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/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
47 static int ext4_ext_truncate_extend_restart(handle_t *handle,
53 if (!ext4_handle_valid(handle))
55 if (handle->h_buffer_credits > needed)
57 err = ext4_journal_extend(handle, needed);
60 err = ext4_truncate_restart_trans(handle, inode, needed);
72 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
73 struct ext4_ext_path *path)
76 /* path points to block */
77 return ext4_journal_get_write_access(handle, path->p_bh);
79 /* path points to leaf/index in inode body */
80 /* we use in-core data, no need to protect them */
90 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
91 struct ext4_ext_path *path)
95 /* path points to block */
96 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
98 /* path points to leaf/index in inode body */
99 err = ext4_mark_inode_dirty(handle, inode);
104 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
105 struct ext4_ext_path *path,
108 struct ext4_inode_info *ei = EXT4_I(inode);
109 ext4_fsblk_t bg_start;
110 ext4_fsblk_t last_block;
111 ext4_grpblk_t colour;
112 ext4_group_t block_group;
113 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
117 struct ext4_extent *ex;
118 depth = path->p_depth;
120 /* try to predict block placement */
121 ex = path[depth].p_ext;
123 return (ext4_ext_pblock(ex) +
124 (block - le32_to_cpu(ex->ee_block)));
126 /* it looks like index is empty;
127 * try to find starting block from index itself */
128 if (path[depth].p_bh)
129 return path[depth].p_bh->b_blocknr;
132 /* OK. use inode's group */
133 block_group = ei->i_block_group;
134 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
136 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
137 * block groups per flexgroup, reserve the first block
138 * group for directories and special files. Regular
139 * files will start at the second block group. This
140 * tends to speed up directory access and improves
143 block_group &= ~(flex_size-1);
144 if (S_ISREG(inode->i_mode))
147 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
148 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
151 * If we are doing delayed allocation, we don't need take
152 * colour into account.
154 if (test_opt(inode->i_sb, DELALLOC))
157 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
158 colour = (current->pid % 16) *
159 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
161 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
162 return bg_start + colour + block;
166 * Allocation for a meta data block
169 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
170 struct ext4_ext_path *path,
171 struct ext4_extent *ex, int *err)
173 ext4_fsblk_t goal, newblock;
175 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
176 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
180 static inline int ext4_ext_space_block(struct inode *inode, int check)
184 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
185 / sizeof(struct ext4_extent);
187 #ifdef AGGRESSIVE_TEST
195 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
199 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
200 / sizeof(struct ext4_extent_idx);
202 #ifdef AGGRESSIVE_TEST
210 static inline int ext4_ext_space_root(struct inode *inode, int check)
214 size = sizeof(EXT4_I(inode)->i_data);
215 size -= sizeof(struct ext4_extent_header);
216 size /= sizeof(struct ext4_extent);
218 #ifdef AGGRESSIVE_TEST
226 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
230 size = sizeof(EXT4_I(inode)->i_data);
231 size -= sizeof(struct ext4_extent_header);
232 size /= sizeof(struct ext4_extent_idx);
234 #ifdef AGGRESSIVE_TEST
243 * Calculate the number of metadata blocks needed
244 * to allocate @blocks
245 * Worse case is one block per extent
247 int ext4_ext_calc_metadata_amount(struct inode *inode, sector_t lblock)
249 struct ext4_inode_info *ei = EXT4_I(inode);
252 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
253 / sizeof(struct ext4_extent_idx));
256 * If the new delayed allocation block is contiguous with the
257 * previous da block, it can share index blocks with the
258 * previous block, so we only need to allocate a new index
259 * block every idxs leaf blocks. At ldxs**2 blocks, we need
260 * an additional index block, and at ldxs**3 blocks, yet
261 * another index blocks.
263 if (ei->i_da_metadata_calc_len &&
264 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
265 if ((ei->i_da_metadata_calc_len % idxs) == 0)
267 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
269 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
271 ei->i_da_metadata_calc_len = 0;
273 ei->i_da_metadata_calc_len++;
274 ei->i_da_metadata_calc_last_lblock++;
279 * In the worst case we need a new set of index blocks at
280 * every level of the inode's extent tree.
282 ei->i_da_metadata_calc_len = 1;
283 ei->i_da_metadata_calc_last_lblock = lblock;
284 return ext_depth(inode) + 1;
288 ext4_ext_max_entries(struct inode *inode, int depth)
292 if (depth == ext_depth(inode)) {
294 max = ext4_ext_space_root(inode, 1);
296 max = ext4_ext_space_root_idx(inode, 1);
299 max = ext4_ext_space_block(inode, 1);
301 max = ext4_ext_space_block_idx(inode, 1);
307 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
309 ext4_fsblk_t block = ext4_ext_pblock(ext);
310 int len = ext4_ext_get_actual_len(ext);
312 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
315 static int ext4_valid_extent_idx(struct inode *inode,
316 struct ext4_extent_idx *ext_idx)
318 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
320 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
323 static int ext4_valid_extent_entries(struct inode *inode,
324 struct ext4_extent_header *eh,
327 struct ext4_extent *ext;
328 struct ext4_extent_idx *ext_idx;
329 unsigned short entries;
330 if (eh->eh_entries == 0)
333 entries = le16_to_cpu(eh->eh_entries);
337 ext = EXT_FIRST_EXTENT(eh);
339 if (!ext4_valid_extent(inode, ext))
345 ext_idx = EXT_FIRST_INDEX(eh);
347 if (!ext4_valid_extent_idx(inode, ext_idx))
356 static int __ext4_ext_check(const char *function, unsigned int line,
357 struct inode *inode, struct ext4_extent_header *eh,
360 const char *error_msg;
363 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
364 error_msg = "invalid magic";
367 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
368 error_msg = "unexpected eh_depth";
371 if (unlikely(eh->eh_max == 0)) {
372 error_msg = "invalid eh_max";
375 max = ext4_ext_max_entries(inode, depth);
376 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
377 error_msg = "too large eh_max";
380 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
381 error_msg = "invalid eh_entries";
384 if (!ext4_valid_extent_entries(inode, eh, depth)) {
385 error_msg = "invalid extent entries";
391 ext4_error_inode(inode, function, line, 0,
392 "bad header/extent: %s - magic %x, "
393 "entries %u, max %u(%u), depth %u(%u)",
394 error_msg, le16_to_cpu(eh->eh_magic),
395 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
396 max, le16_to_cpu(eh->eh_depth), depth);
401 #define ext4_ext_check(inode, eh, depth) \
402 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
404 int ext4_ext_check_inode(struct inode *inode)
406 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
410 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
412 int k, l = path->p_depth;
415 for (k = 0; k <= l; k++, path++) {
417 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
418 ext4_idx_pblock(path->p_idx));
419 } else if (path->p_ext) {
420 ext_debug(" %d:[%d]%d:%llu ",
421 le32_to_cpu(path->p_ext->ee_block),
422 ext4_ext_is_uninitialized(path->p_ext),
423 ext4_ext_get_actual_len(path->p_ext),
424 ext4_ext_pblock(path->p_ext));
431 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
433 int depth = ext_depth(inode);
434 struct ext4_extent_header *eh;
435 struct ext4_extent *ex;
441 eh = path[depth].p_hdr;
442 ex = EXT_FIRST_EXTENT(eh);
444 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
446 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
447 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
448 ext4_ext_is_uninitialized(ex),
449 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
454 #define ext4_ext_show_path(inode, path)
455 #define ext4_ext_show_leaf(inode, path)
458 void ext4_ext_drop_refs(struct ext4_ext_path *path)
460 int depth = path->p_depth;
463 for (i = 0; i <= depth; i++, path++)
471 * ext4_ext_binsearch_idx:
472 * binary search for the closest index of the given block
473 * the header must be checked before calling this
476 ext4_ext_binsearch_idx(struct inode *inode,
477 struct ext4_ext_path *path, ext4_lblk_t block)
479 struct ext4_extent_header *eh = path->p_hdr;
480 struct ext4_extent_idx *r, *l, *m;
483 ext_debug("binsearch for %u(idx): ", block);
485 l = EXT_FIRST_INDEX(eh) + 1;
486 r = EXT_LAST_INDEX(eh);
489 if (block < le32_to_cpu(m->ei_block))
493 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
494 m, le32_to_cpu(m->ei_block),
495 r, le32_to_cpu(r->ei_block));
499 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
500 ext4_idx_pblock(path->p_idx));
502 #ifdef CHECK_BINSEARCH
504 struct ext4_extent_idx *chix, *ix;
507 chix = ix = EXT_FIRST_INDEX(eh);
508 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
510 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
511 printk(KERN_DEBUG "k=%d, ix=0x%p, "
513 ix, EXT_FIRST_INDEX(eh));
514 printk(KERN_DEBUG "%u <= %u\n",
515 le32_to_cpu(ix->ei_block),
516 le32_to_cpu(ix[-1].ei_block));
518 BUG_ON(k && le32_to_cpu(ix->ei_block)
519 <= le32_to_cpu(ix[-1].ei_block));
520 if (block < le32_to_cpu(ix->ei_block))
524 BUG_ON(chix != path->p_idx);
531 * ext4_ext_binsearch:
532 * binary search for closest extent of the given block
533 * the header must be checked before calling this
536 ext4_ext_binsearch(struct inode *inode,
537 struct ext4_ext_path *path, ext4_lblk_t block)
539 struct ext4_extent_header *eh = path->p_hdr;
540 struct ext4_extent *r, *l, *m;
542 if (eh->eh_entries == 0) {
544 * this leaf is empty:
545 * we get such a leaf in split/add case
550 ext_debug("binsearch for %u: ", block);
552 l = EXT_FIRST_EXTENT(eh) + 1;
553 r = EXT_LAST_EXTENT(eh);
557 if (block < le32_to_cpu(m->ee_block))
561 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
562 m, le32_to_cpu(m->ee_block),
563 r, le32_to_cpu(r->ee_block));
567 ext_debug(" -> %d:%llu:[%d]%d ",
568 le32_to_cpu(path->p_ext->ee_block),
569 ext4_ext_pblock(path->p_ext),
570 ext4_ext_is_uninitialized(path->p_ext),
571 ext4_ext_get_actual_len(path->p_ext));
573 #ifdef CHECK_BINSEARCH
575 struct ext4_extent *chex, *ex;
578 chex = ex = EXT_FIRST_EXTENT(eh);
579 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
580 BUG_ON(k && le32_to_cpu(ex->ee_block)
581 <= le32_to_cpu(ex[-1].ee_block));
582 if (block < le32_to_cpu(ex->ee_block))
586 BUG_ON(chex != path->p_ext);
592 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
594 struct ext4_extent_header *eh;
596 eh = ext_inode_hdr(inode);
599 eh->eh_magic = EXT4_EXT_MAGIC;
600 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
601 ext4_mark_inode_dirty(handle, inode);
602 ext4_ext_invalidate_cache(inode);
606 struct ext4_ext_path *
607 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
