the size of the hole.
2103 * @inode: The files inode
2104 * @block: The block to look for in the cache
2105 * @ex: Pointer where the cached extent will be stored
2106 * if it contains block
2108 * Return 0 if cache is invalid; 1 if the cache is valid
2110 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2111 struct ext4_ext_cache *ex){
2112 struct ext4_ext_cache *cex;
2113 struct ext4_sb_info *sbi;
2117 * We borrow i_block_reservation_lock to protect i_cached_extent
2119 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2120 cex = &EXT4_I(inode)->i_cached_extent;
2121 sbi = EXT4_SB(inode->i_sb);
2123 /* has cache valid data? */
2124 if (cex->ec_len == 0)
2127 if (in_range(block, cex->ec_block, cex->ec_len)) {
2128 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2129 ext_debug("%u cached by %u:%u:%llu\n",
2131 cex->ec_block, cex->ec_len, cex->ec_start);
2135 trace_ext4_ext_in_cache(inode, block, ret);
2136 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2141 * ext4_ext_in_cache()
2142 * Checks to see if the given block is in the cache.
2143 * If it is, the cached extent is stored in the given
2146 * @inode: The files inode
2147 * @block: The block to look for in the cache
2148 * @ex: Pointer where the cached extent will be stored
2149 * if it contains block
2151 * Return 0 if cache is invalid; 1 if the cache is valid
2154 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2155 struct ext4_extent *ex)
2157 struct ext4_ext_cache cex;
2160 if (ext4_ext_check_cache(inode, block, &cex)) {
2161 ex->ee_block = cpu_to_le32(cex.ec_block);
2162 ext4_ext_store_pblock(ex, cex.ec_start);
2163 ex->ee_len = cpu_to_le16(cex.ec_len);
2173 * removes index from the index block.
2175 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2176 struct ext4_ext_path *path)
2181 /* free index block */
2183 leaf = ext4_idx_pblock(path->p_idx);
2184 if (unlikely(path->p_hdr->eh_entries == 0)) {
2185 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2188 err = ext4_ext_get_access(handle, inode, path);
2192 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2193 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2194 len *= sizeof(struct ext4_extent_idx);
2195 memmove(path->p_idx, path->p_idx + 1, len);
2198 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2199 err = ext4_ext_dirty(handle, inode, path);
2202 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2203 trace_ext4_ext_rm_idx(inode, leaf);
2205 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2206 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2211 * ext4_ext_calc_credits_for_single_extent:
2212 * This routine returns max. credits that needed to insert an extent
2213 * to the extent tree.
2214 * When pass the actual path, the caller should calculate credits
2217 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2218 struct ext4_ext_path *path)
2221 int depth = ext_depth(inode);
2224 /* probably there is space in leaf? */
2225 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2226 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2229 * There are some space in the leaf tree, no
2230 * need to account for leaf block credit
2232 * bitmaps and block group descriptor blocks
2233 * and other metadata blocks still need to be
2236 /* 1 bitmap, 1 block group descriptor */
2237 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2242 return ext4_chunk_trans_blocks(inode, nrblocks);
2246 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2248 * if nrblocks are fit in a single extent (chunk flag is 1), then
2249 * in the worse case, each tree level index/leaf need to be changed
2250 * if the tree split due to insert a new extent, then the old tree
2251 * index/leaf need to be updated too
2253 * If the nrblocks are discontiguous, they could cause
2254 * the whole tree split more than once, but this is really rare.
2256 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2259 int depth = ext_depth(inode);
2269 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2270 struct ext4_extent *ex,
2271 ext4_fsblk_t *partial_cluster,
2272 ext4_lblk_t from, ext4_lblk_t to)
2274 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2275 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2277 int flags = EXT4_FREE_BLOCKS_FORGET;
2279 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2280 flags |= EXT4_FREE_BLOCKS_METADATA;
2282 * For bigalloc file systems, we never free a partial cluster
2283 * at the beginning of the extent. Instead, we make a note
2284 * that we tried freeing the cluster, and check to see if we
2285 * need to free it on a subsequent call to ext4_remove_blocks,
2286 * or at the end of the ext4_truncate() operation.
2288 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2290 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2292 * If we have a partial cluster, and it's different from the
2293 * cluster of the last block, we need to explicitly free the
2294 * partial cluster here.
2296 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2297 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2298 ext4_free_blocks(handle, inode, NULL,
2299 EXT4_C2B(sbi, *partial_cluster),
2300 sbi->s_cluster_ratio, flags);
2301 *partial_cluster = 0;
2304 #ifdef EXTENTS_STATS
2306 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2307 spin_lock(&sbi->s_ext_stats_lock);
2308 sbi->s_ext_blocks += ee_len;
2309 sbi->s_ext_extents++;
2310 if (ee_len < sbi->s_ext_min)
2311 sbi->s_ext_min = ee_len;
2312 if (ee_len > sbi->s_ext_max)
2313 sbi->s_ext_max = ee_len;
2314 if (ext_depth(inode) > sbi->s_depth_max)
2315 sbi->s_depth_max = ext_depth(inode);
2316 spin_unlock(&sbi->s_ext_stats_lock);
2319 if (from >= le32_to_cpu(ex->ee_block)
2320 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2324 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2325 pblk = ext4_ext_pblock(ex) + ee_len - num;
2326 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2327 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2329 * If the block range to be freed didn't start at the
2330 * beginning of a cluster, and we removed the entire
2331 * extent, save the partial cluster here, since we
2332 * might need to delete if we determine that the
2333 * truncate operation has removed all of the blocks in
2336 if (pblk & (sbi->s_cluster_ratio - 1) &&
2338 *partial_cluster = EXT4_B2C(sbi, pblk);
2340 *partial_cluster = 0;
2341 } else if (from == le32_to_cpu(ex->ee_block)
2342 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2348 start = ext4_ext_pblock(ex);
2350 ext_debug("free first %u blocks starting %llu\n", num, start);
2351 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2354 printk(KERN_INFO "strange request: removal(2) "
2355 "%u-%u from %u:%u\n",
2356 from, to, le32_to_cpu(ex->ee_block), ee_len);
2363 * ext4_ext_rm_leaf() Removes the extents associated with the
2364 * blocks appearing between "start" and "end", and splits the extents
2365 * if "start" and "end" appear in the same extent
2367 * @handle: The journal handle
2368 * @inode: The files inode
2369 * @path: The path to the leaf
2370 * @start: The first block to remove
2371 * @end: The last block to remove
2374 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2375 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2376 ext4_lblk_t start, ext4_lblk_t end)
2378 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2379 int err = 0, correct_index = 0;
2380 int depth = ext_depth(inode), credits;
2381 struct ext4_extent_header *eh;
2384 ext4_lblk_t ex_ee_block;
2385 unsigned short ex_ee_len;
2386 unsigned uninitialized = 0;
2387 struct ext4_extent *ex;
2389 /* the header must be checked already in ext4_ext_remove_space() */
2390 ext_debug("truncate since %u in leaf to %u\n", start, end);
2391 if (!path[depth].p_hdr)
2392 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2393 eh = path[depth].p_hdr;
2394 if (unlikely(path[depth].p_hdr == NULL)) {
2395 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2398 /* find where to start removing */
2399 ex = EXT_LAST_EXTENT(eh);
2401 ex_ee_block = le32_to_cpu(ex->ee_block);
2402 ex_ee_len = ext4_ext_get_actual_len(ex);
2404 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2406 while (ex >= EXT_FIRST_EXTENT(eh) &&
2407 ex_ee_block + ex_ee_len > start) {
2409 if (ext4_ext_is_uninitialized(ex))
2414 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2415 uninitialized, ex_ee_len);
2416 path[depth].p_ext = ex;
2418 a = ex_ee_block > start ? ex_ee_block : start;
2419 b = ex_ee_block+ex_ee_len - 1 < end ?
2420 ex_ee_block+ex_ee_len - 1 : end;
2422 ext_debug(" border %u:%u\n", a, b);
2424 /* If this extent is beyond the end of the hole, skip it */
2425 if (end < ex_ee_block) {
2427 ex_ee_block = le32_to_cpu(ex->ee_block);
2428 ex_ee_len = ext4_ext_get_actual_len(ex);
2430 } else if (b != ex_ee_block + ex_ee_len - 1) {
2431 EXT4_ERROR_INODE(inode,
2432 "can not handle truncate %u:%u "
2434 start, end, ex_ee_block,
2435 ex_ee_block + ex_ee_len - 1);
2438 } else if (a != ex_ee_block) {
2439 /* remove tail of the extent */
2440 num = a - ex_ee_block;
2442 /* remove whole extent: excellent! */
2446 * 3 for leaf, sb, and inode plus 2 (bmap and group
2447 * descriptor) for each block group; assume two block
2448 * groups plus ex_ee_len/blocks_per_block_group for
2451 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2452 if (ex == EXT_FIRST_EXTENT(eh)) {
2454 credits += (ext_depth(inode)) + 1;
2456 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2458 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2462 err = ext4_ext_get_access(handle, inode, path + depth);
2466 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2472 /* this extent is removed; mark slot entirely unused */
2473 ext4_ext_store_pblock(ex, 0);
2475 ex->ee_len = cpu_to_le16(num);
2477 * Do not mark uninitialized if all the blocks in the
2478 * extent have been removed.
