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 does a very simple check to see if we can collapse
1660 * an extent tree with a single extent tree leaf block into the inode.
1662 static void ext4_ext_try_to_merge_up(handle_t *handle,
1663 struct inode *inode,
1664 struct ext4_ext_path *path)
1667 unsigned max_root = ext4_ext_space_root(inode, 0);
1670 if ((path[0].p_depth != 1) ||
1671 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1672 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1676 * We need to modify the block allocation bitmap and the block
1677 * group descriptor to release the extent tree block. If we
1678 * can't get the journal credits, give up.
1680 if (ext4_journal_extend(handle, 2))
1684 * Copy the extent data up to the inode
1686 blk = ext4_idx_pblock(path[0].p_idx);
1687 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1688 sizeof(struct ext4_extent_idx);
1689 s += sizeof(struct ext4_extent_header);
1691 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1692 path[0].p_depth = 0;
1693 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1694 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1695 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1697 brelse(path[1].p_bh);
1698 ext4_free_blocks(handle, inode, NULL, blk, 1,
1699 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1703 * This function tries to merge the @ex extent to neighbours in the tree.
1704 * return 1 if merge left else 0.
1706 static void ext4_ext_try_to_merge(handle_t *handle,
1707 struct inode *inode,
1708 struct ext4_ext_path *path,
1709 struct ext4_extent *ex) {
1710 struct ext4_extent_header *eh;
1714 depth = ext_depth(inode);
1715 BUG_ON(path[depth].p_hdr == NULL);
1716 eh = path[depth].p_hdr;
1718 if (ex > EXT_FIRST_EXTENT(eh))
1719 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1722 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1724 ext4_ext_try_to_merge_up(handle, inode, path);
1728 * check if a portion of the "newext" extent overlaps with an
1731 * If there is an overlap discovered, it updates the length of the newext
1732 * such that there will be no overlap, and then returns 1.
1733 * If there is no overlap found, it returns 0.
1735 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1736 struct inode *inode,
1737 struct ext4_extent *newext,
1738 struct ext4_ext_path *path)
1741 unsigned int depth, len1;
1742 unsigned int ret = 0;
1744 b1 = le32_to_cpu(newext->ee_block);
1745 len1 = ext4_ext_get_actual_len(newext);
1746 depth = ext_depth(inode);
1747 if (!path[depth].p_ext)
1749 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1750 b2 &= ~(sbi->s_cluster_ratio - 1);
1753 * get the next allocated block if the extent in the path
1754 * is before the requested block(s)
1757 b2 = ext4_ext_next_allocated_block(path);
1758 if (b2 == EXT_MAX_BLOCKS)
1760 b2 &= ~(sbi->s_cluster_ratio - 1);
1763 /* check for wrap through zero on extent logical start block*/
1764 if (b1 + len1 < b1) {
1765 len1 = EXT_MAX_BLOCKS - b1;
1766 newext->ee_len = cpu_to_le16(len1);
1770 /* check for overlap */
1771 if (b1 + len1 > b2) {
1772 newext->ee_len = cpu_to_le16(b2 - b1);
1780 * ext4_ext_insert_extent:
1781 * tries to merge requsted extent into the existing extent or
1782 * inserts requested extent as new one into the tree,
1783 * creating new leaf in the no-space case.
1785 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1786 struct ext4_ext_path *path,
1787 struct ext4_extent *newext, int flag)
1789 struct ext4_extent_header *eh;
1790 struct ext4_extent *ex, *fex;
1791 struct ext4_extent *nearex; /* nearest extent */
1792 struct ext4_ext_path *npath = NULL;
1793 int depth, len, err;
1795 unsigned uninitialized = 0;
1798 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1799 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1802 depth = ext_depth(inode);
1803 ex = path[depth].p_ext;
1804 if (unlikely(path[depth].p_hdr == NULL)) {
1805 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1809 /* try to insert block into found extent and return */
1810 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1811 && ext4_can_extents_be_merged(inode, ex, newext)) {
1812 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1813 ext4_ext_is_uninitialized(newext),
1814 ext4_ext_get_actual_len(newext),
1815 le32_to_cpu(ex->ee_block),
1816 ext4_ext_is_uninitialized(ex),
1817 ext4_ext_get_actual_len(ex),
1818 ext4_ext_pblock(ex));
1819 err = ext4_ext_get_access(handle, inode, path + depth);
1824 * ext4_can_extents_be_merged should have checked that either
1825 * both extents are uninitialized, or both aren't. Thus we
1826 * need to check only one of them here.
1828 if (ext4_ext_is_uninitialized(ex))
1830 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1831 + ext4_ext_get_actual_len(newext));
1833 ext4_ext_mark_uninitialized(ex);
1834 eh = path[depth].p_hdr;
1839 depth = ext_depth(inode);
1840 eh = path[depth].p_hdr;
1841 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1844 /* probably next leaf has space for us? */
1845 fex = EXT_LAST_EXTENT(eh);
1846 next = EXT_MAX_BLOCKS;
1847 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1848 next = ext4_ext_next_leaf_block(path);
1849 if (next != EXT_MAX_BLOCKS) {
1850 ext_debug("next leaf block - %u\n", next);
1851 BUG_ON(npath != NULL);
1852 npath = ext4_ext_find_extent(inode, next, NULL);
1854 return PTR_ERR(npath);
1855 BUG_ON(npath->p_depth != path->p_depth);
1856 eh = npath[depth].p_hdr;
1857 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1858 ext_debug("next leaf isn't full(%d)\n",
1859 le16_to_cpu(eh->eh_entries));
1863 ext_debug("next leaf has no free space(%d,%d)\n",
1864 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1868 * There is no free space in the found leaf.
1869 * We're gonna add a new leaf in the tree.
1871 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1872 flags = EXT4_MB_USE_ROOT_BLOCKS;
1873 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1876 depth = ext_depth(inode);
1877 eh = path[depth].p_hdr;
1880 nearex = path[depth].p_ext;
1882 err = ext4_ext_get_access(handle, inode, path + depth);
1887 /* there is no extent in this leaf, create first one */
1888 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1889 le32_to_cpu(newext->ee_block),
1890 ext4_ext_pblock(newext),
1891 ext4_ext_is_uninitialized(newext),
1892 ext4_ext_get_actual_len(newext));
1893 nearex = EXT_FIRST_EXTENT(eh);
1895 if (le32_to_cpu(newext->ee_block)
1896 > le32_to_cpu(nearex->ee_block)) {
1898 ext_debug("insert %u:%llu:[%d]%d before: "
1900 le32_to_cpu(newext->ee_block),
1901 ext4_ext_pblock(newext),
1902 ext4_ext_is_uninitialized(newext),
1903 ext4_ext_get_actual_len(newext),
1908 BUG_ON(newext->ee_block == nearex->ee_block);
1909 ext_debug("insert %u:%llu:[%d]%d after: "
1911 le32_to_cpu(newext->ee_block),
1912 ext4_ext_pblock(newext),
1913 ext4_ext_is_uninitialized(newext),
1914 ext4_ext_get_actual_len(newext),
1917 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1919 ext_debug("insert %u:%llu:[%d]%d: "
1920 "move %d extents from 0x%p to 0x%p\n",
1921 le32_to_cpu(newext->ee_block),
1922 ext4_ext_pblock(newext),
1923 ext4_ext_is_uninitialized(newext),
1924 ext4_ext_get_actual_len(newext),
1925 len, nearex, nearex + 1);
1926 memmove(nearex + 1, nearex,
1927 len * sizeof(struct ext4_extent));
1931 le16_add_cpu(&eh->eh_entries, 1);
1932 path[depth].p_ext = nearex;
1933 nearex->ee_block = newext->ee_block;
1934 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1935 nearex->ee_len = newext->ee_len;
1938 /* try to merge extents */
1939 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1940 ext4_ext_try_to_merge(handle, inode, path, nearex);
1943 /* time to correct all indexes above */
1944 err = ext4_ext_correct_indexes(handle, inode, path);
1948 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
1952 ext4_ext_drop_refs(npath);
1955 ext4_ext_invalidate_cache(inode);
1959 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1960 ext4_lblk_t num, ext_prepare_callback func,
1963 struct ext4_ext_path *path = NULL;
1964 struct ext4_ext_cache cbex;
1965 struct ext4_extent *ex;
1966 ext4_lblk_t next, start = 0, end = 0;
1967 ext4_lblk_t last = block + num;
1968 int depth, exists, err = 0;
1970 BUG_ON(func == NULL);
1971 BUG_ON(inode == NULL);
1973 while (block < last && block != EXT_MAX_BLOCKS) {
1975 /* find extent for this block */
1976 down_read(&EXT4_I(inode)->i_data_sem);
1977 path = ext4_ext_find_extent(inode, block, path);
1978 up_read(&EXT4_I(inode)->i_data_sem);
1980 err = PTR_ERR(path);
1985 depth = ext_depth(inode);
1986 if (unlikely(path[depth].p_hdr == NULL)) {
1987 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1991 ex = path[depth].p_ext;
1992 next = ext4_ext_next_allocated_block(path);
1996 /* there is no extent yet, so try to allocate
1997 * all requested space */
2000 } else if (le32_to_cpu(ex->ee_block) > block) {
2001 /* need to allocate space before found extent */
2003 end = le32_to_cpu(ex->ee_block);
2004 if (block + num < end)
2006 } else if (block >= le32_to_cpu(ex->ee_block)
2007 + ext4_ext_get_actual_len(ex)) {
2008 /* need to allocate space after found extent */
2013 } else if (block >= le32_to_cpu(ex->ee_block)) {
2015 * some part of requested space is covered
2019 end = le32_to_cpu(ex->ee_block)
2020 + ext4_ext_get_actual_len(ex);
2021 if (block + num < end)
2027 BUG_ON(end <= start);
2030 cbex.ec_block = start;
2031 cbex.ec_len = end - start;
2034 cbex.ec_block = le32_to_cpu(ex->ee_block);
2035 cbex.ec_len = ext4_ext_get_actual_len(ex);
2036 cbex.ec_start = ext4_ext_pblock(ex);
2039 if (unlikely(cbex.ec_len == 0)) {
2040 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
2044 err = func(inode, next, &cbex, ex, cbdata);
2045 ext4_ext_drop_refs(path);
2050 if (err == EXT_REPEAT)
2052 else if (err == EXT_BREAK) {
2057 if (ext_depth(inode) != depth) {
2058 /* depth was changed. we have to realloc path */
2063 block = cbex.ec_block + cbex.ec_len;
2067 ext4_ext_drop_refs(path);
2075 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
2076 __u32 len, ext4_fsblk_t start)
2078 struct ext4_ext_cache *cex;
2080 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2081 trace_ext4_ext_put_in_cache(inode, block, len, start);
2082 cex = &EXT4_I(inode)->i_cached_extent;
2083 cex->ec_block = block;
2085 cex->ec_start = start;
2086 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2090 * ext4_ext_put_gap_in_cache:
2091 * calculate boundaries of the gap that the requested block fits into
2092 * and cache this gap
2095 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2098 int depth = ext_depth(inode);
2101 struct ext4_extent *ex;
2103 ex = path[depth].p_ext;
2105 /* there is no extent yet, so gap is [0;-] */
2107 len = EXT_MAX_BLOCKS;
2108 ext_debug("cache gap(whole file):");
2109 } else if (block < le32_to_cpu(ex->ee_block)) {
2111 len = le32_to_cpu(ex->ee_block) - block;
2112 ext_debug("cache gap(before): %u [%u:%u]",
2114 le32_to_cpu(ex->ee_block),
2115 ext4_ext_get_actual_len(ex));
2116 } else if (block >= le32_to_cpu(ex->ee_block)
2117 + ext4_ext_get_actual_len(ex)) {
2119 lblock = le32_to_cpu(ex->ee_block)
2120 + ext4_ext_get_actual_len(ex);
2122 next = ext4_ext_next_allocated_block(path);
2123 ext_debug("cache gap(after): [%u:%u] %u",
2124 le32_to_cpu(ex->ee_block),
2125 ext4_ext_get_actual_len(ex),
2127 BUG_ON(next == lblock);
2128 len = next - lblock;
2134 ext_debug(" -> %u:%lu\n", lblock, len);
2135 ext4_ext_put_in_cache(inode, lblock, len, 0);
2139 * ext4_ext_check_cache()
2140 * Checks to see if the given block is in the cache.