608 struct ext4_ext_path *path)
610 struct ext4_extent_header *eh;
611 struct buffer_head *bh;
612 short int depth, i, ppos = 0, alloc = 0;
614 eh = ext_inode_hdr(inode);
615 depth = ext_depth(inode);
617 /* account possible depth increase */
619 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
622 return ERR_PTR(-ENOMEM);
629 /* walk through the tree */
631 int need_to_validate = 0;
633 ext_debug("depth %d: num %d, max %d\n",
634 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
636 ext4_ext_binsearch_idx(inode, path + ppos, block);
637 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
638 path[ppos].p_depth = i;
639 path[ppos].p_ext = NULL;
641 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
644 if (!bh_uptodate_or_lock(bh)) {
645 if (bh_submit_read(bh) < 0) {
649 /* validate the extent entries */
650 need_to_validate = 1;
652 eh = ext_block_hdr(bh);
654 if (unlikely(ppos > depth)) {
656 EXT4_ERROR_INODE(inode,
657 "ppos %d > depth %d", ppos, depth);
660 path[ppos].p_bh = bh;
661 path[ppos].p_hdr = eh;
664 if (need_to_validate && ext4_ext_check(inode, eh, i))
668 path[ppos].p_depth = i;
669 path[ppos].p_ext = NULL;
670 path[ppos].p_idx = NULL;
673 ext4_ext_binsearch(inode, path + ppos, block);
674 /* if not an empty leaf */
675 if (path[ppos].p_ext)
676 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
678 ext4_ext_show_path(inode, path);
683 ext4_ext_drop_refs(path);
686 return ERR_PTR(-EIO);
690 * ext4_ext_insert_index:
691 * insert new index [@logical;@ptr] into the block at @curp;
692 * check where to insert: before @curp or after @curp
694 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
695 struct ext4_ext_path *curp,
696 int logical, ext4_fsblk_t ptr)
698 struct ext4_extent_idx *ix;
701 err = ext4_ext_get_access(handle, inode, curp);
705 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
706 EXT4_ERROR_INODE(inode,
707 "logical %d == ei_block %d!",
708 logical, le32_to_cpu(curp->p_idx->ei_block));
711 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
712 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
714 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
715 len = (len - 1) * sizeof(struct ext4_extent_idx);
716 len = len < 0 ? 0 : len;
717 ext_debug("insert new index %d after: %llu. "
718 "move %d from 0x%p to 0x%p\n",
720 (curp->p_idx + 1), (curp->p_idx + 2));
721 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
723 ix = curp->p_idx + 1;
726 len = len * sizeof(struct ext4_extent_idx);
727 len = len < 0 ? 0 : len;
728 ext_debug("insert new index %d before: %llu. "
729 "move %d from 0x%p to 0x%p\n",
731 curp->p_idx, (curp->p_idx + 1));
732 memmove(curp->p_idx + 1, curp->p_idx, len);
736 ix->ei_block = cpu_to_le32(logical);
737 ext4_idx_store_pblock(ix, ptr);
738 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
740 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
741 > le16_to_cpu(curp->p_hdr->eh_max))) {
742 EXT4_ERROR_INODE(inode,
743 "logical %d == ei_block %d!",
744 logical, le32_to_cpu(curp->p_idx->ei_block));
747 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
748 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
752 err = ext4_ext_dirty(handle, inode, curp);
753 ext4_std_error(inode->i_sb, err);
760 * inserts new subtree into the path, using free index entry
762 * - allocates all needed blocks (new leaf and all intermediate index blocks)
763 * - makes decision where to split
764 * - moves remaining extents and index entries (right to the split point)
765 * into the newly allocated blocks
766 * - initializes subtree
768 static int ext4_ext_split(handle_t *handle, struct inode *inode,
769 struct ext4_ext_path *path,
770 struct ext4_extent *newext, int at)
772 struct buffer_head *bh = NULL;
773 int depth = ext_depth(inode);
774 struct ext4_extent_header *neh;
775 struct ext4_extent_idx *fidx;
776 struct ext4_extent *ex;
778 ext4_fsblk_t newblock, oldblock;
780 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
783 /* make decision: where to split? */
784 /* FIXME: now decision is simplest: at current extent */
786 /* if current leaf will be split, then we should use
787 * border from split point */
788 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
789 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
792 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
793 border = path[depth].p_ext[1].ee_block;
794 ext_debug("leaf will be split."
795 " next leaf starts at %d\n",
796 le32_to_cpu(border));
798 border = newext->ee_block;
799 ext_debug("leaf will be added."
800 " next leaf starts at %d\n",
801 le32_to_cpu(border));
805 * If error occurs, then we break processing
806 * and mark filesystem read-only. index won't
807 * be inserted and tree will be in consistent
808 * state. Next mount will repair buffers too.
812 * Get array to track all allocated blocks.
813 * We need this to handle errors and free blocks
816 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
820 /* allocate all needed blocks */
821 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
822 for (a = 0; a < depth - at; a++) {
823 newblock = ext4_ext_new_meta_block(handle, inode, path,
827 ablocks[a] = newblock;
830 /* initialize new leaf */
831 newblock = ablocks[--a];
832 if (unlikely(newblock == 0)) {
833 EXT4_ERROR_INODE(inode, "newblock == 0!");
837 bh = sb_getblk(inode->i_sb, newblock);
844 err = ext4_journal_get_create_access(handle, bh);
848 neh = ext_block_hdr(bh);
850 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
851 neh->eh_magic = EXT4_EXT_MAGIC;
853 ex = EXT_FIRST_EXTENT(neh);
855 /* move remainder of path[depth] to the new leaf */
856 if (unlikely(path[depth].p_hdr->eh_entries !=
857 path[depth].p_hdr->eh_max)) {
858 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
859 path[depth].p_hdr->eh_entries,
860 path[depth].p_hdr->eh_max);
864 /* start copy from next extent */
865 /* TODO: we could do it by single memmove */
868 while (path[depth].p_ext <=
869 EXT_MAX_EXTENT(path[depth].p_hdr)) {
870 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
871 le32_to_cpu(path[depth].p_ext->ee_block),
872 ext4_ext_pblock(path[depth].p_ext),
873 ext4_ext_is_uninitialized(path[depth].p_ext),
874 ext4_ext_get_actual_len(path[depth].p_ext),
876 /*memmove(ex++, path[depth].p_ext++,
877 sizeof(struct ext4_extent));
883 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
884 le16_add_cpu(&neh->eh_entries, m);
887 set_buffer_uptodate(bh);
890 err = ext4_handle_dirty_metadata(handle, inode, bh);
896 /* correct old leaf */
898 err = ext4_ext_get_access(handle, inode, path + depth);
901 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
902 err = ext4_ext_dirty(handle, inode, path + depth);
908 /* create intermediate indexes */
910 if (unlikely(k < 0)) {
911 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
916 ext_debug("create %d intermediate indices\n", k);
917 /* insert new index into current index block */
918 /* current depth stored in i var */
922 newblock = ablocks[--a];
923 bh = sb_getblk(inode->i_sb, newblock);
930 err = ext4_journal_get_create_access(handle, bh);
934 neh = ext_block_hdr(bh);
935 neh->eh_entries = cpu_to_le16(1);
936 neh->eh_magic = EXT4_EXT_MAGIC;
937 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
938 neh->eh_depth = cpu_to_le16(depth - i);
939 fidx = EXT_FIRST_INDEX(neh);
940 fidx->ei_block = border;
941 ext4_idx_store_pblock(fidx, oldblock);
943 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
944 i, newblock, le32_to_cpu(border), oldblock);
949 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
950 EXT_MAX_INDEX(path[i].p_hdr));
951 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
952 EXT_LAST_INDEX(path[i].p_hdr))) {
953 EXT4_ERROR_INODE(inode,
954 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
955 le32_to_cpu(path[i].p_ext->ee_block));
959 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
960 ext_debug("%d: move %d:%llu in new index %llu\n", i,
961 le32_to_cpu(path[i].p_idx->ei_block),
962 ext4_idx_pblock(path[i].p_idx),
964 /*memmove(++fidx, path[i].p_idx++,
965 sizeof(struct ext4_extent_idx));
967 BUG_ON(neh->eh_entries > neh->eh_max);*/
972 memmove(++fidx, path[i].p_idx - m,
973 sizeof(struct ext4_extent_idx) * m);
974 le16_add_cpu(&neh->eh_entries, m);
976 set_buffer_uptodate(bh);
979 err = ext4_handle_dirty_metadata(handle, inode, bh);
985 /* correct old index */
987 err = ext4_ext_get_access(handle, inode, path + i);
990 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
991 err = ext4_ext_dirty(handle, inode, path + i);
999 /* insert new index */
1000 err = ext4_ext_insert_index(handle, inode, path + at,
1001 le32_to_cpu(border), newblock);
1005 if (buffer_locked(bh))
1011 /* free all allocated blocks in error case */
1012 for (i = 0; i < depth; i++) {
1015 ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1016 EXT4_FREE_BLOCKS_METADATA);
1025 * ext4_ext_grow_indepth:
1026 * implements tree growing procedure:
1027 * - allocates new block
1028 * - moves top-level data (index block or leaf) into the new block
1029 * - initializes new top-level, creating index that points to the
1030 * just created block
1032 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1033 struct ext4_ext_path *path,
1034 struct ext4_extent *newext)
1036 struct ext4_ext_path *curp = path;
1037 struct ext4_extent_header *neh;
1038 struct buffer_head *bh;
1039 ext4_fsblk_t newblock;
1042 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1046 bh = sb_getblk(inode->i_sb, newblock);
1049 ext4_std_error(inode->i_sb, err);
1054 err = ext4_journal_get_create_access(handle, bh);
1060 /* move top-level index/leaf into new block */
1061 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1063 /* set size of new block */
1064 neh = ext_block_hdr(bh);
1065 /* old root could have indexes or leaves
1066 * so calculate e_max right way */
1067 if (ext_depth(inode))
1068 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1070 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1071 neh->eh_magic = EXT4_EXT_MAGIC;
1072 set_buffer_uptodate(bh);
1075 err = ext4_handle_dirty_metadata(handle, inode, bh);
1079 /* create index in new top-level index: num,max,pointer */
1080 err = ext4_ext_get_access(handle, inode, curp);
1084 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1085 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1086 curp->p_hdr->eh_entries = cpu_to_le16(1);
1087 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1089 if (path[0].p_hdr->eh_depth)
1090 curp->p_idx->ei_block =
1091 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1093 curp->p_idx->ei_block =
1094 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1095 ext4_idx_store_pblock(curp->p_idx, newblock);
1097 neh = ext_inode_hdr(inode);
1098 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1099 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1100 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1101 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1103 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1104 err = ext4_ext_dirty(handle, inode, curp);
1112 * ext4_ext_create_new_leaf:
1113 * finds empty index and adds new leaf.
1114 * if no free index is found, then it requests in-depth growing.
