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
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/ext4_jbd2.h>
36 #include <linux/jbd.h>
37 #include <linux/highuid.h>
38 #include <linux/pagemap.h>
39 #include <linux/quotaops.h>
40 #include <linux/string.h>
41 #include <linux/slab.h>
42 #include <linux/falloc.h>
43 #include <linux/ext4_fs_extents.h>
44 #include <asm/uaccess.h>
49 * combine low and high parts of physical block number into ext4_fsblk_t
51 static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
55 block = le32_to_cpu(ex->ee_start);
56 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
62 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
68 block = le32_to_cpu(ix->ei_leaf);
69 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
74 * ext4_ext_store_pblock:
75 * stores a large physical block number into an extent struct,
76 * breaking it into parts
78 static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
80 ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff));
81 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
85 * ext4_idx_store_pblock:
86 * stores a large physical block number into an index struct,
87 * breaking it into parts
89 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
91 ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff));
92 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
95 static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
99 if (handle->h_buffer_credits > needed)
101 if (!ext4_journal_extend(handle, needed))
103 err = ext4_journal_restart(handle, needed);
113 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
114 struct ext4_ext_path *path)
117 /* path points to block */
118 return ext4_journal_get_write_access(handle, path->p_bh);
120 /* path points to leaf/index in inode body */
121 /* we use in-core data, no need to protect them */
131 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
132 struct ext4_ext_path *path)
136 /* path points to block */
137 err = ext4_journal_dirty_metadata(handle, path->p_bh);
139 /* path points to leaf/index in inode body */
140 err = ext4_mark_inode_dirty(handle, inode);
145 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
146 struct ext4_ext_path *path,
149 struct ext4_inode_info *ei = EXT4_I(inode);
150 ext4_fsblk_t bg_start;
151 ext4_grpblk_t colour;
155 struct ext4_extent *ex;
156 depth = path->p_depth;
158 /* try to predict block placement */
159 ex = path[depth].p_ext;
161 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
163 /* it looks like index is empty;
164 * try to find starting block from index itself */
165 if (path[depth].p_bh)
166 return path[depth].p_bh->b_blocknr;
169 /* OK. use inode's group */
170 bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
171 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
172 colour = (current->pid % 16) *
173 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
174 return bg_start + colour + block;
178 ext4_ext_new_block(handle_t *handle, struct inode *inode,
179 struct ext4_ext_path *path,
180 struct ext4_extent *ex, int *err)
182 ext4_fsblk_t goal, newblock;
184 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
185 newblock = ext4_new_block(handle, inode, goal, err);
189 static int ext4_ext_space_block(struct inode *inode)
193 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
194 / sizeof(struct ext4_extent);
195 #ifdef AGGRESSIVE_TEST
202 static int ext4_ext_space_block_idx(struct inode *inode)
206 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
207 / sizeof(struct ext4_extent_idx);
208 #ifdef AGGRESSIVE_TEST
215 static int ext4_ext_space_root(struct inode *inode)
219 size = sizeof(EXT4_I(inode)->i_data);
220 size -= sizeof(struct ext4_extent_header);
221 size /= sizeof(struct ext4_extent);
222 #ifdef AGGRESSIVE_TEST
229 static int ext4_ext_space_root_idx(struct inode *inode)
233 size = sizeof(EXT4_I(inode)->i_data);
234 size -= sizeof(struct ext4_extent_header);
235 size /= sizeof(struct ext4_extent_idx);
236 #ifdef AGGRESSIVE_TEST
244 ext4_ext_max_entries(struct inode *inode, int depth)
248 if (depth == ext_depth(inode)) {
250 max = ext4_ext_space_root(inode);
252 max = ext4_ext_space_root_idx(inode);
255 max = ext4_ext_space_block(inode);
257 max = ext4_ext_space_block_idx(inode);
263 static int __ext4_ext_check_header(const char *function, struct inode *inode,
264 struct ext4_extent_header *eh,
267 const char *error_msg;
270 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
271 error_msg = "invalid magic";
274 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
275 error_msg = "unexpected eh_depth";
278 if (unlikely(eh->eh_max == 0)) {
279 error_msg = "invalid eh_max";
282 max = ext4_ext_max_entries(inode, depth);
283 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
284 error_msg = "too large eh_max";
287 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
288 error_msg = "invalid eh_entries";
294 ext4_error(inode->i_sb, function,
295 "bad header in inode #%lu: %s - magic %x, "
296 "entries %u, max %u(%u), depth %u(%u)",
297 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
298 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
299 max, le16_to_cpu(eh->eh_depth), depth);
304 #define ext4_ext_check_header(inode, eh, depth) \
305 __ext4_ext_check_header(__FUNCTION__, inode, eh, depth)
308 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
310 int k, l = path->p_depth;
313 for (k = 0; k <= l; k++, path++) {
315 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
316 idx_pblock(path->p_idx));
317 } else if (path->p_ext) {
318 ext_debug(" %d:%d:%llu ",
319 le32_to_cpu(path->p_ext->ee_block),
320 ext4_ext_get_actual_len(path->p_ext),
321 ext_pblock(path->p_ext));
328 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
330 int depth = ext_depth(inode);
331 struct ext4_extent_header *eh;
332 struct ext4_extent *ex;
338 eh = path[depth].p_hdr;
339 ex = EXT_FIRST_EXTENT(eh);
341 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
342 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
343 ext4_ext_get_actual_len(ex), ext_pblock(ex));
348 #define ext4_ext_show_path(inode,path)
349 #define ext4_ext_show_leaf(inode,path)
352 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
354 int depth = path->p_depth;
357 for (i = 0; i <= depth; i++, path++)
365 * ext4_ext_binsearch_idx:
366 * binary search for the closest index of the given block
367 * the header must be checked before calling this
370 ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block)
372 struct ext4_extent_header *eh = path->p_hdr;
373 struct ext4_extent_idx *r, *l, *m;
376 ext_debug("binsearch for %d(idx): ", block);
378 l = EXT_FIRST_INDEX(eh) + 1;
379 r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
382 if (block < le32_to_cpu(m->ei_block))
386 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
387 m, le32_to_cpu(m->ei_block),
388 r, le32_to_cpu(r->ei_block));
392 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
393 idx_pblock(path->p_idx));
395 #ifdef CHECK_BINSEARCH
397 struct ext4_extent_idx *chix, *ix;
400 chix = ix = EXT_FIRST_INDEX(eh);
401 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
403 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
404 printk("k=%d, ix=0x%p, first=0x%p\n", k,
405 ix, EXT_FIRST_INDEX(eh));
407 le32_to_cpu(ix->ei_block),
408 le32_to_cpu(ix[-1].ei_block));
410 BUG_ON(k && le32_to_cpu(ix->ei_block)
411 <= le32_to_cpu(ix[-1].ei_block));
412 if (block < le32_to_cpu(ix->ei_block))
416 BUG_ON(chix != path->p_idx);
423 * ext4_ext_binsearch:
424 * binary search for closest extent of the given block
425 * the header must be checked before calling this
428 ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
430 struct ext4_extent_header *eh = path->p_hdr;
431 struct ext4_extent *r, *l, *m;
433 if (eh->eh_entries == 0) {
435 * this leaf is empty:
436 * we get such a leaf in split/add case
441 ext_debug("binsearch for %d: ", block);