2141 * If it is, the cached extent is stored in the given
2142 * cache extent pointer. If the cached extent is a hole,
2143 * this routine should be used instead of
2144 * ext4_ext_in_cache if the calling function needs to
2145 * know the size of the hole.
2147 * @inode: The files inode
2148 * @block: The block to look for in the cache
2149 * @ex: Pointer where the cached extent will be stored
2150 * if it contains block
2152 * Return 0 if cache is invalid; 1 if the cache is valid
2154 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2155 struct ext4_ext_cache *ex){
2156 struct ext4_ext_cache *cex;
2157 struct ext4_sb_info *sbi;
2161 * We borrow i_block_reservation_lock to protect i_cached_extent
2163 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2164 cex = &EXT4_I(inode)->i_cached_extent;
2165 sbi = EXT4_SB(inode->i_sb);
2167 /* has cache valid data? */
2168 if (cex->ec_len == 0)
2171 if (in_range(block, cex->ec_block, cex->ec_len)) {
2172 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2173 ext_debug("%u cached by %u:%u:%llu\n",
2175 cex->ec_block, cex->ec_len, cex->ec_start);
2179 trace_ext4_ext_in_cache(inode, block, ret);
2180 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2185 * ext4_ext_in_cache()
2186 * Checks to see if the given block is in the cache.
2187 * If it is, the cached extent is stored in the given
2190 * @inode: The files inode
2191 * @block: The block to look for in the cache
2192 * @ex: Pointer where the cached extent will be stored
2193 * if it contains block
2195 * Return 0 if cache is invalid; 1 if the cache is valid
2198 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2199 struct ext4_extent *ex)
2201 struct ext4_ext_cache cex;
2204 if (ext4_ext_check_cache(inode, block, &cex)) {
2205 ex->ee_block = cpu_to_le32(cex.ec_block);
2206 ext4_ext_store_pblock(ex, cex.ec_start);
2207 ex->ee_len = cpu_to_le16(cex.ec_len);
2217 * removes index from the index block.
2219 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2220 struct ext4_ext_path *path)
2225 /* free index block */
2227 leaf = ext4_idx_pblock(path->p_idx);
2228 if (unlikely(path->p_hdr->eh_entries == 0)) {
2229 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2232 err = ext4_ext_get_access(handle, inode, path);
2236 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2237 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2238 len *= sizeof(struct ext4_extent_idx);
2239 memmove(path->p_idx, path->p_idx + 1, len);
2242 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2243 err = ext4_ext_dirty(handle, inode, path);
2246 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2247 trace_ext4_ext_rm_idx(inode, leaf);
2249 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2250 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2255 * ext4_ext_calc_credits_for_single_extent:
2256 * This routine returns max. credits that needed to insert an extent
2257 * to the extent tree.
2258 * When pass the actual path, the caller should calculate credits
2261 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2262 struct ext4_ext_path *path)
2265 int depth = ext_depth(inode);
2268 /* probably there is space in leaf? */
2269 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2270 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2273 * There are some space in the leaf tree, no
2274 * need to account for leaf block credit
2276 * bitmaps and block group descriptor blocks
2277 * and other metadata blocks still need to be
2280 /* 1 bitmap, 1 block group descriptor */
2281 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2286 return ext4_chunk_trans_blocks(inode, nrblocks);
2290 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2292 * if nrblocks are fit in a single extent (chunk flag is 1), then
2293 * in the worse case, each tree level index/leaf need to be changed
2294 * if the tree split due to insert a new extent, then the old tree
2295 * index/leaf need to be updated too
2297 * If the nrblocks are discontiguous, they could cause
2298 * the whole tree split more than once, but this is really rare.
2300 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2303 int depth = ext_depth(inode);
2313 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2314 struct ext4_extent *ex,
2315 ext4_fsblk_t *partial_cluster,
2316 ext4_lblk_t from, ext4_lblk_t to)
2318 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2319 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2323 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2324 flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2325 else if (ext4_should_journal_data(inode))
2326 flags |= EXT4_FREE_BLOCKS_FORGET;
2329 * For bigalloc file systems, we never free a partial cluster
2330 * at the beginning of the extent. Instead, we make a note
2331 * that we tried freeing the cluster, and check to see if we
2332 * need to free it on a subsequent call to ext4_remove_blocks,
2333 * or at the end of the ext4_truncate() operation.
2335 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2337 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2339 * If we have a partial cluster, and it's different from the
2340 * cluster of the last block, we need to explicitly free the
2341 * partial cluster here.
2343 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2344 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2345 ext4_free_blocks(handle, inode, NULL,
2346 EXT4_C2B(sbi, *partial_cluster),
2347 sbi->s_cluster_ratio, flags);
2348 *partial_cluster = 0;
2351 #ifdef EXTENTS_STATS
2353 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2354 spin_lock(&sbi->s_ext_stats_lock);
2355 sbi->s_ext_blocks += ee_len;
2356 sbi->s_ext_extents++;
2357 if (ee_len < sbi->s_ext_min)
2358 sbi->s_ext_min = ee_len;
2359 if (ee_len > sbi->s_ext_max)
2360 sbi->s_ext_max = ee_len;
2361 if (ext_depth(inode) > sbi->s_depth_max)
2362 sbi->s_depth_max = ext_depth(inode);
2363 spin_unlock(&sbi->s_ext_stats_lock);
2366 if (from >= le32_to_cpu(ex->ee_block)
2367 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2371 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2372 pblk = ext4_ext_pblock(ex) + ee_len - num;
2373 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2374 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2376 * If the block range to be freed didn't start at the
2377 * beginning of a cluster, and we removed the entire
2378 * extent, save the partial cluster here, since we
2379 * might need to delete if we determine that the
2380 * truncate operation has removed all of the blocks in
2383 if (pblk & (sbi->s_cluster_ratio - 1) &&
2385 *partial_cluster = EXT4_B2C(sbi, pblk);
2387 *partial_cluster = 0;
2388 } else if (from == le32_to_cpu(ex->ee_block)
2389 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2395 start = ext4_ext_pblock(ex);
2397 ext_debug("free first %u blocks starting %llu\n", num, start);
2398 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2401 printk(KERN_INFO "strange request: removal(2) "
2402 "%u-%u from %u:%u\n",
2403 from, to, le32_to_cpu(ex->ee_block), ee_len);
2410 * ext4_ext_rm_leaf() Removes the extents associated with the
2411 * blocks appearing between "start" and "end", and splits the extents
2412 * if "start" and "end" appear in the same extent
2414 * @handle: The journal handle
2415 * @inode: The files inode
2416 * @path: The path to the leaf
2417 * @start: The first block to remove
2418 * @end: The last block to remove
2421 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2422 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2423 ext4_lblk_t start, ext4_lblk_t end)
2425 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2426 int err = 0, correct_index = 0;
2427 int depth = ext_depth(inode), credits;
2428 struct ext4_extent_header *eh;
2431 ext4_lblk_t ex_ee_block;
2432 unsigned short ex_ee_len;
2433 unsigned uninitialized = 0;
2434 struct ext4_extent *ex;
2436 /* the header must be checked already in ext4_ext_remove_space() */
2437 ext_debug("truncate since %u in leaf to %u\n", start, end);
2438 if (!path[depth].p_hdr)
2439 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2440 eh = path[depth].p_hdr;
2441 if (unlikely(path[depth].p_hdr == NULL)) {
2442 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2445 /* find where to start removing */
2446 ex = EXT_LAST_EXTENT(eh);
2448 ex_ee_block = le32_to_cpu(ex->ee_block);
2449 ex_ee_len = ext4_ext_get_actual_len(ex);
2451 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2453 while (ex >= EXT_FIRST_EXTENT(eh) &&
2454 ex_ee_block + ex_ee_len > start) {
2456 if (ext4_ext_is_uninitialized(ex))
2461 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2462 uninitialized, ex_ee_len);
2463 path[depth].p_ext = ex;
2465 a = ex_ee_block > start ? ex_ee_block : start;
2466 b = ex_ee_block+ex_ee_len - 1 < end ?