1116 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1117 struct ext4_ext_path *path,
1118 struct ext4_extent *newext)
1120 struct ext4_ext_path *curp;
1121 int depth, i, err = 0;
1124 i = depth = ext_depth(inode);
1126 /* walk up to the tree and look for free index entry */
1127 curp = path + depth;
1128 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1133 /* we use already allocated block for index block,
1134 * so subsequent data blocks should be contiguous */
1135 if (EXT_HAS_FREE_INDEX(curp)) {
1136 /* if we found index with free entry, then use that
1137 * entry: create all needed subtree and add new leaf */
1138 err = ext4_ext_split(handle, inode, path, newext, i);
1143 ext4_ext_drop_refs(path);
1144 path = ext4_ext_find_extent(inode,
1145 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1148 err = PTR_ERR(path);
1150 /* tree is full, time to grow in depth */
1151 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1156 ext4_ext_drop_refs(path);
1157 path = ext4_ext_find_extent(inode,
1158 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1161 err = PTR_ERR(path);
1166 * only first (depth 0 -> 1) produces free space;
1167 * in all other cases we have to split the grown tree
1169 depth = ext_depth(inode);
1170 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1171 /* now we need to split */
1181 * search the closest allocated block to the left for *logical
1182 * and returns it at @logical + it's physical address at @phys
1183 * if *logical is the smallest allocated block, the function
1184 * returns 0 at @phys
1185 * return value contains 0 (success) or error code
1187 static int ext4_ext_search_left(struct inode *inode,
1188 struct ext4_ext_path *path,
1189 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1191 struct ext4_extent_idx *ix;
1192 struct ext4_extent *ex;
1195 if (unlikely(path == NULL)) {
1196 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1199 depth = path->p_depth;
1202 if (depth == 0 && path->p_ext == NULL)
1205 /* usually extent in the path covers blocks smaller
1206 * then *logical, but it can be that extent is the
1207 * first one in the file */
1209 ex = path[depth].p_ext;
1210 ee_len = ext4_ext_get_actual_len(ex);
1211 if (*logical < le32_to_cpu(ex->ee_block)) {
1212 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1213 EXT4_ERROR_INODE(inode,
1214 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1215 *logical, le32_to_cpu(ex->ee_block));
1218 while (--depth >= 0) {
1219 ix = path[depth].p_idx;
1220 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1221 EXT4_ERROR_INODE(inode,
1222 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1223 ix != NULL ? ix->ei_block : 0,
1224 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1225 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1233 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1234 EXT4_ERROR_INODE(inode,
1235 "logical %d < ee_block %d + ee_len %d!",
1236 *logical, le32_to_cpu(ex->ee_block), ee_len);
1240 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1241 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1246 * search the closest allocated block to the right for *logical
1247 * and returns it at @logical + it's physical address at @phys
1248 * if *logical is the smallest allocated block, the function
1249 * returns 0 at @phys
1250 * return value contains 0 (success) or error code
1252 static int ext4_ext_search_right(struct inode *inode,
1253 struct ext4_ext_path *path,
1254 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1256 struct buffer_head *bh = NULL;
1257 struct ext4_extent_header *eh;
1258 struct ext4_extent_idx *ix;
1259 struct ext4_extent *ex;
1261 int depth; /* Note, NOT eh_depth; depth from top of tree */
1264 if (unlikely(path == NULL)) {
1265 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1268 depth = path->p_depth;
1271 if (depth == 0 && path->p_ext == NULL)
1274 /* usually extent in the path covers blocks smaller
1275 * then *logical, but it can be that extent is the
1276 * first one in the file */
1278 ex = path[depth].p_ext;
1279 ee_len = ext4_ext_get_actual_len(ex);
1280 if (*logical < le32_to_cpu(ex->ee_block)) {
1281 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1282 EXT4_ERROR_INODE(inode,
1283 "first_extent(path[%d].p_hdr) != ex",
1287 while (--depth >= 0) {
1288 ix = path[depth].p_idx;
1289 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1290 EXT4_ERROR_INODE(inode,
1291 "ix != EXT_FIRST_INDEX *logical %d!",
1296 *logical = le32_to_cpu(ex->ee_block);
1297 *phys = ext4_ext_pblock(ex);
1301 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1302 EXT4_ERROR_INODE(inode,
1303 "logical %d < ee_block %d + ee_len %d!",
1304 *logical, le32_to_cpu(ex->ee_block), ee_len);
1308 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1309 /* next allocated block in this leaf */
1311 *logical = le32_to_cpu(ex->ee_block);
1312 *phys = ext4_ext_pblock(ex);
1316 /* go up and search for index to the right */
1317 while (--depth >= 0) {
1318 ix = path[depth].p_idx;
1319 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1323 /* we've gone up to the root and found no index to the right */
1327 /* we've found index to the right, let's
1328 * follow it and find the closest allocated
1329 * block to the right */
1331 block = ext4_idx_pblock(ix);
1332 while (++depth < path->p_depth) {
1333 bh = sb_bread(inode->i_sb, block);
1336 eh = ext_block_hdr(bh);
1337 /* subtract from p_depth to get proper eh_depth */
1338 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1342 ix = EXT_FIRST_INDEX(eh);
1343 block = ext4_idx_pblock(ix);
1347 bh = sb_bread(inode->i_sb, block);
1350 eh = ext_block_hdr(bh);
1351 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1355 ex = EXT_FIRST_EXTENT(eh);
1356 *logical = le32_to_cpu(ex->ee_block);
1357 *phys = ext4_ext_pblock(ex);
1363 * ext4_ext_next_allocated_block:
1364 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1365 * NOTE: it considers block number from index entry as
1366 * allocated block. Thus, index entries have to be consistent
1370 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1374 BUG_ON(path == NULL);
1375 depth = path->p_depth;
1377 if (depth == 0 && path->p_ext == NULL)
1378 return EXT_MAX_BLOCK;
1380 while (depth >= 0) {
1381 if (depth == path->p_depth) {
1383 if (path[depth].p_ext !=
1384 EXT_LAST_EXTENT(path[depth].p_hdr))
1385 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1388 if (path[depth].p_idx !=
1389 EXT_LAST_INDEX(path[depth].p_hdr))
1390 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1395 return EXT_MAX_BLOCK;
1399 * ext4_ext_next_leaf_block:
1400 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1402 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1403 struct ext4_ext_path *path)
1407 BUG_ON(path == NULL);
1408 depth = path->p_depth;
1410 /* zero-tree has no leaf blocks at all */
1412 return EXT_MAX_BLOCK;
1414 /* go to index block */
1417 while (depth >= 0) {
1418 if (path[depth].p_idx !=
1419 EXT_LAST_INDEX(path[depth].p_hdr))
1420 return (ext4_lblk_t)
1421 le32_to_cpu(path[depth].p_idx[1].ei_block);
1425 return EXT_MAX_BLOCK;
1429 * ext4_ext_correct_indexes:
1430 * if leaf gets modified and modified extent is first in the leaf,
1431 * then we have to correct all indexes above.
1432 * TODO: do we need to correct tree in all cases?
1434 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1435 struct ext4_ext_path *path)
1437 struct ext4_extent_header *eh;
1438 int depth = ext_depth(inode);
1439 struct ext4_extent *ex;
1443 eh = path[depth].p_hdr;
1444 ex = path[depth].p_ext;
1446 if (unlikely(ex == NULL || eh == NULL)) {
1447 EXT4_ERROR_INODE(inode,
1448 "ex %p == NULL or eh %p == NULL", ex, eh);
1453 /* there is no tree at all */
1457 if (ex != EXT_FIRST_EXTENT(eh)) {
1458 /* we correct tree if first leaf got modified only */
1463 * TODO: we need correction if border is smaller than current one
1466 border = path[depth].p_ext->ee_block;
1467 err = ext4_ext_get_access(handle, inode, path + k);
1470 path[k].p_idx->ei_block = border;
1471 err = ext4_ext_dirty(handle, inode, path + k);
1476 /* change all left-side indexes */
1477 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1479 err = ext4_ext_get_access(handle, inode, path + k);
1482 path[k].p_idx->ei_block = border;
1483 err = ext4_ext_dirty(handle, inode, path + k);
1492 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1493 struct ext4_extent *ex2)
1495 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1498 * Make sure that either both extents are uninitialized, or
1501 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1504 if (ext4_ext_is_uninitialized(ex1))
1505 max_len = EXT_UNINIT_MAX_LEN;
1507 max_len = EXT_INIT_MAX_LEN;
1509 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1510 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1512 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1513 le32_to_cpu(ex2->ee_block))
1517 * To allow future support for preallocated extents to be added
1518 * as an RO_COMPAT feature, refuse to merge to extents if
1519 * this can result in the top bit of ee_len being set.
1521 if (ext1_ee_len + ext2_ee_len > max_len)
1523 #ifdef AGGRESSIVE_TEST
1524 if (ext1_ee_len >= 4)
1528 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1534 * This function tries to merge the "ex" extent to the next extent in the tree.
1535 * It always tries to merge towards right. If you want to merge towards
1536 * left, pass "ex - 1" as argument instead of "ex".
1537 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1538 * 1 if they got merged.
1540 static int ext4_ext_try_to_merge(struct inode *inode,
1541 struct ext4_ext_path *path,
1542 struct ext4_extent *ex)
1544 struct ext4_extent_header *eh;
1545 unsigned int depth, len;
1547 int uninitialized = 0;
1549 depth = ext_depth(inode);
1550 BUG_ON(path[depth].p_hdr == NULL);
1551 eh = path[depth].p_hdr;
1553 while (ex < EXT_LAST_EXTENT(eh)) {
1554 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1556 /* merge with next extent! */
1557 if (ext4_ext_is_uninitialized(ex))
1559 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1560 + ext4_ext_get_actual_len(ex + 1));
1562 ext4_ext_mark_uninitialized(ex);
1564 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1565 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1566 * sizeof(struct ext4_extent);
1567 memmove(ex + 1, ex + 2, len);
1569 le16_add_cpu(&eh->eh_entries, -1);
1571 WARN_ON(eh->eh_entries == 0);
1572 if (!eh->eh_entries)
1573 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1580 * check if a portion of the "newext" extent overlaps with an
1583 * If there is an overlap discovered, it updates the length of the newext
1584 * such that there will be no overlap, and then returns 1.
1585 * If there is no overlap found, it returns 0.
1587 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1588 struct ext4_extent *newext,
1589 struct ext4_ext_path *path)
1592 unsigned int depth, len1;
1593 unsigned int ret = 0;
1595 b1 = le32_to_cpu(newext->ee_block);
1596 len1 = ext4_ext_get_actual_len(newext);
1597 depth = ext_depth(inode);
1598 if (!path[depth].p_ext)
1600 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1603 * get the next allocated block if the extent in the path
1604 * is before the requested block(s)
1607 b2 = ext4_ext_next_allocated_block(path);
1608 if (b2 == EXT_MAX_BLOCK)
1612 /* check for wrap through zero on extent logical start block*/
1613 if (b1 + len1 < b1) {
1614 len1 = EXT_MAX_BLOCK - b1;
1615 newext->ee_len = cpu_to_le16(len1);
1619 /* check for overlap */
1620 if (b1 + len1 > b2) {
1621 newext->ee_len = cpu_to_le16(b2 - b1);
1629 * ext4_ext_insert_extent:
1630 * tries to merge requsted extent into the existing extent or
1631 * inserts requested extent as new one into the tree,
1632 * creating new leaf in the no-space case.
1634 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1635 struct ext4_ext_path *path,
1636 struct ext4_extent *newext, int flag)
1638 struct ext4_extent_header *eh;
1639 struct ext4_extent *ex, *fex;
1640 struct ext4_extent *nearex; /* nearest extent */
1641 struct ext4_ext_path *npath = NULL;
1642 int depth, len, err;
1644 unsigned uninitialized = 0;
1646 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1647 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1650 depth = ext_depth(inode);
1651 ex = path[depth].p_ext;
1652 if (unlikely(path[depth].p_hdr == NULL)) {
1653 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1657 /* try to insert block into found extent and return */
1658 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1659 && ext4_can_extents_be_merged(inode, ex, newext)) {
1660 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1661 ext4_ext_is_uninitialized(newext),
1662 ext4_ext_get_actual_len(newext),
1663 le32_to_cpu(ex->ee_block),
1664 ext4_ext_is_uninitialized(ex),
1665 ext4_ext_get_actual_len(ex),
1666 ext4_ext_pblock(ex));
1667 err = ext4_ext_get_access(handle, inode, path + depth);
1672 * ext4_can_extents_be_merged should have checked that either
1673 * both extents are uninitialized, or both aren't. Thus we
1674 * need to check only one of them here.