2480 if (uninitialized && num)
2481 ext4_ext_mark_uninitialized(ex);
2483 * If the extent was completely released,
2484 * we need to remove it from the leaf
2487 if (end != EXT_MAX_BLOCKS - 1) {
2489 * For hole punching, we need to scoot all the
2490 * extents up when an extent is removed so that
2491 * we dont have blank extents in the middle
2493 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2494 sizeof(struct ext4_extent));
2496 /* Now get rid of the one at the end */
2497 memset(EXT_LAST_EXTENT(eh), 0,
2498 sizeof(struct ext4_extent));
2500 le16_add_cpu(&eh->eh_entries, -1);
2502 *partial_cluster = 0;
2504 err = ext4_ext_dirty(handle, inode, path + depth);
2508 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2509 ext4_ext_pblock(ex));
2511 ex_ee_block = le32_to_cpu(ex->ee_block);
2512 ex_ee_len = ext4_ext_get_actual_len(ex);
2515 if (correct_index && eh->eh_entries)
2516 err = ext4_ext_correct_indexes(handle, inode, path);
2519 * If there is still a entry in the leaf node, check to see if
2520 * it references the partial cluster. This is the only place
2521 * where it could; if it doesn't, we can free the cluster.
2523 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2524 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2525 *partial_cluster)) {
2526 int flags = EXT4_FREE_BLOCKS_FORGET;
2528 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2529 flags |= EXT4_FREE_BLOCKS_METADATA;
2531 ext4_free_blocks(handle, inode, NULL,
2532 EXT4_C2B(sbi, *partial_cluster),
2533 sbi->s_cluster_ratio, flags);
2534 *partial_cluster = 0;
2537 /* if this leaf is free, then we should
2538 * remove it from index block above */
2539 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2540 err = ext4_ext_rm_idx(handle, inode, path + depth);
2547 * ext4_ext_more_to_rm:
2548 * returns 1 if current index has to be freed (even partial)
2551 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2553 BUG_ON(path->p_idx == NULL);
2555 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2559 * if truncate on deeper level happened, it wasn't partial,
2560 * so we have to consider current index for truncation
2562 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2567 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2570 struct super_block *sb = inode->i_sb;
2571 int depth = ext_depth(inode);
2572 struct ext4_ext_path *path = NULL;
2573 ext4_fsblk_t partial_cluster = 0;
2577 ext_debug("truncate since %u to %u\n", start, end);
2579 /* probably first extent we're gonna free will be last in block */
2580 handle = ext4_journal_start(inode, depth + 1);
2582 return PTR_ERR(handle);
2585 ext4_ext_invalidate_cache(inode);
2587 trace_ext4_ext_remove_space(inode, start, depth);
2590 * Check if we are removing extents inside the extent tree. If that
2591 * is the case, we are going to punch a hole inside the extent tree
2592 * so we have to check whether we need to split the extent covering
2593 * the last block to remove so we can easily remove the part of it
2594 * in ext4_ext_rm_leaf().
2596 if (end < EXT_MAX_BLOCKS - 1) {
2597 struct ext4_extent *ex;
2598 ext4_lblk_t ee_block;
2600 /* find extent for this block */
2601 path = ext4_ext_find_extent(inode, end, NULL);
2603 ext4_journal_stop(handle);
2604 return PTR_ERR(path);
2606 depth = ext_depth(inode);
2607 ex = path[depth].p_ext;
2609 ext4_ext_drop_refs(path);
2615 ee_block = le32_to_cpu(ex->ee_block);
2618 * See if the last block is inside the extent, if so split
2619 * the extent at 'end' block so we can easily remove the
2620 * tail of the first part of the split extent in
2621 * ext4_ext_rm_leaf().
2623 if (end >= ee_block &&
2624 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2627 if (ext4_ext_is_uninitialized(ex))
2628 split_flag = EXT4_EXT_MARK_UNINIT1 |
2629 EXT4_EXT_MARK_UNINIT2;
2632 * Split the extent in two so that 'end' is the last
2633 * block in the first new extent
2635 err = ext4_split_extent_at(handle, inode, path,
2636 end + 1, split_flag,
2637 EXT4_GET_BLOCKS_PRE_IO |
2638 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2647 * We start scanning from right side, freeing all the blocks
2648 * after i_size and walking into the tree depth-wise.
2650 depth = ext_depth(inode);
2655 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2657 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2660 ext4_journal_stop(handle);
2663 path[0].p_depth = depth;
2664 path[0].p_hdr = ext_inode_hdr(inode);
2666 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2673 while (i >= 0 && err == 0) {
2675 /* this is leaf block */
2676 err = ext4_ext_rm_leaf(handle, inode, path,
2677 &partial_cluster, start,
2679 /* root level has p_bh == NULL, brelse() eats this */
2680 brelse(path[i].p_bh);
2681 path[i].p_bh = NULL;
2686 /* this is index block */
2687 if (!path[i].p_hdr) {
2688 ext_debug("initialize header\n");
2689 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2692 if (!path[i].p_idx) {
2693 /* this level hasn't been touched yet */
2694 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2695 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2696 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2698 le16_to_cpu(path[i].p_hdr->eh_entries));
2700 /* we were already here, see at next index */
2704 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2705 i, EXT_FIRST_INDEX(path[i].p_hdr),
2707 if (ext4_ext_more_to_rm(path + i)) {
2708 struct buffer_head *bh;
2709 /* go to the next level */
2710 ext_debug("move to level %d (block %llu)\n",
2711 i + 1, ext4_idx_pblock(path[i].p_idx));
2712 memset(path + i + 1, 0, sizeof(*path));
2713 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2715 /* should we reset i_size? */
2719 if (WARN_ON(i + 1 > depth)) {
2723 if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2724 depth - i - 1, bh)) {
2728 path[i + 1].p_bh = bh;
2730 /* save actual number of indexes since this
2731 * number is changed at the next iteration */
2732 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2735 /* we finished processing this index, go up */
2736 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2737 /* index is empty, remove it;
2738 * handle must be already prepared by the
2739 * truncatei_leaf() */
2740 err = ext4_ext_rm_idx(handle, inode, path + i);
2742 /* root level has p_bh == NULL, brelse() eats this */
2743 brelse(path[i].p_bh);
2744 path[i].p_bh = NULL;
2746 ext_debug("return to level %d\n", i);
2750 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2751 path->p_hdr->eh_entries);
2753 /* If we still have something in the partial cluster and we have removed
2754 * even the first extent, then we should free the blocks in the partial
2755 * cluster as well. */
2756 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2757 int flags = EXT4_FREE_BLOCKS_FORGET;
2759 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2760 flags |= EXT4_FREE_BLOCKS_METADATA;
2762 ext4_free_blocks(handle, inode, NULL,
2763 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2764 EXT4_SB(sb)->s_cluster_ratio, flags);
2765 partial_cluster = 0;
2768 /* TODO: flexible tree reduction should be here */
2769 if (path->p_hdr->eh_entries == 0) {
2771 * truncate to zero freed all the tree,
2772 * so we need to correct eh_depth
2774 err = ext4_ext_get_access(handle, inode, path);
2776 ext_inode_hdr(inode)->eh_depth = 0;
2777 ext_inode_hdr(inode)->eh_max =
2778 cpu_to_le16(ext4_ext_space_root(inode, 0));
2779 err = ext4_ext_dirty(handle, inode, path);
2783 ext4_ext_drop_refs(path);
2785 if (err == -EAGAIN) {
2789 ext4_journal_stop(handle);
2795 * called at mount time
2797 void ext4_ext_init(struct super_block *sb)
2800 * possible initialization would be here
2803 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2804 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2805 printk(KERN_INFO "EXT4-fs: file extents enabled"
2806 #ifdef AGGRESSIVE_TEST
2807 ", aggressive tests"
2809 #ifdef CHECK_BINSEARCH
2812 #ifdef EXTENTS_STATS
2817 #ifdef EXTENTS_STATS
2818 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2819 EXT4_SB(sb)->s_ext_min = 1 << 30;
2820 EXT4_SB(sb)->s_ext_max = 0;
2826 * called at umount time
2828 void ext4_ext_release(struct super_block *sb)
2830 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2833 #ifdef EXTENTS_STATS
2834 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2835 struct ext4_sb_info *sbi = EXT4_SB(sb);
2836 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2837 sbi->s_ext_blocks, sbi->s_ext_extents,
2838 sbi->s_ext_blocks / sbi->s_ext_extents);
2839 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2840 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2845 /* FIXME!! we need to try to merge to left or right after zero-out */
2846 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2848 ext4_fsblk_t ee_pblock;
2849 unsigned int ee_len;
2852 ee_len = ext4_ext_get_actual_len(ex);
2853 ee_pblock = ext4_ext_pblock(ex);
2855 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2863 * ext4_split_extent_at() splits an extent at given block.