2467 ex_ee_block+ex_ee_len - 1 : end;
2469 ext_debug(" border %u:%u\n", a, b);
2471 /* If this extent is beyond the end of the hole, skip it */
2472 if (end < ex_ee_block) {
2474 ex_ee_block = le32_to_cpu(ex->ee_block);
2475 ex_ee_len = ext4_ext_get_actual_len(ex);
2477 } else if (b != ex_ee_block + ex_ee_len - 1) {
2478 EXT4_ERROR_INODE(inode,
2479 "can not handle truncate %u:%u "
2481 start, end, ex_ee_block,
2482 ex_ee_block + ex_ee_len - 1);
2485 } else if (a != ex_ee_block) {
2486 /* remove tail of the extent */
2487 num = a - ex_ee_block;
2489 /* remove whole extent: excellent! */
2493 * 3 for leaf, sb, and inode plus 2 (bmap and group
2494 * descriptor) for each block group; assume two block
2495 * groups plus ex_ee_len/blocks_per_block_group for
2498 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2499 if (ex == EXT_FIRST_EXTENT(eh)) {
2501 credits += (ext_depth(inode)) + 1;
2503 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2505 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2509 err = ext4_ext_get_access(handle, inode, path + depth);
2513 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2519 /* this extent is removed; mark slot entirely unused */
2520 ext4_ext_store_pblock(ex, 0);
2522 ex->ee_len = cpu_to_le16(num);
2524 * Do not mark uninitialized if all the blocks in the
2525 * extent have been removed.
2527 if (uninitialized && num)
2528 ext4_ext_mark_uninitialized(ex);
2530 * If the extent was completely released,
2531 * we need to remove it from the leaf
2534 if (end != EXT_MAX_BLOCKS - 1) {
2536 * For hole punching, we need to scoot all the
2537 * extents up when an extent is removed so that
2538 * we dont have blank extents in the middle
2540 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2541 sizeof(struct ext4_extent));
2543 /* Now get rid of the one at the end */
2544 memset(EXT_LAST_EXTENT(eh), 0,
2545 sizeof(struct ext4_extent));
2547 le16_add_cpu(&eh->eh_entries, -1);
2549 *partial_cluster = 0;
2551 err = ext4_ext_dirty(handle, inode, path + depth);
2555 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2556 ext4_ext_pblock(ex));
2558 ex_ee_block = le32_to_cpu(ex->ee_block);
2559 ex_ee_len = ext4_ext_get_actual_len(ex);
2562 if (correct_index && eh->eh_entries)
2563 err = ext4_ext_correct_indexes(handle, inode, path);
2566 * If there is still a entry in the leaf node, check to see if
2567 * it references the partial cluster. This is the only place
2568 * where it could; if it doesn't, we can free the cluster.
2570 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2571 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2572 *partial_cluster)) {
2573 int flags = EXT4_FREE_BLOCKS_FORGET;
2575 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2576 flags |= EXT4_FREE_BLOCKS_METADATA;
2578 ext4_free_blocks(handle, inode, NULL,
2579 EXT4_C2B(sbi, *partial_cluster),
2580 sbi->s_cluster_ratio, flags);
2581 *partial_cluster = 0;
2584 /* if this leaf is free, then we should
2585 * remove it from index block above */
2586 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2587 err = ext4_ext_rm_idx(handle, inode, path + depth);
2594 * ext4_ext_more_to_rm:
2595 * returns 1 if current index has to be freed (even partial)
2598 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2600 BUG_ON(path->p_idx == NULL);
2602 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2606 * if truncate on deeper level happened, it wasn't partial,
2607 * so we have to consider current index for truncation
2609 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2614 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2617 struct super_block *sb = inode->i_sb;
2618 int depth = ext_depth(inode);
2619 struct ext4_ext_path *path = NULL;
2620 ext4_fsblk_t partial_cluster = 0;
2624 ext_debug("truncate since %u to %u\n", start, end);
2626 /* probably first extent we're gonna free will be last in block */
2627 handle = ext4_journal_start(inode, depth + 1);
2629 return PTR_ERR(handle);
2632 ext4_ext_invalidate_cache(inode);
2634 trace_ext4_ext_remove_space(inode, start, depth);
2637 * Check if we are removing extents inside the extent tree. If that
2638 * is the case, we are going to punch a hole inside the extent tree
2639 * so we have to check whether we need to split the extent covering
2640 * the last block to remove so we can easily remove the part of it
2641 * in ext4_ext_rm_leaf().
2643 if (end < EXT_MAX_BLOCKS - 1) {
2644 struct ext4_extent *ex;
2645 ext4_lblk_t ee_block;
2647 /* find extent for this block */
2648 path = ext4_ext_find_extent(inode, end, NULL);
2650 ext4_journal_stop(handle);
2651 return PTR_ERR(path);
2653 depth = ext_depth(inode);
2654 ex = path[depth].p_ext;
2656 ext4_ext_drop_refs(path);
2662 ee_block = le32_to_cpu(ex->ee_block);
2665 * See if the last block is inside the extent, if so split
2666 * the extent at 'end' block so we can easily remove the
2667 * tail of the first part of the split extent in
2668 * ext4_ext_rm_leaf().
2670 if (end >= ee_block &&
2671 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2674 if (ext4_ext_is_uninitialized(ex))
2675 split_flag = EXT4_EXT_MARK_UNINIT1 |
2676 EXT4_EXT_MARK_UNINIT2;
2679 * Split the extent in two so that 'end' is the last
2680 * block in the first new extent
2682 err = ext4_split_extent_at(handle, inode, path,
2683 end + 1, split_flag,
2684 EXT4_GET_BLOCKS_PRE_IO |
2685 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2694 * We start scanning from right side, freeing all the blocks
2695 * after i_size and walking into the tree depth-wise.
2697 depth = ext_depth(inode);
2702 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2704 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2707 ext4_journal_stop(handle);
2710 path[0].p_depth = depth;
2711 path[0].p_hdr = ext_inode_hdr(inode);
2714 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2721 while (i >= 0 && err == 0) {
2723 /* this is leaf block */
2724 err = ext4_ext_rm_leaf(handle, inode, path,
2725 &partial_cluster, start,
2727 /* root level has p_bh == NULL, brelse() eats this */
2728 brelse(path[i].p_bh);
2729 path[i].p_bh = NULL;
2734 /* this is index block */
2735 if (!path[i].p_hdr) {
2736 ext_debug("initialize header\n");
2737 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2740 if (!path[i].p_idx) {
2741 /* this level hasn't been touched yet */
2742 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2743 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2744 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2746 le16_to_cpu(path[i].p_hdr->eh_entries));
2748 /* we were already here, see at next index */
2752 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2753 i, EXT_FIRST_INDEX(path[i].p_hdr),
2755 if (ext4_ext_more_to_rm(path + i)) {
2756 struct buffer_head *bh;
2757 /* go to the next level */
2758 ext_debug("move to level %d (block %llu)\n",
2759 i + 1, ext4_idx_pblock(path[i].p_idx));
2760 memset(path + i + 1, 0, sizeof(*path));
2761 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2763 /* should we reset i_size? */
2767 if (WARN_ON(i + 1 > depth)) {
2771 if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2772 depth - i - 1, bh)) {
2776 path[i + 1].p_bh = bh;
2778 /* save actual number of indexes since this
2779 * number is changed at the next iteration */
2780 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2783 /* we finished processing this index, go up */
2784 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2785 /* index is empty, remove it;
2786 * handle must be already prepared by the
2787 * truncatei_leaf() */
2788 err = ext4_ext_rm_idx(handle, inode, path + i);
2790 /* root level has p_bh == NULL, brelse() eats this */
2791 brelse(path[i].p_bh);
2792 path[i].p_bh = NULL;
2794 ext_debug("return to level %d\n", i);
2798 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2799 path->p_hdr->eh_entries);
2801 /* If we still have something in the partial cluster and we have removed
2802 * even the first extent, then we should free the blocks in the partial
2803 * cluster as well. */
2804 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2805 int flags = EXT4_FREE_BLOCKS_FORGET;
2807 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2808 flags |= EXT4_FREE_BLOCKS_METADATA;
2810 ext4_free_blocks(handle, inode, NULL,
2811 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2812 EXT4_SB(sb)->s_cluster_ratio, flags);
2813 partial_cluster = 0;
2816 /* TODO: flexible tree reduction should be here */
2817 if (path->p_hdr->eh_entries == 0) {
2819 * truncate to zero freed all the tree,
2820 * so we need to correct eh_depth
2822 err = ext4_ext_get_access(handle, inode, path);
2824 ext_inode_hdr(inode)->eh_depth = 0;
2825 ext_inode_hdr(inode)->eh_max =
2826 cpu_to_le16(ext4_ext_space_root(inode, 0));
2827 err = ext4_ext_dirty(handle, inode, path);
2831 ext4_ext_drop_refs(path);
2833 if (err == -EAGAIN) {
2837 ext4_journal_stop(handle);
2843 * called at mount time
2845 void ext4_ext_init(struct super_block *sb)
2848 * possible initialization would be here
2851 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2852 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2853 printk(KERN_INFO "EXT4-fs: file extents enabled"
2854 #ifdef AGGRESSIVE_TEST
2855 ", aggressive tests"
2857 #ifdef CHECK_BINSEARCH
2860 #ifdef EXTENTS_STATS
2865 #ifdef EXTENTS_STATS
2866 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2867 EXT4_SB(sb)->s_ext_min = 1 << 30;
2868 EXT4_SB(sb)->s_ext_max = 0;
2874 * called at umount time
2876 void ext4_ext_release(struct super_block *sb)
2878 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2881 #ifdef EXTENTS_STATS
2882 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2883 struct ext4_sb_info *sbi = EXT4_SB(sb);
2884 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2885 sbi->s_ext_blocks, sbi->s_ext_extents,
2886 sbi->s_ext_blocks / sbi->s_ext_extents);
2887 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2888 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2893 /* FIXME!! we need to try to merge to left or right after zero-out */
2894 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2896 ext4_fsblk_t ee_pblock;
2897 unsigned int ee_len;
2900 ee_len = ext4_ext_get_actual_len(ex);
2901 ee_pblock = ext4_ext_pblock(ex);
2903 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2911 * ext4_split_extent_at() splits an extent at given block.
2913 * @handle: the journal handle
2914 * @inode: the file inode
2915 * @path: the path to the extent
2916 * @split: the logical block where the extent is splitted.
2917 * @split_flags: indicates if the extent could be zeroout if split fails, and
2918 * the states(init or uninit) of new extents.
2919 * @flags: flags used to insert new extent to extent tree.
2922 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2923 * of which are deterimined by split_flag.
2925 * There are two cases:
2926 * a> the extent are splitted into two extent.
2927 * b> split is not needed, and just mark the extent.
2929 * return 0 on success.