1676 if (ext4_ext_is_uninitialized(ex))
1678 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1679 + ext4_ext_get_actual_len(newext));
1681 ext4_ext_mark_uninitialized(ex);
1682 eh = path[depth].p_hdr;
1688 depth = ext_depth(inode);
1689 eh = path[depth].p_hdr;
1690 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1693 /* probably next leaf has space for us? */
1694 fex = EXT_LAST_EXTENT(eh);
1695 next = ext4_ext_next_leaf_block(inode, path);
1696 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1697 && next != EXT_MAX_BLOCK) {
1698 ext_debug("next leaf block - %d\n", next);
1699 BUG_ON(npath != NULL);
1700 npath = ext4_ext_find_extent(inode, next, NULL);
1702 return PTR_ERR(npath);
1703 BUG_ON(npath->p_depth != path->p_depth);
1704 eh = npath[depth].p_hdr;
1705 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1706 ext_debug("next leaf isnt full(%d)\n",
1707 le16_to_cpu(eh->eh_entries));
1711 ext_debug("next leaf has no free space(%d,%d)\n",
1712 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1716 * There is no free space in the found leaf.
1717 * We're gonna add a new leaf in the tree.
1719 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1722 depth = ext_depth(inode);
1723 eh = path[depth].p_hdr;
1726 nearex = path[depth].p_ext;
1728 err = ext4_ext_get_access(handle, inode, path + depth);
1733 /* there is no extent in this leaf, create first one */
1734 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1735 le32_to_cpu(newext->ee_block),
1736 ext4_ext_pblock(newext),
1737 ext4_ext_is_uninitialized(newext),
1738 ext4_ext_get_actual_len(newext));
1739 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1740 } else if (le32_to_cpu(newext->ee_block)
1741 > le32_to_cpu(nearex->ee_block)) {
1742 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1743 if (nearex != EXT_LAST_EXTENT(eh)) {
1744 len = EXT_MAX_EXTENT(eh) - nearex;
1745 len = (len - 1) * sizeof(struct ext4_extent);
1746 len = len < 0 ? 0 : len;
1747 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1748 "move %d from 0x%p to 0x%p\n",
1749 le32_to_cpu(newext->ee_block),
1750 ext4_ext_pblock(newext),
1751 ext4_ext_is_uninitialized(newext),
1752 ext4_ext_get_actual_len(newext),
1753 nearex, len, nearex + 1, nearex + 2);
1754 memmove(nearex + 2, nearex + 1, len);
1756 path[depth].p_ext = nearex + 1;
1758 BUG_ON(newext->ee_block == nearex->ee_block);
1759 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1760 len = len < 0 ? 0 : len;
1761 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1762 "move %d from 0x%p to 0x%p\n",
1763 le32_to_cpu(newext->ee_block),
1764 ext4_ext_pblock(newext),
1765 ext4_ext_is_uninitialized(newext),
1766 ext4_ext_get_actual_len(newext),
1767 nearex, len, nearex + 1, nearex + 2);
1768 memmove(nearex + 1, nearex, len);
1769 path[depth].p_ext = nearex;
1772 le16_add_cpu(&eh->eh_entries, 1);
1773 nearex = path[depth].p_ext;
1774 nearex->ee_block = newext->ee_block;
1775 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1776 nearex->ee_len = newext->ee_len;
1779 /* try to merge extents to the right */
1780 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1781 ext4_ext_try_to_merge(inode, path, nearex);
1783 /* try to merge extents to the left */
1785 /* time to correct all indexes above */
1786 err = ext4_ext_correct_indexes(handle, inode, path);
1790 err = ext4_ext_dirty(handle, inode, path + depth);
1794 ext4_ext_drop_refs(npath);
1797 ext4_ext_invalidate_cache(inode);
1801 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1802 ext4_lblk_t num, ext_prepare_callback func,
1805 struct ext4_ext_path *path = NULL;
1806 struct ext4_ext_cache cbex;
1807 struct ext4_extent *ex;
1808 ext4_lblk_t next, start = 0, end = 0;
1809 ext4_lblk_t last = block + num;
1810 int depth, exists, err = 0;
1812 BUG_ON(func == NULL);
1813 BUG_ON(inode == NULL);
1815 while (block < last && block != EXT_MAX_BLOCK) {
1817 /* find extent for this block */
1818 down_read(&EXT4_I(inode)->i_data_sem);
1819 path = ext4_ext_find_extent(inode, block, path);
1820 up_read(&EXT4_I(inode)->i_data_sem);
1822 err = PTR_ERR(path);
1827 depth = ext_depth(inode);
1828 if (unlikely(path[depth].p_hdr == NULL)) {
1829 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1833 ex = path[depth].p_ext;
1834 next = ext4_ext_next_allocated_block(path);
1838 /* there is no extent yet, so try to allocate
1839 * all requested space */
1842 } else if (le32_to_cpu(ex->ee_block) > block) {
1843 /* need to allocate space before found extent */
1845 end = le32_to_cpu(ex->ee_block);
1846 if (block + num < end)
1848 } else if (block >= le32_to_cpu(ex->ee_block)
1849 + ext4_ext_get_actual_len(ex)) {
1850 /* need to allocate space after found extent */
1855 } else if (block >= le32_to_cpu(ex->ee_block)) {
1857 * some part of requested space is covered
1861 end = le32_to_cpu(ex->ee_block)
1862 + ext4_ext_get_actual_len(ex);
1863 if (block + num < end)
1869 BUG_ON(end <= start);
1872 cbex.ec_block = start;
1873 cbex.ec_len = end - start;
1875 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1877 cbex.ec_block = le32_to_cpu(ex->ee_block);
1878 cbex.ec_len = ext4_ext_get_actual_len(ex);
1879 cbex.ec_start = ext4_ext_pblock(ex);
1880 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1883 if (unlikely(cbex.ec_len == 0)) {
1884 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1888 err = func(inode, path, &cbex, ex, cbdata);
1889 ext4_ext_drop_refs(path);
1894 if (err == EXT_REPEAT)
1896 else if (err == EXT_BREAK) {
1901 if (ext_depth(inode) != depth) {
1902 /* depth was changed. we have to realloc path */
1907 block = cbex.ec_block + cbex.ec_len;
1911 ext4_ext_drop_refs(path);
1919 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1920 __u32 len, ext4_fsblk_t start, int type)
1922 struct ext4_ext_cache *cex;
1924 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1925 cex = &EXT4_I(inode)->i_cached_extent;
1926 cex->ec_type = type;
1927 cex->ec_block = block;
1929 cex->ec_start = start;
1930 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1934 * ext4_ext_put_gap_in_cache:
1935 * calculate boundaries of the gap that the requested block fits into
1936 * and cache this gap
1939 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1942 int depth = ext_depth(inode);
1945 struct ext4_extent *ex;
1947 ex = path[depth].p_ext;
1949 /* there is no extent yet, so gap is [0;-] */
1951 len = EXT_MAX_BLOCK;
1952 ext_debug("cache gap(whole file):");
1953 } else if (block < le32_to_cpu(ex->ee_block)) {
1955 len = le32_to_cpu(ex->ee_block) - block;
1956 ext_debug("cache gap(before): %u [%u:%u]",
1958 le32_to_cpu(ex->ee_block),
1959 ext4_ext_get_actual_len(ex));
1960 } else if (block >= le32_to_cpu(ex->ee_block)
1961 + ext4_ext_get_actual_len(ex)) {
1963 lblock = le32_to_cpu(ex->ee_block)
1964 + ext4_ext_get_actual_len(ex);
1966 next = ext4_ext_next_allocated_block(path);
1967 ext_debug("cache gap(after): [%u:%u] %u",
1968 le32_to_cpu(ex->ee_block),
1969 ext4_ext_get_actual_len(ex),
1971 BUG_ON(next == lblock);
1972 len = next - lblock;
1978 ext_debug(" -> %u:%lu\n", lblock, len);
1979 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1983 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
1984 struct ext4_extent *ex)
1986 struct ext4_ext_cache *cex;
1987 int ret = EXT4_EXT_CACHE_NO;
1990 * We borrow i_block_reservation_lock to protect i_cached_extent
1992 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1993 cex = &EXT4_I(inode)->i_cached_extent;
1995 /* has cache valid data? */
1996 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1999 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
2000 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
2001 if (in_range(block, cex->ec_block, cex->ec_len)) {
2002 ex->ee_block = cpu_to_le32(cex->ec_block);
2003 ext4_ext_store_pblock(ex, cex->ec_start);
2004 ex->ee_len = cpu_to_le16(cex->ec_len);
2005 ext_debug("%u cached by %u:%u:%llu\n",
2007 cex->ec_block, cex->ec_len, cex->ec_start);
2011 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2017 * removes index from the index block.
2018 * It's used in truncate case only, thus all requests are for
2019 * last index in the block only.
2021 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2022 struct ext4_ext_path *path)
2027 /* free index block */
2029 leaf = ext4_idx_pblock(path->p_idx);
2030 if (unlikely(path->p_hdr->eh_entries == 0)) {
2031 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2034 err = ext4_ext_get_access(handle, inode, path);
2037 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2038 err = ext4_ext_dirty(handle, inode, path);
2041 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2042 ext4_free_blocks(handle, inode, 0, leaf, 1,
2043 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2048 * ext4_ext_calc_credits_for_single_extent:
2049 * This routine returns max. credits that needed to insert an extent
2050 * to the extent tree.
2051 * When pass the actual path, the caller should calculate credits
2054 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2055 struct ext4_ext_path *path)
2058 int depth = ext_depth(inode);
2061 /* probably there is space in leaf? */
2062 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2063 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2066 * There are some space in the leaf tree, no
2067 * need to account for leaf block credit
2069 * bitmaps and block group descriptor blocks
2070 * and other metadat blocks still need to be
2073 /* 1 bitmap, 1 block group descriptor */
2074 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2079 return ext4_chunk_trans_blocks(inode, nrblocks);
2083 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2085 * if nrblocks are fit in a single extent (chunk flag is 1), then
2086 * in the worse case, each tree level index/leaf need to be changed
2087 * if the tree split due to insert a new extent, then the old tree
2088 * index/leaf need to be updated too
2090 * If the nrblocks are discontiguous, they could cause
2091 * the whole tree split more than once, but this is really rare.