2865 * @handle: the journal handle
2866 * @inode: the file inode
2867 * @path: the path to the extent
2868 * @split: the logical block where the extent is splitted.
2869 * @split_flags: indicates if the extent could be zeroout if split fails, and
2870 * the states(init or uninit) of new extents.
2871 * @flags: flags used to insert new extent to extent tree.
2874 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2875 * of which are deterimined by split_flag.
2877 * There are two cases:
2878 * a> the extent are splitted into two extent.
2879 * b> split is not needed, and just mark the extent.
2881 * return 0 on success.
2883 static int ext4_split_extent_at(handle_t *handle,
2884 struct inode *inode,
2885 struct ext4_ext_path *path,
2890 ext4_fsblk_t newblock;
2891 ext4_lblk_t ee_block;
2892 struct ext4_extent *ex, newex, orig_ex;
2893 struct ext4_extent *ex2 = NULL;
2894 unsigned int ee_len, depth;
2897 ext_debug("ext4_split_extents_at: inode %lu, logical"
2898 "block %llu\n", inode->i_ino, (unsigned long long)split);
2900 ext4_ext_show_leaf(inode, path);
2902 depth = ext_depth(inode);
2903 ex = path[depth].p_ext;
2904 ee_block = le32_to_cpu(ex->ee_block);
2905 ee_len = ext4_ext_get_actual_len(ex);
2906 newblock = split - ee_block + ext4_ext_pblock(ex);
2908 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2910 err = ext4_ext_get_access(handle, inode, path + depth);
2914 if (split == ee_block) {
2916 * case b: block @split is the block that the extent begins with
2917 * then we just change the state of the extent, and splitting
2920 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2921 ext4_ext_mark_uninitialized(ex);
2923 ext4_ext_mark_initialized(ex);
2925 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2926 ext4_ext_try_to_merge(inode, path, ex);
2928 err = ext4_ext_dirty(handle, inode, path + depth);
2933 memcpy(&orig_ex, ex, sizeof(orig_ex));
2934 ex->ee_len = cpu_to_le16(split - ee_block);
2935 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2936 ext4_ext_mark_uninitialized(ex);
2939 * path may lead to new leaf, not to original leaf any more
2940 * after ext4_ext_insert_extent() returns,
2942 err = ext4_ext_dirty(handle, inode, path + depth);
2944 goto fix_extent_len;
2947 ex2->ee_block = cpu_to_le32(split);
2948 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2949 ext4_ext_store_pblock(ex2, newblock);
2950 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2951 ext4_ext_mark_uninitialized(ex2);
2953 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2954 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2955 err = ext4_ext_zeroout(inode, &orig_ex);
2957 goto fix_extent_len;
2958 /* update the extent length and mark as initialized */
2959 ex->ee_len = cpu_to_le16(ee_len);
2960 ext4_ext_try_to_merge(inode, path, ex);
2961 err = ext4_ext_dirty(handle, inode, path + depth);
2964 goto fix_extent_len;
2967 ext4_ext_show_leaf(inode, path);
2971 ex->ee_len = orig_ex.ee_len;
2972 ext4_ext_dirty(handle, inode, path + depth);
2977 * ext4_split_extents() splits an extent and mark extent which is covered
2978 * by @map as split_flags indicates
2980 * It may result in splitting the extent into multiple extents (upto three)
2981 * There are three possibilities:
2982 * a> There is no split required
2983 * b> Splits in two extents: Split is happening at either end of the extent
2984 * c> Splits in three extents: Somone is splitting in middle of the extent
2987 static int ext4_split_extent(handle_t *handle,
2988 struct inode *inode,
2989 struct ext4_ext_path *path,
2990 struct ext4_map_blocks *map,
2994 ext4_lblk_t ee_block;
2995 struct ext4_extent *ex;
2996 unsigned int ee_len, depth;
2999 int split_flag1, flags1;
3001 depth = ext_depth(inode);
3002 ex = path[depth].p_ext;
3003 ee_block = le32_to_cpu(ex->ee_block);
3004 ee_len = ext4_ext_get_actual_len(ex);
3005 uninitialized = ext4_ext_is_uninitialized(ex);
3007 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3008 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
3009 EXT4_EXT_MAY_ZEROOUT : 0;
3010 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3012 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3013 EXT4_EXT_MARK_UNINIT2;
3014 err = ext4_split_extent_at(handle, inode, path,
3015 map->m_lblk + map->m_len, split_flag1, flags1);
3020 ext4_ext_drop_refs(path);
3021 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3023 return PTR_ERR(path);
3025 if (map->m_lblk >= ee_block) {
3026 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
3027 EXT4_EXT_MAY_ZEROOUT : 0;
3029 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3030 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3031 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3032 err = ext4_split_extent_at(handle, inode, path,
3033 map->m_lblk, split_flag1, flags);
3038 ext4_ext_show_leaf(inode, path);
3040 return err ? err : map->m_len;
3043 #define EXT4_EXT_ZERO_LEN 7
3045 * This function is called by ext4_ext_map_blocks() if someone tries to write
3046 * to an uninitialized extent. It may result in splitting the uninitialized
3047 * extent into multiple extents (up to three - one initialized and two
3049 * There are three possibilities:
3050 * a> There is no split required: Entire extent should be initialized
3051 * b> Splits in two extents: Write is happening at either end of the extent
3052 * c> Splits in three extents: Somone is writing in middle of the extent
3055 * - The extent pointed to by 'path' is uninitialized.
3056 * - The extent pointed to by 'path' contains a superset
3057 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3059 * Post-conditions on success:
3060 * - the returned value is the number of blocks beyond map->l_lblk
3061 * that are allocated and initialized.
3062 * It is guaranteed to be >= map->m_len.
3064 static int ext4_ext_convert_to_initialized(handle_t *handle,
3065 struct inode *inode,
3066 struct ext4_map_blocks *map,
3067 struct ext4_ext_path *path)
3069 struct ext4_extent_header *eh;
3070 struct ext4_map_blocks split_map;
3071 struct ext4_extent zero_ex;
3072 struct ext4_extent *ex;
3073 ext4_lblk_t ee_block, eof_block;
3074 unsigned int ee_len, depth;
3079 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3080 "block %llu, max_blocks %u\n", inode->i_ino,
3081 (unsigned long long)map->m_lblk, map->m_len);
3083 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3084 inode->i_sb->s_blocksize_bits;
3085 if (eof_block < map->m_lblk + map->m_len)
3086 eof_block = map->m_lblk + map->m_len;
3088 depth = ext_depth(inode);
3089 eh = path[depth].p_hdr;
3090 ex = path[depth].p_ext;
3091 ee_block = le32_to_cpu(ex->ee_block);
3092 ee_len = ext4_ext_get_actual_len(ex);
3093 allocated = ee_len - (map->m_lblk - ee_block);
3095 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3097 /* Pre-conditions */
3098 BUG_ON(!ext4_ext_is_uninitialized(ex));
3099 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3102 * Attempt to transfer newly initialized blocks from the currently
3103 * uninitialized extent to its left neighbor. This is much cheaper
3104 * than an insertion followed by a merge as those involve costly
3105 * memmove() calls. This is the common case in steady state for
3106 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3109 * Limitations of the current logic:
3110 * - L1: we only deal with writes at the start of the extent.
3111 * The approach could be extended to writes at the end
3112 * of the extent but this scenario was deemed less common.
3113 * - L2: we do not deal with writes covering the whole extent.
3114 * This would require removing the extent if the transfer
3116 * - L3: we only attempt to merge with an extent stored in the
3117 * same extent tree node.
3119 if ((map->m_lblk == ee_block) && /*L1*/
3120 (map->m_len < ee_len) && /*L2*/
3121 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3122 struct ext4_extent *prev_ex;
3123 ext4_lblk_t prev_lblk;
3124 ext4_fsblk_t prev_pblk, ee_pblk;
3125 unsigned int prev_len, write_len;
3128 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3129 prev_len = ext4_ext_get_actual_len(prev_ex);
3130 prev_pblk = ext4_ext_pblock(prev_ex);
3131 ee_pblk = ext4_ext_pblock(ex);
3132 write_len = map->m_len;
3135 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3136 * upon those conditions:
3137 * - C1: prev_ex is initialized,
3138 * - C2: prev_ex is logically abutting ex,
3139 * - C3: prev_ex is physically abutting ex,
3140 * - C4: prev_ex can receive the additional blocks without
3141 * overflowing the (initialized) length limit.