2931 static int ext4_split_extent_at(handle_t *handle,
2932 struct inode *inode,
2933 struct ext4_ext_path *path,
2938 ext4_fsblk_t newblock;
2939 ext4_lblk_t ee_block;
2940 struct ext4_extent *ex, newex, orig_ex;
2941 struct ext4_extent *ex2 = NULL;
2942 unsigned int ee_len, depth;
2945 ext_debug("ext4_split_extents_at: inode %lu, logical"
2946 "block %llu\n", inode->i_ino, (unsigned long long)split);
2948 ext4_ext_show_leaf(inode, path);
2950 depth = ext_depth(inode);
2951 ex = path[depth].p_ext;
2952 ee_block = le32_to_cpu(ex->ee_block);
2953 ee_len = ext4_ext_get_actual_len(ex);
2954 newblock = split - ee_block + ext4_ext_pblock(ex);
2956 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2958 err = ext4_ext_get_access(handle, inode, path + depth);
2962 if (split == ee_block) {
2964 * case b: block @split is the block that the extent begins with
2965 * then we just change the state of the extent, and splitting
2968 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2969 ext4_ext_mark_uninitialized(ex);
2971 ext4_ext_mark_initialized(ex);
2973 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2974 ext4_ext_try_to_merge(handle, inode, path, ex);
2976 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2981 memcpy(&orig_ex, ex, sizeof(orig_ex));
2982 ex->ee_len = cpu_to_le16(split - ee_block);
2983 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2984 ext4_ext_mark_uninitialized(ex);
2987 * path may lead to new leaf, not to original leaf any more
2988 * after ext4_ext_insert_extent() returns,
2990 err = ext4_ext_dirty(handle, inode, path + depth);
2992 goto fix_extent_len;
2995 ex2->ee_block = cpu_to_le32(split);
2996 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2997 ext4_ext_store_pblock(ex2, newblock);
2998 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2999 ext4_ext_mark_uninitialized(ex2);
3001 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3002 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3003 err = ext4_ext_zeroout(inode, &orig_ex);
3005 goto fix_extent_len;
3006 /* update the extent length and mark as initialized */
3007 ex->ee_len = cpu_to_le16(ee_len);
3008 ext4_ext_try_to_merge(handle, inode, path, ex);
3009 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3012 goto fix_extent_len;
3015 ext4_ext_show_leaf(inode, path);
3019 ex->ee_len = orig_ex.ee_len;
3020 ext4_ext_dirty(handle, inode, path + depth);
3025 * ext4_split_extents() splits an extent and mark extent which is covered
3026 * by @map as split_flags indicates
3028 * It may result in splitting the extent into multiple extents (upto three)
3029 * There are three possibilities:
3030 * a> There is no split required
3031 * b> Splits in two extents: Split is happening at either end of the extent
3032 * c> Splits in three extents: Somone is splitting in middle of the extent
3035 static int ext4_split_extent(handle_t *handle,
3036 struct inode *inode,
3037 struct ext4_ext_path *path,
3038 struct ext4_map_blocks *map,
3042 ext4_lblk_t ee_block;
3043 struct ext4_extent *ex;
3044 unsigned int ee_len, depth;
3047 int split_flag1, flags1;
3049 depth = ext_depth(inode);
3050 ex = path[depth].p_ext;
3051 ee_block = le32_to_cpu(ex->ee_block);
3052 ee_len = ext4_ext_get_actual_len(ex);
3053 uninitialized = ext4_ext_is_uninitialized(ex);
3055 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3056 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
3057 EXT4_EXT_MAY_ZEROOUT : 0;
3058 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3060 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3061 EXT4_EXT_MARK_UNINIT2;
3062 err = ext4_split_extent_at(handle, inode, path,
3063 map->m_lblk + map->m_len, split_flag1, flags1);
3068 ext4_ext_drop_refs(path);
3069 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3071 return PTR_ERR(path);
3073 if (map->m_lblk >= ee_block) {
3074 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
3075 EXT4_EXT_MAY_ZEROOUT : 0;
3077 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3078 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3079 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3080 err = ext4_split_extent_at(handle, inode, path,
3081 map->m_lblk, split_flag1, flags);
3086 ext4_ext_show_leaf(inode, path);
3088 return err ? err : map->m_len;
3092 * This function is called by ext4_ext_map_blocks() if someone tries to write
3093 * to an uninitialized extent. It may result in splitting the uninitialized
3094 * extent into multiple extents (up to three - one initialized and two
3096 * There are three possibilities:
3097 * a> There is no split required: Entire extent should be initialized
3098 * b> Splits in two extents: Write is happening at either end of the extent
3099 * c> Splits in three extents: Somone is writing in middle of the extent
3102 * - The extent pointed to by 'path' is uninitialized.
3103 * - The extent pointed to by 'path' contains a superset
3104 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3106 * Post-conditions on success:
3107 * - the returned value is the number of blocks beyond map->l_lblk
3108 * that are allocated and initialized.
3109 * It is guaranteed to be >= map->m_len.
3111 static int ext4_ext_convert_to_initialized(handle_t *handle,
3112 struct inode *inode,
3113 struct ext4_map_blocks *map,
3114 struct ext4_ext_path *path)
3116 struct ext4_sb_info *sbi;
3117 struct ext4_extent_header *eh;
3118 struct ext4_map_blocks split_map;
3119 struct ext4_extent zero_ex;
3120 struct ext4_extent *ex;
3121 ext4_lblk_t ee_block, eof_block;
3122 unsigned int ee_len, depth;
3123 int allocated, max_zeroout = 0;
3127 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3128 "block %llu, max_blocks %u\n", inode->i_ino,
3129 (unsigned long long)map->m_lblk, map->m_len);
3131 sbi = EXT4_SB(inode->i_sb);
3132 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3133 inode->i_sb->s_blocksize_bits;
3134 if (eof_block < map->m_lblk + map->m_len)
3135 eof_block = map->m_lblk + map->m_len;
3137 depth = ext_depth(inode);
3138 eh = path[depth].p_hdr;
3139 ex = path[depth].p_ext;
3140 ee_block = le32_to_cpu(ex->ee_block);
3141 ee_len = ext4_ext_get_actual_len(ex);
3142 allocated = ee_len - (map->m_lblk - ee_block);
3144 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3146 /* Pre-conditions */
3147 BUG_ON(!ext4_ext_is_uninitialized(ex));
3148 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3151 * Attempt to transfer newly initialized blocks from the currently
3152 * uninitialized extent to its left neighbor. This is much cheaper
3153 * than an insertion followed by a merge as those involve costly
3154 * memmove() calls. This is the common case in steady state for
3155 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3158 * Limitations of the current logic:
3159 * - L1: we only deal with writes at the start of the extent.
3160 * The approach could be extended to writes at the end
3161 * of the extent but this scenario was deemed less common.
3162 * - L2: we do not deal with writes covering the whole extent.
3163 * This would require removing the extent if the transfer
3165 * - L3: we only attempt to merge with an extent stored in the
3166 * same extent tree node.
3168 if ((map->m_lblk == ee_block) && /*L1*/
3169 (map->m_len < ee_len) && /*L2*/
3170 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3171 struct ext4_extent *prev_ex;
3172 ext4_lblk_t prev_lblk;
3173 ext4_fsblk_t prev_pblk, ee_pblk;
3174 unsigned int prev_len, write_len;
3177 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3178 prev_len = ext4_ext_get_actual_len(prev_ex);
3179 prev_pblk = ext4_ext_pblock(prev_ex);
3180 ee_pblk = ext4_ext_pblock(ex);
3181 write_len = map->m_len;
3184 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3185 * upon those conditions:
3186 * - C1: prev_ex is initialized,
3187 * - C2: prev_ex is logically abutting ex,
3188 * - C3: prev_ex is physically abutting ex,
3189 * - C4: prev_ex can receive the additional blocks without
3190 * overflowing the (initialized) length limit.
3192 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3193 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3194 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3195 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3196 err = ext4_ext_get_access(handle, inode, path + depth);
3200 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3203 /* Shift the start of ex by 'write_len' blocks */
3204 ex->ee_block = cpu_to_le32(ee_block + write_len);
3205 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3206 ex->ee_len = cpu_to_le16(ee_len - write_len);
3207 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3209 /* Extend prev_ex by 'write_len' blocks */
3210 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3212 /* Mark the block containing both extents as dirty */
3213 ext4_ext_dirty(handle, inode, path + depth);
3215 /* Update path to point to the right extent */
3216 path[depth].p_ext = prev_ex;
3218 /* Result: number of initialized blocks past m_lblk */
3219 allocated = write_len;
3224 WARN_ON(map->m_lblk < ee_block);
3226 * It is safe to convert extent to initialized via explicit
3227 * zeroout only if extent is fully insde i_size or new_size.
3229 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3231 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3232 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3233 inode->i_sb->s_blocksize_bits;
3235 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3236 if (max_zeroout && (ee_len <= max_zeroout)) {
3237 err = ext4_ext_zeroout(inode, ex);
3241 err = ext4_ext_get_access(handle, inode, path + depth);
3244 ext4_ext_mark_initialized(ex);
3245 ext4_ext_try_to_merge(handle, inode, path, ex);
3246 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3252 * 1. split the extent into three extents.
3253 * 2. split the extent into two extents, zeroout the first half.
3254 * 3. split the extent into two extents, zeroout the second half.
3255 * 4. split the extent into two extents with out zeroout.
3257 split_map.m_lblk = map->m_lblk;
3258 split_map.m_len = map->m_len;
3260 if (max_zeroout && (allocated > map->m_len)) {
3261 if (allocated <= max_zeroout) {
3264 cpu_to_le32(map->m_lblk);
3265 zero_ex.ee_len = cpu_to_le16(allocated);
3266 ext4_ext_store_pblock(&zero_ex,
3267 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3268 err = ext4_ext_zeroout(inode, &zero_ex);
3271 split_map.m_lblk = map->m_lblk;
3272 split_map.m_len = allocated;
3273 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3275 if (map->m_lblk != ee_block) {
3276 zero_ex.ee_block = ex->ee_block;
3277 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3279 ext4_ext_store_pblock(&zero_ex,
3280 ext4_ext_pblock(ex));
3281 err = ext4_ext_zeroout(inode, &zero_ex);
3286 split_map.m_lblk = ee_block;
3287 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3288 allocated = map->m_len;
3292 allocated = ext4_split_extent(handle, inode, path,
3293 &split_map, split_flag, 0);
3298 return err ? err : allocated;
3302 * This function is called by ext4_ext_map_blocks() from
3303 * ext4_get_blocks_dio_write() when DIO to write
3304 * to an uninitialized extent.