443 l = EXT_FIRST_EXTENT(eh) + 1;
444 r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
448 if (block < le32_to_cpu(m->ee_block))
452 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
453 m, le32_to_cpu(m->ee_block),
454 r, le32_to_cpu(r->ee_block));
458 ext_debug(" -> %d:%llu:%d ",
459 le32_to_cpu(path->p_ext->ee_block),
460 ext_pblock(path->p_ext),
461 ext4_ext_get_actual_len(path->p_ext));
463 #ifdef CHECK_BINSEARCH
465 struct ext4_extent *chex, *ex;
468 chex = ex = EXT_FIRST_EXTENT(eh);
469 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
470 BUG_ON(k && le32_to_cpu(ex->ee_block)
471 <= le32_to_cpu(ex[-1].ee_block));
472 if (block < le32_to_cpu(ex->ee_block))
476 BUG_ON(chex != path->p_ext);
482 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
484 struct ext4_extent_header *eh;
486 eh = ext_inode_hdr(inode);
489 eh->eh_magic = EXT4_EXT_MAGIC;
490 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
491 ext4_mark_inode_dirty(handle, inode);
492 ext4_ext_invalidate_cache(inode);
496 struct ext4_ext_path *
497 ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
499 struct ext4_extent_header *eh;
500 struct buffer_head *bh;
501 short int depth, i, ppos = 0, alloc = 0;
503 eh = ext_inode_hdr(inode);
504 depth = ext_depth(inode);
505 if (ext4_ext_check_header(inode, eh, depth))
506 return ERR_PTR(-EIO);
509 /* account possible depth increase */
511 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
514 return ERR_PTR(-ENOMEM);
520 /* walk through the tree */
522 ext_debug("depth %d: num %d, max %d\n",
523 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
525 ext4_ext_binsearch_idx(inode, path + ppos, block);
526 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
527 path[ppos].p_depth = i;
528 path[ppos].p_ext = NULL;
530 bh = sb_bread(inode->i_sb, path[ppos].p_block);
534 eh = ext_block_hdr(bh);
536 BUG_ON(ppos > depth);
537 path[ppos].p_bh = bh;
538 path[ppos].p_hdr = eh;
541 if (ext4_ext_check_header(inode, eh, i))
545 path[ppos].p_depth = i;
546 path[ppos].p_hdr = eh;
547 path[ppos].p_ext = NULL;
548 path[ppos].p_idx = NULL;
551 ext4_ext_binsearch(inode, path + ppos, block);
553 ext4_ext_show_path(inode, path);
558 ext4_ext_drop_refs(path);
561 return ERR_PTR(-EIO);
565 * ext4_ext_insert_index:
566 * insert new index [@logical;@ptr] into the block at @curp;
567 * check where to insert: before @curp or after @curp
569 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
570 struct ext4_ext_path *curp,
571 int logical, ext4_fsblk_t ptr)
573 struct ext4_extent_idx *ix;
576 err = ext4_ext_get_access(handle, inode, curp);
580 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
581 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
582 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
584 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
585 len = (len - 1) * sizeof(struct ext4_extent_idx);
586 len = len < 0 ? 0 : len;
587 ext_debug("insert new index %d after: %llu. "
588 "move %d from 0x%p to 0x%p\n",
590 (curp->p_idx + 1), (curp->p_idx + 2));
591 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
593 ix = curp->p_idx + 1;
596 len = len * sizeof(struct ext4_extent_idx);
597 len = len < 0 ? 0 : len;
598 ext_debug("insert new index %d before: %llu. "
599 "move %d from 0x%p to 0x%p\n",
601 curp->p_idx, (curp->p_idx + 1));
602 memmove(curp->p_idx + 1, curp->p_idx, len);
606 ix->ei_block = cpu_to_le32(logical);
607 ext4_idx_store_pblock(ix, ptr);
608 curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
610 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
611 > le16_to_cpu(curp->p_hdr->eh_max));
612 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
614 err = ext4_ext_dirty(handle, inode, curp);
615 ext4_std_error(inode->i_sb, err);
622 * inserts new subtree into the path, using free index entry
624 * - allocates all needed blocks (new leaf and all intermediate index blocks)
625 * - makes decision where to split
626 * - moves remaining extents and index entries (right to the split point)
627 * into the newly allocated blocks
628 * - initializes subtree
630 static int ext4_ext_split(handle_t *handle, struct inode *inode,
631 struct ext4_ext_path *path,
632 struct ext4_extent *newext, int at)
634 struct buffer_head *bh = NULL;
635 int depth = ext_depth(inode);
636 struct ext4_extent_header *neh;
637 struct ext4_extent_idx *fidx;
638 struct ext4_extent *ex;
640 ext4_fsblk_t newblock, oldblock;
642 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
645 /* make decision: where to split? */
646 /* FIXME: now decision is simplest: at current extent */
648 /* if current leaf will be split, then we should use
649 * border from split point */
650 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
651 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
652 border = path[depth].p_ext[1].ee_block;
653 ext_debug("leaf will be split."
654 " next leaf starts at %d\n",
655 le32_to_cpu(border));
657 border = newext->ee_block;
658 ext_debug("leaf will be added."
659 " next leaf starts at %d\n",
660 le32_to_cpu(border));
664 * If error occurs, then we break processing
665 * and mark filesystem read-only. index won't
666 * be inserted and tree will be in consistent
667 * state. Next mount will repair buffers too.
671 * Get array to track all allocated blocks.
672 * We need this to handle errors and free blocks
675 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
679 /* allocate all needed blocks */
680 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
681 for (a = 0; a < depth - at; a++) {
682 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
685 ablocks[a] = newblock;
688 /* initialize new leaf */
689 newblock = ablocks[--a];
690 BUG_ON(newblock == 0);
691 bh = sb_getblk(inode->i_sb, newblock);
698 err = ext4_journal_get_create_access(handle, bh);
702 neh = ext_block_hdr(bh);
704 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
705 neh->eh_magic = EXT4_EXT_MAGIC;
707 ex = EXT_FIRST_EXTENT(neh);
709 /* move remainder of path[depth] to the new leaf */
710 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
711 /* start copy from next extent */
712 /* TODO: we could do it by single memmove */
715 while (path[depth].p_ext <=
716 EXT_MAX_EXTENT(path[depth].p_hdr)) {
717 ext_debug("move %d:%llu:%d in new leaf %llu\n",
718 le32_to_cpu(path[depth].p_ext->ee_block),
719 ext_pblock(path[depth].p_ext),
720 ext4_ext_get_actual_len(path[depth].p_ext),
722 /*memmove(ex++, path[depth].p_ext++,
723 sizeof(struct ext4_extent));
729 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
730 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
733 set_buffer_uptodate(bh);
736 err = ext4_journal_dirty_metadata(handle, bh);
742 /* correct old leaf */
744 err = ext4_ext_get_access(handle, inode, path + depth);
747 path[depth].p_hdr->eh_entries =
748 cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
749 err = ext4_ext_dirty(handle, inode, path + depth);
755 /* create intermediate indexes */
759 ext_debug("create %d intermediate indices\n", k);
760 /* insert new index into current index block */
761 /* current depth stored in i var */
765 newblock = ablocks[--a];
766 bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock);
773 err = ext4_journal_get_create_access(handle, bh);
777 neh = ext_block_hdr(bh);
778 neh->eh_entries = cpu_to_le16(1);
779 neh->eh_magic = EXT4_EXT_MAGIC;
780 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
781 neh->eh_depth = cpu_to_le16(depth - i);
782 fidx = EXT_FIRST_INDEX(neh);
783 fidx->ei_block = border;
784 ext4_idx_store_pblock(fidx, oldblock);
786 ext_debug("int.index at %d (block %llu): %lu -> %llu\n", i,
787 newblock, (unsigned long) le32_to_cpu(border),
793 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
794 EXT_MAX_INDEX(path[i].p_hdr));
795 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
796 EXT_LAST_INDEX(path[i].p_hdr));
797 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