3143 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3144 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3145 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3146 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3147 err = ext4_ext_get_access(handle, inode, path + depth);
3151 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3154 /* Shift the start of ex by 'write_len' blocks */
3155 ex->ee_block = cpu_to_le32(ee_block + write_len);
3156 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3157 ex->ee_len = cpu_to_le16(ee_len - write_len);
3158 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3160 /* Extend prev_ex by 'write_len' blocks */
3161 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3163 /* Mark the block containing both extents as dirty */
3164 ext4_ext_dirty(handle, inode, path + depth);
3166 /* Update path to point to the right extent */
3167 path[depth].p_ext = prev_ex;
3169 /* Result: number of initialized blocks past m_lblk */
3170 allocated = write_len;
3175 WARN_ON(map->m_lblk < ee_block);
3177 * It is safe to convert extent to initialized via explicit
3178 * zeroout only if extent is fully insde i_size or new_size.
3180 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3182 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3183 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3184 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3185 err = ext4_ext_zeroout(inode, ex);
3189 err = ext4_ext_get_access(handle, inode, path + depth);
3192 ext4_ext_mark_initialized(ex);
3193 ext4_ext_try_to_merge(inode, path, ex);
3194 err = ext4_ext_dirty(handle, inode, path + depth);
3200 * 1. split the extent into three extents.
3201 * 2. split the extent into two extents, zeroout the first half.
3202 * 3. split the extent into two extents, zeroout the second half.
3203 * 4. split the extent into two extents with out zeroout.
3205 split_map.m_lblk = map->m_lblk;
3206 split_map.m_len = map->m_len;
3208 if (allocated > map->m_len) {
3209 if (allocated <= EXT4_EXT_ZERO_LEN &&
3210 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3213 cpu_to_le32(map->m_lblk);
3214 zero_ex.ee_len = cpu_to_le16(allocated);
3215 ext4_ext_store_pblock(&zero_ex,
3216 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3217 err = ext4_ext_zeroout(inode, &zero_ex);
3220 split_map.m_lblk = map->m_lblk;
3221 split_map.m_len = allocated;
3222 } else if ((map->m_lblk - ee_block + map->m_len <
3223 EXT4_EXT_ZERO_LEN) &&
3224 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3226 if (map->m_lblk != ee_block) {
3227 zero_ex.ee_block = ex->ee_block;
3228 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3230 ext4_ext_store_pblock(&zero_ex,
3231 ext4_ext_pblock(ex));
3232 err = ext4_ext_zeroout(inode, &zero_ex);
3237 split_map.m_lblk = ee_block;
3238 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3239 allocated = map->m_len;
3243 allocated = ext4_split_extent(handle, inode, path,
3244 &split_map, split_flag, 0);
3249 return err ? err : allocated;
3253 * This function is called by ext4_ext_map_blocks() from
3254 * ext4_get_blocks_dio_write() when DIO to write
3255 * to an uninitialized extent.
3257 * Writing to an uninitialized extent may result in splitting the uninitialized
3258 * extent into multiple /initialized uninitialized extents (up to three)
3259 * There are three possibilities:
3260 * a> There is no split required: Entire extent should be uninitialized
3261 * b> Splits in two extents: Write is happening at either end of the extent
3262 * c> Splits in three extents: Somone is writing in middle of the extent
3264 * One of more index blocks maybe needed if the extent tree grow after
3265 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3266 * complete, we need to split the uninitialized extent before DIO submit
3267 * the IO. The uninitialized extent called at this time will be split
3268 * into three uninitialized extent(at most). After IO complete, the part
3269 * being filled will be convert to initialized by the end_io callback function
3270 * via ext4_convert_unwritten_extents().
3272 * Returns the size of uninitialized extent to be written on success.
3274 static int ext4_split_unwritten_extents(handle_t *handle,
3275 struct inode *inode,
3276 struct ext4_map_blocks *map,
3277 struct ext4_ext_path *path,
3280 ext4_lblk_t eof_block;
3281 ext4_lblk_t ee_block;
3282 struct ext4_extent *ex;
3283 unsigned int ee_len;
3284 int split_flag = 0, depth;
3286 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3287 "block %llu, max_blocks %u\n", inode->i_ino,
3288 (unsigned long long)map->m_lblk, map->m_len);
3290 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3291 inode->i_sb->s_blocksize_bits;
3292 if (eof_block < map->m_lblk + map->m_len)
3293 eof_block = map->m_lblk + map->m_len;
3295 * It is safe to convert extent to initialized via explicit
3296 * zeroout only if extent is fully insde i_size or new_size.
3298 depth = ext_depth(inode);
3299 ex = path[depth].p_ext;
3300 ee_block = le32_to_cpu(ex->ee_block);
3301 ee_len = ext4_ext_get_actual_len(ex);
3303 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3304 split_flag |= EXT4_EXT_MARK_UNINIT2;
3306 flags |= EXT4_GET_BLOCKS_PRE_IO;
3307 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3310 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3311 struct inode *inode,
3312 struct ext4_ext_path *path)
3314 struct ext4_extent *ex;
3318 depth = ext_depth(inode);
3319 ex = path[depth].p_ext;
3321 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3322 "block %llu, max_blocks %u\n", inode->i_ino,
3323 (unsigned long long)le32_to_cpu(ex->ee_block),
3324 ext4_ext_get_actual_len(ex));
3326 err = ext4_ext_get_access(handle, inode, path + depth);
3329 /* first mark the extent as initialized */
3330 ext4_ext_mark_initialized(ex);
3332 /* note: ext4_ext_correct_indexes() isn't needed here because
3333 * borders are not changed
3335 ext4_ext_try_to_merge(inode, path, ex);
3337 /* Mark modified extent as dirty */
3338 err = ext4_ext_dirty(handle, inode, path + depth);
3340 ext4_ext_show_leaf(inode, path);
3344 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3345 sector_t block, int count)
3348 for (i = 0; i < count; i++)
3349 unmap_underlying_metadata(bdev, block + i);
3353 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3355 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3357 struct ext4_ext_path *path,
3361 struct ext4_extent_header *eh;
3362 struct ext4_extent *last_ex;
3364 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3367 depth = ext_depth(inode);
3368 eh = path[depth].p_hdr;
3371 * We're going to remove EOFBLOCKS_FL entirely in future so we
3372 * do not care for this case anymore. Simply remove the flag
3373 * if there are no extents.
3375 if (unlikely(!eh->eh_entries))
3377 last_ex = EXT_LAST_EXTENT(eh);
3379 * We should clear the EOFBLOCKS_FL flag if we are writing the
3380 * last block in the last extent in the file. We test this by
3381 * first checking to see if the caller to
3382 * ext4_ext_get_blocks() was interested in the last block (or
3383 * a block beyond the last block) in the current extent. If
3384 * this turns out to be false, we can bail out from this
3385 * function immediately.
3387 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3388 ext4_ext_get_actual_len(last_ex))
3391 * If the caller does appear to be planning to write at or
3392 * beyond the end of the current extent, we then test to see
3393 * if the current extent is the last extent in the file, by
3394 * checking to make sure it was reached via the rightmost node
3395 * at each level of the tree.
3397 for (i = depth-1; i >= 0; i--)
3398 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3401 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3402 return ext4_mark_inode_dirty(handle, inode);
3406 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3408 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3409 * whether there are any buffers marked for delayed allocation. It returns '1'
3410 * on the first delalloc'ed buffer head found. If no buffer head in the given
3411 * range is marked for delalloc, it returns 0.
3412 * lblk_start should always be <= lblk_end.
3413 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3414 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3415 * block sooner). This is useful when blocks are truncated sequentially from
3416 * lblk_start towards lblk_end.
3418 static int ext4_find_delalloc_range(struct inode *inode,
3419 ext4_lblk_t lblk_start,
3420 ext4_lblk_t lblk_end,
3421 int search_hint_reverse)
3423 struct address_space *mapping = inode->i_mapping;
3424 struct buffer_head *head, *bh = NULL;
3426 ext4_lblk_t i, pg_lblk;
3429 if (!test_opt(inode->i_sb, DELALLOC))
3432 /* reverse search wont work if fs block size is less than page size */
3433 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3434 search_hint_reverse = 0;
3436 if (search_hint_reverse)
3441 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3443 while ((i >= lblk_start) && (i <= lblk_end)) {
3444 page = find_get_page(mapping, index);
3448 if (!page_has_buffers(page))
3451 head = page_buffers(page);
3456 pg_lblk = index << (PAGE_CACHE_SHIFT -
3459 if (unlikely(pg_lblk < lblk_start)) {
3461 * This is possible when fs block size is less
3462 * than page size and our cluster starts/ends in
3463 * middle of the page. So we need to skip the
3464 * initial few blocks till we reach the 'lblk'
3470 /* Check if the buffer is delayed allocated and that it
3471 * is not yet mapped. (when da-buffers are mapped during
3472 * their writeout, their da_mapped bit is set.)