3306 * Writing to an uninitialized extent may result in splitting the uninitialized
3307 * extent into multiple initialized/uninitialized extents (up to three)
3308 * There are three possibilities:
3309 * a> There is no split required: Entire extent should be uninitialized
3310 * b> Splits in two extents: Write is happening at either end of the extent
3311 * c> Splits in three extents: Somone is writing in middle of the extent
3313 * One of more index blocks maybe needed if the extent tree grow after
3314 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3315 * complete, we need to split the uninitialized extent before DIO submit
3316 * the IO. The uninitialized extent called at this time will be split
3317 * into three uninitialized extent(at most). After IO complete, the part
3318 * being filled will be convert to initialized by the end_io callback function
3319 * via ext4_convert_unwritten_extents().
3321 * Returns the size of uninitialized extent to be written on success.
3323 static int ext4_split_unwritten_extents(handle_t *handle,
3324 struct inode *inode,
3325 struct ext4_map_blocks *map,
3326 struct ext4_ext_path *path,
3329 ext4_lblk_t eof_block;
3330 ext4_lblk_t ee_block;
3331 struct ext4_extent *ex;
3332 unsigned int ee_len;
3333 int split_flag = 0, depth;
3335 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3336 "block %llu, max_blocks %u\n", inode->i_ino,
3337 (unsigned long long)map->m_lblk, map->m_len);
3339 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3340 inode->i_sb->s_blocksize_bits;
3341 if (eof_block < map->m_lblk + map->m_len)
3342 eof_block = map->m_lblk + map->m_len;
3344 * It is safe to convert extent to initialized via explicit
3345 * zeroout only if extent is fully insde i_size or new_size.
3347 depth = ext_depth(inode);
3348 ex = path[depth].p_ext;
3349 ee_block = le32_to_cpu(ex->ee_block);
3350 ee_len = ext4_ext_get_actual_len(ex);
3352 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3353 split_flag |= EXT4_EXT_MARK_UNINIT2;
3355 flags |= EXT4_GET_BLOCKS_PRE_IO;
3356 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3359 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3360 struct inode *inode,
3361 struct ext4_ext_path *path)
3363 struct ext4_extent *ex;
3367 depth = ext_depth(inode);
3368 ex = path[depth].p_ext;
3370 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3371 "block %llu, max_blocks %u\n", inode->i_ino,
3372 (unsigned long long)le32_to_cpu(ex->ee_block),
3373 ext4_ext_get_actual_len(ex));
3375 err = ext4_ext_get_access(handle, inode, path + depth);
3378 /* first mark the extent as initialized */
3379 ext4_ext_mark_initialized(ex);
3381 /* note: ext4_ext_correct_indexes() isn't needed here because
3382 * borders are not changed
3384 ext4_ext_try_to_merge(handle, inode, path, ex);
3386 /* Mark modified extent as dirty */
3387 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3389 ext4_ext_show_leaf(inode, path);
3393 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3394 sector_t block, int count)
3397 for (i = 0; i < count; i++)
3398 unmap_underlying_metadata(bdev, block + i);
3402 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3404 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3406 struct ext4_ext_path *path,
3410 struct ext4_extent_header *eh;
3411 struct ext4_extent *last_ex;
3413 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3416 depth = ext_depth(inode);
3417 eh = path[depth].p_hdr;
3420 * We're going to remove EOFBLOCKS_FL entirely in future so we
3421 * do not care for this case anymore. Simply remove the flag
3422 * if there are no extents.
3424 if (unlikely(!eh->eh_entries))
3426 last_ex = EXT_LAST_EXTENT(eh);
3428 * We should clear the EOFBLOCKS_FL flag if we are writing the
3429 * last block in the last extent in the file. We test this by
3430 * first checking to see if the caller to
3431 * ext4_ext_get_blocks() was interested in the last block (or
3432 * a block beyond the last block) in the current extent. If
3433 * this turns out to be false, we can bail out from this
3434 * function immediately.
3436 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3437 ext4_ext_get_actual_len(last_ex))
3440 * If the caller does appear to be planning to write at or
3441 * beyond the end of the current extent, we then test to see
3442 * if the current extent is the last extent in the file, by
3443 * checking to make sure it was reached via the rightmost node
3444 * at each level of the tree.
3446 for (i = depth-1; i >= 0; i--)
3447 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3450 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3451 return ext4_mark_inode_dirty(handle, inode);
3455 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3457 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3458 * whether there are any buffers marked for delayed allocation. It returns '1'
3459 * on the first delalloc'ed buffer head found. If no buffer head in the given
3460 * range is marked for delalloc, it returns 0.
3461 * lblk_start should always be <= lblk_end.
3462 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3463 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3464 * block sooner). This is useful when blocks are truncated sequentially from
3465 * lblk_start towards lblk_end.
3467 static int ext4_find_delalloc_range(struct inode *inode,
3468 ext4_lblk_t lblk_start,
3469 ext4_lblk_t lblk_end,
3470 int search_hint_reverse)
3472 struct address_space *mapping = inode->i_mapping;
3473 struct buffer_head *head, *bh = NULL;
3475 ext4_lblk_t i, pg_lblk;
3478 if (!test_opt(inode->i_sb, DELALLOC))
3481 /* reverse search wont work if fs block size is less than page size */
3482 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3483 search_hint_reverse = 0;
3485 if (search_hint_reverse)
3490 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3492 while ((i >= lblk_start) && (i <= lblk_end)) {
3493 page = find_get_page(mapping, index);
3497 if (!page_has_buffers(page))
3500 head = page_buffers(page);
3505 pg_lblk = index << (PAGE_CACHE_SHIFT -
3508 if (unlikely(pg_lblk < lblk_start)) {
3510 * This is possible when fs block size is less
3511 * than page size and our cluster starts/ends in
3512 * middle of the page. So we need to skip the
3513 * initial few blocks till we reach the 'lblk'
3519 /* Check if the buffer is delayed allocated and that it
3520 * is not yet mapped. (when da-buffers are mapped during
3521 * their writeout, their da_mapped bit is set.)
3523 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3524 page_cache_release(page);
3525 trace_ext4_find_delalloc_range(inode,
3526 lblk_start, lblk_end,
3527 search_hint_reverse,
3531 if (search_hint_reverse)
3535 } while ((i >= lblk_start) && (i <= lblk_end) &&
3536 ((bh = bh->b_this_page) != head));
3539 page_cache_release(page);
3541 * Move to next page. 'i' will be the first lblk in the next
3544 if (search_hint_reverse)
3548 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3551 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3552 search_hint_reverse, 0, 0);
3556 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3557 int search_hint_reverse)
3559 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3560 ext4_lblk_t lblk_start, lblk_end;
3561 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3562 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3564 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3565 search_hint_reverse);
3569 * Determines how many complete clusters (out of those specified by the 'map')
3570 * are under delalloc and were reserved quota for.
3571 * This function is called when we are writing out the blocks that were
3572 * originally written with their allocation delayed, but then the space was
3573 * allocated using fallocate() before the delayed allocation could be resolved.
3574 * The cases to look for are:
3575 * ('=' indicated delayed allocated blocks
3576 * '-' indicates non-delayed allocated blocks)
3577 * (a) partial clusters towards beginning and/or end outside of allocated range
3578 * are not delalloc'ed.
3580 * |----c---=|====c====|====c====|===-c----|
3581 * |++++++ allocated ++++++|
3582 * ==> 4 complete clusters in above example
3584 * (b) partial cluster (outside of allocated range) towards either end is
3585 * marked for delayed allocation. In this case, we will exclude that
3588 * |----====c========|========c========|
3589 * |++++++ allocated ++++++|
3590 * ==> 1 complete clusters in above example
3593 * |================c================|
3594 * |++++++ allocated ++++++|
3595 * ==> 0 complete clusters in above example
3597 * The ext4_da_update_reserve_space will be called only if we
3598 * determine here that there were some "entire" clusters that span
3599 * this 'allocated' range.
3600 * In the non-bigalloc case, this function will just end up returning num_blks
3601 * without ever calling ext4_find_delalloc_range.
3604 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3605 unsigned int num_blks)
3607 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3608 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3609 ext4_lblk_t lblk_from, lblk_to, c_offset;
3610 unsigned int allocated_clusters = 0;
3612 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3613 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3615 /* max possible clusters for this allocation */
3616 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3618 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3620 /* Check towards left side */
3621 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3623 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3624 lblk_to = lblk_from + c_offset - 1;
3626 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3627 allocated_clusters--;
3630 /* Now check towards right. */
3631 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3632 if (allocated_clusters && c_offset) {
3633 lblk_from = lblk_start + num_blks;
3634 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3636 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3637 allocated_clusters--;
3640 return allocated_clusters;
3644 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3645 struct ext4_map_blocks *map,
3646 struct ext4_ext_path *path, int flags,
3647 unsigned int allocated, ext4_fsblk_t newblock)
3651 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3653 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3654 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3655 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3657 ext4_ext_show_leaf(inode, path);
3659 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3662 /* get_block() before submit the IO, split the extent */
3663 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3664 ret = ext4_split_unwritten_extents(handle, inode, map,
3667 * Flag the inode(non aio case) or end_io struct (aio case)
3668 * that this IO needs to conversion to written when IO is
3672 ext4_set_io_unwritten_flag(inode, io);
3674 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3675 if (ext4_should_dioread_nolock(inode))
3676 map->m_flags |= EXT4_MAP_UNINIT;
3679 /* IO end_io complete, convert the filled extent to written */
3680 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3681 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3684 ext4_update_inode_fsync_trans(handle, inode, 1);
3685 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3691 /* buffered IO case */
3693 * repeat fallocate creation request
3694 * we already have an unwritten extent
3696 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3699 /* buffered READ or buffered write_begin() lookup */
3700 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3702 * We have blocks reserved already. We
3703 * return allocated blocks so that delalloc
3704 * won't do block reservation for us. But
3705 * the buffer head will be unmapped so that
3706 * a read from the block returns 0s.
3708 map->m_flags |= EXT4_MAP_UNWRITTEN;
3712 /* buffered write, writepage time, convert*/
3713 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3715 ext4_update_inode_fsync_trans(handle, inode, 1);
3722 map->m_flags |= EXT4_MAP_NEW;
3724 * if we allocated more blocks than requested
3725 * we need to make sure we unmap the extra block
3726 * allocated. The actual needed block will get
3727 * unmapped later when we find the buffer_head marked
3730 if (allocated > map->m_len) {
3731 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3732 newblock + map->m_len,
3733 allocated - map->m_len);
3734 allocated = map->m_len;
3738 * If we have done fallocate with the offset that is already
3739 * delayed allocated, we would have block reservation
3740 * and quota reservation done in the delayed write path.