798 ext_debug("%d: move %d:%llu in new index %llu\n", i,
799 le32_to_cpu(path[i].p_idx->ei_block),
800 idx_pblock(path[i].p_idx),
802 /*memmove(++fidx, path[i].p_idx++,
803 sizeof(struct ext4_extent_idx));
805 BUG_ON(neh->eh_entries > neh->eh_max);*/
810 memmove(++fidx, path[i].p_idx - m,
811 sizeof(struct ext4_extent_idx) * m);
813 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
815 set_buffer_uptodate(bh);
818 err = ext4_journal_dirty_metadata(handle, bh);
824 /* correct old index */
826 err = ext4_ext_get_access(handle, inode, path + i);
829 path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
830 err = ext4_ext_dirty(handle, inode, path + i);
838 /* insert new index */
839 err = ext4_ext_insert_index(handle, inode, path + at,
840 le32_to_cpu(border), newblock);
844 if (buffer_locked(bh))
850 /* free all allocated blocks in error case */
851 for (i = 0; i < depth; i++) {
854 ext4_free_blocks(handle, inode, ablocks[i], 1);
863 * ext4_ext_grow_indepth:
864 * implements tree growing procedure:
865 * - allocates new block
866 * - moves top-level data (index block or leaf) into the new block
867 * - initializes new top-level, creating index that points to the
870 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
871 struct ext4_ext_path *path,
872 struct ext4_extent *newext)
874 struct ext4_ext_path *curp = path;
875 struct ext4_extent_header *neh;
876 struct ext4_extent_idx *fidx;
877 struct buffer_head *bh;
878 ext4_fsblk_t newblock;
881 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
885 bh = sb_getblk(inode->i_sb, newblock);
888 ext4_std_error(inode->i_sb, err);
893 err = ext4_journal_get_create_access(handle, bh);
899 /* move top-level index/leaf into new block */
900 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
902 /* set size of new block */
903 neh = ext_block_hdr(bh);
904 /* old root could have indexes or leaves
905 * so calculate e_max right way */
906 if (ext_depth(inode))
907 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
909 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
910 neh->eh_magic = EXT4_EXT_MAGIC;
911 set_buffer_uptodate(bh);
914 err = ext4_journal_dirty_metadata(handle, bh);
918 /* create index in new top-level index: num,max,pointer */
919 err = ext4_ext_get_access(handle, inode, curp);
923 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
924 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
925 curp->p_hdr->eh_entries = cpu_to_le16(1);
926 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
927 /* FIXME: it works, but actually path[0] can be index */
928 curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
929 ext4_idx_store_pblock(curp->p_idx, newblock);
931 neh = ext_inode_hdr(inode);
932 fidx = EXT_FIRST_INDEX(neh);
933 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
934 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
935 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
937 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
938 err = ext4_ext_dirty(handle, inode, curp);
946 * ext4_ext_create_new_leaf:
947 * finds empty index and adds new leaf.
948 * if no free index is found, then it requests in-depth growing.
950 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
951 struct ext4_ext_path *path,
952 struct ext4_extent *newext)
954 struct ext4_ext_path *curp;
955 int depth, i, err = 0;
958 i = depth = ext_depth(inode);
960 /* walk up to the tree and look for free index entry */
962 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
967 /* we use already allocated block for index block,
968 * so subsequent data blocks should be contiguous */
969 if (EXT_HAS_FREE_INDEX(curp)) {
970 /* if we found index with free entry, then use that
971 * entry: create all needed subtree and add new leaf */
972 err = ext4_ext_split(handle, inode, path, newext, i);
975 ext4_ext_drop_refs(path);
976 path = ext4_ext_find_extent(inode,
977 le32_to_cpu(newext->ee_block),
982 /* tree is full, time to grow in depth */
983 err = ext4_ext_grow_indepth(handle, inode, path, newext);
988 ext4_ext_drop_refs(path);
989 path = ext4_ext_find_extent(inode,
990 le32_to_cpu(newext->ee_block),
998 * only first (depth 0 -> 1) produces free space;
999 * in all other cases we have to split the grown tree
1001 depth = ext_depth(inode);
1002 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1003 /* now we need to split */
1013 * ext4_ext_next_allocated_block:
1014 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1015 * NOTE: it considers block number from index entry as
1016 * allocated block. Thus, index entries have to be consistent
1019 static unsigned long
1020 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1024 BUG_ON(path == NULL);
1025 depth = path->p_depth;
1027 if (depth == 0 && path->p_ext == NULL)
1028 return EXT_MAX_BLOCK;
1030 while (depth >= 0) {
1031 if (depth == path->p_depth) {
1033 if (path[depth].p_ext !=
1034 EXT_LAST_EXTENT(path[depth].p_hdr))
1035 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1038 if (path[depth].p_idx !=
1039 EXT_LAST_INDEX(path[depth].p_hdr))
1040 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1045 return EXT_MAX_BLOCK;
1049 * ext4_ext_next_leaf_block:
1050 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1052 static unsigned ext4_ext_next_leaf_block(struct inode *inode,
1053 struct ext4_ext_path *path)
1057 BUG_ON(path == NULL);
1058 depth = path->p_depth;
1060 /* zero-tree has no leaf blocks at all */
1062 return EXT_MAX_BLOCK;
1064 /* go to index block */
1067 while (depth >= 0) {
1068 if (path[depth].p_idx !=
1069 EXT_LAST_INDEX(path[depth].p_hdr))
1070 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1074 return EXT_MAX_BLOCK;
1078 * ext4_ext_correct_indexes:
1079 * if leaf gets modified and modified extent is first in the leaf,
1080 * then we have to correct all indexes above.
1081 * TODO: do we need to correct tree in all cases?
1083 int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1084 struct ext4_ext_path *path)
1086 struct ext4_extent_header *eh;
1087 int depth = ext_depth(inode);
1088 struct ext4_extent *ex;
1092 eh = path[depth].p_hdr;
1093 ex = path[depth].p_ext;
1098 /* there is no tree at all */
1102 if (ex != EXT_FIRST_EXTENT(eh)) {
1103 /* we correct tree if first leaf got modified only */
1108 * TODO: we need correction if border is smaller than current one
1111 border = path[depth].p_ext->ee_block;
1112 err = ext4_ext_get_access(handle, inode, path + k);
1115 path[k].p_idx->ei_block = border;
1116 err = ext4_ext_dirty(handle, inode, path + k);
1121 /* change all left-side indexes */
1122 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1124 err = ext4_ext_get_access(handle, inode, path + k);
1127 path[k].p_idx->ei_block = border;
1128 err = ext4_ext_dirty(handle, inode, path + k);
1137 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1138 struct ext4_extent *ex2)
1140 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1143 * Make sure that either both extents are uninitialized, or
1146 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1149 if (ext4_ext_is_uninitialized(ex1))
1150 max_len = EXT_UNINIT_MAX_LEN;
1152 max_len = EXT_INIT_MAX_LEN;
1154 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1155 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1157 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1158 le32_to_cpu(ex2->ee_block))
1162 * To allow future support for preallocated extents to be added
1163 * as an RO_COMPAT feature, refuse to merge to extents if
1164 * this can result in the top bit of ee_len being set.
1166 if (ext1_ee_len + ext2_ee_len > max_len)
1168 #ifdef AGGRESSIVE_TEST
1169 if (le16_to_cpu(ex1->ee_len) >= 4)
1173 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1179 * This function tries to merge the "ex" extent to the next extent in the tree.
1180 * It always tries to merge towards right. If you want to merge towards
1181 * left, pass "ex - 1" as argument instead of "ex".