3474 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3475 page_cache_release(page);
3476 trace_ext4_find_delalloc_range(inode,
3477 lblk_start, lblk_end,
3478 search_hint_reverse,
3482 if (search_hint_reverse)
3486 } while ((i >= lblk_start) && (i <= lblk_end) &&
3487 ((bh = bh->b_this_page) != head));
3490 page_cache_release(page);
3492 * Move to next page. 'i' will be the first lblk in the next
3495 if (search_hint_reverse)
3499 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3502 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3503 search_hint_reverse, 0, 0);
3507 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3508 int search_hint_reverse)
3510 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3511 ext4_lblk_t lblk_start, lblk_end;
3512 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3513 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3515 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3516 search_hint_reverse);
3520 * Determines how many complete clusters (out of those specified by the 'map')
3521 * are under delalloc and were reserved quota for.
3522 * This function is called when we are writing out the blocks that were
3523 * originally written with their allocation delayed, but then the space was
3524 * allocated using fallocate() before the delayed allocation could be resolved.
3525 * The cases to look for are:
3526 * ('=' indicated delayed allocated blocks
3527 * '-' indicates non-delayed allocated blocks)
3528 * (a) partial clusters towards beginning and/or end outside of allocated range
3529 * are not delalloc'ed.
3531 * |----c---=|====c====|====c====|===-c----|
3532 * |++++++ allocated ++++++|
3533 * ==> 4 complete clusters in above example
3535 * (b) partial cluster (outside of allocated range) towards either end is
3536 * marked for delayed allocation. In this case, we will exclude that
3539 * |----====c========|========c========|
3540 * |++++++ allocated ++++++|
3541 * ==> 1 complete clusters in above example
3544 * |================c================|
3545 * |++++++ allocated ++++++|
3546 * ==> 0 complete clusters in above example
3548 * The ext4_da_update_reserve_space will be called only if we
3549 * determine here that there were some "entire" clusters that span
3550 * this 'allocated' range.
3551 * In the non-bigalloc case, this function will just end up returning num_blks
3552 * without ever calling ext4_find_delalloc_range.
3555 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3556 unsigned int num_blks)
3558 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3559 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3560 ext4_lblk_t lblk_from, lblk_to, c_offset;
3561 unsigned int allocated_clusters = 0;
3563 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3564 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3566 /* max possible clusters for this allocation */
3567 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3569 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3571 /* Check towards left side */
3572 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3574 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3575 lblk_to = lblk_from + c_offset - 1;
3577 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3578 allocated_clusters--;
3581 /* Now check towards right. */
3582 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3583 if (allocated_clusters && c_offset) {
3584 lblk_from = lblk_start + num_blks;
3585 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3587 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3588 allocated_clusters--;
3591 return allocated_clusters;
3595 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3596 struct ext4_map_blocks *map,
3597 struct ext4_ext_path *path, int flags,
3598 unsigned int allocated, ext4_fsblk_t newblock)
3602 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3604 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3605 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3606 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3608 ext4_ext_show_leaf(inode, path);
3610 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3613 /* get_block() before submit the IO, split the extent */
3614 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3615 ret = ext4_split_unwritten_extents(handle, inode, map,
3618 * Flag the inode(non aio case) or end_io struct (aio case)
3619 * that this IO needs to conversion to written when IO is
3623 ext4_set_io_unwritten_flag(inode, io);
3625 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3626 if (ext4_should_dioread_nolock(inode))
3627 map->m_flags |= EXT4_MAP_UNINIT;
3630 /* IO end_io complete, convert the filled extent to written */
3631 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3632 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3635 ext4_update_inode_fsync_trans(handle, inode, 1);
3636 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3642 /* buffered IO case */
3644 * repeat fallocate creation request
3645 * we already have an unwritten extent
3647 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3650 /* buffered READ or buffered write_begin() lookup */
3651 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3653 * We have blocks reserved already. We
3654 * return allocated blocks so that delalloc
3655 * won't do block reservation for us. But
3656 * the buffer head will be unmapped so that
3657 * a read from the block returns 0s.
3659 map->m_flags |= EXT4_MAP_UNWRITTEN;
3663 /* buffered write, writepage time, convert*/
3664 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3666 ext4_update_inode_fsync_trans(handle, inode, 1);
3673 map->m_flags |= EXT4_MAP_NEW;
3675 * if we allocated more blocks than requested
3676 * we need to make sure we unmap the extra block
3677 * allocated. The actual needed block will get
3678 * unmapped later when we find the buffer_head marked
3681 if (allocated > map->m_len) {
3682 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3683 newblock + map->m_len,
3684 allocated - map->m_len);
3685 allocated = map->m_len;
3689 * If we have done fallocate with the offset that is already
3690 * delayed allocated, we would have block reservation
3691 * and quota reservation done in the delayed write path.
3692 * But fallocate would have already updated quota and block
3693 * count for this offset. So cancel these reservation
3695 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3696 unsigned int reserved_clusters;
3697 reserved_clusters = get_reserved_cluster_alloc(inode,
3698 map->m_lblk, map->m_len);
3699 if (reserved_clusters)
3700 ext4_da_update_reserve_space(inode,
3706 map->m_flags |= EXT4_MAP_MAPPED;
3707 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3708 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3714 if (allocated > map->m_len)
3715 allocated = map->m_len;
3716 ext4_ext_show_leaf(inode, path);
3717 map->m_pblk = newblock;
3718 map->m_len = allocated;
3721 ext4_ext_drop_refs(path);
3724 return err ? err : allocated;
3728 * get_implied_cluster_alloc - check to see if the requested
3729 * allocation (in the map structure) overlaps with a cluster already
3730 * allocated in an extent.
3731 * @sb The filesystem superblock structure
3732 * @map The requested lblk->pblk mapping
3733 * @ex The extent structure which might contain an implied
3734 * cluster allocation
3736 * This function is called by ext4_ext_map_blocks() after we failed to
3737 * find blocks that were already in the inode's extent tree. Hence,
3738 * we know that the beginning of the requested region cannot overlap
3739 * the extent from the inode's extent tree. There are three cases we
3740 * want to catch. The first is this case:
3742 * |--- cluster # N--|
3743 * |--- extent ---| |---- requested region ---|
3746 * The second case that we need to test for is this one:
3748 * |--------- cluster # N ----------------|
3749 * |--- requested region --| |------- extent ----|
3750 * |=======================|
3752 * The third case is when the requested region lies between two extents
3753 * within the same cluster:
3754 * |------------- cluster # N-------------|
3755 * |----- ex -----| |---- ex_right ----|
3756 * |------ requested region ------|
3757 * |================|
3759 * In each of the above cases, we need to set the map->m_pblk and
3760 * map->m_len so it corresponds to the return the extent labelled as
3761 * "|====|" from cluster #N, since it is already in use for data in
3762 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3763 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3764 * as a new "allocated" block region. Otherwise, we will return 0 and
3765 * ext4_ext_map_blocks() will then allocate one or more new clusters
3766 * by calling ext4_mb_new_blocks().
3768 static int get_implied_cluster_alloc(struct super_block *sb,
3769 struct ext4_map_blocks *map,
3770 struct ext4_extent *ex,
3771 struct ext4_ext_path *path)
3773 struct ext4_sb_info *sbi = EXT4_SB(sb);
3774 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3775 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3776 ext4_lblk_t rr_cluster_start;
3777 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3778 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3779 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3781 /* The extent passed in that we are trying to match */
3782 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3783 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3785 /* The requested region passed into ext4_map_blocks() */
3786 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3788 if ((rr_cluster_start == ex_cluster_end) ||
3789 (rr_cluster_start == ex_cluster_start)) {
3790 if (rr_cluster_start == ex_cluster_end)
3791 ee_start += ee_len - 1;
3792 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3794 map->m_len = min(map->m_len,
3795 (unsigned) sbi->s_cluster_ratio - c_offset);
3797 * Check for and handle this case:
3799 * |--------- cluster # N-------------|
3800 * |------- extent ----|
3801 * |--- requested region ---|
3805 if (map->m_lblk < ee_block)
3806 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3809 * Check for the case where there is already another allocated
3810 * block to the right of 'ex' but before the end of the cluster.
3812 * |------------- cluster # N-------------|
3813 * |----- ex -----| |---- ex_right ----|
3814 * |------ requested region ------|
3815 * |================|
3817 if (map->m_lblk > ee_block) {
3818 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3819 map->m_len = min(map->m_len, next - map->m_lblk);
3822 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3826 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3832 * Block allocation/map/preallocation routine for extents based files
3835 * Need to be called with
3836 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3837 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3839 * return > 0, number of of blocks already mapped/allocated
3840 * if create == 0 and these are pre-allocated blocks
3841 * buffer head is unmapped
3842 * otherwise blocks are mapped
3844 * return = 0, if plain look up failed (blocks have not been allocated)
3845 * buffer head is unmapped
3847 * return < 0, error case.