3741 * But fallocate would have already updated quota and block
3742 * count for this offset. So cancel these reservation
3744 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3745 unsigned int reserved_clusters;
3746 reserved_clusters = get_reserved_cluster_alloc(inode,
3747 map->m_lblk, map->m_len);
3748 if (reserved_clusters)
3749 ext4_da_update_reserve_space(inode,
3755 map->m_flags |= EXT4_MAP_MAPPED;
3756 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3757 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3763 if (allocated > map->m_len)
3764 allocated = map->m_len;
3765 ext4_ext_show_leaf(inode, path);
3766 map->m_pblk = newblock;
3767 map->m_len = allocated;
3770 ext4_ext_drop_refs(path);
3773 return err ? err : allocated;
3777 * get_implied_cluster_alloc - check to see if the requested
3778 * allocation (in the map structure) overlaps with a cluster already
3779 * allocated in an extent.
3780 * @sb The filesystem superblock structure
3781 * @map The requested lblk->pblk mapping
3782 * @ex The extent structure which might contain an implied
3783 * cluster allocation
3785 * This function is called by ext4_ext_map_blocks() after we failed to
3786 * find blocks that were already in the inode's extent tree. Hence,
3787 * we know that the beginning of the requested region cannot overlap
3788 * the extent from the inode's extent tree. There are three cases we
3789 * want to catch. The first is this case:
3791 * |--- cluster # N--|
3792 * |--- extent ---| |---- requested region ---|
3795 * The second case that we need to test for is this one:
3797 * |--------- cluster # N ----------------|
3798 * |--- requested region --| |------- extent ----|
3799 * |=======================|
3801 * The third case is when the requested region lies between two extents
3802 * within the same cluster:
3803 * |------------- cluster # N-------------|
3804 * |----- ex -----| |---- ex_right ----|
3805 * |------ requested region ------|
3806 * |================|
3808 * In each of the above cases, we need to set the map->m_pblk and
3809 * map->m_len so it corresponds to the return the extent labelled as
3810 * "|====|" from cluster #N, since it is already in use for data in
3811 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3812 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3813 * as a new "allocated" block region. Otherwise, we will return 0 and
3814 * ext4_ext_map_blocks() will then allocate one or more new clusters
3815 * by calling ext4_mb_new_blocks().
3817 static int get_implied_cluster_alloc(struct super_block *sb,
3818 struct ext4_map_blocks *map,
3819 struct ext4_extent *ex,
3820 struct ext4_ext_path *path)
3822 struct ext4_sb_info *sbi = EXT4_SB(sb);
3823 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3824 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3825 ext4_lblk_t rr_cluster_start;
3826 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3827 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3828 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3830 /* The extent passed in that we are trying to match */
3831 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3832 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3834 /* The requested region passed into ext4_map_blocks() */
3835 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3837 if ((rr_cluster_start == ex_cluster_end) ||
3838 (rr_cluster_start == ex_cluster_start)) {
3839 if (rr_cluster_start == ex_cluster_end)
3840 ee_start += ee_len - 1;
3841 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3843 map->m_len = min(map->m_len,
3844 (unsigned) sbi->s_cluster_ratio - c_offset);
3846 * Check for and handle this case:
3848 * |--------- cluster # N-------------|
3849 * |------- extent ----|
3850 * |--- requested region ---|
3854 if (map->m_lblk < ee_block)
3855 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3858 * Check for the case where there is already another allocated
3859 * block to the right of 'ex' but before the end of the cluster.
3861 * |------------- cluster # N-------------|
3862 * |----- ex -----| |---- ex_right ----|
3863 * |------ requested region ------|
3864 * |================|
3866 if (map->m_lblk > ee_block) {
3867 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3868 map->m_len = min(map->m_len, next - map->m_lblk);
3871 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3875 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3881 * Block allocation/map/preallocation routine for extents based files
3884 * Need to be called with
3885 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3886 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3888 * return > 0, number of of blocks already mapped/allocated
3889 * if create == 0 and these are pre-allocated blocks
3890 * buffer head is unmapped
3891 * otherwise blocks are mapped
3893 * return = 0, if plain look up failed (blocks have not been allocated)
3894 * buffer head is unmapped
3896 * return < 0, error case.
3898 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3899 struct ext4_map_blocks *map, int flags)
3901 struct ext4_ext_path *path = NULL;
3902 struct ext4_extent newex, *ex, *ex2;
3903 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3904 ext4_fsblk_t newblock = 0;
3905 int free_on_err = 0, err = 0, depth, ret;
3906 unsigned int allocated = 0, offset = 0;
3907 unsigned int allocated_clusters = 0;
3908 struct ext4_allocation_request ar;
3909 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3910 ext4_lblk_t cluster_offset;
3912 ext_debug("blocks %u/%u requested for inode %lu\n",
3913 map->m_lblk, map->m_len, inode->i_ino);
3914 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3916 /* check in cache */
3917 if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3918 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3919 if ((sbi->s_cluster_ratio > 1) &&
3920 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3921 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3923 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3925 * block isn't allocated yet and
3926 * user doesn't want to allocate it
3930 /* we should allocate requested block */
3932 /* block is already allocated */
3933 if (sbi->s_cluster_ratio > 1)
3934 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3935 newblock = map->m_lblk
3936 - le32_to_cpu(newex.ee_block)
3937 + ext4_ext_pblock(&newex);
3938 /* number of remaining blocks in the extent */
3939 allocated = ext4_ext_get_actual_len(&newex) -
3940 (map->m_lblk - le32_to_cpu(newex.ee_block));
3945 /* find extent for this block */
3946 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3948 err = PTR_ERR(path);
3953 depth = ext_depth(inode);
3956 * consistent leaf must not be empty;
3957 * this situation is possible, though, _during_ tree modification;
3958 * this is why assert can't be put in ext4_ext_find_extent()
3960 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3961 EXT4_ERROR_INODE(inode, "bad extent address "
3962 "lblock: %lu, depth: %d pblock %lld",
3963 (unsigned long) map->m_lblk, depth,
3964 path[depth].p_block);
3969 ex = path[depth].p_ext;
3971 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3972 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3973 unsigned short ee_len;
3976 * Uninitialized extents are treated as holes, except that
3977 * we split out initialized portions during a write.
3979 ee_len = ext4_ext_get_actual_len(ex);
3981 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3983 /* if found extent covers block, simply return it */
3984 if (in_range(map->m_lblk, ee_block, ee_len)) {
3985 newblock = map->m_lblk - ee_block + ee_start;
3986 /* number of remaining blocks in the extent */
3987 allocated = ee_len - (map->m_lblk - ee_block);
3988 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3989 ee_block, ee_len, newblock);
3992 * Do not put uninitialized extent
3995 if (!ext4_ext_is_uninitialized(ex)) {
3996 ext4_ext_put_in_cache(inode, ee_block,
4000 ret = ext4_ext_handle_uninitialized_extents(
4001 handle, inode, map, path, flags,
4002 allocated, newblock);
4007 if ((sbi->s_cluster_ratio > 1) &&
4008 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
4009 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4012 * requested block isn't allocated yet;
4013 * we couldn't try to create block if create flag is zero
4015 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4017 * put just found gap into cache to speed up
4018 * subsequent requests
4020 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4025 * Okay, we need to do block allocation.
4027 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4028 newex.ee_block = cpu_to_le32(map->m_lblk);
4029 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4032 * If we are doing bigalloc, check to see if the extent returned
4033 * by ext4_ext_find_extent() implies a cluster we can use.
4035 if (cluster_offset && ex &&
4036 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4037 ar.len = allocated = map->m_len;
4038 newblock = map->m_pblk;
4039 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4040 goto got_allocated_blocks;
4043 /* find neighbour allocated blocks */
4044 ar.lleft = map->m_lblk;
4045 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4048 ar.lright = map->m_lblk;
4050 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4054 /* Check if the extent after searching to the right implies a
4055 * cluster we can use. */
4056 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4057 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4058 ar.len = allocated = map->m_len;
4059 newblock = map->m_pblk;
4060 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4061 goto got_allocated_blocks;
4065 * See if request is beyond maximum number of blocks we can have in
4066 * a single extent. For an initialized extent this limit is
4067 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4068 * EXT_UNINIT_MAX_LEN.
4070 if (map->m_len > EXT_INIT_MAX_LEN &&
4071 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4072 map->m_len = EXT_INIT_MAX_LEN;
4073 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4074 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4075 map->m_len = EXT_UNINIT_MAX_LEN;
4077 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4078 newex.ee_len = cpu_to_le16(map->m_len);
4079 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4081 allocated = ext4_ext_get_actual_len(&newex);
4083 allocated = map->m_len;
4085 /* allocate new block */
4087 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4088 ar.logical = map->m_lblk;
4090 * We calculate the offset from the beginning of the cluster
4091 * for the logical block number, since when we allocate a
4092 * physical cluster, the physical block should start at the
4093 * same offset from the beginning of the cluster. This is
4094 * needed so that future calls to get_implied_cluster_alloc()
4097 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4098 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4100 ar.logical -= offset;
4101 if (S_ISREG(inode->i_mode))
4102 ar.flags = EXT4_MB_HINT_DATA;
4104 /* disable in-core preallocation for non-regular files */
4106 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4107 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4108 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4111 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4112 ar.goal, newblock, allocated);
4114 allocated_clusters = ar.len;
4115 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4116 if (ar.len > allocated)
4119 got_allocated_blocks:
4120 /* try to insert new extent into found leaf and return */
4121 ext4_ext_store_pblock(&newex, newblock + offset);
4122 newex.ee_len = cpu_to_le16(ar.len);
4123 /* Mark uninitialized */
4124 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4125 ext4_ext_mark_uninitialized(&newex);
4127 * io_end structure was created for every IO write to an
4128 * uninitialized extent. To avoid unnecessary conversion,
4129 * here we flag the IO that really needs the conversion.
4130 * For non asycn direct IO case, flag the inode state
4131 * that we need to perform conversion when IO is done.