2093 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2096 int depth = ext_depth(inode);
2106 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2107 struct ext4_extent *ex,
2108 ext4_lblk_t from, ext4_lblk_t to)
2110 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2111 int flags = EXT4_FREE_BLOCKS_FORGET;
2113 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2114 flags |= EXT4_FREE_BLOCKS_METADATA;
2115 #ifdef EXTENTS_STATS
2117 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2118 spin_lock(&sbi->s_ext_stats_lock);
2119 sbi->s_ext_blocks += ee_len;
2120 sbi->s_ext_extents++;
2121 if (ee_len < sbi->s_ext_min)
2122 sbi->s_ext_min = ee_len;
2123 if (ee_len > sbi->s_ext_max)
2124 sbi->s_ext_max = ee_len;
2125 if (ext_depth(inode) > sbi->s_depth_max)
2126 sbi->s_depth_max = ext_depth(inode);
2127 spin_unlock(&sbi->s_ext_stats_lock);
2130 if (from >= le32_to_cpu(ex->ee_block)
2131 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2136 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2137 start = ext4_ext_pblock(ex) + ee_len - num;
2138 ext_debug("free last %u blocks starting %llu\n", num, start);
2139 ext4_free_blocks(handle, inode, 0, start, num, flags);
2140 } else if (from == le32_to_cpu(ex->ee_block)
2141 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2142 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2143 from, to, le32_to_cpu(ex->ee_block), ee_len);
2145 printk(KERN_INFO "strange request: removal(2) "
2146 "%u-%u from %u:%u\n",
2147 from, to, le32_to_cpu(ex->ee_block), ee_len);
2153 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2154 struct ext4_ext_path *path, ext4_lblk_t start)
2156 int err = 0, correct_index = 0;
2157 int depth = ext_depth(inode), credits;
2158 struct ext4_extent_header *eh;
2159 ext4_lblk_t a, b, block;
2161 ext4_lblk_t ex_ee_block;
2162 unsigned short ex_ee_len;
2163 unsigned uninitialized = 0;
2164 struct ext4_extent *ex;
2166 /* the header must be checked already in ext4_ext_remove_space() */
2167 ext_debug("truncate since %u in leaf\n", start);
2168 if (!path[depth].p_hdr)
2169 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2170 eh = path[depth].p_hdr;
2171 if (unlikely(path[depth].p_hdr == NULL)) {
2172 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2175 /* find where to start removing */
2176 ex = EXT_LAST_EXTENT(eh);
2178 ex_ee_block = le32_to_cpu(ex->ee_block);
2179 ex_ee_len = ext4_ext_get_actual_len(ex);
2181 while (ex >= EXT_FIRST_EXTENT(eh) &&
2182 ex_ee_block + ex_ee_len > start) {
2184 if (ext4_ext_is_uninitialized(ex))
2189 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2190 uninitialized, ex_ee_len);
2191 path[depth].p_ext = ex;
2193 a = ex_ee_block > start ? ex_ee_block : start;
2194 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2195 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2197 ext_debug(" border %u:%u\n", a, b);
2199 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2203 } else if (a != ex_ee_block) {
2204 /* remove tail of the extent */
2205 block = ex_ee_block;
2207 } else if (b != ex_ee_block + ex_ee_len - 1) {
2208 /* remove head of the extent */
2211 /* there is no "make a hole" API yet */
2214 /* remove whole extent: excellent! */
2215 block = ex_ee_block;
2217 BUG_ON(a != ex_ee_block);
2218 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2222 * 3 for leaf, sb, and inode plus 2 (bmap and group
2223 * descriptor) for each block group; assume two block
2224 * groups plus ex_ee_len/blocks_per_block_group for
2227 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2228 if (ex == EXT_FIRST_EXTENT(eh)) {
2230 credits += (ext_depth(inode)) + 1;
2232 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2234 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2238 err = ext4_ext_get_access(handle, inode, path + depth);
2242 err = ext4_remove_blocks(handle, inode, ex, a, b);
2247 /* this extent is removed; mark slot entirely unused */
2248 ext4_ext_store_pblock(ex, 0);
2249 le16_add_cpu(&eh->eh_entries, -1);
2252 ex->ee_block = cpu_to_le32(block);
2253 ex->ee_len = cpu_to_le16(num);
2255 * Do not mark uninitialized if all the blocks in the
2256 * extent have been removed.
2258 if (uninitialized && num)
2259 ext4_ext_mark_uninitialized(ex);
2261 err = ext4_ext_dirty(handle, inode, path + depth);
2265 ext_debug("new extent: %u:%u:%llu\n", block, num,
2266 ext4_ext_pblock(ex));
2268 ex_ee_block = le32_to_cpu(ex->ee_block);
2269 ex_ee_len = ext4_ext_get_actual_len(ex);
2272 if (correct_index && eh->eh_entries)
2273 err = ext4_ext_correct_indexes(handle, inode, path);
2275 /* if this leaf is free, then we should
2276 * remove it from index block above */
2277 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2278 err = ext4_ext_rm_idx(handle, inode, path + depth);
2285 * ext4_ext_more_to_rm:
2286 * returns 1 if current index has to be freed (even partial)
2289 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2291 BUG_ON(path->p_idx == NULL);
2293 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2297 * if truncate on deeper level happened, it wasn't partial,
2298 * so we have to consider current index for truncation
2300 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2305 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2307 struct super_block *sb = inode->i_sb;
2308 int depth = ext_depth(inode);
2309 struct ext4_ext_path *path;
2313 ext_debug("truncate since %u\n", start);
2315 /* probably first extent we're gonna free will be last in block */
2316 handle = ext4_journal_start(inode, depth + 1);
2318 return PTR_ERR(handle);
2321 ext4_ext_invalidate_cache(inode);
2324 * We start scanning from right side, freeing all the blocks
2325 * after i_size and walking into the tree depth-wise.
2327 depth = ext_depth(inode);
2328 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2330 ext4_journal_stop(handle);
2333 path[0].p_depth = depth;
2334 path[0].p_hdr = ext_inode_hdr(inode);
2335 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2341 while (i >= 0 && err == 0) {
2343 /* this is leaf block */
2344 err = ext4_ext_rm_leaf(handle, inode, path, start);
2345 /* root level has p_bh == NULL, brelse() eats this */
2346 brelse(path[i].p_bh);
2347 path[i].p_bh = NULL;
2352 /* this is index block */
2353 if (!path[i].p_hdr) {
2354 ext_debug("initialize header\n");
2355 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2358 if (!path[i].p_idx) {
2359 /* this level hasn't been touched yet */
2360 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2361 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2362 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2364 le16_to_cpu(path[i].p_hdr->eh_entries));
2366 /* we were already here, see at next index */
2370 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2371 i, EXT_FIRST_INDEX(path[i].p_hdr),
2373 if (ext4_ext_more_to_rm(path + i)) {
2374 struct buffer_head *bh;
2375 /* go to the next level */
2376 ext_debug("move to level %d (block %llu)\n",
2377 i + 1, ext4_idx_pblock(path[i].p_idx));
2378 memset(path + i + 1, 0, sizeof(*path));
2379 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2381 /* should we reset i_size? */
2385 if (WARN_ON(i + 1 > depth)) {
2389 if (ext4_ext_check(inode, ext_block_hdr(bh),
2394 path[i + 1].p_bh = bh;
2396 /* save actual number of indexes since this
2397 * number is changed at the next iteration */
2398 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2401 /* we finished processing this index, go up */
2402 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2403 /* index is empty, remove it;
2404 * handle must be already prepared by the
2405 * truncatei_leaf() */
2406 err = ext4_ext_rm_idx(handle, inode, path + i);
2408 /* root level has p_bh == NULL, brelse() eats this */
2409 brelse(path[i].p_bh);
2410 path[i].p_bh = NULL;
2412 ext_debug("return to level %d\n", i);
2416 /* TODO: flexible tree reduction should be here */
2417 if (path->p_hdr->eh_entries == 0) {
2419 * truncate to zero freed all the tree,
2420 * so we need to correct eh_depth
2422 err = ext4_ext_get_access(handle, inode, path);
2424 ext_inode_hdr(inode)->eh_depth = 0;
2425 ext_inode_hdr(inode)->eh_max =
2426 cpu_to_le16(ext4_ext_space_root(inode, 0));
2427 err = ext4_ext_dirty(handle, inode, path);
2431 ext4_ext_drop_refs(path);
2435 ext4_journal_stop(handle);
2441 * called at mount time
2443 void ext4_ext_init(struct super_block *sb)
2446 * possible initialization would be here
2449 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2450 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2451 printk(KERN_INFO "EXT4-fs: file extents enabled");
2452 #ifdef AGGRESSIVE_TEST
2453 printk(", aggressive tests");
2455 #ifdef CHECK_BINSEARCH
2456 printk(", check binsearch");
2458 #ifdef EXTENTS_STATS
2463 #ifdef EXTENTS_STATS
2464 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2465 EXT4_SB(sb)->s_ext_min = 1 << 30;
2466 EXT4_SB(sb)->s_ext_max = 0;
2472 * called at umount time
2474 void ext4_ext_release(struct super_block *sb)
2476 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2479 #ifdef EXTENTS_STATS
2480 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2481 struct ext4_sb_info *sbi = EXT4_SB(sb);
2482 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2483 sbi->s_ext_blocks, sbi->s_ext_extents,
2484 sbi->s_ext_blocks / sbi->s_ext_extents);
2485 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2486 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2491 /* FIXME!! we need to try to merge to left or right after zero-out */
2492 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2494 ext4_fsblk_t ee_pblock;
2495 unsigned int ee_len;
2498 ee_len = ext4_ext_get_actual_len(ex);
2499 ee_pblock = ext4_ext_pblock(ex);
2501 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len,
2502 GFP_NOFS, BLKDEV_IFL_WAIT);
2509 #define EXT4_EXT_ZERO_LEN 7
2511 * This function is called by ext4_ext_map_blocks() if someone tries to write
2512 * to an uninitialized extent. It may result in splitting the uninitialized
2513 * extent into multiple extents (upto three - one initialized and two
2515 * There are three possibilities:
2516 * a> There is no split required: Entire extent should be initialized
2517 * b> Splits in two extents: Write is happening at either end of the extent
2518 * c> Splits in three extents: Somone is writing in middle of the extent
2520 static int ext4_ext_convert_to_initialized(handle_t *handle,
2521 struct inode *inode,
2522 struct ext4_map_blocks *map,
2523 struct ext4_ext_path *path)
2525 struct ext4_extent *ex, newex, orig_ex;
2526 struct ext4_extent *ex1 = NULL;
2527 struct ext4_extent *ex2 = NULL;
2528 struct ext4_extent *ex3 = NULL;
2529 struct ext4_extent_header *eh;
2530 ext4_lblk_t ee_block, eof_block;
2531 unsigned int allocated, ee_len, depth;
2532 ext4_fsblk_t newblock;
2537 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2538 "block %llu, max_blocks %u\n", inode->i_ino,
2539 (unsigned long long)map->m_lblk, map->m_len);
2541 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2542 inode->i_sb->s_blocksize_bits;
2543 if (eof_block < map->m_lblk + map->m_len)
2544 eof_block = map->m_lblk + map->m_len;
2546 depth = ext_depth(inode);
2547 eh = path[depth].p_hdr;
2548 ex = path[depth].p_ext;
2549 ee_block = le32_to_cpu(ex->ee_block);
2550 ee_len = ext4_ext_get_actual_len(ex);
2551 allocated = ee_len - (map->m_lblk - ee_block);
2552 newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2555 orig_ex.ee_block = ex->ee_block;
2556 orig_ex.ee_len = cpu_to_le16(ee_len);
2557 ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2560 * It is safe to convert extent to initialized via explicit
2561 * zeroout only if extent is fully insde i_size or new_size.
2563 may_zeroout = ee_block + ee_len <= eof_block;
2565 err = ext4_ext_get_access(handle, inode, path + depth);
2568 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2569 if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2570 err = ext4_ext_zeroout(inode, &orig_ex);
2572 goto fix_extent_len;
2573 /* update the extent length and mark as initialized */
2574 ex->ee_block = orig_ex.ee_block;
2575 ex->ee_len = orig_ex.ee_len;
2576 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2577 ext4_ext_dirty(handle, inode, path + depth);
2578 /* zeroed the full extent */
2582 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2583 if (map->m_lblk > ee_block) {
2585 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2586 ext4_ext_mark_uninitialized(ex1);
2590 * for sanity, update the length of the ex2 extent before
2591 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2592 * overlap of blocks.