1182 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1183 * 1 if they got merged.
1185 int ext4_ext_try_to_merge(struct inode *inode,
1186 struct ext4_ext_path *path,
1187 struct ext4_extent *ex)
1189 struct ext4_extent_header *eh;
1190 unsigned int depth, len;
1192 int uninitialized = 0;
1194 depth = ext_depth(inode);
1195 BUG_ON(path[depth].p_hdr == NULL);
1196 eh = path[depth].p_hdr;
1198 while (ex < EXT_LAST_EXTENT(eh)) {
1199 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1201 /* merge with next extent! */
1202 if (ext4_ext_is_uninitialized(ex))
1204 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1205 + ext4_ext_get_actual_len(ex + 1));
1207 ext4_ext_mark_uninitialized(ex);
1209 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1210 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1211 * sizeof(struct ext4_extent);
1212 memmove(ex + 1, ex + 2, len);
1214 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
1216 WARN_ON(eh->eh_entries == 0);
1217 if (!eh->eh_entries)
1218 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1219 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1226 * check if a portion of the "newext" extent overlaps with an
1229 * If there is an overlap discovered, it updates the length of the newext
1230 * such that there will be no overlap, and then returns 1.
1231 * If there is no overlap found, it returns 0.
1233 unsigned int ext4_ext_check_overlap(struct inode *inode,
1234 struct ext4_extent *newext,
1235 struct ext4_ext_path *path)
1237 unsigned long b1, b2;
1238 unsigned int depth, len1;
1239 unsigned int ret = 0;
1241 b1 = le32_to_cpu(newext->ee_block);
1242 len1 = ext4_ext_get_actual_len(newext);
1243 depth = ext_depth(inode);
1244 if (!path[depth].p_ext)
1246 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1249 * get the next allocated block if the extent in the path
1250 * is before the requested block(s)
1253 b2 = ext4_ext_next_allocated_block(path);
1254 if (b2 == EXT_MAX_BLOCK)
1258 /* check for wrap through zero */
1259 if (b1 + len1 < b1) {
1260 len1 = EXT_MAX_BLOCK - b1;
1261 newext->ee_len = cpu_to_le16(len1);
1265 /* check for overlap */
1266 if (b1 + len1 > b2) {
1267 newext->ee_len = cpu_to_le16(b2 - b1);
1275 * ext4_ext_insert_extent:
1276 * tries to merge requsted extent into the existing extent or
1277 * inserts requested extent as new one into the tree,
1278 * creating new leaf in the no-space case.
1280 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1281 struct ext4_ext_path *path,
1282 struct ext4_extent *newext)
1284 struct ext4_extent_header * eh;
1285 struct ext4_extent *ex, *fex;
1286 struct ext4_extent *nearex; /* nearest extent */
1287 struct ext4_ext_path *npath = NULL;
1288 int depth, len, err, next;
1289 unsigned uninitialized = 0;
1291 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
1292 depth = ext_depth(inode);
1293 ex = path[depth].p_ext;
1294 BUG_ON(path[depth].p_hdr == NULL);
1296 /* try to insert block into found extent and return */
1297 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1298 ext_debug("append %d block to %d:%d (from %llu)\n",
1299 ext4_ext_get_actual_len(newext),
1300 le32_to_cpu(ex->ee_block),
1301 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1302 err = ext4_ext_get_access(handle, inode, path + depth);
1307 * ext4_can_extents_be_merged should have checked that either
1308 * both extents are uninitialized, or both aren't. Thus we
1309 * need to check only one of them here.
1311 if (ext4_ext_is_uninitialized(ex))
1313 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1314 + ext4_ext_get_actual_len(newext));
1316 ext4_ext_mark_uninitialized(ex);
1317 eh = path[depth].p_hdr;
1323 depth = ext_depth(inode);
1324 eh = path[depth].p_hdr;
1325 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1328 /* probably next leaf has space for us? */
1329 fex = EXT_LAST_EXTENT(eh);
1330 next = ext4_ext_next_leaf_block(inode, path);
1331 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1332 && next != EXT_MAX_BLOCK) {
1333 ext_debug("next leaf block - %d\n", next);
1334 BUG_ON(npath != NULL);
1335 npath = ext4_ext_find_extent(inode, next, NULL);
1337 return PTR_ERR(npath);
1338 BUG_ON(npath->p_depth != path->p_depth);
1339 eh = npath[depth].p_hdr;
1340 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1341 ext_debug("next leaf isnt full(%d)\n",
1342 le16_to_cpu(eh->eh_entries));
1346 ext_debug("next leaf has no free space(%d,%d)\n",
1347 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1351 * There is no free space in the found leaf.
1352 * We're gonna add a new leaf in the tree.
1354 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1357 depth = ext_depth(inode);
1358 eh = path[depth].p_hdr;
1361 nearex = path[depth].p_ext;
1363 err = ext4_ext_get_access(handle, inode, path + depth);
1368 /* there is no extent in this leaf, create first one */
1369 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1370 le32_to_cpu(newext->ee_block),
1372 ext4_ext_get_actual_len(newext));
1373 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1374 } else if (le32_to_cpu(newext->ee_block)
1375 > le32_to_cpu(nearex->ee_block)) {
1376 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1377 if (nearex != EXT_LAST_EXTENT(eh)) {
1378 len = EXT_MAX_EXTENT(eh) - nearex;
1379 len = (len - 1) * sizeof(struct ext4_extent);
1380 len = len < 0 ? 0 : len;
1381 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1382 "move %d from 0x%p to 0x%p\n",
1383 le32_to_cpu(newext->ee_block),
1385 ext4_ext_get_actual_len(newext),
1386 nearex, len, nearex + 1, nearex + 2);
1387 memmove(nearex + 2, nearex + 1, len);
1389 path[depth].p_ext = nearex + 1;
1391 BUG_ON(newext->ee_block == nearex->ee_block);
1392 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1393 len = len < 0 ? 0 : len;
1394 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1395 "move %d from 0x%p to 0x%p\n",
1396 le32_to_cpu(newext->ee_block),
1398 ext4_ext_get_actual_len(newext),
1399 nearex, len, nearex + 1, nearex + 2);
1400 memmove(nearex + 1, nearex, len);
1401 path[depth].p_ext = nearex;
1404 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1405 nearex = path[depth].p_ext;
1406 nearex->ee_block = newext->ee_block;
1407 nearex->ee_start = newext->ee_start;
1408 nearex->ee_start_hi = newext->ee_start_hi;