4133 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4135 ext4_set_io_unwritten_flag(inode, io);
4137 ext4_set_inode_state(inode,
4138 EXT4_STATE_DIO_UNWRITTEN);
4140 if (ext4_should_dioread_nolock(inode))
4141 map->m_flags |= EXT4_MAP_UNINIT;
4145 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4146 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4149 err = ext4_ext_insert_extent(handle, inode, path,
4151 if (err && free_on_err) {
4152 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4153 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4154 /* free data blocks we just allocated */
4155 /* not a good idea to call discard here directly,
4156 * but otherwise we'd need to call it every free() */
4157 ext4_discard_preallocations(inode);
4158 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4159 ext4_ext_get_actual_len(&newex), fb_flags);
4163 /* previous routine could use block we allocated */
4164 newblock = ext4_ext_pblock(&newex);
4165 allocated = ext4_ext_get_actual_len(&newex);
4166 if (allocated > map->m_len)
4167 allocated = map->m_len;
4168 map->m_flags |= EXT4_MAP_NEW;
4171 * Update reserved blocks/metadata blocks after successful
4172 * block allocation which had been deferred till now.
4174 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4175 unsigned int reserved_clusters;
4177 * Check how many clusters we had reserved this allocated range
4179 reserved_clusters = get_reserved_cluster_alloc(inode,
4180 map->m_lblk, allocated);
4181 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4182 if (reserved_clusters) {
4184 * We have clusters reserved for this range.
4185 * But since we are not doing actual allocation
4186 * and are simply using blocks from previously
4187 * allocated cluster, we should release the
4188 * reservation and not claim quota.
4190 ext4_da_update_reserve_space(inode,
4191 reserved_clusters, 0);
4194 BUG_ON(allocated_clusters < reserved_clusters);
4195 /* We will claim quota for all newly allocated blocks.*/
4196 ext4_da_update_reserve_space(inode, allocated_clusters,
4198 if (reserved_clusters < allocated_clusters) {
4199 struct ext4_inode_info *ei = EXT4_I(inode);
4200 int reservation = allocated_clusters -
4203 * It seems we claimed few clusters outside of
4204 * the range of this allocation. We should give
4205 * it back to the reservation pool. This can
4206 * happen in the following case:
4208 * * Suppose s_cluster_ratio is 4 (i.e., each
4209 * cluster has 4 blocks. Thus, the clusters
4210 * are [0-3],[4-7],[8-11]...
4211 * * First comes delayed allocation write for
4212 * logical blocks 10 & 11. Since there were no
4213 * previous delayed allocated blocks in the
4214 * range [8-11], we would reserve 1 cluster
4216 * * Next comes write for logical blocks 3 to 8.
4217 * In this case, we will reserve 2 clusters
4218 * (for [0-3] and [4-7]; and not for [8-11] as
4219 * that range has a delayed allocated blocks.
4220 * Thus total reserved clusters now becomes 3.
4221 * * Now, during the delayed allocation writeout
4222 * time, we will first write blocks [3-8] and
4223 * allocate 3 clusters for writing these
4224 * blocks. Also, we would claim all these
4225 * three clusters above.
4226 * * Now when we come here to writeout the
4227 * blocks [10-11], we would expect to claim
4228 * the reservation of 1 cluster we had made
4229 * (and we would claim it since there are no
4230 * more delayed allocated blocks in the range
4231 * [8-11]. But our reserved cluster count had
4232 * already gone to 0.
4234 * Thus, at the step 4 above when we determine
4235 * that there are still some unwritten delayed
4236 * allocated blocks outside of our current
4237 * block range, we should increment the
4238 * reserved clusters count so that when the
4239 * remaining blocks finally gets written, we
4242 dquot_reserve_block(inode,
4243 EXT4_C2B(sbi, reservation));
4244 spin_lock(&ei->i_block_reservation_lock);
4245 ei->i_reserved_data_blocks += reservation;
4246 spin_unlock(&ei->i_block_reservation_lock);
4252 * Cache the extent and update transaction to commit on fdatasync only
4253 * when it is _not_ an uninitialized extent.
4255 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4256 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4257 ext4_update_inode_fsync_trans(handle, inode, 1);
4259 ext4_update_inode_fsync_trans(handle, inode, 0);
4261 if (allocated > map->m_len)
4262 allocated = map->m_len;
4263 ext4_ext_show_leaf(inode, path);
4264 map->m_flags |= EXT4_MAP_MAPPED;
4265 map->m_pblk = newblock;
4266 map->m_len = allocated;
4269 ext4_ext_drop_refs(path);
4273 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4274 newblock, map->m_len, err ? err : allocated);
4276 return err ? err : allocated;
4279 void ext4_ext_truncate(struct inode *inode)
4281 struct address_space *mapping = inode->i_mapping;
4282 struct super_block *sb = inode->i_sb;
4283 ext4_lblk_t last_block;
4289 * finish any pending end_io work so we won't run the risk of
4290 * converting any truncated blocks to initialized later
4292 ext4_flush_completed_IO(inode);
4295 * probably first extent we're gonna free will be last in block
4297 err = ext4_writepage_trans_blocks(inode);
4298 handle = ext4_journal_start(inode, err);
4302 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4303 page_len = PAGE_CACHE_SIZE -
4304 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4306 err = ext4_discard_partial_page_buffers(handle,
4307 mapping, inode->i_size, page_len, 0);
4313 if (ext4_orphan_add(handle, inode))
4316 down_write(&EXT4_I(inode)->i_data_sem);
4317 ext4_ext_invalidate_cache(inode);
4319 ext4_discard_preallocations(inode);
4322 * TODO: optimization is possible here.
4323 * Probably we need not scan at all,
4324 * because page truncation is enough.
4327 /* we have to know where to truncate from in crash case */
4328 EXT4_I(inode)->i_disksize = inode->i_size;
4329 ext4_mark_inode_dirty(handle, inode);
4331 last_block = (inode->i_size + sb->s_blocksize - 1)
4332 >> EXT4_BLOCK_SIZE_BITS(sb);
4333 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4335 /* In a multi-transaction truncate, we only make the final
4336 * transaction synchronous.
4339 ext4_handle_sync(handle);
4341 up_write(&EXT4_I(inode)->i_data_sem);
4345 * If this was a simple ftruncate() and the file will remain alive,
4346 * then we need to clear up the orphan record which we created above.
4347 * However, if this was a real unlink then we were called by
4348 * ext4_delete_inode(), and we allow that function to clean up the
4349 * orphan info for us.
4352 ext4_orphan_del(handle, inode);
4354 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4355 ext4_mark_inode_dirty(handle, inode);
4356 ext4_journal_stop(handle);
4359 static void ext4_falloc_update_inode(struct inode *inode,
4360 int mode, loff_t new_size, int update_ctime)
4362 struct timespec now;
4365 now = current_fs_time(inode->i_sb);
4366 if (!timespec_equal(&inode->i_ctime, &now))
4367 inode->i_ctime = now;
4370 * Update only when preallocation was requested beyond
4373 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4374 if (new_size > i_size_read(inode))
4375 i_size_write(inode, new_size);
4376 if (new_size > EXT4_I(inode)->i_disksize)
4377 ext4_update_i_disksize(inode, new_size);
4380 * Mark that we allocate beyond EOF so the subsequent truncate
4381 * can proceed even if the new size is the same as i_size.
4383 if (new_size > i_size_read(inode))
4384 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4390 * preallocate space for a file. This implements ext4's fallocate file
4391 * operation, which gets called from sys_fallocate system call.
4392 * For block-mapped files, posix_fallocate should fall back to the method
4393 * of writing zeroes to the required new blocks (the same behavior which is
4394 * expected for file systems which do not support fallocate() system call).
4396 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4398 struct inode *inode = file->f_path.dentry->d_inode;
4401 unsigned int max_blocks;
4406 struct ext4_map_blocks map;
4407 unsigned int credits, blkbits = inode->i_blkbits;
4410 * currently supporting (pre)allocate mode for extent-based
4413 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4416 /* Return error if mode is not supported */
4417 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4420 if (mode & FALLOC_FL_PUNCH_HOLE)
4421 return ext4_punch_hole(file, offset, len);
4423 trace_ext4_fallocate_enter(inode, offset, len, mode);
4424 map.m_lblk = offset >> blkbits;
4426 * We can't just convert len to max_blocks because
4427 * If blocksize = 4096 offset = 3072 and len = 2048
4429 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4432 * credits to insert 1 extent into extent tree
4434 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4435 mutex_lock(&inode->i_mutex);
4436 ret = inode_newsize_ok(inode, (len + offset));
4438 mutex_unlock(&inode->i_mutex);
4439 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4442 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4443 if (mode & FALLOC_FL_KEEP_SIZE)
4444 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4446 * Don't normalize the request if it can fit in one extent so
4447 * that it doesn't get unnecessarily split into multiple
4450 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4451 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4453 while (ret >= 0 && ret < max_blocks) {
4454 map.m_lblk = map.m_lblk + ret;
4455 map.m_len = max_blocks = max_blocks - ret;
4456 handle = ext4_journal_start(inode, credits);
4457 if (IS_ERR(handle)) {
4458 ret = PTR_ERR(handle);
4461 ret = ext4_map_blocks(handle, inode, &map, flags);
4465 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4466 "returned error inode#%lu, block=%u, "
4467 "max_blocks=%u", __func__,
4468 inode->i_ino, map.m_lblk, max_blocks);
4470 ext4_mark_inode_dirty(handle, inode);
4471 ret2 = ext4_journal_stop(handle);
4474 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4475 blkbits) >> blkbits))
4476 new_size = offset + len;
4478 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4480 ext4_falloc_update_inode(inode, mode, new_size,
4481 (map.m_flags & EXT4_MAP_NEW));
4482 ext4_mark_inode_dirty(handle, inode);
4483 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4484 ext4_handle_sync(handle);
4485 ret2 = ext4_journal_stop(handle);
4489 if (ret == -ENOSPC &&
4490 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4494 mutex_unlock(&inode->i_mutex);
4495 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4496 ret > 0 ? ret2 : ret);
4497 return ret > 0 ? ret2 : ret;
4501 * This function convert a range of blocks to written extents
4502 * The caller of this function will pass the start offset and the size.
4503 * all unwritten extents within this range will be converted to
4506 * This function is called from the direct IO end io call back
4507 * function, to convert the fallocated extents after IO is completed.
4508 * Returns 0 on success.