3849 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3850 struct ext4_map_blocks *map, int flags)
3852 struct ext4_ext_path *path = NULL;
3853 struct ext4_extent newex, *ex, *ex2;
3854 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3855 ext4_fsblk_t newblock = 0;
3856 int free_on_err = 0, err = 0, depth, ret;
3857 unsigned int allocated = 0, offset = 0;
3858 unsigned int allocated_clusters = 0;
3859 struct ext4_allocation_request ar;
3860 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3861 ext4_lblk_t cluster_offset;
3863 ext_debug("blocks %u/%u requested for inode %lu\n",
3864 map->m_lblk, map->m_len, inode->i_ino);
3865 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3867 /* check in cache */
3868 if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3869 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3870 if ((sbi->s_cluster_ratio > 1) &&
3871 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3872 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3874 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3876 * block isn't allocated yet and
3877 * user doesn't want to allocate it
3881 /* we should allocate requested block */
3883 /* block is already allocated */
3884 if (sbi->s_cluster_ratio > 1)
3885 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3886 newblock = map->m_lblk
3887 - le32_to_cpu(newex.ee_block)
3888 + ext4_ext_pblock(&newex);
3889 /* number of remaining blocks in the extent */
3890 allocated = ext4_ext_get_actual_len(&newex) -
3891 (map->m_lblk - le32_to_cpu(newex.ee_block));
3896 /* find extent for this block */
3897 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3899 err = PTR_ERR(path);
3904 depth = ext_depth(inode);
3907 * consistent leaf must not be empty;
3908 * this situation is possible, though, _during_ tree modification;
3909 * this is why assert can't be put in ext4_ext_find_extent()
3911 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3912 EXT4_ERROR_INODE(inode, "bad extent address "
3913 "lblock: %lu, depth: %d pblock %lld",
3914 (unsigned long) map->m_lblk, depth,
3915 path[depth].p_block);
3920 ex = path[depth].p_ext;
3922 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3923 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3924 unsigned short ee_len;
3927 * Uninitialized extents are treated as holes, except that
3928 * we split out initialized portions during a write.
3930 ee_len = ext4_ext_get_actual_len(ex);
3932 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3934 /* if found extent covers block, simply return it */
3935 if (in_range(map->m_lblk, ee_block, ee_len)) {
3936 newblock = map->m_lblk - ee_block + ee_start;
3937 /* number of remaining blocks in the extent */
3938 allocated = ee_len - (map->m_lblk - ee_block);
3939 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3940 ee_block, ee_len, newblock);
3943 * Do not put uninitialized extent
3946 if (!ext4_ext_is_uninitialized(ex)) {
3947 ext4_ext_put_in_cache(inode, ee_block,
3951 ret = ext4_ext_handle_uninitialized_extents(
3952 handle, inode, map, path, flags,
3953 allocated, newblock);
3958 if ((sbi->s_cluster_ratio > 1) &&
3959 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3960 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3963 * requested block isn't allocated yet;
3964 * we couldn't try to create block if create flag is zero
3966 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3968 * put just found gap into cache to speed up
3969 * subsequent requests
3971 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3976 * Okay, we need to do block allocation.
3978 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
3979 newex.ee_block = cpu_to_le32(map->m_lblk);
3980 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3983 * If we are doing bigalloc, check to see if the extent returned
3984 * by ext4_ext_find_extent() implies a cluster we can use.
3986 if (cluster_offset && ex &&
3987 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
3988 ar.len = allocated = map->m_len;
3989 newblock = map->m_pblk;
3990 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3991 goto got_allocated_blocks;
3994 /* find neighbour allocated blocks */
3995 ar.lleft = map->m_lblk;
3996 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3999 ar.lright = map->m_lblk;
4001 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4005 /* Check if the extent after searching to the right implies a
4006 * cluster we can use. */
4007 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4008 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4009 ar.len = allocated = map->m_len;
4010 newblock = map->m_pblk;
4011 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4012 goto got_allocated_blocks;
4016 * See if request is beyond maximum number of blocks we can have in
4017 * a single extent. For an initialized extent this limit is
4018 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4019 * EXT_UNINIT_MAX_LEN.
4021 if (map->m_len > EXT_INIT_MAX_LEN &&
4022 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4023 map->m_len = EXT_INIT_MAX_LEN;
4024 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4025 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4026 map->m_len = EXT_UNINIT_MAX_LEN;
4028 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4029 newex.ee_len = cpu_to_le16(map->m_len);
4030 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4032 allocated = ext4_ext_get_actual_len(&newex);
4034 allocated = map->m_len;
4036 /* allocate new block */
4038 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4039 ar.logical = map->m_lblk;
4041 * We calculate the offset from the beginning of the cluster
4042 * for the logical block number, since when we allocate a
4043 * physical cluster, the physical block should start at the
4044 * same offset from the beginning of the cluster. This is
4045 * needed so that future calls to get_implied_cluster_alloc()
4048 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4049 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4051 ar.logical -= offset;
4052 if (S_ISREG(inode->i_mode))
4053 ar.flags = EXT4_MB_HINT_DATA;
4055 /* disable in-core preallocation for non-regular files */
4057 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4058 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4059 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4062 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4063 ar.goal, newblock, allocated);
4065 allocated_clusters = ar.len;
4066 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4067 if (ar.len > allocated)
4070 got_allocated_blocks:
4071 /* try to insert new extent into found leaf and return */
4072 ext4_ext_store_pblock(&newex, newblock + offset);
4073 newex.ee_len = cpu_to_le16(ar.len);
4074 /* Mark uninitialized */
4075 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4076 ext4_ext_mark_uninitialized(&newex);
4078 * io_end structure was created for every IO write to an
4079 * uninitialized extent. To avoid unnecessary conversion,
4080 * here we flag the IO that really needs the conversion.
4081 * For non asycn direct IO case, flag the inode state
4082 * that we need to perform conversion when IO is done.
4084 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4086 ext4_set_io_unwritten_flag(inode, io);
4088 ext4_set_inode_state(inode,
4089 EXT4_STATE_DIO_UNWRITTEN);
4091 if (ext4_should_dioread_nolock(inode))
4092 map->m_flags |= EXT4_MAP_UNINIT;
4096 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4097 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4100 err = ext4_ext_insert_extent(handle, inode, path,
4102 if (err && free_on_err) {
4103 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4104 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4105 /* free data blocks we just allocated */
4106 /* not a good idea to call discard here directly,
4107 * but otherwise we'd need to call it every free() */
4108 ext4_discard_preallocations(inode);
4109 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4110 ext4_ext_get_actual_len(&newex), fb_flags);
4114 /* previous routine could use block we allocated */
4115 newblock = ext4_ext_pblock(&newex);
4116 allocated = ext4_ext_get_actual_len(&newex);
4117 if (allocated > map->m_len)
4118 allocated = map->m_len;
4119 map->m_flags |= EXT4_MAP_NEW;
4122 * Update reserved blocks/metadata blocks after successful
4123 * block allocation which had been deferred till now.
4125 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4126 unsigned int reserved_clusters;
4128 * Check how many clusters we had reserved this allocated range
4130 reserved_clusters = get_reserved_cluster_alloc(inode,
4131 map->m_lblk, allocated);
4132 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4133 if (reserved_clusters) {
4135 * We have clusters reserved for this range.
4136 * But since we are not doing actual allocation
4137 * and are simply using blocks from previously
4138 * allocated cluster, we should release the
4139 * reservation and not claim quota.
4141 ext4_da_update_reserve_space(inode,
4142 reserved_clusters, 0);
4145 BUG_ON(allocated_clusters < reserved_clusters);
4146 /* We will claim quota for all newly allocated blocks.*/
4147 ext4_da_update_reserve_space(inode, allocated_clusters,
4149 if (reserved_clusters < allocated_clusters) {
4150 struct ext4_inode_info *ei = EXT4_I(inode);
4151 int reservation = allocated_clusters -
4154 * It seems we claimed few clusters outside of
4155 * the range of this allocation. We should give
4156 * it back to the reservation pool. This can
4157 * happen in the following case:
4159 * * Suppose s_cluster_ratio is 4 (i.e., each
4160 * cluster has 4 blocks. Thus, the clusters
4161 * are [0-3],[4-7],[8-11]...
4162 * * First comes delayed allocation write for
4163 * logical blocks 10 & 11. Since there were no
4164 * previous delayed allocated blocks in the
4165 * range [8-11], we would reserve 1 cluster
4167 * * Next comes write for logical blocks 3 to 8.
4168 * In this case, we will reserve 2 clusters
4169 * (for [0-3] and [4-7]; and not for [8-11] as
4170 * that range has a delayed allocated blocks.
4171 * Thus total reserved clusters now becomes 3.
4172 * * Now, during the delayed allocation writeout
4173 * time, we will first write blocks [3-8] and
4174 * allocate 3 clusters for writing these
4175 * blocks. Also, we would claim all these
4176 * three clusters above.
4177 * * Now when we come here to writeout the
4178 * blocks [10-11], we would expect to claim
4179 * the reservation of 1 cluster we had made
4180 * (and we would claim it since there are no
4181 * more delayed allocated blocks in the range
4182 * [8-11]. But our reserved cluster count had
4183 * already gone to 0.