1409 nearex->ee_len = newext->ee_len;
1412 /* try to merge extents to the right */
1413 ext4_ext_try_to_merge(inode, path, nearex);
1415 /* try to merge extents to the left */
1417 /* time to correct all indexes above */
1418 err = ext4_ext_correct_indexes(handle, inode, path);
1422 err = ext4_ext_dirty(handle, inode, path + depth);
1426 ext4_ext_drop_refs(npath);
1429 ext4_ext_tree_changed(inode);
1430 ext4_ext_invalidate_cache(inode);
1434 int ext4_ext_walk_space(struct inode *inode, unsigned long block,
1435 unsigned long num, ext_prepare_callback func,
1438 struct ext4_ext_path *path = NULL;
1439 struct ext4_ext_cache cbex;
1440 struct ext4_extent *ex;
1441 unsigned long next, start = 0, end = 0;
1442 unsigned long last = block + num;
1443 int depth, exists, err = 0;
1445 BUG_ON(func == NULL);
1446 BUG_ON(inode == NULL);
1448 while (block < last && block != EXT_MAX_BLOCK) {
1450 /* find extent for this block */
1451 path = ext4_ext_find_extent(inode, block, path);
1453 err = PTR_ERR(path);
1458 depth = ext_depth(inode);
1459 BUG_ON(path[depth].p_hdr == NULL);
1460 ex = path[depth].p_ext;
1461 next = ext4_ext_next_allocated_block(path);
1465 /* there is no extent yet, so try to allocate
1466 * all requested space */
1469 } else if (le32_to_cpu(ex->ee_block) > block) {
1470 /* need to allocate space before found extent */
1472 end = le32_to_cpu(ex->ee_block);
1473 if (block + num < end)
1475 } else if (block >= le32_to_cpu(ex->ee_block)
1476 + ext4_ext_get_actual_len(ex)) {
1477 /* need to allocate space after found extent */
1482 } else if (block >= le32_to_cpu(ex->ee_block)) {
1484 * some part of requested space is covered
1488 end = le32_to_cpu(ex->ee_block)
1489 + ext4_ext_get_actual_len(ex);
1490 if (block + num < end)
1496 BUG_ON(end <= start);
1499 cbex.ec_block = start;
1500 cbex.ec_len = end - start;
1502 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1504 cbex.ec_block = le32_to_cpu(ex->ee_block);
1505 cbex.ec_len = ext4_ext_get_actual_len(ex);
1506 cbex.ec_start = ext_pblock(ex);
1507 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1510 BUG_ON(cbex.ec_len == 0);
1511 err = func(inode, path, &cbex, cbdata);
1512 ext4_ext_drop_refs(path);
1516 if (err == EXT_REPEAT)
1518 else if (err == EXT_BREAK) {
1523 if (ext_depth(inode) != depth) {
1524 /* depth was changed. we have to realloc path */
1529 block = cbex.ec_block + cbex.ec_len;
1533 ext4_ext_drop_refs(path);
1541 ext4_ext_put_in_cache(struct inode *inode, __u32 block,
1542 __u32 len, __u32 start, int type)
1544 struct ext4_ext_cache *cex;
1546 cex = &EXT4_I(inode)->i_cached_extent;
1547 cex->ec_type = type;
1548 cex->ec_block = block;
1550 cex->ec_start = start;
1554 * ext4_ext_put_gap_in_cache:
1555 * calculate boundaries of the gap that the requested block fits into
1556 * and cache this gap
1559 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1560 unsigned long block)
1562 int depth = ext_depth(inode);
1563 unsigned long lblock, len;
1564 struct ext4_extent *ex;
1566 ex = path[depth].p_ext;
1568 /* there is no extent yet, so gap is [0;-] */
1570 len = EXT_MAX_BLOCK;
1571 ext_debug("cache gap(whole file):");
1572 } else if (block < le32_to_cpu(ex->ee_block)) {
1574 len = le32_to_cpu(ex->ee_block) - block;
1575 ext_debug("cache gap(before): %lu [%lu:%lu]",
1576 (unsigned long) block,
1577 (unsigned long) le32_to_cpu(ex->ee_block),
1578 (unsigned long) ext4_ext_get_actual_len(ex));
1579 } else if (block >= le32_to_cpu(ex->ee_block)
1580 + ext4_ext_get_actual_len(ex)) {
1581 lblock = le32_to_cpu(ex->ee_block)
1582 + ext4_ext_get_actual_len(ex);
1583 len = ext4_ext_next_allocated_block(path);
1584 ext_debug("cache gap(after): [%lu:%lu] %lu",
1585 (unsigned long) le32_to_cpu(ex->ee_block),
1586 (unsigned long) ext4_ext_get_actual_len(ex),
1587 (unsigned long) block);
1588 BUG_ON(len == lblock);
1595 ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len);
1596 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1600 ext4_ext_in_cache(struct inode *inode, unsigned long block,
1601 struct ext4_extent *ex)
1603 struct ext4_ext_cache *cex;
1605 cex = &EXT4_I(inode)->i_cached_extent;
1607 /* has cache valid data? */
1608 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1609 return EXT4_EXT_CACHE_NO;
1611 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1612 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1613 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1614 ex->ee_block = cpu_to_le32(cex->ec_block);
1615 ext4_ext_store_pblock(ex, cex->ec_start);
1616 ex->ee_len = cpu_to_le16(cex->ec_len);
1617 ext_debug("%lu cached by %lu:%lu:%llu\n",
1618 (unsigned long) block,
1619 (unsigned long) cex->ec_block,
1620 (unsigned long) cex->ec_len,
1622 return cex->ec_type;
1626 return EXT4_EXT_CACHE_NO;
1631 * removes index from the index block.
1632 * It's used in truncate case only, thus all requests are for
1633 * last index in the block only.
1635 int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1636 struct ext4_ext_path *path)
1638 struct buffer_head *bh;
1642 /* free index block */
1644 leaf = idx_pblock(path->p_idx);
1645 BUG_ON(path->p_hdr->eh_entries == 0);
1646 err = ext4_ext_get_access(handle, inode, path);
1649 path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1650 err = ext4_ext_dirty(handle, inode, path);
1653 ext_debug("index is empty, remove it, free block %llu\n", leaf);
1654 bh = sb_find_get_block(inode->i_sb, leaf);
1655 ext4_forget(handle, 1, inode, bh, leaf);
1656 ext4_free_blocks(handle, inode, leaf, 1);
1661 * ext4_ext_calc_credits_for_insert:
1662 * This routine returns max. credits that the extent tree can consume.
1663 * It should be OK for low-performance paths like ->writepage()
1664 * To allow many writing processes to fit into a single transaction,
1665 * the caller should calculate credits under truncate_mutex and
1666 * pass the actual path.
1668 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1669 struct ext4_ext_path *path)
1674 /* probably there is space in leaf? */
1675 depth = ext_depth(inode);
1676 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1677 < le16_to_cpu(path[depth].p_hdr->eh_max))
1682 * given 32-bit logical block (4294967296 blocks), max. tree
1683 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1684 * Let's also add one more level for imbalance.