4510 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4514 unsigned int max_blocks;
4517 struct ext4_map_blocks map;
4518 unsigned int credits, blkbits = inode->i_blkbits;
4520 map.m_lblk = offset >> blkbits;
4522 * We can't just convert len to max_blocks because
4523 * If blocksize = 4096 offset = 3072 and len = 2048
4525 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4528 * credits to insert 1 extent into extent tree
4530 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4531 while (ret >= 0 && ret < max_blocks) {
4533 map.m_len = (max_blocks -= ret);
4534 handle = ext4_journal_start(inode, credits);
4535 if (IS_ERR(handle)) {
4536 ret = PTR_ERR(handle);
4539 ret = ext4_map_blocks(handle, inode, &map,
4540 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4543 ext4_msg(inode->i_sb, KERN_ERR,
4544 "%s:%d: inode #%lu: block %u: len %u: "
4545 "ext4_ext_map_blocks returned %d",
4546 __func__, __LINE__, inode->i_ino, map.m_lblk,
4549 ext4_mark_inode_dirty(handle, inode);
4550 ret2 = ext4_journal_stop(handle);
4551 if (ret <= 0 || ret2 )
4554 return ret > 0 ? ret2 : ret;
4558 * Callback function called for each extent to gather FIEMAP information.
4560 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4561 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4569 struct fiemap_extent_info *fieinfo = data;
4570 unsigned char blksize_bits;
4572 blksize_bits = inode->i_sb->s_blocksize_bits;
4573 logical = (__u64)newex->ec_block << blksize_bits;
4575 if (newex->ec_start == 0) {
4577 * No extent in extent-tree contains block @newex->ec_start,
4578 * then the block may stay in 1)a hole or 2)delayed-extent.
4580 * Holes or delayed-extents are processed as follows.
4581 * 1. lookup dirty pages with specified range in pagecache.
4582 * If no page is got, then there is no delayed-extent and
4583 * return with EXT_CONTINUE.
4584 * 2. find the 1st mapped buffer,
4585 * 3. check if the mapped buffer is both in the request range
4586 * and a delayed buffer. If not, there is no delayed-extent,
4588 * 4. a delayed-extent is found, the extent will be collected.
4590 ext4_lblk_t end = 0;
4591 pgoff_t last_offset;
4594 pgoff_t start_index = 0;
4595 struct page **pages = NULL;
4596 struct buffer_head *bh = NULL;
4597 struct buffer_head *head = NULL;
4598 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4600 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4604 offset = logical >> PAGE_SHIFT;
4606 last_offset = offset;
4608 ret = find_get_pages_tag(inode->i_mapping, &offset,
4609 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4611 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4612 /* First time, try to find a mapped buffer. */
4615 for (index = 0; index < ret; index++)
4616 page_cache_release(pages[index]);
4619 return EXT_CONTINUE;
4624 /* Try to find the 1st mapped buffer. */
4625 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4627 if (!page_has_buffers(pages[index]))
4629 head = page_buffers(pages[index]);
4636 if (end >= newex->ec_block +
4638 /* The buffer is out of
4639 * the request range.
4643 if (buffer_mapped(bh) &&
4644 end >= newex->ec_block) {
4645 start_index = index - 1;
4646 /* get the 1st mapped buffer. */
4647 goto found_mapped_buffer;
4650 bh = bh->b_this_page;
4652 } while (bh != head);
4654 /* No mapped buffer in the range found in this page,
4655 * We need to look up next page.
4658 /* There is no page left, but we need to limit
4661 newex->ec_len = end - newex->ec_block;
4666 /*Find contiguous delayed buffers. */
4667 if (ret > 0 && pages[0]->index == last_offset)
4668 head = page_buffers(pages[0]);
4674 found_mapped_buffer:
4675 if (bh != NULL && buffer_delay(bh)) {
4676 /* 1st or contiguous delayed buffer found. */
4677 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4679 * 1st delayed buffer found, record
4680 * the start of extent.
4682 flags |= FIEMAP_EXTENT_DELALLOC;
4683 newex->ec_block = end;
4684 logical = (__u64)end << blksize_bits;
4686 /* Find contiguous delayed buffers. */
4688 if (!buffer_delay(bh))
4689 goto found_delayed_extent;
4690 bh = bh->b_this_page;
4692 } while (bh != head);
4694 for (; index < ret; index++) {
4695 if (!page_has_buffers(pages[index])) {
4699 head = page_buffers(pages[index]);
4705 if (pages[index]->index !=
4706 pages[start_index]->index + index
4708 /* Blocks are not contiguous. */
4714 if (!buffer_delay(bh))
4715 /* Delayed-extent ends. */
4716 goto found_delayed_extent;
4717 bh = bh->b_this_page;
4719 } while (bh != head);
4721 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4725 found_delayed_extent:
4726 newex->ec_len = min(end - newex->ec_block,
4727 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4728 if (ret == nr_pages && bh != NULL &&
4729 newex->ec_len < EXT_INIT_MAX_LEN &&
4731 /* Have not collected an extent and continue. */
4732 for (index = 0; index < ret; index++)
4733 page_cache_release(pages[index]);
4737 for (index = 0; index < ret; index++)
4738 page_cache_release(pages[index]);
4742 physical = (__u64)newex->ec_start << blksize_bits;
4743 length = (__u64)newex->ec_len << blksize_bits;
4745 if (ex && ext4_ext_is_uninitialized(ex))
4746 flags |= FIEMAP_EXTENT_UNWRITTEN;
4748 if (next == EXT_MAX_BLOCKS)
4749 flags |= FIEMAP_EXTENT_LAST;
4751 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4757 return EXT_CONTINUE;
4759 /* fiemap flags we can handle specified here */
4760 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4762 static int ext4_xattr_fiemap(struct inode *inode,
4763 struct fiemap_extent_info *fieinfo)
4767 __u32 flags = FIEMAP_EXTENT_LAST;
4768 int blockbits = inode->i_sb->s_blocksize_bits;
4772 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4773 struct ext4_iloc iloc;
4774 int offset; /* offset of xattr in inode */
4776 error = ext4_get_inode_loc(inode, &iloc);
4779 physical = iloc.bh->b_blocknr << blockbits;
4780 offset = EXT4_GOOD_OLD_INODE_SIZE +
4781 EXT4_I(inode)->i_extra_isize;
4783 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4784 flags |= FIEMAP_EXTENT_DATA_INLINE;
4786 } else { /* external block */
4787 physical = EXT4_I(inode)->i_file_acl << blockbits;
4788 length = inode->i_sb->s_blocksize;
4792 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4794 return (error < 0 ? error : 0);
4798 * ext4_ext_punch_hole
4800 * Punches a hole of "length" bytes in a file starting
4803 * @inode: The inode of the file to punch a hole in
4804 * @offset: The starting byte offset of the hole
4805 * @length: The length of the hole
4807 * Returns the number of blocks removed or negative on err
4809 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4811 struct inode *inode = file->f_path.dentry->d_inode;
4812 struct super_block *sb = inode->i_sb;
4813 ext4_lblk_t first_block, stop_block;
4814 struct address_space *mapping = inode->i_mapping;
4816 loff_t first_page, last_page, page_len;
4817 loff_t first_page_offset, last_page_offset;
4818 int credits, err = 0;
4820 /* No need to punch hole beyond i_size */
4821 if (offset >= inode->i_size)
4825 * If the hole extends beyond i_size, set the hole
4826 * to end after the page that contains i_size
4828 if (offset + length > inode->i_size) {
4829 length = inode->i_size +
4830 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4834 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4835 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4837 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4838 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4841 * Write out all dirty pages to avoid race conditions
4842 * Then release them.
4844 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4845 err = filemap_write_and_wait_range(mapping,
4846 offset, offset + length - 1);
4852 /* Now release the pages */
4853 if (last_page_offset > first_page_offset) {
4854 truncate_pagecache_range(inode, first_page_offset,
4855 last_page_offset - 1);
4858 /* finish any pending end_io work */
4859 ext4_flush_completed_IO(inode);
4861 credits = ext4_writepage_trans_blocks(inode);
4862 handle = ext4_journal_start(inode, credits);
4864 return PTR_ERR(handle);
4868 * Now we need to zero out the non-page-aligned data in the
4869 * pages at the start and tail of the hole, and unmap the buffer
4870 * heads for the block aligned regions of the page that were
4871 * completely zeroed.
4873 if (first_page > last_page) {
4875 * If the file space being truncated is contained within a page
4876 * just zero out and unmap the middle of that page
4878 err = ext4_discard_partial_page_buffers(handle,
4879 mapping, offset, length, 0);
4885 * zero out and unmap the partial page that contains
4886 * the start of the hole
4888 page_len = first_page_offset - offset;
4890 err = ext4_discard_partial_page_buffers(handle, mapping,
4891 offset, page_len, 0);
4897 * zero out and unmap the partial page that contains
4898 * the end of the hole
4900 page_len = offset + length - last_page_offset;
4902 err = ext4_discard_partial_page_buffers(handle, mapping,
4903 last_page_offset, page_len, 0);
4910 * If i_size is contained in the last page, we need to
4911 * unmap and zero the partial page after i_size
4913 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4914 inode->i_size % PAGE_CACHE_SIZE != 0) {
4916 page_len = PAGE_CACHE_SIZE -
4917 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4920 err = ext4_discard_partial_page_buffers(handle,
4921 mapping, inode->i_size, page_len, 0);
4928 first_block = (offset + sb->s_blocksize - 1) >>
4929 EXT4_BLOCK_SIZE_BITS(sb);
4930 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4932 /* If there are no blocks to remove, return now */
4933 if (first_block >= stop_block)
4936 down_write(&EXT4_I(inode)->i_data_sem);
4937 ext4_ext_invalidate_cache(inode);
4938 ext4_discard_preallocations(inode);
4940 err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
4942 ext4_ext_invalidate_cache(inode);
4943 ext4_discard_preallocations(inode);
4946 ext4_handle_sync(handle);
4948 up_write(&EXT4_I(inode)->i_data_sem);
4951 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4952 ext4_mark_inode_dirty(handle, inode);
4953 ext4_journal_stop(handle);
4956 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4957 __u64 start, __u64 len)
4959 ext4_lblk_t start_blk;
4962 /* fallback to generic here if not in extents fmt */
4963 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4964 return generic_block_fiemap(inode, fieinfo, start, len,
4967 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4970 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4971 error = ext4_xattr_fiemap(inode, fieinfo);
4973 ext4_lblk_t len_blks;
4976 start_blk = start >> inode->i_sb->s_blocksize_bits;
4977 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4978 if (last_blk >= EXT_MAX_BLOCKS)
4979 last_blk = EXT_MAX_BLOCKS-1;
4980 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4983 * Walk the extent tree gathering extent information.
4984 * ext4_ext_fiemap_cb will push extents back to user.
4986 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4987 ext4_ext_fiemap_cb, fieinfo);