4185 * Thus, at the step 4 above when we determine
4186 * that there are still some unwritten delayed
4187 * allocated blocks outside of our current
4188 * block range, we should increment the
4189 * reserved clusters count so that when the
4190 * remaining blocks finally gets written, we
4193 dquot_reserve_block(inode,
4194 EXT4_C2B(sbi, reservation));
4195 spin_lock(&ei->i_block_reservation_lock);
4196 ei->i_reserved_data_blocks += reservation;
4197 spin_unlock(&ei->i_block_reservation_lock);
4203 * Cache the extent and update transaction to commit on fdatasync only
4204 * when it is _not_ an uninitialized extent.
4206 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4207 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4208 ext4_update_inode_fsync_trans(handle, inode, 1);
4210 ext4_update_inode_fsync_trans(handle, inode, 0);
4212 if (allocated > map->m_len)
4213 allocated = map->m_len;
4214 ext4_ext_show_leaf(inode, path);
4215 map->m_flags |= EXT4_MAP_MAPPED;
4216 map->m_pblk = newblock;
4217 map->m_len = allocated;
4220 ext4_ext_drop_refs(path);
4224 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4225 newblock, map->m_len, err ? err : allocated);
4227 return err ? err : allocated;
4230 void ext4_ext_truncate(struct inode *inode)
4232 struct address_space *mapping = inode->i_mapping;
4233 struct super_block *sb = inode->i_sb;
4234 ext4_lblk_t last_block;
4240 * finish any pending end_io work so we won't run the risk of
4241 * converting any truncated blocks to initialized later
4243 ext4_flush_completed_IO(inode);
4246 * probably first extent we're gonna free will be last in block
4248 err = ext4_writepage_trans_blocks(inode);
4249 handle = ext4_journal_start(inode, err);
4253 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4254 page_len = PAGE_CACHE_SIZE -
4255 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4257 err = ext4_discard_partial_page_buffers(handle,
4258 mapping, inode->i_size, page_len, 0);
4264 if (ext4_orphan_add(handle, inode))
4267 down_write(&EXT4_I(inode)->i_data_sem);
4268 ext4_ext_invalidate_cache(inode);
4270 ext4_discard_preallocations(inode);
4273 * TODO: optimization is possible here.
4274 * Probably we need not scan at all,
4275 * because page truncation is enough.
4278 /* we have to know where to truncate from in crash case */
4279 EXT4_I(inode)->i_disksize = inode->i_size;
4280 ext4_mark_inode_dirty(handle, inode);
4282 last_block = (inode->i_size + sb->s_blocksize - 1)
4283 >> EXT4_BLOCK_SIZE_BITS(sb);
4284 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4286 /* In a multi-transaction truncate, we only make the final
4287 * transaction synchronous.
4290 ext4_handle_sync(handle);
4292 up_write(&EXT4_I(inode)->i_data_sem);
4296 * If this was a simple ftruncate() and the file will remain alive,
4297 * then we need to clear up the orphan record which we created above.
4298 * However, if this was a real unlink then we were called by
4299 * ext4_delete_inode(), and we allow that function to clean up the
4300 * orphan info for us.
4303 ext4_orphan_del(handle, inode);
4305 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4306 ext4_mark_inode_dirty(handle, inode);
4307 ext4_journal_stop(handle);
4310 static void ext4_falloc_update_inode(struct inode *inode,
4311 int mode, loff_t new_size, int update_ctime)
4313 struct timespec now;
4316 now = current_fs_time(inode->i_sb);
4317 if (!timespec_equal(&inode->i_ctime, &now))
4318 inode->i_ctime = now;
4321 * Update only when preallocation was requested beyond
4324 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4325 if (new_size > i_size_read(inode))
4326 i_size_write(inode, new_size);
4327 if (new_size > EXT4_I(inode)->i_disksize)
4328 ext4_update_i_disksize(inode, new_size);
4331 * Mark that we allocate beyond EOF so the subsequent truncate
4332 * can proceed even if the new size is the same as i_size.
4334 if (new_size > i_size_read(inode))
4335 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4341 * preallocate space for a file. This implements ext4's fallocate file
4342 * operation, which gets called from sys_fallocate system call.
4343 * For block-mapped files, posix_fallocate should fall back to the method
4344 * of writing zeroes to the required new blocks (the same behavior which is
4345 * expected for file systems which do not support fallocate() system call).
4347 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4349 struct inode *inode = file->f_path.dentry->d_inode;
4352 unsigned int max_blocks;
4357 struct ext4_map_blocks map;
4358 unsigned int credits, blkbits = inode->i_blkbits;
4361 * currently supporting (pre)allocate mode for extent-based
4364 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4367 /* Return error if mode is not supported */
4368 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4371 if (mode & FALLOC_FL_PUNCH_HOLE)
4372 return ext4_punch_hole(file, offset, len);
4374 trace_ext4_fallocate_enter(inode, offset, len, mode);
4375 map.m_lblk = offset >> blkbits;
4377 * We can't just convert len to max_blocks because
4378 * If blocksize = 4096 offset = 3072 and len = 2048
4380 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4383 * credits to insert 1 extent into extent tree
4385 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4386 mutex_lock(&inode->i_mutex);
4387 ret = inode_newsize_ok(inode, (len + offset));
4389 mutex_unlock(&inode->i_mutex);
4390 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4393 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4394 if (mode & FALLOC_FL_KEEP_SIZE)
4395 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4397 * Don't normalize the request if it can fit in one extent so
4398 * that it doesn't get unnecessarily split into multiple
4401 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4402 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4404 while (ret >= 0 && ret < max_blocks) {
4405 map.m_lblk = map.m_lblk + ret;
4406 map.m_len = max_blocks = max_blocks - ret;
4407 handle = ext4_journal_start(inode, credits);
4408 if (IS_ERR(handle)) {
4409 ret = PTR_ERR(handle);
4412 ret = ext4_map_blocks(handle, inode, &map, flags);
4416 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4417 "returned error inode#%lu, block=%u, "
4418 "max_blocks=%u", __func__,
4419 inode->i_ino, map.m_lblk, max_blocks);
4421 ext4_mark_inode_dirty(handle, inode);
4422 ret2 = ext4_journal_stop(handle);
4425 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4426 blkbits) >> blkbits))
4427 new_size = offset + len;
4429 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4431 ext4_falloc_update_inode(inode, mode, new_size,
4432 (map.m_flags & EXT4_MAP_NEW));
4433 ext4_mark_inode_dirty(handle, inode);
4434 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4435 ext4_handle_sync(handle);
4436 ret2 = ext4_journal_stop(handle);
4440 if (ret == -ENOSPC &&
4441 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4445 mutex_unlock(&inode->i_mutex);
4446 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4447 ret > 0 ? ret2 : ret);
4448 return ret > 0 ? ret2 : ret;
4452 * This function convert a range of blocks to written extents
4453 * The caller of this function will pass the start offset and the size.
4454 * all unwritten extents within this range will be converted to
4457 * This function is called from the direct IO end io call back
4458 * function, to convert the fallocated extents after IO is completed.
4459 * Returns 0 on success.
4461 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4465 unsigned int max_blocks;
4468 struct ext4_map_blocks map;
4469 unsigned int credits, blkbits = inode->i_blkbits;
4471 map.m_lblk = offset >> blkbits;
4473 * We can't just convert len to max_blocks because
4474 * If blocksize = 4096 offset = 3072 and len = 2048
4476 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4479 * credits to insert 1 extent into extent tree
4481 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4482 while (ret >= 0 && ret < max_blocks) {
4484 map.m_len = (max_blocks -= ret);
4485 handle = ext4_journal_start(inode, credits);
4486 if (IS_ERR(handle)) {
4487 ret = PTR_ERR(handle);
4490 ret = ext4_map_blocks(handle, inode, &map,
4491 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4494 ext4_msg(inode->i_sb, KERN_ERR,
4495 "%s:%d: inode #%lu: block %u: len %u: "
4496 "ext4_ext_map_blocks returned %d",
4497 __func__, __LINE__, inode->i_ino, map.m_lblk,
4500 ext4_mark_inode_dirty(handle, inode);
4501 ret2 = ext4_journal_stop(handle);
4502 if (ret <= 0 || ret2 )
4505 return ret > 0 ? ret2 : ret;
4509 * Callback function called for each extent to gather FIEMAP information.
4511 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4512 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4520 struct fiemap_extent_info *fieinfo = data;
4521 unsigned char blksize_bits;
4523 blksize_bits = inode->i_sb->s_blocksize_bits;
4524 logical = (__u64)newex->ec_block << blksize_bits;
4526 if (newex->ec_start == 0) {
4528 * No extent in extent-tree contains block @newex->ec_start,
4529 * then the block may stay in 1)a hole or 2)delayed-extent.
4531 * Holes or delayed-extents are processed as follows.
4532 * 1. lookup dirty pages with specified range in pagecache.
4533 * If no page is got, then there is no delayed-extent and
4534 * return with EXT_CONTINUE.
4535 * 2. find the 1st mapped buffer,
4536 * 3. check if the mapped buffer is both in the request range
4537 * and a delayed buffer. If not, there is no delayed-extent,
4539 * 4. a delayed-extent is found, the extent will be collected.