1688 /* allocation of new data block(s) */
1692 * tree can be full, so it would need to grow in depth:
1693 * we need one credit to modify old root, credits for
1694 * new root will be added in split accounting
1699 * Index split can happen, we would need:
1700 * allocate intermediate indexes (bitmap + group)
1701 * + change two blocks at each level, but root (already included)
1703 needed += (depth * 2) + (depth * 2);
1705 /* any allocation modifies superblock */
1711 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1712 struct ext4_extent *ex,
1713 unsigned long from, unsigned long to)
1715 struct buffer_head *bh;
1716 unsigned short ee_len = ext4_ext_get_actual_len(ex);
1719 #ifdef EXTENTS_STATS
1721 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1722 spin_lock(&sbi->s_ext_stats_lock);
1723 sbi->s_ext_blocks += ee_len;
1724 sbi->s_ext_extents++;
1725 if (ee_len < sbi->s_ext_min)
1726 sbi->s_ext_min = ee_len;
1727 if (ee_len > sbi->s_ext_max)
1728 sbi->s_ext_max = ee_len;
1729 if (ext_depth(inode) > sbi->s_depth_max)
1730 sbi->s_depth_max = ext_depth(inode);
1731 spin_unlock(&sbi->s_ext_stats_lock);
1734 if (from >= le32_to_cpu(ex->ee_block)
1735 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
1739 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1740 start = ext_pblock(ex) + ee_len - num;
1741 ext_debug("free last %lu blocks starting %llu\n", num, start);
1742 for (i = 0; i < num; i++) {
1743 bh = sb_find_get_block(inode->i_sb, start + i);
1744 ext4_forget(handle, 0, inode, bh, start + i);
1746 ext4_free_blocks(handle, inode, start, num);
1747 } else if (from == le32_to_cpu(ex->ee_block)
1748 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
1749 printk("strange request: removal %lu-%lu from %u:%u\n",
1750 from, to, le32_to_cpu(ex->ee_block), ee_len);
1752 printk("strange request: removal(2) %lu-%lu from %u:%u\n",
1753 from, to, le32_to_cpu(ex->ee_block), ee_len);
1759 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1760 struct ext4_ext_path *path, unsigned long start)
1762 int err = 0, correct_index = 0;
1763 int depth = ext_depth(inode), credits;
1764 struct ext4_extent_header *eh;
1765 unsigned a, b, block, num;
1766 unsigned long ex_ee_block;
1767 unsigned short ex_ee_len;
1768 unsigned uninitialized = 0;
1769 struct ext4_extent *ex;
1771 /* the header must be checked already in ext4_ext_remove_space() */
1772 ext_debug("truncate since %lu in leaf\n", start);
1773 if (!path[depth].p_hdr)
1774 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1775 eh = path[depth].p_hdr;
1778 /* find where to start removing */
1779 ex = EXT_LAST_EXTENT(eh);
1781 ex_ee_block = le32_to_cpu(ex->ee_block);
1782 if (ext4_ext_is_uninitialized(ex))
1784 ex_ee_len = ext4_ext_get_actual_len(ex);
1786 while (ex >= EXT_FIRST_EXTENT(eh) &&
1787 ex_ee_block + ex_ee_len > start) {
1788 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1789 path[depth].p_ext = ex;
1791 a = ex_ee_block > start ? ex_ee_block : start;
1792 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1793 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1795 ext_debug(" border %u:%u\n", a, b);
1797 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1801 } else if (a != ex_ee_block) {
1802 /* remove tail of the extent */
1803 block = ex_ee_block;
1805 } else if (b != ex_ee_block + ex_ee_len - 1) {
1806 /* remove head of the extent */
1809 /* there is no "make a hole" API yet */
1812 /* remove whole extent: excellent! */
1813 block = ex_ee_block;
1815 BUG_ON(a != ex_ee_block);
1816 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1819 /* at present, extent can't cross block group: */
1820 /* leaf + bitmap + group desc + sb + inode */
1822 if (ex == EXT_FIRST_EXTENT(eh)) {
1824 credits += (ext_depth(inode)) + 1;
1827 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1830 handle = ext4_ext_journal_restart(handle, credits);
1831 if (IS_ERR(handle)) {
1832 err = PTR_ERR(handle);
1836 err = ext4_ext_get_access(handle, inode, path + depth);
1840 err = ext4_remove_blocks(handle, inode, ex, a, b);
1845 /* this extent is removed; mark slot entirely unused */
1846 ext4_ext_store_pblock(ex, 0);
1847 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1850 ex->ee_block = cpu_to_le32(block);
1851 ex->ee_len = cpu_to_le16(num);
1853 * Do not mark uninitialized if all the blocks in the
1854 * extent have been removed.
1856 if (uninitialized && num)
1857 ext4_ext_mark_uninitialized(ex);
1859 err = ext4_ext_dirty(handle, inode, path + depth);
1863 ext_debug("new extent: %u:%u:%llu\n", block, num,
1866 ex_ee_block = le32_to_cpu(ex->ee_block);
1867 ex_ee_len = ext4_ext_get_actual_len(ex);
1870 if (correct_index && eh->eh_entries)
1871 err = ext4_ext_correct_indexes(handle, inode, path);
1873 /* if this leaf is free, then we should
1874 * remove it from index block above */
1875 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1876 err = ext4_ext_rm_idx(handle, inode, path + depth);
1883 * ext4_ext_more_to_rm:
1884 * returns 1 if current index has to be freed (even partial)
1887 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1889 BUG_ON(path->p_idx == NULL);
1891 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1895 * if truncate on deeper level happened, it wasn't partial,
1896 * so we have to consider current index for truncation
1898 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1903 int ext4_ext_remove_space(struct inode *inode, unsigned long start)
1905 struct super_block *sb = inode->i_sb;
1906 int depth = ext_depth(inode);
1907 struct ext4_ext_path *path;
1911 ext_debug("truncate since %lu\n", start);
1913 /* probably first extent we're gonna free will be last in block */
1914 handle = ext4_journal_start(inode, depth + 1);
1916 return PTR_ERR(handle);
1918 ext4_ext_invalidate_cache(inode);
1921 * We start scanning from right side, freeing all the blocks
1922 * after i_size and walking into the tree depth-wise.
1924 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1926 ext4_journal_stop(handle);
1929 path[0].p_hdr = ext_inode_hdr(inode);
1930 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
1934 path[0].p_depth = depth;
1936 while (i >= 0 && err == 0) {
1938 /* this is leaf block */
1939 err = ext4_ext_rm_leaf(handle, inode, path, start);
1940 /* root level has p_bh == NULL, brelse() eats this */
1941 brelse(path[i].p_bh);
1942 path[i].p_bh = NULL;
1947 /* this is index block */
1948 if (!path[i].p_hdr) {
1949 ext_debug("initialize header\n");
1950 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1953 if (!path[i].p_idx) {
1954 /* this level hasn't been touched yet */
1955 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1956 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1957 ext_debug("init index ptr: hdr 0x%p, num %d\n",
1959 le16_to_cpu(path[i].p_hdr->eh_entries));
1961 /* we were already here, see at next index */
1965 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
1966 i, EXT_FIRST_INDEX(path[i].p_hdr),
1968 if (ext4_ext_more_to_rm(path + i)) {
1969 struct buffer_head *bh;
1970 /* go to the next level */
1971 ext_debug("move to level %d (block %llu)\n",
1972 i + 1, idx_pblock(path[i].p_idx));
1973 memset(path + i + 1, 0, sizeof(*path));
1974 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
1976 /* should we reset i_size? */
1980 if (WARN_ON(i + 1 > depth)) {
1984 if (ext4_ext_check_header(inode, ext_block_hdr(bh),
1989 path[i + 1].p_bh = bh;
1991 /* save actual number of indexes since this
1992 * number is changed at the next iteration */
1993 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1996 /* we finished processing this index, go up */
1997 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
1998 /* index is empty, remove it;
1999 * handle must be already prepared by the
2000 * truncatei_leaf() */
2001 err = ext4_ext_rm_idx(handle, inode, path + i);
2003 /* root level has p_bh == NULL, brelse() eats this */
2004 brelse(path[i].p_bh);
2005 path[i].p_bh = NULL;
2007 ext_debug("return to level %d\n", i);
2011 /* TODO: flexible tree reduction should be here */
2012 if (path->p_hdr->eh_entries == 0) {