2594 if (!ex1 && allocated > map->m_len)
2595 ex2->ee_len = cpu_to_le16(map->m_len);
2596 /* ex3: to ee_block + ee_len : uninitialised */
2597 if (allocated > map->m_len) {
2598 unsigned int newdepth;
2599 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2600 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2602 * map->m_lblk == ee_block is handled by the zerouout
2604 * Mark first half uninitialized.
2605 * Mark second half initialized and zero out the
2606 * initialized extent
2608 ex->ee_block = orig_ex.ee_block;
2609 ex->ee_len = cpu_to_le16(ee_len - allocated);
2610 ext4_ext_mark_uninitialized(ex);
2611 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2612 ext4_ext_dirty(handle, inode, path + depth);
2615 ex3->ee_block = cpu_to_le32(map->m_lblk);
2616 ext4_ext_store_pblock(ex3, newblock);
2617 ex3->ee_len = cpu_to_le16(allocated);
2618 err = ext4_ext_insert_extent(handle, inode, path,
2620 if (err == -ENOSPC) {
2621 err = ext4_ext_zeroout(inode, &orig_ex);
2623 goto fix_extent_len;
2624 ex->ee_block = orig_ex.ee_block;
2625 ex->ee_len = orig_ex.ee_len;
2626 ext4_ext_store_pblock(ex,
2627 ext4_ext_pblock(&orig_ex));
2628 ext4_ext_dirty(handle, inode, path + depth);
2629 /* blocks available from map->m_lblk */
2633 goto fix_extent_len;
2636 * We need to zero out the second half because
2637 * an fallocate request can update file size and
2638 * converting the second half to initialized extent
2639 * implies that we can leak some junk data to user
2642 err = ext4_ext_zeroout(inode, ex3);
2645 * We should actually mark the
2646 * second half as uninit and return error
2647 * Insert would have changed the extent
2649 depth = ext_depth(inode);
2650 ext4_ext_drop_refs(path);
2651 path = ext4_ext_find_extent(inode, map->m_lblk,
2654 err = PTR_ERR(path);
2657 /* get the second half extent details */
2658 ex = path[depth].p_ext;
2659 err = ext4_ext_get_access(handle, inode,
2663 ext4_ext_mark_uninitialized(ex);
2664 ext4_ext_dirty(handle, inode, path + depth);
2668 /* zeroed the second half */
2672 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2673 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2674 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2675 ext4_ext_mark_uninitialized(ex3);
2676 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2677 if (err == -ENOSPC && may_zeroout) {
2678 err = ext4_ext_zeroout(inode, &orig_ex);
2680 goto fix_extent_len;
2681 /* update the extent length and mark as initialized */
2682 ex->ee_block = orig_ex.ee_block;
2683 ex->ee_len = orig_ex.ee_len;
2684 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2685 ext4_ext_dirty(handle, inode, path + depth);
2686 /* zeroed the full extent */
2687 /* blocks available from map->m_lblk */
2691 goto fix_extent_len;
2693 * The depth, and hence eh & ex might change
2694 * as part of the insert above.
2696 newdepth = ext_depth(inode);
2698 * update the extent length after successful insert of the
2701 ee_len -= ext4_ext_get_actual_len(ex3);
2702 orig_ex.ee_len = cpu_to_le16(ee_len);
2703 may_zeroout = ee_block + ee_len <= eof_block;
2706 ext4_ext_drop_refs(path);
2707 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2709 err = PTR_ERR(path);
2712 eh = path[depth].p_hdr;
2713 ex = path[depth].p_ext;
2717 err = ext4_ext_get_access(handle, inode, path + depth);
2721 allocated = map->m_len;
2723 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2724 * to insert a extent in the middle zerout directly
2725 * otherwise give the extent a chance to merge to left
2727 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2728 map->m_lblk != ee_block && may_zeroout) {
2729 err = ext4_ext_zeroout(inode, &orig_ex);
2731 goto fix_extent_len;
2732 /* update the extent length and mark as initialized */
2733 ex->ee_block = orig_ex.ee_block;
2734 ex->ee_len = orig_ex.ee_len;
2735 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2736 ext4_ext_dirty(handle, inode, path + depth);
2737 /* zero out the first half */
2738 /* blocks available from map->m_lblk */
2743 * If there was a change of depth as part of the
2744 * insertion of ex3 above, we need to update the length
2745 * of the ex1 extent again here
2747 if (ex1 && ex1 != ex) {
2749 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2750 ext4_ext_mark_uninitialized(ex1);
2753 /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2754 ex2->ee_block = cpu_to_le32(map->m_lblk);
2755 ext4_ext_store_pblock(ex2, newblock);
2756 ex2->ee_len = cpu_to_le16(allocated);
2760 * New (initialized) extent starts from the first block
2761 * in the current extent. i.e., ex2 == ex
2762 * We have to see if it can be merged with the extent
2765 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2767 * To merge left, pass "ex2 - 1" to try_to_merge(),
2768 * since it merges towards right _only_.
2770 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2772 err = ext4_ext_correct_indexes(handle, inode, path);
2775 depth = ext_depth(inode);
2780 * Try to Merge towards right. This might be required
2781 * only when the whole extent is being written to.
2782 * i.e. ex2 == ex and ex3 == NULL.
2785 ret = ext4_ext_try_to_merge(inode, path, ex2);
2787 err = ext4_ext_correct_indexes(handle, inode, path);
2792 /* Mark modified extent as dirty */
2793 err = ext4_ext_dirty(handle, inode, path + depth);
2796 err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2797 if (err == -ENOSPC && may_zeroout) {
2798 err = ext4_ext_zeroout(inode, &orig_ex);
2800 goto fix_extent_len;
2801 /* update the extent length and mark as initialized */
2802 ex->ee_block = orig_ex.ee_block;
2803 ex->ee_len = orig_ex.ee_len;
2804 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2805 ext4_ext_dirty(handle, inode, path + depth);
2806 /* zero out the first half */
2809 goto fix_extent_len;
2811 ext4_ext_show_leaf(inode, path);
2812 return err ? err : allocated;
2815 ex->ee_block = orig_ex.ee_block;
2816 ex->ee_len = orig_ex.ee_len;
2817 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2818 ext4_ext_mark_uninitialized(ex);
2819 ext4_ext_dirty(handle, inode, path + depth);
2824 * This function is called by ext4_ext_map_blocks() from
2825 * ext4_get_blocks_dio_write() when DIO to write
2826 * to an uninitialized extent.
2828 * Writing to an uninitized extent may result in splitting the uninitialized
2829 * extent into multiple /intialized unintialized extents (up to three)
2830 * There are three possibilities:
2831 * a> There is no split required: Entire extent should be uninitialized
2832 * b> Splits in two extents: Write is happening at either end of the extent
2833 * c> Splits in three extents: Somone is writing in middle of the extent
2835 * One of more index blocks maybe needed if the extent tree grow after
2836 * the unintialized extent split. To prevent ENOSPC occur at the IO
2837 * complete, we need to split the uninitialized extent before DIO submit
2838 * the IO. The uninitialized extent called at this time will be split
2839 * into three uninitialized extent(at most). After IO complete, the part
2840 * being filled will be convert to initialized by the end_io callback function
2841 * via ext4_convert_unwritten_extents().
2843 * Returns the size of uninitialized extent to be written on success.
2845 static int ext4_split_unwritten_extents(handle_t *handle,
2846 struct inode *inode,
2847 struct ext4_map_blocks *map,
2848 struct ext4_ext_path *path,
2851 struct ext4_extent *ex, newex, orig_ex;
2852 struct ext4_extent *ex1 = NULL;
2853 struct ext4_extent *ex2 = NULL;
2854 struct ext4_extent *ex3 = NULL;
2855 ext4_lblk_t ee_block, eof_block;
2856 unsigned int allocated, ee_len, depth;
2857 ext4_fsblk_t newblock;
2861 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2862 "block %llu, max_blocks %u\n", inode->i_ino,
2863 (unsigned long long)map->m_lblk, map->m_len);
2865 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2866 inode->i_sb->s_blocksize_bits;
2867 if (eof_block < map->m_lblk + map->m_len)
2868 eof_block = map->m_lblk + map->m_len;
2870 depth = ext_depth(inode);
2871 ex = path[depth].p_ext;
2872 ee_block = le32_to_cpu(ex->ee_block);
2873 ee_len = ext4_ext_get_actual_len(ex);
2874 allocated = ee_len - (map->m_lblk - ee_block);
2875 newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2878 orig_ex.ee_block = ex->ee_block;
2879 orig_ex.ee_len = cpu_to_le16(ee_len);
2880 ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2883 * It is safe to convert extent to initialized via explicit
2884 * zeroout only if extent is fully insde i_size or new_size.
2886 may_zeroout = ee_block + ee_len <= eof_block;
2889 * If the uninitialized extent begins at the same logical
2890 * block where the write begins, and the write completely
2891 * covers the extent, then we don't need to split it.
2893 if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2896 err = ext4_ext_get_access(handle, inode, path + depth);
2899 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2900 if (map->m_lblk > ee_block) {
2902 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2903 ext4_ext_mark_uninitialized(ex1);
2907 * for sanity, update the length of the ex2 extent before
2908 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2909 * overlap of blocks.
2911 if (!ex1 && allocated > map->m_len)
2912 ex2->ee_len = cpu_to_le16(map->m_len);
2913 /* ex3: to ee_block + ee_len : uninitialised */
2914 if (allocated > map->m_len) {
2915 unsigned int newdepth;
2917 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2918 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2919 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2920 ext4_ext_mark_uninitialized(ex3);
2921 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2922 if (err == -ENOSPC && may_zeroout) {
2923 err = ext4_ext_zeroout(inode, &orig_ex);
2925 goto fix_extent_len;
2926 /* update the extent length and mark as initialized */
2927 ex->ee_block = orig_ex.ee_block;
2928 ex->ee_len = orig_ex.ee_len;
2929 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2930 ext4_ext_dirty(handle, inode, path + depth);
2931 /* zeroed the full extent */
2932 /* blocks available from map->m_lblk */
2936 goto fix_extent_len;
2938 * The depth, and hence eh & ex might change
2939 * as part of the insert above.
2941 newdepth = ext_depth(inode);
2943 * update the extent length after successful insert of the
2946 ee_len -= ext4_ext_get_actual_len(ex3);
2947 orig_ex.ee_len = cpu_to_le16(ee_len);
2948 may_zeroout = ee_block + ee_len <= eof_block;
2951 ext4_ext_drop_refs(path);
2952 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2954 err = PTR_ERR(path);
2957 ex = path[depth].p_ext;
2961 err = ext4_ext_get_access(handle, inode, path + depth);
2965 allocated = map->m_len;
2968 * If there was a change of depth as part of the
2969 * insertion of ex3 above, we need to update the length
2970 * of the ex1 extent again here
2972 if (ex1 && ex1 != ex) {
2974 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2975 ext4_ext_mark_uninitialized(ex1);
2979 * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
2980 * using direct I/O, uninitialised still.