4541 ext4_lblk_t end = 0;
4542 pgoff_t last_offset;
4545 pgoff_t start_index = 0;
4546 struct page **pages = NULL;
4547 struct buffer_head *bh = NULL;
4548 struct buffer_head *head = NULL;
4549 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4551 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4555 offset = logical >> PAGE_SHIFT;
4557 last_offset = offset;
4559 ret = find_get_pages_tag(inode->i_mapping, &offset,
4560 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4562 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4563 /* First time, try to find a mapped buffer. */
4566 for (index = 0; index < ret; index++)
4567 page_cache_release(pages[index]);
4570 return EXT_CONTINUE;
4575 /* Try to find the 1st mapped buffer. */
4576 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4578 if (!page_has_buffers(pages[index]))
4580 head = page_buffers(pages[index]);
4587 if (end >= newex->ec_block +
4589 /* The buffer is out of
4590 * the request range.
4594 if (buffer_mapped(bh) &&
4595 end >= newex->ec_block) {
4596 start_index = index - 1;
4597 /* get the 1st mapped buffer. */
4598 goto found_mapped_buffer;
4601 bh = bh->b_this_page;
4603 } while (bh != head);
4605 /* No mapped buffer in the range found in this page,
4606 * We need to look up next page.
4609 /* There is no page left, but we need to limit
4612 newex->ec_len = end - newex->ec_block;
4617 /*Find contiguous delayed buffers. */
4618 if (ret > 0 && pages[0]->index == last_offset)
4619 head = page_buffers(pages[0]);
4625 found_mapped_buffer:
4626 if (bh != NULL && buffer_delay(bh)) {
4627 /* 1st or contiguous delayed buffer found. */
4628 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4630 * 1st delayed buffer found, record
4631 * the start of extent.
4633 flags |= FIEMAP_EXTENT_DELALLOC;
4634 newex->ec_block = end;
4635 logical = (__u64)end << blksize_bits;
4637 /* Find contiguous delayed buffers. */
4639 if (!buffer_delay(bh))
4640 goto found_delayed_extent;
4641 bh = bh->b_this_page;
4643 } while (bh != head);
4645 for (; index < ret; index++) {
4646 if (!page_has_buffers(pages[index])) {
4650 head = page_buffers(pages[index]);
4656 if (pages[index]->index !=
4657 pages[start_index]->index + index
4659 /* Blocks are not contiguous. */
4665 if (!buffer_delay(bh))
4666 /* Delayed-extent ends. */
4667 goto found_delayed_extent;
4668 bh = bh->b_this_page;
4670 } while (bh != head);
4672 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4676 found_delayed_extent:
4677 newex->ec_len = min(end - newex->ec_block,
4678 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4679 if (ret == nr_pages && bh != NULL &&
4680 newex->ec_len < EXT_INIT_MAX_LEN &&
4682 /* Have not collected an extent and continue. */
4683 for (index = 0; index < ret; index++)
4684 page_cache_release(pages[index]);
4688 for (index = 0; index < ret; index++)
4689 page_cache_release(pages[index]);
4693 physical = (__u64)newex->ec_start << blksize_bits;
4694 length = (__u64)newex->ec_len << blksize_bits;
4696 if (ex && ext4_ext_is_uninitialized(ex))
4697 flags |= FIEMAP_EXTENT_UNWRITTEN;
4699 if (next == EXT_MAX_BLOCKS)
4700 flags |= FIEMAP_EXTENT_LAST;
4702 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4708 return EXT_CONTINUE;
4710 /* fiemap flags we can handle specified here */
4711 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4713 static int ext4_xattr_fiemap(struct inode *inode,
4714 struct fiemap_extent_info *fieinfo)
4718 __u32 flags = FIEMAP_EXTENT_LAST;
4719 int blockbits = inode->i_sb->s_blocksize_bits;
4723 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4724 struct ext4_iloc iloc;
4725 int offset; /* offset of xattr in inode */
4727 error = ext4_get_inode_loc(inode, &iloc);
4730 physical = iloc.bh->b_blocknr << blockbits;
4731 offset = EXT4_GOOD_OLD_INODE_SIZE +
4732 EXT4_I(inode)->i_extra_isize;
4734 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4735 flags |= FIEMAP_EXTENT_DATA_INLINE;
4737 } else { /* external block */
4738 physical = EXT4_I(inode)->i_file_acl << blockbits;
4739 length = inode->i_sb->s_blocksize;
4743 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4745 return (error < 0 ? error : 0);
4749 * ext4_ext_punch_hole
4751 * Punches a hole of "length" bytes in a file starting
4754 * @inode: The inode of the file to punch a hole in
4755 * @offset: The starting byte offset of the hole
4756 * @length: The length of the hole
4758 * Returns the number of blocks removed or negative on err
4760 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4762 struct inode *inode = file->f_path.dentry->d_inode;
4763 struct super_block *sb = inode->i_sb;
4764 ext4_lblk_t first_block, stop_block;
4765 struct address_space *mapping = inode->i_mapping;
4767 loff_t first_page, last_page, page_len;
4768 loff_t first_page_offset, last_page_offset;
4769 int credits, err = 0;
4771 /* No need to punch hole beyond i_size */
4772 if (offset >= inode->i_size)
4776 * If the hole extends beyond i_size, set the hole
4777 * to end after the page that contains i_size
4779 if (offset + length > inode->i_size) {
4780 length = inode->i_size +
4781 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4785 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4786 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4788 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4789 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4792 * Write out all dirty pages to avoid race conditions
4793 * Then release them.
4795 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4796 err = filemap_write_and_wait_range(mapping,
4797 offset, offset + length - 1);
4803 /* Now release the pages */
4804 if (last_page_offset > first_page_offset) {
4805 truncate_pagecache_range(inode, first_page_offset,
4806 last_page_offset - 1);
4809 /* finish any pending end_io work */
4810 ext4_flush_completed_IO(inode);
4812 credits = ext4_writepage_trans_blocks(inode);
4813 handle = ext4_journal_start(inode, credits);
4815 return PTR_ERR(handle);
4817 err = ext4_orphan_add(handle, inode);
4822 * Now we need to zero out the non-page-aligned data in the
4823 * pages at the start and tail of the hole, and unmap the buffer
4824 * heads for the block aligned regions of the page that were
4825 * completely zeroed.
4827 if (first_page > last_page) {
4829 * If the file space being truncated is contained within a page
4830 * just zero out and unmap the middle of that page
4832 err = ext4_discard_partial_page_buffers(handle,
4833 mapping, offset, length, 0);
4839 * zero out and unmap the partial page that contains
4840 * the start of the hole
4842 page_len = first_page_offset - offset;
4844 err = ext4_discard_partial_page_buffers(handle, mapping,
4845 offset, page_len, 0);
4851 * zero out and unmap the partial page that contains
4852 * the end of the hole
4854 page_len = offset + length - last_page_offset;
4856 err = ext4_discard_partial_page_buffers(handle, mapping,
4857 last_page_offset, page_len, 0);
4864 * If i_size is contained in the last page, we need to
4865 * unmap and zero the partial page after i_size
4867 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4868 inode->i_size % PAGE_CACHE_SIZE != 0) {
4870 page_len = PAGE_CACHE_SIZE -
4871 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4874 err = ext4_discard_partial_page_buffers(handle,
4875 mapping, inode->i_size, page_len, 0);
4882 first_block = (offset + sb->s_blocksize - 1) >>
4883 EXT4_BLOCK_SIZE_BITS(sb);
4884 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4886 /* If there are no blocks to remove, return now */
4887 if (first_block >= stop_block)
4890 down_write(&EXT4_I(inode)->i_data_sem);
4891 ext4_ext_invalidate_cache(inode);
4892 ext4_discard_preallocations(inode);
4894 err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
4896 ext4_ext_invalidate_cache(inode);
4897 ext4_discard_preallocations(inode);
4900 ext4_handle_sync(handle);
4902 up_write(&EXT4_I(inode)->i_data_sem);
4905 ext4_orphan_del(handle, inode);
4906 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4907 ext4_mark_inode_dirty(handle, inode);
4908 ext4_journal_stop(handle);
4911 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4912 __u64 start, __u64 len)
4914 ext4_lblk_t start_blk;
4917 /* fallback to generic here if not in extents fmt */
4918 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4919 return generic_block_fiemap(inode, fieinfo, start, len,
4922 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4925 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4926 error = ext4_xattr_fiemap(inode, fieinfo);
4928 ext4_lblk_t len_blks;
4931 start_blk = start >> inode->i_sb->s_blocksize_bits;
4932 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4933 if (last_blk >= EXT_MAX_BLOCKS)
4934 last_blk = EXT_MAX_BLOCKS-1;
4935 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4938 * Walk the extent tree gathering extent information.
4939 * ext4_ext_fiemap_cb will push extents back to user.
4941 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4942 ext4_ext_fiemap_cb, fieinfo);