2014 * truncate to zero freed all the tree,
2015 * so we need to correct eh_depth
2017 err = ext4_ext_get_access(handle, inode, path);
2019 ext_inode_hdr(inode)->eh_depth = 0;
2020 ext_inode_hdr(inode)->eh_max =
2021 cpu_to_le16(ext4_ext_space_root(inode));
2022 err = ext4_ext_dirty(handle, inode, path);
2026 ext4_ext_tree_changed(inode);
2027 ext4_ext_drop_refs(path);
2029 ext4_journal_stop(handle);
2035 * called at mount time
2037 void ext4_ext_init(struct super_block *sb)
2040 * possible initialization would be here
2043 if (test_opt(sb, EXTENTS)) {
2044 printk("EXT4-fs: file extents enabled");
2045 #ifdef AGGRESSIVE_TEST
2046 printk(", aggressive tests");
2048 #ifdef CHECK_BINSEARCH
2049 printk(", check binsearch");
2051 #ifdef EXTENTS_STATS
2055 #ifdef EXTENTS_STATS
2056 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2057 EXT4_SB(sb)->s_ext_min = 1 << 30;
2058 EXT4_SB(sb)->s_ext_max = 0;
2064 * called at umount time
2066 void ext4_ext_release(struct super_block *sb)
2068 if (!test_opt(sb, EXTENTS))
2071 #ifdef EXTENTS_STATS
2072 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2073 struct ext4_sb_info *sbi = EXT4_SB(sb);
2074 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2075 sbi->s_ext_blocks, sbi->s_ext_extents,
2076 sbi->s_ext_blocks / sbi->s_ext_extents);
2077 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2078 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2084 * This function is called by ext4_ext_get_blocks() if someone tries to write
2085 * to an uninitialized extent. It may result in splitting the uninitialized
2086 * extent into multiple extents (upto three - one initialized and two
2088 * There are three possibilities:
2089 * a> There is no split required: Entire extent should be initialized
2090 * b> Splits in two extents: Write is happening at either end of the extent
2091 * c> Splits in three extents: Somone is writing in middle of the extent
2093 int ext4_ext_convert_to_initialized(handle_t *handle, struct inode *inode,
2094 struct ext4_ext_path *path,
2095 ext4_fsblk_t iblock,
2096 unsigned long max_blocks)
2098 struct ext4_extent *ex, newex;
2099 struct ext4_extent *ex1 = NULL;
2100 struct ext4_extent *ex2 = NULL;
2101 struct ext4_extent *ex3 = NULL;
2102 struct ext4_extent_header *eh;
2103 unsigned int allocated, ee_block, ee_len, depth;
2104 ext4_fsblk_t newblock;
2108 depth = ext_depth(inode);
2109 eh = path[depth].p_hdr;
2110 ex = path[depth].p_ext;
2111 ee_block = le32_to_cpu(ex->ee_block);
2112 ee_len = ext4_ext_get_actual_len(ex);
2113 allocated = ee_len - (iblock - ee_block);
2114 newblock = iblock - ee_block + ext_pblock(ex);
2117 /* ex1: ee_block to iblock - 1 : uninitialized */
2118 if (iblock > ee_block) {
2120 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2121 ext4_ext_mark_uninitialized(ex1);
2125 * for sanity, update the length of the ex2 extent before
2126 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2127 * overlap of blocks.
2129 if (!ex1 && allocated > max_blocks)
2130 ex2->ee_len = cpu_to_le16(max_blocks);
2131 /* ex3: to ee_block + ee_len : uninitialised */
2132 if (allocated > max_blocks) {
2133 unsigned int newdepth;
2135 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2136 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2137 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2138 ext4_ext_mark_uninitialized(ex3);
2139 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2143 * The depth, and hence eh & ex might change
2144 * as part of the insert above.
2146 newdepth = ext_depth(inode);
2147 if (newdepth != depth) {
2149 path = ext4_ext_find_extent(inode, iblock, NULL);
2151 err = PTR_ERR(path);
2155 eh = path[depth].p_hdr;
2156 ex = path[depth].p_ext;
2160 allocated = max_blocks;
2163 * If there was a change of depth as part of the
2164 * insertion of ex3 above, we need to update the length
2165 * of the ex1 extent again here
2167 if (ex1 && ex1 != ex) {
2169 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2170 ext4_ext_mark_uninitialized(ex1);
2173 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2174 ex2->ee_block = cpu_to_le32(iblock);
2175 ex2->ee_start = cpu_to_le32(newblock);
2176 ext4_ext_store_pblock(ex2, newblock);
2177 ex2->ee_len = cpu_to_le16(allocated);
2180 err = ext4_ext_get_access(handle, inode, path + depth);
2184 * New (initialized) extent starts from the first block
2185 * in the current extent. i.e., ex2 == ex
2186 * We have to see if it can be merged with the extent
2189 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2191 * To merge left, pass "ex2 - 1" to try_to_merge(),
2192 * since it merges towards right _only_.
2194 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2196 err = ext4_ext_correct_indexes(handle, inode, path);
2199 depth = ext_depth(inode);
2204 * Try to Merge towards right. This might be required
2205 * only when the whole extent is being written to.
2206 * i.e. ex2 == ex and ex3 == NULL.
2209 ret = ext4_ext_try_to_merge(inode, path, ex2);
2211 err = ext4_ext_correct_indexes(handle, inode, path);
2216 /* Mark modified extent as dirty */
2217 err = ext4_ext_dirty(handle, inode, path + depth);
2220 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2222 return err ? err : allocated;
2225 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2226 ext4_fsblk_t iblock,
2227 unsigned long max_blocks, struct buffer_head *bh_result,
2228 int create, int extend_disksize)
2230 struct ext4_ext_path *path = NULL;
2231 struct ext4_extent_header *eh;
2232 struct ext4_extent newex, *ex;
2233 ext4_fsblk_t goal, newblock;
2234 int err = 0, depth, ret;
2235 unsigned long allocated = 0;
2237 __clear_bit(BH_New, &bh_result->b_state);
2238 ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock,
2239 max_blocks, (unsigned) inode->i_ino);
2240 mutex_lock(&EXT4_I(inode)->truncate_mutex);
2242 /* check in cache */
2243 goal = ext4_ext_in_cache(inode, iblock, &newex);
2245 if (goal == EXT4_EXT_CACHE_GAP) {
2248 * block isn't allocated yet and
2249 * user doesn't want to allocate it
2253 /* we should allocate requested block */
2254 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2255 /* block is already allocated */
2257 - le32_to_cpu(newex.ee_block)
2258 + ext_pblock(&newex);
2259 /* number of remaining blocks in the extent */
2260 allocated = le16_to_cpu(newex.ee_len) -
2261 (iblock - le32_to_cpu(newex.ee_block));
2268 /* find extent for this block */
2269 path = ext4_ext_find_extent(inode, iblock, NULL);
2271 err = PTR_ERR(path);
2276 depth = ext_depth(inode);
2279 * consistent leaf must not be empty;
2280 * this situation is possible, though, _during_ tree modification;
2281 * this is why assert can't be put in ext4_ext_find_extent()
2283 BUG_ON(path[depth].p_ext == NULL && depth != 0);
2284 eh = path[depth].p_hdr;
2286 ex = path[depth].p_ext;
2288 unsigned long ee_block = le32_to_cpu(ex->ee_block);
2289 ext4_fsblk_t ee_start = ext_pblock(ex);
2290 unsigned short ee_len;
2293 * Uninitialized extents are treated as holes, except that
2294 * we split out initialized portions during a write.
2296 ee_len = ext4_ext_get_actual_len(ex);
2297 /* if found extent covers block, simply return it */
2298 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2299 newblock = iblock - ee_block + ee_start;
2300 /* number of remaining blocks in the extent */
2301 allocated = ee_len - (iblock - ee_block);
2302 ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock,
2303 ee_block, ee_len, newblock);
2305 /* Do not put uninitialized extent in the cache */
2306 if (!ext4_ext_is_uninitialized(ex)) {
2307 ext4_ext_put_in_cache(inode, ee_block,
2309 EXT4_EXT_CACHE_EXTENT);
2312 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2317 ret = ext4_ext_convert_to_initialized(handle, inode,
2329 * requested block isn't allocated yet;
2330 * we couldn't try to create block if create flag is zero
2334 * put just found gap into cache to speed up
2335 * subsequent requests
2337 ext4_ext_put_gap_in_cache(inode, path, iblock);
2341 * Okay, we need to do block allocation. Lazily initialize the block
2342 * allocation info here if necessary.