2982 ex2->ee_block = cpu_to_le32(map->m_lblk);
2983 ext4_ext_store_pblock(ex2, newblock);
2984 ex2->ee_len = cpu_to_le16(allocated);
2985 ext4_ext_mark_uninitialized(ex2);
2988 /* Mark modified extent as dirty */
2989 err = ext4_ext_dirty(handle, inode, path + depth);
2990 ext_debug("out here\n");
2993 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2994 if (err == -ENOSPC && may_zeroout) {
2995 err = ext4_ext_zeroout(inode, &orig_ex);
2997 goto fix_extent_len;
2998 /* update the extent length and mark as initialized */
2999 ex->ee_block = orig_ex.ee_block;
3000 ex->ee_len = orig_ex.ee_len;
3001 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3002 ext4_ext_dirty(handle, inode, path + depth);
3003 /* zero out the first half */
3006 goto fix_extent_len;
3008 ext4_ext_show_leaf(inode, path);
3009 return err ? err : allocated;
3012 ex->ee_block = orig_ex.ee_block;
3013 ex->ee_len = orig_ex.ee_len;
3014 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3015 ext4_ext_mark_uninitialized(ex);
3016 ext4_ext_dirty(handle, inode, path + depth);
3019 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3020 struct inode *inode,
3021 struct ext4_ext_path *path)
3023 struct ext4_extent *ex;
3024 struct ext4_extent_header *eh;
3029 depth = ext_depth(inode);
3030 eh = path[depth].p_hdr;
3031 ex = path[depth].p_ext;
3033 err = ext4_ext_get_access(handle, inode, path + depth);
3036 /* first mark the extent as initialized */
3037 ext4_ext_mark_initialized(ex);
3040 * We have to see if it can be merged with the extent
3043 if (ex > EXT_FIRST_EXTENT(eh)) {
3045 * To merge left, pass "ex - 1" to try_to_merge(),
3046 * since it merges towards right _only_.
3048 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3050 err = ext4_ext_correct_indexes(handle, inode, path);
3053 depth = ext_depth(inode);
3058 * Try to Merge towards right.
3060 ret = ext4_ext_try_to_merge(inode, path, ex);
3062 err = ext4_ext_correct_indexes(handle, inode, path);
3065 depth = ext_depth(inode);
3067 /* Mark modified extent as dirty */
3068 err = ext4_ext_dirty(handle, inode, path + depth);
3070 ext4_ext_show_leaf(inode, path);
3074 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3075 sector_t block, int count)
3078 for (i = 0; i < count; i++)
3079 unmap_underlying_metadata(bdev, block + i);
3083 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3085 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3086 struct ext4_map_blocks *map,
3087 struct ext4_ext_path *path,
3091 struct ext4_extent_header *eh;
3092 struct ext4_extent *ex, *last_ex;
3094 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3097 depth = ext_depth(inode);
3098 eh = path[depth].p_hdr;
3099 ex = path[depth].p_ext;
3101 if (unlikely(!eh->eh_entries)) {
3102 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3103 "EOFBLOCKS_FL set");
3106 last_ex = EXT_LAST_EXTENT(eh);
3108 * We should clear the EOFBLOCKS_FL flag if we are writing the
3109 * last block in the last extent in the file. We test this by
3110 * first checking to see if the caller to
3111 * ext4_ext_get_blocks() was interested in the last block (or
3112 * a block beyond the last block) in the current extent. If
3113 * this turns out to be false, we can bail out from this
3114 * function immediately.
3116 if (map->m_lblk + len < le32_to_cpu(last_ex->ee_block) +
3117 ext4_ext_get_actual_len(last_ex))
3120 * If the caller does appear to be planning to write at or
3121 * beyond the end of the current extent, we then test to see
3122 * if the current extent is the last extent in the file, by
3123 * checking to make sure it was reached via the rightmost node
3124 * at each level of the tree.
3126 for (i = depth-1; i >= 0; i--)
3127 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3129 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3130 return ext4_mark_inode_dirty(handle, inode);
3134 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3135 struct ext4_map_blocks *map,
3136 struct ext4_ext_path *path, int flags,
3137 unsigned int allocated, ext4_fsblk_t newblock)
3141 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3143 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3144 "block %llu, max_blocks %u, flags %d, allocated %u",
3145 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3147 ext4_ext_show_leaf(inode, path);
3149 /* get_block() before submit the IO, split the extent */
3150 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3151 ret = ext4_split_unwritten_extents(handle, inode, map,
3154 * Flag the inode(non aio case) or end_io struct (aio case)
3155 * that this IO needs to convertion to written when IO is
3159 io->flag = EXT4_IO_END_UNWRITTEN;
3161 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3162 if (ext4_should_dioread_nolock(inode))
3163 map->m_flags |= EXT4_MAP_UNINIT;
3166 /* IO end_io complete, convert the filled extent to written */
3167 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3168 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3171 ext4_update_inode_fsync_trans(handle, inode, 1);
3172 err = check_eofblocks_fl(handle, inode, map, path,
3178 /* buffered IO case */
3180 * repeat fallocate creation request
3181 * we already have an unwritten extent
3183 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3186 /* buffered READ or buffered write_begin() lookup */
3187 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3189 * We have blocks reserved already. We
3190 * return allocated blocks so that delalloc
3191 * won't do block reservation for us. But
3192 * the buffer head will be unmapped so that
3193 * a read from the block returns 0s.
3195 map->m_flags |= EXT4_MAP_UNWRITTEN;
3199 /* buffered write, writepage time, convert*/
3200 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3202 ext4_update_inode_fsync_trans(handle, inode, 1);
3203 err = check_eofblocks_fl(handle, inode, map, path, map->m_len);
3214 map->m_flags |= EXT4_MAP_NEW;
3216 * if we allocated more blocks than requested
3217 * we need to make sure we unmap the extra block
3218 * allocated. The actual needed block will get
3219 * unmapped later when we find the buffer_head marked
3222 if (allocated > map->m_len) {
3223 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3224 newblock + map->m_len,
3225 allocated - map->m_len);
3226 allocated = map->m_len;
3230 * If we have done fallocate with the offset that is already
3231 * delayed allocated, we would have block reservation
3232 * and quota reservation done in the delayed write path.
3233 * But fallocate would have already updated quota and block
3234 * count for this offset. So cancel these reservation
3236 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3237 ext4_da_update_reserve_space(inode, allocated, 0);
3240 map->m_flags |= EXT4_MAP_MAPPED;
3242 if (allocated > map->m_len)
3243 allocated = map->m_len;
3244 ext4_ext_show_leaf(inode, path);
3245 map->m_pblk = newblock;
3246 map->m_len = allocated;
3249 ext4_ext_drop_refs(path);
3252 return err ? err : allocated;
3256 * Block allocation/map/preallocation routine for extents based files
3259 * Need to be called with
3260 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3261 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3263 * return > 0, number of of blocks already mapped/allocated
3264 * if create == 0 and these are pre-allocated blocks
3265 * buffer head is unmapped
3266 * otherwise blocks are mapped
3268 * return = 0, if plain look up failed (blocks have not been allocated)
3269 * buffer head is unmapped
3271 * return < 0, error case.
3273 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3274 struct ext4_map_blocks *map, int flags)
3276 struct ext4_ext_path *path = NULL;
3277 struct ext4_extent_header *eh;
3278 struct ext4_extent newex, *ex;
3279 ext4_fsblk_t newblock;
3280 int err = 0, depth, ret, cache_type;
3281 unsigned int allocated = 0;
3282 struct ext4_allocation_request ar;
3283 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3285 ext_debug("blocks %u/%u requested for inode %lu\n",
3286 map->m_lblk, map->m_len, inode->i_ino);
3288 /* check in cache */
3289 cache_type = ext4_ext_in_cache(inode, map->m_lblk, &newex);
3291 if (cache_type == EXT4_EXT_CACHE_GAP) {
3292 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3294 * block isn't allocated yet and
3295 * user doesn't want to allocate it
3299 /* we should allocate requested block */
3300 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3301 /* block is already allocated */
3302 newblock = map->m_lblk
3303 - le32_to_cpu(newex.ee_block)
3304 + ext4_ext_pblock(&newex);
3305 /* number of remaining blocks in the extent */
3306 allocated = ext4_ext_get_actual_len(&newex) -
3307 (map->m_lblk - le32_to_cpu(newex.ee_block));
3314 /* find extent for this block */
3315 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3317 err = PTR_ERR(path);
3322 depth = ext_depth(inode);
3325 * consistent leaf must not be empty;
3326 * this situation is possible, though, _during_ tree modification;
3327 * this is why assert can't be put in ext4_ext_find_extent()
3329 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3330 EXT4_ERROR_INODE(inode, "bad extent address "
3331 "lblock: %lu, depth: %d pblock %lld",
3332 (unsigned long) map->m_lblk, depth,
3333 path[depth].p_block);
3337 eh = path[depth].p_hdr;
3339 ex = path[depth].p_ext;
3341 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3342 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3343 unsigned short ee_len;
3346 * Uninitialized extents are treated as holes, except that
3347 * we split out initialized portions during a write.
3349 ee_len = ext4_ext_get_actual_len(ex);
3350 /* if found extent covers block, simply return it */
3351 if (in_range(map->m_lblk, ee_block, ee_len)) {
3352 newblock = map->m_lblk - ee_block + ee_start;
3353 /* number of remaining blocks in the extent */
3354 allocated = ee_len - (map->m_lblk - ee_block);
3355 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3356 ee_block, ee_len, newblock);
3358 /* Do not put uninitialized extent in the cache */
3359 if (!ext4_ext_is_uninitialized(ex)) {
3360 ext4_ext_put_in_cache(inode, ee_block,
3362 EXT4_EXT_CACHE_EXTENT);
3365 ret = ext4_ext_handle_uninitialized_extents(handle,
3366 inode, map, path, flags, allocated,
3373 * requested block isn't allocated yet;
3374 * we couldn't try to create block if create flag is zero
3376 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3378 * put just found gap into cache to speed up
3379 * subsequent requests
3381 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3385 * Okay, we need to do block allocation.
3388 /* find neighbour allocated blocks */
3389 ar.lleft = map->m_lblk;
3390 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3393 ar.lright = map->m_lblk;
3394 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3399 * See if request is beyond maximum number of blocks we can have in
3400 * a single extent. For an initialized extent this limit is
3401 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3402 * EXT_UNINIT_MAX_LEN.
3404 if (map->m_len > EXT_INIT_MAX_LEN &&
3405 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3406 map->m_len = EXT_INIT_MAX_LEN;
3407 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3408 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3409 map->m_len = EXT_UNINIT_MAX_LEN;
3411 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3412 newex.ee_block = cpu_to_le32(map->m_lblk);
3413 newex.ee_len = cpu_to_le16(map->m_len);
3414 err = ext4_ext_check_overlap(inode, &newex, path);
3416 allocated = ext4_ext_get_actual_len(&newex);
3418 allocated = map->m_len;
3420 /* allocate new block */
3422 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3423 ar.logical = map->m_lblk;
3425 if (S_ISREG(inode->i_mode))
3426 ar.flags = EXT4_MB_HINT_DATA;
3428 /* disable in-core preallocation for non-regular files */
3430 newblock = ext4_mb_new_blocks(handle, &ar, &err);
3433 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3434 ar.goal, newblock, allocated);
3436 /* try to insert new extent into found leaf and return */
3437 ext4_ext_store_pblock(&newex, newblock);
3438 newex.ee_len = cpu_to_le16(ar.len);
3439 /* Mark uninitialized */
3440 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3441 ext4_ext_mark_uninitialized(&newex);
3443 * io_end structure was created for every IO write to an
3444 * uninitialized extent. To avoid unecessary conversion,
3445 * here we flag the IO that really needs the conversion.