2344 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2345 ext4_init_block_alloc_info(inode);
2347 /* allocate new block */
2348 goal = ext4_ext_find_goal(inode, path, iblock);
2351 * See if request is beyond maximum number of blocks we can have in
2352 * a single extent. For an initialized extent this limit is
2353 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2354 * EXT_UNINIT_MAX_LEN.
2356 if (max_blocks > EXT_INIT_MAX_LEN &&
2357 create != EXT4_CREATE_UNINITIALIZED_EXT)
2358 max_blocks = EXT_INIT_MAX_LEN;
2359 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2360 create == EXT4_CREATE_UNINITIALIZED_EXT)
2361 max_blocks = EXT_UNINIT_MAX_LEN;
2363 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2364 newex.ee_block = cpu_to_le32(iblock);
2365 newex.ee_len = cpu_to_le16(max_blocks);
2366 err = ext4_ext_check_overlap(inode, &newex, path);
2368 allocated = le16_to_cpu(newex.ee_len);
2370 allocated = max_blocks;
2371 newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2374 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2375 goal, newblock, allocated);
2377 /* try to insert new extent into found leaf and return */
2378 ext4_ext_store_pblock(&newex, newblock);
2379 newex.ee_len = cpu_to_le16(allocated);
2380 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2381 ext4_ext_mark_uninitialized(&newex);
2382 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2384 /* free data blocks we just allocated */
2385 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2386 le16_to_cpu(newex.ee_len));
2390 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2391 EXT4_I(inode)->i_disksize = inode->i_size;
2393 /* previous routine could use block we allocated */
2394 newblock = ext_pblock(&newex);
2396 __set_bit(BH_New, &bh_result->b_state);
2398 /* Cache only when it is _not_ an uninitialized extent */
2399 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2400 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2401 EXT4_EXT_CACHE_EXTENT);
2403 if (allocated > max_blocks)
2404 allocated = max_blocks;
2405 ext4_ext_show_leaf(inode, path);
2406 __set_bit(BH_Mapped, &bh_result->b_state);
2407 bh_result->b_bdev = inode->i_sb->s_bdev;
2408 bh_result->b_blocknr = newblock;
2411 ext4_ext_drop_refs(path);
2414 mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2416 return err ? err : allocated;
2419 void ext4_ext_truncate(struct inode * inode, struct page *page)
2421 struct address_space *mapping = inode->i_mapping;
2422 struct super_block *sb = inode->i_sb;
2423 unsigned long last_block;
2428 * probably first extent we're gonna free will be last in block
2430 err = ext4_writepage_trans_blocks(inode) + 3;
2431 handle = ext4_journal_start(inode, err);
2432 if (IS_ERR(handle)) {
2434 clear_highpage(page);
2435 flush_dcache_page(page);
2437 page_cache_release(page);
2443 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2445 mutex_lock(&EXT4_I(inode)->truncate_mutex);
2446 ext4_ext_invalidate_cache(inode);
2449 * TODO: optimization is possible here.
2450 * Probably we need not scan at all,
2451 * because page truncation is enough.
2453 if (ext4_orphan_add(handle, inode))
2456 /* we have to know where to truncate from in crash case */
2457 EXT4_I(inode)->i_disksize = inode->i_size;
2458 ext4_mark_inode_dirty(handle, inode);
2460 last_block = (inode->i_size + sb->s_blocksize - 1)
2461 >> EXT4_BLOCK_SIZE_BITS(sb);
2462 err = ext4_ext_remove_space(inode, last_block);
2464 /* In a multi-transaction truncate, we only make the final
2465 * transaction synchronous.
2472 * If this was a simple ftruncate() and the file will remain alive,
2473 * then we need to clear up the orphan record which we created above.
2474 * However, if this was a real unlink then we were called by
2475 * ext4_delete_inode(), and we allow that function to clean up the
2476 * orphan info for us.
2479 ext4_orphan_del(handle, inode);
2481 mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2482 ext4_journal_stop(handle);
2486 * ext4_ext_writepage_trans_blocks:
2487 * calculate max number of blocks we could modify
2488 * in order to allocate new block for an inode
2490 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2494 needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2496 /* caller wants to allocate num blocks, but note it includes sb */
2497 needed = needed * num - (num - 1);
2500 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2507 * preallocate space for a file. This implements ext4's fallocate inode
2508 * operation, which gets called from sys_fallocate system call.
2509 * For block-mapped files, posix_fallocate should fall back to the method
2510 * of writing zeroes to the required new blocks (the same behavior which is
2511 * expected for file systems which do not support fallocate() system call).
2513 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2516 ext4_fsblk_t block, max_blocks;
2517 ext4_fsblk_t nblocks = 0;
2521 struct buffer_head map_bh;
2522 unsigned int credits, blkbits = inode->i_blkbits;
2525 * currently supporting (pre)allocate mode for extent-based
2528 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2531 /* preallocation to directories is currently not supported */
2532 if (S_ISDIR(inode->i_mode))
2535 block = offset >> blkbits;
2536 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2540 * credits to insert 1 extent into extent tree + buffers to be able to
2541 * modify 1 super block, 1 block bitmap and 1 group descriptor.
2543 credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
2545 while (ret >= 0 && ret < max_blocks) {
2546 block = block + ret;
2547 max_blocks = max_blocks - ret;
2548 handle = ext4_journal_start(inode, credits);
2549 if (IS_ERR(handle)) {
2550 ret = PTR_ERR(handle);
2554 ret = ext4_ext_get_blocks(handle, inode, block,
2555 max_blocks, &map_bh,
2556 EXT4_CREATE_UNINITIALIZED_EXT, 0);
2559 ext4_error(inode->i_sb, "ext4_fallocate",
2560 "ext4_ext_get_blocks returned 0! inode#%lu"
2561 ", block=%llu, max_blocks=%llu",
2562 inode->i_ino, block, max_blocks);
2564 ext4_mark_inode_dirty(handle, inode);
2565 ret2 = ext4_journal_stop(handle);
2569 /* check wrap through sign-bit/zero here */
2570 if ((block + ret) < 0 || (block + ret) < block) {
2572 ext4_mark_inode_dirty(handle, inode);
2573 ret2 = ext4_journal_stop(handle);
2576 if (buffer_new(&map_bh) && ((block + ret) >
2577 (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits)
2579 nblocks = nblocks + ret;
2582 /* Update ctime if new blocks get allocated */
2584 struct timespec now;
2586 now = current_fs_time(inode->i_sb);
2587 if (!timespec_equal(&inode->i_ctime, &now))
2588 inode->i_ctime = now;
2591 ext4_mark_inode_dirty(handle, inode);
2592 ret2 = ext4_journal_stop(handle);
2597 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2601 * Time to update the file size.
2602 * Update only when preallocation was requested beyond the file size.
2604 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2605 (offset + len) > i_size_read(inode)) {
2608 * if no error, we assume preallocation succeeded
2611 mutex_lock(&inode->i_mutex);
2612 i_size_write(inode, offset + len);
2613 EXT4_I(inode)->i_disksize = i_size_read(inode);
2614 mutex_unlock(&inode->i_mutex);
2615 } else if (ret < 0 && nblocks) {
2616 /* Handle partial allocation scenario */
2619 mutex_lock(&inode->i_mutex);
2620 newsize = (nblocks << blkbits) + i_size_read(inode);
2621 i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits));
2622 EXT4_I(inode)->i_disksize = i_size_read(inode);
2623 mutex_unlock(&inode->i_mutex);
2627 return ret > 0 ? ret2 : ret;