3446 * For non asycn direct IO case, flag the inode state
3447 * that we need to perform convertion when IO is done.
3449 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3451 io->flag = EXT4_IO_END_UNWRITTEN;
3453 ext4_set_inode_state(inode,
3454 EXT4_STATE_DIO_UNWRITTEN);
3456 if (ext4_should_dioread_nolock(inode))
3457 map->m_flags |= EXT4_MAP_UNINIT;
3460 err = check_eofblocks_fl(handle, inode, map, path, ar.len);
3464 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3466 /* free data blocks we just allocated */
3467 /* not a good idea to call discard here directly,
3468 * but otherwise we'd need to call it every free() */
3469 ext4_discard_preallocations(inode);
3470 ext4_free_blocks(handle, inode, 0, ext4_ext_pblock(&newex),
3471 ext4_ext_get_actual_len(&newex), 0);
3475 /* previous routine could use block we allocated */
3476 newblock = ext4_ext_pblock(&newex);
3477 allocated = ext4_ext_get_actual_len(&newex);
3478 if (allocated > map->m_len)
3479 allocated = map->m_len;
3480 map->m_flags |= EXT4_MAP_NEW;
3483 * Update reserved blocks/metadata blocks after successful
3484 * block allocation which had been deferred till now.
3486 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3487 ext4_da_update_reserve_space(inode, allocated, 1);
3490 * Cache the extent and update transaction to commit on fdatasync only
3491 * when it is _not_ an uninitialized extent.
3493 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3494 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock,
3495 EXT4_EXT_CACHE_EXTENT);
3496 ext4_update_inode_fsync_trans(handle, inode, 1);
3498 ext4_update_inode_fsync_trans(handle, inode, 0);
3500 if (allocated > map->m_len)
3501 allocated = map->m_len;
3502 ext4_ext_show_leaf(inode, path);
3503 map->m_flags |= EXT4_MAP_MAPPED;
3504 map->m_pblk = newblock;
3505 map->m_len = allocated;
3508 ext4_ext_drop_refs(path);
3511 return err ? err : allocated;
3514 void ext4_ext_truncate(struct inode *inode)
3516 struct address_space *mapping = inode->i_mapping;
3517 struct super_block *sb = inode->i_sb;
3518 ext4_lblk_t last_block;
3523 * probably first extent we're gonna free will be last in block
3525 err = ext4_writepage_trans_blocks(inode);
3526 handle = ext4_journal_start(inode, err);
3530 if (inode->i_size & (sb->s_blocksize - 1))
3531 ext4_block_truncate_page(handle, mapping, inode->i_size);
3533 if (ext4_orphan_add(handle, inode))
3536 down_write(&EXT4_I(inode)->i_data_sem);
3537 ext4_ext_invalidate_cache(inode);
3539 ext4_discard_preallocations(inode);
3542 * TODO: optimization is possible here.
3543 * Probably we need not scan at all,
3544 * because page truncation is enough.
3547 /* we have to know where to truncate from in crash case */
3548 EXT4_I(inode)->i_disksize = inode->i_size;
3549 ext4_mark_inode_dirty(handle, inode);
3551 last_block = (inode->i_size + sb->s_blocksize - 1)
3552 >> EXT4_BLOCK_SIZE_BITS(sb);
3553 err = ext4_ext_remove_space(inode, last_block);
3555 /* In a multi-transaction truncate, we only make the final
3556 * transaction synchronous.
3559 ext4_handle_sync(handle);
3562 up_write(&EXT4_I(inode)->i_data_sem);
3564 * If this was a simple ftruncate() and the file will remain alive,
3565 * then we need to clear up the orphan record which we created above.
3566 * However, if this was a real unlink then we were called by
3567 * ext4_delete_inode(), and we allow that function to clean up the
3568 * orphan info for us.
3571 ext4_orphan_del(handle, inode);
3573 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3574 ext4_mark_inode_dirty(handle, inode);
3575 ext4_journal_stop(handle);
3578 static void ext4_falloc_update_inode(struct inode *inode,
3579 int mode, loff_t new_size, int update_ctime)
3581 struct timespec now;
3584 now = current_fs_time(inode->i_sb);
3585 if (!timespec_equal(&inode->i_ctime, &now))
3586 inode->i_ctime = now;
3589 * Update only when preallocation was requested beyond
3592 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3593 if (new_size > i_size_read(inode))
3594 i_size_write(inode, new_size);
3595 if (new_size > EXT4_I(inode)->i_disksize)
3596 ext4_update_i_disksize(inode, new_size);
3599 * Mark that we allocate beyond EOF so the subsequent truncate
3600 * can proceed even if the new size is the same as i_size.
3602 if (new_size > i_size_read(inode))
3603 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3609 * preallocate space for a file. This implements ext4's fallocate inode
3610 * operation, which gets called from sys_fallocate system call.
3611 * For block-mapped files, posix_fallocate should fall back to the method
3612 * of writing zeroes to the required new blocks (the same behavior which is
3613 * expected for file systems which do not support fallocate() system call).
3615 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3619 unsigned int max_blocks;
3623 struct ext4_map_blocks map;
3624 unsigned int credits, blkbits = inode->i_blkbits;
3627 * currently supporting (pre)allocate mode for extent-based
3630 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3633 /* preallocation to directories is currently not supported */
3634 if (S_ISDIR(inode->i_mode))
3637 map.m_lblk = offset >> blkbits;
3639 * We can't just convert len to max_blocks because
3640 * If blocksize = 4096 offset = 3072 and len = 2048
3642 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3645 * credits to insert 1 extent into extent tree
3647 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3648 mutex_lock(&inode->i_mutex);
3649 ret = inode_newsize_ok(inode, (len + offset));
3651 mutex_unlock(&inode->i_mutex);
3655 while (ret >= 0 && ret < max_blocks) {
3656 map.m_lblk = map.m_lblk + ret;
3657 map.m_len = max_blocks = max_blocks - ret;
3658 handle = ext4_journal_start(inode, credits);
3659 if (IS_ERR(handle)) {
3660 ret = PTR_ERR(handle);
3663 ret = ext4_map_blocks(handle, inode, &map,
3664 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3668 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3669 "returned error inode#%lu, block=%u, "
3670 "max_blocks=%u", __func__,
3671 inode->i_ino, block, max_blocks);
3673 ext4_mark_inode_dirty(handle, inode);
3674 ret2 = ext4_journal_stop(handle);
3677 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3678 blkbits) >> blkbits))
3679 new_size = offset + len;
3681 new_size = (map.m_lblk + ret) << blkbits;
3683 ext4_falloc_update_inode(inode, mode, new_size,
3684 (map.m_flags & EXT4_MAP_NEW));
3685 ext4_mark_inode_dirty(handle, inode);
3686 ret2 = ext4_journal_stop(handle);
3690 if (ret == -ENOSPC &&
3691 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3695 mutex_unlock(&inode->i_mutex);
3696 return ret > 0 ? ret2 : ret;
3700 * This function convert a range of blocks to written extents
3701 * The caller of this function will pass the start offset and the size.
3702 * all unwritten extents within this range will be converted to
3705 * This function is called from the direct IO end io call back
3706 * function, to convert the fallocated extents after IO is completed.
3707 * Returns 0 on success.
3709 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3713 unsigned int max_blocks;
3716 struct ext4_map_blocks map;
3717 unsigned int credits, blkbits = inode->i_blkbits;
3719 map.m_lblk = offset >> blkbits;
3721 * We can't just convert len to max_blocks because
3722 * If blocksize = 4096 offset = 3072 and len = 2048
3724 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3727 * credits to insert 1 extent into extent tree
3729 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3730 while (ret >= 0 && ret < max_blocks) {
3732 map.m_len = (max_blocks -= ret);
3733 handle = ext4_journal_start(inode, credits);
3734 if (IS_ERR(handle)) {
3735 ret = PTR_ERR(handle);
3738 ret = ext4_map_blocks(handle, inode, &map,
3739 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3742 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3743 "returned error inode#%lu, block=%u, "
3744 "max_blocks=%u", __func__,
3745 inode->i_ino, map.m_lblk, map.m_len);
3747 ext4_mark_inode_dirty(handle, inode);
3748 ret2 = ext4_journal_stop(handle);
3749 if (ret <= 0 || ret2 )
3752 return ret > 0 ? ret2 : ret;
3755 * Callback function called for each extent to gather FIEMAP information.
3757 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3758 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3761 struct fiemap_extent_info *fieinfo = data;
3762 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3769 logical = (__u64)newex->ec_block << blksize_bits;
3771 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3774 struct buffer_head *bh = NULL;
3776 offset = logical >> PAGE_SHIFT;
3777 page = find_get_page(inode->i_mapping, offset);
3778 if (!page || !page_has_buffers(page))
3779 return EXT_CONTINUE;
3781 bh = page_buffers(page);
3784 return EXT_CONTINUE;
3786 if (buffer_delay(bh)) {
3787 flags |= FIEMAP_EXTENT_DELALLOC;
3788 page_cache_release(page);
3790 page_cache_release(page);
3791 return EXT_CONTINUE;
3795 physical = (__u64)newex->ec_start << blksize_bits;
3796 length = (__u64)newex->ec_len << blksize_bits;
3798 if (ex && ext4_ext_is_uninitialized(ex))
3799 flags |= FIEMAP_EXTENT_UNWRITTEN;
3802 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3804 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3805 * this also indicates no more allocated blocks.
3807 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3809 if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3810 newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3811 loff_t size = i_size_read(inode);
3812 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3814 flags |= FIEMAP_EXTENT_LAST;
3815 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3816 logical+length > size)
3817 length = (size - logical + bs - 1) & ~(bs-1);
3820 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3827 return EXT_CONTINUE;
3830 /* fiemap flags we can handle specified here */
3831 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3833 static int ext4_xattr_fiemap(struct inode *inode,
3834 struct fiemap_extent_info *fieinfo)
3838 __u32 flags = FIEMAP_EXTENT_LAST;
3839 int blockbits = inode->i_sb->s_blocksize_bits;
3843 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3844 struct ext4_iloc iloc;
3845 int offset; /* offset of xattr in inode */
3847 error = ext4_get_inode_loc(inode, &iloc);
3850 physical = iloc.bh->b_blocknr << blockbits;
3851 offset = EXT4_GOOD_OLD_INODE_SIZE +
3852 EXT4_I(inode)->i_extra_isize;
3854 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3855 flags |= FIEMAP_EXTENT_DATA_INLINE;
3857 } else { /* external block */
3858 physical = EXT4_I(inode)->i_file_acl << blockbits;
3859 length = inode->i_sb->s_blocksize;
3863 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3865 return (error < 0 ? error : 0);
3868 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3869 __u64 start, __u64 len)
3871 ext4_lblk_t start_blk;
3874 /* fallback to generic here if not in extents fmt */
3875 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3876 return generic_block_fiemap(inode, fieinfo, start, len,
3879 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3882 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3883 error = ext4_xattr_fiemap(inode, fieinfo);
3885 ext4_lblk_t len_blks;
3888 start_blk = start >> inode->i_sb->s_blocksize_bits;
3889 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3890 if (last_blk >= EXT_MAX_BLOCK)
3891 last_blk = EXT_MAX_BLOCK-1;
3892 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3895 * Walk the extent tree gathering extent information.
3896 * ext4_ext_fiemap_cb will push extents back to user.
3898 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3899 ext4_ext_fiemap_cb, fieinfo);